265 files changed, 112060 insertions, 12543 deletions

diff --git a/drivers/infiniband/hw/Makefile b/drivers/infiniband/hw/Makefile
new file mode 100644
index 00000000000..e900b03531a
--- /dev/null
+++ b/drivers/infiniband/hw/Makefile
@@ -0,0 +1,12 @@
+obj-$(CONFIG_INFINIBAND_MTHCA)		+= mthca/
+obj-$(CONFIG_INFINIBAND_IPATH)		+= ipath/
+obj-$(CONFIG_INFINIBAND_QIB)		+= qib/
+obj-$(CONFIG_INFINIBAND_EHCA)		+= ehca/
+obj-$(CONFIG_INFINIBAND_AMSO1100)	+= amso1100/
+obj-$(CONFIG_INFINIBAND_CXGB3)		+= cxgb3/
+obj-$(CONFIG_INFINIBAND_CXGB4)		+= cxgb4/
+obj-$(CONFIG_MLX4_INFINIBAND)		+= mlx4/
+obj-$(CONFIG_MLX5_INFINIBAND)		+= mlx5/
+obj-$(CONFIG_INFINIBAND_NES)		+= nes/
+obj-$(CONFIG_INFINIBAND_OCRDMA)		+= ocrdma/
+obj-$(CONFIG_INFINIBAND_USNIC)		+= usnic/
diff --git a/drivers/infiniband/hw/amso1100/Kbuild b/drivers/infiniband/hw/amso1100/Kbuild
index 06964c4af84..950dfabcd89 100644
--- a/drivers/infiniband/hw/amso1100/Kbuild
+++ b/drivers/infiniband/hw/amso1100/Kbuild
@@ -1,6 +1,4 @@
-ifdef CONFIG_INFINIBAND_AMSO1100_DEBUG
-EXTRA_CFLAGS += -DDEBUG
-endif
+ccflags-$(CONFIG_INFINIBAND_AMSO1100_DEBUG) := -DDEBUG
 
 obj-$(CONFIG_INFINIBAND_AMSO1100) += iw_c2.o
 
diff --git a/drivers/infiniband/hw/amso1100/c2.c b/drivers/infiniband/hw/amso1100/c2.c
index 113f3c03c5b..00400c352c1 100644
--- a/drivers/infiniband/hw/amso1100/c2.c
+++ b/drivers/infiniband/hw/amso1100/c2.c
@@ -36,6 +36,7 @@
 #include <linux/netdevice.h>
 #include <linux/etherdevice.h>
 #include <linux/inetdevice.h>
+#include <linux/interrupt.h>
 #include <linux/delay.h>
 #include <linux/ethtool.h>
 #include <linux/mii.h>
@@ -46,6 +47,8 @@
 #include <linux/tcp.h>
 #include <linux/init.h>
 #include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/prefetch.h>
 
 #include <asm/io.h>
 #include <asm/irq.h>
@@ -76,7 +79,6 @@ static irqreturn_t c2_interrupt(int irq, void *dev_id);
 static void c2_tx_timeout(struct net_device *netdev);
 static int c2_change_mtu(struct net_device *netdev, int new_mtu);
 static void c2_reset(struct c2_port *c2_port);
-static struct net_device_stats *c2_get_stats(struct net_device *netdev);
 
 static struct pci_device_id c2_pci_table[] = {
 	{ PCI_DEVICE(0x18b8, 0xb001) },
@@ -87,11 +89,7 @@ MODULE_DEVICE_TABLE(pci, c2_pci_table);
 
 static void c2_print_macaddr(struct net_device *netdev)
 {
-	pr_debug("%s: MAC %02X:%02X:%02X:%02X:%02X:%02X, "
-		"IRQ %u\n", netdev->name,
-		netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
-		netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5],
-		netdev->irq);
+	pr_debug("%s: MAC %pM, IRQ %u\n", netdev->name, netdev->dev_addr, netdev->irq);
 }
 
 static void c2_set_rxbufsize(struct c2_port *c2_port)
@@ -349,7 +347,7 @@ static void c2_tx_clean(struct c2_port *c2_port)
 					     elem->hw_desc + C2_TXP_ADDR);
 				__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_DONE),
 					     elem->hw_desc + C2_TXP_FLAGS);
-				c2_port->netstats.tx_dropped++;
+				c2_port->netdev->stats.tx_dropped++;
 				break;
 			} else {
 				__raw_writew(0,
@@ -457,7 +455,7 @@ static void c2_rx_error(struct c2_port *c2_port, struct c2_element *elem)
 		     elem->hw_desc + C2_RXP_FLAGS);
 
 	pr_debug("packet dropped\n");
-	c2_port->netstats.rx_dropped++;
+	c2_port->netdev->stats.rx_dropped++;
 }
 
 static void c2_rx_interrupt(struct net_device *netdev)
@@ -531,9 +529,8 @@ static void c2_rx_interrupt(struct net_device *netdev)
 
 		netif_rx(skb);
 
-		netdev->last_rx = jiffies;
-		c2_port->netstats.rx_packets++;
-		c2_port->netstats.rx_bytes += buflen;
+		netdev->stats.rx_packets++;
+		netdev->stats.rx_bytes += buflen;
 	}
 
 	/* Save where we left off */
@@ -797,19 +794,16 @@ static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY),
 		     elem->hw_desc + C2_TXP_FLAGS);
 
-	c2_port->netstats.tx_packets++;
-	c2_port->netstats.tx_bytes += maplen;
+	netdev->stats.tx_packets++;
+	netdev->stats.tx_bytes += maplen;
 
 	/* Loop thru additional data fragments and queue them */
 	if (skb_shinfo(skb)->nr_frags) {
 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
-			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
-			maplen = frag->size;
-			mapaddr =
-			    pci_map_page(c2dev->pcidev, frag->page,
-					 frag->page_offset, maplen,
-					 PCI_DMA_TODEVICE);
-
+			const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+			maplen = skb_frag_size(frag);
+			mapaddr = skb_frag_dma_map(&c2dev->pcidev->dev, frag,
+						   0, maplen, DMA_TO_DEVICE);
 			elem = elem->next;
 			elem->skb = NULL;
 			elem->mapaddr = mapaddr;
@@ -823,8 +817,8 @@ static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 			__raw_writew((__force u16) cpu_to_be16(TXP_HTXD_READY),
 				     elem->hw_desc + C2_TXP_FLAGS);
 
-			c2_port->netstats.tx_packets++;
-			c2_port->netstats.tx_bytes += maplen;
+			netdev->stats.tx_packets++;
+			netdev->stats.tx_bytes += maplen;
 		}
 	}
 
@@ -845,13 +839,6 @@ static int c2_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 	return NETDEV_TX_OK;
 }
 
-static struct net_device_stats *c2_get_stats(struct net_device *netdev)
-{
-	struct c2_port *c2_port = netdev_priv(netdev);
-
-	return &c2_port->netstats;
-}
-
 static void c2_tx_timeout(struct net_device *netdev)
 {
 	struct c2_port *c2_port = netdev_priv(netdev);
@@ -880,6 +867,16 @@ static int c2_change_mtu(struct net_device *netdev, int new_mtu)
 	return ret;
 }
 
+static const struct net_device_ops c2_netdev = {
+	.ndo_open 		= c2_up,
+	.ndo_stop 		= c2_down,
+	.ndo_start_xmit		= c2_xmit_frame,
+	.ndo_tx_timeout		= c2_tx_timeout,
+	.ndo_change_mtu		= c2_change_mtu,
+	.ndo_set_mac_address 	= eth_mac_addr,
+	.ndo_validate_addr	= eth_validate_addr,
+};
+
 /* Initialize network device */
 static struct net_device *c2_devinit(struct c2_dev *c2dev,
 				     void __iomem * mmio_addr)
@@ -894,12 +891,7 @@ static struct net_device *c2_devinit(struct c2_dev *c2dev,
 
 	SET_NETDEV_DEV(netdev, &c2dev->pcidev->dev);
 
-	netdev->open = c2_up;
-	netdev->stop = c2_down;
-	netdev->hard_start_xmit = c2_xmit_frame;
-	netdev->get_stats = c2_get_stats;
-	netdev->tx_timeout = c2_tx_timeout;
-	netdev->change_mtu = c2_change_mtu;
+	netdev->netdev_ops = &c2_netdev;
 	netdev->watchdog_timeo = C2_TX_TIMEOUT;
 	netdev->irq = c2dev->pcidev->irq;
 
@@ -928,8 +920,7 @@ static struct net_device *c2_devinit(struct c2_dev *c2dev,
 	return netdev;
 }
 
-static int __devinit c2_probe(struct pci_dev *pcidev,
-			      const struct pci_device_id *ent)
+static int c2_probe(struct pci_dev *pcidev, const struct pci_device_id *ent)
 {
 	int ret = 0, i;
 	unsigned long reg0_start, reg0_flags, reg0_len;
@@ -992,13 +983,13 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 	}
 
 	if ((sizeof(dma_addr_t) > 4)) {
-		ret = pci_set_dma_mask(pcidev, DMA_64BIT_MASK);
+		ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(64));
 		if (ret < 0) {
 			printk(KERN_ERR PFX "64b DMA configuration failed\n");
 			goto bail2;
 		}
 	} else {
-		ret = pci_set_dma_mask(pcidev, DMA_32BIT_MASK);
+		ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(32));
 		if (ret < 0) {
 			printk(KERN_ERR PFX "32b DMA configuration failed\n");
 			goto bail2;
@@ -1091,6 +1082,7 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 
 	/* Initialize network device */
 	if ((netdev = c2_devinit(c2dev, mmio_regs)) == NULL) {
+		ret = -ENOMEM;
 		iounmap(mmio_regs);
 		goto bail4;
 	}
@@ -1160,7 +1152,8 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 		goto bail10;
 	}
 
-	if (c2_register_device(c2dev))
+	ret = c2_register_device(c2dev);
+	if (ret)
 		goto bail10;
 
 	return 0;
@@ -1199,7 +1192,7 @@ static int __devinit c2_probe(struct pci_dev *pcidev,
 	return ret;
 }
 
-static void __devexit c2_remove(struct pci_dev *pcidev)
+static void c2_remove(struct pci_dev *pcidev)
 {
 	struct c2_dev *c2dev = pci_get_drvdata(pcidev);
 	struct net_device *netdev = c2dev->netdev;
@@ -1244,18 +1237,7 @@ static struct pci_driver c2_pci_driver = {
 	.name = DRV_NAME,
 	.id_table = c2_pci_table,
 	.probe = c2_probe,
-	.remove = __devexit_p(c2_remove),
+	.remove = c2_remove,
 };
 
-static int __init c2_init_module(void)
-{
-	return pci_register_driver(&c2_pci_driver);
-}
-
-static void __exit c2_exit_module(void)
-{
-	pci_unregister_driver(&c2_pci_driver);
-}
-
-module_init(c2_init_module);
-module_exit(c2_exit_module);
+module_pci_driver(c2_pci_driver);
diff --git a/drivers/infiniband/hw/amso1100/c2.h b/drivers/infiniband/hw/amso1100/c2.h
index d12a24a84fd..d619d735838 100644
--- a/drivers/infiniband/hw/amso1100/c2.h
+++ b/drivers/infiniband/hw/amso1100/c2.h
@@ -250,7 +250,7 @@ struct c2_array {
 struct sp_chunk {
 	struct sp_chunk *next;
 	dma_addr_t dma_addr;
-	DECLARE_PCI_UNMAP_ADDR(mapping);
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 	u16 head;
 	u16 shared_ptr[0];
 };
@@ -265,7 +265,6 @@ struct c2_pd_table {
 struct c2_qp_table {
 	struct idr idr;
 	spinlock_t lock;
-	int last;
 };
 
 struct c2_element {
@@ -369,8 +368,6 @@ struct c2_port {
 	unsigned long mem_size;
 
 	u32 rx_buf_size;
-
-	struct net_device_stats netstats;
 };
 
 /*
@@ -500,16 +497,16 @@ extern int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 			struct ib_send_wr **bad_wr);
 extern int c2_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr,
 			   struct ib_recv_wr **bad_wr);
-extern void __devinit c2_init_qp_table(struct c2_dev *c2dev);
-extern void __devexit c2_cleanup_qp_table(struct c2_dev *c2dev);
+extern void c2_init_qp_table(struct c2_dev *c2dev);
+extern void c2_cleanup_qp_table(struct c2_dev *c2dev);
 extern void c2_set_qp_state(struct c2_qp *, int);
 extern struct c2_qp *c2_find_qpn(struct c2_dev *c2dev, int qpn);
 
 /* PDs */
 extern int c2_pd_alloc(struct c2_dev *c2dev, int privileged, struct c2_pd *pd);
 extern void c2_pd_free(struct c2_dev *c2dev, struct c2_pd *pd);
-extern int __devinit c2_init_pd_table(struct c2_dev *c2dev);
-extern void __devexit c2_cleanup_pd_table(struct c2_dev *c2dev);
+extern int c2_init_pd_table(struct c2_dev *c2dev);
+extern void c2_cleanup_pd_table(struct c2_dev *c2dev);
 
 /* CQs */
 extern int c2_init_cq(struct c2_dev *c2dev, int entries,
diff --git a/drivers/infiniband/hw/amso1100/c2_ae.c b/drivers/infiniband/hw/amso1100/c2_ae.c
index 62af74295db..cedda25232b 100644
--- a/drivers/infiniband/hw/amso1100/c2_ae.c
+++ b/drivers/infiniband/hw/amso1100/c2_ae.c
@@ -141,7 +141,7 @@ static const char *to_qp_state_str(int state)
 		return "C2_QP_STATE_ERROR";
 	default:
 		return "<invalid QP state>";
-	};
+	}
 }
 
 void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
@@ -155,9 +155,11 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 	enum c2_event_id event_id;
 	unsigned long flags;
 	int status;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_event.local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_event.remote_addr;
 
 	/*
-	 * retreive the message
+	 * retrieve the message
 	 */
 	wr = c2_mq_consume(mq);
 	if (!wr)
@@ -206,10 +208,10 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 		case CCAE_ACTIVE_CONNECT_RESULTS:
 			res = &wr->ae.ae_active_connect_results;
 			cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
-			cm_event.local_addr.sin_addr.s_addr = res->laddr;
-			cm_event.remote_addr.sin_addr.s_addr = res->raddr;
-			cm_event.local_addr.sin_port = res->lport;
-			cm_event.remote_addr.sin_port =	res->rport;
+			laddr->sin_addr.s_addr = res->laddr;
+			raddr->sin_addr.s_addr = res->raddr;
+			laddr->sin_port = res->lport;
+			raddr->sin_port = res->rport;
 			if (status == 0) {
 				cm_event.private_data_len =
 					be32_to_cpu(res->private_data_length);
@@ -281,13 +283,18 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 		}
 		cm_event.event = IW_CM_EVENT_CONNECT_REQUEST;
 		cm_event.provider_data = (void*)(unsigned long)req->cr_handle;
-		cm_event.local_addr.sin_addr.s_addr = req->laddr;
-		cm_event.remote_addr.sin_addr.s_addr = req->raddr;
-		cm_event.local_addr.sin_port = req->lport;
-		cm_event.remote_addr.sin_port = req->rport;
+		laddr->sin_addr.s_addr = req->laddr;
+		raddr->sin_addr.s_addr = req->raddr;
+		laddr->sin_port = req->lport;
+		raddr->sin_port = req->rport;
 		cm_event.private_data_len =
 			be32_to_cpu(req->private_data_length);
 		cm_event.private_data = req->private_data;
+		/*
+		 * Until ird/ord negotiation via MPAv2 support is added, send
+		 * max supported values
+		 */
+		cm_event.ird = cm_event.ord = 128;
 
 		if (cm_id->event_handler)
 			cm_id->event_handler(cm_id, &cm_event);
@@ -306,6 +313,7 @@ void c2_ae_event(struct c2_dev *c2dev, u32 mq_index)
 		if (cq->ibcq.event_handler)
 			cq->ibcq.event_handler(&ib_event,
 					       cq->ibcq.cq_context);
+		break;
 	}
 
 	default:
diff --git a/drivers/infiniband/hw/amso1100/c2_alloc.c b/drivers/infiniband/hw/amso1100/c2_alloc.c
index e9110163aef..78d247ec696 100644
--- a/drivers/infiniband/hw/amso1100/c2_alloc.c
+++ b/drivers/infiniband/hw/amso1100/c2_alloc.c
@@ -32,7 +32,6 @@
  */
 
 #include <linux/errno.h>
-#include <linux/slab.h>
 #include <linux/bitmap.h>
 
 #include "c2.h"
@@ -50,7 +49,7 @@ static int c2_alloc_mqsp_chunk(struct c2_dev *c2dev, gfp_t gfp_mask,
 		return -ENOMEM;
 
 	new_head->dma_addr = dma_addr;
-	pci_unmap_addr_set(new_head, mapping, new_head->dma_addr);
+	dma_unmap_addr_set(new_head, mapping, new_head->dma_addr);
 
 	new_head->next = NULL;
 	new_head->head = 0;
@@ -82,7 +81,7 @@ void c2_free_mqsp_pool(struct c2_dev *c2dev, struct sp_chunk *root)
 	while (root) {
 		next = root->next;
 		dma_free_coherent(&c2dev->pcidev->dev, PAGE_SIZE, root,
-				  pci_unmap_addr(root, mapping));
+				  dma_unmap_addr(root, mapping));
 		root = next;
 	}
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_cm.c b/drivers/infiniband/hw/amso1100/c2_cm.c
index 75b93e9b881..23bfa94fbd4 100644
--- a/drivers/infiniband/hw/amso1100/c2_cm.c
+++ b/drivers/infiniband/hw/amso1100/c2_cm.c
@@ -31,6 +31,8 @@
  * SOFTWARE.
  *
  */
+#include <linux/slab.h>
+
 #include "c2.h"
 #include "c2_wr.h"
 #include "c2_vq.h"
@@ -44,6 +46,10 @@ int c2_llp_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
 	struct c2wr_qp_connect_req *wr;	/* variable size needs a malloc. */
 	struct c2_vq_req *vq_req;
 	int err;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
+
+	if (cm_id->remote_addr.ss_family != AF_INET)
+		return -ENOSYS;
 
 	ibqp = c2_get_qp(cm_id->device, iw_param->qpn);
 	if (!ibqp)
@@ -89,8 +95,8 @@ int c2_llp_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *iw_param)
 	wr->rnic_handle = c2dev->adapter_handle;
 	wr->qp_handle = qp->adapter_handle;
 
-	wr->remote_addr = cm_id->remote_addr.sin_addr.s_addr;
-	wr->remote_port = cm_id->remote_addr.sin_port;
+	wr->remote_addr = raddr->sin_addr.s_addr;
+	wr->remote_port = raddr->sin_port;
 
 	/*
 	 * Move any private data from the callers's buf into
@@ -133,6 +139,10 @@ int c2_llp_service_create(struct iw_cm_id *cm_id, int backlog)
 	struct c2wr_ep_listen_create_rep *reply;
 	struct c2_vq_req *vq_req;
 	int err;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
+
+	if (cm_id->local_addr.ss_family != AF_INET)
+		return -ENOSYS;
 
 	c2dev = to_c2dev(cm_id->device);
 	if (c2dev == NULL)
@@ -151,8 +161,8 @@ int c2_llp_service_create(struct iw_cm_id *cm_id, int backlog)
 	c2_wr_set_id(&wr, CCWR_EP_LISTEN_CREATE);
 	wr.hdr.context = (u64) (unsigned long) vq_req;
 	wr.rnic_handle = c2dev->adapter_handle;
-	wr.local_addr = cm_id->local_addr.sin_addr.s_addr;
-	wr.local_port = cm_id->local_addr.sin_port;
+	wr.local_addr = laddr->sin_addr.s_addr;
+	wr.local_port = laddr->sin_port;
 	wr.backlog = cpu_to_be32(backlog);
 	wr.user_context = (u64) (unsigned long) cm_id;
 
diff --git a/drivers/infiniband/hw/amso1100/c2_cq.c b/drivers/infiniband/hw/amso1100/c2_cq.c
index bb17cce3cb5..49e0e8533f7 100644
--- a/drivers/infiniband/hw/amso1100/c2_cq.c
+++ b/drivers/infiniband/hw/amso1100/c2_cq.c
@@ -35,6 +35,8 @@
  * SOFTWARE.
  *
  */
+#include <linux/gfp.h>
+
 #include "c2.h"
 #include "c2_vq.h"
 #include "c2_status.h"
@@ -133,7 +135,7 @@ static inline int c2_poll_one(struct c2_dev *c2dev,
 	struct c2_qp *qp;
 	int is_recv = 0;
 
-	ce = (struct c2wr_ce *) c2_mq_consume(&cq->mq);
+	ce = c2_mq_consume(&cq->mq);
 	if (!ce) {
 		return -EAGAIN;
 	}
@@ -146,7 +148,7 @@ static inline int c2_poll_one(struct c2_dev *c2dev,
 	while ((qp =
 		(struct c2_qp *) (unsigned long) ce->qp_user_context) == NULL) {
 		c2_mq_free(&cq->mq);
-		ce = (struct c2wr_ce *) c2_mq_consume(&cq->mq);
+		ce = c2_mq_consume(&cq->mq);
 		if (!ce)
 			return -EAGAIN;
 	}
@@ -255,7 +257,7 @@ int c2_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
 static void c2_free_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq)
 {
 	dma_free_coherent(&c2dev->pcidev->dev, mq->q_size * mq->msg_size,
-			  mq->msg_pool.host, pci_unmap_addr(mq, mapping));
+			  mq->msg_pool.host, dma_unmap_addr(mq, mapping));
 }
 
 static int c2_alloc_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq, int q_size,
@@ -276,7 +278,7 @@ static int c2_alloc_cq_buf(struct c2_dev *c2dev, struct c2_mq *mq, int q_size,
 		       NULL,	/* peer (currently unknown) */
 		       C2_MQ_HOST_TARGET);
 
-	pci_unmap_addr_set(mq, mapping, mq->host_dma);
+	dma_unmap_addr_set(mq, mapping, mq->host_dma);
 
 	return 0;
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_intr.c b/drivers/infiniband/hw/amso1100/c2_intr.c
index 3b5095470cb..3a17d9b36db 100644
--- a/drivers/infiniband/hw/amso1100/c2_intr.c
+++ b/drivers/infiniband/hw/amso1100/c2_intr.c
@@ -62,8 +62,8 @@ void c2_rnic_interrupt(struct c2_dev *c2dev)
 static void handle_mq(struct c2_dev *c2dev, u32 mq_index)
 {
 	if (c2dev->qptr_array[mq_index] == NULL) {
-		pr_debug(KERN_INFO "handle_mq: stray activity for mq_index=%d\n",
-			mq_index);
+		pr_debug("handle_mq: stray activity for mq_index=%d\n",
+			 mq_index);
 		return;
 	}
 
@@ -169,7 +169,8 @@ static void handle_vq(struct c2_dev *c2dev, u32 mq_index)
 		 * We should never get here, as the adapter should
 		 * never send us a reply that we're not expecting.
 		 */
-		vq_repbuf_free(c2dev, host_msg);
+		if (reply_msg != NULL)
+			vq_repbuf_free(c2dev, host_msg);
 		pr_debug("handle_vq: UNEXPECTEDLY got NULL req\n");
 		return;
 	}
@@ -183,6 +184,11 @@ static void handle_vq(struct c2_dev *c2dev, u32 mq_index)
 	case IW_CM_EVENT_ESTABLISHED:
 		c2_set_qp_state(req->qp,
 				C2_QP_STATE_RTS);
+		/*
+		 * Until ird/ord negotiation via MPAv2 support is added, send
+		 * max supported values
+		 */
+		cm_event.ird = cm_event.ord = 128;
 	case IW_CM_EVENT_CLOSE:
 
 		/*
diff --git a/drivers/infiniband/hw/amso1100/c2_mm.c b/drivers/infiniband/hw/amso1100/c2_mm.c
index b506fe22b4d..119c4f3d979 100644
--- a/drivers/infiniband/hw/amso1100/c2_mm.c
+++ b/drivers/infiniband/hw/amso1100/c2_mm.c
@@ -30,6 +30,8 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include <linux/slab.h>
+
 #include "c2.h"
 #include "c2_vq.h"
 
diff --git a/drivers/infiniband/hw/amso1100/c2_mq.h b/drivers/infiniband/hw/amso1100/c2_mq.h
index acede007b94..fc1b9a7cec4 100644
--- a/drivers/infiniband/hw/amso1100/c2_mq.h
+++ b/drivers/infiniband/hw/amso1100/c2_mq.h
@@ -71,7 +71,7 @@ struct c2_mq {
 		u8 __iomem *adapter;
 	} msg_pool;
 	dma_addr_t host_dma;
-	DECLARE_PCI_UNMAP_ADDR(mapping);
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 	u16 hint_count;
 	u16 priv;
 	struct c2_mq_shared __iomem *peer;
diff --git a/drivers/infiniband/hw/amso1100/c2_pd.c b/drivers/infiniband/hw/amso1100/c2_pd.c
index 00c709926c8..f3e81dc357b 100644
--- a/drivers/infiniband/hw/amso1100/c2_pd.c
+++ b/drivers/infiniband/hw/amso1100/c2_pd.c
@@ -34,6 +34,7 @@
  */
 
 #include <linux/init.h>
+#include <linux/slab.h>
 #include <linux/errno.h>
 
 #include "c2.h"
@@ -69,7 +70,7 @@ void c2_pd_free(struct c2_dev *c2dev, struct c2_pd *pd)
 	spin_unlock(&c2dev->pd_table.lock);
 }
 
-int __devinit c2_init_pd_table(struct c2_dev *c2dev)
+int c2_init_pd_table(struct c2_dev *c2dev)
 {
 
 	c2dev->pd_table.last = 0;
@@ -83,7 +84,7 @@ int __devinit c2_init_pd_table(struct c2_dev *c2dev)
 	return 0;
 }
 
-void __devexit c2_cleanup_pd_table(struct c2_dev *c2dev)
+void c2_cleanup_pd_table(struct c2_dev *c2dev)
 {
 	kfree(c2dev->pd_table.table);
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.c b/drivers/infiniband/hw/amso1100/c2_provider.c
index 2acf9b62cf9..8af33cf1fc4 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.c
+++ b/drivers/infiniband/hw/amso1100/c2_provider.c
@@ -50,6 +50,7 @@
 #include <linux/dma-mapping.h>
 #include <linux/if_arp.h>
 #include <linux/vmalloc.h>
+#include <linux/slab.h>
 
 #include <asm/io.h>
 #include <asm/irq.h>
@@ -93,19 +94,11 @@ static int c2_query_port(struct ib_device *ibdev,
 	props->pkey_tbl_len = 1;
 	props->qkey_viol_cntr = 0;
 	props->active_width = 1;
-	props->active_speed = 1;
+	props->active_speed = IB_SPEED_SDR;
 
 	return 0;
 }
 
-static int c2_modify_port(struct ib_device *ibdev,
-			  u8 port, int port_modify_mask,
-			  struct ib_port_modify *props)
-{
-	pr_debug("%s:%u\n", __func__, __LINE__);
-	return 0;
-}
-
 static int c2_query_pkey(struct ib_device *ibdev,
 			 u8 port, u16 index, u16 * pkey)
 {
@@ -272,7 +265,6 @@ static struct ib_qp *c2_create_qp(struct ib_pd *pd,
 		pr_debug("%s: Invalid QP type: %d\n", __func__,
 			init_attr->qp_type);
 		return ERR_PTR(-EINVAL);
-		break;
 	}
 
 	if (err) {
@@ -439,9 +431,9 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	u64 *pages;
 	u64 kva = 0;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct c2_pd *c2pd = to_c2pd(pd);
 	struct c2_mr *c2mr;
 
@@ -460,10 +452,7 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	shift = ffs(c2mr->umem->page_size) - 1;
-
-	n = 0;
-	list_for_each_entry(chunk, &c2mr->umem->chunk_list, list)
-		n += chunk->nents;
+	n = c2mr->umem->nmap;
 
 	pages = kmalloc(n * sizeof(u64), GFP_KERNEL);
 	if (!pages) {
@@ -472,14 +461,12 @@ static struct ib_mr *c2_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	i = 0;
-	list_for_each_entry(chunk, &c2mr->umem->chunk_list, list) {
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] =
-					sg_dma_address(&chunk->page_list[j]) +
-					(c2mr->umem->page_size * k);
-			}
+	for_each_sg(c2mr->umem->sg_head.sgl, sg, c2mr->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] =
+				sg_dma_address(sg) +
+				(c2mr->umem->page_size * k);
 		}
 	}
 
@@ -654,7 +641,7 @@ static int c2_service_destroy(struct iw_cm_id *cm_id)
 static int c2_pseudo_up(struct net_device *netdev)
 {
 	struct in_device *ind;
-	struct c2_dev *c2dev = netdev->priv;
+	struct c2_dev *c2dev = netdev->ml_priv;
 
 	ind = in_dev_get(netdev);
 	if (!ind)
@@ -679,7 +666,7 @@ static int c2_pseudo_up(struct net_device *netdev)
 static int c2_pseudo_down(struct net_device *netdev)
 {
 	struct in_device *ind;
-	struct c2_dev *c2dev = netdev->priv;
+	struct c2_dev *c2dev = netdev->ml_priv;
 
 	ind = in_dev_get(netdev);
 	if (!ind)
@@ -709,26 +696,27 @@ static int c2_pseudo_xmit_frame(struct sk_buff *skb, struct net_device *netdev)
 
 static int c2_pseudo_change_mtu(struct net_device *netdev, int new_mtu)
 {
-	int ret = 0;
-
 	if (new_mtu < ETH_ZLEN || new_mtu > ETH_JUMBO_MTU)
 		return -EINVAL;
 
 	netdev->mtu = new_mtu;
 
 	/* TODO: Tell rnic about new rmda interface mtu */
-	return ret;
+	return 0;
 }
 
+static const struct net_device_ops c2_pseudo_netdev_ops = {
+	.ndo_open 		= c2_pseudo_up,
+	.ndo_stop 		= c2_pseudo_down,
+	.ndo_start_xmit 	= c2_pseudo_xmit_frame,
+	.ndo_change_mtu 	= c2_pseudo_change_mtu,
+	.ndo_validate_addr	= eth_validate_addr,
+};
+
 static void setup(struct net_device *netdev)
 {
-	netdev->open = c2_pseudo_up;
-	netdev->stop = c2_pseudo_down;
-	netdev->hard_start_xmit = c2_pseudo_xmit_frame;
-	netdev->get_stats = NULL;
-	netdev->tx_timeout = NULL;
-	netdev->set_mac_address = NULL;
-	netdev->change_mtu = c2_pseudo_change_mtu;
+	netdev->netdev_ops = &c2_pseudo_netdev_ops;
+
 	netdev->watchdog_timeo = 0;
 	netdev->type = ARPHRD_ETHER;
 	netdev->mtu = 1500;
@@ -736,7 +724,6 @@ static void setup(struct net_device *netdev)
 	netdev->addr_len = ETH_ALEN;
 	netdev->tx_queue_len = 0;
 	netdev->flags |= IFF_NOARP;
-	return;
 }
 
 static struct net_device *c2_pseudo_netdev_init(struct c2_dev *c2dev)
@@ -747,24 +734,21 @@ static struct net_device *c2_pseudo_netdev_init(struct c2_dev *c2dev)
 	/* change ethxxx to iwxxx */
 	strcpy(name, "iw");
 	strcat(name, &c2dev->netdev->name[3]);
-	netdev = alloc_netdev(sizeof(*netdev), name, setup);
+	netdev = alloc_netdev(0, name, setup);
 	if (!netdev) {
 		printk(KERN_ERR PFX "%s -  etherdev alloc failed",
 			__func__);
 		return NULL;
 	}
 
-	netdev->priv = c2dev;
+	netdev->ml_priv = c2dev;
 
 	SET_NETDEV_DEV(netdev, &c2dev->pcidev->dev);
 
 	memcpy_fromio(netdev->dev_addr, c2dev->kva + C2_REGS_RDMA_ENADDR, 6);
 
 	/* Print out the MAC address */
-	pr_debug("%s: MAC %02X:%02X:%02X:%02X:%02X:%02X\n",
-		netdev->name,
-		netdev->dev_addr[0], netdev->dev_addr[1], netdev->dev_addr[2],
-		netdev->dev_addr[3], netdev->dev_addr[4], netdev->dev_addr[5]);
+	pr_debug("%s: MAC %pM\n", netdev->name, netdev->dev_addr);
 
 #if 0
 	/* Disable network packets */
@@ -781,11 +765,11 @@ int c2_register_device(struct c2_dev *dev)
 	/* Register pseudo network device */
 	dev->pseudo_netdev = c2_pseudo_netdev_init(dev);
 	if (!dev->pseudo_netdev)
-		goto out3;
+		goto out;
 
 	ret = register_netdev(dev->pseudo_netdev);
 	if (ret)
-		goto out2;
+		goto out_free_netdev;
 
 	pr_debug("%s:%u\n", __func__, __LINE__);
 	strlcpy(dev->ibdev.name, "amso%d", IB_DEVICE_NAME_MAX);
@@ -817,7 +801,6 @@ int c2_register_device(struct c2_dev *dev)
 	dev->ibdev.dma_device = &dev->pcidev->dev;
 	dev->ibdev.query_device = c2_query_device;
 	dev->ibdev.query_port = c2_query_port;
-	dev->ibdev.modify_port = c2_modify_port;
 	dev->ibdev.query_pkey = c2_query_pkey;
 	dev->ibdev.query_gid = c2_query_gid;
 	dev->ibdev.alloc_ucontext = c2_alloc_ucontext;
@@ -852,6 +835,10 @@ int c2_register_device(struct c2_dev *dev)
 	dev->ibdev.post_recv = c2_post_receive;
 
 	dev->ibdev.iwcm = kmalloc(sizeof(*dev->ibdev.iwcm), GFP_KERNEL);
+	if (dev->ibdev.iwcm == NULL) {
+		ret = -ENOMEM;
+		goto out_unregister_netdev;
+	}
 	dev->ibdev.iwcm->add_ref = c2_add_ref;
 	dev->ibdev.iwcm->rem_ref = c2_rem_ref;
 	dev->ibdev.iwcm->get_qp = c2_get_qp;
@@ -861,25 +848,27 @@ int c2_register_device(struct c2_dev *dev)
 	dev->ibdev.iwcm->create_listen = c2_service_create;
 	dev->ibdev.iwcm->destroy_listen = c2_service_destroy;
 
-	ret = ib_register_device(&dev->ibdev);
+	ret = ib_register_device(&dev->ibdev, NULL);
 	if (ret)
-		goto out1;
+		goto out_free_iwcm;
 
 	for (i = 0; i < ARRAY_SIZE(c2_dev_attributes); ++i) {
 		ret = device_create_file(&dev->ibdev.dev,
 					       c2_dev_attributes[i]);
 		if (ret)
-			goto out0;
+			goto out_unregister_ibdev;
 	}
-	goto out3;
+	goto out;
 
-out0:
+out_unregister_ibdev:
 	ib_unregister_device(&dev->ibdev);
-out1:
+out_free_iwcm:
+	kfree(dev->ibdev.iwcm);
+out_unregister_netdev:
 	unregister_netdev(dev->pseudo_netdev);
-out2:
+out_free_netdev:
 	free_netdev(dev->pseudo_netdev);
-out3:
+out:
 	pr_debug("%s:%u ret=%d\n", __func__, __LINE__, ret);
 	return ret;
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_provider.h b/drivers/infiniband/hw/amso1100/c2_provider.h
index 1076df2ee96..bf189987711 100644
--- a/drivers/infiniband/hw/amso1100/c2_provider.h
+++ b/drivers/infiniband/hw/amso1100/c2_provider.h
@@ -50,7 +50,7 @@
 
 struct c2_buf_list {
 	void *buf;
-	 DECLARE_PCI_UNMAP_ADDR(mapping)
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 };
 
 
diff --git a/drivers/infiniband/hw/amso1100/c2_qp.c b/drivers/infiniband/hw/amso1100/c2_qp.c
index a6d89440ad2..86708dee58b 100644
--- a/drivers/infiniband/hw/amso1100/c2_qp.c
+++ b/drivers/infiniband/hw/amso1100/c2_qp.c
@@ -36,6 +36,7 @@
  */
 
 #include <linux/delay.h>
+#include <linux/gfp.h>
 
 #include "c2.h"
 #include "c2_vq.h"
@@ -381,14 +382,16 @@ static int c2_alloc_qpn(struct c2_dev *c2dev, struct c2_qp *qp)
 {
 	int ret;
 
-        do {
-		spin_lock_irq(&c2dev->qp_table.lock);
-		ret = idr_get_new_above(&c2dev->qp_table.idr, qp,
-					c2dev->qp_table.last++, &qp->qpn);
-		spin_unlock_irq(&c2dev->qp_table.lock);
-        } while ((ret == -EAGAIN) &&
-	 	 idr_pre_get(&c2dev->qp_table.idr, GFP_KERNEL));
-	return ret;
+	idr_preload(GFP_KERNEL);
+	spin_lock_irq(&c2dev->qp_table.lock);
+
+	ret = idr_alloc_cyclic(&c2dev->qp_table.idr, qp, 0, 0, GFP_NOWAIT);
+	if (ret >= 0)
+		qp->qpn = ret;
+
+	spin_unlock_irq(&c2dev->qp_table.lock);
+	idr_preload_end();
+	return ret < 0 ? ret : 0;
 }
 
 static void c2_free_qpn(struct c2_dev *c2dev, int qpn)
@@ -611,7 +614,7 @@ void c2_free_qp(struct c2_dev *c2dev, struct c2_qp *qp)
 	c2_unlock_cqs(send_cq, recv_cq);
 
 	/*
-	 * Destory qp in the rnic...
+	 * Destroy qp in the rnic...
 	 */
 	destroy_qp(c2dev, qp);
 
@@ -798,8 +801,10 @@ int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 	u8 actual_sge_count;
 	u32 msg_size;
 
-	if (qp->state > IB_QPS_RTS)
-		return -EINVAL;
+	if (qp->state > IB_QPS_RTS) {
+		err = -EINVAL;
+		goto out;
+	}
 
 	while (ib_wr) {
 
@@ -930,6 +935,7 @@ int c2_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 		ib_wr = ib_wr->next;
 	}
 
+out:
 	if (err)
 		*bad_wr = ib_wr;
 	return err;
@@ -944,8 +950,10 @@ int c2_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr,
 	unsigned long lock_flags;
 	int err = 0;
 
-	if (qp->state > IB_QPS_RTS)
-		return -EINVAL;
+	if (qp->state > IB_QPS_RTS) {
+		err = -EINVAL;
+		goto out;
+	}
 
 	/*
 	 * Try and post each work request
@@ -998,18 +1006,19 @@ int c2_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr,
 		ib_wr = ib_wr->next;
 	}
 
+out:
 	if (err)
 		*bad_wr = ib_wr;
 	return err;
 }
 
-void __devinit c2_init_qp_table(struct c2_dev *c2dev)
+void c2_init_qp_table(struct c2_dev *c2dev)
 {
 	spin_lock_init(&c2dev->qp_table.lock);
 	idr_init(&c2dev->qp_table.idr);
 }
 
-void __devexit c2_cleanup_qp_table(struct c2_dev *c2dev)
+void c2_cleanup_qp_table(struct c2_dev *c2dev)
 {
 	idr_destroy(&c2dev->qp_table.idr);
 }
diff --git a/drivers/infiniband/hw/amso1100/c2_rnic.c b/drivers/infiniband/hw/amso1100/c2_rnic.c
index dd05c483564..d2a6d961344 100644
--- a/drivers/infiniband/hw/amso1100/c2_rnic.c
+++ b/drivers/infiniband/hw/amso1100/c2_rnic.c
@@ -51,6 +51,7 @@
 #include <linux/mm.h>
 #include <linux/inet.h>
 #include <linux/vmalloc.h>
+#include <linux/slab.h>
 
 #include <linux/route.h>
 
@@ -438,10 +439,10 @@ static int c2_rnic_close(struct c2_dev *c2dev)
 
 /*
  * Called by c2_probe to initialize the RNIC. This principally
- * involves initalizing the various limits and resouce pools that
+ * involves initializing the various limits and resource pools that
  * comprise the RNIC instance.
  */
-int __devinit c2_rnic_init(struct c2_dev *c2dev)
+int c2_rnic_init(struct c2_dev *c2dev)
 {
 	int err;
 	u32 qsize, msgsize;
@@ -458,13 +459,12 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 	     IB_DEVICE_MEM_WINDOW);
 
 	/* Allocate the qptr_array */
-	c2dev->qptr_array = vmalloc(C2_MAX_CQS * sizeof(void *));
+	c2dev->qptr_array = vzalloc(C2_MAX_CQS * sizeof(void *));
 	if (!c2dev->qptr_array) {
 		return -ENOMEM;
 	}
 
-	/* Inialize the qptr_array */
-	memset(c2dev->qptr_array, 0, C2_MAX_CQS * sizeof(void *));
+	/* Initialize the qptr_array */
 	c2dev->qptr_array[0] = (void *) &c2dev->req_vq;
 	c2dev->qptr_array[1] = (void *) &c2dev->rep_vq;
 	c2dev->qptr_array[2] = (void *) &c2dev->aeq;
@@ -523,7 +523,7 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 		err = -ENOMEM;
 		goto bail1;
 	}
-	pci_unmap_addr_set(&c2dev->rep_vq, mapping, c2dev->rep_vq.host_dma);
+	dma_unmap_addr_set(&c2dev->rep_vq, mapping, c2dev->rep_vq.host_dma);
 	pr_debug("%s rep_vq va %p dma %llx\n", __func__, q1_pages,
 		 (unsigned long long) c2dev->rep_vq.host_dma);
 	c2_mq_rep_init(&c2dev->rep_vq,
@@ -544,7 +544,7 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 		err = -ENOMEM;
 		goto bail2;
 	}
-	pci_unmap_addr_set(&c2dev->aeq, mapping, c2dev->aeq.host_dma);
+	dma_unmap_addr_set(&c2dev->aeq, mapping, c2dev->aeq.host_dma);
 	pr_debug("%s aeq va %p dma %llx\n", __func__, q2_pages,
 		 (unsigned long long) c2dev->aeq.host_dma);
 	c2_mq_rep_init(&c2dev->aeq,
@@ -576,7 +576,8 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 		goto bail4;
 
 	/* Initialize cached the adapter limits */
-	if (c2_rnic_query(c2dev, &c2dev->props))
+	err = c2_rnic_query(c2dev, &c2dev->props);
+	if (err)
 		goto bail5;
 
 	/* Initialize the PD pool */
@@ -595,11 +596,11 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
       bail3:
 	dma_free_coherent(&c2dev->pcidev->dev,
 			  c2dev->aeq.q_size * c2dev->aeq.msg_size,
-			  q2_pages, pci_unmap_addr(&c2dev->aeq, mapping));
+			  q2_pages, dma_unmap_addr(&c2dev->aeq, mapping));
       bail2:
 	dma_free_coherent(&c2dev->pcidev->dev,
 			  c2dev->rep_vq.q_size * c2dev->rep_vq.msg_size,
-			  q1_pages, pci_unmap_addr(&c2dev->rep_vq, mapping));
+			  q1_pages, dma_unmap_addr(&c2dev->rep_vq, mapping));
       bail1:
 	c2_free_mqsp_pool(c2dev, c2dev->kern_mqsp_pool);
       bail0:
@@ -611,7 +612,7 @@ int __devinit c2_rnic_init(struct c2_dev *c2dev)
 /*
  * Called by c2_remove to cleanup the RNIC resources.
  */
-void __devexit c2_rnic_term(struct c2_dev *c2dev)
+void c2_rnic_term(struct c2_dev *c2dev)
 {
 
 	/* Close the open adapter instance */
@@ -636,13 +637,13 @@ void __devexit c2_rnic_term(struct c2_dev *c2dev)
 	dma_free_coherent(&c2dev->pcidev->dev,
 			  c2dev->aeq.q_size * c2dev->aeq.msg_size,
 			  c2dev->aeq.msg_pool.host,
-			  pci_unmap_addr(&c2dev->aeq, mapping));
+			  dma_unmap_addr(&c2dev->aeq, mapping));
 
 	/* Free the verbs reply queue */
 	dma_free_coherent(&c2dev->pcidev->dev,
 			  c2dev->rep_vq.q_size * c2dev->rep_vq.msg_size,
 			  c2dev->rep_vq.msg_pool.host,
-			  pci_unmap_addr(&c2dev->rep_vq, mapping));
+			  dma_unmap_addr(&c2dev->rep_vq, mapping));
 
 	/* Free the MQ shared pointer pool */
 	c2_free_mqsp_pool(c2dev, c2dev->kern_mqsp_pool);
diff --git a/drivers/infiniband/hw/amso1100/c2_vq.c b/drivers/infiniband/hw/amso1100/c2_vq.c
index 9ce7819b7b2..2ec716fb2ed 100644
--- a/drivers/infiniband/hw/amso1100/c2_vq.c
+++ b/drivers/infiniband/hw/amso1100/c2_vq.c
@@ -107,7 +107,7 @@ struct c2_vq_req *vq_req_alloc(struct c2_dev *c2dev)
 	r = kmalloc(sizeof(struct c2_vq_req), GFP_KERNEL);
 	if (r) {
 		init_waitqueue_head(&r->wait_object);
-		r->reply_msg = (u64) NULL;
+		r->reply_msg = 0;
 		r->event = 0;
 		r->cm_id = NULL;
 		r->qp = NULL;
@@ -123,7 +123,7 @@ struct c2_vq_req *vq_req_alloc(struct c2_dev *c2dev)
  */
 void vq_req_free(struct c2_dev *c2dev, struct c2_vq_req *r)
 {
-	r->reply_msg = (u64) NULL;
+	r->reply_msg = 0;
 	if (atomic_dec_and_test(&r->refcnt)) {
 		kfree(r);
 	}
@@ -151,7 +151,7 @@ void vq_req_get(struct c2_dev *c2dev, struct c2_vq_req *r)
 void vq_req_put(struct c2_dev *c2dev, struct c2_vq_req *r)
 {
 	if (atomic_dec_and_test(&r->refcnt)) {
-		if (r->reply_msg != (u64) NULL)
+		if (r->reply_msg != 0)
 			vq_repbuf_free(c2dev,
 				       (void *) (unsigned long) r->reply_msg);
 		kfree(r);
diff --git a/drivers/infiniband/hw/amso1100/c2_wr.h b/drivers/infiniband/hw/amso1100/c2_wr.h
index c65fbdd6e46..8d4b4ca463c 100644
--- a/drivers/infiniband/hw/amso1100/c2_wr.h
+++ b/drivers/infiniband/hw/amso1100/c2_wr.h
@@ -131,7 +131,7 @@ enum c2wr_ids {
 	 *          All the preceding IDs are fixed, and must not change.
 	 *          You can add new IDs, but must not remove or reorder
 	 *          any IDs. If you do, YOU will ruin any hope of
-	 *          compatability between versions.
+	 *          compatibility between versions.
 	 */
 	CCWR_LAST,
 
@@ -242,7 +242,7 @@ enum c2_acf {
 /*
  *  to fix bug 1815 we define the max size allowable of the
  *  terminate message (per the IETF spec).Refer to the IETF
- *  protocal specification, section 12.1.6, page 64)
+ *  protocol specification, section 12.1.6, page 64)
  *  The message is prefixed by 20 types of DDP info.
  *
  *  Then the message has 6 bytes for the terminate control
diff --git a/drivers/infiniband/hw/cxgb3/Kconfig b/drivers/infiniband/hw/cxgb3/Kconfig
index 2acec3fadf6..2b6352b8548 100644
--- a/drivers/infiniband/hw/cxgb3/Kconfig
+++ b/drivers/infiniband/hw/cxgb3/Kconfig
@@ -10,7 +10,7 @@ config INFINIBAND_CXGB3
 	  our website at <http://www.chelsio.com>.
 
 	  For customer support, please visit our customer support page at
-	  <http://www.chelsio.com/support.htm>.
+	  <http://www.chelsio.com/support.html>.
 
 	  Please send feedback to <linux-bugs@chelsio.com>.
 
diff --git a/drivers/infiniband/hw/cxgb3/Makefile b/drivers/infiniband/hw/cxgb3/Makefile
index 7e7b5a66f04..2761364185a 100644
--- a/drivers/infiniband/hw/cxgb3/Makefile
+++ b/drivers/infiniband/hw/cxgb3/Makefile
@@ -1,10 +1,8 @@
-EXTRA_CFLAGS += -Idrivers/net/cxgb3
+ccflags-y := -Idrivers/net/ethernet/chelsio/cxgb3
 
 obj-$(CONFIG_INFINIBAND_CXGB3) += iw_cxgb3.o
 
 iw_cxgb3-y :=  iwch_cm.o iwch_ev.o iwch_cq.o iwch_qp.o iwch_mem.o \
 	       iwch_provider.o iwch.o cxio_hal.o cxio_resource.o
 
-ifdef CONFIG_INFINIBAND_CXGB3_DEBUG
-EXTRA_CFLAGS += -DDEBUG
-endif
+ccflags-$(CONFIG_INFINIBAND_CXGB3_DEBUG) += -DDEBUG
diff --git a/drivers/infiniband/hw/cxgb3/cxio_dbg.c b/drivers/infiniband/hw/cxgb3/cxio_dbg.c
index a8d24d53f30..8bca6b4ec9a 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_dbg.c
+++ b/drivers/infiniband/hw/cxgb3/cxio_dbg.c
@@ -31,6 +31,7 @@
  */
 #ifdef DEBUG
 #include <linux/types.h>
+#include <linux/slab.h>
 #include "common.h"
 #include "cxgb3_ioctl.h"
 #include "cxio_hal.h"
diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.c b/drivers/infiniband/hw/cxgb3/cxio_hal.c
index f6d5747153a..de1c61b417d 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_hal.c
+++ b/drivers/infiniband/hw/cxgb3/cxio_hal.c
@@ -37,6 +37,7 @@
 #include <linux/spinlock.h>
 #include <linux/pci.h>
 #include <linux/dma-mapping.h>
+#include <linux/slab.h>
 #include <net/net_namespace.h>
 
 #include "cxio_resource.h"
@@ -109,7 +110,6 @@ int cxio_hal_cq_op(struct cxio_rdev *rdev_p, struct t3_cq *cq,
 		while (!CQ_VLD_ENTRY(rptr, cq->size_log2, cqe)) {
 			udelay(1);
 			if (i++ > 1000000) {
-				BUG_ON(1);
 				printk(KERN_ERR "%s: stalled rnic\n",
 				       rdev_p->dev_name);
 				return -EIO;
@@ -152,29 +152,30 @@ static int cxio_hal_clear_qp_ctx(struct cxio_rdev *rdev_p, u32 qpid)
 	sge_cmd = qpid << 8 | 3;
 	wqe->sge_cmd = cpu_to_be64(sge_cmd);
 	skb->priority = CPL_PRIORITY_CONTROL;
-	return (cxgb3_ofld_send(rdev_p->t3cdev_p, skb));
+	return iwch_cxgb3_ofld_send(rdev_p->t3cdev_p, skb);
 }
 
-int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq)
+int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq, int kernel)
 {
 	struct rdma_cq_setup setup;
 	int size = (1UL << (cq->size_log2)) * sizeof(struct t3_cqe);
 
+	size += 1; /* one extra page for storing cq-in-err state */
 	cq->cqid = cxio_hal_get_cqid(rdev_p->rscp);
 	if (!cq->cqid)
 		return -ENOMEM;
-	cq->sw_queue = kzalloc(size, GFP_KERNEL);
-	if (!cq->sw_queue)
-		return -ENOMEM;
-	cq->queue = dma_alloc_coherent(&(rdev_p->rnic_info.pdev->dev),
-					     (1UL << (cq->size_log2)) *
-					     sizeof(struct t3_cqe),
+	if (kernel) {
+		cq->sw_queue = kzalloc(size, GFP_KERNEL);
+		if (!cq->sw_queue)
+			return -ENOMEM;
+	}
+	cq->queue = dma_alloc_coherent(&(rdev_p->rnic_info.pdev->dev), size,
 					     &(cq->dma_addr), GFP_KERNEL);
 	if (!cq->queue) {
 		kfree(cq->sw_queue);
 		return -ENOMEM;
 	}
-	pci_unmap_addr_set(cq, mapping, cq->dma_addr);
+	dma_unmap_addr_set(cq, mapping, cq->dma_addr);
 	memset(cq->queue, 0, size);
 	setup.id = cq->cqid;
 	setup.base_addr = (u64) (cq->dma_addr);
@@ -188,6 +189,7 @@ int cxio_create_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq)
 	return (rdev_p->t3cdev_p->ctl(rdev_p->t3cdev_p, RDMA_CQ_SETUP, &setup));
 }
 
+#ifdef notyet
 int cxio_resize_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq)
 {
 	struct rdma_cq_setup setup;
@@ -199,6 +201,7 @@ int cxio_resize_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq)
 	setup.ovfl_mode = 1;
 	return (rdev_p->t3cdev_p->ctl(rdev_p->t3cdev_p, RDMA_CQ_SETUP, &setup));
 }
+#endif
 
 static u32 get_qpid(struct cxio_rdev *rdev_p, struct cxio_ucontext *uctx)
 {
@@ -297,7 +300,7 @@ int cxio_create_qp(struct cxio_rdev *rdev_p, u32 kernel_domain,
 		goto err4;
 
 	memset(wq->queue, 0, depth * sizeof(union t3_wr));
-	pci_unmap_addr_set(wq, mapping, wq->dma_addr);
+	dma_unmap_addr_set(wq, mapping, wq->dma_addr);
 	wq->doorbell = (void __iomem *)rdev_p->rnic_info.kdb_addr;
 	if (!kernel_domain)
 		wq->udb = (u64)rdev_p->rnic_info.udbell_physbase +
@@ -325,7 +328,7 @@ int cxio_destroy_cq(struct cxio_rdev *rdev_p, struct t3_cq *cq)
 	dma_free_coherent(&(rdev_p->rnic_info.pdev->dev),
 			  (1UL << (cq->size_log2))
 			  * sizeof(struct t3_cqe), cq->queue,
-			  pci_unmap_addr(cq, mapping));
+			  dma_unmap_addr(cq, mapping));
 	cxio_hal_put_cqid(rdev_p->rscp, cq->cqid);
 	return err;
 }
@@ -336,7 +339,7 @@ int cxio_destroy_qp(struct cxio_rdev *rdev_p, struct t3_wq *wq,
 	dma_free_coherent(&(rdev_p->rnic_info.pdev->dev),
 			  (1UL << (wq->size_log2))
 			  * sizeof(union t3_wr), wq->queue,
-			  pci_unmap_addr(wq, mapping));
+			  dma_unmap_addr(wq, mapping));
 	kfree(wq->sq);
 	cxio_hal_rqtpool_free(rdev_p, wq->rq_addr, (1UL << wq->rq_size_log2));
 	kfree(wq->rq);
@@ -410,6 +413,7 @@ int cxio_flush_sq(struct t3_wq *wq, struct t3_cq *cq, int count)
 	ptr = wq->sq_rptr + count;
 	sqp = wq->sq + Q_PTR2IDX(ptr, wq->sq_size_log2);
 	while (ptr != wq->sq_wptr) {
+		sqp->signaled = 0;
 		insert_sq_cqe(wq, cq, sqp);
 		ptr++;
 		sqp = wq->sq + Q_PTR2IDX(ptr, wq->sq_size_log2);
@@ -450,7 +454,7 @@ static int cqe_completes_wr(struct t3_cqe *cqe, struct t3_wq *wq)
 	if ((CQE_OPCODE(*cqe) == T3_READ_RESP) && SQ_TYPE(*cqe))
 		return 0;
 
-	if ((CQE_OPCODE(*cqe) == T3_SEND) && RQ_TYPE(*cqe) &&
+	if (CQE_SEND_OPCODE(*cqe) && RQ_TYPE(*cqe) &&
 	    Q_EMPTY(wq->rq_rptr, wq->rq_wptr))
 		return 0;
 
@@ -536,7 +540,7 @@ static int cxio_hal_init_ctrl_qp(struct cxio_rdev *rdev_p)
 		err = -ENOMEM;
 		goto err;
 	}
-	pci_unmap_addr_set(&rdev_p->ctrl_qp, mapping,
+	dma_unmap_addr_set(&rdev_p->ctrl_qp, mapping,
 			   rdev_p->ctrl_qp.dma_addr);
 	rdev_p->ctrl_qp.doorbell = (void __iomem *)rdev_p->rnic_info.kdb_addr;
 	memset(rdev_p->ctrl_qp.workq, 0,
@@ -571,7 +575,7 @@ static int cxio_hal_init_ctrl_qp(struct cxio_rdev *rdev_p)
 	     (unsigned long long) rdev_p->ctrl_qp.dma_addr,
 	     rdev_p->ctrl_qp.workq, 1 << T3_CTRL_QP_SIZE_LOG2);
 	skb->priority = CPL_PRIORITY_CONTROL;
-	return (cxgb3_ofld_send(rdev_p->t3cdev_p, skb));
+	return iwch_cxgb3_ofld_send(rdev_p->t3cdev_p, skb);
 err:
 	kfree_skb(skb);
 	return err;
@@ -582,13 +586,13 @@ static int cxio_hal_destroy_ctrl_qp(struct cxio_rdev *rdev_p)
 	dma_free_coherent(&(rdev_p->rnic_info.pdev->dev),
 			  (1UL << T3_CTRL_QP_SIZE_LOG2)
 			  * sizeof(union t3_wr), rdev_p->ctrl_qp.workq,
-			  pci_unmap_addr(&rdev_p->ctrl_qp, mapping));
+			  dma_unmap_addr(&rdev_p->ctrl_qp, mapping));
 	return cxio_hal_clear_qp_ctx(rdev_p, T3_CTRL_QP_ID);
 }
 
 /* write len bytes of data into addr (32B aligned address)
  * If data is NULL, clear len byte of memory to zero.
- * caller aquires the ctrl_qp lock before the call
+ * caller acquires the ctrl_qp lock before the call
  */
 static int cxio_hal_ctrl_qp_write_mem(struct cxio_rdev *rdev_p, u32 addr,
 				      u32 len, void *data)
@@ -701,6 +705,9 @@ static int __cxio_tpt_op(struct cxio_rdev *rdev_p, u32 reset_tpt_entry,
 	u32 stag_idx;
 	u32 wptr;
 
+	if (cxio_fatal_error(rdev_p))
+		return -EIO;
+
 	stag_state = stag_state > 0;
 	stag_idx = (*stag) >> 8;
 
@@ -725,17 +732,15 @@ static int __cxio_tpt_op(struct cxio_rdev *rdev_p, u32 reset_tpt_entry,
 				V_TPT_STAG_TYPE(type) | V_TPT_PDID(pdid));
 		BUG_ON(page_size >= 28);
 		tpt.flags_pagesize_qpid = cpu_to_be32(V_TPT_PERM(perm) |
-				F_TPT_MW_BIND_ENABLE |
-				V_TPT_ADDR_TYPE((zbva ? TPT_ZBTO : TPT_VATO)) |
-				V_TPT_PAGE_SIZE(page_size));
-		tpt.rsvd_pbl_addr = reset_tpt_entry ? 0 :
-				    cpu_to_be32(V_TPT_PBL_ADDR(PBL_OFF(rdev_p, pbl_addr)>>3));
+			((perm & TPT_MW_BIND) ? F_TPT_MW_BIND_ENABLE : 0) |
+			V_TPT_ADDR_TYPE((zbva ? TPT_ZBTO : TPT_VATO)) |
+			V_TPT_PAGE_SIZE(page_size));
+		tpt.rsvd_pbl_addr = cpu_to_be32(V_TPT_PBL_ADDR(PBL_OFF(rdev_p, pbl_addr)>>3));
 		tpt.len = cpu_to_be32(len);
 		tpt.va_hi = cpu_to_be32((u32) (to >> 32));
 		tpt.va_low_or_fbo = cpu_to_be32((u32) (to & 0xFFFFFFFFULL));
 		tpt.rsvd_bind_cnt_or_pstag = 0;
-		tpt.rsvd_pbl_size = reset_tpt_entry ? 0 :
-				  cpu_to_be32(V_TPT_PBL_SIZE(pbl_size >> 2));
+		tpt.rsvd_pbl_size = cpu_to_be32(V_TPT_PBL_SIZE(pbl_size >> 2));
 	}
 	err = cxio_hal_ctrl_qp_write_mem(rdev_p,
 				       stag_idx +
@@ -848,14 +853,16 @@ int cxio_rdma_init(struct cxio_rdev *rdev_p, struct t3_rdma_init_attr *attr)
 	wqe->qpcaps = attr->qpcaps;
 	wqe->ulpdu_size = cpu_to_be16(attr->tcp_emss);
 	wqe->rqe_count = cpu_to_be16(attr->rqe_count);
-	wqe->flags_rtr_type = cpu_to_be16(attr->flags|V_RTR_TYPE(attr->rtr_type));
+	wqe->flags_rtr_type = cpu_to_be16(attr->flags |
+					  V_RTR_TYPE(attr->rtr_type) |
+					  V_CHAN(attr->chan));
 	wqe->ord = cpu_to_be32(attr->ord);
 	wqe->ird = cpu_to_be32(attr->ird);
 	wqe->qp_dma_addr = cpu_to_be64(attr->qp_dma_addr);
 	wqe->qp_dma_size = cpu_to_be32(attr->qp_dma_size);
 	wqe->irs = cpu_to_be32(attr->irs);
 	skb->priority = 0;	/* 0=>ToeQ; 1=>CtrlQ */
-	return (cxgb3_ofld_send(rdev_p->t3cdev_p, skb));
+	return iwch_cxgb3_ofld_send(rdev_p->t3cdev_p, skb);
 }
 
 void cxio_register_ev_cb(cxio_hal_ev_callback_func_t ev_cb)
@@ -938,6 +945,23 @@ int cxio_rdev_open(struct cxio_rdev *rdev_p)
 	if (!rdev_p->t3cdev_p)
 		rdev_p->t3cdev_p = dev2t3cdev(netdev_p);
 	rdev_p->t3cdev_p->ulp = (void *) rdev_p;
+
+	err = rdev_p->t3cdev_p->ctl(rdev_p->t3cdev_p, GET_EMBEDDED_INFO,
+					 &(rdev_p->fw_info));
+	if (err) {
+		printk(KERN_ERR "%s t3cdev_p(%p)->ctl returned error %d.\n",
+		     __func__, rdev_p->t3cdev_p, err);
+		goto err1;
+	}
+	if (G_FW_VERSION_MAJOR(rdev_p->fw_info.fw_vers) != CXIO_FW_MAJ) {
+		printk(KERN_ERR MOD "fatal firmware version mismatch: "
+		       "need version %u but adapter has version %u\n",
+		       CXIO_FW_MAJ,
+		       G_FW_VERSION_MAJOR(rdev_p->fw_info.fw_vers));
+		err = -EINVAL;
+		goto err1;
+	}
+
 	err = rdev_p->t3cdev_p->ctl(rdev_p->t3cdev_p, RDMA_GET_PARAMS,
 					 &(rdev_p->rnic_info));
 	if (err) {
@@ -1011,6 +1035,7 @@ err3:
 err2:
 	cxio_hal_destroy_ctrl_qp(rdev_p);
 err1:
+	rdev_p->t3cdev_p->ulp = NULL;
 	list_del(&rdev_p->entry);
 	return err;
 }
@@ -1021,9 +1046,9 @@ void cxio_rdev_close(struct cxio_rdev *rdev_p)
 		cxio_hal_pblpool_destroy(rdev_p);
 		cxio_hal_rqtpool_destroy(rdev_p);
 		list_del(&rdev_p->entry);
-		rdev_p->t3cdev_p->ulp = NULL;
 		cxio_hal_destroy_ctrl_qp(rdev_p);
 		cxio_hal_destroy_resource(rdev_p->rscp);
+		rdev_p->t3cdev_p->ulp = NULL;
 	}
 }
 
@@ -1204,11 +1229,12 @@ int cxio_poll_cq(struct t3_wq *wq, struct t3_cq *cq, struct t3_cqe *cqe,
 		}
 
 		/* incoming SEND with no receive posted failures */
-		if ((CQE_OPCODE(*hw_cqe) == T3_SEND) && RQ_TYPE(*hw_cqe) &&
+		if (CQE_SEND_OPCODE(*hw_cqe) && RQ_TYPE(*hw_cqe) &&
 		    Q_EMPTY(wq->rq_rptr, wq->rq_wptr)) {
 			ret = -1;
 			goto skip_cqe;
 		}
+		BUG_ON((*cqe_flushed == 0) && !SW_CQE(*hw_cqe));
 		goto proc_cqe;
 	}
 
@@ -1223,6 +1249,13 @@ int cxio_poll_cq(struct t3_wq *wq, struct t3_cq *cq, struct t3_cqe *cqe,
 		 * then we complete this with TPT_ERR_MSN and mark the wq in
 		 * error.
 		 */
+
+		if (Q_EMPTY(wq->rq_rptr, wq->rq_wptr)) {
+			wq->error = 1;
+			ret = -1;
+			goto skip_cqe;
+		}
+
 		if (unlikely((CQE_WRID_MSN(*hw_cqe) != (wq->rq_rptr + 1)))) {
 			wq->error = 1;
 			hw_cqe->header |= htonl(V_CQE_STATUS(TPT_ERR_MSN));
@@ -1277,6 +1310,7 @@ proc_cqe:
 			cxio_hal_pblpool_free(wq->rdev,
 				wq->rq[Q_PTR2IDX(wq->rq_rptr,
 				wq->rq_size_log2)].pbl_addr, T3_STAG0_PBL_SIZE);
+		BUG_ON(Q_EMPTY(wq->rq_rptr, wq->rq_wptr));
 		wq->rq_rptr++;
 	}
 
diff --git a/drivers/infiniband/hw/cxgb3/cxio_hal.h b/drivers/infiniband/hw/cxgb3/cxio_hal.h
index 656fe47bc84..78fbe9ffe7f 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_hal.h
+++ b/drivers/infiniband/hw/cxgb3/cxio_hal.h
@@ -34,6 +34,7 @@
 
 #include <linux/list.h>
 #include <linux/mutex.h>
+#include <linux/kfifo.h>
 
 #include "t3_cpl.h"
 #include "t3cdev.h"
@@ -52,7 +53,7 @@
 #define T3_MAX_PBL_SIZE 256
 #define T3_MAX_RQ_SIZE 1024
 #define T3_MAX_QP_DEPTH (T3_MAX_RQ_SIZE-1)
-#define T3_MAX_CQ_DEPTH 8192
+#define T3_MAX_CQ_DEPTH 65536
 #define T3_MAX_NUM_STAG (1<<15)
 #define T3_MAX_MR_SIZE 0x100000000ULL
 #define T3_PAGESIZE_MASK 0xffff000  /* 4KB-128MB */
@@ -61,6 +62,8 @@
 
 #define T3_MAX_DEV_NAME_LEN 32
 
+#define CXIO_FW_MAJ 7
+
 struct cxio_hal_ctrl_qp {
 	u32 wptr;
 	u32 rptr;
@@ -68,18 +71,18 @@ struct cxio_hal_ctrl_qp {
 	wait_queue_head_t waitq;/* wait for RspQ/CQE msg */
 	union t3_wr *workq;	/* the work request queue */
 	dma_addr_t dma_addr;	/* pci bus address of the workq */
-	DECLARE_PCI_UNMAP_ADDR(mapping)
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 	void __iomem *doorbell;
 };
 
 struct cxio_hal_resource {
-	struct kfifo *tpt_fifo;
+	struct kfifo tpt_fifo;
 	spinlock_t tpt_fifo_lock;
-	struct kfifo *qpid_fifo;
+	struct kfifo qpid_fifo;
 	spinlock_t qpid_fifo_lock;
-	struct kfifo *cqid_fifo;
+	struct kfifo cqid_fifo;
 	spinlock_t cqid_fifo_lock;
-	struct kfifo *pdid_fifo;
+	struct kfifo pdid_fifo;
 	spinlock_t pdid_fifo_lock;
 };
 
@@ -108,8 +111,16 @@ struct cxio_rdev {
 	struct gen_pool *pbl_pool;
 	struct gen_pool *rqt_pool;
 	struct list_head entry;
+	struct ch_embedded_info fw_info;
+	u32	flags;
+#define	CXIO_ERROR_FATAL	1
 };
 
+static inline int cxio_fatal_error(struct cxio_rdev *rdev_p)
+{
+	return rdev_p->flags & CXIO_ERROR_FATAL;
+}
+
 static inline int cxio_num_stags(struct cxio_rdev *rdev_p)
 {
 	return min((int)T3_MAX_NUM_STAG, (int)((rdev_p->rnic_info.tpt_top - rdev_p->rnic_info.tpt_base) >> 5));
@@ -146,7 +157,7 @@ int cxio_rdev_open(struct cxio_rdev *rdev);
 void cxio_rdev_close(struct cxio_rdev *rdev);
 int cxio_hal_cq_op(struct cxio_rdev *rdev, struct t3_cq *cq,
 		   enum t3_cq_opcode op, u32 credit);
-int cxio_create_cq(struct cxio_rdev *rdev, struct t3_cq *cq);
+int cxio_create_cq(struct cxio_rdev *rdev, struct t3_cq *cq, int kernel);
 int cxio_destroy_cq(struct cxio_rdev *rdev, struct t3_cq *cq);
 int cxio_resize_cq(struct cxio_rdev *rdev, struct t3_cq *cq);
 void cxio_release_ucontext(struct cxio_rdev *rdev, struct cxio_ucontext *uctx);
@@ -183,6 +194,7 @@ void cxio_count_scqes(struct t3_cq *cq, struct t3_wq *wq, int *count);
 void cxio_flush_hw_cq(struct t3_cq *cq);
 int cxio_poll_cq(struct t3_wq *wq, struct t3_cq *cq, struct t3_cqe *cqe,
 		     u8 *cqe_flushed, u64 *cookie, u32 *credit);
+int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb);
 
 #define MOD "iw_cxgb3: "
 #define PDBG(fmt, args...) pr_debug(MOD fmt, ## args)
diff --git a/drivers/infiniband/hw/cxgb3/cxio_resource.c b/drivers/infiniband/hw/cxgb3/cxio_resource.c
index bd233c08765..c40088ecf9f 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_resource.c
+++ b/drivers/infiniband/hw/cxgb3/cxio_resource.c
@@ -39,12 +39,12 @@
 #include "cxio_resource.h"
 #include "cxio_hal.h"
 
-static struct kfifo *rhdl_fifo;
+static struct kfifo rhdl_fifo;
 static spinlock_t rhdl_fifo_lock;
 
 #define RANDOM_SIZE 16
 
-static int __cxio_init_resource_fifo(struct kfifo **fifo,
+static int __cxio_init_resource_fifo(struct kfifo *fifo,
 				   spinlock_t *fifo_lock,
 				   u32 nr, u32 skip_low,
 				   u32 skip_high,
@@ -55,50 +55,51 @@ static int __cxio_init_resource_fifo(struct kfifo **fifo,
 	u32 rarray[16];
 	spin_lock_init(fifo_lock);
 
-	*fifo = kfifo_alloc(nr * sizeof(u32), GFP_KERNEL, fifo_lock);
-	if (IS_ERR(*fifo))
+	if (kfifo_alloc(fifo, nr * sizeof(u32), GFP_KERNEL))
 		return -ENOMEM;
 
 	for (i = 0; i < skip_low + skip_high; i++)
-		__kfifo_put(*fifo, (unsigned char *) &entry, sizeof(u32));
+		kfifo_in(fifo, (unsigned char *) &entry, sizeof(u32));
 	if (random) {
 		j = 0;
-		random_bytes = random32();
+		random_bytes = prandom_u32();
 		for (i = 0; i < RANDOM_SIZE; i++)
 			rarray[i] = i + skip_low;
 		for (i = skip_low + RANDOM_SIZE; i < nr - skip_high; i++) {
 			if (j >= RANDOM_SIZE) {
 				j = 0;
-				random_bytes = random32();
+				random_bytes = prandom_u32();
 			}
 			idx = (random_bytes >> (j * 2)) & 0xF;
-			__kfifo_put(*fifo,
+			kfifo_in(fifo,
 				(unsigned char *) &rarray[idx],
 				sizeof(u32));
 			rarray[idx] = i;
 			j++;
 		}
 		for (i = 0; i < RANDOM_SIZE; i++)
-			__kfifo_put(*fifo,
+			kfifo_in(fifo,
 				(unsigned char *) &rarray[i],
 				sizeof(u32));
 	} else
 		for (i = skip_low; i < nr - skip_high; i++)
-			__kfifo_put(*fifo, (unsigned char *) &i, sizeof(u32));
+			kfifo_in(fifo, (unsigned char *) &i, sizeof(u32));
 
 	for (i = 0; i < skip_low + skip_high; i++)
-		kfifo_get(*fifo, (unsigned char *) &entry, sizeof(u32));
+		if (kfifo_out_locked(fifo, (unsigned char *) &entry,
+				sizeof(u32), fifo_lock) != sizeof(u32))
+					break;
 	return 0;
 }
 
-static int cxio_init_resource_fifo(struct kfifo **fifo, spinlock_t * fifo_lock,
+static int cxio_init_resource_fifo(struct kfifo *fifo, spinlock_t * fifo_lock,
 				   u32 nr, u32 skip_low, u32 skip_high)
 {
 	return (__cxio_init_resource_fifo(fifo, fifo_lock, nr, skip_low,
 					  skip_high, 0));
 }
 
-static int cxio_init_resource_fifo_random(struct kfifo **fifo,
+static int cxio_init_resource_fifo_random(struct kfifo *fifo,
 				   spinlock_t * fifo_lock,
 				   u32 nr, u32 skip_low, u32 skip_high)
 {
@@ -113,15 +114,13 @@ static int cxio_init_qpid_fifo(struct cxio_rdev *rdev_p)
 
 	spin_lock_init(&rdev_p->rscp->qpid_fifo_lock);
 
-	rdev_p->rscp->qpid_fifo = kfifo_alloc(T3_MAX_NUM_QP * sizeof(u32),
-					      GFP_KERNEL,
-					      &rdev_p->rscp->qpid_fifo_lock);
-	if (IS_ERR(rdev_p->rscp->qpid_fifo))
+	if (kfifo_alloc(&rdev_p->rscp->qpid_fifo, T3_MAX_NUM_QP * sizeof(u32),
+					      GFP_KERNEL))
 		return -ENOMEM;
 
 	for (i = 16; i < T3_MAX_NUM_QP; i++)
 		if (!(i & rdev_p->qpmask))
-			__kfifo_put(rdev_p->rscp->qpid_fifo,
+			kfifo_in(&rdev_p->rscp->qpid_fifo,
 				    (unsigned char *) &i, sizeof(u32));
 	return 0;
 }
@@ -134,7 +133,7 @@ int cxio_hal_init_rhdl_resource(u32 nr_rhdl)
 
 void cxio_hal_destroy_rhdl_resource(void)
 {
-	kfifo_free(rhdl_fifo);
+	kfifo_free(&rhdl_fifo);
 }
 
 /* nr_* must be power of 2 */
@@ -167,11 +166,11 @@ int cxio_hal_init_resource(struct cxio_rdev *rdev_p,
 		goto pdid_err;
 	return 0;
 pdid_err:
-	kfifo_free(rscp->cqid_fifo);
+	kfifo_free(&rscp->cqid_fifo);
 cqid_err:
-	kfifo_free(rscp->qpid_fifo);
+	kfifo_free(&rscp->qpid_fifo);
 qpid_err:
-	kfifo_free(rscp->tpt_fifo);
+	kfifo_free(&rscp->tpt_fifo);
 tpt_err:
 	return -ENOMEM;
 }
@@ -179,33 +178,37 @@ tpt_err:
 /*
  * returns 0 if no resource available
  */
-static u32 cxio_hal_get_resource(struct kfifo *fifo)
+static u32 cxio_hal_get_resource(struct kfifo *fifo, spinlock_t * lock)
 {
 	u32 entry;
-	if (kfifo_get(fifo, (unsigned char *) &entry, sizeof(u32)))
+	if (kfifo_out_locked(fifo, (unsigned char *) &entry, sizeof(u32), lock))
 		return entry;
 	else
 		return 0;	/* fifo emptry */
 }
 
-static void cxio_hal_put_resource(struct kfifo *fifo, u32 entry)
+static void cxio_hal_put_resource(struct kfifo *fifo, spinlock_t * lock,
+		u32 entry)
 {
-	BUG_ON(kfifo_put(fifo, (unsigned char *) &entry, sizeof(u32)) == 0);
+	BUG_ON(
+	kfifo_in_locked(fifo, (unsigned char *) &entry, sizeof(u32), lock)
+	== 0);
 }
 
 u32 cxio_hal_get_stag(struct cxio_hal_resource *rscp)
 {
-	return cxio_hal_get_resource(rscp->tpt_fifo);
+	return cxio_hal_get_resource(&rscp->tpt_fifo, &rscp->tpt_fifo_lock);
 }
 
 void cxio_hal_put_stag(struct cxio_hal_resource *rscp, u32 stag)
 {
-	cxio_hal_put_resource(rscp->tpt_fifo, stag);
+	cxio_hal_put_resource(&rscp->tpt_fifo, &rscp->tpt_fifo_lock, stag);
 }
 
 u32 cxio_hal_get_qpid(struct cxio_hal_resource *rscp)
 {
-	u32 qpid = cxio_hal_get_resource(rscp->qpid_fifo);
+	u32 qpid = cxio_hal_get_resource(&rscp->qpid_fifo,
+			&rscp->qpid_fifo_lock);
 	PDBG("%s qpid 0x%x\n", __func__, qpid);
 	return qpid;
 }
@@ -213,35 +216,35 @@ u32 cxio_hal_get_qpid(struct cxio_hal_resource *rscp)
 void cxio_hal_put_qpid(struct cxio_hal_resource *rscp, u32 qpid)
 {
 	PDBG("%s qpid 0x%x\n", __func__, qpid);
-	cxio_hal_put_resource(rscp->qpid_fifo, qpid);
+	cxio_hal_put_resource(&rscp->qpid_fifo, &rscp->qpid_fifo_lock, qpid);
 }
 
 u32 cxio_hal_get_cqid(struct cxio_hal_resource *rscp)
 {
-	return cxio_hal_get_resource(rscp->cqid_fifo);
+	return cxio_hal_get_resource(&rscp->cqid_fifo, &rscp->cqid_fifo_lock);
 }
 
 void cxio_hal_put_cqid(struct cxio_hal_resource *rscp, u32 cqid)
 {
-	cxio_hal_put_resource(rscp->cqid_fifo, cqid);
+	cxio_hal_put_resource(&rscp->cqid_fifo, &rscp->cqid_fifo_lock, cqid);
 }
 
 u32 cxio_hal_get_pdid(struct cxio_hal_resource *rscp)
 {
-	return cxio_hal_get_resource(rscp->pdid_fifo);
+	return cxio_hal_get_resource(&rscp->pdid_fifo, &rscp->pdid_fifo_lock);
 }
 
 void cxio_hal_put_pdid(struct cxio_hal_resource *rscp, u32 pdid)
 {
-	cxio_hal_put_resource(rscp->pdid_fifo, pdid);
+	cxio_hal_put_resource(&rscp->pdid_fifo, &rscp->pdid_fifo_lock, pdid);
 }
 
 void cxio_hal_destroy_resource(struct cxio_hal_resource *rscp)
 {
-	kfifo_free(rscp->tpt_fifo);
-	kfifo_free(rscp->cqid_fifo);
-	kfifo_free(rscp->qpid_fifo);
-	kfifo_free(rscp->pdid_fifo);
+	kfifo_free(&rscp->tpt_fifo);
+	kfifo_free(&rscp->cqid_fifo);
+	kfifo_free(&rscp->qpid_fifo);
+	kfifo_free(&rscp->pdid_fifo);
 	kfree(rscp);
 }
 
diff --git a/drivers/infiniband/hw/cxgb3/cxio_wr.h b/drivers/infiniband/hw/cxgb3/cxio_wr.h
index 04618f7bfbb..83d2e19d31a 100644
--- a/drivers/infiniband/hw/cxgb3/cxio_wr.h
+++ b/drivers/infiniband/hw/cxgb3/cxio_wr.h
@@ -176,7 +176,7 @@ struct t3_send_wr {
 	struct t3_sge sgl[T3_MAX_SGE];	/* 4+ */
 };
 
-#define T3_MAX_FASTREG_DEPTH 24
+#define T3_MAX_FASTREG_DEPTH 10
 #define T3_MAX_FASTREG_FRAG 10
 
 struct t3_fastreg_wr {
@@ -327,6 +327,11 @@ enum rdma_init_rtr_types {
 #define V_RTR_TYPE(x)	((x) << S_RTR_TYPE)
 #define G_RTR_TYPE(x)	((((x) >> S_RTR_TYPE)) & M_RTR_TYPE)
 
+#define S_CHAN		4
+#define M_CHAN		0x3
+#define V_CHAN(x)	((x) << S_CHAN)
+#define G_CHAN(x)	((((x) >> S_CHAN)) & M_CHAN)
+
 struct t3_rdma_init_attr {
 	u32 tid;
 	u32 qpid;
@@ -346,6 +351,7 @@ struct t3_rdma_init_attr {
 	u16 flags;
 	u16 rqe_count;
 	u32 irs;
+	u32 chan;
 };
 
 struct t3_rdma_init_wr {
@@ -604,6 +610,12 @@ struct t3_cqe {
 #define CQE_STATUS(x)     (G_CQE_STATUS(be32_to_cpu((x).header)))
 #define CQE_OPCODE(x)     (G_CQE_OPCODE(be32_to_cpu((x).header)))
 
+#define CQE_SEND_OPCODE(x)( \
+	(G_CQE_OPCODE(be32_to_cpu((x).header)) == T3_SEND) || \
+	(G_CQE_OPCODE(be32_to_cpu((x).header)) == T3_SEND_WITH_SE) || \
+	(G_CQE_OPCODE(be32_to_cpu((x).header)) == T3_SEND_WITH_INV) || \
+	(G_CQE_OPCODE(be32_to_cpu((x).header)) == T3_SEND_WITH_SE_INV))
+
 #define CQE_LEN(x)        (be32_to_cpu((x).len))
 
 /* used for RQ completion processing */
@@ -677,9 +689,9 @@ struct t3_swrq {
  * A T3 WQ implements both the SQ and RQ.
  */
 struct t3_wq {
-	union t3_wr *queue;		/* DMA accessable memory */
+	union t3_wr *queue;		/* DMA accessible memory */
 	dma_addr_t dma_addr;		/* DMA address for HW */
-	DECLARE_PCI_UNMAP_ADDR(mapping)	/* unmap kruft */
+	DEFINE_DMA_UNMAP_ADDR(mapping); /* unmap kruft */
 	u32 error;			/* 1 once we go to ERROR */
 	u32 qpid;
 	u32 wptr;			/* idx to next available WR slot */
@@ -706,7 +718,7 @@ struct t3_cq {
 	u32 wptr;
 	u32 size_log2;
 	dma_addr_t dma_addr;
-	DECLARE_PCI_UNMAP_ADDR(mapping)
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 	struct t3_cqe *queue;
 	struct t3_cqe *sw_queue;
 	u32 sw_rptr;
@@ -716,9 +728,40 @@ struct t3_cq {
 #define CQ_VLD_ENTRY(ptr,size_log2,cqe) (Q_GENBIT(ptr,size_log2) == \
 					 CQE_GENBIT(*cqe))
 
+struct t3_cq_status_page {
+	u32 cq_err;
+};
+
+static inline int cxio_cq_in_error(struct t3_cq *cq)
+{
+	return ((struct t3_cq_status_page *)
+		&cq->queue[1 << cq->size_log2])->cq_err;
+}
+
+static inline void cxio_set_cq_in_error(struct t3_cq *cq)
+{
+	((struct t3_cq_status_page *)
+	 &cq->queue[1 << cq->size_log2])->cq_err = 1;
+}
+
 static inline void cxio_set_wq_in_error(struct t3_wq *wq)
 {
-	wq->queue->wq_in_err.err = 1;
+	wq->queue->wq_in_err.err |= 1;
+}
+
+static inline void cxio_disable_wq_db(struct t3_wq *wq)
+{
+	wq->queue->wq_in_err.err |= 2;
+}
+
+static inline void cxio_enable_wq_db(struct t3_wq *wq)
+{
+	wq->queue->wq_in_err.err &= ~2;
+}
+
+static inline int cxio_wq_db_enabled(struct t3_wq *wq)
+{
+	return !(wq->queue->wq_in_err.err & 2);
 }
 
 static inline struct t3_cqe *cxio_next_hw_cqe(struct t3_cq *cq)
diff --git a/drivers/infiniband/hw/cxgb3/iwch.c b/drivers/infiniband/hw/cxgb3/iwch.c
index 4489c89d671..8e77dc543dd 100644
--- a/drivers/infiniband/hw/cxgb3/iwch.c
+++ b/drivers/infiniband/hw/cxgb3/iwch.c
@@ -47,22 +47,62 @@ MODULE_DESCRIPTION("Chelsio T3 RDMA Driver");
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRV_VERSION);
 
-cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS];
-
 static void open_rnic_dev(struct t3cdev *);
 static void close_rnic_dev(struct t3cdev *);
+static void iwch_event_handler(struct t3cdev *, u32, u32);
 
 struct cxgb3_client t3c_client = {
 	.name = "iw_cxgb3",
 	.add = open_rnic_dev,
 	.remove = close_rnic_dev,
 	.handlers = t3c_handlers,
-	.redirect = iwch_ep_redirect
+	.redirect = iwch_ep_redirect,
+	.event_handler = iwch_event_handler
 };
 
 static LIST_HEAD(dev_list);
 static DEFINE_MUTEX(dev_mutex);
 
+static int disable_qp_db(int id, void *p, void *data)
+{
+	struct iwch_qp *qhp = p;
+
+	cxio_disable_wq_db(&qhp->wq);
+	return 0;
+}
+
+static int enable_qp_db(int id, void *p, void *data)
+{
+	struct iwch_qp *qhp = p;
+
+	if (data)
+		ring_doorbell(qhp->rhp->rdev.ctrl_qp.doorbell, qhp->wq.qpid);
+	cxio_enable_wq_db(&qhp->wq);
+	return 0;
+}
+
+static void disable_dbs(struct iwch_dev *rnicp)
+{
+	spin_lock_irq(&rnicp->lock);
+	idr_for_each(&rnicp->qpidr, disable_qp_db, NULL);
+	spin_unlock_irq(&rnicp->lock);
+}
+
+static void enable_dbs(struct iwch_dev *rnicp, int ring_db)
+{
+	spin_lock_irq(&rnicp->lock);
+	idr_for_each(&rnicp->qpidr, enable_qp_db,
+		     (void *)(unsigned long)ring_db);
+	spin_unlock_irq(&rnicp->lock);
+}
+
+static void iwch_db_drop_task(struct work_struct *work)
+{
+	struct iwch_dev *rnicp = container_of(work, struct iwch_dev,
+					      db_drop_task.work);
+	enable_dbs(rnicp, 1);
+}
+
 static void rnic_init(struct iwch_dev *rnicp)
 {
 	PDBG("%s iwch_dev %p\n", __func__,  rnicp);
@@ -70,6 +110,7 @@ static void rnic_init(struct iwch_dev *rnicp)
 	idr_init(&rnicp->qpidr);
 	idr_init(&rnicp->mmidr);
 	spin_lock_init(&rnicp->lock);
+	INIT_DELAYED_WORK(&rnicp->db_drop_task, iwch_db_drop_task);
 
 	rnicp->attr.max_qps = T3_MAX_NUM_QP - 32;
 	rnicp->attr.max_wrs = T3_MAX_QP_DEPTH;
@@ -103,11 +144,9 @@ static void rnic_init(struct iwch_dev *rnicp)
 static void open_rnic_dev(struct t3cdev *tdev)
 {
 	struct iwch_dev *rnicp;
-	static int vers_printed;
 
 	PDBG("%s t3cdev %p\n", __func__,  tdev);
-	if (!vers_printed++)
-		printk(KERN_INFO MOD "Chelsio T3 RDMA Driver - version %s\n",
+	printk_once(KERN_INFO MOD "Chelsio T3 RDMA Driver - version %s\n",
 		       DRV_VERSION);
 	rnicp = (struct iwch_dev *)ib_alloc_device(sizeof(*rnicp));
 	if (!rnicp) {
@@ -147,6 +186,9 @@ static void close_rnic_dev(struct t3cdev *tdev)
 	mutex_lock(&dev_mutex);
 	list_for_each_entry_safe(dev, tmp, &dev_list, entry) {
 		if (dev->rdev.t3cdev_p == tdev) {
+			dev->rdev.flags = CXIO_ERROR_FATAL;
+			synchronize_net();
+			cancel_delayed_work_sync(&dev->db_drop_task);
 			list_del(&dev->entry);
 			iwch_unregister_device(dev);
 			cxio_rdev_close(&dev->rdev);
@@ -160,6 +202,69 @@ static void close_rnic_dev(struct t3cdev *tdev)
 	mutex_unlock(&dev_mutex);
 }
 
+static void iwch_event_handler(struct t3cdev *tdev, u32 evt, u32 port_id)
+{
+	struct cxio_rdev *rdev = tdev->ulp;
+	struct iwch_dev *rnicp;
+	struct ib_event event;
+	u32 portnum = port_id + 1;
+	int dispatch = 0;
+
+	if (!rdev)
+		return;
+	rnicp = rdev_to_iwch_dev(rdev);
+	switch (evt) {
+	case OFFLOAD_STATUS_DOWN: {
+		rdev->flags = CXIO_ERROR_FATAL;
+		synchronize_net();
+		event.event  = IB_EVENT_DEVICE_FATAL;
+		dispatch = 1;
+		break;
+		}
+	case OFFLOAD_PORT_DOWN: {
+		event.event  = IB_EVENT_PORT_ERR;
+		dispatch = 1;
+		break;
+		}
+	case OFFLOAD_PORT_UP: {
+		event.event  = IB_EVENT_PORT_ACTIVE;
+		dispatch = 1;
+		break;
+		}
+	case OFFLOAD_DB_FULL: {
+		disable_dbs(rnicp);
+		break;
+		}
+	case OFFLOAD_DB_EMPTY: {
+		enable_dbs(rnicp, 1);
+		break;
+		}
+	case OFFLOAD_DB_DROP: {
+		unsigned long delay = 1000;
+		unsigned short r;
+
+		disable_dbs(rnicp);
+		get_random_bytes(&r, 2);
+		delay += r & 1023;
+
+		/*
+		 * delay is between 1000-2023 usecs.
+		 */
+		schedule_delayed_work(&rnicp->db_drop_task,
+			usecs_to_jiffies(delay));
+		break;
+		}
+	}
+
+	if (dispatch) {
+		event.device = &rnicp->ibdev;
+		event.element.port_num = portnum;
+		ib_dispatch_event(&event);
+	}
+
+	return;
+}
+
 static int __init iwch_init_module(void)
 {
 	int err;
diff --git a/drivers/infiniband/hw/cxgb3/iwch.h b/drivers/infiniband/hw/cxgb3/iwch.h
index 3773453b2cf..837862287a2 100644
--- a/drivers/infiniband/hw/cxgb3/iwch.h
+++ b/drivers/infiniband/hw/cxgb3/iwch.h
@@ -36,6 +36,7 @@
 #include <linux/list.h>
 #include <linux/spinlock.h>
 #include <linux/idr.h>
+#include <linux/workqueue.h>
 
 #include <rdma/ib_verbs.h>
 
@@ -110,6 +111,7 @@ struct iwch_dev {
 	struct idr mmidr;
 	spinlock_t lock;
 	struct list_head entry;
+	struct delayed_work db_drop_task;
 };
 
 static inline struct iwch_dev *to_iwch_dev(struct ib_device *ibdev)
@@ -117,6 +119,11 @@ static inline struct iwch_dev *to_iwch_dev(struct ib_device *ibdev)
 	return container_of(ibdev, struct iwch_dev, ibdev);
 }
 
+static inline struct iwch_dev *rdev_to_iwch_dev(struct cxio_rdev *rdev)
+{
+	return container_of(rdev, struct iwch_dev, rdev);
+}
+
 static inline int t3b_device(const struct iwch_dev *rhp)
 {
 	return rhp->rdev.t3cdev_p->type == T3B;
@@ -146,19 +153,17 @@ static inline int insert_handle(struct iwch_dev *rhp, struct idr *idr,
 				void *handle, u32 id)
 {
 	int ret;
-	int newid;
-
-	do {
-		if (!idr_pre_get(idr, GFP_KERNEL)) {
-			return -ENOMEM;
-		}
-		spin_lock_irq(&rhp->lock);
-		ret = idr_get_new_above(idr, handle, id, &newid);
-		BUG_ON(newid != id);
-		spin_unlock_irq(&rhp->lock);
-	} while (ret == -EAGAIN);
-
-	return ret;
+
+	idr_preload(GFP_KERNEL);
+	spin_lock_irq(&rhp->lock);
+
+	ret = idr_alloc(idr, handle, id, id + 1, GFP_NOWAIT);
+
+	spin_unlock_irq(&rhp->lock);
+	idr_preload_end();
+
+	BUG_ON(ret == -ENOSPC);
+	return ret < 0 ? ret : 0;
 }
 
 static inline void remove_handle(struct iwch_dev *rhp, struct idr *idr, u32 id)
diff --git a/drivers/infiniband/hw/cxgb3/iwch_cm.c b/drivers/infiniband/hw/cxgb3/iwch_cm.c
index c325c44807e..cb78b1e9bcd 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_cm.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_cm.c
@@ -31,6 +31,7 @@
  */
 #include <linux/module.h>
 #include <linux/list.h>
+#include <linux/slab.h>
 #include <linux/workqueue.h>
 #include <linux/skbuff.h>
 #include <linux/timer.h>
@@ -101,12 +102,9 @@ static unsigned int cong_flavor = 1;
 module_param(cong_flavor, uint, 0644);
 MODULE_PARM_DESC(cong_flavor, "TCP Congestion control flavor (default=1)");
 
-static void process_work(struct work_struct *work);
 static struct workqueue_struct *workq;
-static DECLARE_WORK(skb_work, process_work);
 
 static struct sk_buff_head rxq;
-static cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS];
 
 static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
 static void ep_timeout(unsigned long arg);
@@ -130,15 +128,46 @@ static void stop_ep_timer(struct iwch_ep *ep)
 {
 	PDBG("%s ep %p\n", __func__, ep);
 	if (!timer_pending(&ep->timer)) {
-		printk(KERN_ERR "%s timer stopped when its not running!  ep %p state %u\n",
+		WARN(1, "%s timer stopped when its not running!  ep %p state %u\n",
 			__func__, ep, ep->com.state);
-		WARN_ON(1);
 		return;
 	}
 	del_timer_sync(&ep->timer);
 	put_ep(&ep->com);
 }
 
+static int iwch_l2t_send(struct t3cdev *tdev, struct sk_buff *skb, struct l2t_entry *l2e)
+{
+	int	error = 0;
+	struct cxio_rdev *rdev;
+
+	rdev = (struct cxio_rdev *)tdev->ulp;
+	if (cxio_fatal_error(rdev)) {
+		kfree_skb(skb);
+		return -EIO;
+	}
+	error = l2t_send(tdev, skb, l2e);
+	if (error < 0)
+		kfree_skb(skb);
+	return error;
+}
+
+int iwch_cxgb3_ofld_send(struct t3cdev *tdev, struct sk_buff *skb)
+{
+	int	error = 0;
+	struct cxio_rdev *rdev;
+
+	rdev = (struct cxio_rdev *)tdev->ulp;
+	if (cxio_fatal_error(rdev)) {
+		kfree_skb(skb);
+		return -EIO;
+	}
+	error = cxgb3_ofld_send(tdev, skb);
+	if (error < 0)
+		kfree_skb(skb);
+	return error;
+}
+
 static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
 {
 	struct cpl_tid_release *req;
@@ -150,7 +179,7 @@ static void release_tid(struct t3cdev *tdev, u32 hwtid, struct sk_buff *skb)
 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
 	skb->priority = CPL_PRIORITY_SETUP;
-	cxgb3_ofld_send(tdev, skb);
+	iwch_cxgb3_ofld_send(tdev, skb);
 	return;
 }
 
@@ -172,8 +201,7 @@ int iwch_quiesce_tid(struct iwch_ep *ep)
 	req->val = cpu_to_be64(1 << S_TCB_RX_QUIESCE);
 
 	skb->priority = CPL_PRIORITY_DATA;
-	cxgb3_ofld_send(ep->com.tdev, skb);
-	return 0;
+	return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 }
 
 int iwch_resume_tid(struct iwch_ep *ep)
@@ -194,8 +222,7 @@ int iwch_resume_tid(struct iwch_ep *ep)
 	req->val = 0;
 
 	skb->priority = CPL_PRIORITY_DATA;
-	cxgb3_ofld_send(ep->com.tdev, skb);
-	return 0;
+	return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 }
 
 static void set_emss(struct iwch_ep *ep, u16 opt)
@@ -252,42 +279,25 @@ static void *alloc_ep(int size, gfp_t gfp)
 
 void __free_ep(struct kref *kref)
 {
-	struct iwch_ep_common *epc;
-	epc = container_of(kref, struct iwch_ep_common, kref);
-	PDBG("%s ep %p state %s\n", __func__, epc, states[state_read(epc)]);
-	kfree(epc);
+	struct iwch_ep *ep;
+	ep = container_of(container_of(kref, struct iwch_ep_common, kref),
+			  struct iwch_ep, com);
+	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
+	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
+		cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
+		dst_release(ep->dst);
+		l2t_release(ep->com.tdev, ep->l2t);
+	}
+	kfree(ep);
 }
 
 static void release_ep_resources(struct iwch_ep *ep)
 {
 	PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
-	cxgb3_remove_tid(ep->com.tdev, (void *)ep, ep->hwtid);
-	dst_release(ep->dst);
-	l2t_release(L2DATA(ep->com.tdev), ep->l2t);
+	set_bit(RELEASE_RESOURCES, &ep->com.flags);
 	put_ep(&ep->com);
 }
 
-static void process_work(struct work_struct *work)
-{
-	struct sk_buff *skb = NULL;
-	void *ep;
-	struct t3cdev *tdev;
-	int ret;
-
-	while ((skb = skb_dequeue(&rxq))) {
-		ep = *((void **) (skb->cb));
-		tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
-		ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
-		if (ret & CPL_RET_BUF_DONE)
-			kfree_skb(skb);
-
-		/*
-		 * ep was referenced in sched(), and is freed here.
-		 */
-		put_ep((struct iwch_ep_common *)ep);
-	}
-}
-
 static int status2errno(int status)
 {
 	switch (status) {
@@ -327,23 +337,12 @@ static struct rtable *find_route(struct t3cdev *dev, __be32 local_ip,
 				 __be16 peer_port, u8 tos)
 {
 	struct rtable *rt;
-	struct flowi fl = {
-		.oif = 0,
-		.nl_u = {
-			 .ip4_u = {
-				   .daddr = peer_ip,
-				   .saddr = local_ip,
-				   .tos = tos}
-			 },
-		.proto = IPPROTO_TCP,
-		.uli_u = {
-			  .ports = {
-				    .sport = local_port,
-				    .dport = peer_port}
-			  }
-	};
-
-	if (ip_route_output_flow(&init_net, &rt, &fl, NULL, 0))
+	struct flowi4 fl4;
+
+	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
+				   peer_port, local_port, IPPROTO_TCP,
+				   tos, 0);
+	if (IS_ERR(rt))
 		return NULL;
 	return rt;
 }
@@ -382,7 +381,7 @@ static void abort_arp_failure(struct t3cdev *dev, struct sk_buff *skb)
 
 	PDBG("%s t3cdev %p\n", __func__, dev);
 	req->cmd = CPL_ABORT_NO_RST;
-	cxgb3_ofld_send(dev, skb);
+	iwch_cxgb3_ofld_send(dev, skb);
 }
 
 static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
@@ -402,8 +401,7 @@ static int send_halfclose(struct iwch_ep *ep, gfp_t gfp)
 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_CLOSE_CON));
 	req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, ep->hwtid));
-	l2t_send(ep->com.tdev, skb, ep->l2t);
-	return 0;
+	return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 }
 
 static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
@@ -420,12 +418,12 @@ static int send_abort(struct iwch_ep *ep, struct sk_buff *skb, gfp_t gfp)
 	skb->priority = CPL_PRIORITY_DATA;
 	set_arp_failure_handler(skb, abort_arp_failure);
 	req = (struct cpl_abort_req *) skb_put(skb, sizeof(*req));
+	memset(req, 0, sizeof(*req));
 	req->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_OFLD_HOST_ABORT_CON_REQ));
 	req->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
 	req->cmd = CPL_ABORT_SEND_RST;
-	l2t_send(ep->com.tdev, skb, ep->l2t);
-	return 0;
+	return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 }
 
 static int send_connect(struct iwch_ep *ep)
@@ -454,7 +452,8 @@ static int send_connect(struct iwch_ep *ep)
 	    V_MSS_IDX(mtu_idx) |
 	    V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
 	opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
-	opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
+	opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
+	       V_CONG_CONTROL_FLAVOR(cong_flavor);
 	skb->priority = CPL_PRIORITY_SETUP;
 	set_arp_failure_handler(skb, act_open_req_arp_failure);
 
@@ -469,8 +468,7 @@ static int send_connect(struct iwch_ep *ep)
 	req->opt0l = htonl(opt0l);
 	req->params = 0;
 	req->opt2 = htonl(opt2);
-	l2t_send(ep->com.tdev, skb, ep->l2t);
-	return 0;
+	return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 }
 
 static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
@@ -527,7 +525,7 @@ static void send_mpa_req(struct iwch_ep *ep, struct sk_buff *skb)
 	req->sndseq = htonl(ep->snd_seq);
 	BUG_ON(ep->mpa_skb);
 	ep->mpa_skb = skb;
-	l2t_send(ep->com.tdev, skb, ep->l2t);
+	iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 	start_ep_timer(ep);
 	state_set(&ep->com, MPA_REQ_SENT);
 	return;
@@ -578,8 +576,7 @@ static int send_mpa_reject(struct iwch_ep *ep, const void *pdata, u8 plen)
 	req->sndseq = htonl(ep->snd_seq);
 	BUG_ON(ep->mpa_skb);
 	ep->mpa_skb = skb;
-	l2t_send(ep->com.tdev, skb, ep->l2t);
-	return 0;
+	return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 }
 
 static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
@@ -630,8 +627,7 @@ static int send_mpa_reply(struct iwch_ep *ep, const void *pdata, u8 plen)
 	req->sndseq = htonl(ep->snd_seq);
 	ep->mpa_skb = skb;
 	state_set(&ep->com, MPA_REP_SENT);
-	l2t_send(ep->com.tdev, skb, ep->l2t);
-	return 0;
+	return iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 }
 
 static int act_establish(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
@@ -726,8 +722,10 @@ static void connect_reply_upcall(struct iwch_ep *ep, int status)
 	memset(&event, 0, sizeof(event));
 	event.event = IW_CM_EVENT_CONNECT_REPLY;
 	event.status = status;
-	event.local_addr = ep->com.local_addr;
-	event.remote_addr = ep->com.remote_addr;
+	memcpy(&event.local_addr, &ep->com.local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&event.remote_addr, &ep->com.remote_addr,
+	       sizeof(ep->com.remote_addr));
 
 	if ((status == 0) || (status == -ECONNREFUSED)) {
 		event.private_data_len = ep->plen;
@@ -752,15 +750,24 @@ static void connect_request_upcall(struct iwch_ep *ep)
 	PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 	memset(&event, 0, sizeof(event));
 	event.event = IW_CM_EVENT_CONNECT_REQUEST;
-	event.local_addr = ep->com.local_addr;
-	event.remote_addr = ep->com.remote_addr;
+	memcpy(&event.local_addr, &ep->com.local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&event.remote_addr, &ep->com.remote_addr,
+	       sizeof(ep->com.local_addr));
 	event.private_data_len = ep->plen;
 	event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
 	event.provider_data = ep;
-	if (state_read(&ep->parent_ep->com) != DEAD)
+	/*
+	 * Until ird/ord negotiation via MPAv2 support is added, send max
+	 * supported values
+	 */
+	event.ird = event.ord = 8;
+	if (state_read(&ep->parent_ep->com) != DEAD) {
+		get_ep(&ep->com);
 		ep->parent_ep->com.cm_id->event_handler(
 						ep->parent_ep->com.cm_id,
 						&event);
+	}
 	put_ep(&ep->parent_ep->com);
 	ep->parent_ep = NULL;
 }
@@ -772,6 +779,11 @@ static void established_upcall(struct iwch_ep *ep)
 	PDBG("%s ep %p\n", __func__, ep);
 	memset(&event, 0, sizeof(event));
 	event.event = IW_CM_EVENT_ESTABLISHED;
+	/*
+	 * Until ird/ord negotiation via MPAv2 support is added, send max
+	 * supported values
+	 */
+	event.ird = event.ord = 8;
 	if (ep->com.cm_id) {
 		PDBG("%s ep %p tid %d\n", __func__, ep, ep->hwtid);
 		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
@@ -795,7 +807,7 @@ static int update_rx_credits(struct iwch_ep *ep, u32 credits)
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_RX_DATA_ACK, ep->hwtid));
 	req->credit_dack = htonl(V_RX_CREDITS(credits) | V_RX_FORCE_ACK(1));
 	skb->priority = CPL_PRIORITY_ACK;
-	cxgb3_ofld_send(ep->com.tdev, skb);
+	iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 	return credits;
 }
 
@@ -916,7 +928,7 @@ static void process_mpa_reply(struct iwch_ep *ep, struct sk_buff *skb)
 		goto err;
 
 	if (peer2peer && iwch_rqes_posted(ep->com.qp) == 0) {
-		iwch_post_zb_read(ep->com.qp);
+		iwch_post_zb_read(ep);
 	}
 
 	goto out;
@@ -1080,37 +1092,45 @@ static int tx_ack(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	struct iwch_ep *ep = ctx;
 	struct cpl_wr_ack *hdr = cplhdr(skb);
 	unsigned int credits = ntohs(hdr->credits);
+	unsigned long flags;
+	int post_zb = 0;
 
 	PDBG("%s ep %p credits %u\n", __func__, ep, credits);
 
 	if (credits == 0) {
-		PDBG(KERN_ERR "%s 0 credit ack  ep %p state %u\n",
-			__func__, ep, state_read(&ep->com));
+		PDBG("%s 0 credit ack  ep %p state %u\n",
+		     __func__, ep, state_read(&ep->com));
 		return CPL_RET_BUF_DONE;
 	}
 
+	spin_lock_irqsave(&ep->com.lock, flags);
 	BUG_ON(credits != 1);
 	dst_confirm(ep->dst);
 	if (!ep->mpa_skb) {
 		PDBG("%s rdma_init wr_ack ep %p state %u\n",
-			__func__, ep, state_read(&ep->com));
+			__func__, ep, ep->com.state);
 		if (ep->mpa_attr.initiator) {
 			PDBG("%s initiator ep %p state %u\n",
-				__func__, ep, state_read(&ep->com));
-			if (peer2peer)
-				iwch_post_zb_read(ep->com.qp);
+				__func__, ep, ep->com.state);
+			if (peer2peer && ep->com.state == FPDU_MODE)
+				post_zb = 1;
 		} else {
 			PDBG("%s responder ep %p state %u\n",
-				__func__, ep, state_read(&ep->com));
-			ep->com.rpl_done = 1;
-			wake_up(&ep->com.waitq);
+				__func__, ep, ep->com.state);
+			if (ep->com.state == MPA_REQ_RCVD) {
+				ep->com.rpl_done = 1;
+				wake_up(&ep->com.waitq);
+			}
 		}
 	} else {
 		PDBG("%s lsm ack ep %p state %u freeing skb\n",
-			__func__, ep, state_read(&ep->com));
+			__func__, ep, ep->com.state);
 		kfree_skb(ep->mpa_skb);
 		ep->mpa_skb = NULL;
 	}
+	spin_unlock_irqrestore(&ep->com.lock, flags);
+	if (post_zb)
+		iwch_post_zb_read(ep);
 	return CPL_RET_BUF_DONE;
 }
 
@@ -1127,8 +1147,7 @@ static int abort_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	 * We get 2 abort replies from the HW.  The first one must
 	 * be ignored except for scribbling that we need one more.
 	 */
-	if (!(ep->flags & ABORT_REQ_IN_PROGRESS)) {
-		ep->flags |= ABORT_REQ_IN_PROGRESS;
+	if (!test_and_set_bit(ABORT_REQ_IN_PROGRESS, &ep->com.flags)) {
 		return CPL_RET_BUF_DONE;
 	}
 
@@ -1173,7 +1192,7 @@ static int act_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 		release_tid(ep->com.tdev, GET_TID(rpl), NULL);
 	cxgb3_free_atid(ep->com.tdev, ep->atid);
 	dst_release(ep->dst);
-	l2t_release(L2DATA(ep->com.tdev), ep->l2t);
+	l2t_release(ep->com.tdev, ep->l2t);
 	put_ep(&ep->com);
 	return CPL_RET_BUF_DONE;
 }
@@ -1203,8 +1222,7 @@ static int listen_start(struct iwch_listen_ep *ep)
 	req->opt1 = htonl(V_CONN_POLICY(CPL_CONN_POLICY_ASK));
 
 	skb->priority = 1;
-	cxgb3_ofld_send(ep->com.tdev, skb);
-	return 0;
+	return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 }
 
 static int pass_open_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
@@ -1237,8 +1255,7 @@ static int listen_stop(struct iwch_listen_ep *ep)
 	req->cpu_idx = 0;
 	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_LISTSRV_REQ, ep->stid));
 	skb->priority = 1;
-	cxgb3_ofld_send(ep->com.tdev, skb);
-	return 0;
+	return iwch_cxgb3_ofld_send(ep->com.tdev, skb);
 }
 
 static int close_listsrv_rpl(struct t3cdev *tdev, struct sk_buff *skb,
@@ -1275,7 +1292,8 @@ static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
 	    V_MSS_IDX(mtu_idx) |
 	    V_L2T_IDX(ep->l2t->idx) | V_TX_CHANNEL(ep->l2t->smt_idx);
 	opt0l = V_TOS((ep->tos >> 2) & M_TOS) | V_RCV_BUFSIZ(rcv_win>>10);
-	opt2 = V_FLAVORS_VALID(1) | V_CONG_CONTROL_FLAVOR(cong_flavor);
+	opt2 = F_RX_COALESCE_VALID | V_RX_COALESCE(0) | V_FLAVORS_VALID(1) |
+	       V_CONG_CONTROL_FLAVOR(cong_flavor);
 
 	rpl = cplhdr(skb);
 	rpl->wr.wr_hi = htonl(V_WR_OP(FW_WROPCODE_FORWARD));
@@ -1286,7 +1304,7 @@ static void accept_cr(struct iwch_ep *ep, __be32 peer_ip, struct sk_buff *skb)
 	rpl->opt2 = htonl(opt2);
 	rpl->rsvd = rpl->opt2;	/* workaround for HW bug */
 	skb->priority = CPL_PRIORITY_SETUP;
-	l2t_send(ep->com.tdev, skb, ep->l2t);
+	iwch_l2t_send(ep->com.tdev, skb, ep->l2t);
 
 	return;
 }
@@ -1315,7 +1333,7 @@ static void reject_cr(struct t3cdev *tdev, u32 hwtid, __be32 peer_ip,
 		rpl->opt0l_status = htonl(CPL_PASS_OPEN_REJECT);
 		rpl->opt2 = 0;
 		rpl->rsvd = rpl->opt2;
-		cxgb3_ofld_send(tdev, skb);
+		iwch_cxgb3_ofld_send(tdev, skb);
 	}
 }
 
@@ -1343,15 +1361,8 @@ static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	tim.mac_addr = req->dst_mac;
 	tim.vlan_tag = ntohs(req->vlan_tag);
 	if (tdev->ctl(tdev, GET_IFF_FROM_MAC, &tim) < 0 || !tim.dev) {
-		printk(KERN_ERR
-			"%s bad dst mac %02x %02x %02x %02x %02x %02x\n",
-			__func__,
-			req->dst_mac[0],
-			req->dst_mac[1],
-			req->dst_mac[2],
-			req->dst_mac[3],
-			req->dst_mac[4],
-			req->dst_mac[5]);
+		printk(KERN_ERR "%s bad dst mac %pM\n",
+			__func__, req->dst_mac);
 		goto reject;
 	}
 
@@ -1366,8 +1377,8 @@ static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 		       __func__);
 		goto reject;
 	}
-	dst = &rt->u.dst;
-	l2t = t3_l2t_get(tdev, dst->neighbour, dst->neighbour->dev);
+	dst = &rt->dst;
+	l2t = t3_l2t_get(tdev, dst, NULL, &req->peer_ip);
 	if (!l2t) {
 		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
 		       __func__);
@@ -1378,7 +1389,7 @@ static int pass_accept_req(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	if (!child_ep) {
 		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
 		       __func__);
-		l2t_release(L2DATA(tdev), l2t);
+		l2t_release(tdev, l2t);
 		dst_release(dst);
 		goto reject;
 	}
@@ -1450,10 +1461,14 @@ static int peer_close(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 		/*
 		 * We're gonna mark this puppy DEAD, but keep
 		 * the reference on it until the ULP accepts or
-		 * rejects the CR.
+		 * rejects the CR. Also wake up anyone waiting
+		 * in rdma connection migration (see iwch_accept_cr()).
 		 */
 		__state_set(&ep->com, CLOSING);
-		get_ep(&ep->com);
+		ep->com.rpl_done = 1;
+		ep->com.rpl_err = -ECONNRESET;
+		PDBG("waking up ep %p\n", ep);
+		wake_up(&ep->com.waitq);
 		break;
 	case MPA_REP_SENT:
 		__state_set(&ep->com, CLOSING);
@@ -1534,8 +1549,7 @@ static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	 * We get 2 peer aborts from the HW.  The first one must
 	 * be ignored except for scribbling that we need one more.
 	 */
-	if (!(ep->flags & PEER_ABORT_IN_PROGRESS)) {
-		ep->flags |= PEER_ABORT_IN_PROGRESS;
+	if (!test_and_set_bit(PEER_ABORT_IN_PROGRESS, &ep->com.flags)) {
 		return CPL_RET_BUF_DONE;
 	}
 
@@ -1562,9 +1576,13 @@ static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 		/*
 		 * We're gonna mark this puppy DEAD, but keep
 		 * the reference on it until the ULP accepts or
-		 * rejects the CR.
+		 * rejects the CR. Also wake up anyone waiting
+		 * in rdma connection migration (see iwch_accept_cr()).
 		 */
-		get_ep(&ep->com);
+		ep->com.rpl_done = 1;
+		ep->com.rpl_err = -ECONNRESET;
+		PDBG("waking up ep %p\n", ep);
+		wake_up(&ep->com.waitq);
 		break;
 	case MORIBUND:
 	case CLOSING:
@@ -1613,7 +1631,7 @@ static int peer_abort(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	rpl->wr.wr_lo = htonl(V_WR_TID(ep->hwtid));
 	OPCODE_TID(rpl) = htonl(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
 	rpl->cmd = CPL_ABORT_NO_RST;
-	cxgb3_ofld_send(ep->com.tdev, rpl_skb);
+	iwch_cxgb3_ofld_send(ep->com.tdev, rpl_skb);
 out:
 	if (release)
 		release_ep_resources(ep);
@@ -1666,7 +1684,7 @@ static int close_con_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
  * T3A does 3 things when a TERM is received:
  * 1) send up a CPL_RDMA_TERMINATE message with the TERM packet
  * 2) generate an async event on the QP with the TERMINATE opcode
- * 3) post a TERMINATE opcde cqe into the associated CQ.
+ * 3) post a TERMINATE opcode cqe into the associated CQ.
  *
  * For (1), we save the message in the qp for later consumer consumption.
  * For (2), we move the QP into TERMINATE, post a QP event and disconnect.
@@ -1678,6 +1696,9 @@ static int terminate(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 {
 	struct iwch_ep *ep = ctx;
 
+	if (state_read(&ep->com) != FPDU_MODE)
+		return CPL_RET_BUF_DONE;
+
 	PDBG("%s ep %p\n", __func__, ep);
 	skb_pull(skb, sizeof(struct cpl_rdma_terminate));
 	PDBG("%s saving %d bytes of term msg\n", __func__, skb->len);
@@ -1739,9 +1760,8 @@ static void ep_timeout(unsigned long arg)
 		__state_set(&ep->com, ABORTING);
 		break;
 	default:
-		printk(KERN_ERR "%s unexpected state ep %p state %u\n",
+		WARN(1, "%s unexpected state ep %p state %u\n",
 			__func__, ep, ep->com.state);
-		WARN_ON(1);
 		abort = 0;
 	}
 	spin_unlock_irqrestore(&ep->com.lock, flags);
@@ -1767,6 +1787,7 @@ int iwch_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
 		err = send_mpa_reject(ep, pdata, pdata_len);
 		err = iwch_ep_disconnect(ep, 0, GFP_KERNEL);
 	}
+	put_ep(&ep->com);
 	return 0;
 }
 
@@ -1780,8 +1801,10 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	struct iwch_qp *qp = get_qhp(h, conn_param->qpn);
 
 	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
-	if (state_read(&ep->com) == DEAD)
-		return -ECONNRESET;
+	if (state_read(&ep->com) == DEAD) {
+		err = -ECONNRESET;
+		goto err;
+	}
 
 	BUG_ON(state_read(&ep->com) != MPA_REQ_RCVD);
 	BUG_ON(!qp);
@@ -1789,20 +1812,21 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	if ((conn_param->ord > qp->rhp->attr.max_rdma_read_qp_depth) ||
 	    (conn_param->ird > qp->rhp->attr.max_rdma_reads_per_qp)) {
 		abort_connection(ep, NULL, GFP_KERNEL);
-		return -EINVAL;
+		err = -EINVAL;
+		goto err;
 	}
 
 	cm_id->add_ref(cm_id);
 	ep->com.cm_id = cm_id;
 	ep->com.qp = qp;
 
-	ep->com.rpl_done = 0;
-	ep->com.rpl_err = 0;
 	ep->ird = conn_param->ird;
 	ep->ord = conn_param->ord;
-	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
 
-	get_ep(&ep->com);
+	if (peer2peer && ep->ird == 0)
+		ep->ird = 1;
+
+	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
 
 	/* bind QP to EP and move to RTS */
 	attrs.mpa_attr = ep->mpa_attr;
@@ -1821,30 +1845,31 @@ int iwch_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	err = iwch_modify_qp(ep->com.qp->rhp,
 			     ep->com.qp, mask, &attrs, 1);
 	if (err)
-		goto err;
+		goto err1;
 
 	/* if needed, wait for wr_ack */
 	if (iwch_rqes_posted(qp)) {
 		wait_event(ep->com.waitq, ep->com.rpl_done);
 		err = ep->com.rpl_err;
 		if (err)
-			goto err;
+			goto err1;
 	}
 
 	err = send_mpa_reply(ep, conn_param->private_data,
 			     conn_param->private_data_len);
 	if (err)
-		goto err;
+		goto err1;
 
 
 	state_set(&ep->com, FPDU_MODE);
 	established_upcall(ep);
 	put_ep(&ep->com);
 	return 0;
-err:
+err1:
 	ep->com.cm_id = NULL;
 	ep->com.qp = NULL;
 	cm_id->rem_ref(cm_id);
+err:
 	put_ep(&ep->com);
 	return err;
 }
@@ -1852,8 +1877,9 @@ err:
 static int is_loopback_dst(struct iw_cm_id *cm_id)
 {
 	struct net_device *dev;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
 
-	dev = ip_dev_find(&init_net, cm_id->remote_addr.sin_addr.s_addr);
+	dev = ip_dev_find(&init_net, raddr->sin_addr.s_addr);
 	if (!dev)
 		return 0;
 	dev_put(dev);
@@ -1862,10 +1888,17 @@ static int is_loopback_dst(struct iw_cm_id *cm_id)
 
 int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 {
-	int err = 0;
 	struct iwch_dev *h = to_iwch_dev(cm_id->device);
 	struct iwch_ep *ep;
 	struct rtable *rt;
+	int err = 0;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
+
+	if (cm_id->remote_addr.ss_family != PF_INET) {
+		err = -ENOSYS;
+		goto out;
+	}
 
 	if (is_loopback_dst(cm_id)) {
 		err = -ENOSYS;
@@ -1885,6 +1918,10 @@ int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 		       conn_param->private_data, ep->plen);
 	ep->ird = conn_param->ird;
 	ep->ord = conn_param->ord;
+
+	if (peer2peer && ep->ord == 0)
+		ep->ord = 1;
+
 	ep->com.tdev = h->rdev.t3cdev_p;
 
 	cm_id->add_ref(cm_id);
@@ -1905,21 +1942,17 @@ int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	}
 
 	/* find a route */
-	rt = find_route(h->rdev.t3cdev_p,
-			cm_id->local_addr.sin_addr.s_addr,
-			cm_id->remote_addr.sin_addr.s_addr,
-			cm_id->local_addr.sin_port,
-			cm_id->remote_addr.sin_port, IPTOS_LOWDELAY);
+	rt = find_route(h->rdev.t3cdev_p, laddr->sin_addr.s_addr,
+			raddr->sin_addr.s_addr, laddr->sin_port,
+			raddr->sin_port, IPTOS_LOWDELAY);
 	if (!rt) {
 		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
 		err = -EHOSTUNREACH;
 		goto fail3;
 	}
-	ep->dst = &rt->u.dst;
-
-	/* get a l2t entry */
-	ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst->neighbour,
-			     ep->dst->neighbour->dev);
+	ep->dst = &rt->dst;
+	ep->l2t = t3_l2t_get(ep->com.tdev, ep->dst, NULL,
+			     &raddr->sin_addr.s_addr);
 	if (!ep->l2t) {
 		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
 		err = -ENOMEM;
@@ -1928,20 +1961,23 @@ int iwch_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 
 	state_set(&ep->com, CONNECTING);
 	ep->tos = IPTOS_LOWDELAY;
-	ep->com.local_addr = cm_id->local_addr;
-	ep->com.remote_addr = cm_id->remote_addr;
+	memcpy(&ep->com.local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
+	       sizeof(ep->com.remote_addr));
 
 	/* send connect request to rnic */
 	err = send_connect(ep);
 	if (!err)
 		goto out;
 
-	l2t_release(L2DATA(h->rdev.t3cdev_p), ep->l2t);
+	l2t_release(h->rdev.t3cdev_p, ep->l2t);
 fail4:
 	dst_release(ep->dst);
 fail3:
 	cxgb3_free_atid(ep->com.tdev, ep->atid);
 fail2:
+	cm_id->rem_ref(cm_id);
 	put_ep(&ep->com);
 out:
 	return err;
@@ -1956,6 +1992,11 @@ int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
 
 	might_sleep();
 
+	if (cm_id->local_addr.ss_family != PF_INET) {
+		err = -ENOSYS;
+		goto fail1;
+	}
+
 	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
 	if (!ep) {
 		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
@@ -1967,7 +2008,8 @@ int iwch_create_listen(struct iw_cm_id *cm_id, int backlog)
 	cm_id->add_ref(cm_id);
 	ep->com.cm_id = cm_id;
 	ep->backlog = backlog;
-	ep->com.local_addr = cm_id->local_addr;
+	memcpy(&ep->com.local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.local_addr));
 
 	/*
 	 * Allocate a server TID.
@@ -2013,8 +2055,11 @@ int iwch_destroy_listen(struct iw_cm_id *cm_id)
 	ep->com.rpl_done = 0;
 	ep->com.rpl_err = 0;
 	err = listen_stop(ep);
+	if (err)
+		goto done;
 	wait_event(ep->com.waitq, ep->com.rpl_done);
 	cxgb3_free_stid(ep->com.tdev, ep->stid);
+done:
 	err = ep->com.rpl_err;
 	cm_id->rem_ref(cm_id);
 	put_ep(&ep->com);
@@ -2026,12 +2071,22 @@ int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
 	int ret=0;
 	unsigned long flags;
 	int close = 0;
+	int fatal = 0;
+	struct t3cdev *tdev;
+	struct cxio_rdev *rdev;
 
 	spin_lock_irqsave(&ep->com.lock, flags);
 
 	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
 	     states[ep->com.state], abrupt);
 
+	tdev = (struct t3cdev *)ep->com.tdev;
+	rdev = (struct cxio_rdev *)tdev->ulp;
+	if (cxio_fatal_error(rdev)) {
+		fatal = 1;
+		close_complete_upcall(ep);
+		ep->com.state = DEAD;
+	}
 	switch (ep->com.state) {
 	case MPA_REQ_WAIT:
 	case MPA_REQ_SENT:
@@ -2045,14 +2100,17 @@ int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
 			ep->com.state = CLOSING;
 			start_ep_timer(ep);
 		}
+		set_bit(CLOSE_SENT, &ep->com.flags);
 		break;
 	case CLOSING:
-		close = 1;
-		if (abrupt) {
-			stop_ep_timer(ep);
-			ep->com.state = ABORTING;
-		} else
-			ep->com.state = MORIBUND;
+		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
+			close = 1;
+			if (abrupt) {
+				stop_ep_timer(ep);
+				ep->com.state = ABORTING;
+			} else
+				ep->com.state = MORIBUND;
+		}
 		break;
 	case MORIBUND:
 	case ABORTING:
@@ -2071,7 +2129,11 @@ int iwch_ep_disconnect(struct iwch_ep *ep, int abrupt, gfp_t gfp)
 			ret = send_abort(ep, NULL, gfp);
 		else
 			ret = send_halfclose(ep, gfp);
+		if (ret)
+			fatal = 1;
 	}
+	if (fatal)
+		release_ep_resources(ep);
 	return ret;
 }
 
@@ -2086,7 +2148,7 @@ int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
 	PDBG("%s ep %p redirect to dst %p l2t %p\n", __func__, ep, new,
 	     l2t);
 	dst_hold(new);
-	l2t_release(L2DATA(ep->com.tdev), ep->l2t);
+	l2t_release(ep->com.tdev, ep->l2t);
 	ep->l2t = l2t;
 	dst_release(old);
 	ep->dst = new;
@@ -2095,7 +2157,49 @@ int iwch_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
 
 /*
  * All the CM events are handled on a work queue to have a safe context.
+ * These are the real handlers that are called from the work queue.
  */
+static const cxgb3_cpl_handler_func work_handlers[NUM_CPL_CMDS] = {
+	[CPL_ACT_ESTABLISH]	= act_establish,
+	[CPL_ACT_OPEN_RPL]	= act_open_rpl,
+	[CPL_RX_DATA]		= rx_data,
+	[CPL_TX_DMA_ACK]	= tx_ack,
+	[CPL_ABORT_RPL_RSS]	= abort_rpl,
+	[CPL_ABORT_RPL]		= abort_rpl,
+	[CPL_PASS_OPEN_RPL]	= pass_open_rpl,
+	[CPL_CLOSE_LISTSRV_RPL]	= close_listsrv_rpl,
+	[CPL_PASS_ACCEPT_REQ]	= pass_accept_req,
+	[CPL_PASS_ESTABLISH]	= pass_establish,
+	[CPL_PEER_CLOSE]	= peer_close,
+	[CPL_ABORT_REQ_RSS]	= peer_abort,
+	[CPL_CLOSE_CON_RPL]	= close_con_rpl,
+	[CPL_RDMA_TERMINATE]	= terminate,
+	[CPL_RDMA_EC_STATUS]	= ec_status,
+};
+
+static void process_work(struct work_struct *work)
+{
+	struct sk_buff *skb = NULL;
+	void *ep;
+	struct t3cdev *tdev;
+	int ret;
+
+	while ((skb = skb_dequeue(&rxq))) {
+		ep = *((void **) (skb->cb));
+		tdev = *((struct t3cdev **) (skb->cb + sizeof(void *)));
+		ret = work_handlers[G_OPCODE(ntohl((__force __be32)skb->csum))](tdev, skb, ep);
+		if (ret & CPL_RET_BUF_DONE)
+			kfree_skb(skb);
+
+		/*
+		 * ep was referenced in sched(), and is freed here.
+		 */
+		put_ep((struct iwch_ep_common *)ep);
+	}
+}
+
+static DECLARE_WORK(skb_work, process_work);
+
 static int sched(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 {
 	struct iwch_ep_common *epc = ctx;
@@ -2127,6 +2231,29 @@ static int set_tcb_rpl(struct t3cdev *tdev, struct sk_buff *skb, void *ctx)
 	return CPL_RET_BUF_DONE;
 }
 
+/*
+ * All upcalls from the T3 Core go to sched() to schedule the
+ * processing on a work queue.
+ */
+cxgb3_cpl_handler_func t3c_handlers[NUM_CPL_CMDS] = {
+	[CPL_ACT_ESTABLISH]	= sched,
+	[CPL_ACT_OPEN_RPL]	= sched,
+	[CPL_RX_DATA]		= sched,
+	[CPL_TX_DMA_ACK]	= sched,
+	[CPL_ABORT_RPL_RSS]	= sched,
+	[CPL_ABORT_RPL]		= sched,
+	[CPL_PASS_OPEN_RPL]	= sched,
+	[CPL_CLOSE_LISTSRV_RPL]	= sched,
+	[CPL_PASS_ACCEPT_REQ]	= sched,
+	[CPL_PASS_ESTABLISH]	= sched,
+	[CPL_PEER_CLOSE]	= sched,
+	[CPL_CLOSE_CON_RPL]	= sched,
+	[CPL_ABORT_REQ_RSS]	= sched,
+	[CPL_RDMA_TERMINATE]	= sched,
+	[CPL_RDMA_EC_STATUS]	= sched,
+	[CPL_SET_TCB_RPL]	= set_tcb_rpl,
+};
+
 int __init iwch_cm_init(void)
 {
 	skb_queue_head_init(&rxq);
@@ -2135,46 +2262,6 @@ int __init iwch_cm_init(void)
 	if (!workq)
 		return -ENOMEM;
 
-	/*
-	 * All upcalls from the T3 Core go to sched() to
-	 * schedule the processing on a work queue.
-	 */
-	t3c_handlers[CPL_ACT_ESTABLISH] = sched;
-	t3c_handlers[CPL_ACT_OPEN_RPL] = sched;
-	t3c_handlers[CPL_RX_DATA] = sched;
-	t3c_handlers[CPL_TX_DMA_ACK] = sched;
-	t3c_handlers[CPL_ABORT_RPL_RSS] = sched;
-	t3c_handlers[CPL_ABORT_RPL] = sched;
-	t3c_handlers[CPL_PASS_OPEN_RPL] = sched;
-	t3c_handlers[CPL_CLOSE_LISTSRV_RPL] = sched;
-	t3c_handlers[CPL_PASS_ACCEPT_REQ] = sched;
-	t3c_handlers[CPL_PASS_ESTABLISH] = sched;
-	t3c_handlers[CPL_PEER_CLOSE] = sched;
-	t3c_handlers[CPL_CLOSE_CON_RPL] = sched;
-	t3c_handlers[CPL_ABORT_REQ_RSS] = sched;
-	t3c_handlers[CPL_RDMA_TERMINATE] = sched;
-	t3c_handlers[CPL_RDMA_EC_STATUS] = sched;
-	t3c_handlers[CPL_SET_TCB_RPL] = set_tcb_rpl;
-
-	/*
-	 * These are the real handlers that are called from a
-	 * work queue.
-	 */
-	work_handlers[CPL_ACT_ESTABLISH] = act_establish;
-	work_handlers[CPL_ACT_OPEN_RPL] = act_open_rpl;
-	work_handlers[CPL_RX_DATA] = rx_data;
-	work_handlers[CPL_TX_DMA_ACK] = tx_ack;
-	work_handlers[CPL_ABORT_RPL_RSS] = abort_rpl;
-	work_handlers[CPL_ABORT_RPL] = abort_rpl;
-	work_handlers[CPL_PASS_OPEN_RPL] = pass_open_rpl;
-	work_handlers[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl;
-	work_handlers[CPL_PASS_ACCEPT_REQ] = pass_accept_req;
-	work_handlers[CPL_PASS_ESTABLISH] = pass_establish;
-	work_handlers[CPL_PEER_CLOSE] = peer_close;
-	work_handlers[CPL_ABORT_REQ_RSS] = peer_abort;
-	work_handlers[CPL_CLOSE_CON_RPL] = close_con_rpl;
-	work_handlers[CPL_RDMA_TERMINATE] = terminate;
-	work_handlers[CPL_RDMA_EC_STATUS] = ec_status;
 	return 0;
 }
 
diff --git a/drivers/infiniband/hw/cxgb3/iwch_cm.h b/drivers/infiniband/hw/cxgb3/iwch_cm.h
index d7c7e09f099..b9efadfffb4 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_cm.h
+++ b/drivers/infiniband/hw/cxgb3/iwch_cm.h
@@ -145,8 +145,10 @@ enum iwch_ep_state {
 };
 
 enum iwch_ep_flags {
-	PEER_ABORT_IN_PROGRESS	= (1 << 0),
-	ABORT_REQ_IN_PROGRESS	= (1 << 1),
+	PEER_ABORT_IN_PROGRESS	= 0,
+	ABORT_REQ_IN_PROGRESS	= 1,
+	RELEASE_RESOURCES	= 2,
+	CLOSE_SENT		= 3,
 };
 
 struct iwch_ep_common {
@@ -161,6 +163,7 @@ struct iwch_ep_common {
 	wait_queue_head_t waitq;
 	int rpl_done;
 	int rpl_err;
+	unsigned long flags;
 };
 
 struct iwch_listen_ep {
@@ -188,7 +191,6 @@ struct iwch_ep {
 	u16 plen;
 	u32 ird;
 	u32 ord;
-	u32 flags;
 };
 
 static inline struct iwch_ep *to_ep(struct iw_cm_id *cm_id)
diff --git a/drivers/infiniband/hw/cxgb3/iwch_ev.c b/drivers/infiniband/hw/cxgb3/iwch_ev.c
index 7b67a677172..abcc9e76962 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_ev.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_ev.c
@@ -29,7 +29,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
-#include <linux/slab.h>
+#include <linux/gfp.h>
 #include <linux/mman.h>
 #include <net/sock.h>
 #include "iwch_provider.h"
@@ -46,6 +46,7 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
 	struct ib_event event;
 	struct iwch_qp_attributes attrs;
 	struct iwch_qp *qhp;
+	unsigned long flag;
 
 	spin_lock(&rnicp->lock);
 	qhp = get_qhp(rnicp, CQE_QPID(rsp_msg->cqe));
@@ -76,6 +77,14 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
 	atomic_inc(&qhp->refcnt);
 	spin_unlock(&rnicp->lock);
 
+	if (qhp->attr.state == IWCH_QP_STATE_RTS) {
+		attrs.next_state = IWCH_QP_STATE_TERMINATE;
+		iwch_modify_qp(qhp->rhp, qhp, IWCH_QP_ATTR_NEXT_STATE,
+			       &attrs, 1);
+		if (send_term)
+			iwch_post_terminate(qhp, rsp_msg);
+	}
+
 	event.event = ib_event;
 	event.device = chp->ibcq.device;
 	if (ib_event == IB_EVENT_CQ_ERR)
@@ -86,13 +95,9 @@ static void post_qp_event(struct iwch_dev *rnicp, struct iwch_cq *chp,
 	if (qhp->ibqp.event_handler)
 		(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
 
-	if (qhp->attr.state == IWCH_QP_STATE_RTS) {
-		attrs.next_state = IWCH_QP_STATE_TERMINATE;
-		iwch_modify_qp(qhp->rhp, qhp, IWCH_QP_ATTR_NEXT_STATE,
-			       &attrs, 1);
-		if (send_term)
-			iwch_post_terminate(qhp, rsp_msg);
-	}
+	spin_lock_irqsave(&chp->comp_handler_lock, flag);
+	(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+	spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 
 	if (atomic_dec_and_test(&qhp->refcnt))
 		wake_up(&qhp->wait);
@@ -105,6 +110,7 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
 	struct iwch_cq *chp;
 	struct iwch_qp *qhp;
 	u32 cqid = RSPQ_CQID(rsp_msg);
+	unsigned long flag;
 
 	rnicp = (struct iwch_dev *) rdev_p->ulp;
 	spin_lock(&rnicp->lock);
@@ -168,7 +174,9 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
 		 */
 		if (qhp->ep && SQ_TYPE(rsp_msg->cqe))
 			dst_confirm(qhp->ep->dst);
+		spin_lock_irqsave(&chp->comp_handler_lock, flag);
 		(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+		spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
 		break;
 
 	case TPT_ERR_STAG:
@@ -179,12 +187,6 @@ void iwch_ev_dispatch(struct cxio_rdev *rdev_p, struct sk_buff *skb)
 	case TPT_ERR_BOUND:
 	case TPT_ERR_INVALIDATE_SHARED_MR:
 	case TPT_ERR_INVALIDATE_MR_WITH_MW_BOUND:
-		printk(KERN_ERR "%s - CQE Err qpid 0x%x opcode %d status 0x%x "
-		       "type %d wrid.hi 0x%x wrid.lo 0x%x \n", __func__,
-		       CQE_QPID(rsp_msg->cqe), CQE_OPCODE(rsp_msg->cqe),
-		       CQE_STATUS(rsp_msg->cqe), CQE_TYPE(rsp_msg->cqe),
-		       CQE_WRID_HI(rsp_msg->cqe), CQE_WRID_LOW(rsp_msg->cqe));
-		(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
 		post_qp_event(rnicp, chp, rsp_msg, IB_EVENT_QP_ACCESS_ERR, 1);
 		break;
 
diff --git a/drivers/infiniband/hw/cxgb3/iwch_mem.c b/drivers/infiniband/hw/cxgb3/iwch_mem.c
index ec49a5cbdeb..5c36ee2809a 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_mem.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_mem.c
@@ -29,6 +29,7 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include <linux/slab.h>
 #include <asm/byteorder.h>
 
 #include <rdma/iw_cm.h>
@@ -39,7 +40,7 @@
 #include "iwch.h"
 #include "iwch_provider.h"
 
-static void iwch_finish_mem_reg(struct iwch_mr *mhp, u32 stag)
+static int iwch_finish_mem_reg(struct iwch_mr *mhp, u32 stag)
 {
 	u32 mmid;
 
@@ -47,14 +48,15 @@ static void iwch_finish_mem_reg(struct iwch_mr *mhp, u32 stag)
 	mhp->attr.stag = stag;
 	mmid = stag >> 8;
 	mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
-	insert_handle(mhp->rhp, &mhp->rhp->mmidr, mhp, mmid);
 	PDBG("%s mmid 0x%x mhp %p\n", __func__, mmid, mhp);
+	return insert_handle(mhp->rhp, &mhp->rhp->mmidr, mhp, mmid);
 }
 
 int iwch_register_mem(struct iwch_dev *rhp, struct iwch_pd *php,
 		      struct iwch_mr *mhp, int shift)
 {
 	u32 stag;
+	int ret;
 
 	if (cxio_register_phys_mem(&rhp->rdev,
 				   &stag, mhp->attr.pdid,
@@ -66,9 +68,11 @@ int iwch_register_mem(struct iwch_dev *rhp, struct iwch_pd *php,
 				   mhp->attr.pbl_size, mhp->attr.pbl_addr))
 		return -ENOMEM;
 
-	iwch_finish_mem_reg(mhp, stag);
-
-	return 0;
+	ret = iwch_finish_mem_reg(mhp, stag);
+	if (ret)
+		cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+	return ret;
 }
 
 int iwch_reregister_mem(struct iwch_dev *rhp, struct iwch_pd *php,
@@ -77,6 +81,7 @@ int iwch_reregister_mem(struct iwch_dev *rhp, struct iwch_pd *php,
 					int npages)
 {
 	u32 stag;
+	int ret;
 
 	/* We could support this... */
 	if (npages > mhp->attr.pbl_size)
@@ -93,9 +98,12 @@ int iwch_reregister_mem(struct iwch_dev *rhp, struct iwch_pd *php,
 				   mhp->attr.pbl_size, mhp->attr.pbl_addr))
 		return -ENOMEM;
 
-	iwch_finish_mem_reg(mhp, stag);
+	ret = iwch_finish_mem_reg(mhp, stag);
+	if (ret)
+		cxio_dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
 
-	return 0;
+	return ret;
 }
 
 int iwch_alloc_pbl(struct iwch_mr *mhp, int npages)
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.c b/drivers/infiniband/hw/cxgb3/iwch_provider.c
index b89640aa6e1..811b24a539c 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.c
@@ -37,9 +37,12 @@
 #include <linux/delay.h>
 #include <linux/errno.h>
 #include <linux/list.h>
+#include <linux/sched.h>
 #include <linux/spinlock.h>
 #include <linux/ethtool.h>
 #include <linux/rtnetlink.h>
+#include <linux/inetdevice.h>
+#include <linux/slab.h>
 
 #include <asm/io.h>
 #include <asm/irq.h>
@@ -58,13 +61,6 @@
 #include "iwch_user.h"
 #include "common.h"
 
-static int iwch_modify_port(struct ib_device *ibdev,
-			    u8 port, int port_modify_mask,
-			    struct ib_port_modify *props)
-{
-	return -ENOSYS;
-}
-
 static struct ib_ah *iwch_ah_create(struct ib_pd *pd,
 				    struct ib_ah_attr *ah_attr)
 {
@@ -151,6 +147,8 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve
 	struct iwch_create_cq_resp uresp;
 	struct iwch_create_cq_req ureq;
 	struct iwch_ucontext *ucontext = NULL;
+	static int warned;
+	size_t resplen;
 
 	PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
 	rhp = to_iwch_dev(ibdev);
@@ -185,16 +183,21 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve
 	entries = roundup_pow_of_two(entries);
 	chp->cq.size_log2 = ilog2(entries);
 
-	if (cxio_create_cq(&rhp->rdev, &chp->cq)) {
+	if (cxio_create_cq(&rhp->rdev, &chp->cq, !ucontext)) {
 		kfree(chp);
 		return ERR_PTR(-ENOMEM);
 	}
 	chp->rhp = rhp;
 	chp->ibcq.cqe = 1 << chp->cq.size_log2;
 	spin_lock_init(&chp->lock);
+	spin_lock_init(&chp->comp_handler_lock);
 	atomic_set(&chp->refcnt, 1);
 	init_waitqueue_head(&chp->wait);
-	insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
+	if (insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid)) {
+		cxio_destroy_cq(&chp->rhp->rdev, &chp->cq);
+		kfree(chp);
+		return ERR_PTR(-ENOMEM);
+	}
 
 	if (ucontext) {
 		struct iwch_mm_entry *mm;
@@ -210,15 +213,27 @@ static struct ib_cq *iwch_create_cq(struct ib_device *ibdev, int entries, int ve
 		uresp.key = ucontext->key;
 		ucontext->key += PAGE_SIZE;
 		spin_unlock(&ucontext->mmap_lock);
-		if (ib_copy_to_udata(udata, &uresp, sizeof (uresp))) {
+		mm->key = uresp.key;
+		mm->addr = virt_to_phys(chp->cq.queue);
+		if (udata->outlen < sizeof uresp) {
+			if (!warned++)
+				printk(KERN_WARNING MOD "Warning - "
+				       "downlevel libcxgb3 (non-fatal).\n");
+			mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
+					     sizeof(struct t3_cqe));
+			resplen = sizeof(struct iwch_create_cq_resp_v0);
+		} else {
+			mm->len = PAGE_ALIGN(((1UL << uresp.size_log2) + 1) *
+					     sizeof(struct t3_cqe));
+			uresp.memsize = mm->len;
+			uresp.reserved = 0;
+			resplen = sizeof uresp;
+		}
+		if (ib_copy_to_udata(udata, &uresp, resplen)) {
 			kfree(mm);
 			iwch_destroy_cq(&chp->ibcq);
 			return ERR_PTR(-EFAULT);
 		}
-		mm->key = uresp.key;
-		mm->addr = virt_to_phys(chp->cq.queue);
-		mm->len = PAGE_ALIGN((1UL << uresp.size_log2) *
-					     sizeof (struct t3_cqe));
 		insert_mmap(ucontext, mm);
 	}
 	PDBG("created cqid 0x%0x chp %p size 0x%0x, dma_addr 0x%0llx\n",
@@ -545,7 +560,7 @@ static int iwch_reregister_phys_mem(struct ib_mr *mr,
 	__be64 *page_list = NULL;
 	int shift = 0;
 	u64 total_size;
-	int npages;
+	int npages = 0;
 	int ret;
 
 	PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
@@ -603,14 +618,13 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	__be64 *pages;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
-	struct ib_umem_chunk *chunk;
 	struct iwch_dev *rhp;
 	struct iwch_pd *php;
 	struct iwch_mr *mhp;
 	struct iwch_reg_user_mr_resp uresp;
-
+	struct scatterlist *sg;
 	PDBG("%s ib_pd %p\n", __func__, pd);
 
 	php = to_iwch_pd(pd);
@@ -630,9 +644,7 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	shift = ffs(mhp->umem->page_size) - 1;
 
-	n = 0;
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		n += chunk->nents;
+	n = mhp->umem->nmap;
 
 	err = iwch_alloc_pbl(mhp, n);
 	if (err)
@@ -646,12 +658,10 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	i = n = 0;
 
-	list_for_each_entry(chunk, &mhp->umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
+	for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
+			len = sg_dma_len(sg) >> shift;
 			for (k = 0; k < len; ++k) {
-				pages[i++] = cpu_to_be64(sg_dma_address(
-					&chunk->page_list[j]) +
+				pages[i++] = cpu_to_be64(sg_dma_address(sg) +
 					mhp->umem->page_size * k);
 				if (i == PAGE_SIZE / sizeof *pages) {
 					err = iwch_write_pbl(mhp, pages, i, n);
@@ -661,7 +671,7 @@ static struct ib_mr *iwch_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 					i = 0;
 				}
 			}
-		}
+	}
 
 	if (i)
 		err = iwch_write_pbl(mhp, pages, i, n);
@@ -724,7 +734,7 @@ static struct ib_mr *iwch_get_dma_mr(struct ib_pd *pd, int acc)
 	return ibmr;
 }
 
-static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
+static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
 {
 	struct iwch_dev *rhp;
 	struct iwch_pd *php;
@@ -733,6 +743,9 @@ static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 	u32 stag = 0;
 	int ret;
 
+	if (type != IB_MW_TYPE_1)
+		return ERR_PTR(-EINVAL);
+
 	php = to_iwch_pd(pd);
 	rhp = php->rhp;
 	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
@@ -749,7 +762,11 @@ static struct ib_mw *iwch_alloc_mw(struct ib_pd *pd)
 	mhp->attr.stag = stag;
 	mmid = (stag) >> 8;
 	mhp->ibmw.rkey = stag;
-	insert_handle(rhp, &rhp->mmidr, mhp, mmid);
+	if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
+		cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
+		kfree(mhp);
+		return ERR_PTR(-ENOMEM);
+	}
 	PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 	return &(mhp->ibmw);
 }
@@ -765,8 +782,8 @@ static int iwch_dealloc_mw(struct ib_mw *mw)
 	mmid = (mw->rkey) >> 8;
 	cxio_deallocate_window(&rhp->rdev, mhp->attr.stag);
 	remove_handle(rhp, &rhp->mmidr, mmid);
-	kfree(mhp);
 	PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
+	kfree(mhp);
 	return 0;
 }
 
@@ -777,37 +794,43 @@ static struct ib_mr *iwch_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
 	struct iwch_mr *mhp;
 	u32 mmid;
 	u32 stag = 0;
-	int ret;
+	int ret = 0;
 
 	php = to_iwch_pd(pd);
 	rhp = php->rhp;
 	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
 	if (!mhp)
-		return ERR_PTR(-ENOMEM);
+		goto err;
 
 	mhp->rhp = rhp;
 	ret = iwch_alloc_pbl(mhp, pbl_depth);
-	if (ret) {
-		kfree(mhp);
-		return ERR_PTR(ret);
-	}
+	if (ret)
+		goto err1;
 	mhp->attr.pbl_size = pbl_depth;
 	ret = cxio_allocate_stag(&rhp->rdev, &stag, php->pdid,
 				 mhp->attr.pbl_size, mhp->attr.pbl_addr);
-	if (ret) {
-		iwch_free_pbl(mhp);
-		kfree(mhp);
-		return ERR_PTR(ret);
-	}
+	if (ret)
+		goto err2;
 	mhp->attr.pdid = php->pdid;
 	mhp->attr.type = TPT_NON_SHARED_MR;
 	mhp->attr.stag = stag;
 	mhp->attr.state = 1;
 	mmid = (stag) >> 8;
 	mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
-	insert_handle(rhp, &rhp->mmidr, mhp, mmid);
+	if (insert_handle(rhp, &rhp->mmidr, mhp, mmid))
+		goto err3;
+
 	PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
 	return &(mhp->ibmr);
+err3:
+	cxio_dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+err2:
+	iwch_free_pbl(mhp);
+err1:
+	kfree(mhp);
+err:
+	return ERR_PTR(ret);
 }
 
 static struct ib_fast_reg_page_list *iwch_alloc_fastreg_pbl(
@@ -960,7 +983,13 @@ static struct ib_qp *iwch_create_qp(struct ib_pd *pd,
 	spin_lock_init(&qhp->lock);
 	init_waitqueue_head(&qhp->wait);
 	atomic_set(&qhp->refcnt, 1);
-	insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid);
+
+	if (insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.qpid)) {
+		cxio_destroy_qp(&rhp->rdev, &qhp->wq,
+			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+		kfree(qhp);
+		return ERR_PTR(-ENOMEM);
+	}
 
 	if (udata) {
 
@@ -1102,9 +1131,7 @@ static u64 fw_vers_string_to_u64(struct iwch_dev *iwch_dev)
 	char *cp, *next;
 	unsigned fw_maj, fw_min, fw_mic;
 
-	rtnl_lock();
 	lldev->ethtool_ops->get_drvinfo(lldev, &info);
-	rtnl_unlock();
 
 	next = info.fw_version + 1;
 	cp = strsep(&next, ".");
@@ -1154,14 +1181,42 @@ static int iwch_query_device(struct ib_device *ibdev,
 static int iwch_query_port(struct ib_device *ibdev,
 			   u8 port, struct ib_port_attr *props)
 {
+	struct iwch_dev *dev;
+	struct net_device *netdev;
+	struct in_device *inetdev;
+
 	PDBG("%s ibdev %p\n", __func__, ibdev);
+
+	dev = to_iwch_dev(ibdev);
+	netdev = dev->rdev.port_info.lldevs[port-1];
+
+	memset(props, 0, sizeof(struct ib_port_attr));
 	props->max_mtu = IB_MTU_4096;
-	props->lid = 0;
-	props->lmc = 0;
-	props->sm_lid = 0;
-	props->sm_sl = 0;
-	props->state = IB_PORT_ACTIVE;
-	props->phys_state = 0;
+	if (netdev->mtu >= 4096)
+		props->active_mtu = IB_MTU_4096;
+	else if (netdev->mtu >= 2048)
+		props->active_mtu = IB_MTU_2048;
+	else if (netdev->mtu >= 1024)
+		props->active_mtu = IB_MTU_1024;
+	else if (netdev->mtu >= 512)
+		props->active_mtu = IB_MTU_512;
+	else
+		props->active_mtu = IB_MTU_256;
+
+	if (!netif_carrier_ok(netdev))
+		props->state = IB_PORT_DOWN;
+	else {
+		inetdev = in_dev_get(netdev);
+		if (inetdev) {
+			if (inetdev->ifa_list)
+				props->state = IB_PORT_ACTIVE;
+			else
+				props->state = IB_PORT_INIT;
+			in_dev_put(inetdev);
+		} else
+			props->state = IB_PORT_INIT;
+	}
+
 	props->port_cap_flags =
 	    IB_PORT_CM_SUP |
 	    IB_PORT_SNMP_TUNNEL_SUP |
@@ -1170,9 +1225,8 @@ static int iwch_query_port(struct ib_device *ibdev,
 	    IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
 	props->gid_tbl_len = 1;
 	props->pkey_tbl_len = 1;
-	props->qkey_viol_cntr = 0;
 	props->active_width = 2;
-	props->active_speed = 2;
+	props->active_speed = IB_SPEED_DDR;
 	props->max_msg_sz = -1;
 
 	return 0;
@@ -1187,28 +1241,6 @@ static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
 	return sprintf(buf, "%d\n", iwch_dev->rdev.t3cdev_p->type);
 }
 
-static int fw_supports_fastreg(struct iwch_dev *iwch_dev)
-{
-	struct ethtool_drvinfo info;
-	struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
-	char *cp, *next;
-	unsigned fw_maj, fw_min;
-
-	rtnl_lock();
-	lldev->ethtool_ops->get_drvinfo(lldev, &info);
-	rtnl_unlock();
-
-	next = info.fw_version+1;
-	cp = strsep(&next, ".");
-	sscanf(cp, "%i", &fw_maj);
-	cp = strsep(&next, ".");
-	sscanf(cp, "%i", &fw_min);
-
-	PDBG("%s maj %u min %u\n", __func__, fw_maj, fw_min);
-
-	return fw_maj > 6 || (fw_maj == 6 && fw_min > 0);
-}
-
 static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, char *buf)
 {
 	struct iwch_dev *iwch_dev = container_of(dev, struct iwch_dev,
@@ -1217,9 +1249,7 @@ static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr, ch
 	struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
 
 	PDBG("%s dev 0x%p\n", __func__, dev);
-	rtnl_lock();
 	lldev->ethtool_ops->get_drvinfo(lldev, &info);
-	rtnl_unlock();
 	return sprintf(buf, "%s\n", info.fw_version);
 }
 
@@ -1232,9 +1262,7 @@ static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
 	struct net_device *lldev = iwch_dev->rdev.t3cdev_p->lldev;
 
 	PDBG("%s dev 0x%p\n", __func__, dev);
-	rtnl_lock();
 	lldev->ethtool_ops->get_drvinfo(lldev, &info);
-	rtnl_unlock();
 	return sprintf(buf, "%s\n", info.driver);
 }
 
@@ -1325,12 +1353,12 @@ int iwch_register_device(struct iwch_dev *dev)
 	memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
 	memcpy(&dev->ibdev.node_guid, dev->rdev.t3cdev_p->lldev->dev_addr, 6);
 	dev->ibdev.owner = THIS_MODULE;
-	dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_WINDOW;
+	dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
+				IB_DEVICE_MEM_WINDOW |
+				IB_DEVICE_MEM_MGT_EXTENSIONS;
 
 	/* cxgb3 supports STag 0. */
 	dev->ibdev.local_dma_lkey = 0;
-	if (fw_supports_fastreg(dev))
-		dev->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
 
 	dev->ibdev.uverbs_cmd_mask =
 	    (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
@@ -1357,7 +1385,6 @@ int iwch_register_device(struct iwch_dev *dev)
 	dev->ibdev.dma_device = &(dev->rdev.rnic_info.pdev->dev);
 	dev->ibdev.query_device = iwch_query_device;
 	dev->ibdev.query_port = iwch_query_port;
-	dev->ibdev.modify_port = iwch_modify_port;
 	dev->ibdev.query_pkey = iwch_query_pkey;
 	dev->ibdev.query_gid = iwch_query_gid;
 	dev->ibdev.alloc_ucontext = iwch_alloc_ucontext;
@@ -1392,6 +1419,7 @@ int iwch_register_device(struct iwch_dev *dev)
 	dev->ibdev.post_send = iwch_post_send;
 	dev->ibdev.post_recv = iwch_post_receive;
 	dev->ibdev.get_protocol_stats = iwch_get_mib;
+	dev->ibdev.uverbs_abi_ver = IWCH_UVERBS_ABI_VERSION;
 
 	dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
 	if (!dev->ibdev.iwcm)
@@ -1406,7 +1434,7 @@ int iwch_register_device(struct iwch_dev *dev)
 	dev->ibdev.iwcm->rem_ref = iwch_qp_rem_ref;
 	dev->ibdev.iwcm->get_qp = iwch_get_qp;
 
-	ret = ib_register_device(&dev->ibdev);
+	ret = ib_register_device(&dev->ibdev, NULL);
 	if (ret)
 		goto bail1;
 
@@ -1421,6 +1449,7 @@ int iwch_register_device(struct iwch_dev *dev)
 bail2:
 	ib_unregister_device(&dev->ibdev);
 bail1:
+	kfree(dev->ibdev.iwcm);
 	return ret;
 }
 
@@ -1433,5 +1462,6 @@ void iwch_unregister_device(struct iwch_dev *dev)
 		device_remove_file(&dev->ibdev.dev,
 				   iwch_class_attributes[i]);
 	ib_unregister_device(&dev->ibdev);
+	kfree(dev->ibdev.iwcm);
 	return;
 }
diff --git a/drivers/infiniband/hw/cxgb3/iwch_provider.h b/drivers/infiniband/hw/cxgb3/iwch_provider.h
index f5ceca05c43..87c14b0c5ac 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_provider.h
+++ b/drivers/infiniband/hw/cxgb3/iwch_provider.h
@@ -103,6 +103,7 @@ struct iwch_cq {
 	struct iwch_dev *rhp;
 	struct t3_cq cq;
 	spinlock_t lock;
+	spinlock_t comp_handler_lock;
 	atomic_t refcnt;
 	wait_queue_head_t wait;
 	u32 __user *user_rptr_addr;
@@ -293,9 +294,16 @@ static inline u32 iwch_ib_to_tpt_access(int acc)
 	return (acc & IB_ACCESS_REMOTE_WRITE ? TPT_REMOTE_WRITE : 0) |
 	       (acc & IB_ACCESS_REMOTE_READ ? TPT_REMOTE_READ : 0) |
 	       (acc & IB_ACCESS_LOCAL_WRITE ? TPT_LOCAL_WRITE : 0) |
+	       (acc & IB_ACCESS_MW_BIND ? TPT_MW_BIND : 0) |
 	       TPT_LOCAL_READ;
 }
 
+static inline u32 iwch_ib_to_tpt_bind_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_WRITE ? TPT_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ ? TPT_REMOTE_READ : 0);
+}
+
 enum iwch_mmid_state {
 	IWCH_STAG_STATE_VALID,
 	IWCH_STAG_STATE_INVALID
@@ -325,11 +333,9 @@ int iwch_bind_mw(struct ib_qp *qp,
 			     struct ib_mw_bind *mw_bind);
 int iwch_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
 int iwch_post_terminate(struct iwch_qp *qhp, struct respQ_msg_t *rsp_msg);
-int iwch_post_zb_read(struct iwch_qp *qhp);
+int iwch_post_zb_read(struct iwch_ep *ep);
 int iwch_register_device(struct iwch_dev *dev);
 void iwch_unregister_device(struct iwch_dev *dev);
-int iwch_quiesce_qps(struct iwch_cq *chp);
-int iwch_resume_qps(struct iwch_cq *chp);
 void stop_read_rep_timer(struct iwch_qp *qhp);
 int iwch_register_mem(struct iwch_dev *rhp, struct iwch_pd *php,
 		      struct iwch_mr *mhp, int shift);
diff --git a/drivers/infiniband/hw/cxgb3/iwch_qp.c b/drivers/infiniband/hw/cxgb3/iwch_qp.c
index 9a3be3a9d5d..b57c0befd96 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_qp.c
+++ b/drivers/infiniband/hw/cxgb3/iwch_qp.c
@@ -29,6 +29,8 @@
  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  * SOFTWARE.
  */
+#include <linux/sched.h>
+#include <linux/gfp.h>
 #include "iwch_provider.h"
 #include "iwch.h"
 #include "iwch_cm.h"
@@ -99,8 +101,8 @@ static int build_rdma_write(union t3_wr *wqe, struct ib_send_wr *wr,
 	if (wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
 		plen = 4;
 		wqe->write.sgl[0].stag = wr->ex.imm_data;
-		wqe->write.sgl[0].len = __constant_cpu_to_be32(0);
-		wqe->write.num_sgle = __constant_cpu_to_be32(0);
+		wqe->write.sgl[0].len = cpu_to_be32(0);
+		wqe->write.num_sgle = cpu_to_be32(0);
 		*flit_cnt = 6;
 	} else {
 		plen = 0;
@@ -195,15 +197,12 @@ static int build_inv_stag(union t3_wr *wqe, struct ib_send_wr *wr,
 	return 0;
 }
 
-/*
- * TBD: this is going to be moved to firmware. Missing pdid/qpid check for now.
- */
 static int iwch_sgl2pbl_map(struct iwch_dev *rhp, struct ib_sge *sg_list,
 			    u32 num_sgle, u32 * pbl_addr, u8 * page_size)
 {
 	int i;
 	struct iwch_mr *mhp;
-	u32 offset;
+	u64 offset;
 	for (i = 0; i < num_sgle; i++) {
 
 		mhp = get_mhp(rhp, (sg_list[i].lkey) >> 8);
@@ -235,8 +234,8 @@ static int iwch_sgl2pbl_map(struct iwch_dev *rhp, struct ib_sge *sg_list,
 			return -EINVAL;
 		}
 		offset = sg_list[i].addr - mhp->attr.va_fbo;
-		offset += ((u32) mhp->attr.va_fbo) %
-		          (1UL << (12 + mhp->attr.page_size));
+		offset += mhp->attr.va_fbo &
+			  ((1UL << (12 + mhp->attr.page_size)) - 1);
 		pbl_addr[i] = ((mhp->attr.pbl_addr -
 			        rhp->rdev.rnic_info.pbl_base) >> 3) +
 			      (offset >> (12 + mhp->attr.page_size));
@@ -266,8 +265,8 @@ static int build_rdma_recv(struct iwch_qp *qhp, union t3_wr *wqe,
 		wqe->recv.sgl[i].len = cpu_to_be32(wr->sg_list[i].length);
 
 		/* to in the WQE == the offset into the page */
-		wqe->recv.sgl[i].to = cpu_to_be64(((u32) wr->sg_list[i].addr) %
-				(1UL << (12 + page_size[i])));
+		wqe->recv.sgl[i].to = cpu_to_be64(((u32)wr->sg_list[i].addr) &
+				((1UL << (12 + page_size[i])) - 1));
 
 		/* pbl_addr is the adapters address in the PBL */
 		wqe->recv.pbl_addr[i] = cpu_to_be32(pbl_addr[i]);
@@ -367,18 +366,19 @@ int iwch_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	spin_lock_irqsave(&qhp->lock, flag);
 	if (qhp->attr.state > IWCH_QP_STATE_RTS) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
-		return -EINVAL;
+		err = -EINVAL;
+		goto out;
 	}
 	num_wrs = Q_FREECNT(qhp->wq.sq_rptr, qhp->wq.sq_wptr,
 		  qhp->wq.sq_size_log2);
-	if (num_wrs <= 0) {
+	if (num_wrs == 0) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
-		return -ENOMEM;
+		err = -ENOMEM;
+		goto out;
 	}
 	while (wr) {
 		if (num_wrs == 0) {
 			err = -ENOMEM;
-			*bad_wr = wr;
 			break;
 		}
 		idx = Q_PTR2IDX(qhp->wq.wptr, qhp->wq.size_log2);
@@ -430,10 +430,8 @@ int iwch_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			     wr->opcode);
 			err = -EINVAL;
 		}
-		if (err) {
-			*bad_wr = wr;
+		if (err)
 			break;
-		}
 		wqe->send.wrid.id0.hi = qhp->wq.sq_wptr;
 		sqp->wr_id = wr->wr_id;
 		sqp->opcode = wr2opcode(t3_wr_opcode);
@@ -455,7 +453,12 @@ int iwch_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		++(qhp->wq.sq_wptr);
 	}
 	spin_unlock_irqrestore(&qhp->lock, flag);
-	ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
+	if (cxio_wq_db_enabled(&qhp->wq))
+		ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
+
+out:
+	if (err)
+		*bad_wr = wr;
 	return err;
 }
 
@@ -473,18 +476,19 @@ int iwch_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 	spin_lock_irqsave(&qhp->lock, flag);
 	if (qhp->attr.state > IWCH_QP_STATE_RTS) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
-		return -EINVAL;
+		err = -EINVAL;
+		goto out;
 	}
 	num_wrs = Q_FREECNT(qhp->wq.rq_rptr, qhp->wq.rq_wptr,
 			    qhp->wq.rq_size_log2) - 1;
 	if (!wr) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
-		return -EINVAL;
+		err = -ENOMEM;
+		goto out;
 	}
 	while (wr) {
 		if (wr->num_sge > T3_MAX_SGE) {
 			err = -EINVAL;
-			*bad_wr = wr;
 			break;
 		}
 		idx = Q_PTR2IDX(qhp->wq.wptr, qhp->wq.size_log2);
@@ -496,10 +500,10 @@ int iwch_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 				err = build_zero_stag_recv(qhp, wqe, wr);
 		else
 			err = -ENOMEM;
-		if (err) {
-			*bad_wr = wr;
+
+		if (err)
 			break;
-		}
+
 		build_fw_riwrh((void *) wqe, T3_WR_RCV, T3_COMPLETION_FLAG,
 			       Q_GENBIT(qhp->wq.wptr, qhp->wq.size_log2),
 			       0, sizeof(struct t3_receive_wr) >> 3, T3_SOPEOP);
@@ -512,7 +516,12 @@ int iwch_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 		num_wrs--;
 	}
 	spin_unlock_irqrestore(&qhp->lock, flag);
-	ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
+	if (cxio_wq_db_enabled(&qhp->wq))
+		ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
+
+out:
+	if (err)
+		*bad_wr = wr;
 	return err;
 }
 
@@ -545,7 +554,7 @@ int iwch_bind_mw(struct ib_qp *qp,
 	}
 	num_wrs = Q_FREECNT(qhp->wq.sq_rptr, qhp->wq.sq_wptr,
 			    qhp->wq.sq_size_log2);
-	if ((num_wrs) <= 0) {
+	if (num_wrs == 0) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
 		return -ENOMEM;
 	}
@@ -558,18 +567,19 @@ int iwch_bind_mw(struct ib_qp *qp,
 	if (mw_bind->send_flags & IB_SEND_SIGNALED)
 		t3_wr_flags = T3_COMPLETION_FLAG;
 
-	sgl.addr = mw_bind->addr;
-	sgl.lkey = mw_bind->mr->lkey;
-	sgl.length = mw_bind->length;
+	sgl.addr = mw_bind->bind_info.addr;
+	sgl.lkey = mw_bind->bind_info.mr->lkey;
+	sgl.length = mw_bind->bind_info.length;
 	wqe->bind.reserved = 0;
 	wqe->bind.type = TPT_VATO;
 
 	/* TBD: check perms */
-	wqe->bind.perms = iwch_ib_to_tpt_access(mw_bind->mw_access_flags);
-	wqe->bind.mr_stag = cpu_to_be32(mw_bind->mr->lkey);
+	wqe->bind.perms = iwch_ib_to_tpt_bind_access(
+		mw_bind->bind_info.mw_access_flags);
+	wqe->bind.mr_stag = cpu_to_be32(mw_bind->bind_info.mr->lkey);
 	wqe->bind.mw_stag = cpu_to_be32(mw->rkey);
-	wqe->bind.mw_len = cpu_to_be32(mw_bind->length);
-	wqe->bind.mw_va = cpu_to_be64(mw_bind->addr);
+	wqe->bind.mw_len = cpu_to_be32(mw_bind->bind_info.length);
+	wqe->bind.mw_va = cpu_to_be64(mw_bind->bind_info.addr);
 	err = iwch_sgl2pbl_map(rhp, &sgl, 1, &pbl_addr, &page_size);
 	if (err) {
 		spin_unlock_irqrestore(&qhp->lock, flag);
@@ -591,7 +601,8 @@ int iwch_bind_mw(struct ib_qp *qp,
 	++(qhp->wq.sq_wptr);
 	spin_unlock_irqrestore(&qhp->lock, flag);
 
-	ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
+	if (cxio_wq_db_enabled(&qhp->wq))
+		ring_doorbell(qhp->wq.doorbell, qhp->wq.qpid);
 
 	return err;
 }
@@ -728,7 +739,7 @@ static inline void build_term_codes(struct respQ_msg_t *rsp_msg,
 	}
 }
 
-int iwch_post_zb_read(struct iwch_qp *qhp)
+int iwch_post_zb_read(struct iwch_ep *ep)
 {
 	union t3_wr *wqe;
 	struct sk_buff *skb;
@@ -745,17 +756,16 @@ int iwch_post_zb_read(struct iwch_qp *qhp)
 	wqe->read.rdmaop = T3_READ_REQ;
 	wqe->read.reserved[0] = 0;
 	wqe->read.reserved[1] = 0;
-	wqe->read.reserved[2] = 0;
 	wqe->read.rem_stag = cpu_to_be32(1);
 	wqe->read.rem_to = cpu_to_be64(1);
 	wqe->read.local_stag = cpu_to_be32(1);
 	wqe->read.local_len = cpu_to_be32(0);
 	wqe->read.local_to = cpu_to_be64(1);
 	wqe->send.wrh.op_seop_flags = cpu_to_be32(V_FW_RIWR_OP(T3_WR_READ));
-	wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(qhp->ep->hwtid)|
+	wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(ep->hwtid)|
 						V_FW_RIWR_LEN(flit_cnt));
 	skb->priority = CPL_PRIORITY_DATA;
-	return cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
+	return iwch_cxgb3_ofld_send(ep->com.qp->rhp->rdev.t3cdev_p, skb);
 }
 
 /*
@@ -787,61 +797,82 @@ int iwch_post_terminate(struct iwch_qp *qhp, struct respQ_msg_t *rsp_msg)
 			 V_FW_RIWR_FLAGS(T3_COMPLETION_FLAG | T3_NOTIFY_FLAG));
 	wqe->send.wrh.gen_tid_len = cpu_to_be32(V_FW_RIWR_TID(qhp->ep->hwtid));
 	skb->priority = CPL_PRIORITY_DATA;
-	return cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
+	return iwch_cxgb3_ofld_send(qhp->rhp->rdev.t3cdev_p, skb);
 }
 
 /*
  * Assumes qhp lock is held.
  */
-static void __flush_qp(struct iwch_qp *qhp, unsigned long *flag)
+static void __flush_qp(struct iwch_qp *qhp, struct iwch_cq *rchp,
+				struct iwch_cq *schp)
 {
-	struct iwch_cq *rchp, *schp;
 	int count;
 	int flushed;
 
-	rchp = get_chp(qhp->rhp, qhp->attr.rcq);
-	schp = get_chp(qhp->rhp, qhp->attr.scq);
 
 	PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
 	/* take a ref on the qhp since we must release the lock */
 	atomic_inc(&qhp->refcnt);
-	spin_unlock_irqrestore(&qhp->lock, *flag);
+	spin_unlock(&qhp->lock);
 
-	/* locking heirarchy: cq lock first, then qp lock. */
-	spin_lock_irqsave(&rchp->lock, *flag);
+	/* locking hierarchy: cq lock first, then qp lock. */
+	spin_lock(&rchp->lock);
 	spin_lock(&qhp->lock);
 	cxio_flush_hw_cq(&rchp->cq);
 	cxio_count_rcqes(&rchp->cq, &qhp->wq, &count);
 	flushed = cxio_flush_rq(&qhp->wq, &rchp->cq, count);
 	spin_unlock(&qhp->lock);
-	spin_unlock_irqrestore(&rchp->lock, *flag);
-	if (flushed)
+	spin_unlock(&rchp->lock);
+	if (flushed) {
+		spin_lock(&rchp->comp_handler_lock);
 		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock(&rchp->comp_handler_lock);
+	}
 
-	/* locking heirarchy: cq lock first, then qp lock. */
-	spin_lock_irqsave(&schp->lock, *flag);
+	/* locking hierarchy: cq lock first, then qp lock. */
+	spin_lock(&schp->lock);
 	spin_lock(&qhp->lock);
 	cxio_flush_hw_cq(&schp->cq);
 	cxio_count_scqes(&schp->cq, &qhp->wq, &count);
 	flushed = cxio_flush_sq(&qhp->wq, &schp->cq, count);
 	spin_unlock(&qhp->lock);
-	spin_unlock_irqrestore(&schp->lock, *flag);
-	if (flushed)
+	spin_unlock(&schp->lock);
+	if (flushed) {
+		spin_lock(&schp->comp_handler_lock);
 		(*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
+		spin_unlock(&schp->comp_handler_lock);
+	}
 
 	/* deref */
 	if (atomic_dec_and_test(&qhp->refcnt))
 	        wake_up(&qhp->wait);
 
-	spin_lock_irqsave(&qhp->lock, *flag);
+	spin_lock(&qhp->lock);
 }
 
-static void flush_qp(struct iwch_qp *qhp, unsigned long *flag)
+static void flush_qp(struct iwch_qp *qhp)
 {
-	if (qhp->ibqp.uobject)
+	struct iwch_cq *rchp, *schp;
+
+	rchp = get_chp(qhp->rhp, qhp->attr.rcq);
+	schp = get_chp(qhp->rhp, qhp->attr.scq);
+
+	if (qhp->ibqp.uobject) {
 		cxio_set_wq_in_error(&qhp->wq);
-	else
-		__flush_qp(qhp, flag);
+		cxio_set_cq_in_error(&rchp->cq);
+		spin_lock(&rchp->comp_handler_lock);
+		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock(&rchp->comp_handler_lock);
+		if (schp != rchp) {
+			cxio_set_cq_in_error(&schp->cq);
+			spin_lock(&schp->comp_handler_lock);
+			(*schp->ibcq.comp_handler)(&schp->ibcq,
+						   schp->ibcq.cq_context);
+			spin_unlock(&schp->comp_handler_lock);
+		}
+		return;
+	}
+	__flush_qp(qhp, rchp, schp);
 }
 
 
@@ -852,7 +883,8 @@ u16 iwch_rqes_posted(struct iwch_qp *qhp)
 {
 	union t3_wr *wqe = qhp->wq.queue;
 	u16 count = 0;
-	while ((count+1) != 0 && fw_riwrh_opcode((struct fw_riwrh *)wqe) == T3_WR_RCV) {
+
+	while (count < USHRT_MAX && fw_riwrh_opcode((struct fw_riwrh *)wqe) == T3_WR_RCV) {
 		count++;
 		wqe++;
 	}
@@ -879,20 +911,13 @@ static int rdma_init(struct iwch_dev *rhp, struct iwch_qp *qhp,
 		(qhp->attr.mpa_attr.xmit_marker_enabled << 1) |
 		(qhp->attr.mpa_attr.crc_enabled << 2);
 
-	/*
-	 * XXX - The IWCM doesn't quite handle getting these
-	 * attrs set before going into RTS.  For now, just turn
-	 * them on always...
-	 */
-#if 0
-	init_attr.qpcaps = qhp->attr.enableRdmaRead |
-		(qhp->attr.enableRdmaWrite << 1) |
-		(qhp->attr.enableBind << 2) |
-		(qhp->attr.enable_stag0_fastreg << 3) |
-		(qhp->attr.enable_stag0_fastreg << 4);
-#else
-	init_attr.qpcaps = 0x1f;
-#endif
+	init_attr.qpcaps = uP_RI_QP_RDMA_READ_ENABLE |
+			   uP_RI_QP_RDMA_WRITE_ENABLE |
+			   uP_RI_QP_BIND_ENABLE;
+	if (!qhp->ibqp.uobject)
+		init_attr.qpcaps |= uP_RI_QP_STAG0_ENABLE |
+				    uP_RI_QP_FAST_REGISTER_ENABLE;
+
 	init_attr.tcp_emss = qhp->ep->emss;
 	init_attr.ord = qhp->attr.max_ord;
 	init_attr.ird = qhp->attr.max_ird;
@@ -900,8 +925,7 @@ static int rdma_init(struct iwch_dev *rhp, struct iwch_qp *qhp,
 	init_attr.qp_dma_size = (1UL << qhp->wq.size_log2);
 	init_attr.rqe_count = iwch_rqes_posted(qhp);
 	init_attr.flags = qhp->attr.mpa_attr.initiator ? MPA_INITIATOR : 0;
-	if (!qhp->ibqp.uobject)
-		init_attr.flags |= PRIV_QP;
+	init_attr.chan = qhp->ep->l2t->smt_idx;
 	if (peer2peer) {
 		init_attr.rtr_type = RTR_READ;
 		if (init_attr.ord == 0 && qhp->attr.mpa_attr.initiator)
@@ -1008,7 +1032,7 @@ int iwch_modify_qp(struct iwch_dev *rhp, struct iwch_qp *qhp,
 			break;
 		case IWCH_QP_STATE_ERROR:
 			qhp->attr.state = IWCH_QP_STATE_ERROR;
-			flush_qp(qhp, &flag);
+			flush_qp(qhp);
 			break;
 		default:
 			ret = -EINVAL;
@@ -1056,7 +1080,7 @@ int iwch_modify_qp(struct iwch_dev *rhp, struct iwch_qp *qhp,
 		}
 		switch (attrs->next_state) {
 			case IWCH_QP_STATE_IDLE:
-				flush_qp(qhp, &flag);
+				flush_qp(qhp);
 				qhp->attr.state = IWCH_QP_STATE_IDLE;
 				qhp->attr.llp_stream_handle = NULL;
 				put_ep(&qhp->ep->com);
@@ -1082,7 +1106,6 @@ int iwch_modify_qp(struct iwch_dev *rhp, struct iwch_qp *qhp,
 			goto out;
 		}
 		qhp->attr.state = IWCH_QP_STATE_IDLE;
-		memset(&qhp->attr, 0, sizeof(qhp->attr));
 		break;
 	case IWCH_QP_STATE_TERMINATE:
 		if (!internal) {
@@ -1111,7 +1134,7 @@ err:
 	free=1;
 	wake_up(&qhp->wait);
 	BUG_ON(!ep);
-	flush_qp(qhp, &flag);
+	flush_qp(qhp);
 out:
 	spin_unlock_irqrestore(&qhp->lock, flag);
 
@@ -1138,59 +1161,3 @@ out:
 	PDBG("%s exit state %d\n", __func__, qhp->attr.state);
 	return ret;
 }
-
-static int quiesce_qp(struct iwch_qp *qhp)
-{
-	spin_lock_irq(&qhp->lock);
-	iwch_quiesce_tid(qhp->ep);
-	qhp->flags |= QP_QUIESCED;
-	spin_unlock_irq(&qhp->lock);
-	return 0;
-}
-
-static int resume_qp(struct iwch_qp *qhp)
-{
-	spin_lock_irq(&qhp->lock);
-	iwch_resume_tid(qhp->ep);
-	qhp->flags &= ~QP_QUIESCED;
-	spin_unlock_irq(&qhp->lock);
-	return 0;
-}
-
-int iwch_quiesce_qps(struct iwch_cq *chp)
-{
-	int i;
-	struct iwch_qp *qhp;
-
-	for (i=0; i < T3_MAX_NUM_QP; i++) {
-		qhp = get_qhp(chp->rhp, i);
-		if (!qhp)
-			continue;
-		if ((qhp->attr.rcq == chp->cq.cqid) && !qp_quiesced(qhp)) {
-			quiesce_qp(qhp);
-			continue;
-		}
-		if ((qhp->attr.scq == chp->cq.cqid) && !qp_quiesced(qhp))
-			quiesce_qp(qhp);
-	}
-	return 0;
-}
-
-int iwch_resume_qps(struct iwch_cq *chp)
-{
-	int i;
-	struct iwch_qp *qhp;
-
-	for (i=0; i < T3_MAX_NUM_QP; i++) {
-		qhp = get_qhp(chp->rhp, i);
-		if (!qhp)
-			continue;
-		if ((qhp->attr.rcq == chp->cq.cqid) && qp_quiesced(qhp)) {
-			resume_qp(qhp);
-			continue;
-		}
-		if ((qhp->attr.scq == chp->cq.cqid) && qp_quiesced(qhp))
-			resume_qp(qhp);
-	}
-	return 0;
-}
diff --git a/drivers/infiniband/hw/cxgb3/iwch_user.h b/drivers/infiniband/hw/cxgb3/iwch_user.h
index cb7086f558c..a277c31fcaf 100644
--- a/drivers/infiniband/hw/cxgb3/iwch_user.h
+++ b/drivers/infiniband/hw/cxgb3/iwch_user.h
@@ -45,10 +45,18 @@ struct iwch_create_cq_req {
 	__u64 user_rptr_addr;
 };
 
+struct iwch_create_cq_resp_v0 {
+	__u64 key;
+	__u32 cqid;
+	__u32 size_log2;
+};
+
 struct iwch_create_cq_resp {
 	__u64 key;
 	__u32 cqid;
 	__u32 size_log2;
+	__u32 memsize;
+	__u32 reserved;
 };
 
 struct iwch_create_qp_resp {
diff --git a/drivers/infiniband/hw/cxgb4/Kconfig b/drivers/infiniband/hw/cxgb4/Kconfig
new file mode 100644
index 00000000000..23f38cf2c5c
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/Kconfig
@@ -0,0 +1,18 @@
+config INFINIBAND_CXGB4
+	tristate "Chelsio T4/T5 RDMA Driver"
+	depends on CHELSIO_T4 && INET && (IPV6 || IPV6=n)
+	select GENERIC_ALLOCATOR
+	---help---
+	  This is an iWARP/RDMA driver for the Chelsio T4 and T5
+	  1GbE, 10GbE adapters and T5 40GbE adapter.
+
+	  For general information about Chelsio and our products, visit
+	  our website at <http://www.chelsio.com>.
+
+	  For customer support, please visit our customer support page at
+	  <http://www.chelsio.com/support.html>.
+
+	  Please send feedback to <linux-bugs@chelsio.com>.
+
+	  To compile this driver as a module, choose M here: the module
+	  will be called iw_cxgb4.
diff --git a/drivers/infiniband/hw/cxgb4/Makefile b/drivers/infiniband/hw/cxgb4/Makefile
new file mode 100644
index 00000000000..e11cf729994
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/Makefile
@@ -0,0 +1,5 @@
+ccflags-y := -Idrivers/net/ethernet/chelsio/cxgb4
+
+obj-$(CONFIG_INFINIBAND_CXGB4) += iw_cxgb4.o
+
+iw_cxgb4-y :=  device.o cm.o provider.o mem.o cq.o qp.o resource.o ev.o id_table.o
diff --git a/drivers/infiniband/hw/cxgb4/cm.c b/drivers/infiniband/hw/cxgb4/cm.c
new file mode 100644
index 00000000000..768a0fb67dd
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/cm.c
@@ -0,0 +1,3933 @@
+/*
+ * Copyright (c) 2009-2014 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <linux/list.h>
+#include <linux/workqueue.h>
+#include <linux/skbuff.h>
+#include <linux/timer.h>
+#include <linux/notifier.h>
+#include <linux/inetdevice.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
+#include <linux/if_vlan.h>
+
+#include <net/neighbour.h>
+#include <net/netevent.h>
+#include <net/route.h>
+#include <net/tcp.h>
+#include <net/ip6_route.h>
+#include <net/addrconf.h>
+
+#include <rdma/ib_addr.h>
+
+#include "iw_cxgb4.h"
+
+static char *states[] = {
+	"idle",
+	"listen",
+	"connecting",
+	"mpa_wait_req",
+	"mpa_req_sent",
+	"mpa_req_rcvd",
+	"mpa_rep_sent",
+	"fpdu_mode",
+	"aborting",
+	"closing",
+	"moribund",
+	"dead",
+	NULL,
+};
+
+static int nocong;
+module_param(nocong, int, 0644);
+MODULE_PARM_DESC(nocong, "Turn of congestion control (default=0)");
+
+static int enable_ecn;
+module_param(enable_ecn, int, 0644);
+MODULE_PARM_DESC(enable_ecn, "Enable ECN (default=0/disabled)");
+
+static int dack_mode = 1;
+module_param(dack_mode, int, 0644);
+MODULE_PARM_DESC(dack_mode, "Delayed ack mode (default=1)");
+
+int c4iw_max_read_depth = 8;
+module_param(c4iw_max_read_depth, int, 0644);
+MODULE_PARM_DESC(c4iw_max_read_depth, "Per-connection max ORD/IRD (default=8)");
+
+static int enable_tcp_timestamps;
+module_param(enable_tcp_timestamps, int, 0644);
+MODULE_PARM_DESC(enable_tcp_timestamps, "Enable tcp timestamps (default=0)");
+
+static int enable_tcp_sack;
+module_param(enable_tcp_sack, int, 0644);
+MODULE_PARM_DESC(enable_tcp_sack, "Enable tcp SACK (default=0)");
+
+static int enable_tcp_window_scaling = 1;
+module_param(enable_tcp_window_scaling, int, 0644);
+MODULE_PARM_DESC(enable_tcp_window_scaling,
+		 "Enable tcp window scaling (default=1)");
+
+int c4iw_debug;
+module_param(c4iw_debug, int, 0644);
+MODULE_PARM_DESC(c4iw_debug, "Enable debug logging (default=0)");
+
+static int peer2peer = 1;
+module_param(peer2peer, int, 0644);
+MODULE_PARM_DESC(peer2peer, "Support peer2peer ULPs (default=1)");
+
+static int p2p_type = FW_RI_INIT_P2PTYPE_READ_REQ;
+module_param(p2p_type, int, 0644);
+MODULE_PARM_DESC(p2p_type, "RDMAP opcode to use for the RTR message: "
+			   "1=RDMA_READ 0=RDMA_WRITE (default 1)");
+
+static int ep_timeout_secs = 60;
+module_param(ep_timeout_secs, int, 0644);
+MODULE_PARM_DESC(ep_timeout_secs, "CM Endpoint operation timeout "
+				   "in seconds (default=60)");
+
+static int mpa_rev = 1;
+module_param(mpa_rev, int, 0644);
+MODULE_PARM_DESC(mpa_rev, "MPA Revision, 0 supports amso1100, "
+		"1 is RFC0544 spec compliant, 2 is IETF MPA Peer Connect Draft"
+		" compliant (default=1)");
+
+static int markers_enabled;
+module_param(markers_enabled, int, 0644);
+MODULE_PARM_DESC(markers_enabled, "Enable MPA MARKERS (default(0)=disabled)");
+
+static int crc_enabled = 1;
+module_param(crc_enabled, int, 0644);
+MODULE_PARM_DESC(crc_enabled, "Enable MPA CRC (default(1)=enabled)");
+
+static int rcv_win = 256 * 1024;
+module_param(rcv_win, int, 0644);
+MODULE_PARM_DESC(rcv_win, "TCP receive window in bytes (default=256KB)");
+
+static int snd_win = 128 * 1024;
+module_param(snd_win, int, 0644);
+MODULE_PARM_DESC(snd_win, "TCP send window in bytes (default=128KB)");
+
+static struct workqueue_struct *workq;
+
+static struct sk_buff_head rxq;
+
+static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp);
+static void ep_timeout(unsigned long arg);
+static void connect_reply_upcall(struct c4iw_ep *ep, int status);
+
+static LIST_HEAD(timeout_list);
+static spinlock_t timeout_lock;
+
+static void deref_qp(struct c4iw_ep *ep)
+{
+	c4iw_qp_rem_ref(&ep->com.qp->ibqp);
+	clear_bit(QP_REFERENCED, &ep->com.flags);
+}
+
+static void ref_qp(struct c4iw_ep *ep)
+{
+	set_bit(QP_REFERENCED, &ep->com.flags);
+	c4iw_qp_add_ref(&ep->com.qp->ibqp);
+}
+
+static void start_ep_timer(struct c4iw_ep *ep)
+{
+	PDBG("%s ep %p\n", __func__, ep);
+	if (timer_pending(&ep->timer)) {
+		pr_err("%s timer already started! ep %p\n",
+		       __func__, ep);
+		return;
+	}
+	clear_bit(TIMEOUT, &ep->com.flags);
+	c4iw_get_ep(&ep->com);
+	ep->timer.expires = jiffies + ep_timeout_secs * HZ;
+	ep->timer.data = (unsigned long)ep;
+	ep->timer.function = ep_timeout;
+	add_timer(&ep->timer);
+}
+
+static int stop_ep_timer(struct c4iw_ep *ep)
+{
+	PDBG("%s ep %p stopping\n", __func__, ep);
+	del_timer_sync(&ep->timer);
+	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
+		c4iw_put_ep(&ep->com);
+		return 0;
+	}
+	return 1;
+}
+
+static int c4iw_l2t_send(struct c4iw_rdev *rdev, struct sk_buff *skb,
+		  struct l2t_entry *l2e)
+{
+	int	error = 0;
+
+	if (c4iw_fatal_error(rdev)) {
+		kfree_skb(skb);
+		PDBG("%s - device in error state - dropping\n", __func__);
+		return -EIO;
+	}
+	error = cxgb4_l2t_send(rdev->lldi.ports[0], skb, l2e);
+	if (error < 0)
+		kfree_skb(skb);
+	return error < 0 ? error : 0;
+}
+
+int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb)
+{
+	int	error = 0;
+
+	if (c4iw_fatal_error(rdev)) {
+		kfree_skb(skb);
+		PDBG("%s - device in error state - dropping\n", __func__);
+		return -EIO;
+	}
+	error = cxgb4_ofld_send(rdev->lldi.ports[0], skb);
+	if (error < 0)
+		kfree_skb(skb);
+	return error < 0 ? error : 0;
+}
+
+static void release_tid(struct c4iw_rdev *rdev, u32 hwtid, struct sk_buff *skb)
+{
+	struct cpl_tid_release *req;
+
+	skb = get_skb(skb, sizeof *req, GFP_KERNEL);
+	if (!skb)
+		return;
+	req = (struct cpl_tid_release *) skb_put(skb, sizeof(*req));
+	INIT_TP_WR(req, hwtid);
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_TID_RELEASE, hwtid));
+	set_wr_txq(skb, CPL_PRIORITY_SETUP, 0);
+	c4iw_ofld_send(rdev, skb);
+	return;
+}
+
+static void set_emss(struct c4iw_ep *ep, u16 opt)
+{
+	ep->emss = ep->com.dev->rdev.lldi.mtus[GET_TCPOPT_MSS(opt)] -
+		   sizeof(struct iphdr) - sizeof(struct tcphdr);
+	ep->mss = ep->emss;
+	if (GET_TCPOPT_TSTAMP(opt))
+		ep->emss -= 12;
+	if (ep->emss < 128)
+		ep->emss = 128;
+	if (ep->emss & 7)
+		PDBG("Warning: misaligned mtu idx %u mss %u emss=%u\n",
+		     GET_TCPOPT_MSS(opt), ep->mss, ep->emss);
+	PDBG("%s mss_idx %u mss %u emss=%u\n", __func__, GET_TCPOPT_MSS(opt),
+	     ep->mss, ep->emss);
+}
+
+static enum c4iw_ep_state state_read(struct c4iw_ep_common *epc)
+{
+	enum c4iw_ep_state state;
+
+	mutex_lock(&epc->mutex);
+	state = epc->state;
+	mutex_unlock(&epc->mutex);
+	return state;
+}
+
+static void __state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
+{
+	epc->state = new;
+}
+
+static void state_set(struct c4iw_ep_common *epc, enum c4iw_ep_state new)
+{
+	mutex_lock(&epc->mutex);
+	PDBG("%s - %s -> %s\n", __func__, states[epc->state], states[new]);
+	__state_set(epc, new);
+	mutex_unlock(&epc->mutex);
+	return;
+}
+
+static void *alloc_ep(int size, gfp_t gfp)
+{
+	struct c4iw_ep_common *epc;
+
+	epc = kzalloc(size, gfp);
+	if (epc) {
+		kref_init(&epc->kref);
+		mutex_init(&epc->mutex);
+		c4iw_init_wr_wait(&epc->wr_wait);
+	}
+	PDBG("%s alloc ep %p\n", __func__, epc);
+	return epc;
+}
+
+void _c4iw_free_ep(struct kref *kref)
+{
+	struct c4iw_ep *ep;
+
+	ep = container_of(kref, struct c4iw_ep, com.kref);
+	PDBG("%s ep %p state %s\n", __func__, ep, states[state_read(&ep->com)]);
+	if (test_bit(QP_REFERENCED, &ep->com.flags))
+		deref_qp(ep);
+	if (test_bit(RELEASE_RESOURCES, &ep->com.flags)) {
+		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
+		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
+		dst_release(ep->dst);
+		cxgb4_l2t_release(ep->l2t);
+	}
+	if (test_bit(RELEASE_MAPINFO, &ep->com.flags)) {
+		print_addr(&ep->com, __func__, "remove_mapinfo/mapping");
+		iwpm_remove_mapinfo(&ep->com.local_addr,
+				    &ep->com.mapped_local_addr);
+		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
+	}
+	kfree(ep);
+}
+
+static void release_ep_resources(struct c4iw_ep *ep)
+{
+	set_bit(RELEASE_RESOURCES, &ep->com.flags);
+	c4iw_put_ep(&ep->com);
+}
+
+static int status2errno(int status)
+{
+	switch (status) {
+	case CPL_ERR_NONE:
+		return 0;
+	case CPL_ERR_CONN_RESET:
+		return -ECONNRESET;
+	case CPL_ERR_ARP_MISS:
+		return -EHOSTUNREACH;
+	case CPL_ERR_CONN_TIMEDOUT:
+		return -ETIMEDOUT;
+	case CPL_ERR_TCAM_FULL:
+		return -ENOMEM;
+	case CPL_ERR_CONN_EXIST:
+		return -EADDRINUSE;
+	default:
+		return -EIO;
+	}
+}
+
+/*
+ * Try and reuse skbs already allocated...
+ */
+static struct sk_buff *get_skb(struct sk_buff *skb, int len, gfp_t gfp)
+{
+	if (skb && !skb_is_nonlinear(skb) && !skb_cloned(skb)) {
+		skb_trim(skb, 0);
+		skb_get(skb);
+		skb_reset_transport_header(skb);
+	} else {
+		skb = alloc_skb(len, gfp);
+	}
+	t4_set_arp_err_handler(skb, NULL, NULL);
+	return skb;
+}
+
+static struct net_device *get_real_dev(struct net_device *egress_dev)
+{
+	return rdma_vlan_dev_real_dev(egress_dev) ? : egress_dev;
+}
+
+static int our_interface(struct c4iw_dev *dev, struct net_device *egress_dev)
+{
+	int i;
+
+	egress_dev = get_real_dev(egress_dev);
+	for (i = 0; i < dev->rdev.lldi.nports; i++)
+		if (dev->rdev.lldi.ports[i] == egress_dev)
+			return 1;
+	return 0;
+}
+
+static struct dst_entry *find_route6(struct c4iw_dev *dev, __u8 *local_ip,
+				     __u8 *peer_ip, __be16 local_port,
+				     __be16 peer_port, u8 tos,
+				     __u32 sin6_scope_id)
+{
+	struct dst_entry *dst = NULL;
+
+	if (IS_ENABLED(CONFIG_IPV6)) {
+		struct flowi6 fl6;
+
+		memset(&fl6, 0, sizeof(fl6));
+		memcpy(&fl6.daddr, peer_ip, 16);
+		memcpy(&fl6.saddr, local_ip, 16);
+		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
+			fl6.flowi6_oif = sin6_scope_id;
+		dst = ip6_route_output(&init_net, NULL, &fl6);
+		if (!dst)
+			goto out;
+		if (!our_interface(dev, ip6_dst_idev(dst)->dev) &&
+		    !(ip6_dst_idev(dst)->dev->flags & IFF_LOOPBACK)) {
+			dst_release(dst);
+			dst = NULL;
+		}
+	}
+
+out:
+	return dst;
+}
+
+static struct dst_entry *find_route(struct c4iw_dev *dev, __be32 local_ip,
+				 __be32 peer_ip, __be16 local_port,
+				 __be16 peer_port, u8 tos)
+{
+	struct rtable *rt;
+	struct flowi4 fl4;
+	struct neighbour *n;
+
+	rt = ip_route_output_ports(&init_net, &fl4, NULL, peer_ip, local_ip,
+				   peer_port, local_port, IPPROTO_TCP,
+				   tos, 0);
+	if (IS_ERR(rt))
+		return NULL;
+	n = dst_neigh_lookup(&rt->dst, &peer_ip);
+	if (!n)
+		return NULL;
+	if (!our_interface(dev, n->dev) &&
+	    !(n->dev->flags & IFF_LOOPBACK)) {
+		dst_release(&rt->dst);
+		return NULL;
+	}
+	neigh_release(n);
+	return &rt->dst;
+}
+
+static void arp_failure_discard(void *handle, struct sk_buff *skb)
+{
+	PDBG("%s c4iw_dev %p\n", __func__, handle);
+	kfree_skb(skb);
+}
+
+/*
+ * Handle an ARP failure for an active open.
+ */
+static void act_open_req_arp_failure(void *handle, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep = handle;
+
+	printk(KERN_ERR MOD "ARP failure duing connect\n");
+	kfree_skb(skb);
+	connect_reply_upcall(ep, -EHOSTUNREACH);
+	state_set(&ep->com, DEAD);
+	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
+	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
+	dst_release(ep->dst);
+	cxgb4_l2t_release(ep->l2t);
+	c4iw_put_ep(&ep->com);
+}
+
+/*
+ * Handle an ARP failure for a CPL_ABORT_REQ.  Change it into a no RST variant
+ * and send it along.
+ */
+static void abort_arp_failure(void *handle, struct sk_buff *skb)
+{
+	struct c4iw_rdev *rdev = handle;
+	struct cpl_abort_req *req = cplhdr(skb);
+
+	PDBG("%s rdev %p\n", __func__, rdev);
+	req->cmd = CPL_ABORT_NO_RST;
+	c4iw_ofld_send(rdev, skb);
+}
+
+static void send_flowc(struct c4iw_ep *ep, struct sk_buff *skb)
+{
+	unsigned int flowclen = 80;
+	struct fw_flowc_wr *flowc;
+	int i;
+
+	skb = get_skb(skb, flowclen, GFP_KERNEL);
+	flowc = (struct fw_flowc_wr *)__skb_put(skb, flowclen);
+
+	flowc->op_to_nparams = cpu_to_be32(FW_WR_OP(FW_FLOWC_WR) |
+					   FW_FLOWC_WR_NPARAMS(8));
+	flowc->flowid_len16 = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(flowclen,
+					  16)) | FW_WR_FLOWID(ep->hwtid));
+
+	flowc->mnemval[0].mnemonic = FW_FLOWC_MNEM_PFNVFN;
+	flowc->mnemval[0].val = cpu_to_be32(PCI_FUNC(ep->com.dev->rdev.lldi.pdev->devfn) << 8);
+	flowc->mnemval[1].mnemonic = FW_FLOWC_MNEM_CH;
+	flowc->mnemval[1].val = cpu_to_be32(ep->tx_chan);
+	flowc->mnemval[2].mnemonic = FW_FLOWC_MNEM_PORT;
+	flowc->mnemval[2].val = cpu_to_be32(ep->tx_chan);
+	flowc->mnemval[3].mnemonic = FW_FLOWC_MNEM_IQID;
+	flowc->mnemval[3].val = cpu_to_be32(ep->rss_qid);
+	flowc->mnemval[4].mnemonic = FW_FLOWC_MNEM_SNDNXT;
+	flowc->mnemval[4].val = cpu_to_be32(ep->snd_seq);
+	flowc->mnemval[5].mnemonic = FW_FLOWC_MNEM_RCVNXT;
+	flowc->mnemval[5].val = cpu_to_be32(ep->rcv_seq);
+	flowc->mnemval[6].mnemonic = FW_FLOWC_MNEM_SNDBUF;
+	flowc->mnemval[6].val = cpu_to_be32(ep->snd_win);
+	flowc->mnemval[7].mnemonic = FW_FLOWC_MNEM_MSS;
+	flowc->mnemval[7].val = cpu_to_be32(ep->emss);
+	/* Pad WR to 16 byte boundary */
+	flowc->mnemval[8].mnemonic = 0;
+	flowc->mnemval[8].val = 0;
+	for (i = 0; i < 9; i++) {
+		flowc->mnemval[i].r4[0] = 0;
+		flowc->mnemval[i].r4[1] = 0;
+		flowc->mnemval[i].r4[2] = 0;
+	}
+
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	c4iw_ofld_send(&ep->com.dev->rdev, skb);
+}
+
+static int send_halfclose(struct c4iw_ep *ep, gfp_t gfp)
+{
+	struct cpl_close_con_req *req;
+	struct sk_buff *skb;
+	int wrlen = roundup(sizeof *req, 16);
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	skb = get_skb(NULL, wrlen, gfp);
+	if (!skb) {
+		printk(KERN_ERR MOD "%s - failed to alloc skb\n", __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
+	req = (struct cpl_close_con_req *) skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	INIT_TP_WR(req, ep->hwtid);
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_CLOSE_CON_REQ,
+						    ep->hwtid));
+	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+static int send_abort(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
+{
+	struct cpl_abort_req *req;
+	int wrlen = roundup(sizeof *req, 16);
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	skb = get_skb(skb, wrlen, gfp);
+	if (!skb) {
+		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
+		       __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	t4_set_arp_err_handler(skb, &ep->com.dev->rdev, abort_arp_failure);
+	req = (struct cpl_abort_req *) skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	INIT_TP_WR(req, ep->hwtid);
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_REQ, ep->hwtid));
+	req->cmd = CPL_ABORT_SEND_RST;
+	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+/*
+ * c4iw_form_pm_msg - Form a port mapper message with mapping info
+ */
+static void c4iw_form_pm_msg(struct c4iw_ep *ep,
+				struct iwpm_sa_data *pm_msg)
+{
+	memcpy(&pm_msg->loc_addr, &ep->com.local_addr,
+		sizeof(ep->com.local_addr));
+	memcpy(&pm_msg->rem_addr, &ep->com.remote_addr,
+		sizeof(ep->com.remote_addr));
+}
+
+/*
+ * c4iw_form_reg_msg - Form a port mapper message with dev info
+ */
+static void c4iw_form_reg_msg(struct c4iw_dev *dev,
+				struct iwpm_dev_data *pm_msg)
+{
+	memcpy(pm_msg->dev_name, dev->ibdev.name, IWPM_DEVNAME_SIZE);
+	memcpy(pm_msg->if_name, dev->rdev.lldi.ports[0]->name,
+				IWPM_IFNAME_SIZE);
+}
+
+static void c4iw_record_pm_msg(struct c4iw_ep *ep,
+			struct iwpm_sa_data *pm_msg)
+{
+	memcpy(&ep->com.mapped_local_addr, &pm_msg->mapped_loc_addr,
+		sizeof(ep->com.mapped_local_addr));
+	memcpy(&ep->com.mapped_remote_addr, &pm_msg->mapped_rem_addr,
+		sizeof(ep->com.mapped_remote_addr));
+}
+
+static void best_mtu(const unsigned short *mtus, unsigned short mtu,
+		     unsigned int *idx, int use_ts)
+{
+	unsigned short hdr_size = sizeof(struct iphdr) +
+				  sizeof(struct tcphdr) +
+				  (use_ts ? 12 : 0);
+	unsigned short data_size = mtu - hdr_size;
+
+	cxgb4_best_aligned_mtu(mtus, hdr_size, data_size, 8, idx);
+}
+
+static int send_connect(struct c4iw_ep *ep)
+{
+	struct cpl_act_open_req *req;
+	struct cpl_t5_act_open_req *t5_req;
+	struct cpl_act_open_req6 *req6;
+	struct cpl_t5_act_open_req6 *t5_req6;
+	struct sk_buff *skb;
+	u64 opt0;
+	u32 opt2;
+	unsigned int mtu_idx;
+	int wscale;
+	int wrlen;
+	int sizev4 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
+				sizeof(struct cpl_act_open_req) :
+				sizeof(struct cpl_t5_act_open_req);
+	int sizev6 = is_t4(ep->com.dev->rdev.lldi.adapter_type) ?
+				sizeof(struct cpl_act_open_req6) :
+				sizeof(struct cpl_t5_act_open_req6);
+	struct sockaddr_in *la = (struct sockaddr_in *)
+				 &ep->com.mapped_local_addr;
+	struct sockaddr_in *ra = (struct sockaddr_in *)
+				 &ep->com.mapped_remote_addr;
+	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)
+				   &ep->com.mapped_local_addr;
+	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)
+				   &ep->com.mapped_remote_addr;
+	int win;
+
+	wrlen = (ep->com.remote_addr.ss_family == AF_INET) ?
+			roundup(sizev4, 16) :
+			roundup(sizev6, 16);
+
+	PDBG("%s ep %p atid %u\n", __func__, ep, ep->atid);
+
+	skb = get_skb(NULL, wrlen, GFP_KERNEL);
+	if (!skb) {
+		printk(KERN_ERR MOD "%s - failed to alloc skb.\n",
+		       __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
+
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps);
+	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
+
+	opt0 = (nocong ? NO_CONG(1) : 0) |
+	       KEEP_ALIVE(1) |
+	       DELACK(1) |
+	       WND_SCALE(wscale) |
+	       MSS_IDX(mtu_idx) |
+	       L2T_IDX(ep->l2t->idx) |
+	       TX_CHAN(ep->tx_chan) |
+	       SMAC_SEL(ep->smac_idx) |
+	       DSCP(ep->tos) |
+	       ULP_MODE(ULP_MODE_TCPDDP) |
+	       RCV_BUFSIZ(win);
+	opt2 = RX_CHANNEL(0) |
+	       CCTRL_ECN(enable_ecn) |
+	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
+	if (enable_tcp_timestamps)
+		opt2 |= TSTAMPS_EN(1);
+	if (enable_tcp_sack)
+		opt2 |= SACK_EN(1);
+	if (wscale && enable_tcp_window_scaling)
+		opt2 |= WND_SCALE_EN(1);
+	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
+		opt2 |= T5_OPT_2_VALID;
+		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
+	}
+	t4_set_arp_err_handler(skb, ep, act_open_req_arp_failure);
+
+	if (is_t4(ep->com.dev->rdev.lldi.adapter_type)) {
+		if (ep->com.remote_addr.ss_family == AF_INET) {
+			req = (struct cpl_act_open_req *) skb_put(skb, wrlen);
+			INIT_TP_WR(req, 0);
+			OPCODE_TID(req) = cpu_to_be32(
+					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
+					((ep->rss_qid << 14) | ep->atid)));
+			req->local_port = la->sin_port;
+			req->peer_port = ra->sin_port;
+			req->local_ip = la->sin_addr.s_addr;
+			req->peer_ip = ra->sin_addr.s_addr;
+			req->opt0 = cpu_to_be64(opt0);
+			req->params = cpu_to_be32(cxgb4_select_ntuple(
+						ep->com.dev->rdev.lldi.ports[0],
+						ep->l2t));
+			req->opt2 = cpu_to_be32(opt2);
+		} else {
+			req6 = (struct cpl_act_open_req6 *)skb_put(skb, wrlen);
+
+			INIT_TP_WR(req6, 0);
+			OPCODE_TID(req6) = cpu_to_be32(
+					   MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
+					   ((ep->rss_qid<<14)|ep->atid)));
+			req6->local_port = la6->sin6_port;
+			req6->peer_port = ra6->sin6_port;
+			req6->local_ip_hi = *((__be64 *)
+						(la6->sin6_addr.s6_addr));
+			req6->local_ip_lo = *((__be64 *)
+						(la6->sin6_addr.s6_addr + 8));
+			req6->peer_ip_hi = *((__be64 *)
+						(ra6->sin6_addr.s6_addr));
+			req6->peer_ip_lo = *((__be64 *)
+						(ra6->sin6_addr.s6_addr + 8));
+			req6->opt0 = cpu_to_be64(opt0);
+			req6->params = cpu_to_be32(cxgb4_select_ntuple(
+						ep->com.dev->rdev.lldi.ports[0],
+						ep->l2t));
+			req6->opt2 = cpu_to_be32(opt2);
+		}
+	} else {
+		u32 isn = (prandom_u32() & ~7UL) - 1;
+
+		opt2 |= T5_OPT_2_VALID;
+		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
+		if (peer2peer)
+			isn += 4;
+
+		if (ep->com.remote_addr.ss_family == AF_INET) {
+			t5_req = (struct cpl_t5_act_open_req *)
+				 skb_put(skb, wrlen);
+			INIT_TP_WR(t5_req, 0);
+			OPCODE_TID(t5_req) = cpu_to_be32(
+					MK_OPCODE_TID(CPL_ACT_OPEN_REQ,
+					((ep->rss_qid << 14) | ep->atid)));
+			t5_req->local_port = la->sin_port;
+			t5_req->peer_port = ra->sin_port;
+			t5_req->local_ip = la->sin_addr.s_addr;
+			t5_req->peer_ip = ra->sin_addr.s_addr;
+			t5_req->opt0 = cpu_to_be64(opt0);
+			t5_req->params = cpu_to_be64(V_FILTER_TUPLE(
+						     cxgb4_select_ntuple(
+					     ep->com.dev->rdev.lldi.ports[0],
+					     ep->l2t)));
+			t5_req->rsvd = cpu_to_be32(isn);
+			PDBG("%s snd_isn %u\n", __func__,
+			     be32_to_cpu(t5_req->rsvd));
+			t5_req->opt2 = cpu_to_be32(opt2);
+		} else {
+			t5_req6 = (struct cpl_t5_act_open_req6 *)
+				  skb_put(skb, wrlen);
+			INIT_TP_WR(t5_req6, 0);
+			OPCODE_TID(t5_req6) = cpu_to_be32(
+					      MK_OPCODE_TID(CPL_ACT_OPEN_REQ6,
+					      ((ep->rss_qid<<14)|ep->atid)));
+			t5_req6->local_port = la6->sin6_port;
+			t5_req6->peer_port = ra6->sin6_port;
+			t5_req6->local_ip_hi = *((__be64 *)
+						(la6->sin6_addr.s6_addr));
+			t5_req6->local_ip_lo = *((__be64 *)
+						(la6->sin6_addr.s6_addr + 8));
+			t5_req6->peer_ip_hi = *((__be64 *)
+						(ra6->sin6_addr.s6_addr));
+			t5_req6->peer_ip_lo = *((__be64 *)
+						(ra6->sin6_addr.s6_addr + 8));
+			t5_req6->opt0 = cpu_to_be64(opt0);
+			t5_req6->params = (__force __be64)cpu_to_be32(
+							cxgb4_select_ntuple(
+						ep->com.dev->rdev.lldi.ports[0],
+						ep->l2t));
+			t5_req6->rsvd = cpu_to_be32(isn);
+			PDBG("%s snd_isn %u\n", __func__,
+			     be32_to_cpu(t5_req6->rsvd));
+			t5_req6->opt2 = cpu_to_be32(opt2);
+		}
+	}
+
+	set_bit(ACT_OPEN_REQ, &ep->com.history);
+	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+static void send_mpa_req(struct c4iw_ep *ep, struct sk_buff *skb,
+		u8 mpa_rev_to_use)
+{
+	int mpalen, wrlen;
+	struct fw_ofld_tx_data_wr *req;
+	struct mpa_message *mpa;
+	struct mpa_v2_conn_params mpa_v2_params;
+
+	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
+
+	BUG_ON(skb_cloned(skb));
+
+	mpalen = sizeof(*mpa) + ep->plen;
+	if (mpa_rev_to_use == 2)
+		mpalen += sizeof(struct mpa_v2_conn_params);
+	wrlen = roundup(mpalen + sizeof *req, 16);
+	skb = get_skb(skb, wrlen, GFP_KERNEL);
+	if (!skb) {
+		connect_reply_upcall(ep, -ENOMEM);
+		return;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+
+	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	req->op_to_immdlen = cpu_to_be32(
+		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
+		FW_WR_COMPL(1) |
+		FW_WR_IMMDLEN(mpalen));
+	req->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(ep->hwtid) |
+		FW_WR_LEN16(wrlen >> 4));
+	req->plen = cpu_to_be32(mpalen);
+	req->tunnel_to_proxy = cpu_to_be32(
+		FW_OFLD_TX_DATA_WR_FLUSH(1) |
+		FW_OFLD_TX_DATA_WR_SHOVE(1));
+
+	mpa = (struct mpa_message *)(req + 1);
+	memcpy(mpa->key, MPA_KEY_REQ, sizeof(mpa->key));
+	mpa->flags = (crc_enabled ? MPA_CRC : 0) |
+		     (markers_enabled ? MPA_MARKERS : 0) |
+		     (mpa_rev_to_use == 2 ? MPA_ENHANCED_RDMA_CONN : 0);
+	mpa->private_data_size = htons(ep->plen);
+	mpa->revision = mpa_rev_to_use;
+	if (mpa_rev_to_use == 1) {
+		ep->tried_with_mpa_v1 = 1;
+		ep->retry_with_mpa_v1 = 0;
+	}
+
+	if (mpa_rev_to_use == 2) {
+		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
+					       sizeof (struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons((u16)ep->ird);
+		mpa_v2_params.ord = htons((u16)ep->ord);
+
+		if (peer2peer) {
+			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
+			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_WRITE_RTR);
+			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_READ_RTR);
+		}
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
+
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params),
+			       ep->mpa_pkt + sizeof(*mpa), ep->plen);
+	} else
+		if (ep->plen)
+			memcpy(mpa->private_data,
+					ep->mpa_pkt + sizeof(*mpa), ep->plen);
+
+	/*
+	 * Reference the mpa skb.  This ensures the data area
+	 * will remain in memory until the hw acks the tx.
+	 * Function fw4_ack() will deref it.
+	 */
+	skb_get(skb);
+	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
+	BUG_ON(ep->mpa_skb);
+	ep->mpa_skb = skb;
+	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+	start_ep_timer(ep);
+	__state_set(&ep->com, MPA_REQ_SENT);
+	ep->mpa_attr.initiator = 1;
+	ep->snd_seq += mpalen;
+	return;
+}
+
+static int send_mpa_reject(struct c4iw_ep *ep, const void *pdata, u8 plen)
+{
+	int mpalen, wrlen;
+	struct fw_ofld_tx_data_wr *req;
+	struct mpa_message *mpa;
+	struct sk_buff *skb;
+	struct mpa_v2_conn_params mpa_v2_params;
+
+	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
+
+	mpalen = sizeof(*mpa) + plen;
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
+		mpalen += sizeof(struct mpa_v2_conn_params);
+	wrlen = roundup(mpalen + sizeof *req, 16);
+
+	skb = get_skb(NULL, wrlen, GFP_KERNEL);
+	if (!skb) {
+		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+
+	req = (struct fw_ofld_tx_data_wr *)skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	req->op_to_immdlen = cpu_to_be32(
+		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
+		FW_WR_COMPL(1) |
+		FW_WR_IMMDLEN(mpalen));
+	req->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(ep->hwtid) |
+		FW_WR_LEN16(wrlen >> 4));
+	req->plen = cpu_to_be32(mpalen);
+	req->tunnel_to_proxy = cpu_to_be32(
+		FW_OFLD_TX_DATA_WR_FLUSH(1) |
+		FW_OFLD_TX_DATA_WR_SHOVE(1));
+
+	mpa = (struct mpa_message *)(req + 1);
+	memset(mpa, 0, sizeof(*mpa));
+	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
+	mpa->flags = MPA_REJECT;
+	mpa->revision = ep->mpa_attr.version;
+	mpa->private_data_size = htons(plen);
+
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
+		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
+					       sizeof (struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons(((u16)ep->ird) |
+					  (peer2peer ? MPA_V2_PEER2PEER_MODEL :
+					   0));
+		mpa_v2_params.ord = htons(((u16)ep->ord) | (peer2peer ?
+					  (p2p_type ==
+					   FW_RI_INIT_P2PTYPE_RDMA_WRITE ?
+					   MPA_V2_RDMA_WRITE_RTR : p2p_type ==
+					   FW_RI_INIT_P2PTYPE_READ_REQ ?
+					   MPA_V2_RDMA_READ_RTR : 0) : 0));
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
+
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params), pdata, plen);
+	} else
+		if (plen)
+			memcpy(mpa->private_data, pdata, plen);
+
+	/*
+	 * Reference the mpa skb again.  This ensures the data area
+	 * will remain in memory until the hw acks the tx.
+	 * Function fw4_ack() will deref it.
+	 */
+	skb_get(skb);
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
+	BUG_ON(ep->mpa_skb);
+	ep->mpa_skb = skb;
+	ep->snd_seq += mpalen;
+	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+static int send_mpa_reply(struct c4iw_ep *ep, const void *pdata, u8 plen)
+{
+	int mpalen, wrlen;
+	struct fw_ofld_tx_data_wr *req;
+	struct mpa_message *mpa;
+	struct sk_buff *skb;
+	struct mpa_v2_conn_params mpa_v2_params;
+
+	PDBG("%s ep %p tid %u pd_len %d\n", __func__, ep, ep->hwtid, ep->plen);
+
+	mpalen = sizeof(*mpa) + plen;
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn)
+		mpalen += sizeof(struct mpa_v2_conn_params);
+	wrlen = roundup(mpalen + sizeof *req, 16);
+
+	skb = get_skb(NULL, wrlen, GFP_KERNEL);
+	if (!skb) {
+		printk(KERN_ERR MOD "%s - cannot alloc skb!\n", __func__);
+		return -ENOMEM;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+
+	req = (struct fw_ofld_tx_data_wr *) skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	req->op_to_immdlen = cpu_to_be32(
+		FW_WR_OP(FW_OFLD_TX_DATA_WR) |
+		FW_WR_COMPL(1) |
+		FW_WR_IMMDLEN(mpalen));
+	req->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(ep->hwtid) |
+		FW_WR_LEN16(wrlen >> 4));
+	req->plen = cpu_to_be32(mpalen);
+	req->tunnel_to_proxy = cpu_to_be32(
+		FW_OFLD_TX_DATA_WR_FLUSH(1) |
+		FW_OFLD_TX_DATA_WR_SHOVE(1));
+
+	mpa = (struct mpa_message *)(req + 1);
+	memset(mpa, 0, sizeof(*mpa));
+	memcpy(mpa->key, MPA_KEY_REP, sizeof(mpa->key));
+	mpa->flags = (ep->mpa_attr.crc_enabled ? MPA_CRC : 0) |
+		     (markers_enabled ? MPA_MARKERS : 0);
+	mpa->revision = ep->mpa_attr.version;
+	mpa->private_data_size = htons(plen);
+
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		mpa->flags |= MPA_ENHANCED_RDMA_CONN;
+		mpa->private_data_size = htons(ntohs(mpa->private_data_size) +
+					       sizeof (struct mpa_v2_conn_params));
+		mpa_v2_params.ird = htons((u16)ep->ird);
+		mpa_v2_params.ord = htons((u16)ep->ord);
+		if (peer2peer && (ep->mpa_attr.p2p_type !=
+					FW_RI_INIT_P2PTYPE_DISABLED)) {
+			mpa_v2_params.ird |= htons(MPA_V2_PEER2PEER_MODEL);
+
+			if (p2p_type == FW_RI_INIT_P2PTYPE_RDMA_WRITE)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_WRITE_RTR);
+			else if (p2p_type == FW_RI_INIT_P2PTYPE_READ_REQ)
+				mpa_v2_params.ord |=
+					htons(MPA_V2_RDMA_READ_RTR);
+		}
+
+		memcpy(mpa->private_data, &mpa_v2_params,
+		       sizeof(struct mpa_v2_conn_params));
+
+		if (ep->plen)
+			memcpy(mpa->private_data +
+			       sizeof(struct mpa_v2_conn_params), pdata, plen);
+	} else
+		if (plen)
+			memcpy(mpa->private_data, pdata, plen);
+
+	/*
+	 * Reference the mpa skb.  This ensures the data area
+	 * will remain in memory until the hw acks the tx.
+	 * Function fw4_ack() will deref it.
+	 */
+	skb_get(skb);
+	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
+	ep->mpa_skb = skb;
+	__state_set(&ep->com, MPA_REP_SENT);
+	ep->snd_seq += mpalen;
+	return c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+static int act_establish(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_act_establish *req = cplhdr(skb);
+	unsigned int tid = GET_TID(req);
+	unsigned int atid = GET_TID_TID(ntohl(req->tos_atid));
+	struct tid_info *t = dev->rdev.lldi.tids;
+
+	ep = lookup_atid(t, atid);
+
+	PDBG("%s ep %p tid %u snd_isn %u rcv_isn %u\n", __func__, ep, tid,
+	     be32_to_cpu(req->snd_isn), be32_to_cpu(req->rcv_isn));
+
+	mutex_lock(&ep->com.mutex);
+	dst_confirm(ep->dst);
+
+	/* setup the hwtid for this connection */
+	ep->hwtid = tid;
+	cxgb4_insert_tid(t, ep, tid);
+	insert_handle(dev, &dev->hwtid_idr, ep, ep->hwtid);
+
+	ep->snd_seq = be32_to_cpu(req->snd_isn);
+	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
+
+	set_emss(ep, ntohs(req->tcp_opt));
+
+	/* dealloc the atid */
+	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
+	cxgb4_free_atid(t, atid);
+	set_bit(ACT_ESTAB, &ep->com.history);
+
+	/* start MPA negotiation */
+	send_flowc(ep, NULL);
+	if (ep->retry_with_mpa_v1)
+		send_mpa_req(ep, skb, 1);
+	else
+		send_mpa_req(ep, skb, mpa_rev);
+	mutex_unlock(&ep->com.mutex);
+	return 0;
+}
+
+static void close_complete_upcall(struct c4iw_ep *ep, int status)
+{
+	struct iw_cm_event event;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_CLOSE;
+	event.status = status;
+	if (ep->com.cm_id) {
+		PDBG("close complete delivered ep %p cm_id %p tid %u\n",
+		     ep, ep->com.cm_id, ep->hwtid);
+		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
+		ep->com.cm_id->rem_ref(ep->com.cm_id);
+		ep->com.cm_id = NULL;
+		set_bit(CLOSE_UPCALL, &ep->com.history);
+	}
+}
+
+static int abort_connection(struct c4iw_ep *ep, struct sk_buff *skb, gfp_t gfp)
+{
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	__state_set(&ep->com, ABORTING);
+	set_bit(ABORT_CONN, &ep->com.history);
+	return send_abort(ep, skb, gfp);
+}
+
+static void peer_close_upcall(struct c4iw_ep *ep)
+{
+	struct iw_cm_event event;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_DISCONNECT;
+	if (ep->com.cm_id) {
+		PDBG("peer close delivered ep %p cm_id %p tid %u\n",
+		     ep, ep->com.cm_id, ep->hwtid);
+		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
+		set_bit(DISCONN_UPCALL, &ep->com.history);
+	}
+}
+
+static void peer_abort_upcall(struct c4iw_ep *ep)
+{
+	struct iw_cm_event event;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_CLOSE;
+	event.status = -ECONNRESET;
+	if (ep->com.cm_id) {
+		PDBG("abort delivered ep %p cm_id %p tid %u\n", ep,
+		     ep->com.cm_id, ep->hwtid);
+		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
+		ep->com.cm_id->rem_ref(ep->com.cm_id);
+		ep->com.cm_id = NULL;
+		set_bit(ABORT_UPCALL, &ep->com.history);
+	}
+}
+
+static void connect_reply_upcall(struct c4iw_ep *ep, int status)
+{
+	struct iw_cm_event event;
+
+	PDBG("%s ep %p tid %u status %d\n", __func__, ep, ep->hwtid, status);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_CONNECT_REPLY;
+	event.status = status;
+	memcpy(&event.local_addr, &ep->com.local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&event.remote_addr, &ep->com.remote_addr,
+	       sizeof(ep->com.remote_addr));
+
+	if ((status == 0) || (status == -ECONNREFUSED)) {
+		if (!ep->tried_with_mpa_v1) {
+			/* this means MPA_v2 is used */
+			event.private_data_len = ep->plen -
+				sizeof(struct mpa_v2_conn_params);
+			event.private_data = ep->mpa_pkt +
+				sizeof(struct mpa_message) +
+				sizeof(struct mpa_v2_conn_params);
+		} else {
+			/* this means MPA_v1 is used */
+			event.private_data_len = ep->plen;
+			event.private_data = ep->mpa_pkt +
+				sizeof(struct mpa_message);
+		}
+	}
+
+	PDBG("%s ep %p tid %u status %d\n", __func__, ep,
+	     ep->hwtid, status);
+	set_bit(CONN_RPL_UPCALL, &ep->com.history);
+	ep->com.cm_id->event_handler(ep->com.cm_id, &event);
+
+	if (status < 0) {
+		ep->com.cm_id->rem_ref(ep->com.cm_id);
+		ep->com.cm_id = NULL;
+	}
+}
+
+static int connect_request_upcall(struct c4iw_ep *ep)
+{
+	struct iw_cm_event event;
+	int ret;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_CONNECT_REQUEST;
+	memcpy(&event.local_addr, &ep->com.local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&event.remote_addr, &ep->com.remote_addr,
+	       sizeof(ep->com.remote_addr));
+	event.provider_data = ep;
+	if (!ep->tried_with_mpa_v1) {
+		/* this means MPA_v2 is used */
+		event.ord = ep->ord;
+		event.ird = ep->ird;
+		event.private_data_len = ep->plen -
+			sizeof(struct mpa_v2_conn_params);
+		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message) +
+			sizeof(struct mpa_v2_conn_params);
+	} else {
+		/* this means MPA_v1 is used. Send max supported */
+		event.ord = c4iw_max_read_depth;
+		event.ird = c4iw_max_read_depth;
+		event.private_data_len = ep->plen;
+		event.private_data = ep->mpa_pkt + sizeof(struct mpa_message);
+	}
+	c4iw_get_ep(&ep->com);
+	ret = ep->parent_ep->com.cm_id->event_handler(ep->parent_ep->com.cm_id,
+						      &event);
+	if (ret)
+		c4iw_put_ep(&ep->com);
+	set_bit(CONNREQ_UPCALL, &ep->com.history);
+	c4iw_put_ep(&ep->parent_ep->com);
+	return ret;
+}
+
+static void established_upcall(struct c4iw_ep *ep)
+{
+	struct iw_cm_event event;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	memset(&event, 0, sizeof(event));
+	event.event = IW_CM_EVENT_ESTABLISHED;
+	event.ird = ep->ird;
+	event.ord = ep->ord;
+	if (ep->com.cm_id) {
+		PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+		ep->com.cm_id->event_handler(ep->com.cm_id, &event);
+		set_bit(ESTAB_UPCALL, &ep->com.history);
+	}
+}
+
+static int update_rx_credits(struct c4iw_ep *ep, u32 credits)
+{
+	struct cpl_rx_data_ack *req;
+	struct sk_buff *skb;
+	int wrlen = roundup(sizeof *req, 16);
+
+	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
+	skb = get_skb(NULL, wrlen, GFP_KERNEL);
+	if (!skb) {
+		printk(KERN_ERR MOD "update_rx_credits - cannot alloc skb!\n");
+		return 0;
+	}
+
+	/*
+	 * If we couldn't specify the entire rcv window at connection setup
+	 * due to the limit in the number of bits in the RCV_BUFSIZ field,
+	 * then add the overage in to the credits returned.
+	 */
+	if (ep->rcv_win > RCV_BUFSIZ_MASK * 1024)
+		credits += ep->rcv_win - RCV_BUFSIZ_MASK * 1024;
+
+	req = (struct cpl_rx_data_ack *) skb_put(skb, wrlen);
+	memset(req, 0, wrlen);
+	INIT_TP_WR(req, ep->hwtid);
+	OPCODE_TID(req) = cpu_to_be32(MK_OPCODE_TID(CPL_RX_DATA_ACK,
+						    ep->hwtid));
+	req->credit_dack = cpu_to_be32(credits | RX_FORCE_ACK(1) |
+				       F_RX_DACK_CHANGE |
+				       V_RX_DACK_MODE(dack_mode));
+	set_wr_txq(skb, CPL_PRIORITY_ACK, ep->ctrlq_idx);
+	c4iw_ofld_send(&ep->com.dev->rdev, skb);
+	return credits;
+}
+
+static int process_mpa_reply(struct c4iw_ep *ep, struct sk_buff *skb)
+{
+	struct mpa_message *mpa;
+	struct mpa_v2_conn_params *mpa_v2_params;
+	u16 plen;
+	u16 resp_ird, resp_ord;
+	u8 rtr_mismatch = 0, insuff_ird = 0;
+	struct c4iw_qp_attributes attrs;
+	enum c4iw_qp_attr_mask mask;
+	int err;
+	int disconnect = 0;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+
+	/*
+	 * Stop mpa timer.  If it expired, then
+	 * we ignore the MPA reply.  process_timeout()
+	 * will abort the connection.
+	 */
+	if (stop_ep_timer(ep))
+		return 0;
+
+	/*
+	 * If we get more than the supported amount of private data
+	 * then we must fail this connection.
+	 */
+	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
+		err = -EINVAL;
+		goto err;
+	}
+
+	/*
+	 * copy the new data into our accumulation buffer.
+	 */
+	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
+				  skb->len);
+	ep->mpa_pkt_len += skb->len;
+
+	/*
+	 * if we don't even have the mpa message, then bail.
+	 */
+	if (ep->mpa_pkt_len < sizeof(*mpa))
+		return 0;
+	mpa = (struct mpa_message *) ep->mpa_pkt;
+
+	/* Validate MPA header. */
+	if (mpa->revision > mpa_rev) {
+		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
+		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
+		err = -EPROTO;
+		goto err;
+	}
+	if (memcmp(mpa->key, MPA_KEY_REP, sizeof(mpa->key))) {
+		err = -EPROTO;
+		goto err;
+	}
+
+	plen = ntohs(mpa->private_data_size);
+
+	/*
+	 * Fail if there's too much private data.
+	 */
+	if (plen > MPA_MAX_PRIVATE_DATA) {
+		err = -EPROTO;
+		goto err;
+	}
+
+	/*
+	 * If plen does not account for pkt size
+	 */
+	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
+		err = -EPROTO;
+		goto err;
+	}
+
+	ep->plen = (u8) plen;
+
+	/*
+	 * If we don't have all the pdata yet, then bail.
+	 * We'll continue process when more data arrives.
+	 */
+	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
+		return 0;
+
+	if (mpa->flags & MPA_REJECT) {
+		err = -ECONNREFUSED;
+		goto err;
+	}
+
+	/*
+	 * If we get here we have accumulated the entire mpa
+	 * start reply message including private data. And
+	 * the MPA header is valid.
+	 */
+	__state_set(&ep->com, FPDU_MODE);
+	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
+	ep->mpa_attr.recv_marker_enabled = markers_enabled;
+	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
+	ep->mpa_attr.version = mpa->revision;
+	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+
+	if (mpa->revision == 2) {
+		ep->mpa_attr.enhanced_rdma_conn =
+			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
+		if (ep->mpa_attr.enhanced_rdma_conn) {
+			mpa_v2_params = (struct mpa_v2_conn_params *)
+				(ep->mpa_pkt + sizeof(*mpa));
+			resp_ird = ntohs(mpa_v2_params->ird) &
+				MPA_V2_IRD_ORD_MASK;
+			resp_ord = ntohs(mpa_v2_params->ord) &
+				MPA_V2_IRD_ORD_MASK;
+
+			/*
+			 * This is a double-check. Ideally, below checks are
+			 * not required since ird/ord stuff has been taken
+			 * care of in c4iw_accept_cr
+			 */
+			if ((ep->ird < resp_ord) || (ep->ord > resp_ird)) {
+				err = -ENOMEM;
+				ep->ird = resp_ord;
+				ep->ord = resp_ird;
+				insuff_ird = 1;
+			}
+
+			if (ntohs(mpa_v2_params->ird) &
+					MPA_V2_PEER2PEER_MODEL) {
+				if (ntohs(mpa_v2_params->ord) &
+						MPA_V2_RDMA_WRITE_RTR)
+					ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
+				else if (ntohs(mpa_v2_params->ord) &
+						MPA_V2_RDMA_READ_RTR)
+					ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_READ_REQ;
+			}
+		}
+	} else if (mpa->revision == 1)
+		if (peer2peer)
+			ep->mpa_attr.p2p_type = p2p_type;
+
+	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
+	     "xmit_marker_enabled=%d, version=%d p2p_type=%d local-p2p_type = "
+	     "%d\n", __func__, ep->mpa_attr.crc_enabled,
+	     ep->mpa_attr.recv_marker_enabled,
+	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
+	     ep->mpa_attr.p2p_type, p2p_type);
+
+	/*
+	 * If responder's RTR does not match with that of initiator, assign
+	 * FW_RI_INIT_P2PTYPE_DISABLED in mpa attributes so that RTR is not
+	 * generated when moving QP to RTS state.
+	 * A TERM message will be sent after QP has moved to RTS state
+	 */
+	if ((ep->mpa_attr.version == 2) && peer2peer &&
+			(ep->mpa_attr.p2p_type != p2p_type)) {
+		ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+		rtr_mismatch = 1;
+	}
+
+	attrs.mpa_attr = ep->mpa_attr;
+	attrs.max_ird = ep->ird;
+	attrs.max_ord = ep->ord;
+	attrs.llp_stream_handle = ep;
+	attrs.next_state = C4IW_QP_STATE_RTS;
+
+	mask = C4IW_QP_ATTR_NEXT_STATE |
+	    C4IW_QP_ATTR_LLP_STREAM_HANDLE | C4IW_QP_ATTR_MPA_ATTR |
+	    C4IW_QP_ATTR_MAX_IRD | C4IW_QP_ATTR_MAX_ORD;
+
+	/* bind QP and TID with INIT_WR */
+	err = c4iw_modify_qp(ep->com.qp->rhp,
+			     ep->com.qp, mask, &attrs, 1);
+	if (err)
+		goto err;
+
+	/*
+	 * If responder's RTR requirement did not match with what initiator
+	 * supports, generate TERM message
+	 */
+	if (rtr_mismatch) {
+		printk(KERN_ERR "%s: RTR mismatch, sending TERM\n", __func__);
+		attrs.layer_etype = LAYER_MPA | DDP_LLP;
+		attrs.ecode = MPA_NOMATCH_RTR;
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		attrs.send_term = 1;
+		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		err = -ENOMEM;
+		disconnect = 1;
+		goto out;
+	}
+
+	/*
+	 * Generate TERM if initiator IRD is not sufficient for responder
+	 * provided ORD. Currently, we do the same behaviour even when
+	 * responder provided IRD is also not sufficient as regards to
+	 * initiator ORD.
+	 */
+	if (insuff_ird) {
+		printk(KERN_ERR "%s: Insufficient IRD, sending TERM\n",
+				__func__);
+		attrs.layer_etype = LAYER_MPA | DDP_LLP;
+		attrs.ecode = MPA_INSUFF_IRD;
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		attrs.send_term = 1;
+		err = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		err = -ENOMEM;
+		disconnect = 1;
+		goto out;
+	}
+	goto out;
+err:
+	__state_set(&ep->com, ABORTING);
+	send_abort(ep, skb, GFP_KERNEL);
+out:
+	connect_reply_upcall(ep, err);
+	return disconnect;
+}
+
+static void process_mpa_request(struct c4iw_ep *ep, struct sk_buff *skb)
+{
+	struct mpa_message *mpa;
+	struct mpa_v2_conn_params *mpa_v2_params;
+	u16 plen;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+
+	/*
+	 * If we get more than the supported amount of private data
+	 * then we must fail this connection.
+	 */
+	if (ep->mpa_pkt_len + skb->len > sizeof(ep->mpa_pkt)) {
+		(void)stop_ep_timer(ep);
+		abort_connection(ep, skb, GFP_KERNEL);
+		return;
+	}
+
+	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
+
+	/*
+	 * Copy the new data into our accumulation buffer.
+	 */
+	skb_copy_from_linear_data(skb, &(ep->mpa_pkt[ep->mpa_pkt_len]),
+				  skb->len);
+	ep->mpa_pkt_len += skb->len;
+
+	/*
+	 * If we don't even have the mpa message, then bail.
+	 * We'll continue process when more data arrives.
+	 */
+	if (ep->mpa_pkt_len < sizeof(*mpa))
+		return;
+
+	PDBG("%s enter (%s line %u)\n", __func__, __FILE__, __LINE__);
+	mpa = (struct mpa_message *) ep->mpa_pkt;
+
+	/*
+	 * Validate MPA Header.
+	 */
+	if (mpa->revision > mpa_rev) {
+		printk(KERN_ERR MOD "%s MPA version mismatch. Local = %d,"
+		       " Received = %d\n", __func__, mpa_rev, mpa->revision);
+		(void)stop_ep_timer(ep);
+		abort_connection(ep, skb, GFP_KERNEL);
+		return;
+	}
+
+	if (memcmp(mpa->key, MPA_KEY_REQ, sizeof(mpa->key))) {
+		(void)stop_ep_timer(ep);
+		abort_connection(ep, skb, GFP_KERNEL);
+		return;
+	}
+
+	plen = ntohs(mpa->private_data_size);
+
+	/*
+	 * Fail if there's too much private data.
+	 */
+	if (plen > MPA_MAX_PRIVATE_DATA) {
+		(void)stop_ep_timer(ep);
+		abort_connection(ep, skb, GFP_KERNEL);
+		return;
+	}
+
+	/*
+	 * If plen does not account for pkt size
+	 */
+	if (ep->mpa_pkt_len > (sizeof(*mpa) + plen)) {
+		(void)stop_ep_timer(ep);
+		abort_connection(ep, skb, GFP_KERNEL);
+		return;
+	}
+	ep->plen = (u8) plen;
+
+	/*
+	 * If we don't have all the pdata yet, then bail.
+	 */
+	if (ep->mpa_pkt_len < (sizeof(*mpa) + plen))
+		return;
+
+	/*
+	 * If we get here we have accumulated the entire mpa
+	 * start reply message including private data.
+	 */
+	ep->mpa_attr.initiator = 0;
+	ep->mpa_attr.crc_enabled = (mpa->flags & MPA_CRC) | crc_enabled ? 1 : 0;
+	ep->mpa_attr.recv_marker_enabled = markers_enabled;
+	ep->mpa_attr.xmit_marker_enabled = mpa->flags & MPA_MARKERS ? 1 : 0;
+	ep->mpa_attr.version = mpa->revision;
+	if (mpa->revision == 1)
+		ep->tried_with_mpa_v1 = 1;
+	ep->mpa_attr.p2p_type = FW_RI_INIT_P2PTYPE_DISABLED;
+
+	if (mpa->revision == 2) {
+		ep->mpa_attr.enhanced_rdma_conn =
+			mpa->flags & MPA_ENHANCED_RDMA_CONN ? 1 : 0;
+		if (ep->mpa_attr.enhanced_rdma_conn) {
+			mpa_v2_params = (struct mpa_v2_conn_params *)
+				(ep->mpa_pkt + sizeof(*mpa));
+			ep->ird = ntohs(mpa_v2_params->ird) &
+				MPA_V2_IRD_ORD_MASK;
+			ep->ord = ntohs(mpa_v2_params->ord) &
+				MPA_V2_IRD_ORD_MASK;
+			if (ntohs(mpa_v2_params->ird) & MPA_V2_PEER2PEER_MODEL)
+				if (peer2peer) {
+					if (ntohs(mpa_v2_params->ord) &
+							MPA_V2_RDMA_WRITE_RTR)
+						ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_RDMA_WRITE;
+					else if (ntohs(mpa_v2_params->ord) &
+							MPA_V2_RDMA_READ_RTR)
+						ep->mpa_attr.p2p_type =
+						FW_RI_INIT_P2PTYPE_READ_REQ;
+				}
+		}
+	} else if (mpa->revision == 1)
+		if (peer2peer)
+			ep->mpa_attr.p2p_type = p2p_type;
+
+	PDBG("%s - crc_enabled=%d, recv_marker_enabled=%d, "
+	     "xmit_marker_enabled=%d, version=%d p2p_type=%d\n", __func__,
+	     ep->mpa_attr.crc_enabled, ep->mpa_attr.recv_marker_enabled,
+	     ep->mpa_attr.xmit_marker_enabled, ep->mpa_attr.version,
+	     ep->mpa_attr.p2p_type);
+
+	/*
+	 * If the endpoint timer already expired, then we ignore
+	 * the start request.  process_timeout() will abort
+	 * the connection.
+	 */
+	if (!stop_ep_timer(ep)) {
+		__state_set(&ep->com, MPA_REQ_RCVD);
+
+		/* drive upcall */
+		mutex_lock(&ep->parent_ep->com.mutex);
+		if (ep->parent_ep->com.state != DEAD) {
+			if (connect_request_upcall(ep))
+				abort_connection(ep, skb, GFP_KERNEL);
+		} else {
+			abort_connection(ep, skb, GFP_KERNEL);
+		}
+		mutex_unlock(&ep->parent_ep->com.mutex);
+	}
+	return;
+}
+
+static int rx_data(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_rx_data *hdr = cplhdr(skb);
+	unsigned int dlen = ntohs(hdr->len);
+	unsigned int tid = GET_TID(hdr);
+	struct tid_info *t = dev->rdev.lldi.tids;
+	__u8 status = hdr->status;
+	int disconnect = 0;
+
+	ep = lookup_tid(t, tid);
+	if (!ep)
+		return 0;
+	PDBG("%s ep %p tid %u dlen %u\n", __func__, ep, ep->hwtid, dlen);
+	skb_pull(skb, sizeof(*hdr));
+	skb_trim(skb, dlen);
+	mutex_lock(&ep->com.mutex);
+
+	/* update RX credits */
+	update_rx_credits(ep, dlen);
+
+	switch (ep->com.state) {
+	case MPA_REQ_SENT:
+		ep->rcv_seq += dlen;
+		disconnect = process_mpa_reply(ep, skb);
+		break;
+	case MPA_REQ_WAIT:
+		ep->rcv_seq += dlen;
+		process_mpa_request(ep, skb);
+		break;
+	case FPDU_MODE: {
+		struct c4iw_qp_attributes attrs;
+		BUG_ON(!ep->com.qp);
+		if (status)
+			pr_err("%s Unexpected streaming data." \
+			       " qpid %u ep %p state %d tid %u status %d\n",
+			       __func__, ep->com.qp->wq.sq.qid, ep,
+			       ep->com.state, ep->hwtid, status);
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		disconnect = 1;
+		break;
+	}
+	default:
+		break;
+	}
+	mutex_unlock(&ep->com.mutex);
+	if (disconnect)
+		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
+	return 0;
+}
+
+static int abort_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_abort_rpl_rss *rpl = cplhdr(skb);
+	int release = 0;
+	unsigned int tid = GET_TID(rpl);
+	struct tid_info *t = dev->rdev.lldi.tids;
+
+	ep = lookup_tid(t, tid);
+	if (!ep) {
+		printk(KERN_WARNING MOD "Abort rpl to freed endpoint\n");
+		return 0;
+	}
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	mutex_lock(&ep->com.mutex);
+	switch (ep->com.state) {
+	case ABORTING:
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+		__state_set(&ep->com, DEAD);
+		release = 1;
+		break;
+	default:
+		printk(KERN_ERR "%s ep %p state %d\n",
+		     __func__, ep, ep->com.state);
+		break;
+	}
+	mutex_unlock(&ep->com.mutex);
+
+	if (release)
+		release_ep_resources(ep);
+	return 0;
+}
+
+static void send_fw_act_open_req(struct c4iw_ep *ep, unsigned int atid)
+{
+	struct sk_buff *skb;
+	struct fw_ofld_connection_wr *req;
+	unsigned int mtu_idx;
+	int wscale;
+	struct sockaddr_in *sin;
+	int win;
+
+	skb = get_skb(NULL, sizeof(*req), GFP_KERNEL);
+	req = (struct fw_ofld_connection_wr *)__skb_put(skb, sizeof(*req));
+	memset(req, 0, sizeof(*req));
+	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR));
+	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
+	req->le.filter = cpu_to_be32(cxgb4_select_ntuple(
+				     ep->com.dev->rdev.lldi.ports[0],
+				     ep->l2t));
+	sin = (struct sockaddr_in *)&ep->com.mapped_local_addr;
+	req->le.lport = sin->sin_port;
+	req->le.u.ipv4.lip = sin->sin_addr.s_addr;
+	sin = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
+	req->le.pport = sin->sin_port;
+	req->le.u.ipv4.pip = sin->sin_addr.s_addr;
+	req->tcb.t_state_to_astid =
+			htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_SENT) |
+			V_FW_OFLD_CONNECTION_WR_ASTID(atid));
+	req->tcb.cplrxdataack_cplpassacceptrpl =
+			htons(F_FW_OFLD_CONNECTION_WR_CPLRXDATAACK);
+	req->tcb.tx_max = (__force __be32) jiffies;
+	req->tcb.rcv_adv = htons(1);
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps);
+	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
+
+	req->tcb.opt0 = (__force __be64) (TCAM_BYPASS(1) |
+		(nocong ? NO_CONG(1) : 0) |
+		KEEP_ALIVE(1) |
+		DELACK(1) |
+		WND_SCALE(wscale) |
+		MSS_IDX(mtu_idx) |
+		L2T_IDX(ep->l2t->idx) |
+		TX_CHAN(ep->tx_chan) |
+		SMAC_SEL(ep->smac_idx) |
+		DSCP(ep->tos) |
+		ULP_MODE(ULP_MODE_TCPDDP) |
+		RCV_BUFSIZ(win));
+	req->tcb.opt2 = (__force __be32) (PACE(1) |
+		TX_QUEUE(ep->com.dev->rdev.lldi.tx_modq[ep->tx_chan]) |
+		RX_CHANNEL(0) |
+		CCTRL_ECN(enable_ecn) |
+		RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid));
+	if (enable_tcp_timestamps)
+		req->tcb.opt2 |= (__force __be32) TSTAMPS_EN(1);
+	if (enable_tcp_sack)
+		req->tcb.opt2 |= (__force __be32) SACK_EN(1);
+	if (wscale && enable_tcp_window_scaling)
+		req->tcb.opt2 |= (__force __be32) WND_SCALE_EN(1);
+	req->tcb.opt0 = cpu_to_be64((__force u64) req->tcb.opt0);
+	req->tcb.opt2 = cpu_to_be32((__force u32) req->tcb.opt2);
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, ep->ctrlq_idx);
+	set_bit(ACT_OFLD_CONN, &ep->com.history);
+	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+}
+
+/*
+ * Return whether a failed active open has allocated a TID
+ */
+static inline int act_open_has_tid(int status)
+{
+	return status != CPL_ERR_TCAM_FULL && status != CPL_ERR_CONN_EXIST &&
+	       status != CPL_ERR_ARP_MISS;
+}
+
+/* Returns whether a CPL status conveys negative advice.
+ */
+static int is_neg_adv(unsigned int status)
+{
+	return status == CPL_ERR_RTX_NEG_ADVICE ||
+	       status == CPL_ERR_PERSIST_NEG_ADVICE ||
+	       status == CPL_ERR_KEEPALV_NEG_ADVICE;
+}
+
+static void set_tcp_window(struct c4iw_ep *ep, struct port_info *pi)
+{
+	ep->snd_win = snd_win;
+	ep->rcv_win = rcv_win;
+	PDBG("%s snd_win %d rcv_win %d\n", __func__, ep->snd_win, ep->rcv_win);
+}
+
+#define ACT_OPEN_RETRY_COUNT 2
+
+static int import_ep(struct c4iw_ep *ep, int iptype, __u8 *peer_ip,
+		     struct dst_entry *dst, struct c4iw_dev *cdev,
+		     bool clear_mpa_v1)
+{
+	struct neighbour *n;
+	int err, step;
+	struct net_device *pdev;
+
+	n = dst_neigh_lookup(dst, peer_ip);
+	if (!n)
+		return -ENODEV;
+
+	rcu_read_lock();
+	err = -ENOMEM;
+	if (n->dev->flags & IFF_LOOPBACK) {
+		if (iptype == 4)
+			pdev = ip_dev_find(&init_net, *(__be32 *)peer_ip);
+		else if (IS_ENABLED(CONFIG_IPV6))
+			for_each_netdev(&init_net, pdev) {
+				if (ipv6_chk_addr(&init_net,
+						  (struct in6_addr *)peer_ip,
+						  pdev, 1))
+					break;
+			}
+		else
+			pdev = NULL;
+
+		if (!pdev) {
+			err = -ENODEV;
+			goto out;
+		}
+		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
+					n, pdev, 0);
+		if (!ep->l2t)
+			goto out;
+		ep->mtu = pdev->mtu;
+		ep->tx_chan = cxgb4_port_chan(pdev);
+		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
+		step = cdev->rdev.lldi.ntxq /
+			cdev->rdev.lldi.nchan;
+		ep->txq_idx = cxgb4_port_idx(pdev) * step;
+		step = cdev->rdev.lldi.nrxq /
+			cdev->rdev.lldi.nchan;
+		ep->ctrlq_idx = cxgb4_port_idx(pdev);
+		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
+			cxgb4_port_idx(pdev) * step];
+		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
+		dev_put(pdev);
+	} else {
+		pdev = get_real_dev(n->dev);
+		ep->l2t = cxgb4_l2t_get(cdev->rdev.lldi.l2t,
+					n, pdev, 0);
+		if (!ep->l2t)
+			goto out;
+		ep->mtu = dst_mtu(dst);
+		ep->tx_chan = cxgb4_port_chan(pdev);
+		ep->smac_idx = (cxgb4_port_viid(pdev) & 0x7F) << 1;
+		step = cdev->rdev.lldi.ntxq /
+			cdev->rdev.lldi.nchan;
+		ep->txq_idx = cxgb4_port_idx(pdev) * step;
+		ep->ctrlq_idx = cxgb4_port_idx(pdev);
+		step = cdev->rdev.lldi.nrxq /
+			cdev->rdev.lldi.nchan;
+		ep->rss_qid = cdev->rdev.lldi.rxq_ids[
+			cxgb4_port_idx(pdev) * step];
+		set_tcp_window(ep, (struct port_info *)netdev_priv(pdev));
+
+		if (clear_mpa_v1) {
+			ep->retry_with_mpa_v1 = 0;
+			ep->tried_with_mpa_v1 = 0;
+		}
+	}
+	err = 0;
+out:
+	rcu_read_unlock();
+
+	neigh_release(n);
+
+	return err;
+}
+
+static int c4iw_reconnect(struct c4iw_ep *ep)
+{
+	int err = 0;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)
+				    &ep->com.cm_id->local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)
+				    &ep->com.cm_id->remote_addr;
+	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)
+				      &ep->com.cm_id->local_addr;
+	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)
+				      &ep->com.cm_id->remote_addr;
+	int iptype;
+	__u8 *ra;
+
+	PDBG("%s qp %p cm_id %p\n", __func__, ep->com.qp, ep->com.cm_id);
+	init_timer(&ep->timer);
+
+	/*
+	 * Allocate an active TID to initiate a TCP connection.
+	 */
+	ep->atid = cxgb4_alloc_atid(ep->com.dev->rdev.lldi.tids, ep);
+	if (ep->atid == -1) {
+		pr_err("%s - cannot alloc atid.\n", __func__);
+		err = -ENOMEM;
+		goto fail2;
+	}
+	insert_handle(ep->com.dev, &ep->com.dev->atid_idr, ep, ep->atid);
+
+	/* find a route */
+	if (ep->com.cm_id->local_addr.ss_family == AF_INET) {
+		ep->dst = find_route(ep->com.dev, laddr->sin_addr.s_addr,
+				     raddr->sin_addr.s_addr, laddr->sin_port,
+				     raddr->sin_port, 0);
+		iptype = 4;
+		ra = (__u8 *)&raddr->sin_addr;
+	} else {
+		ep->dst = find_route6(ep->com.dev, laddr6->sin6_addr.s6_addr,
+				      raddr6->sin6_addr.s6_addr,
+				      laddr6->sin6_port, raddr6->sin6_port, 0,
+				      raddr6->sin6_scope_id);
+		iptype = 6;
+		ra = (__u8 *)&raddr6->sin6_addr;
+	}
+	if (!ep->dst) {
+		pr_err("%s - cannot find route.\n", __func__);
+		err = -EHOSTUNREACH;
+		goto fail3;
+	}
+	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, false);
+	if (err) {
+		pr_err("%s - cannot alloc l2e.\n", __func__);
+		goto fail4;
+	}
+
+	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
+	     __func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
+	     ep->l2t->idx);
+
+	state_set(&ep->com, CONNECTING);
+	ep->tos = 0;
+
+	/* send connect request to rnic */
+	err = send_connect(ep);
+	if (!err)
+		goto out;
+
+	cxgb4_l2t_release(ep->l2t);
+fail4:
+	dst_release(ep->dst);
+fail3:
+	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
+	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
+fail2:
+	/*
+	 * remember to send notification to upper layer.
+	 * We are in here so the upper layer is not aware that this is
+	 * re-connect attempt and so, upper layer is still waiting for
+	 * response of 1st connect request.
+	 */
+	connect_reply_upcall(ep, -ECONNRESET);
+	c4iw_put_ep(&ep->com);
+out:
+	return err;
+}
+
+static int act_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_act_open_rpl *rpl = cplhdr(skb);
+	unsigned int atid = GET_TID_TID(GET_AOPEN_ATID(
+					ntohl(rpl->atid_status)));
+	struct tid_info *t = dev->rdev.lldi.tids;
+	int status = GET_AOPEN_STATUS(ntohl(rpl->atid_status));
+	struct sockaddr_in *la;
+	struct sockaddr_in *ra;
+	struct sockaddr_in6 *la6;
+	struct sockaddr_in6 *ra6;
+
+	ep = lookup_atid(t, atid);
+	la = (struct sockaddr_in *)&ep->com.mapped_local_addr;
+	ra = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
+	la6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
+	ra6 = (struct sockaddr_in6 *)&ep->com.mapped_remote_addr;
+
+	PDBG("%s ep %p atid %u status %u errno %d\n", __func__, ep, atid,
+	     status, status2errno(status));
+
+	if (is_neg_adv(status)) {
+		printk(KERN_WARNING MOD "Connection problems for atid %u\n",
+			atid);
+		return 0;
+	}
+
+	set_bit(ACT_OPEN_RPL, &ep->com.history);
+
+	/*
+	 * Log interesting failures.
+	 */
+	switch (status) {
+	case CPL_ERR_CONN_RESET:
+	case CPL_ERR_CONN_TIMEDOUT:
+		break;
+	case CPL_ERR_TCAM_FULL:
+		mutex_lock(&dev->rdev.stats.lock);
+		dev->rdev.stats.tcam_full++;
+		mutex_unlock(&dev->rdev.stats.lock);
+		if (ep->com.local_addr.ss_family == AF_INET &&
+		    dev->rdev.lldi.enable_fw_ofld_conn) {
+			send_fw_act_open_req(ep,
+					     GET_TID_TID(GET_AOPEN_ATID(
+					     ntohl(rpl->atid_status))));
+			return 0;
+		}
+		break;
+	case CPL_ERR_CONN_EXIST:
+		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
+			set_bit(ACT_RETRY_INUSE, &ep->com.history);
+			remove_handle(ep->com.dev, &ep->com.dev->atid_idr,
+					atid);
+			cxgb4_free_atid(t, atid);
+			dst_release(ep->dst);
+			cxgb4_l2t_release(ep->l2t);
+			c4iw_reconnect(ep);
+			return 0;
+		}
+		break;
+	default:
+		if (ep->com.local_addr.ss_family == AF_INET) {
+			pr_info("Active open failure - atid %u status %u errno %d %pI4:%u->%pI4:%u\n",
+				atid, status, status2errno(status),
+				&la->sin_addr.s_addr, ntohs(la->sin_port),
+				&ra->sin_addr.s_addr, ntohs(ra->sin_port));
+		} else {
+			pr_info("Active open failure - atid %u status %u errno %d %pI6:%u->%pI6:%u\n",
+				atid, status, status2errno(status),
+				la6->sin6_addr.s6_addr, ntohs(la6->sin6_port),
+				ra6->sin6_addr.s6_addr, ntohs(ra6->sin6_port));
+		}
+		break;
+	}
+
+	connect_reply_upcall(ep, status2errno(status));
+	state_set(&ep->com, DEAD);
+
+	if (status && act_open_has_tid(status))
+		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, GET_TID(rpl));
+
+	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, atid);
+	cxgb4_free_atid(t, atid);
+	dst_release(ep->dst);
+	cxgb4_l2t_release(ep->l2t);
+	c4iw_put_ep(&ep->com);
+
+	return 0;
+}
+
+static int pass_open_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_pass_open_rpl *rpl = cplhdr(skb);
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int stid = GET_TID(rpl);
+	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
+
+	if (!ep) {
+		PDBG("%s stid %d lookup failure!\n", __func__, stid);
+		goto out;
+	}
+	PDBG("%s ep %p status %d error %d\n", __func__, ep,
+	     rpl->status, status2errno(rpl->status));
+	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
+
+out:
+	return 0;
+}
+
+static int close_listsrv_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_close_listsvr_rpl *rpl = cplhdr(skb);
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int stid = GET_TID(rpl);
+	struct c4iw_listen_ep *ep = lookup_stid(t, stid);
+
+	PDBG("%s ep %p\n", __func__, ep);
+	c4iw_wake_up(&ep->com.wr_wait, status2errno(rpl->status));
+	return 0;
+}
+
+static void accept_cr(struct c4iw_ep *ep, struct sk_buff *skb,
+		      struct cpl_pass_accept_req *req)
+{
+	struct cpl_pass_accept_rpl *rpl;
+	unsigned int mtu_idx;
+	u64 opt0;
+	u32 opt2;
+	int wscale;
+	struct cpl_t5_pass_accept_rpl *rpl5 = NULL;
+	int win;
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	BUG_ON(skb_cloned(skb));
+
+	skb_get(skb);
+	rpl = cplhdr(skb);
+	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
+		skb_trim(skb, roundup(sizeof(*rpl5), 16));
+		rpl5 = (void *)rpl;
+		INIT_TP_WR(rpl5, ep->hwtid);
+	} else {
+		skb_trim(skb, sizeof(*rpl));
+		INIT_TP_WR(rpl, ep->hwtid);
+	}
+	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
+						    ep->hwtid));
+
+	best_mtu(ep->com.dev->rdev.lldi.mtus, ep->mtu, &mtu_idx,
+		 enable_tcp_timestamps && req->tcpopt.tstamp);
+	wscale = compute_wscale(rcv_win);
+
+	/*
+	 * Specify the largest window that will fit in opt0. The
+	 * remainder will be specified in the rx_data_ack.
+	 */
+	win = ep->rcv_win >> 10;
+	if (win > RCV_BUFSIZ_MASK)
+		win = RCV_BUFSIZ_MASK;
+	opt0 = (nocong ? NO_CONG(1) : 0) |
+	       KEEP_ALIVE(1) |
+	       DELACK(1) |
+	       WND_SCALE(wscale) |
+	       MSS_IDX(mtu_idx) |
+	       L2T_IDX(ep->l2t->idx) |
+	       TX_CHAN(ep->tx_chan) |
+	       SMAC_SEL(ep->smac_idx) |
+	       DSCP(ep->tos >> 2) |
+	       ULP_MODE(ULP_MODE_TCPDDP) |
+	       RCV_BUFSIZ(win);
+	opt2 = RX_CHANNEL(0) |
+	       RSS_QUEUE_VALID | RSS_QUEUE(ep->rss_qid);
+
+	if (enable_tcp_timestamps && req->tcpopt.tstamp)
+		opt2 |= TSTAMPS_EN(1);
+	if (enable_tcp_sack && req->tcpopt.sack)
+		opt2 |= SACK_EN(1);
+	if (wscale && enable_tcp_window_scaling)
+		opt2 |= WND_SCALE_EN(1);
+	if (enable_ecn) {
+		const struct tcphdr *tcph;
+		u32 hlen = ntohl(req->hdr_len);
+
+		tcph = (const void *)(req + 1) + G_ETH_HDR_LEN(hlen) +
+			G_IP_HDR_LEN(hlen);
+		if (tcph->ece && tcph->cwr)
+			opt2 |= CCTRL_ECN(1);
+	}
+	if (is_t5(ep->com.dev->rdev.lldi.adapter_type)) {
+		u32 isn = (prandom_u32() & ~7UL) - 1;
+		opt2 |= T5_OPT_2_VALID;
+		opt2 |= V_CONG_CNTRL(CONG_ALG_TAHOE);
+		opt2 |= CONG_CNTRL_VALID; /* OPT_2_ISS for T5 */
+		rpl5 = (void *)rpl;
+		memset(&rpl5->iss, 0, roundup(sizeof(*rpl5)-sizeof(*rpl), 16));
+		if (peer2peer)
+			isn += 4;
+		rpl5->iss = cpu_to_be32(isn);
+		PDBG("%s iss %u\n", __func__, be32_to_cpu(rpl5->iss));
+	}
+
+	rpl->opt0 = cpu_to_be64(opt0);
+	rpl->opt2 = cpu_to_be32(opt2);
+	set_wr_txq(skb, CPL_PRIORITY_SETUP, ep->ctrlq_idx);
+	t4_set_arp_err_handler(skb, NULL, arp_failure_discard);
+	c4iw_l2t_send(&ep->com.dev->rdev, skb, ep->l2t);
+
+	return;
+}
+
+static void reject_cr(struct c4iw_dev *dev, u32 hwtid, struct sk_buff *skb)
+{
+	PDBG("%s c4iw_dev %p tid %u\n", __func__, dev, hwtid);
+	BUG_ON(skb_cloned(skb));
+	skb_trim(skb, sizeof(struct cpl_tid_release));
+	release_tid(&dev->rdev, hwtid, skb);
+	return;
+}
+
+static void get_4tuple(struct cpl_pass_accept_req *req, int *iptype,
+		       __u8 *local_ip, __u8 *peer_ip,
+		       __be16 *local_port, __be16 *peer_port)
+{
+	int eth_len = G_ETH_HDR_LEN(be32_to_cpu(req->hdr_len));
+	int ip_len = G_IP_HDR_LEN(be32_to_cpu(req->hdr_len));
+	struct iphdr *ip = (struct iphdr *)((u8 *)(req + 1) + eth_len);
+	struct ipv6hdr *ip6 = (struct ipv6hdr *)((u8 *)(req + 1) + eth_len);
+	struct tcphdr *tcp = (struct tcphdr *)
+			     ((u8 *)(req + 1) + eth_len + ip_len);
+
+	if (ip->version == 4) {
+		PDBG("%s saddr 0x%x daddr 0x%x sport %u dport %u\n", __func__,
+		     ntohl(ip->saddr), ntohl(ip->daddr), ntohs(tcp->source),
+		     ntohs(tcp->dest));
+		*iptype = 4;
+		memcpy(peer_ip, &ip->saddr, 4);
+		memcpy(local_ip, &ip->daddr, 4);
+	} else {
+		PDBG("%s saddr %pI6 daddr %pI6 sport %u dport %u\n", __func__,
+		     ip6->saddr.s6_addr, ip6->daddr.s6_addr, ntohs(tcp->source),
+		     ntohs(tcp->dest));
+		*iptype = 6;
+		memcpy(peer_ip, ip6->saddr.s6_addr, 16);
+		memcpy(local_ip, ip6->daddr.s6_addr, 16);
+	}
+	*peer_port = tcp->source;
+	*local_port = tcp->dest;
+
+	return;
+}
+
+static int pass_accept_req(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *child_ep = NULL, *parent_ep;
+	struct cpl_pass_accept_req *req = cplhdr(skb);
+	unsigned int stid = GET_POPEN_TID(ntohl(req->tos_stid));
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int hwtid = GET_TID(req);
+	struct dst_entry *dst;
+	__u8 local_ip[16], peer_ip[16];
+	__be16 local_port, peer_port;
+	int err;
+	u16 peer_mss = ntohs(req->tcpopt.mss);
+	int iptype;
+	unsigned short hdrs;
+
+	parent_ep = lookup_stid(t, stid);
+	if (!parent_ep) {
+		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
+		goto reject;
+	}
+
+	if (state_read(&parent_ep->com) != LISTEN) {
+		printk(KERN_ERR "%s - listening ep not in LISTEN\n",
+		       __func__);
+		goto reject;
+	}
+
+	get_4tuple(req, &iptype, local_ip, peer_ip, &local_port, &peer_port);
+
+	/* Find output route */
+	if (iptype == 4)  {
+		PDBG("%s parent ep %p hwtid %u laddr %pI4 raddr %pI4 lport %d rport %d peer_mss %d\n"
+		     , __func__, parent_ep, hwtid,
+		     local_ip, peer_ip, ntohs(local_port),
+		     ntohs(peer_port), peer_mss);
+		dst = find_route(dev, *(__be32 *)local_ip, *(__be32 *)peer_ip,
+				 local_port, peer_port,
+				 GET_POPEN_TOS(ntohl(req->tos_stid)));
+	} else {
+		PDBG("%s parent ep %p hwtid %u laddr %pI6 raddr %pI6 lport %d rport %d peer_mss %d\n"
+		     , __func__, parent_ep, hwtid,
+		     local_ip, peer_ip, ntohs(local_port),
+		     ntohs(peer_port), peer_mss);
+		dst = find_route6(dev, local_ip, peer_ip, local_port, peer_port,
+				  PASS_OPEN_TOS(ntohl(req->tos_stid)),
+				  ((struct sockaddr_in6 *)
+				  &parent_ep->com.local_addr)->sin6_scope_id);
+	}
+	if (!dst) {
+		printk(KERN_ERR MOD "%s - failed to find dst entry!\n",
+		       __func__);
+		goto reject;
+	}
+
+	child_ep = alloc_ep(sizeof(*child_ep), GFP_KERNEL);
+	if (!child_ep) {
+		printk(KERN_ERR MOD "%s - failed to allocate ep entry!\n",
+		       __func__);
+		dst_release(dst);
+		goto reject;
+	}
+
+	err = import_ep(child_ep, iptype, peer_ip, dst, dev, false);
+	if (err) {
+		printk(KERN_ERR MOD "%s - failed to allocate l2t entry!\n",
+		       __func__);
+		dst_release(dst);
+		kfree(child_ep);
+		goto reject;
+	}
+
+	hdrs = sizeof(struct iphdr) + sizeof(struct tcphdr) +
+	       ((enable_tcp_timestamps && req->tcpopt.tstamp) ? 12 : 0);
+	if (peer_mss && child_ep->mtu > (peer_mss + hdrs))
+		child_ep->mtu = peer_mss + hdrs;
+
+	state_set(&child_ep->com, CONNECTING);
+	child_ep->com.dev = dev;
+	child_ep->com.cm_id = NULL;
+	if (iptype == 4) {
+		struct sockaddr_in *sin = (struct sockaddr_in *)
+			&child_ep->com.local_addr;
+		sin->sin_family = PF_INET;
+		sin->sin_port = local_port;
+		sin->sin_addr.s_addr = *(__be32 *)local_ip;
+		sin = (struct sockaddr_in *)&child_ep->com.remote_addr;
+		sin->sin_family = PF_INET;
+		sin->sin_port = peer_port;
+		sin->sin_addr.s_addr = *(__be32 *)peer_ip;
+	} else {
+		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
+			&child_ep->com.local_addr;
+		sin6->sin6_family = PF_INET6;
+		sin6->sin6_port = local_port;
+		memcpy(sin6->sin6_addr.s6_addr, local_ip, 16);
+		sin6 = (struct sockaddr_in6 *)&child_ep->com.remote_addr;
+		sin6->sin6_family = PF_INET6;
+		sin6->sin6_port = peer_port;
+		memcpy(sin6->sin6_addr.s6_addr, peer_ip, 16);
+	}
+	c4iw_get_ep(&parent_ep->com);
+	child_ep->parent_ep = parent_ep;
+	child_ep->tos = GET_POPEN_TOS(ntohl(req->tos_stid));
+	child_ep->dst = dst;
+	child_ep->hwtid = hwtid;
+
+	PDBG("%s tx_chan %u smac_idx %u rss_qid %u\n", __func__,
+	     child_ep->tx_chan, child_ep->smac_idx, child_ep->rss_qid);
+
+	init_timer(&child_ep->timer);
+	cxgb4_insert_tid(t, child_ep, hwtid);
+	insert_handle(dev, &dev->hwtid_idr, child_ep, child_ep->hwtid);
+	accept_cr(child_ep, skb, req);
+	set_bit(PASS_ACCEPT_REQ, &child_ep->com.history);
+	goto out;
+reject:
+	reject_cr(dev, hwtid, skb);
+out:
+	return 0;
+}
+
+static int pass_establish(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_pass_establish *req = cplhdr(skb);
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(req);
+
+	ep = lookup_tid(t, tid);
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	ep->snd_seq = be32_to_cpu(req->snd_isn);
+	ep->rcv_seq = be32_to_cpu(req->rcv_isn);
+
+	PDBG("%s ep %p hwtid %u tcp_opt 0x%02x\n", __func__, ep, tid,
+	     ntohs(req->tcp_opt));
+
+	set_emss(ep, ntohs(req->tcp_opt));
+
+	dst_confirm(ep->dst);
+	state_set(&ep->com, MPA_REQ_WAIT);
+	start_ep_timer(ep);
+	send_flowc(ep, skb);
+	set_bit(PASS_ESTAB, &ep->com.history);
+
+	return 0;
+}
+
+static int peer_close(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_peer_close *hdr = cplhdr(skb);
+	struct c4iw_ep *ep;
+	struct c4iw_qp_attributes attrs;
+	int disconnect = 1;
+	int release = 0;
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(hdr);
+	int ret;
+
+	ep = lookup_tid(t, tid);
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	dst_confirm(ep->dst);
+
+	set_bit(PEER_CLOSE, &ep->com.history);
+	mutex_lock(&ep->com.mutex);
+	switch (ep->com.state) {
+	case MPA_REQ_WAIT:
+		__state_set(&ep->com, CLOSING);
+		break;
+	case MPA_REQ_SENT:
+		__state_set(&ep->com, CLOSING);
+		connect_reply_upcall(ep, -ECONNRESET);
+		break;
+	case MPA_REQ_RCVD:
+
+		/*
+		 * We're gonna mark this puppy DEAD, but keep
+		 * the reference on it until the ULP accepts or
+		 * rejects the CR. Also wake up anyone waiting
+		 * in rdma connection migration (see c4iw_accept_cr()).
+		 */
+		__state_set(&ep->com, CLOSING);
+		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+		break;
+	case MPA_REP_SENT:
+		__state_set(&ep->com, CLOSING);
+		PDBG("waking up ep %p tid %u\n", ep, ep->hwtid);
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+		break;
+	case FPDU_MODE:
+		start_ep_timer(ep);
+		__state_set(&ep->com, CLOSING);
+		attrs.next_state = C4IW_QP_STATE_CLOSING;
+		ret = c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		if (ret != -ECONNRESET) {
+			peer_close_upcall(ep);
+			disconnect = 1;
+		}
+		break;
+	case ABORTING:
+		disconnect = 0;
+		break;
+	case CLOSING:
+		__state_set(&ep->com, MORIBUND);
+		disconnect = 0;
+		break;
+	case MORIBUND:
+		(void)stop_ep_timer(ep);
+		if (ep->com.cm_id && ep->com.qp) {
+			attrs.next_state = C4IW_QP_STATE_IDLE;
+			c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+				       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+		}
+		close_complete_upcall(ep, 0);
+		__state_set(&ep->com, DEAD);
+		release = 1;
+		disconnect = 0;
+		break;
+	case DEAD:
+		disconnect = 0;
+		break;
+	default:
+		BUG_ON(1);
+	}
+	mutex_unlock(&ep->com.mutex);
+	if (disconnect)
+		c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
+	if (release)
+		release_ep_resources(ep);
+	return 0;
+}
+
+static int peer_abort(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_abort_req_rss *req = cplhdr(skb);
+	struct c4iw_ep *ep;
+	struct cpl_abort_rpl *rpl;
+	struct sk_buff *rpl_skb;
+	struct c4iw_qp_attributes attrs;
+	int ret;
+	int release = 0;
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(req);
+
+	ep = lookup_tid(t, tid);
+	if (is_neg_adv(req->status)) {
+		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
+		     ep->hwtid);
+		return 0;
+	}
+	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
+	     ep->com.state);
+	set_bit(PEER_ABORT, &ep->com.history);
+
+	/*
+	 * Wake up any threads in rdma_init() or rdma_fini().
+	 * However, this is not needed if com state is just
+	 * MPA_REQ_SENT
+	 */
+	if (ep->com.state != MPA_REQ_SENT)
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+
+	mutex_lock(&ep->com.mutex);
+	switch (ep->com.state) {
+	case CONNECTING:
+		break;
+	case MPA_REQ_WAIT:
+		(void)stop_ep_timer(ep);
+		break;
+	case MPA_REQ_SENT:
+		(void)stop_ep_timer(ep);
+		if (mpa_rev == 1 || (mpa_rev == 2 && ep->tried_with_mpa_v1))
+			connect_reply_upcall(ep, -ECONNRESET);
+		else {
+			/*
+			 * we just don't send notification upwards because we
+			 * want to retry with mpa_v1 without upper layers even
+			 * knowing it.
+			 *
+			 * do some housekeeping so as to re-initiate the
+			 * connection
+			 */
+			PDBG("%s: mpa_rev=%d. Retrying with mpav1\n", __func__,
+			     mpa_rev);
+			ep->retry_with_mpa_v1 = 1;
+		}
+		break;
+	case MPA_REP_SENT:
+		break;
+	case MPA_REQ_RCVD:
+		break;
+	case MORIBUND:
+	case CLOSING:
+		stop_ep_timer(ep);
+		/*FALLTHROUGH*/
+	case FPDU_MODE:
+		if (ep->com.cm_id && ep->com.qp) {
+			attrs.next_state = C4IW_QP_STATE_ERROR;
+			ret = c4iw_modify_qp(ep->com.qp->rhp,
+				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
+				     &attrs, 1);
+			if (ret)
+				printk(KERN_ERR MOD
+				       "%s - qp <- error failed!\n",
+				       __func__);
+		}
+		peer_abort_upcall(ep);
+		break;
+	case ABORTING:
+		break;
+	case DEAD:
+		PDBG("%s PEER_ABORT IN DEAD STATE!!!!\n", __func__);
+		mutex_unlock(&ep->com.mutex);
+		return 0;
+	default:
+		BUG_ON(1);
+		break;
+	}
+	dst_confirm(ep->dst);
+	if (ep->com.state != ABORTING) {
+		__state_set(&ep->com, DEAD);
+		/* we don't release if we want to retry with mpa_v1 */
+		if (!ep->retry_with_mpa_v1)
+			release = 1;
+	}
+	mutex_unlock(&ep->com.mutex);
+
+	rpl_skb = get_skb(skb, sizeof(*rpl), GFP_KERNEL);
+	if (!rpl_skb) {
+		printk(KERN_ERR MOD "%s - cannot allocate skb!\n",
+		       __func__);
+		release = 1;
+		goto out;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+	rpl = (struct cpl_abort_rpl *) skb_put(rpl_skb, sizeof(*rpl));
+	INIT_TP_WR(rpl, ep->hwtid);
+	OPCODE_TID(rpl) = cpu_to_be32(MK_OPCODE_TID(CPL_ABORT_RPL, ep->hwtid));
+	rpl->cmd = CPL_ABORT_NO_RST;
+	c4iw_ofld_send(&ep->com.dev->rdev, rpl_skb);
+out:
+	if (release)
+		release_ep_resources(ep);
+	else if (ep->retry_with_mpa_v1) {
+		remove_handle(ep->com.dev, &ep->com.dev->hwtid_idr, ep->hwtid);
+		cxgb4_remove_tid(ep->com.dev->rdev.lldi.tids, 0, ep->hwtid);
+		dst_release(ep->dst);
+		cxgb4_l2t_release(ep->l2t);
+		c4iw_reconnect(ep);
+	}
+
+	return 0;
+}
+
+static int close_con_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct c4iw_qp_attributes attrs;
+	struct cpl_close_con_rpl *rpl = cplhdr(skb);
+	int release = 0;
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(rpl);
+
+	ep = lookup_tid(t, tid);
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+	BUG_ON(!ep);
+
+	/* The cm_id may be null if we failed to connect */
+	mutex_lock(&ep->com.mutex);
+	switch (ep->com.state) {
+	case CLOSING:
+		__state_set(&ep->com, MORIBUND);
+		break;
+	case MORIBUND:
+		(void)stop_ep_timer(ep);
+		if ((ep->com.cm_id) && (ep->com.qp)) {
+			attrs.next_state = C4IW_QP_STATE_IDLE;
+			c4iw_modify_qp(ep->com.qp->rhp,
+					     ep->com.qp,
+					     C4IW_QP_ATTR_NEXT_STATE,
+					     &attrs, 1);
+		}
+		close_complete_upcall(ep, 0);
+		__state_set(&ep->com, DEAD);
+		release = 1;
+		break;
+	case ABORTING:
+	case DEAD:
+		break;
+	default:
+		BUG_ON(1);
+		break;
+	}
+	mutex_unlock(&ep->com.mutex);
+	if (release)
+		release_ep_resources(ep);
+	return 0;
+}
+
+static int terminate(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_rdma_terminate *rpl = cplhdr(skb);
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(rpl);
+	struct c4iw_ep *ep;
+	struct c4iw_qp_attributes attrs;
+
+	ep = lookup_tid(t, tid);
+	BUG_ON(!ep);
+
+	if (ep && ep->com.qp) {
+		printk(KERN_WARNING MOD "TERM received tid %u qpid %u\n", tid,
+		       ep->com.qp->wq.sq.qid);
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		c4iw_modify_qp(ep->com.qp->rhp, ep->com.qp,
+			       C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+	} else
+		printk(KERN_WARNING MOD "TERM received tid %u no ep/qp\n", tid);
+
+	return 0;
+}
+
+/*
+ * Upcall from the adapter indicating data has been transmitted.
+ * For us its just the single MPA request or reply.  We can now free
+ * the skb holding the mpa message.
+ */
+static int fw4_ack(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct c4iw_ep *ep;
+	struct cpl_fw4_ack *hdr = cplhdr(skb);
+	u8 credits = hdr->credits;
+	unsigned int tid = GET_TID(hdr);
+	struct tid_info *t = dev->rdev.lldi.tids;
+
+
+	ep = lookup_tid(t, tid);
+	PDBG("%s ep %p tid %u credits %u\n", __func__, ep, ep->hwtid, credits);
+	if (credits == 0) {
+		PDBG("%s 0 credit ack ep %p tid %u state %u\n",
+		     __func__, ep, ep->hwtid, state_read(&ep->com));
+		return 0;
+	}
+
+	dst_confirm(ep->dst);
+	if (ep->mpa_skb) {
+		PDBG("%s last streaming msg ack ep %p tid %u state %u "
+		     "initiator %u freeing skb\n", __func__, ep, ep->hwtid,
+		     state_read(&ep->com), ep->mpa_attr.initiator ? 1 : 0);
+		kfree_skb(ep->mpa_skb);
+		ep->mpa_skb = NULL;
+	}
+	return 0;
+}
+
+int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
+{
+	int err = 0;
+	int disconnect = 0;
+	struct c4iw_ep *ep = to_ep(cm_id);
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+
+	mutex_lock(&ep->com.mutex);
+	if (ep->com.state == DEAD) {
+		mutex_unlock(&ep->com.mutex);
+		c4iw_put_ep(&ep->com);
+		return -ECONNRESET;
+	}
+	set_bit(ULP_REJECT, &ep->com.history);
+	BUG_ON(ep->com.state != MPA_REQ_RCVD);
+	if (mpa_rev == 0)
+		abort_connection(ep, NULL, GFP_KERNEL);
+	else {
+		err = send_mpa_reject(ep, pdata, pdata_len);
+		disconnect = 1;
+	}
+	mutex_unlock(&ep->com.mutex);
+	if (disconnect)
+		err = c4iw_ep_disconnect(ep, 0, GFP_KERNEL);
+	c4iw_put_ep(&ep->com);
+	return 0;
+}
+
+int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
+{
+	int err;
+	struct c4iw_qp_attributes attrs;
+	enum c4iw_qp_attr_mask mask;
+	struct c4iw_ep *ep = to_ep(cm_id);
+	struct c4iw_dev *h = to_c4iw_dev(cm_id->device);
+	struct c4iw_qp *qp = get_qhp(h, conn_param->qpn);
+
+	PDBG("%s ep %p tid %u\n", __func__, ep, ep->hwtid);
+
+	mutex_lock(&ep->com.mutex);
+	if (ep->com.state == DEAD) {
+		err = -ECONNRESET;
+		goto err;
+	}
+
+	BUG_ON(ep->com.state != MPA_REQ_RCVD);
+	BUG_ON(!qp);
+
+	set_bit(ULP_ACCEPT, &ep->com.history);
+	if ((conn_param->ord > c4iw_max_read_depth) ||
+	    (conn_param->ird > c4iw_max_read_depth)) {
+		abort_connection(ep, NULL, GFP_KERNEL);
+		err = -EINVAL;
+		goto err;
+	}
+
+	if (ep->mpa_attr.version == 2 && ep->mpa_attr.enhanced_rdma_conn) {
+		if (conn_param->ord > ep->ird) {
+			ep->ird = conn_param->ird;
+			ep->ord = conn_param->ord;
+			send_mpa_reject(ep, conn_param->private_data,
+					conn_param->private_data_len);
+			abort_connection(ep, NULL, GFP_KERNEL);
+			err = -ENOMEM;
+			goto err;
+		}
+		if (conn_param->ird > ep->ord) {
+			if (!ep->ord)
+				conn_param->ird = 1;
+			else {
+				abort_connection(ep, NULL, GFP_KERNEL);
+				err = -ENOMEM;
+				goto err;
+			}
+		}
+
+	}
+	ep->ird = conn_param->ird;
+	ep->ord = conn_param->ord;
+
+	if (ep->mpa_attr.version != 2)
+		if (peer2peer && ep->ird == 0)
+			ep->ird = 1;
+
+	PDBG("%s %d ird %d ord %d\n", __func__, __LINE__, ep->ird, ep->ord);
+
+	cm_id->add_ref(cm_id);
+	ep->com.cm_id = cm_id;
+	ep->com.qp = qp;
+	ref_qp(ep);
+
+	/* bind QP to EP and move to RTS */
+	attrs.mpa_attr = ep->mpa_attr;
+	attrs.max_ird = ep->ird;
+	attrs.max_ord = ep->ord;
+	attrs.llp_stream_handle = ep;
+	attrs.next_state = C4IW_QP_STATE_RTS;
+
+	/* bind QP and TID with INIT_WR */
+	mask = C4IW_QP_ATTR_NEXT_STATE |
+			     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
+			     C4IW_QP_ATTR_MPA_ATTR |
+			     C4IW_QP_ATTR_MAX_IRD |
+			     C4IW_QP_ATTR_MAX_ORD;
+
+	err = c4iw_modify_qp(ep->com.qp->rhp,
+			     ep->com.qp, mask, &attrs, 1);
+	if (err)
+		goto err1;
+	err = send_mpa_reply(ep, conn_param->private_data,
+			     conn_param->private_data_len);
+	if (err)
+		goto err1;
+
+	__state_set(&ep->com, FPDU_MODE);
+	established_upcall(ep);
+	mutex_unlock(&ep->com.mutex);
+	c4iw_put_ep(&ep->com);
+	return 0;
+err1:
+	ep->com.cm_id = NULL;
+	cm_id->rem_ref(cm_id);
+err:
+	mutex_unlock(&ep->com.mutex);
+	c4iw_put_ep(&ep->com);
+	return err;
+}
+
+static int pick_local_ipaddrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
+{
+	struct in_device *ind;
+	int found = 0;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
+
+	ind = in_dev_get(dev->rdev.lldi.ports[0]);
+	if (!ind)
+		return -EADDRNOTAVAIL;
+	for_primary_ifa(ind) {
+		laddr->sin_addr.s_addr = ifa->ifa_address;
+		raddr->sin_addr.s_addr = ifa->ifa_address;
+		found = 1;
+		break;
+	}
+	endfor_ifa(ind);
+	in_dev_put(ind);
+	return found ? 0 : -EADDRNOTAVAIL;
+}
+
+static int get_lladdr(struct net_device *dev, struct in6_addr *addr,
+		      unsigned char banned_flags)
+{
+	struct inet6_dev *idev;
+	int err = -EADDRNOTAVAIL;
+
+	rcu_read_lock();
+	idev = __in6_dev_get(dev);
+	if (idev != NULL) {
+		struct inet6_ifaddr *ifp;
+
+		read_lock_bh(&idev->lock);
+		list_for_each_entry(ifp, &idev->addr_list, if_list) {
+			if (ifp->scope == IFA_LINK &&
+			    !(ifp->flags & banned_flags)) {
+				memcpy(addr, &ifp->addr, 16);
+				err = 0;
+				break;
+			}
+		}
+		read_unlock_bh(&idev->lock);
+	}
+	rcu_read_unlock();
+	return err;
+}
+
+static int pick_local_ip6addrs(struct c4iw_dev *dev, struct iw_cm_id *cm_id)
+{
+	struct in6_addr uninitialized_var(addr);
+	struct sockaddr_in6 *la6 = (struct sockaddr_in6 *)&cm_id->local_addr;
+	struct sockaddr_in6 *ra6 = (struct sockaddr_in6 *)&cm_id->remote_addr;
+
+	if (get_lladdr(dev->rdev.lldi.ports[0], &addr, IFA_F_TENTATIVE)) {
+		memcpy(la6->sin6_addr.s6_addr, &addr, 16);
+		memcpy(ra6->sin6_addr.s6_addr, &addr, 16);
+		return 0;
+	}
+	return -EADDRNOTAVAIL;
+}
+
+int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
+{
+	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
+	struct c4iw_ep *ep;
+	int err = 0;
+	struct sockaddr_in *laddr;
+	struct sockaddr_in *raddr;
+	struct sockaddr_in6 *laddr6;
+	struct sockaddr_in6 *raddr6;
+	struct iwpm_dev_data pm_reg_msg;
+	struct iwpm_sa_data pm_msg;
+	__u8 *ra;
+	int iptype;
+	int iwpm_err = 0;
+
+	if ((conn_param->ord > c4iw_max_read_depth) ||
+	    (conn_param->ird > c4iw_max_read_depth)) {
+		err = -EINVAL;
+		goto out;
+	}
+	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
+	if (!ep) {
+		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
+		err = -ENOMEM;
+		goto out;
+	}
+	init_timer(&ep->timer);
+	ep->plen = conn_param->private_data_len;
+	if (ep->plen)
+		memcpy(ep->mpa_pkt + sizeof(struct mpa_message),
+		       conn_param->private_data, ep->plen);
+	ep->ird = conn_param->ird;
+	ep->ord = conn_param->ord;
+
+	if (peer2peer && ep->ord == 0)
+		ep->ord = 1;
+
+	cm_id->add_ref(cm_id);
+	ep->com.dev = dev;
+	ep->com.cm_id = cm_id;
+	ep->com.qp = get_qhp(dev, conn_param->qpn);
+	if (!ep->com.qp) {
+		PDBG("%s qpn 0x%x not found!\n", __func__, conn_param->qpn);
+		err = -EINVAL;
+		goto fail1;
+	}
+	ref_qp(ep);
+	PDBG("%s qpn 0x%x qp %p cm_id %p\n", __func__, conn_param->qpn,
+	     ep->com.qp, cm_id);
+
+	/*
+	 * Allocate an active TID to initiate a TCP connection.
+	 */
+	ep->atid = cxgb4_alloc_atid(dev->rdev.lldi.tids, ep);
+	if (ep->atid == -1) {
+		printk(KERN_ERR MOD "%s - cannot alloc atid.\n", __func__);
+		err = -ENOMEM;
+		goto fail1;
+	}
+	insert_handle(dev, &dev->atid_idr, ep, ep->atid);
+
+	memcpy(&ep->com.local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.local_addr));
+	memcpy(&ep->com.remote_addr, &cm_id->remote_addr,
+	       sizeof(ep->com.remote_addr));
+
+	/* No port mapper available, go with the specified peer information */
+	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.mapped_local_addr));
+	memcpy(&ep->com.mapped_remote_addr, &cm_id->remote_addr,
+	       sizeof(ep->com.mapped_remote_addr));
+
+	c4iw_form_reg_msg(dev, &pm_reg_msg);
+	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
+	if (iwpm_err) {
+		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
+			__func__, iwpm_err);
+	}
+	if (iwpm_valid_pid() && !iwpm_err) {
+		c4iw_form_pm_msg(ep, &pm_msg);
+		iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_C4IW);
+		if (iwpm_err)
+			PDBG("%s: Port Mapper query fail (err = %d).\n",
+				__func__, iwpm_err);
+		else
+			c4iw_record_pm_msg(ep, &pm_msg);
+	}
+	if (iwpm_create_mapinfo(&ep->com.local_addr,
+				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
+		iwpm_remove_mapping(&ep->com.local_addr, RDMA_NL_C4IW);
+		err = -ENOMEM;
+		goto fail1;
+	}
+	print_addr(&ep->com, __func__, "add_query/create_mapinfo");
+	set_bit(RELEASE_MAPINFO, &ep->com.flags);
+
+	laddr = (struct sockaddr_in *)&ep->com.mapped_local_addr;
+	raddr = (struct sockaddr_in *)&ep->com.mapped_remote_addr;
+	laddr6 = (struct sockaddr_in6 *)&ep->com.mapped_local_addr;
+	raddr6 = (struct sockaddr_in6 *) &ep->com.mapped_remote_addr;
+
+	if (cm_id->remote_addr.ss_family == AF_INET) {
+		iptype = 4;
+		ra = (__u8 *)&raddr->sin_addr;
+
+		/*
+		 * Handle loopback requests to INADDR_ANY.
+		 */
+		if ((__force int)raddr->sin_addr.s_addr == INADDR_ANY) {
+			err = pick_local_ipaddrs(dev, cm_id);
+			if (err)
+				goto fail1;
+		}
+
+		/* find a route */
+		PDBG("%s saddr %pI4 sport 0x%x raddr %pI4 rport 0x%x\n",
+		     __func__, &laddr->sin_addr, ntohs(laddr->sin_port),
+		     ra, ntohs(raddr->sin_port));
+		ep->dst = find_route(dev, laddr->sin_addr.s_addr,
+				     raddr->sin_addr.s_addr, laddr->sin_port,
+				     raddr->sin_port, 0);
+	} else {
+		iptype = 6;
+		ra = (__u8 *)&raddr6->sin6_addr;
+
+		/*
+		 * Handle loopback requests to INADDR_ANY.
+		 */
+		if (ipv6_addr_type(&raddr6->sin6_addr) == IPV6_ADDR_ANY) {
+			err = pick_local_ip6addrs(dev, cm_id);
+			if (err)
+				goto fail1;
+		}
+
+		/* find a route */
+		PDBG("%s saddr %pI6 sport 0x%x raddr %pI6 rport 0x%x\n",
+		     __func__, laddr6->sin6_addr.s6_addr,
+		     ntohs(laddr6->sin6_port),
+		     raddr6->sin6_addr.s6_addr, ntohs(raddr6->sin6_port));
+		ep->dst = find_route6(dev, laddr6->sin6_addr.s6_addr,
+				      raddr6->sin6_addr.s6_addr,
+				      laddr6->sin6_port, raddr6->sin6_port, 0,
+				      raddr6->sin6_scope_id);
+	}
+	if (!ep->dst) {
+		printk(KERN_ERR MOD "%s - cannot find route.\n", __func__);
+		err = -EHOSTUNREACH;
+		goto fail2;
+	}
+
+	err = import_ep(ep, iptype, ra, ep->dst, ep->com.dev, true);
+	if (err) {
+		printk(KERN_ERR MOD "%s - cannot alloc l2e.\n", __func__);
+		goto fail3;
+	}
+
+	PDBG("%s txq_idx %u tx_chan %u smac_idx %u rss_qid %u l2t_idx %u\n",
+		__func__, ep->txq_idx, ep->tx_chan, ep->smac_idx, ep->rss_qid,
+		ep->l2t->idx);
+
+	state_set(&ep->com, CONNECTING);
+	ep->tos = 0;
+
+	/* send connect request to rnic */
+	err = send_connect(ep);
+	if (!err)
+		goto out;
+
+	cxgb4_l2t_release(ep->l2t);
+fail3:
+	dst_release(ep->dst);
+fail2:
+	remove_handle(ep->com.dev, &ep->com.dev->atid_idr, ep->atid);
+	cxgb4_free_atid(ep->com.dev->rdev.lldi.tids, ep->atid);
+fail1:
+	cm_id->rem_ref(cm_id);
+	c4iw_put_ep(&ep->com);
+out:
+	return err;
+}
+
+static int create_server6(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
+{
+	int err;
+	struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)
+				    &ep->com.mapped_local_addr;
+
+	c4iw_init_wr_wait(&ep->com.wr_wait);
+	err = cxgb4_create_server6(ep->com.dev->rdev.lldi.ports[0],
+				   ep->stid, &sin6->sin6_addr,
+				   sin6->sin6_port,
+				   ep->com.dev->rdev.lldi.rxq_ids[0]);
+	if (!err)
+		err = c4iw_wait_for_reply(&ep->com.dev->rdev,
+					  &ep->com.wr_wait,
+					  0, 0, __func__);
+	if (err)
+		pr_err("cxgb4_create_server6/filter failed err %d stid %d laddr %pI6 lport %d\n",
+		       err, ep->stid,
+		       sin6->sin6_addr.s6_addr, ntohs(sin6->sin6_port));
+	return err;
+}
+
+static int create_server4(struct c4iw_dev *dev, struct c4iw_listen_ep *ep)
+{
+	int err;
+	struct sockaddr_in *sin = (struct sockaddr_in *)
+				  &ep->com.mapped_local_addr;
+
+	if (dev->rdev.lldi.enable_fw_ofld_conn) {
+		do {
+			err = cxgb4_create_server_filter(
+				ep->com.dev->rdev.lldi.ports[0], ep->stid,
+				sin->sin_addr.s_addr, sin->sin_port, 0,
+				ep->com.dev->rdev.lldi.rxq_ids[0], 0, 0);
+			if (err == -EBUSY) {
+				set_current_state(TASK_UNINTERRUPTIBLE);
+				schedule_timeout(usecs_to_jiffies(100));
+			}
+		} while (err == -EBUSY);
+	} else {
+		c4iw_init_wr_wait(&ep->com.wr_wait);
+		err = cxgb4_create_server(ep->com.dev->rdev.lldi.ports[0],
+				ep->stid, sin->sin_addr.s_addr, sin->sin_port,
+				0, ep->com.dev->rdev.lldi.rxq_ids[0]);
+		if (!err)
+			err = c4iw_wait_for_reply(&ep->com.dev->rdev,
+						  &ep->com.wr_wait,
+						  0, 0, __func__);
+	}
+	if (err)
+		pr_err("cxgb4_create_server/filter failed err %d stid %d laddr %pI4 lport %d\n"
+		       , err, ep->stid,
+		       &sin->sin_addr, ntohs(sin->sin_port));
+	return err;
+}
+
+int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog)
+{
+	int err = 0;
+	struct c4iw_dev *dev = to_c4iw_dev(cm_id->device);
+	struct c4iw_listen_ep *ep;
+	struct iwpm_dev_data pm_reg_msg;
+	struct iwpm_sa_data pm_msg;
+	int iwpm_err = 0;
+
+	might_sleep();
+
+	ep = alloc_ep(sizeof(*ep), GFP_KERNEL);
+	if (!ep) {
+		printk(KERN_ERR MOD "%s - cannot alloc ep.\n", __func__);
+		err = -ENOMEM;
+		goto fail1;
+	}
+	PDBG("%s ep %p\n", __func__, ep);
+	cm_id->add_ref(cm_id);
+	ep->com.cm_id = cm_id;
+	ep->com.dev = dev;
+	ep->backlog = backlog;
+	memcpy(&ep->com.local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.local_addr));
+
+	/*
+	 * Allocate a server TID.
+	 */
+	if (dev->rdev.lldi.enable_fw_ofld_conn &&
+	    ep->com.local_addr.ss_family == AF_INET)
+		ep->stid = cxgb4_alloc_sftid(dev->rdev.lldi.tids,
+					     cm_id->local_addr.ss_family, ep);
+	else
+		ep->stid = cxgb4_alloc_stid(dev->rdev.lldi.tids,
+					    cm_id->local_addr.ss_family, ep);
+
+	if (ep->stid == -1) {
+		printk(KERN_ERR MOD "%s - cannot alloc stid.\n", __func__);
+		err = -ENOMEM;
+		goto fail2;
+	}
+	insert_handle(dev, &dev->stid_idr, ep, ep->stid);
+
+	/* No port mapper available, go with the specified info */
+	memcpy(&ep->com.mapped_local_addr, &cm_id->local_addr,
+	       sizeof(ep->com.mapped_local_addr));
+
+	c4iw_form_reg_msg(dev, &pm_reg_msg);
+	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_C4IW);
+	if (iwpm_err) {
+		PDBG("%s: Port Mapper reg pid fail (err = %d).\n",
+			__func__, iwpm_err);
+	}
+	if (iwpm_valid_pid() && !iwpm_err) {
+		memcpy(&pm_msg.loc_addr, &ep->com.local_addr,
+				sizeof(ep->com.local_addr));
+		iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_C4IW);
+		if (iwpm_err)
+			PDBG("%s: Port Mapper query fail (err = %d).\n",
+				__func__, iwpm_err);
+		else
+			memcpy(&ep->com.mapped_local_addr,
+				&pm_msg.mapped_loc_addr,
+				sizeof(ep->com.mapped_local_addr));
+	}
+	if (iwpm_create_mapinfo(&ep->com.local_addr,
+				&ep->com.mapped_local_addr, RDMA_NL_C4IW)) {
+		err = -ENOMEM;
+		goto fail3;
+	}
+	print_addr(&ep->com, __func__, "add_mapping/create_mapinfo");
+
+	set_bit(RELEASE_MAPINFO, &ep->com.flags);
+	state_set(&ep->com, LISTEN);
+	if (ep->com.local_addr.ss_family == AF_INET)
+		err = create_server4(dev, ep);
+	else
+		err = create_server6(dev, ep);
+	if (!err) {
+		cm_id->provider_data = ep;
+		goto out;
+	}
+
+fail3:
+	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
+			ep->com.local_addr.ss_family);
+fail2:
+	cm_id->rem_ref(cm_id);
+	c4iw_put_ep(&ep->com);
+fail1:
+out:
+	return err;
+}
+
+int c4iw_destroy_listen(struct iw_cm_id *cm_id)
+{
+	int err;
+	struct c4iw_listen_ep *ep = to_listen_ep(cm_id);
+
+	PDBG("%s ep %p\n", __func__, ep);
+
+	might_sleep();
+	state_set(&ep->com, DEAD);
+	if (ep->com.dev->rdev.lldi.enable_fw_ofld_conn &&
+	    ep->com.local_addr.ss_family == AF_INET) {
+		err = cxgb4_remove_server_filter(
+			ep->com.dev->rdev.lldi.ports[0], ep->stid,
+			ep->com.dev->rdev.lldi.rxq_ids[0], 0);
+	} else {
+		c4iw_init_wr_wait(&ep->com.wr_wait);
+		err = cxgb4_remove_server(
+				ep->com.dev->rdev.lldi.ports[0], ep->stid,
+				ep->com.dev->rdev.lldi.rxq_ids[0], 0);
+		if (err)
+			goto done;
+		err = c4iw_wait_for_reply(&ep->com.dev->rdev, &ep->com.wr_wait,
+					  0, 0, __func__);
+	}
+	remove_handle(ep->com.dev, &ep->com.dev->stid_idr, ep->stid);
+	cxgb4_free_stid(ep->com.dev->rdev.lldi.tids, ep->stid,
+			ep->com.local_addr.ss_family);
+done:
+	cm_id->rem_ref(cm_id);
+	c4iw_put_ep(&ep->com);
+	return err;
+}
+
+int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp)
+{
+	int ret = 0;
+	int close = 0;
+	int fatal = 0;
+	struct c4iw_rdev *rdev;
+
+	mutex_lock(&ep->com.mutex);
+
+	PDBG("%s ep %p state %s, abrupt %d\n", __func__, ep,
+	     states[ep->com.state], abrupt);
+
+	rdev = &ep->com.dev->rdev;
+	if (c4iw_fatal_error(rdev)) {
+		fatal = 1;
+		close_complete_upcall(ep, -EIO);
+		ep->com.state = DEAD;
+	}
+	switch (ep->com.state) {
+	case MPA_REQ_WAIT:
+	case MPA_REQ_SENT:
+	case MPA_REQ_RCVD:
+	case MPA_REP_SENT:
+	case FPDU_MODE:
+		close = 1;
+		if (abrupt)
+			ep->com.state = ABORTING;
+		else {
+			ep->com.state = CLOSING;
+			start_ep_timer(ep);
+		}
+		set_bit(CLOSE_SENT, &ep->com.flags);
+		break;
+	case CLOSING:
+		if (!test_and_set_bit(CLOSE_SENT, &ep->com.flags)) {
+			close = 1;
+			if (abrupt) {
+				(void)stop_ep_timer(ep);
+				ep->com.state = ABORTING;
+			} else
+				ep->com.state = MORIBUND;
+		}
+		break;
+	case MORIBUND:
+	case ABORTING:
+	case DEAD:
+		PDBG("%s ignoring disconnect ep %p state %u\n",
+		     __func__, ep, ep->com.state);
+		break;
+	default:
+		BUG();
+		break;
+	}
+
+	if (close) {
+		if (abrupt) {
+			set_bit(EP_DISC_ABORT, &ep->com.history);
+			close_complete_upcall(ep, -ECONNRESET);
+			ret = send_abort(ep, NULL, gfp);
+		} else {
+			set_bit(EP_DISC_CLOSE, &ep->com.history);
+			ret = send_halfclose(ep, gfp);
+		}
+		if (ret)
+			fatal = 1;
+	}
+	mutex_unlock(&ep->com.mutex);
+	if (fatal)
+		release_ep_resources(ep);
+	return ret;
+}
+
+static void active_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
+			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
+{
+	struct c4iw_ep *ep;
+	int atid = be32_to_cpu(req->tid);
+
+	ep = (struct c4iw_ep *)lookup_atid(dev->rdev.lldi.tids,
+					   (__force u32) req->tid);
+	if (!ep)
+		return;
+
+	switch (req->retval) {
+	case FW_ENOMEM:
+		set_bit(ACT_RETRY_NOMEM, &ep->com.history);
+		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
+			send_fw_act_open_req(ep, atid);
+			return;
+		}
+	case FW_EADDRINUSE:
+		set_bit(ACT_RETRY_INUSE, &ep->com.history);
+		if (ep->retry_count++ < ACT_OPEN_RETRY_COUNT) {
+			send_fw_act_open_req(ep, atid);
+			return;
+		}
+		break;
+	default:
+		pr_info("%s unexpected ofld conn wr retval %d\n",
+		       __func__, req->retval);
+		break;
+	}
+	pr_err("active ofld_connect_wr failure %d atid %d\n",
+	       req->retval, atid);
+	mutex_lock(&dev->rdev.stats.lock);
+	dev->rdev.stats.act_ofld_conn_fails++;
+	mutex_unlock(&dev->rdev.stats.lock);
+	connect_reply_upcall(ep, status2errno(req->retval));
+	state_set(&ep->com, DEAD);
+	remove_handle(dev, &dev->atid_idr, atid);
+	cxgb4_free_atid(dev->rdev.lldi.tids, atid);
+	dst_release(ep->dst);
+	cxgb4_l2t_release(ep->l2t);
+	c4iw_put_ep(&ep->com);
+}
+
+static void passive_ofld_conn_reply(struct c4iw_dev *dev, struct sk_buff *skb,
+			struct cpl_fw6_msg_ofld_connection_wr_rpl *req)
+{
+	struct sk_buff *rpl_skb;
+	struct cpl_pass_accept_req *cpl;
+	int ret;
+
+	rpl_skb = (struct sk_buff *)(unsigned long)req->cookie;
+	BUG_ON(!rpl_skb);
+	if (req->retval) {
+		PDBG("%s passive open failure %d\n", __func__, req->retval);
+		mutex_lock(&dev->rdev.stats.lock);
+		dev->rdev.stats.pas_ofld_conn_fails++;
+		mutex_unlock(&dev->rdev.stats.lock);
+		kfree_skb(rpl_skb);
+	} else {
+		cpl = (struct cpl_pass_accept_req *)cplhdr(rpl_skb);
+		OPCODE_TID(cpl) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ,
+					(__force u32) htonl(
+					(__force u32) req->tid)));
+		ret = pass_accept_req(dev, rpl_skb);
+		if (!ret)
+			kfree_skb(rpl_skb);
+	}
+	return;
+}
+
+static int deferred_fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_fw6_msg *rpl = cplhdr(skb);
+	struct cpl_fw6_msg_ofld_connection_wr_rpl *req;
+
+	switch (rpl->type) {
+	case FW6_TYPE_CQE:
+		c4iw_ev_dispatch(dev, (struct t4_cqe *)&rpl->data[0]);
+		break;
+	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
+		req = (struct cpl_fw6_msg_ofld_connection_wr_rpl *)rpl->data;
+		switch (req->t_state) {
+		case TCP_SYN_SENT:
+			active_ofld_conn_reply(dev, skb, req);
+			break;
+		case TCP_SYN_RECV:
+			passive_ofld_conn_reply(dev, skb, req);
+			break;
+		default:
+			pr_err("%s unexpected ofld conn wr state %d\n",
+			       __func__, req->t_state);
+			break;
+		}
+		break;
+	}
+	return 0;
+}
+
+static void build_cpl_pass_accept_req(struct sk_buff *skb, int stid , u8 tos)
+{
+	u32 l2info;
+	u16 vlantag, len, hdr_len, eth_hdr_len;
+	u8 intf;
+	struct cpl_rx_pkt *cpl = cplhdr(skb);
+	struct cpl_pass_accept_req *req;
+	struct tcp_options_received tmp_opt;
+	struct c4iw_dev *dev;
+
+	dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
+	/* Store values from cpl_rx_pkt in temporary location. */
+	vlantag = (__force u16) cpl->vlan;
+	len = (__force u16) cpl->len;
+	l2info  = (__force u32) cpl->l2info;
+	hdr_len = (__force u16) cpl->hdr_len;
+	intf = cpl->iff;
+
+	__skb_pull(skb, sizeof(*req) + sizeof(struct rss_header));
+
+	/*
+	 * We need to parse the TCP options from SYN packet.
+	 * to generate cpl_pass_accept_req.
+	 */
+	memset(&tmp_opt, 0, sizeof(tmp_opt));
+	tcp_clear_options(&tmp_opt);
+	tcp_parse_options(skb, &tmp_opt, 0, NULL);
+
+	req = (struct cpl_pass_accept_req *)__skb_push(skb, sizeof(*req));
+	memset(req, 0, sizeof(*req));
+	req->l2info = cpu_to_be16(V_SYN_INTF(intf) |
+			 V_SYN_MAC_IDX(G_RX_MACIDX(
+			 (__force int) htonl(l2info))) |
+			 F_SYN_XACT_MATCH);
+	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
+			    G_RX_ETHHDR_LEN((__force int) htonl(l2info)) :
+			    G_RX_T5_ETHHDR_LEN((__force int) htonl(l2info));
+	req->hdr_len = cpu_to_be32(V_SYN_RX_CHAN(G_RX_CHAN(
+					(__force int) htonl(l2info))) |
+				   V_TCP_HDR_LEN(G_RX_TCPHDR_LEN(
+					(__force int) htons(hdr_len))) |
+				   V_IP_HDR_LEN(G_RX_IPHDR_LEN(
+					(__force int) htons(hdr_len))) |
+				   V_ETH_HDR_LEN(G_RX_ETHHDR_LEN(eth_hdr_len)));
+	req->vlan = (__force __be16) vlantag;
+	req->len = (__force __be16) len;
+	req->tos_stid = cpu_to_be32(PASS_OPEN_TID(stid) |
+				    PASS_OPEN_TOS(tos));
+	req->tcpopt.mss = htons(tmp_opt.mss_clamp);
+	if (tmp_opt.wscale_ok)
+		req->tcpopt.wsf = tmp_opt.snd_wscale;
+	req->tcpopt.tstamp = tmp_opt.saw_tstamp;
+	if (tmp_opt.sack_ok)
+		req->tcpopt.sack = 1;
+	OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_PASS_ACCEPT_REQ, 0));
+	return;
+}
+
+static void send_fw_pass_open_req(struct c4iw_dev *dev, struct sk_buff *skb,
+				  __be32 laddr, __be16 lport,
+				  __be32 raddr, __be16 rport,
+				  u32 rcv_isn, u32 filter, u16 window,
+				  u32 rss_qid, u8 port_id)
+{
+	struct sk_buff *req_skb;
+	struct fw_ofld_connection_wr *req;
+	struct cpl_pass_accept_req *cpl = cplhdr(skb);
+	int ret;
+
+	req_skb = alloc_skb(sizeof(struct fw_ofld_connection_wr), GFP_KERNEL);
+	req = (struct fw_ofld_connection_wr *)__skb_put(req_skb, sizeof(*req));
+	memset(req, 0, sizeof(*req));
+	req->op_compl = htonl(V_WR_OP(FW_OFLD_CONNECTION_WR) | FW_WR_COMPL(1));
+	req->len16_pkd = htonl(FW_WR_LEN16(DIV_ROUND_UP(sizeof(*req), 16)));
+	req->le.version_cpl = htonl(F_FW_OFLD_CONNECTION_WR_CPL);
+	req->le.filter = (__force __be32) filter;
+	req->le.lport = lport;
+	req->le.pport = rport;
+	req->le.u.ipv4.lip = laddr;
+	req->le.u.ipv4.pip = raddr;
+	req->tcb.rcv_nxt = htonl(rcv_isn + 1);
+	req->tcb.rcv_adv = htons(window);
+	req->tcb.t_state_to_astid =
+		 htonl(V_FW_OFLD_CONNECTION_WR_T_STATE(TCP_SYN_RECV) |
+			V_FW_OFLD_CONNECTION_WR_RCV_SCALE(cpl->tcpopt.wsf) |
+			V_FW_OFLD_CONNECTION_WR_ASTID(
+			GET_PASS_OPEN_TID(ntohl(cpl->tos_stid))));
+
+	/*
+	 * We store the qid in opt2 which will be used by the firmware
+	 * to send us the wr response.
+	 */
+	req->tcb.opt2 = htonl(V_RSS_QUEUE(rss_qid));
+
+	/*
+	 * We initialize the MSS index in TCB to 0xF.
+	 * So that when driver sends cpl_pass_accept_rpl
+	 * TCB picks up the correct value. If this was 0
+	 * TP will ignore any value > 0 for MSS index.
+	 */
+	req->tcb.opt0 = cpu_to_be64(V_MSS_IDX(0xF));
+	req->cookie = (unsigned long)skb;
+
+	set_wr_txq(req_skb, CPL_PRIORITY_CONTROL, port_id);
+	ret = cxgb4_ofld_send(dev->rdev.lldi.ports[0], req_skb);
+	if (ret < 0) {
+		pr_err("%s - cxgb4_ofld_send error %d - dropping\n", __func__,
+		       ret);
+		kfree_skb(skb);
+		kfree_skb(req_skb);
+	}
+}
+
+/*
+ * Handler for CPL_RX_PKT message. Need to handle cpl_rx_pkt
+ * messages when a filter is being used instead of server to
+ * redirect a syn packet. When packets hit filter they are redirected
+ * to the offload queue and driver tries to establish the connection
+ * using firmware work request.
+ */
+static int rx_pkt(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	int stid;
+	unsigned int filter;
+	struct ethhdr *eh = NULL;
+	struct vlan_ethhdr *vlan_eh = NULL;
+	struct iphdr *iph;
+	struct tcphdr *tcph;
+	struct rss_header *rss = (void *)skb->data;
+	struct cpl_rx_pkt *cpl = (void *)skb->data;
+	struct cpl_pass_accept_req *req = (void *)(rss + 1);
+	struct l2t_entry *e;
+	struct dst_entry *dst;
+	struct c4iw_ep *lep;
+	u16 window;
+	struct port_info *pi;
+	struct net_device *pdev;
+	u16 rss_qid, eth_hdr_len;
+	int step;
+	u32 tx_chan;
+	struct neighbour *neigh;
+
+	/* Drop all non-SYN packets */
+	if (!(cpl->l2info & cpu_to_be32(F_RXF_SYN)))
+		goto reject;
+
+	/*
+	 * Drop all packets which did not hit the filter.
+	 * Unlikely to happen.
+	 */
+	if (!(rss->filter_hit && rss->filter_tid))
+		goto reject;
+
+	/*
+	 * Calculate the server tid from filter hit index from cpl_rx_pkt.
+	 */
+	stid = (__force int) cpu_to_be32((__force u32) rss->hash_val);
+
+	lep = (struct c4iw_ep *)lookup_stid(dev->rdev.lldi.tids, stid);
+	if (!lep) {
+		PDBG("%s connect request on invalid stid %d\n", __func__, stid);
+		goto reject;
+	}
+
+	eth_hdr_len = is_t4(dev->rdev.lldi.adapter_type) ?
+			    G_RX_ETHHDR_LEN(htonl(cpl->l2info)) :
+			    G_RX_T5_ETHHDR_LEN(htonl(cpl->l2info));
+	if (eth_hdr_len == ETH_HLEN) {
+		eh = (struct ethhdr *)(req + 1);
+		iph = (struct iphdr *)(eh + 1);
+	} else {
+		vlan_eh = (struct vlan_ethhdr *)(req + 1);
+		iph = (struct iphdr *)(vlan_eh + 1);
+		skb->vlan_tci = ntohs(cpl->vlan);
+	}
+
+	if (iph->version != 0x4)
+		goto reject;
+
+	tcph = (struct tcphdr *)(iph + 1);
+	skb_set_network_header(skb, (void *)iph - (void *)rss);
+	skb_set_transport_header(skb, (void *)tcph - (void *)rss);
+	skb_get(skb);
+
+	PDBG("%s lip 0x%x lport %u pip 0x%x pport %u tos %d\n", __func__,
+	     ntohl(iph->daddr), ntohs(tcph->dest), ntohl(iph->saddr),
+	     ntohs(tcph->source), iph->tos);
+
+	dst = find_route(dev, iph->daddr, iph->saddr, tcph->dest, tcph->source,
+			 iph->tos);
+	if (!dst) {
+		pr_err("%s - failed to find dst entry!\n",
+		       __func__);
+		goto reject;
+	}
+	neigh = dst_neigh_lookup_skb(dst, skb);
+
+	if (!neigh) {
+		pr_err("%s - failed to allocate neigh!\n",
+		       __func__);
+		goto free_dst;
+	}
+
+	if (neigh->dev->flags & IFF_LOOPBACK) {
+		pdev = ip_dev_find(&init_net, iph->daddr);
+		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
+				    pdev, 0);
+		pi = (struct port_info *)netdev_priv(pdev);
+		tx_chan = cxgb4_port_chan(pdev);
+		dev_put(pdev);
+	} else {
+		pdev = get_real_dev(neigh->dev);
+		e = cxgb4_l2t_get(dev->rdev.lldi.l2t, neigh,
+					pdev, 0);
+		pi = (struct port_info *)netdev_priv(pdev);
+		tx_chan = cxgb4_port_chan(pdev);
+	}
+	neigh_release(neigh);
+	if (!e) {
+		pr_err("%s - failed to allocate l2t entry!\n",
+		       __func__);
+		goto free_dst;
+	}
+
+	step = dev->rdev.lldi.nrxq / dev->rdev.lldi.nchan;
+	rss_qid = dev->rdev.lldi.rxq_ids[pi->port_id * step];
+	window = (__force u16) htons((__force u16)tcph->window);
+
+	/* Calcuate filter portion for LE region. */
+	filter = (__force unsigned int) cpu_to_be32(cxgb4_select_ntuple(
+						    dev->rdev.lldi.ports[0],
+						    e));
+
+	/*
+	 * Synthesize the cpl_pass_accept_req. We have everything except the
+	 * TID. Once firmware sends a reply with TID we update the TID field
+	 * in cpl and pass it through the regular cpl_pass_accept_req path.
+	 */
+	build_cpl_pass_accept_req(skb, stid, iph->tos);
+	send_fw_pass_open_req(dev, skb, iph->daddr, tcph->dest, iph->saddr,
+			      tcph->source, ntohl(tcph->seq), filter, window,
+			      rss_qid, pi->port_id);
+	cxgb4_l2t_release(e);
+free_dst:
+	dst_release(dst);
+reject:
+	return 0;
+}
+
+/*
+ * These are the real handlers that are called from a
+ * work queue.
+ */
+static c4iw_handler_func work_handlers[NUM_CPL_CMDS] = {
+	[CPL_ACT_ESTABLISH] = act_establish,
+	[CPL_ACT_OPEN_RPL] = act_open_rpl,
+	[CPL_RX_DATA] = rx_data,
+	[CPL_ABORT_RPL_RSS] = abort_rpl,
+	[CPL_ABORT_RPL] = abort_rpl,
+	[CPL_PASS_OPEN_RPL] = pass_open_rpl,
+	[CPL_CLOSE_LISTSRV_RPL] = close_listsrv_rpl,
+	[CPL_PASS_ACCEPT_REQ] = pass_accept_req,
+	[CPL_PASS_ESTABLISH] = pass_establish,
+	[CPL_PEER_CLOSE] = peer_close,
+	[CPL_ABORT_REQ_RSS] = peer_abort,
+	[CPL_CLOSE_CON_RPL] = close_con_rpl,
+	[CPL_RDMA_TERMINATE] = terminate,
+	[CPL_FW4_ACK] = fw4_ack,
+	[CPL_FW6_MSG] = deferred_fw6_msg,
+	[CPL_RX_PKT] = rx_pkt
+};
+
+static void process_timeout(struct c4iw_ep *ep)
+{
+	struct c4iw_qp_attributes attrs;
+	int abort = 1;
+
+	mutex_lock(&ep->com.mutex);
+	PDBG("%s ep %p tid %u state %d\n", __func__, ep, ep->hwtid,
+	     ep->com.state);
+	set_bit(TIMEDOUT, &ep->com.history);
+	switch (ep->com.state) {
+	case MPA_REQ_SENT:
+		__state_set(&ep->com, ABORTING);
+		connect_reply_upcall(ep, -ETIMEDOUT);
+		break;
+	case MPA_REQ_WAIT:
+		__state_set(&ep->com, ABORTING);
+		break;
+	case CLOSING:
+	case MORIBUND:
+		if (ep->com.cm_id && ep->com.qp) {
+			attrs.next_state = C4IW_QP_STATE_ERROR;
+			c4iw_modify_qp(ep->com.qp->rhp,
+				     ep->com.qp, C4IW_QP_ATTR_NEXT_STATE,
+				     &attrs, 1);
+		}
+		__state_set(&ep->com, ABORTING);
+		close_complete_upcall(ep, -ETIMEDOUT);
+		break;
+	case ABORTING:
+	case DEAD:
+
+		/*
+		 * These states are expected if the ep timed out at the same
+		 * time as another thread was calling stop_ep_timer().
+		 * So we silently do nothing for these states.
+		 */
+		abort = 0;
+		break;
+	default:
+		WARN(1, "%s unexpected state ep %p tid %u state %u\n",
+			__func__, ep, ep->hwtid, ep->com.state);
+		abort = 0;
+	}
+	if (abort)
+		abort_connection(ep, NULL, GFP_KERNEL);
+	mutex_unlock(&ep->com.mutex);
+	c4iw_put_ep(&ep->com);
+}
+
+static void process_timedout_eps(void)
+{
+	struct c4iw_ep *ep;
+
+	spin_lock_irq(&timeout_lock);
+	while (!list_empty(&timeout_list)) {
+		struct list_head *tmp;
+
+		tmp = timeout_list.next;
+		list_del(tmp);
+		tmp->next = NULL;
+		tmp->prev = NULL;
+		spin_unlock_irq(&timeout_lock);
+		ep = list_entry(tmp, struct c4iw_ep, entry);
+		process_timeout(ep);
+		spin_lock_irq(&timeout_lock);
+	}
+	spin_unlock_irq(&timeout_lock);
+}
+
+static void process_work(struct work_struct *work)
+{
+	struct sk_buff *skb = NULL;
+	struct c4iw_dev *dev;
+	struct cpl_act_establish *rpl;
+	unsigned int opcode;
+	int ret;
+
+	process_timedout_eps();
+	while ((skb = skb_dequeue(&rxq))) {
+		rpl = cplhdr(skb);
+		dev = *((struct c4iw_dev **) (skb->cb + sizeof(void *)));
+		opcode = rpl->ot.opcode;
+
+		BUG_ON(!work_handlers[opcode]);
+		ret = work_handlers[opcode](dev, skb);
+		if (!ret)
+			kfree_skb(skb);
+		process_timedout_eps();
+	}
+}
+
+static DECLARE_WORK(skb_work, process_work);
+
+static void ep_timeout(unsigned long arg)
+{
+	struct c4iw_ep *ep = (struct c4iw_ep *)arg;
+	int kickit = 0;
+
+	spin_lock(&timeout_lock);
+	if (!test_and_set_bit(TIMEOUT, &ep->com.flags)) {
+		/*
+		 * Only insert if it is not already on the list.
+		 */
+		if (!ep->entry.next) {
+			list_add_tail(&ep->entry, &timeout_list);
+			kickit = 1;
+		}
+	}
+	spin_unlock(&timeout_lock);
+	if (kickit)
+		queue_work(workq, &skb_work);
+}
+
+/*
+ * All the CM events are handled on a work queue to have a safe context.
+ */
+static int sched(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+
+	/*
+	 * Save dev in the skb->cb area.
+	 */
+	*((struct c4iw_dev **) (skb->cb + sizeof(void *))) = dev;
+
+	/*
+	 * Queue the skb and schedule the worker thread.
+	 */
+	skb_queue_tail(&rxq, skb);
+	queue_work(workq, &skb_work);
+	return 0;
+}
+
+static int set_tcb_rpl(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_set_tcb_rpl *rpl = cplhdr(skb);
+
+	if (rpl->status != CPL_ERR_NONE) {
+		printk(KERN_ERR MOD "Unexpected SET_TCB_RPL status %u "
+		       "for tid %u\n", rpl->status, GET_TID(rpl));
+	}
+	kfree_skb(skb);
+	return 0;
+}
+
+static int fw6_msg(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_fw6_msg *rpl = cplhdr(skb);
+	struct c4iw_wr_wait *wr_waitp;
+	int ret;
+
+	PDBG("%s type %u\n", __func__, rpl->type);
+
+	switch (rpl->type) {
+	case FW6_TYPE_WR_RPL:
+		ret = (int)((be64_to_cpu(rpl->data[0]) >> 8) & 0xff);
+		wr_waitp = (struct c4iw_wr_wait *)(__force unsigned long) rpl->data[1];
+		PDBG("%s wr_waitp %p ret %u\n", __func__, wr_waitp, ret);
+		if (wr_waitp)
+			c4iw_wake_up(wr_waitp, ret ? -ret : 0);
+		kfree_skb(skb);
+		break;
+	case FW6_TYPE_CQE:
+	case FW6_TYPE_OFLD_CONNECTION_WR_RPL:
+		sched(dev, skb);
+		break;
+	default:
+		printk(KERN_ERR MOD "%s unexpected fw6 msg type %u\n", __func__,
+		       rpl->type);
+		kfree_skb(skb);
+		break;
+	}
+	return 0;
+}
+
+static int peer_abort_intr(struct c4iw_dev *dev, struct sk_buff *skb)
+{
+	struct cpl_abort_req_rss *req = cplhdr(skb);
+	struct c4iw_ep *ep;
+	struct tid_info *t = dev->rdev.lldi.tids;
+	unsigned int tid = GET_TID(req);
+
+	ep = lookup_tid(t, tid);
+	if (!ep) {
+		printk(KERN_WARNING MOD
+		       "Abort on non-existent endpoint, tid %d\n", tid);
+		kfree_skb(skb);
+		return 0;
+	}
+	if (is_neg_adv(req->status)) {
+		PDBG("%s neg_adv_abort ep %p tid %u\n", __func__, ep,
+		     ep->hwtid);
+		kfree_skb(skb);
+		return 0;
+	}
+	PDBG("%s ep %p tid %u state %u\n", __func__, ep, ep->hwtid,
+	     ep->com.state);
+
+	/*
+	 * Wake up any threads in rdma_init() or rdma_fini().
+	 * However, if we are on MPAv2 and want to retry with MPAv1
+	 * then, don't wake up yet.
+	 */
+	if (mpa_rev == 2 && !ep->tried_with_mpa_v1) {
+		if (ep->com.state != MPA_REQ_SENT)
+			c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+	} else
+		c4iw_wake_up(&ep->com.wr_wait, -ECONNRESET);
+	sched(dev, skb);
+	return 0;
+}
+
+/*
+ * Most upcalls from the T4 Core go to sched() to
+ * schedule the processing on a work queue.
+ */
+c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS] = {
+	[CPL_ACT_ESTABLISH] = sched,
+	[CPL_ACT_OPEN_RPL] = sched,
+	[CPL_RX_DATA] = sched,
+	[CPL_ABORT_RPL_RSS] = sched,
+	[CPL_ABORT_RPL] = sched,
+	[CPL_PASS_OPEN_RPL] = sched,
+	[CPL_CLOSE_LISTSRV_RPL] = sched,
+	[CPL_PASS_ACCEPT_REQ] = sched,
+	[CPL_PASS_ESTABLISH] = sched,
+	[CPL_PEER_CLOSE] = sched,
+	[CPL_CLOSE_CON_RPL] = sched,
+	[CPL_ABORT_REQ_RSS] = peer_abort_intr,
+	[CPL_RDMA_TERMINATE] = sched,
+	[CPL_FW4_ACK] = sched,
+	[CPL_SET_TCB_RPL] = set_tcb_rpl,
+	[CPL_FW6_MSG] = fw6_msg,
+	[CPL_RX_PKT] = sched
+};
+
+int __init c4iw_cm_init(void)
+{
+	spin_lock_init(&timeout_lock);
+	skb_queue_head_init(&rxq);
+
+	workq = create_singlethread_workqueue("iw_cxgb4");
+	if (!workq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void c4iw_cm_term(void)
+{
+	WARN_ON(!list_empty(&timeout_list));
+	flush_workqueue(workq);
+	destroy_workqueue(workq);
+}
diff --git a/drivers/infiniband/hw/cxgb4/cq.c b/drivers/infiniband/hw/cxgb4/cq.c
new file mode 100644
index 00000000000..c04292c950f
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/cq.c
@@ -0,0 +1,1011 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "iw_cxgb4.h"
+
+static int destroy_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
+		      struct c4iw_dev_ucontext *uctx)
+{
+	struct fw_ri_res_wr *res_wr;
+	struct fw_ri_res *res;
+	int wr_len;
+	struct c4iw_wr_wait wr_wait;
+	struct sk_buff *skb;
+	int ret;
+
+	wr_len = sizeof *res_wr + sizeof *res;
+	skb = alloc_skb(wr_len, GFP_KERNEL);
+	if (!skb)
+		return -ENOMEM;
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+	res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
+	memset(res_wr, 0, wr_len);
+	res_wr->op_nres = cpu_to_be32(
+			FW_WR_OP(FW_RI_RES_WR) |
+			V_FW_RI_RES_WR_NRES(1) |
+			FW_WR_COMPL(1));
+	res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
+	res_wr->cookie = (unsigned long) &wr_wait;
+	res = res_wr->res;
+	res->u.cq.restype = FW_RI_RES_TYPE_CQ;
+	res->u.cq.op = FW_RI_RES_OP_RESET;
+	res->u.cq.iqid = cpu_to_be32(cq->cqid);
+
+	c4iw_init_wr_wait(&wr_wait);
+	ret = c4iw_ofld_send(rdev, skb);
+	if (!ret) {
+		ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
+	}
+
+	kfree(cq->sw_queue);
+	dma_free_coherent(&(rdev->lldi.pdev->dev),
+			  cq->memsize, cq->queue,
+			  dma_unmap_addr(cq, mapping));
+	c4iw_put_cqid(rdev, cq->cqid, uctx);
+	return ret;
+}
+
+static int create_cq(struct c4iw_rdev *rdev, struct t4_cq *cq,
+		     struct c4iw_dev_ucontext *uctx)
+{
+	struct fw_ri_res_wr *res_wr;
+	struct fw_ri_res *res;
+	int wr_len;
+	int user = (uctx != &rdev->uctx);
+	struct c4iw_wr_wait wr_wait;
+	int ret;
+	struct sk_buff *skb;
+
+	cq->cqid = c4iw_get_cqid(rdev, uctx);
+	if (!cq->cqid) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
+	if (!user) {
+		cq->sw_queue = kzalloc(cq->memsize, GFP_KERNEL);
+		if (!cq->sw_queue) {
+			ret = -ENOMEM;
+			goto err2;
+		}
+	}
+	cq->queue = dma_alloc_coherent(&rdev->lldi.pdev->dev, cq->memsize,
+				       &cq->dma_addr, GFP_KERNEL);
+	if (!cq->queue) {
+		ret = -ENOMEM;
+		goto err3;
+	}
+	dma_unmap_addr_set(cq, mapping, cq->dma_addr);
+	memset(cq->queue, 0, cq->memsize);
+
+	/* build fw_ri_res_wr */
+	wr_len = sizeof *res_wr + sizeof *res;
+
+	skb = alloc_skb(wr_len, GFP_KERNEL);
+	if (!skb) {
+		ret = -ENOMEM;
+		goto err4;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+	res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
+	memset(res_wr, 0, wr_len);
+	res_wr->op_nres = cpu_to_be32(
+			FW_WR_OP(FW_RI_RES_WR) |
+			V_FW_RI_RES_WR_NRES(1) |
+			FW_WR_COMPL(1));
+	res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
+	res_wr->cookie = (unsigned long) &wr_wait;
+	res = res_wr->res;
+	res->u.cq.restype = FW_RI_RES_TYPE_CQ;
+	res->u.cq.op = FW_RI_RES_OP_WRITE;
+	res->u.cq.iqid = cpu_to_be32(cq->cqid);
+	res->u.cq.iqandst_to_iqandstindex = cpu_to_be32(
+			V_FW_RI_RES_WR_IQANUS(0) |
+			V_FW_RI_RES_WR_IQANUD(1) |
+			F_FW_RI_RES_WR_IQANDST |
+			V_FW_RI_RES_WR_IQANDSTINDEX(
+				rdev->lldi.ciq_ids[cq->vector]));
+	res->u.cq.iqdroprss_to_iqesize = cpu_to_be16(
+			F_FW_RI_RES_WR_IQDROPRSS |
+			V_FW_RI_RES_WR_IQPCIECH(2) |
+			V_FW_RI_RES_WR_IQINTCNTTHRESH(0) |
+			F_FW_RI_RES_WR_IQO |
+			V_FW_RI_RES_WR_IQESIZE(1));
+	res->u.cq.iqsize = cpu_to_be16(cq->size);
+	res->u.cq.iqaddr = cpu_to_be64(cq->dma_addr);
+
+	c4iw_init_wr_wait(&wr_wait);
+
+	ret = c4iw_ofld_send(rdev, skb);
+	if (ret)
+		goto err4;
+	PDBG("%s wait_event wr_wait %p\n", __func__, &wr_wait);
+	ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
+	if (ret)
+		goto err4;
+
+	cq->gen = 1;
+	cq->gts = rdev->lldi.gts_reg;
+	cq->rdev = rdev;
+	if (user) {
+		cq->ugts = (u64)pci_resource_start(rdev->lldi.pdev, 2) +
+					(cq->cqid << rdev->cqshift);
+		cq->ugts &= PAGE_MASK;
+	}
+	return 0;
+err4:
+	dma_free_coherent(&rdev->lldi.pdev->dev, cq->memsize, cq->queue,
+			  dma_unmap_addr(cq, mapping));
+err3:
+	kfree(cq->sw_queue);
+err2:
+	c4iw_put_cqid(rdev, cq->cqid, uctx);
+err1:
+	return ret;
+}
+
+static void insert_recv_cqe(struct t4_wq *wq, struct t4_cq *cq)
+{
+	struct t4_cqe cqe;
+
+	PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
+	     wq, cq, cq->sw_cidx, cq->sw_pidx);
+	memset(&cqe, 0, sizeof(cqe));
+	cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) |
+				 V_CQE_OPCODE(FW_RI_SEND) |
+				 V_CQE_TYPE(0) |
+				 V_CQE_SWCQE(1) |
+				 V_CQE_QPID(wq->sq.qid));
+	cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
+	cq->sw_queue[cq->sw_pidx] = cqe;
+	t4_swcq_produce(cq);
+}
+
+int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count)
+{
+	int flushed = 0;
+	int in_use = wq->rq.in_use - count;
+
+	BUG_ON(in_use < 0);
+	PDBG("%s wq %p cq %p rq.in_use %u skip count %u\n", __func__,
+	     wq, cq, wq->rq.in_use, count);
+	while (in_use--) {
+		insert_recv_cqe(wq, cq);
+		flushed++;
+	}
+	return flushed;
+}
+
+static void insert_sq_cqe(struct t4_wq *wq, struct t4_cq *cq,
+			  struct t4_swsqe *swcqe)
+{
+	struct t4_cqe cqe;
+
+	PDBG("%s wq %p cq %p sw_cidx %u sw_pidx %u\n", __func__,
+	     wq, cq, cq->sw_cidx, cq->sw_pidx);
+	memset(&cqe, 0, sizeof(cqe));
+	cqe.header = cpu_to_be32(V_CQE_STATUS(T4_ERR_SWFLUSH) |
+				 V_CQE_OPCODE(swcqe->opcode) |
+				 V_CQE_TYPE(1) |
+				 V_CQE_SWCQE(1) |
+				 V_CQE_QPID(wq->sq.qid));
+	CQE_WRID_SQ_IDX(&cqe) = swcqe->idx;
+	cqe.bits_type_ts = cpu_to_be64(V_CQE_GENBIT((u64)cq->gen));
+	cq->sw_queue[cq->sw_pidx] = cqe;
+	t4_swcq_produce(cq);
+}
+
+static void advance_oldest_read(struct t4_wq *wq);
+
+int c4iw_flush_sq(struct c4iw_qp *qhp)
+{
+	int flushed = 0;
+	struct t4_wq *wq = &qhp->wq;
+	struct c4iw_cq *chp = to_c4iw_cq(qhp->ibqp.send_cq);
+	struct t4_cq *cq = &chp->cq;
+	int idx;
+	struct t4_swsqe *swsqe;
+
+	if (wq->sq.flush_cidx == -1)
+		wq->sq.flush_cidx = wq->sq.cidx;
+	idx = wq->sq.flush_cidx;
+	BUG_ON(idx >= wq->sq.size);
+	while (idx != wq->sq.pidx) {
+		swsqe = &wq->sq.sw_sq[idx];
+		BUG_ON(swsqe->flushed);
+		swsqe->flushed = 1;
+		insert_sq_cqe(wq, cq, swsqe);
+		if (wq->sq.oldest_read == swsqe) {
+			BUG_ON(swsqe->opcode != FW_RI_READ_REQ);
+			advance_oldest_read(wq);
+		}
+		flushed++;
+		if (++idx == wq->sq.size)
+			idx = 0;
+	}
+	wq->sq.flush_cidx += flushed;
+	if (wq->sq.flush_cidx >= wq->sq.size)
+		wq->sq.flush_cidx -= wq->sq.size;
+	return flushed;
+}
+
+static void flush_completed_wrs(struct t4_wq *wq, struct t4_cq *cq)
+{
+	struct t4_swsqe *swsqe;
+	int cidx;
+
+	if (wq->sq.flush_cidx == -1)
+		wq->sq.flush_cidx = wq->sq.cidx;
+	cidx = wq->sq.flush_cidx;
+	BUG_ON(cidx > wq->sq.size);
+
+	while (cidx != wq->sq.pidx) {
+		swsqe = &wq->sq.sw_sq[cidx];
+		if (!swsqe->signaled) {
+			if (++cidx == wq->sq.size)
+				cidx = 0;
+		} else if (swsqe->complete) {
+
+			BUG_ON(swsqe->flushed);
+
+			/*
+			 * Insert this completed cqe into the swcq.
+			 */
+			PDBG("%s moving cqe into swcq sq idx %u cq idx %u\n",
+					__func__, cidx, cq->sw_pidx);
+			swsqe->cqe.header |= htonl(V_CQE_SWCQE(1));
+			cq->sw_queue[cq->sw_pidx] = swsqe->cqe;
+			t4_swcq_produce(cq);
+			swsqe->flushed = 1;
+			if (++cidx == wq->sq.size)
+				cidx = 0;
+			wq->sq.flush_cidx = cidx;
+		} else
+			break;
+	}
+}
+
+static void create_read_req_cqe(struct t4_wq *wq, struct t4_cqe *hw_cqe,
+		struct t4_cqe *read_cqe)
+{
+	read_cqe->u.scqe.cidx = wq->sq.oldest_read->idx;
+	read_cqe->len = htonl(wq->sq.oldest_read->read_len);
+	read_cqe->header = htonl(V_CQE_QPID(CQE_QPID(hw_cqe)) |
+			V_CQE_SWCQE(SW_CQE(hw_cqe)) |
+			V_CQE_OPCODE(FW_RI_READ_REQ) |
+			V_CQE_TYPE(1));
+	read_cqe->bits_type_ts = hw_cqe->bits_type_ts;
+}
+
+static void advance_oldest_read(struct t4_wq *wq)
+{
+
+	u32 rptr = wq->sq.oldest_read - wq->sq.sw_sq + 1;
+
+	if (rptr == wq->sq.size)
+		rptr = 0;
+	while (rptr != wq->sq.pidx) {
+		wq->sq.oldest_read = &wq->sq.sw_sq[rptr];
+
+		if (wq->sq.oldest_read->opcode == FW_RI_READ_REQ)
+			return;
+		if (++rptr == wq->sq.size)
+			rptr = 0;
+	}
+	wq->sq.oldest_read = NULL;
+}
+
+/*
+ * Move all CQEs from the HWCQ into the SWCQ.
+ * Deal with out-of-order and/or completions that complete
+ * prior unsignalled WRs.
+ */
+void c4iw_flush_hw_cq(struct c4iw_cq *chp)
+{
+	struct t4_cqe *hw_cqe, *swcqe, read_cqe;
+	struct c4iw_qp *qhp;
+	struct t4_swsqe *swsqe;
+	int ret;
+
+	PDBG("%s  cqid 0x%x\n", __func__, chp->cq.cqid);
+	ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
+
+	/*
+	 * This logic is similar to poll_cq(), but not quite the same
+	 * unfortunately.  Need to move pertinent HW CQEs to the SW CQ but
+	 * also do any translation magic that poll_cq() normally does.
+	 */
+	while (!ret) {
+		qhp = get_qhp(chp->rhp, CQE_QPID(hw_cqe));
+
+		/*
+		 * drop CQEs with no associated QP
+		 */
+		if (qhp == NULL)
+			goto next_cqe;
+
+		if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE)
+			goto next_cqe;
+
+		if (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP) {
+
+			/* If we have reached here because of async
+			 * event or other error, and have egress error
+			 * then drop
+			 */
+			if (CQE_TYPE(hw_cqe) == 1)
+				goto next_cqe;
+
+			/* drop peer2peer RTR reads.
+			 */
+			if (CQE_WRID_STAG(hw_cqe) == 1)
+				goto next_cqe;
+
+			/*
+			 * Eat completions for unsignaled read WRs.
+			 */
+			if (!qhp->wq.sq.oldest_read->signaled) {
+				advance_oldest_read(&qhp->wq);
+				goto next_cqe;
+			}
+
+			/*
+			 * Don't write to the HWCQ, create a new read req CQE
+			 * in local memory and move it into the swcq.
+			 */
+			create_read_req_cqe(&qhp->wq, hw_cqe, &read_cqe);
+			hw_cqe = &read_cqe;
+			advance_oldest_read(&qhp->wq);
+		}
+
+		/* if its a SQ completion, then do the magic to move all the
+		 * unsignaled and now in-order completions into the swcq.
+		 */
+		if (SQ_TYPE(hw_cqe)) {
+			swsqe = &qhp->wq.sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
+			swsqe->cqe = *hw_cqe;
+			swsqe->complete = 1;
+			flush_completed_wrs(&qhp->wq, &chp->cq);
+		} else {
+			swcqe = &chp->cq.sw_queue[chp->cq.sw_pidx];
+			*swcqe = *hw_cqe;
+			swcqe->header |= cpu_to_be32(V_CQE_SWCQE(1));
+			t4_swcq_produce(&chp->cq);
+		}
+next_cqe:
+		t4_hwcq_consume(&chp->cq);
+		ret = t4_next_hw_cqe(&chp->cq, &hw_cqe);
+	}
+}
+
+static int cqe_completes_wr(struct t4_cqe *cqe, struct t4_wq *wq)
+{
+	if (CQE_OPCODE(cqe) == FW_RI_TERMINATE)
+		return 0;
+
+	if ((CQE_OPCODE(cqe) == FW_RI_RDMA_WRITE) && RQ_TYPE(cqe))
+		return 0;
+
+	if ((CQE_OPCODE(cqe) == FW_RI_READ_RESP) && SQ_TYPE(cqe))
+		return 0;
+
+	if (CQE_SEND_OPCODE(cqe) && RQ_TYPE(cqe) && t4_rq_empty(wq))
+		return 0;
+	return 1;
+}
+
+void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count)
+{
+	struct t4_cqe *cqe;
+	u32 ptr;
+
+	*count = 0;
+	PDBG("%s count zero %d\n", __func__, *count);
+	ptr = cq->sw_cidx;
+	while (ptr != cq->sw_pidx) {
+		cqe = &cq->sw_queue[ptr];
+		if (RQ_TYPE(cqe) && (CQE_OPCODE(cqe) != FW_RI_READ_RESP) &&
+		    (CQE_QPID(cqe) == wq->sq.qid) && cqe_completes_wr(cqe, wq))
+			(*count)++;
+		if (++ptr == cq->size)
+			ptr = 0;
+	}
+	PDBG("%s cq %p count %d\n", __func__, cq, *count);
+}
+
+/*
+ * poll_cq
+ *
+ * Caller must:
+ *     check the validity of the first CQE,
+ *     supply the wq assicated with the qpid.
+ *
+ * credit: cq credit to return to sge.
+ * cqe_flushed: 1 iff the CQE is flushed.
+ * cqe: copy of the polled CQE.
+ *
+ * return value:
+ *    0		    CQE returned ok.
+ *    -EAGAIN       CQE skipped, try again.
+ *    -EOVERFLOW    CQ overflow detected.
+ */
+static int poll_cq(struct t4_wq *wq, struct t4_cq *cq, struct t4_cqe *cqe,
+		   u8 *cqe_flushed, u64 *cookie, u32 *credit)
+{
+	int ret = 0;
+	struct t4_cqe *hw_cqe, read_cqe;
+
+	*cqe_flushed = 0;
+	*credit = 0;
+	ret = t4_next_cqe(cq, &hw_cqe);
+	if (ret)
+		return ret;
+
+	PDBG("%s CQE OVF %u qpid 0x%0x genbit %u type %u status 0x%0x"
+	     " opcode 0x%0x len 0x%0x wrid_hi_stag 0x%x wrid_low_msn 0x%x\n",
+	     __func__, CQE_OVFBIT(hw_cqe), CQE_QPID(hw_cqe),
+	     CQE_GENBIT(hw_cqe), CQE_TYPE(hw_cqe), CQE_STATUS(hw_cqe),
+	     CQE_OPCODE(hw_cqe), CQE_LEN(hw_cqe), CQE_WRID_HI(hw_cqe),
+	     CQE_WRID_LOW(hw_cqe));
+
+	/*
+	 * skip cqe's not affiliated with a QP.
+	 */
+	if (wq == NULL) {
+		ret = -EAGAIN;
+		goto skip_cqe;
+	}
+
+	/*
+	* skip hw cqe's if the wq is flushed.
+	*/
+	if (wq->flushed && !SW_CQE(hw_cqe)) {
+		ret = -EAGAIN;
+		goto skip_cqe;
+	}
+
+	/*
+	 * skip TERMINATE cqes...
+	 */
+	if (CQE_OPCODE(hw_cqe) == FW_RI_TERMINATE) {
+		ret = -EAGAIN;
+		goto skip_cqe;
+	}
+
+	/*
+	 * Gotta tweak READ completions:
+	 *	1) the cqe doesn't contain the sq_wptr from the wr.
+	 *	2) opcode not reflected from the wr.
+	 *	3) read_len not reflected from the wr.
+	 *	4) cq_type is RQ_TYPE not SQ_TYPE.
+	 */
+	if (RQ_TYPE(hw_cqe) && (CQE_OPCODE(hw_cqe) == FW_RI_READ_RESP)) {
+
+		/* If we have reached here because of async
+		 * event or other error, and have egress error
+		 * then drop
+		 */
+		if (CQE_TYPE(hw_cqe) == 1) {
+			if (CQE_STATUS(hw_cqe))
+				t4_set_wq_in_error(wq);
+			ret = -EAGAIN;
+			goto skip_cqe;
+		}
+
+		/* If this is an unsolicited read response, then the read
+		 * was generated by the kernel driver as part of peer-2-peer
+		 * connection setup.  So ignore the completion.
+		 */
+		if (CQE_WRID_STAG(hw_cqe) == 1) {
+			if (CQE_STATUS(hw_cqe))
+				t4_set_wq_in_error(wq);
+			ret = -EAGAIN;
+			goto skip_cqe;
+		}
+
+		/*
+		 * Eat completions for unsignaled read WRs.
+		 */
+		if (!wq->sq.oldest_read->signaled) {
+			advance_oldest_read(wq);
+			ret = -EAGAIN;
+			goto skip_cqe;
+		}
+
+		/*
+		 * Don't write to the HWCQ, so create a new read req CQE
+		 * in local memory.
+		 */
+		create_read_req_cqe(wq, hw_cqe, &read_cqe);
+		hw_cqe = &read_cqe;
+		advance_oldest_read(wq);
+	}
+
+	if (CQE_STATUS(hw_cqe) || t4_wq_in_error(wq)) {
+		*cqe_flushed = (CQE_STATUS(hw_cqe) == T4_ERR_SWFLUSH);
+		t4_set_wq_in_error(wq);
+	}
+
+	/*
+	 * RECV completion.
+	 */
+	if (RQ_TYPE(hw_cqe)) {
+
+		/*
+		 * HW only validates 4 bits of MSN.  So we must validate that
+		 * the MSN in the SEND is the next expected MSN.  If its not,
+		 * then we complete this with T4_ERR_MSN and mark the wq in
+		 * error.
+		 */
+
+		if (t4_rq_empty(wq)) {
+			t4_set_wq_in_error(wq);
+			ret = -EAGAIN;
+			goto skip_cqe;
+		}
+		if (unlikely((CQE_WRID_MSN(hw_cqe) != (wq->rq.msn)))) {
+			t4_set_wq_in_error(wq);
+			hw_cqe->header |= htonl(V_CQE_STATUS(T4_ERR_MSN));
+			goto proc_cqe;
+		}
+		goto proc_cqe;
+	}
+
+	/*
+	 * If we get here its a send completion.
+	 *
+	 * Handle out of order completion. These get stuffed
+	 * in the SW SQ. Then the SW SQ is walked to move any
+	 * now in-order completions into the SW CQ.  This handles
+	 * 2 cases:
+	 *	1) reaping unsignaled WRs when the first subsequent
+	 *	   signaled WR is completed.
+	 *	2) out of order read completions.
+	 */
+	if (!SW_CQE(hw_cqe) && (CQE_WRID_SQ_IDX(hw_cqe) != wq->sq.cidx)) {
+		struct t4_swsqe *swsqe;
+
+		PDBG("%s out of order completion going in sw_sq at idx %u\n",
+		     __func__, CQE_WRID_SQ_IDX(hw_cqe));
+		swsqe = &wq->sq.sw_sq[CQE_WRID_SQ_IDX(hw_cqe)];
+		swsqe->cqe = *hw_cqe;
+		swsqe->complete = 1;
+		ret = -EAGAIN;
+		goto flush_wq;
+	}
+
+proc_cqe:
+	*cqe = *hw_cqe;
+
+	/*
+	 * Reap the associated WR(s) that are freed up with this
+	 * completion.
+	 */
+	if (SQ_TYPE(hw_cqe)) {
+		int idx = CQE_WRID_SQ_IDX(hw_cqe);
+		BUG_ON(idx >= wq->sq.size);
+
+		/*
+		* Account for any unsignaled completions completed by
+		* this signaled completion.  In this case, cidx points
+		* to the first unsignaled one, and idx points to the
+		* signaled one.  So adjust in_use based on this delta.
+		* if this is not completing any unsigned wrs, then the
+		* delta will be 0. Handle wrapping also!
+		*/
+		if (idx < wq->sq.cidx)
+			wq->sq.in_use -= wq->sq.size + idx - wq->sq.cidx;
+		else
+			wq->sq.in_use -= idx - wq->sq.cidx;
+		BUG_ON(wq->sq.in_use <= 0 && wq->sq.in_use >= wq->sq.size);
+
+		wq->sq.cidx = (uint16_t)idx;
+		PDBG("%s completing sq idx %u\n", __func__, wq->sq.cidx);
+		*cookie = wq->sq.sw_sq[wq->sq.cidx].wr_id;
+		t4_sq_consume(wq);
+	} else {
+		PDBG("%s completing rq idx %u\n", __func__, wq->rq.cidx);
+		*cookie = wq->rq.sw_rq[wq->rq.cidx].wr_id;
+		BUG_ON(t4_rq_empty(wq));
+		t4_rq_consume(wq);
+		goto skip_cqe;
+	}
+
+flush_wq:
+	/*
+	 * Flush any completed cqes that are now in-order.
+	 */
+	flush_completed_wrs(wq, cq);
+
+skip_cqe:
+	if (SW_CQE(hw_cqe)) {
+		PDBG("%s cq %p cqid 0x%x skip sw cqe cidx %u\n",
+		     __func__, cq, cq->cqid, cq->sw_cidx);
+		t4_swcq_consume(cq);
+	} else {
+		PDBG("%s cq %p cqid 0x%x skip hw cqe cidx %u\n",
+		     __func__, cq, cq->cqid, cq->cidx);
+		t4_hwcq_consume(cq);
+	}
+	return ret;
+}
+
+/*
+ * Get one cq entry from c4iw and map it to openib.
+ *
+ * Returns:
+ *	0			cqe returned
+ *	-ENODATA		EMPTY;
+ *	-EAGAIN			caller must try again
+ *	any other -errno	fatal error
+ */
+static int c4iw_poll_cq_one(struct c4iw_cq *chp, struct ib_wc *wc)
+{
+	struct c4iw_qp *qhp = NULL;
+	struct t4_cqe uninitialized_var(cqe), *rd_cqe;
+	struct t4_wq *wq;
+	u32 credit = 0;
+	u8 cqe_flushed;
+	u64 cookie = 0;
+	int ret;
+
+	ret = t4_next_cqe(&chp->cq, &rd_cqe);
+
+	if (ret)
+		return ret;
+
+	qhp = get_qhp(chp->rhp, CQE_QPID(rd_cqe));
+	if (!qhp)
+		wq = NULL;
+	else {
+		spin_lock(&qhp->lock);
+		wq = &(qhp->wq);
+	}
+	ret = poll_cq(wq, &(chp->cq), &cqe, &cqe_flushed, &cookie, &credit);
+	if (ret)
+		goto out;
+
+	wc->wr_id = cookie;
+	wc->qp = &qhp->ibqp;
+	wc->vendor_err = CQE_STATUS(&cqe);
+	wc->wc_flags = 0;
+
+	PDBG("%s qpid 0x%x type %d opcode %d status 0x%x len %u wrid hi 0x%x "
+	     "lo 0x%x cookie 0x%llx\n", __func__, CQE_QPID(&cqe),
+	     CQE_TYPE(&cqe), CQE_OPCODE(&cqe), CQE_STATUS(&cqe), CQE_LEN(&cqe),
+	     CQE_WRID_HI(&cqe), CQE_WRID_LOW(&cqe), (unsigned long long)cookie);
+
+	if (CQE_TYPE(&cqe) == 0) {
+		if (!CQE_STATUS(&cqe))
+			wc->byte_len = CQE_LEN(&cqe);
+		else
+			wc->byte_len = 0;
+		wc->opcode = IB_WC_RECV;
+		if (CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_INV ||
+		    CQE_OPCODE(&cqe) == FW_RI_SEND_WITH_SE_INV) {
+			wc->ex.invalidate_rkey = CQE_WRID_STAG(&cqe);
+			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
+		}
+	} else {
+		switch (CQE_OPCODE(&cqe)) {
+		case FW_RI_RDMA_WRITE:
+			wc->opcode = IB_WC_RDMA_WRITE;
+			break;
+		case FW_RI_READ_REQ:
+			wc->opcode = IB_WC_RDMA_READ;
+			wc->byte_len = CQE_LEN(&cqe);
+			break;
+		case FW_RI_SEND_WITH_INV:
+		case FW_RI_SEND_WITH_SE_INV:
+			wc->opcode = IB_WC_SEND;
+			wc->wc_flags |= IB_WC_WITH_INVALIDATE;
+			break;
+		case FW_RI_SEND:
+		case FW_RI_SEND_WITH_SE:
+			wc->opcode = IB_WC_SEND;
+			break;
+		case FW_RI_BIND_MW:
+			wc->opcode = IB_WC_BIND_MW;
+			break;
+
+		case FW_RI_LOCAL_INV:
+			wc->opcode = IB_WC_LOCAL_INV;
+			break;
+		case FW_RI_FAST_REGISTER:
+			wc->opcode = IB_WC_FAST_REG_MR;
+			break;
+		default:
+			printk(KERN_ERR MOD "Unexpected opcode %d "
+			       "in the CQE received for QPID=0x%0x\n",
+			       CQE_OPCODE(&cqe), CQE_QPID(&cqe));
+			ret = -EINVAL;
+			goto out;
+		}
+	}
+
+	if (cqe_flushed)
+		wc->status = IB_WC_WR_FLUSH_ERR;
+	else {
+
+		switch (CQE_STATUS(&cqe)) {
+		case T4_ERR_SUCCESS:
+			wc->status = IB_WC_SUCCESS;
+			break;
+		case T4_ERR_STAG:
+			wc->status = IB_WC_LOC_ACCESS_ERR;
+			break;
+		case T4_ERR_PDID:
+			wc->status = IB_WC_LOC_PROT_ERR;
+			break;
+		case T4_ERR_QPID:
+		case T4_ERR_ACCESS:
+			wc->status = IB_WC_LOC_ACCESS_ERR;
+			break;
+		case T4_ERR_WRAP:
+			wc->status = IB_WC_GENERAL_ERR;
+			break;
+		case T4_ERR_BOUND:
+			wc->status = IB_WC_LOC_LEN_ERR;
+			break;
+		case T4_ERR_INVALIDATE_SHARED_MR:
+		case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
+			wc->status = IB_WC_MW_BIND_ERR;
+			break;
+		case T4_ERR_CRC:
+		case T4_ERR_MARKER:
+		case T4_ERR_PDU_LEN_ERR:
+		case T4_ERR_OUT_OF_RQE:
+		case T4_ERR_DDP_VERSION:
+		case T4_ERR_RDMA_VERSION:
+		case T4_ERR_DDP_QUEUE_NUM:
+		case T4_ERR_MSN:
+		case T4_ERR_TBIT:
+		case T4_ERR_MO:
+		case T4_ERR_MSN_RANGE:
+		case T4_ERR_IRD_OVERFLOW:
+		case T4_ERR_OPCODE:
+		case T4_ERR_INTERNAL_ERR:
+			wc->status = IB_WC_FATAL_ERR;
+			break;
+		case T4_ERR_SWFLUSH:
+			wc->status = IB_WC_WR_FLUSH_ERR;
+			break;
+		default:
+			printk(KERN_ERR MOD
+			       "Unexpected cqe_status 0x%x for QPID=0x%0x\n",
+			       CQE_STATUS(&cqe), CQE_QPID(&cqe));
+			ret = -EINVAL;
+		}
+	}
+out:
+	if (wq)
+		spin_unlock(&qhp->lock);
+	return ret;
+}
+
+int c4iw_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	struct c4iw_cq *chp;
+	unsigned long flags;
+	int npolled;
+	int err = 0;
+
+	chp = to_c4iw_cq(ibcq);
+
+	spin_lock_irqsave(&chp->lock, flags);
+	for (npolled = 0; npolled < num_entries; ++npolled) {
+		do {
+			err = c4iw_poll_cq_one(chp, wc + npolled);
+		} while (err == -EAGAIN);
+		if (err)
+			break;
+	}
+	spin_unlock_irqrestore(&chp->lock, flags);
+	return !err || err == -ENODATA ? npolled : err;
+}
+
+int c4iw_destroy_cq(struct ib_cq *ib_cq)
+{
+	struct c4iw_cq *chp;
+	struct c4iw_ucontext *ucontext;
+
+	PDBG("%s ib_cq %p\n", __func__, ib_cq);
+	chp = to_c4iw_cq(ib_cq);
+
+	remove_handle(chp->rhp, &chp->rhp->cqidr, chp->cq.cqid);
+	atomic_dec(&chp->refcnt);
+	wait_event(chp->wait, !atomic_read(&chp->refcnt));
+
+	ucontext = ib_cq->uobject ? to_c4iw_ucontext(ib_cq->uobject->context)
+				  : NULL;
+	destroy_cq(&chp->rhp->rdev, &chp->cq,
+		   ucontext ? &ucontext->uctx : &chp->cq.rdev->uctx);
+	kfree(chp);
+	return 0;
+}
+
+struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
+			     int vector, struct ib_ucontext *ib_context,
+			     struct ib_udata *udata)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_cq *chp;
+	struct c4iw_create_cq_resp uresp;
+	struct c4iw_ucontext *ucontext = NULL;
+	int ret;
+	size_t memsize, hwentries;
+	struct c4iw_mm_entry *mm, *mm2;
+
+	PDBG("%s ib_dev %p entries %d\n", __func__, ibdev, entries);
+
+	rhp = to_c4iw_dev(ibdev);
+
+	if (vector >= rhp->rdev.lldi.nciq)
+		return ERR_PTR(-EINVAL);
+
+	chp = kzalloc(sizeof(*chp), GFP_KERNEL);
+	if (!chp)
+		return ERR_PTR(-ENOMEM);
+
+	if (ib_context)
+		ucontext = to_c4iw_ucontext(ib_context);
+
+	/* account for the status page. */
+	entries++;
+
+	/* IQ needs one extra entry to differentiate full vs empty. */
+	entries++;
+
+	/*
+	 * entries must be multiple of 16 for HW.
+	 */
+	entries = roundup(entries, 16);
+
+	/*
+	 * Make actual HW queue 2x to avoid cdix_inc overflows.
+	 */
+	hwentries = min(entries * 2, T4_MAX_IQ_SIZE);
+
+	/*
+	 * Make HW queue at least 64 entries so GTS updates aren't too
+	 * frequent.
+	 */
+	if (hwentries < 64)
+		hwentries = 64;
+
+	memsize = hwentries * sizeof *chp->cq.queue;
+
+	/*
+	 * memsize must be a multiple of the page size if its a user cq.
+	 */
+	if (ucontext) {
+		memsize = roundup(memsize, PAGE_SIZE);
+		hwentries = memsize / sizeof *chp->cq.queue;
+		while (hwentries > T4_MAX_IQ_SIZE) {
+			memsize -= PAGE_SIZE;
+			hwentries = memsize / sizeof *chp->cq.queue;
+		}
+	}
+	chp->cq.size = hwentries;
+	chp->cq.memsize = memsize;
+	chp->cq.vector = vector;
+
+	ret = create_cq(&rhp->rdev, &chp->cq,
+			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+	if (ret)
+		goto err1;
+
+	chp->rhp = rhp;
+	chp->cq.size--;				/* status page */
+	chp->ibcq.cqe = entries - 2;
+	spin_lock_init(&chp->lock);
+	spin_lock_init(&chp->comp_handler_lock);
+	atomic_set(&chp->refcnt, 1);
+	init_waitqueue_head(&chp->wait);
+	ret = insert_handle(rhp, &rhp->cqidr, chp, chp->cq.cqid);
+	if (ret)
+		goto err2;
+
+	if (ucontext) {
+		mm = kmalloc(sizeof *mm, GFP_KERNEL);
+		if (!mm)
+			goto err3;
+		mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
+		if (!mm2)
+			goto err4;
+
+		uresp.qid_mask = rhp->rdev.cqmask;
+		uresp.cqid = chp->cq.cqid;
+		uresp.size = chp->cq.size;
+		uresp.memsize = chp->cq.memsize;
+		spin_lock(&ucontext->mmap_lock);
+		uresp.key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		uresp.gts_key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		spin_unlock(&ucontext->mmap_lock);
+		ret = ib_copy_to_udata(udata, &uresp,
+				       sizeof(uresp) - sizeof(uresp.reserved));
+		if (ret)
+			goto err5;
+
+		mm->key = uresp.key;
+		mm->addr = virt_to_phys(chp->cq.queue);
+		mm->len = chp->cq.memsize;
+		insert_mmap(ucontext, mm);
+
+		mm2->key = uresp.gts_key;
+		mm2->addr = chp->cq.ugts;
+		mm2->len = PAGE_SIZE;
+		insert_mmap(ucontext, mm2);
+	}
+	PDBG("%s cqid 0x%0x chp %p size %u memsize %zu, dma_addr 0x%0llx\n",
+	     __func__, chp->cq.cqid, chp, chp->cq.size,
+	     chp->cq.memsize,
+	     (unsigned long long) chp->cq.dma_addr);
+	return &chp->ibcq;
+err5:
+	kfree(mm2);
+err4:
+	kfree(mm);
+err3:
+	remove_handle(rhp, &rhp->cqidr, chp->cq.cqid);
+err2:
+	destroy_cq(&chp->rhp->rdev, &chp->cq,
+		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+err1:
+	kfree(chp);
+	return ERR_PTR(ret);
+}
+
+int c4iw_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata)
+{
+	return -ENOSYS;
+}
+
+int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
+{
+	struct c4iw_cq *chp;
+	int ret;
+	unsigned long flag;
+
+	chp = to_c4iw_cq(ibcq);
+	spin_lock_irqsave(&chp->lock, flag);
+	ret = t4_arm_cq(&chp->cq,
+			(flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED);
+	spin_unlock_irqrestore(&chp->lock, flag);
+	if (ret && !(flags & IB_CQ_REPORT_MISSED_EVENTS))
+		ret = 0;
+	return ret;
+}
diff --git a/drivers/infiniband/hw/cxgb4/device.c b/drivers/infiniband/hw/cxgb4/device.c
new file mode 100644
index 00000000000..7db82b24302
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/device.c
@@ -0,0 +1,1362 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/debugfs.h>
+#include <linux/vmalloc.h>
+
+#include <rdma/ib_verbs.h>
+
+#include "iw_cxgb4.h"
+
+#define DRV_VERSION "0.1"
+
+MODULE_AUTHOR("Steve Wise");
+MODULE_DESCRIPTION("Chelsio T4/T5 RDMA Driver");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRV_VERSION);
+
+static int allow_db_fc_on_t5;
+module_param(allow_db_fc_on_t5, int, 0644);
+MODULE_PARM_DESC(allow_db_fc_on_t5,
+		 "Allow DB Flow Control on T5 (default = 0)");
+
+static int allow_db_coalescing_on_t5;
+module_param(allow_db_coalescing_on_t5, int, 0644);
+MODULE_PARM_DESC(allow_db_coalescing_on_t5,
+		 "Allow DB Coalescing on T5 (default = 0)");
+
+struct uld_ctx {
+	struct list_head entry;
+	struct cxgb4_lld_info lldi;
+	struct c4iw_dev *dev;
+};
+
+static LIST_HEAD(uld_ctx_list);
+static DEFINE_MUTEX(dev_mutex);
+
+#define DB_FC_RESUME_SIZE 64
+#define DB_FC_RESUME_DELAY 1
+#define DB_FC_DRAIN_THRESH 0
+
+static struct dentry *c4iw_debugfs_root;
+
+struct c4iw_debugfs_data {
+	struct c4iw_dev *devp;
+	char *buf;
+	int bufsize;
+	int pos;
+};
+
+/* registered cxgb4 netlink callbacks */
+static struct ibnl_client_cbs c4iw_nl_cb_table[] = {
+	[RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
+	[RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
+	[RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
+	[RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
+	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
+	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
+};
+
+static int count_idrs(int id, void *p, void *data)
+{
+	int *countp = data;
+
+	*countp = *countp + 1;
+	return 0;
+}
+
+static ssize_t debugfs_read(struct file *file, char __user *buf, size_t count,
+			    loff_t *ppos)
+{
+	struct c4iw_debugfs_data *d = file->private_data;
+
+	return simple_read_from_buffer(buf, count, ppos, d->buf, d->pos);
+}
+
+static int dump_qp(int id, void *p, void *data)
+{
+	struct c4iw_qp *qp = p;
+	struct c4iw_debugfs_data *qpd = data;
+	int space;
+	int cc;
+
+	if (id != qp->wq.sq.qid)
+		return 0;
+
+	space = qpd->bufsize - qpd->pos - 1;
+	if (space == 0)
+		return 1;
+
+	if (qp->ep) {
+		if (qp->ep->com.local_addr.ss_family == AF_INET) {
+			struct sockaddr_in *lsin = (struct sockaddr_in *)
+				&qp->ep->com.local_addr;
+			struct sockaddr_in *rsin = (struct sockaddr_in *)
+				&qp->ep->com.remote_addr;
+			struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
+				&qp->ep->com.mapped_local_addr;
+			struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
+				&qp->ep->com.mapped_remote_addr;
+
+			cc = snprintf(qpd->buf + qpd->pos, space,
+				      "rc qp sq id %u rq id %u state %u "
+				      "onchip %u ep tid %u state %u "
+				      "%pI4:%u/%u->%pI4:%u/%u\n",
+				      qp->wq.sq.qid, qp->wq.rq.qid,
+				      (int)qp->attr.state,
+				      qp->wq.sq.flags & T4_SQ_ONCHIP,
+				      qp->ep->hwtid, (int)qp->ep->com.state,
+				      &lsin->sin_addr, ntohs(lsin->sin_port),
+				      ntohs(mapped_lsin->sin_port),
+				      &rsin->sin_addr, ntohs(rsin->sin_port),
+				      ntohs(mapped_rsin->sin_port));
+		} else {
+			struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
+				&qp->ep->com.local_addr;
+			struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
+				&qp->ep->com.remote_addr;
+			struct sockaddr_in6 *mapped_lsin6 =
+				(struct sockaddr_in6 *)
+				&qp->ep->com.mapped_local_addr;
+			struct sockaddr_in6 *mapped_rsin6 =
+				(struct sockaddr_in6 *)
+				&qp->ep->com.mapped_remote_addr;
+
+			cc = snprintf(qpd->buf + qpd->pos, space,
+				      "rc qp sq id %u rq id %u state %u "
+				      "onchip %u ep tid %u state %u "
+				      "%pI6:%u/%u->%pI6:%u/%u\n",
+				      qp->wq.sq.qid, qp->wq.rq.qid,
+				      (int)qp->attr.state,
+				      qp->wq.sq.flags & T4_SQ_ONCHIP,
+				      qp->ep->hwtid, (int)qp->ep->com.state,
+				      &lsin6->sin6_addr,
+				      ntohs(lsin6->sin6_port),
+				      ntohs(mapped_lsin6->sin6_port),
+				      &rsin6->sin6_addr,
+				      ntohs(rsin6->sin6_port),
+				      ntohs(mapped_rsin6->sin6_port));
+		}
+	} else
+		cc = snprintf(qpd->buf + qpd->pos, space,
+			     "qp sq id %u rq id %u state %u onchip %u\n",
+			      qp->wq.sq.qid, qp->wq.rq.qid,
+			      (int)qp->attr.state,
+			      qp->wq.sq.flags & T4_SQ_ONCHIP);
+	if (cc < space)
+		qpd->pos += cc;
+	return 0;
+}
+
+static int qp_release(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *qpd = file->private_data;
+	if (!qpd) {
+		printk(KERN_INFO "%s null qpd?\n", __func__);
+		return 0;
+	}
+	vfree(qpd->buf);
+	kfree(qpd);
+	return 0;
+}
+
+static int qp_open(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *qpd;
+	int ret = 0;
+	int count = 1;
+
+	qpd = kmalloc(sizeof *qpd, GFP_KERNEL);
+	if (!qpd) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	qpd->devp = inode->i_private;
+	qpd->pos = 0;
+
+	spin_lock_irq(&qpd->devp->lock);
+	idr_for_each(&qpd->devp->qpidr, count_idrs, &count);
+	spin_unlock_irq(&qpd->devp->lock);
+
+	qpd->bufsize = count * 128;
+	qpd->buf = vmalloc(qpd->bufsize);
+	if (!qpd->buf) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
+	spin_lock_irq(&qpd->devp->lock);
+	idr_for_each(&qpd->devp->qpidr, dump_qp, qpd);
+	spin_unlock_irq(&qpd->devp->lock);
+
+	qpd->buf[qpd->pos++] = 0;
+	file->private_data = qpd;
+	goto out;
+err1:
+	kfree(qpd);
+out:
+	return ret;
+}
+
+static const struct file_operations qp_debugfs_fops = {
+	.owner   = THIS_MODULE,
+	.open    = qp_open,
+	.release = qp_release,
+	.read    = debugfs_read,
+	.llseek  = default_llseek,
+};
+
+static int dump_stag(int id, void *p, void *data)
+{
+	struct c4iw_debugfs_data *stagd = data;
+	int space;
+	int cc;
+
+	space = stagd->bufsize - stagd->pos - 1;
+	if (space == 0)
+		return 1;
+
+	cc = snprintf(stagd->buf + stagd->pos, space, "0x%x\n", id<<8);
+	if (cc < space)
+		stagd->pos += cc;
+	return 0;
+}
+
+static int stag_release(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *stagd = file->private_data;
+	if (!stagd) {
+		printk(KERN_INFO "%s null stagd?\n", __func__);
+		return 0;
+	}
+	kfree(stagd->buf);
+	kfree(stagd);
+	return 0;
+}
+
+static int stag_open(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *stagd;
+	int ret = 0;
+	int count = 1;
+
+	stagd = kmalloc(sizeof *stagd, GFP_KERNEL);
+	if (!stagd) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	stagd->devp = inode->i_private;
+	stagd->pos = 0;
+
+	spin_lock_irq(&stagd->devp->lock);
+	idr_for_each(&stagd->devp->mmidr, count_idrs, &count);
+	spin_unlock_irq(&stagd->devp->lock);
+
+	stagd->bufsize = count * sizeof("0x12345678\n");
+	stagd->buf = kmalloc(stagd->bufsize, GFP_KERNEL);
+	if (!stagd->buf) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
+	spin_lock_irq(&stagd->devp->lock);
+	idr_for_each(&stagd->devp->mmidr, dump_stag, stagd);
+	spin_unlock_irq(&stagd->devp->lock);
+
+	stagd->buf[stagd->pos++] = 0;
+	file->private_data = stagd;
+	goto out;
+err1:
+	kfree(stagd);
+out:
+	return ret;
+}
+
+static const struct file_operations stag_debugfs_fops = {
+	.owner   = THIS_MODULE,
+	.open    = stag_open,
+	.release = stag_release,
+	.read    = debugfs_read,
+	.llseek  = default_llseek,
+};
+
+static char *db_state_str[] = {"NORMAL", "FLOW_CONTROL", "RECOVERY", "STOPPED"};
+
+static int stats_show(struct seq_file *seq, void *v)
+{
+	struct c4iw_dev *dev = seq->private;
+
+	seq_printf(seq, "   Object: %10s %10s %10s %10s\n", "Total", "Current",
+		   "Max", "Fail");
+	seq_printf(seq, "     PDID: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.pd.total, dev->rdev.stats.pd.cur,
+			dev->rdev.stats.pd.max, dev->rdev.stats.pd.fail);
+	seq_printf(seq, "      QID: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.qid.total, dev->rdev.stats.qid.cur,
+			dev->rdev.stats.qid.max, dev->rdev.stats.qid.fail);
+	seq_printf(seq, "   TPTMEM: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.stag.total, dev->rdev.stats.stag.cur,
+			dev->rdev.stats.stag.max, dev->rdev.stats.stag.fail);
+	seq_printf(seq, "   PBLMEM: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.pbl.total, dev->rdev.stats.pbl.cur,
+			dev->rdev.stats.pbl.max, dev->rdev.stats.pbl.fail);
+	seq_printf(seq, "   RQTMEM: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.rqt.total, dev->rdev.stats.rqt.cur,
+			dev->rdev.stats.rqt.max, dev->rdev.stats.rqt.fail);
+	seq_printf(seq, "  OCQPMEM: %10llu %10llu %10llu %10llu\n",
+			dev->rdev.stats.ocqp.total, dev->rdev.stats.ocqp.cur,
+			dev->rdev.stats.ocqp.max, dev->rdev.stats.ocqp.fail);
+	seq_printf(seq, "  DB FULL: %10llu\n", dev->rdev.stats.db_full);
+	seq_printf(seq, " DB EMPTY: %10llu\n", dev->rdev.stats.db_empty);
+	seq_printf(seq, "  DB DROP: %10llu\n", dev->rdev.stats.db_drop);
+	seq_printf(seq, " DB State: %s Transitions %llu FC Interruptions %llu\n",
+		   db_state_str[dev->db_state],
+		   dev->rdev.stats.db_state_transitions,
+		   dev->rdev.stats.db_fc_interruptions);
+	seq_printf(seq, "TCAM_FULL: %10llu\n", dev->rdev.stats.tcam_full);
+	seq_printf(seq, "ACT_OFLD_CONN_FAILS: %10llu\n",
+		   dev->rdev.stats.act_ofld_conn_fails);
+	seq_printf(seq, "PAS_OFLD_CONN_FAILS: %10llu\n",
+		   dev->rdev.stats.pas_ofld_conn_fails);
+	return 0;
+}
+
+static int stats_open(struct inode *inode, struct file *file)
+{
+	return single_open(file, stats_show, inode->i_private);
+}
+
+static ssize_t stats_clear(struct file *file, const char __user *buf,
+		size_t count, loff_t *pos)
+{
+	struct c4iw_dev *dev = ((struct seq_file *)file->private_data)->private;
+
+	mutex_lock(&dev->rdev.stats.lock);
+	dev->rdev.stats.pd.max = 0;
+	dev->rdev.stats.pd.fail = 0;
+	dev->rdev.stats.qid.max = 0;
+	dev->rdev.stats.qid.fail = 0;
+	dev->rdev.stats.stag.max = 0;
+	dev->rdev.stats.stag.fail = 0;
+	dev->rdev.stats.pbl.max = 0;
+	dev->rdev.stats.pbl.fail = 0;
+	dev->rdev.stats.rqt.max = 0;
+	dev->rdev.stats.rqt.fail = 0;
+	dev->rdev.stats.ocqp.max = 0;
+	dev->rdev.stats.ocqp.fail = 0;
+	dev->rdev.stats.db_full = 0;
+	dev->rdev.stats.db_empty = 0;
+	dev->rdev.stats.db_drop = 0;
+	dev->rdev.stats.db_state_transitions = 0;
+	dev->rdev.stats.tcam_full = 0;
+	dev->rdev.stats.act_ofld_conn_fails = 0;
+	dev->rdev.stats.pas_ofld_conn_fails = 0;
+	mutex_unlock(&dev->rdev.stats.lock);
+	return count;
+}
+
+static const struct file_operations stats_debugfs_fops = {
+	.owner   = THIS_MODULE,
+	.open    = stats_open,
+	.release = single_release,
+	.read    = seq_read,
+	.llseek  = seq_lseek,
+	.write   = stats_clear,
+};
+
+static int dump_ep(int id, void *p, void *data)
+{
+	struct c4iw_ep *ep = p;
+	struct c4iw_debugfs_data *epd = data;
+	int space;
+	int cc;
+
+	space = epd->bufsize - epd->pos - 1;
+	if (space == 0)
+		return 1;
+
+	if (ep->com.local_addr.ss_family == AF_INET) {
+		struct sockaddr_in *lsin = (struct sockaddr_in *)
+			&ep->com.local_addr;
+		struct sockaddr_in *rsin = (struct sockaddr_in *)
+			&ep->com.remote_addr;
+		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
+			&ep->com.mapped_local_addr;
+		struct sockaddr_in *mapped_rsin = (struct sockaddr_in *)
+			&ep->com.mapped_remote_addr;
+
+		cc = snprintf(epd->buf + epd->pos, space,
+			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
+			      "history 0x%lx hwtid %d atid %d "
+			      "%pI4:%d/%d <-> %pI4:%d/%d\n",
+			      ep, ep->com.cm_id, ep->com.qp,
+			      (int)ep->com.state, ep->com.flags,
+			      ep->com.history, ep->hwtid, ep->atid,
+			      &lsin->sin_addr, ntohs(lsin->sin_port),
+			      ntohs(mapped_lsin->sin_port),
+			      &rsin->sin_addr, ntohs(rsin->sin_port),
+			      ntohs(mapped_rsin->sin_port));
+	} else {
+		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
+			&ep->com.local_addr;
+		struct sockaddr_in6 *rsin6 = (struct sockaddr_in6 *)
+			&ep->com.remote_addr;
+		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
+			&ep->com.mapped_local_addr;
+		struct sockaddr_in6 *mapped_rsin6 = (struct sockaddr_in6 *)
+			&ep->com.mapped_remote_addr;
+
+		cc = snprintf(epd->buf + epd->pos, space,
+			      "ep %p cm_id %p qp %p state %d flags 0x%lx "
+			      "history 0x%lx hwtid %d atid %d "
+			      "%pI6:%d/%d <-> %pI6:%d/%d\n",
+			      ep, ep->com.cm_id, ep->com.qp,
+			      (int)ep->com.state, ep->com.flags,
+			      ep->com.history, ep->hwtid, ep->atid,
+			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
+			      ntohs(mapped_lsin6->sin6_port),
+			      &rsin6->sin6_addr, ntohs(rsin6->sin6_port),
+			      ntohs(mapped_rsin6->sin6_port));
+	}
+	if (cc < space)
+		epd->pos += cc;
+	return 0;
+}
+
+static int dump_listen_ep(int id, void *p, void *data)
+{
+	struct c4iw_listen_ep *ep = p;
+	struct c4iw_debugfs_data *epd = data;
+	int space;
+	int cc;
+
+	space = epd->bufsize - epd->pos - 1;
+	if (space == 0)
+		return 1;
+
+	if (ep->com.local_addr.ss_family == AF_INET) {
+		struct sockaddr_in *lsin = (struct sockaddr_in *)
+			&ep->com.local_addr;
+		struct sockaddr_in *mapped_lsin = (struct sockaddr_in *)
+			&ep->com.mapped_local_addr;
+
+		cc = snprintf(epd->buf + epd->pos, space,
+			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
+			      "backlog %d %pI4:%d/%d\n",
+			      ep, ep->com.cm_id, (int)ep->com.state,
+			      ep->com.flags, ep->stid, ep->backlog,
+			      &lsin->sin_addr, ntohs(lsin->sin_port),
+			      ntohs(mapped_lsin->sin_port));
+	} else {
+		struct sockaddr_in6 *lsin6 = (struct sockaddr_in6 *)
+			&ep->com.local_addr;
+		struct sockaddr_in6 *mapped_lsin6 = (struct sockaddr_in6 *)
+			&ep->com.mapped_local_addr;
+
+		cc = snprintf(epd->buf + epd->pos, space,
+			      "ep %p cm_id %p state %d flags 0x%lx stid %d "
+			      "backlog %d %pI6:%d/%d\n",
+			      ep, ep->com.cm_id, (int)ep->com.state,
+			      ep->com.flags, ep->stid, ep->backlog,
+			      &lsin6->sin6_addr, ntohs(lsin6->sin6_port),
+			      ntohs(mapped_lsin6->sin6_port));
+	}
+	if (cc < space)
+		epd->pos += cc;
+	return 0;
+}
+
+static int ep_release(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *epd = file->private_data;
+	if (!epd) {
+		pr_info("%s null qpd?\n", __func__);
+		return 0;
+	}
+	vfree(epd->buf);
+	kfree(epd);
+	return 0;
+}
+
+static int ep_open(struct inode *inode, struct file *file)
+{
+	struct c4iw_debugfs_data *epd;
+	int ret = 0;
+	int count = 1;
+
+	epd = kmalloc(sizeof(*epd), GFP_KERNEL);
+	if (!epd) {
+		ret = -ENOMEM;
+		goto out;
+	}
+	epd->devp = inode->i_private;
+	epd->pos = 0;
+
+	spin_lock_irq(&epd->devp->lock);
+	idr_for_each(&epd->devp->hwtid_idr, count_idrs, &count);
+	idr_for_each(&epd->devp->atid_idr, count_idrs, &count);
+	idr_for_each(&epd->devp->stid_idr, count_idrs, &count);
+	spin_unlock_irq(&epd->devp->lock);
+
+	epd->bufsize = count * 160;
+	epd->buf = vmalloc(epd->bufsize);
+	if (!epd->buf) {
+		ret = -ENOMEM;
+		goto err1;
+	}
+
+	spin_lock_irq(&epd->devp->lock);
+	idr_for_each(&epd->devp->hwtid_idr, dump_ep, epd);
+	idr_for_each(&epd->devp->atid_idr, dump_ep, epd);
+	idr_for_each(&epd->devp->stid_idr, dump_listen_ep, epd);
+	spin_unlock_irq(&epd->devp->lock);
+
+	file->private_data = epd;
+	goto out;
+err1:
+	kfree(epd);
+out:
+	return ret;
+}
+
+static const struct file_operations ep_debugfs_fops = {
+	.owner   = THIS_MODULE,
+	.open    = ep_open,
+	.release = ep_release,
+	.read    = debugfs_read,
+};
+
+static int setup_debugfs(struct c4iw_dev *devp)
+{
+	struct dentry *de;
+
+	if (!devp->debugfs_root)
+		return -1;
+
+	de = debugfs_create_file("qps", S_IWUSR, devp->debugfs_root,
+				 (void *)devp, &qp_debugfs_fops);
+	if (de && de->d_inode)
+		de->d_inode->i_size = 4096;
+
+	de = debugfs_create_file("stags", S_IWUSR, devp->debugfs_root,
+				 (void *)devp, &stag_debugfs_fops);
+	if (de && de->d_inode)
+		de->d_inode->i_size = 4096;
+
+	de = debugfs_create_file("stats", S_IWUSR, devp->debugfs_root,
+			(void *)devp, &stats_debugfs_fops);
+	if (de && de->d_inode)
+		de->d_inode->i_size = 4096;
+
+	de = debugfs_create_file("eps", S_IWUSR, devp->debugfs_root,
+			(void *)devp, &ep_debugfs_fops);
+	if (de && de->d_inode)
+		de->d_inode->i_size = 4096;
+
+	return 0;
+}
+
+void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
+			       struct c4iw_dev_ucontext *uctx)
+{
+	struct list_head *pos, *nxt;
+	struct c4iw_qid_list *entry;
+
+	mutex_lock(&uctx->lock);
+	list_for_each_safe(pos, nxt, &uctx->qpids) {
+		entry = list_entry(pos, struct c4iw_qid_list, entry);
+		list_del_init(&entry->entry);
+		if (!(entry->qid & rdev->qpmask)) {
+			c4iw_put_resource(&rdev->resource.qid_table,
+					  entry->qid);
+			mutex_lock(&rdev->stats.lock);
+			rdev->stats.qid.cur -= rdev->qpmask + 1;
+			mutex_unlock(&rdev->stats.lock);
+		}
+		kfree(entry);
+	}
+
+	list_for_each_safe(pos, nxt, &uctx->qpids) {
+		entry = list_entry(pos, struct c4iw_qid_list, entry);
+		list_del_init(&entry->entry);
+		kfree(entry);
+	}
+	mutex_unlock(&uctx->lock);
+}
+
+void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
+			    struct c4iw_dev_ucontext *uctx)
+{
+	INIT_LIST_HEAD(&uctx->qpids);
+	INIT_LIST_HEAD(&uctx->cqids);
+	mutex_init(&uctx->lock);
+}
+
+/* Caller takes care of locking if needed */
+static int c4iw_rdev_open(struct c4iw_rdev *rdev)
+{
+	int err;
+
+	c4iw_init_dev_ucontext(rdev, &rdev->uctx);
+
+	/*
+	 * qpshift is the number of bits to shift the qpid left in order
+	 * to get the correct address of the doorbell for that qp.
+	 */
+	rdev->qpshift = PAGE_SHIFT - ilog2(rdev->lldi.udb_density);
+	rdev->qpmask = rdev->lldi.udb_density - 1;
+	rdev->cqshift = PAGE_SHIFT - ilog2(rdev->lldi.ucq_density);
+	rdev->cqmask = rdev->lldi.ucq_density - 1;
+	PDBG("%s dev %s stag start 0x%0x size 0x%0x num stags %d "
+	     "pbl start 0x%0x size 0x%0x rq start 0x%0x size 0x%0x "
+	     "qp qid start %u size %u cq qid start %u size %u\n",
+	     __func__, pci_name(rdev->lldi.pdev), rdev->lldi.vr->stag.start,
+	     rdev->lldi.vr->stag.size, c4iw_num_stags(rdev),
+	     rdev->lldi.vr->pbl.start,
+	     rdev->lldi.vr->pbl.size, rdev->lldi.vr->rq.start,
+	     rdev->lldi.vr->rq.size,
+	     rdev->lldi.vr->qp.start,
+	     rdev->lldi.vr->qp.size,
+	     rdev->lldi.vr->cq.start,
+	     rdev->lldi.vr->cq.size);
+	PDBG("udb len 0x%x udb base %llx db_reg %p gts_reg %p qpshift %lu "
+	     "qpmask 0x%x cqshift %lu cqmask 0x%x\n",
+	     (unsigned)pci_resource_len(rdev->lldi.pdev, 2),
+	     (u64)pci_resource_start(rdev->lldi.pdev, 2),
+	     rdev->lldi.db_reg,
+	     rdev->lldi.gts_reg,
+	     rdev->qpshift, rdev->qpmask,
+	     rdev->cqshift, rdev->cqmask);
+
+	if (c4iw_num_stags(rdev) == 0) {
+		err = -EINVAL;
+		goto err1;
+	}
+
+	rdev->stats.pd.total = T4_MAX_NUM_PD;
+	rdev->stats.stag.total = rdev->lldi.vr->stag.size;
+	rdev->stats.pbl.total = rdev->lldi.vr->pbl.size;
+	rdev->stats.rqt.total = rdev->lldi.vr->rq.size;
+	rdev->stats.ocqp.total = rdev->lldi.vr->ocq.size;
+	rdev->stats.qid.total = rdev->lldi.vr->qp.size;
+
+	err = c4iw_init_resource(rdev, c4iw_num_stags(rdev), T4_MAX_NUM_PD);
+	if (err) {
+		printk(KERN_ERR MOD "error %d initializing resources\n", err);
+		goto err1;
+	}
+	err = c4iw_pblpool_create(rdev);
+	if (err) {
+		printk(KERN_ERR MOD "error %d initializing pbl pool\n", err);
+		goto err2;
+	}
+	err = c4iw_rqtpool_create(rdev);
+	if (err) {
+		printk(KERN_ERR MOD "error %d initializing rqt pool\n", err);
+		goto err3;
+	}
+	err = c4iw_ocqp_pool_create(rdev);
+	if (err) {
+		printk(KERN_ERR MOD "error %d initializing ocqp pool\n", err);
+		goto err4;
+	}
+	rdev->status_page = (struct t4_dev_status_page *)
+			    __get_free_page(GFP_KERNEL);
+	if (!rdev->status_page) {
+		pr_err(MOD "error allocating status page\n");
+		goto err4;
+	}
+	rdev->status_page->db_off = 0;
+	return 0;
+err4:
+	c4iw_rqtpool_destroy(rdev);
+err3:
+	c4iw_pblpool_destroy(rdev);
+err2:
+	c4iw_destroy_resource(&rdev->resource);
+err1:
+	return err;
+}
+
+static void c4iw_rdev_close(struct c4iw_rdev *rdev)
+{
+	free_page((unsigned long)rdev->status_page);
+	c4iw_pblpool_destroy(rdev);
+	c4iw_rqtpool_destroy(rdev);
+	c4iw_destroy_resource(&rdev->resource);
+}
+
+static void c4iw_dealloc(struct uld_ctx *ctx)
+{
+	c4iw_rdev_close(&ctx->dev->rdev);
+	idr_destroy(&ctx->dev->cqidr);
+	idr_destroy(&ctx->dev->qpidr);
+	idr_destroy(&ctx->dev->mmidr);
+	idr_destroy(&ctx->dev->hwtid_idr);
+	idr_destroy(&ctx->dev->stid_idr);
+	idr_destroy(&ctx->dev->atid_idr);
+	if (ctx->dev->rdev.bar2_kva)
+		iounmap(ctx->dev->rdev.bar2_kva);
+	if (ctx->dev->rdev.oc_mw_kva)
+		iounmap(ctx->dev->rdev.oc_mw_kva);
+	ib_dealloc_device(&ctx->dev->ibdev);
+	ctx->dev = NULL;
+}
+
+static void c4iw_remove(struct uld_ctx *ctx)
+{
+	PDBG("%s c4iw_dev %p\n", __func__,  ctx->dev);
+	c4iw_unregister_device(ctx->dev);
+	c4iw_dealloc(ctx);
+}
+
+static int rdma_supported(const struct cxgb4_lld_info *infop)
+{
+	return infop->vr->stag.size > 0 && infop->vr->pbl.size > 0 &&
+	       infop->vr->rq.size > 0 && infop->vr->qp.size > 0 &&
+	       infop->vr->cq.size > 0;
+}
+
+static struct c4iw_dev *c4iw_alloc(const struct cxgb4_lld_info *infop)
+{
+	struct c4iw_dev *devp;
+	int ret;
+
+	if (!rdma_supported(infop)) {
+		printk(KERN_INFO MOD "%s: RDMA not supported on this device.\n",
+		       pci_name(infop->pdev));
+		return ERR_PTR(-ENOSYS);
+	}
+	if (!ocqp_supported(infop))
+		pr_info("%s: On-Chip Queues not supported on this device.\n",
+			pci_name(infop->pdev));
+
+	devp = (struct c4iw_dev *)ib_alloc_device(sizeof(*devp));
+	if (!devp) {
+		printk(KERN_ERR MOD "Cannot allocate ib device\n");
+		return ERR_PTR(-ENOMEM);
+	}
+	devp->rdev.lldi = *infop;
+
+	/*
+	 * For T5 devices, we map all of BAR2 with WC.
+	 * For T4 devices with onchip qp mem, we map only that part
+	 * of BAR2 with WC.
+	 */
+	devp->rdev.bar2_pa = pci_resource_start(devp->rdev.lldi.pdev, 2);
+	if (is_t5(devp->rdev.lldi.adapter_type)) {
+		devp->rdev.bar2_kva = ioremap_wc(devp->rdev.bar2_pa,
+			pci_resource_len(devp->rdev.lldi.pdev, 2));
+		if (!devp->rdev.bar2_kva) {
+			pr_err(MOD "Unable to ioremap BAR2\n");
+			ib_dealloc_device(&devp->ibdev);
+			return ERR_PTR(-EINVAL);
+		}
+	} else if (ocqp_supported(infop)) {
+		devp->rdev.oc_mw_pa =
+			pci_resource_start(devp->rdev.lldi.pdev, 2) +
+			pci_resource_len(devp->rdev.lldi.pdev, 2) -
+			roundup_pow_of_two(devp->rdev.lldi.vr->ocq.size);
+		devp->rdev.oc_mw_kva = ioremap_wc(devp->rdev.oc_mw_pa,
+			devp->rdev.lldi.vr->ocq.size);
+		if (!devp->rdev.oc_mw_kva) {
+			pr_err(MOD "Unable to ioremap onchip mem\n");
+			ib_dealloc_device(&devp->ibdev);
+			return ERR_PTR(-EINVAL);
+		}
+	}
+
+	PDBG(KERN_INFO MOD "ocq memory: "
+	       "hw_start 0x%x size %u mw_pa 0x%lx mw_kva %p\n",
+	       devp->rdev.lldi.vr->ocq.start, devp->rdev.lldi.vr->ocq.size,
+	       devp->rdev.oc_mw_pa, devp->rdev.oc_mw_kva);
+
+	ret = c4iw_rdev_open(&devp->rdev);
+	if (ret) {
+		printk(KERN_ERR MOD "Unable to open CXIO rdev err %d\n", ret);
+		ib_dealloc_device(&devp->ibdev);
+		return ERR_PTR(ret);
+	}
+
+	idr_init(&devp->cqidr);
+	idr_init(&devp->qpidr);
+	idr_init(&devp->mmidr);
+	idr_init(&devp->hwtid_idr);
+	idr_init(&devp->stid_idr);
+	idr_init(&devp->atid_idr);
+	spin_lock_init(&devp->lock);
+	mutex_init(&devp->rdev.stats.lock);
+	mutex_init(&devp->db_mutex);
+	INIT_LIST_HEAD(&devp->db_fc_list);
+
+	if (c4iw_debugfs_root) {
+		devp->debugfs_root = debugfs_create_dir(
+					pci_name(devp->rdev.lldi.pdev),
+					c4iw_debugfs_root);
+		setup_debugfs(devp);
+	}
+
+
+	return devp;
+}
+
+static void *c4iw_uld_add(const struct cxgb4_lld_info *infop)
+{
+	struct uld_ctx *ctx;
+	static int vers_printed;
+	int i;
+
+	if (!vers_printed++)
+		pr_info("Chelsio T4/T5 RDMA Driver - version %s\n",
+			DRV_VERSION);
+
+	ctx = kzalloc(sizeof *ctx, GFP_KERNEL);
+	if (!ctx) {
+		ctx = ERR_PTR(-ENOMEM);
+		goto out;
+	}
+	ctx->lldi = *infop;
+
+	PDBG("%s found device %s nchan %u nrxq %u ntxq %u nports %u\n",
+	     __func__, pci_name(ctx->lldi.pdev),
+	     ctx->lldi.nchan, ctx->lldi.nrxq,
+	     ctx->lldi.ntxq, ctx->lldi.nports);
+
+	mutex_lock(&dev_mutex);
+	list_add_tail(&ctx->entry, &uld_ctx_list);
+	mutex_unlock(&dev_mutex);
+
+	for (i = 0; i < ctx->lldi.nrxq; i++)
+		PDBG("rxqid[%u] %u\n", i, ctx->lldi.rxq_ids[i]);
+out:
+	return ctx;
+}
+
+static inline struct sk_buff *copy_gl_to_skb_pkt(const struct pkt_gl *gl,
+						 const __be64 *rsp,
+						 u32 pktshift)
+{
+	struct sk_buff *skb;
+
+	/*
+	 * Allocate space for cpl_pass_accept_req which will be synthesized by
+	 * driver. Once the driver synthesizes the request the skb will go
+	 * through the regular cpl_pass_accept_req processing.
+	 * The math here assumes sizeof cpl_pass_accept_req >= sizeof
+	 * cpl_rx_pkt.
+	 */
+	skb = alloc_skb(gl->tot_len + sizeof(struct cpl_pass_accept_req) +
+			sizeof(struct rss_header) - pktshift, GFP_ATOMIC);
+	if (unlikely(!skb))
+		return NULL;
+
+	 __skb_put(skb, gl->tot_len + sizeof(struct cpl_pass_accept_req) +
+		   sizeof(struct rss_header) - pktshift);
+
+	/*
+	 * This skb will contain:
+	 *   rss_header from the rspq descriptor (1 flit)
+	 *   cpl_rx_pkt struct from the rspq descriptor (2 flits)
+	 *   space for the difference between the size of an
+	 *      rx_pkt and pass_accept_req cpl (1 flit)
+	 *   the packet data from the gl
+	 */
+	skb_copy_to_linear_data(skb, rsp, sizeof(struct cpl_pass_accept_req) +
+				sizeof(struct rss_header));
+	skb_copy_to_linear_data_offset(skb, sizeof(struct rss_header) +
+				       sizeof(struct cpl_pass_accept_req),
+				       gl->va + pktshift,
+				       gl->tot_len - pktshift);
+	return skb;
+}
+
+static inline int recv_rx_pkt(struct c4iw_dev *dev, const struct pkt_gl *gl,
+			   const __be64 *rsp)
+{
+	unsigned int opcode = *(u8 *)rsp;
+	struct sk_buff *skb;
+
+	if (opcode != CPL_RX_PKT)
+		goto out;
+
+	skb = copy_gl_to_skb_pkt(gl , rsp, dev->rdev.lldi.sge_pktshift);
+	if (skb == NULL)
+		goto out;
+
+	if (c4iw_handlers[opcode] == NULL) {
+		pr_info("%s no handler opcode 0x%x...\n", __func__,
+		       opcode);
+		kfree_skb(skb);
+		goto out;
+	}
+	c4iw_handlers[opcode](dev, skb);
+	return 1;
+out:
+	return 0;
+}
+
+static int c4iw_uld_rx_handler(void *handle, const __be64 *rsp,
+			const struct pkt_gl *gl)
+{
+	struct uld_ctx *ctx = handle;
+	struct c4iw_dev *dev = ctx->dev;
+	struct sk_buff *skb;
+	u8 opcode;
+
+	if (gl == NULL) {
+		/* omit RSS and rsp_ctrl at end of descriptor */
+		unsigned int len = 64 - sizeof(struct rsp_ctrl) - 8;
+
+		skb = alloc_skb(256, GFP_ATOMIC);
+		if (!skb)
+			goto nomem;
+		__skb_put(skb, len);
+		skb_copy_to_linear_data(skb, &rsp[1], len);
+	} else if (gl == CXGB4_MSG_AN) {
+		const struct rsp_ctrl *rc = (void *)rsp;
+
+		u32 qid = be32_to_cpu(rc->pldbuflen_qid);
+		c4iw_ev_handler(dev, qid);
+		return 0;
+	} else if (unlikely(*(u8 *)rsp != *(u8 *)gl->va)) {
+		if (recv_rx_pkt(dev, gl, rsp))
+			return 0;
+
+		pr_info("%s: unexpected FL contents at %p, " \
+		       "RSS %#llx, FL %#llx, len %u\n",
+		       pci_name(ctx->lldi.pdev), gl->va,
+		       (unsigned long long)be64_to_cpu(*rsp),
+		       (unsigned long long)be64_to_cpu(
+		       *(__force __be64 *)gl->va),
+		       gl->tot_len);
+
+		return 0;
+	} else {
+		skb = cxgb4_pktgl_to_skb(gl, 128, 128);
+		if (unlikely(!skb))
+			goto nomem;
+	}
+
+	opcode = *(u8 *)rsp;
+	if (c4iw_handlers[opcode]) {
+		c4iw_handlers[opcode](dev, skb);
+	} else {
+		pr_info("%s no handler opcode 0x%x...\n", __func__,
+		       opcode);
+		kfree_skb(skb);
+	}
+
+	return 0;
+nomem:
+	return -1;
+}
+
+static int c4iw_uld_state_change(void *handle, enum cxgb4_state new_state)
+{
+	struct uld_ctx *ctx = handle;
+
+	PDBG("%s new_state %u\n", __func__, new_state);
+	switch (new_state) {
+	case CXGB4_STATE_UP:
+		printk(KERN_INFO MOD "%s: Up\n", pci_name(ctx->lldi.pdev));
+		if (!ctx->dev) {
+			int ret;
+
+			ctx->dev = c4iw_alloc(&ctx->lldi);
+			if (IS_ERR(ctx->dev)) {
+				printk(KERN_ERR MOD
+				       "%s: initialization failed: %ld\n",
+				       pci_name(ctx->lldi.pdev),
+				       PTR_ERR(ctx->dev));
+				ctx->dev = NULL;
+				break;
+			}
+			ret = c4iw_register_device(ctx->dev);
+			if (ret) {
+				printk(KERN_ERR MOD
+				       "%s: RDMA registration failed: %d\n",
+				       pci_name(ctx->lldi.pdev), ret);
+				c4iw_dealloc(ctx);
+			}
+		}
+		break;
+	case CXGB4_STATE_DOWN:
+		printk(KERN_INFO MOD "%s: Down\n",
+		       pci_name(ctx->lldi.pdev));
+		if (ctx->dev)
+			c4iw_remove(ctx);
+		break;
+	case CXGB4_STATE_START_RECOVERY:
+		printk(KERN_INFO MOD "%s: Fatal Error\n",
+		       pci_name(ctx->lldi.pdev));
+		if (ctx->dev) {
+			struct ib_event event;
+
+			ctx->dev->rdev.flags |= T4_FATAL_ERROR;
+			memset(&event, 0, sizeof event);
+			event.event  = IB_EVENT_DEVICE_FATAL;
+			event.device = &ctx->dev->ibdev;
+			ib_dispatch_event(&event);
+			c4iw_remove(ctx);
+		}
+		break;
+	case CXGB4_STATE_DETACH:
+		printk(KERN_INFO MOD "%s: Detach\n",
+		       pci_name(ctx->lldi.pdev));
+		if (ctx->dev)
+			c4iw_remove(ctx);
+		break;
+	}
+	return 0;
+}
+
+static int disable_qp_db(int id, void *p, void *data)
+{
+	struct c4iw_qp *qp = p;
+
+	t4_disable_wq_db(&qp->wq);
+	return 0;
+}
+
+static void stop_queues(struct uld_ctx *ctx)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ctx->dev->lock, flags);
+	ctx->dev->rdev.stats.db_state_transitions++;
+	ctx->dev->db_state = STOPPED;
+	if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED)
+		idr_for_each(&ctx->dev->qpidr, disable_qp_db, NULL);
+	else
+		ctx->dev->rdev.status_page->db_off = 1;
+	spin_unlock_irqrestore(&ctx->dev->lock, flags);
+}
+
+static int enable_qp_db(int id, void *p, void *data)
+{
+	struct c4iw_qp *qp = p;
+
+	t4_enable_wq_db(&qp->wq);
+	return 0;
+}
+
+static void resume_rc_qp(struct c4iw_qp *qp)
+{
+	spin_lock(&qp->lock);
+	t4_ring_sq_db(&qp->wq, qp->wq.sq.wq_pidx_inc,
+		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
+	qp->wq.sq.wq_pidx_inc = 0;
+	t4_ring_rq_db(&qp->wq, qp->wq.rq.wq_pidx_inc,
+		      is_t5(qp->rhp->rdev.lldi.adapter_type), NULL);
+	qp->wq.rq.wq_pidx_inc = 0;
+	spin_unlock(&qp->lock);
+}
+
+static void resume_a_chunk(struct uld_ctx *ctx)
+{
+	int i;
+	struct c4iw_qp *qp;
+
+	for (i = 0; i < DB_FC_RESUME_SIZE; i++) {
+		qp = list_first_entry(&ctx->dev->db_fc_list, struct c4iw_qp,
+				      db_fc_entry);
+		list_del_init(&qp->db_fc_entry);
+		resume_rc_qp(qp);
+		if (list_empty(&ctx->dev->db_fc_list))
+			break;
+	}
+}
+
+static void resume_queues(struct uld_ctx *ctx)
+{
+	spin_lock_irq(&ctx->dev->lock);
+	if (ctx->dev->db_state != STOPPED)
+		goto out;
+	ctx->dev->db_state = FLOW_CONTROL;
+	while (1) {
+		if (list_empty(&ctx->dev->db_fc_list)) {
+			WARN_ON(ctx->dev->db_state != FLOW_CONTROL);
+			ctx->dev->db_state = NORMAL;
+			ctx->dev->rdev.stats.db_state_transitions++;
+			if (ctx->dev->rdev.flags & T4_STATUS_PAGE_DISABLED) {
+				idr_for_each(&ctx->dev->qpidr, enable_qp_db,
+					     NULL);
+			} else {
+				ctx->dev->rdev.status_page->db_off = 0;
+			}
+			break;
+		} else {
+			if (cxgb4_dbfifo_count(ctx->dev->rdev.lldi.ports[0], 1)
+			    < (ctx->dev->rdev.lldi.dbfifo_int_thresh <<
+			       DB_FC_DRAIN_THRESH)) {
+				resume_a_chunk(ctx);
+			}
+			if (!list_empty(&ctx->dev->db_fc_list)) {
+				spin_unlock_irq(&ctx->dev->lock);
+				if (DB_FC_RESUME_DELAY) {
+					set_current_state(TASK_UNINTERRUPTIBLE);
+					schedule_timeout(DB_FC_RESUME_DELAY);
+				}
+				spin_lock_irq(&ctx->dev->lock);
+				if (ctx->dev->db_state != FLOW_CONTROL)
+					break;
+			}
+		}
+	}
+out:
+	if (ctx->dev->db_state != NORMAL)
+		ctx->dev->rdev.stats.db_fc_interruptions++;
+	spin_unlock_irq(&ctx->dev->lock);
+}
+
+struct qp_list {
+	unsigned idx;
+	struct c4iw_qp **qps;
+};
+
+static int add_and_ref_qp(int id, void *p, void *data)
+{
+	struct qp_list *qp_listp = data;
+	struct c4iw_qp *qp = p;
+
+	c4iw_qp_add_ref(&qp->ibqp);
+	qp_listp->qps[qp_listp->idx++] = qp;
+	return 0;
+}
+
+static int count_qps(int id, void *p, void *data)
+{
+	unsigned *countp = data;
+	(*countp)++;
+	return 0;
+}
+
+static void deref_qps(struct qp_list *qp_list)
+{
+	int idx;
+
+	for (idx = 0; idx < qp_list->idx; idx++)
+		c4iw_qp_rem_ref(&qp_list->qps[idx]->ibqp);
+}
+
+static void recover_lost_dbs(struct uld_ctx *ctx, struct qp_list *qp_list)
+{
+	int idx;
+	int ret;
+
+	for (idx = 0; idx < qp_list->idx; idx++) {
+		struct c4iw_qp *qp = qp_list->qps[idx];
+
+		spin_lock_irq(&qp->rhp->lock);
+		spin_lock(&qp->lock);
+		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
+					  qp->wq.sq.qid,
+					  t4_sq_host_wq_pidx(&qp->wq),
+					  t4_sq_wq_size(&qp->wq));
+		if (ret) {
+			pr_err(KERN_ERR MOD "%s: Fatal error - "
+			       "DB overflow recovery failed - "
+			       "error syncing SQ qid %u\n",
+			       pci_name(ctx->lldi.pdev), qp->wq.sq.qid);
+			spin_unlock(&qp->lock);
+			spin_unlock_irq(&qp->rhp->lock);
+			return;
+		}
+		qp->wq.sq.wq_pidx_inc = 0;
+
+		ret = cxgb4_sync_txq_pidx(qp->rhp->rdev.lldi.ports[0],
+					  qp->wq.rq.qid,
+					  t4_rq_host_wq_pidx(&qp->wq),
+					  t4_rq_wq_size(&qp->wq));
+
+		if (ret) {
+			pr_err(KERN_ERR MOD "%s: Fatal error - "
+			       "DB overflow recovery failed - "
+			       "error syncing RQ qid %u\n",
+			       pci_name(ctx->lldi.pdev), qp->wq.rq.qid);
+			spin_unlock(&qp->lock);
+			spin_unlock_irq(&qp->rhp->lock);
+			return;
+		}
+		qp->wq.rq.wq_pidx_inc = 0;
+		spin_unlock(&qp->lock);
+		spin_unlock_irq(&qp->rhp->lock);
+
+		/* Wait for the dbfifo to drain */
+		while (cxgb4_dbfifo_count(qp->rhp->rdev.lldi.ports[0], 1) > 0) {
+			set_current_state(TASK_UNINTERRUPTIBLE);
+			schedule_timeout(usecs_to_jiffies(10));
+		}
+	}
+}
+
+static void recover_queues(struct uld_ctx *ctx)
+{
+	int count = 0;
+	struct qp_list qp_list;
+	int ret;
+
+	/* slow everybody down */
+	set_current_state(TASK_UNINTERRUPTIBLE);
+	schedule_timeout(usecs_to_jiffies(1000));
+
+	/* flush the SGE contexts */
+	ret = cxgb4_flush_eq_cache(ctx->dev->rdev.lldi.ports[0]);
+	if (ret) {
+		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
+		       pci_name(ctx->lldi.pdev));
+		return;
+	}
+
+	/* Count active queues so we can build a list of queues to recover */
+	spin_lock_irq(&ctx->dev->lock);
+	WARN_ON(ctx->dev->db_state != STOPPED);
+	ctx->dev->db_state = RECOVERY;
+	idr_for_each(&ctx->dev->qpidr, count_qps, &count);
+
+	qp_list.qps = kzalloc(count * sizeof *qp_list.qps, GFP_ATOMIC);
+	if (!qp_list.qps) {
+		printk(KERN_ERR MOD "%s: Fatal error - DB overflow recovery failed\n",
+		       pci_name(ctx->lldi.pdev));
+		spin_unlock_irq(&ctx->dev->lock);
+		return;
+	}
+	qp_list.idx = 0;
+
+	/* add and ref each qp so it doesn't get freed */
+	idr_for_each(&ctx->dev->qpidr, add_and_ref_qp, &qp_list);
+
+	spin_unlock_irq(&ctx->dev->lock);
+
+	/* now traverse the list in a safe context to recover the db state*/
+	recover_lost_dbs(ctx, &qp_list);
+
+	/* we're almost done!  deref the qps and clean up */
+	deref_qps(&qp_list);
+	kfree(qp_list.qps);
+
+	spin_lock_irq(&ctx->dev->lock);
+	WARN_ON(ctx->dev->db_state != RECOVERY);
+	ctx->dev->db_state = STOPPED;
+	spin_unlock_irq(&ctx->dev->lock);
+}
+
+static int c4iw_uld_control(void *handle, enum cxgb4_control control, ...)
+{
+	struct uld_ctx *ctx = handle;
+
+	switch (control) {
+	case CXGB4_CONTROL_DB_FULL:
+		stop_queues(ctx);
+		ctx->dev->rdev.stats.db_full++;
+		break;
+	case CXGB4_CONTROL_DB_EMPTY:
+		resume_queues(ctx);
+		mutex_lock(&ctx->dev->rdev.stats.lock);
+		ctx->dev->rdev.stats.db_empty++;
+		mutex_unlock(&ctx->dev->rdev.stats.lock);
+		break;
+	case CXGB4_CONTROL_DB_DROP:
+		recover_queues(ctx);
+		mutex_lock(&ctx->dev->rdev.stats.lock);
+		ctx->dev->rdev.stats.db_drop++;
+		mutex_unlock(&ctx->dev->rdev.stats.lock);
+		break;
+	default:
+		printk(KERN_WARNING MOD "%s: unknown control cmd %u\n",
+		       pci_name(ctx->lldi.pdev), control);
+		break;
+	}
+	return 0;
+}
+
+static struct cxgb4_uld_info c4iw_uld_info = {
+	.name = DRV_NAME,
+	.add = c4iw_uld_add,
+	.rx_handler = c4iw_uld_rx_handler,
+	.state_change = c4iw_uld_state_change,
+	.control = c4iw_uld_control,
+};
+
+static int __init c4iw_init_module(void)
+{
+	int err;
+
+	err = c4iw_cm_init();
+	if (err)
+		return err;
+
+	c4iw_debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
+	if (!c4iw_debugfs_root)
+		printk(KERN_WARNING MOD
+		       "could not create debugfs entry, continuing\n");
+
+	if (ibnl_add_client(RDMA_NL_C4IW, RDMA_NL_IWPM_NUM_OPS,
+			    c4iw_nl_cb_table))
+		pr_err("%s[%u]: Failed to add netlink callback\n"
+		       , __func__, __LINE__);
+
+	err = iwpm_init(RDMA_NL_C4IW);
+	if (err) {
+		pr_err("port mapper initialization failed with %d\n", err);
+		ibnl_remove_client(RDMA_NL_C4IW);
+		c4iw_cm_term();
+		debugfs_remove_recursive(c4iw_debugfs_root);
+		return err;
+	}
+
+	cxgb4_register_uld(CXGB4_ULD_RDMA, &c4iw_uld_info);
+
+	return 0;
+}
+
+static void __exit c4iw_exit_module(void)
+{
+	struct uld_ctx *ctx, *tmp;
+
+	mutex_lock(&dev_mutex);
+	list_for_each_entry_safe(ctx, tmp, &uld_ctx_list, entry) {
+		if (ctx->dev)
+			c4iw_remove(ctx);
+		kfree(ctx);
+	}
+	mutex_unlock(&dev_mutex);
+	cxgb4_unregister_uld(CXGB4_ULD_RDMA);
+	iwpm_exit(RDMA_NL_C4IW);
+	ibnl_remove_client(RDMA_NL_C4IW);
+	c4iw_cm_term();
+	debugfs_remove_recursive(c4iw_debugfs_root);
+}
+
+module_init(c4iw_init_module);
+module_exit(c4iw_exit_module);
diff --git a/drivers/infiniband/hw/cxgb4/ev.c b/drivers/infiniband/hw/cxgb4/ev.c
new file mode 100644
index 00000000000..d61d0a18f78
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/ev.c
@@ -0,0 +1,191 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/slab.h>
+#include <linux/mman.h>
+#include <net/sock.h>
+
+#include "iw_cxgb4.h"
+
+static void post_qp_event(struct c4iw_dev *dev, struct c4iw_cq *chp,
+			  struct c4iw_qp *qhp,
+			  struct t4_cqe *err_cqe,
+			  enum ib_event_type ib_event)
+{
+	struct ib_event event;
+	struct c4iw_qp_attributes attrs;
+	unsigned long flag;
+
+	printk(KERN_ERR MOD "AE qpid 0x%x opcode %d status 0x%x "
+	       "type %d wrid.hi 0x%x wrid.lo 0x%x\n",
+	       CQE_QPID(err_cqe), CQE_OPCODE(err_cqe),
+	       CQE_STATUS(err_cqe), CQE_TYPE(err_cqe),
+	       CQE_WRID_HI(err_cqe), CQE_WRID_LOW(err_cqe));
+
+	if (qhp->attr.state == C4IW_QP_STATE_RTS) {
+		attrs.next_state = C4IW_QP_STATE_TERMINATE;
+		c4iw_modify_qp(qhp->rhp, qhp, C4IW_QP_ATTR_NEXT_STATE,
+			       &attrs, 0);
+	}
+
+	event.event = ib_event;
+	event.device = chp->ibcq.device;
+	if (ib_event == IB_EVENT_CQ_ERR)
+		event.element.cq = &chp->ibcq;
+	else
+		event.element.qp = &qhp->ibqp;
+	if (qhp->ibqp.event_handler)
+		(*qhp->ibqp.event_handler)(&event, qhp->ibqp.qp_context);
+
+	spin_lock_irqsave(&chp->comp_handler_lock, flag);
+	(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+	spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
+}
+
+void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe)
+{
+	struct c4iw_cq *chp;
+	struct c4iw_qp *qhp;
+	u32 cqid;
+
+	spin_lock_irq(&dev->lock);
+	qhp = get_qhp(dev, CQE_QPID(err_cqe));
+	if (!qhp) {
+		printk(KERN_ERR MOD "BAD AE qpid 0x%x opcode %d "
+		       "status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
+		       CQE_QPID(err_cqe),
+		       CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
+		       CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
+		       CQE_WRID_LOW(err_cqe));
+		spin_unlock_irq(&dev->lock);
+		goto out;
+	}
+
+	if (SQ_TYPE(err_cqe))
+		cqid = qhp->attr.scq;
+	else
+		cqid = qhp->attr.rcq;
+	chp = get_chp(dev, cqid);
+	if (!chp) {
+		printk(KERN_ERR MOD "BAD AE cqid 0x%x qpid 0x%x opcode %d "
+		       "status 0x%x type %d wrid.hi 0x%x wrid.lo 0x%x\n",
+		       cqid, CQE_QPID(err_cqe),
+		       CQE_OPCODE(err_cqe), CQE_STATUS(err_cqe),
+		       CQE_TYPE(err_cqe), CQE_WRID_HI(err_cqe),
+		       CQE_WRID_LOW(err_cqe));
+		spin_unlock_irq(&dev->lock);
+		goto out;
+	}
+
+	c4iw_qp_add_ref(&qhp->ibqp);
+	atomic_inc(&chp->refcnt);
+	spin_unlock_irq(&dev->lock);
+
+	/* Bad incoming write */
+	if (RQ_TYPE(err_cqe) &&
+	    (CQE_OPCODE(err_cqe) == FW_RI_RDMA_WRITE)) {
+		post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_REQ_ERR);
+		goto done;
+	}
+
+	switch (CQE_STATUS(err_cqe)) {
+
+	/* Completion Events */
+	case T4_ERR_SUCCESS:
+		printk(KERN_ERR MOD "AE with status 0!\n");
+		break;
+
+	case T4_ERR_STAG:
+	case T4_ERR_PDID:
+	case T4_ERR_QPID:
+	case T4_ERR_ACCESS:
+	case T4_ERR_WRAP:
+	case T4_ERR_BOUND:
+	case T4_ERR_INVALIDATE_SHARED_MR:
+	case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
+		post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_ACCESS_ERR);
+		break;
+
+	/* Device Fatal Errors */
+	case T4_ERR_ECC:
+	case T4_ERR_ECC_PSTAG:
+	case T4_ERR_INTERNAL_ERR:
+		post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_DEVICE_FATAL);
+		break;
+
+	/* QP Fatal Errors */
+	case T4_ERR_OUT_OF_RQE:
+	case T4_ERR_PBL_ADDR_BOUND:
+	case T4_ERR_CRC:
+	case T4_ERR_MARKER:
+	case T4_ERR_PDU_LEN_ERR:
+	case T4_ERR_DDP_VERSION:
+	case T4_ERR_RDMA_VERSION:
+	case T4_ERR_OPCODE:
+	case T4_ERR_DDP_QUEUE_NUM:
+	case T4_ERR_MSN:
+	case T4_ERR_TBIT:
+	case T4_ERR_MO:
+	case T4_ERR_MSN_GAP:
+	case T4_ERR_MSN_RANGE:
+	case T4_ERR_RQE_ADDR_BOUND:
+	case T4_ERR_IRD_OVERFLOW:
+		post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
+		break;
+
+	default:
+		printk(KERN_ERR MOD "Unknown T4 status 0x%x QPID 0x%x\n",
+		       CQE_STATUS(err_cqe), qhp->wq.sq.qid);
+		post_qp_event(dev, chp, qhp, err_cqe, IB_EVENT_QP_FATAL);
+		break;
+	}
+done:
+	if (atomic_dec_and_test(&chp->refcnt))
+		wake_up(&chp->wait);
+	c4iw_qp_rem_ref(&qhp->ibqp);
+out:
+	return;
+}
+
+int c4iw_ev_handler(struct c4iw_dev *dev, u32 qid)
+{
+	struct c4iw_cq *chp;
+	unsigned long flag;
+
+	chp = get_chp(dev, qid);
+	if (chp) {
+		spin_lock_irqsave(&chp->comp_handler_lock, flag);
+		(*chp->ibcq.comp_handler)(&chp->ibcq, chp->ibcq.cq_context);
+		spin_unlock_irqrestore(&chp->comp_handler_lock, flag);
+	} else
+		PDBG("%s unknown cqid 0x%x\n", __func__, qid);
+	return 0;
+}
diff --git a/drivers/infiniband/hw/cxgb4/id_table.c b/drivers/infiniband/hw/cxgb4/id_table.c
new file mode 100644
index 00000000000..0161ae6ad62
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/id_table.c
@@ -0,0 +1,112 @@
+/*
+ * Copyright (c) 2011 Chelsio Communications.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/kernel.h>
+#include <linux/random.h>
+#include "iw_cxgb4.h"
+
+#define RANDOM_SKIP 16
+
+/*
+ * Trivial bitmap-based allocator. If the random flag is set, the
+ * allocator is designed to:
+ * - pseudo-randomize the id returned such that it is not trivially predictable.
+ * - avoid reuse of recently used id (at the expense of predictability)
+ */
+u32 c4iw_id_alloc(struct c4iw_id_table *alloc)
+{
+	unsigned long flags;
+	u32 obj;
+
+	spin_lock_irqsave(&alloc->lock, flags);
+
+	obj = find_next_zero_bit(alloc->table, alloc->max, alloc->last);
+	if (obj >= alloc->max)
+		obj = find_first_zero_bit(alloc->table, alloc->max);
+
+	if (obj < alloc->max) {
+		if (alloc->flags & C4IW_ID_TABLE_F_RANDOM)
+			alloc->last += prandom_u32() % RANDOM_SKIP;
+		else
+			alloc->last = obj + 1;
+		if (alloc->last >= alloc->max)
+			alloc->last = 0;
+		set_bit(obj, alloc->table);
+		obj += alloc->start;
+	} else
+		obj = -1;
+
+	spin_unlock_irqrestore(&alloc->lock, flags);
+	return obj;
+}
+
+void c4iw_id_free(struct c4iw_id_table *alloc, u32 obj)
+{
+	unsigned long flags;
+
+	obj -= alloc->start;
+	BUG_ON((int)obj < 0);
+
+	spin_lock_irqsave(&alloc->lock, flags);
+	clear_bit(obj, alloc->table);
+	spin_unlock_irqrestore(&alloc->lock, flags);
+}
+
+int c4iw_id_table_alloc(struct c4iw_id_table *alloc, u32 start, u32 num,
+			u32 reserved, u32 flags)
+{
+	int i;
+
+	alloc->start = start;
+	alloc->flags = flags;
+	if (flags & C4IW_ID_TABLE_F_RANDOM)
+		alloc->last = prandom_u32() % RANDOM_SKIP;
+	else
+		alloc->last = 0;
+	alloc->max  = num;
+	spin_lock_init(&alloc->lock);
+	alloc->table = kmalloc(BITS_TO_LONGS(num) * sizeof(long),
+				GFP_KERNEL);
+	if (!alloc->table)
+		return -ENOMEM;
+
+	bitmap_zero(alloc->table, num);
+	if (!(alloc->flags & C4IW_ID_TABLE_F_EMPTY))
+		for (i = 0; i < reserved; ++i)
+			set_bit(i, alloc->table);
+
+	return 0;
+}
+
+void c4iw_id_table_free(struct c4iw_id_table *alloc)
+{
+	kfree(alloc->table);
+}
diff --git a/drivers/infiniband/hw/cxgb4/iw_cxgb4.h b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
new file mode 100644
index 00000000000..361fff7a074
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/iw_cxgb4.h
@@ -0,0 +1,1000 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer.
+ *      - Redistributions in binary form must reproduce the above
+ *	  copyright notice, this list of conditions and the following
+ *	  disclaimer in the documentation and/or other materials
+ *	  provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __IW_CXGB4_H__
+#define __IW_CXGB4_H__
+
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/idr.h>
+#include <linux/completion.h>
+#include <linux/netdevice.h>
+#include <linux/sched.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/inet.h>
+#include <linux/wait.h>
+#include <linux/kref.h>
+#include <linux/timer.h>
+#include <linux/io.h>
+
+#include <asm/byteorder.h>
+
+#include <net/net_namespace.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/rdma_netlink.h>
+#include <rdma/iw_portmap.h>
+
+#include "cxgb4.h"
+#include "cxgb4_uld.h"
+#include "l2t.h"
+#include "user.h"
+
+#define DRV_NAME "iw_cxgb4"
+#define MOD DRV_NAME ":"
+
+extern int c4iw_debug;
+#define PDBG(fmt, args...) \
+do { \
+	if (c4iw_debug) \
+		printk(MOD fmt, ## args); \
+} while (0)
+
+#include "t4.h"
+
+#define PBL_OFF(rdev_p, a) ((a) - (rdev_p)->lldi.vr->pbl.start)
+#define RQT_OFF(rdev_p, a) ((a) - (rdev_p)->lldi.vr->rq.start)
+
+static inline void *cplhdr(struct sk_buff *skb)
+{
+	return skb->data;
+}
+
+#define C4IW_ID_TABLE_F_RANDOM 1       /* Pseudo-randomize the id's returned */
+#define C4IW_ID_TABLE_F_EMPTY  2       /* Table is initially empty */
+
+struct c4iw_id_table {
+	u32 flags;
+	u32 start;              /* logical minimal id */
+	u32 last;               /* hint for find */
+	u32 max;
+	spinlock_t lock;
+	unsigned long *table;
+};
+
+struct c4iw_resource {
+	struct c4iw_id_table tpt_table;
+	struct c4iw_id_table qid_table;
+	struct c4iw_id_table pdid_table;
+};
+
+struct c4iw_qid_list {
+	struct list_head entry;
+	u32 qid;
+};
+
+struct c4iw_dev_ucontext {
+	struct list_head qpids;
+	struct list_head cqids;
+	struct mutex lock;
+};
+
+enum c4iw_rdev_flags {
+	T4_FATAL_ERROR = (1<<0),
+	T4_STATUS_PAGE_DISABLED = (1<<1),
+};
+
+struct c4iw_stat {
+	u64 total;
+	u64 cur;
+	u64 max;
+	u64 fail;
+};
+
+struct c4iw_stats {
+	struct mutex lock;
+	struct c4iw_stat qid;
+	struct c4iw_stat pd;
+	struct c4iw_stat stag;
+	struct c4iw_stat pbl;
+	struct c4iw_stat rqt;
+	struct c4iw_stat ocqp;
+	u64  db_full;
+	u64  db_empty;
+	u64  db_drop;
+	u64  db_state_transitions;
+	u64  db_fc_interruptions;
+	u64  tcam_full;
+	u64  act_ofld_conn_fails;
+	u64  pas_ofld_conn_fails;
+};
+
+struct c4iw_rdev {
+	struct c4iw_resource resource;
+	unsigned long qpshift;
+	u32 qpmask;
+	unsigned long cqshift;
+	u32 cqmask;
+	struct c4iw_dev_ucontext uctx;
+	struct gen_pool *pbl_pool;
+	struct gen_pool *rqt_pool;
+	struct gen_pool *ocqp_pool;
+	u32 flags;
+	struct cxgb4_lld_info lldi;
+	unsigned long bar2_pa;
+	void __iomem *bar2_kva;
+	unsigned long oc_mw_pa;
+	void __iomem *oc_mw_kva;
+	struct c4iw_stats stats;
+	struct t4_dev_status_page *status_page;
+};
+
+static inline int c4iw_fatal_error(struct c4iw_rdev *rdev)
+{
+	return rdev->flags & T4_FATAL_ERROR;
+}
+
+static inline int c4iw_num_stags(struct c4iw_rdev *rdev)
+{
+	return min((int)T4_MAX_NUM_STAG, (int)(rdev->lldi.vr->stag.size >> 5));
+}
+
+#define C4IW_WR_TO (30*HZ)
+
+struct c4iw_wr_wait {
+	struct completion completion;
+	int ret;
+};
+
+static inline void c4iw_init_wr_wait(struct c4iw_wr_wait *wr_waitp)
+{
+	wr_waitp->ret = 0;
+	init_completion(&wr_waitp->completion);
+}
+
+static inline void c4iw_wake_up(struct c4iw_wr_wait *wr_waitp, int ret)
+{
+	wr_waitp->ret = ret;
+	complete(&wr_waitp->completion);
+}
+
+static inline int c4iw_wait_for_reply(struct c4iw_rdev *rdev,
+				 struct c4iw_wr_wait *wr_waitp,
+				 u32 hwtid, u32 qpid,
+				 const char *func)
+{
+	unsigned to = C4IW_WR_TO;
+	int ret;
+
+	do {
+		ret = wait_for_completion_timeout(&wr_waitp->completion, to);
+		if (!ret) {
+			printk(KERN_ERR MOD "%s - Device %s not responding - "
+			       "tid %u qpid %u\n", func,
+			       pci_name(rdev->lldi.pdev), hwtid, qpid);
+			if (c4iw_fatal_error(rdev)) {
+				wr_waitp->ret = -EIO;
+				break;
+			}
+			to = to << 2;
+		}
+	} while (!ret);
+	if (wr_waitp->ret)
+		PDBG("%s: FW reply %d tid %u qpid %u\n",
+		     pci_name(rdev->lldi.pdev), wr_waitp->ret, hwtid, qpid);
+	return wr_waitp->ret;
+}
+
+enum db_state {
+	NORMAL = 0,
+	FLOW_CONTROL = 1,
+	RECOVERY = 2,
+	STOPPED = 3
+};
+
+struct c4iw_dev {
+	struct ib_device ibdev;
+	struct c4iw_rdev rdev;
+	u32 device_cap_flags;
+	struct idr cqidr;
+	struct idr qpidr;
+	struct idr mmidr;
+	spinlock_t lock;
+	struct mutex db_mutex;
+	struct dentry *debugfs_root;
+	enum db_state db_state;
+	struct idr hwtid_idr;
+	struct idr atid_idr;
+	struct idr stid_idr;
+	struct list_head db_fc_list;
+};
+
+static inline struct c4iw_dev *to_c4iw_dev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct c4iw_dev, ibdev);
+}
+
+static inline struct c4iw_dev *rdev_to_c4iw_dev(struct c4iw_rdev *rdev)
+{
+	return container_of(rdev, struct c4iw_dev, rdev);
+}
+
+static inline struct c4iw_cq *get_chp(struct c4iw_dev *rhp, u32 cqid)
+{
+	return idr_find(&rhp->cqidr, cqid);
+}
+
+static inline struct c4iw_qp *get_qhp(struct c4iw_dev *rhp, u32 qpid)
+{
+	return idr_find(&rhp->qpidr, qpid);
+}
+
+static inline struct c4iw_mr *get_mhp(struct c4iw_dev *rhp, u32 mmid)
+{
+	return idr_find(&rhp->mmidr, mmid);
+}
+
+static inline int _insert_handle(struct c4iw_dev *rhp, struct idr *idr,
+				 void *handle, u32 id, int lock)
+{
+	int ret;
+
+	if (lock) {
+		idr_preload(GFP_KERNEL);
+		spin_lock_irq(&rhp->lock);
+	}
+
+	ret = idr_alloc(idr, handle, id, id + 1, GFP_ATOMIC);
+
+	if (lock) {
+		spin_unlock_irq(&rhp->lock);
+		idr_preload_end();
+	}
+
+	BUG_ON(ret == -ENOSPC);
+	return ret < 0 ? ret : 0;
+}
+
+static inline int insert_handle(struct c4iw_dev *rhp, struct idr *idr,
+				void *handle, u32 id)
+{
+	return _insert_handle(rhp, idr, handle, id, 1);
+}
+
+static inline int insert_handle_nolock(struct c4iw_dev *rhp, struct idr *idr,
+				       void *handle, u32 id)
+{
+	return _insert_handle(rhp, idr, handle, id, 0);
+}
+
+static inline void _remove_handle(struct c4iw_dev *rhp, struct idr *idr,
+				   u32 id, int lock)
+{
+	if (lock)
+		spin_lock_irq(&rhp->lock);
+	idr_remove(idr, id);
+	if (lock)
+		spin_unlock_irq(&rhp->lock);
+}
+
+static inline void remove_handle(struct c4iw_dev *rhp, struct idr *idr, u32 id)
+{
+	_remove_handle(rhp, idr, id, 1);
+}
+
+static inline void remove_handle_nolock(struct c4iw_dev *rhp,
+					 struct idr *idr, u32 id)
+{
+	_remove_handle(rhp, idr, id, 0);
+}
+
+struct c4iw_pd {
+	struct ib_pd ibpd;
+	u32 pdid;
+	struct c4iw_dev *rhp;
+};
+
+static inline struct c4iw_pd *to_c4iw_pd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct c4iw_pd, ibpd);
+}
+
+struct tpt_attributes {
+	u64 len;
+	u64 va_fbo;
+	enum fw_ri_mem_perms perms;
+	u32 stag;
+	u32 pdid;
+	u32 qpid;
+	u32 pbl_addr;
+	u32 pbl_size;
+	u32 state:1;
+	u32 type:2;
+	u32 rsvd:1;
+	u32 remote_invaliate_disable:1;
+	u32 zbva:1;
+	u32 mw_bind_enable:1;
+	u32 page_size:5;
+};
+
+struct c4iw_mr {
+	struct ib_mr ibmr;
+	struct ib_umem *umem;
+	struct c4iw_dev *rhp;
+	u64 kva;
+	struct tpt_attributes attr;
+};
+
+static inline struct c4iw_mr *to_c4iw_mr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct c4iw_mr, ibmr);
+}
+
+struct c4iw_mw {
+	struct ib_mw ibmw;
+	struct c4iw_dev *rhp;
+	u64 kva;
+	struct tpt_attributes attr;
+};
+
+static inline struct c4iw_mw *to_c4iw_mw(struct ib_mw *ibmw)
+{
+	return container_of(ibmw, struct c4iw_mw, ibmw);
+}
+
+struct c4iw_fr_page_list {
+	struct ib_fast_reg_page_list ibpl;
+	DEFINE_DMA_UNMAP_ADDR(mapping);
+	dma_addr_t dma_addr;
+	struct c4iw_dev *dev;
+	int pll_len;
+};
+
+static inline struct c4iw_fr_page_list *to_c4iw_fr_page_list(
+					struct ib_fast_reg_page_list *ibpl)
+{
+	return container_of(ibpl, struct c4iw_fr_page_list, ibpl);
+}
+
+struct c4iw_cq {
+	struct ib_cq ibcq;
+	struct c4iw_dev *rhp;
+	struct t4_cq cq;
+	spinlock_t lock;
+	spinlock_t comp_handler_lock;
+	atomic_t refcnt;
+	wait_queue_head_t wait;
+};
+
+static inline struct c4iw_cq *to_c4iw_cq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct c4iw_cq, ibcq);
+}
+
+struct c4iw_mpa_attributes {
+	u8 initiator;
+	u8 recv_marker_enabled;
+	u8 xmit_marker_enabled;
+	u8 crc_enabled;
+	u8 enhanced_rdma_conn;
+	u8 version;
+	u8 p2p_type;
+};
+
+struct c4iw_qp_attributes {
+	u32 scq;
+	u32 rcq;
+	u32 sq_num_entries;
+	u32 rq_num_entries;
+	u32 sq_max_sges;
+	u32 sq_max_sges_rdma_write;
+	u32 rq_max_sges;
+	u32 state;
+	u8 enable_rdma_read;
+	u8 enable_rdma_write;
+	u8 enable_bind;
+	u8 enable_mmid0_fastreg;
+	u32 max_ord;
+	u32 max_ird;
+	u32 pd;
+	u32 next_state;
+	char terminate_buffer[52];
+	u32 terminate_msg_len;
+	u8 is_terminate_local;
+	struct c4iw_mpa_attributes mpa_attr;
+	struct c4iw_ep *llp_stream_handle;
+	u8 layer_etype;
+	u8 ecode;
+	u16 sq_db_inc;
+	u16 rq_db_inc;
+	u8 send_term;
+};
+
+struct c4iw_qp {
+	struct ib_qp ibqp;
+	struct list_head db_fc_entry;
+	struct c4iw_dev *rhp;
+	struct c4iw_ep *ep;
+	struct c4iw_qp_attributes attr;
+	struct t4_wq wq;
+	spinlock_t lock;
+	struct mutex mutex;
+	atomic_t refcnt;
+	wait_queue_head_t wait;
+	struct timer_list timer;
+	int sq_sig_all;
+};
+
+static inline struct c4iw_qp *to_c4iw_qp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct c4iw_qp, ibqp);
+}
+
+struct c4iw_ucontext {
+	struct ib_ucontext ibucontext;
+	struct c4iw_dev_ucontext uctx;
+	u32 key;
+	spinlock_t mmap_lock;
+	struct list_head mmaps;
+};
+
+static inline struct c4iw_ucontext *to_c4iw_ucontext(struct ib_ucontext *c)
+{
+	return container_of(c, struct c4iw_ucontext, ibucontext);
+}
+
+struct c4iw_mm_entry {
+	struct list_head entry;
+	u64 addr;
+	u32 key;
+	unsigned len;
+};
+
+static inline struct c4iw_mm_entry *remove_mmap(struct c4iw_ucontext *ucontext,
+						u32 key, unsigned len)
+{
+	struct list_head *pos, *nxt;
+	struct c4iw_mm_entry *mm;
+
+	spin_lock(&ucontext->mmap_lock);
+	list_for_each_safe(pos, nxt, &ucontext->mmaps) {
+
+		mm = list_entry(pos, struct c4iw_mm_entry, entry);
+		if (mm->key == key && mm->len == len) {
+			list_del_init(&mm->entry);
+			spin_unlock(&ucontext->mmap_lock);
+			PDBG("%s key 0x%x addr 0x%llx len %d\n", __func__,
+			     key, (unsigned long long) mm->addr, mm->len);
+			return mm;
+		}
+	}
+	spin_unlock(&ucontext->mmap_lock);
+	return NULL;
+}
+
+static inline void insert_mmap(struct c4iw_ucontext *ucontext,
+			       struct c4iw_mm_entry *mm)
+{
+	spin_lock(&ucontext->mmap_lock);
+	PDBG("%s key 0x%x addr 0x%llx len %d\n", __func__,
+	     mm->key, (unsigned long long) mm->addr, mm->len);
+	list_add_tail(&mm->entry, &ucontext->mmaps);
+	spin_unlock(&ucontext->mmap_lock);
+}
+
+enum c4iw_qp_attr_mask {
+	C4IW_QP_ATTR_NEXT_STATE = 1 << 0,
+	C4IW_QP_ATTR_SQ_DB = 1<<1,
+	C4IW_QP_ATTR_RQ_DB = 1<<2,
+	C4IW_QP_ATTR_ENABLE_RDMA_READ = 1 << 7,
+	C4IW_QP_ATTR_ENABLE_RDMA_WRITE = 1 << 8,
+	C4IW_QP_ATTR_ENABLE_RDMA_BIND = 1 << 9,
+	C4IW_QP_ATTR_MAX_ORD = 1 << 11,
+	C4IW_QP_ATTR_MAX_IRD = 1 << 12,
+	C4IW_QP_ATTR_LLP_STREAM_HANDLE = 1 << 22,
+	C4IW_QP_ATTR_STREAM_MSG_BUFFER = 1 << 23,
+	C4IW_QP_ATTR_MPA_ATTR = 1 << 24,
+	C4IW_QP_ATTR_QP_CONTEXT_ACTIVATE = 1 << 25,
+	C4IW_QP_ATTR_VALID_MODIFY = (C4IW_QP_ATTR_ENABLE_RDMA_READ |
+				     C4IW_QP_ATTR_ENABLE_RDMA_WRITE |
+				     C4IW_QP_ATTR_MAX_ORD |
+				     C4IW_QP_ATTR_MAX_IRD |
+				     C4IW_QP_ATTR_LLP_STREAM_HANDLE |
+				     C4IW_QP_ATTR_STREAM_MSG_BUFFER |
+				     C4IW_QP_ATTR_MPA_ATTR |
+				     C4IW_QP_ATTR_QP_CONTEXT_ACTIVATE)
+};
+
+int c4iw_modify_qp(struct c4iw_dev *rhp,
+				struct c4iw_qp *qhp,
+				enum c4iw_qp_attr_mask mask,
+				struct c4iw_qp_attributes *attrs,
+				int internal);
+
+enum c4iw_qp_state {
+	C4IW_QP_STATE_IDLE,
+	C4IW_QP_STATE_RTS,
+	C4IW_QP_STATE_ERROR,
+	C4IW_QP_STATE_TERMINATE,
+	C4IW_QP_STATE_CLOSING,
+	C4IW_QP_STATE_TOT
+};
+
+static inline int c4iw_convert_state(enum ib_qp_state ib_state)
+{
+	switch (ib_state) {
+	case IB_QPS_RESET:
+	case IB_QPS_INIT:
+		return C4IW_QP_STATE_IDLE;
+	case IB_QPS_RTS:
+		return C4IW_QP_STATE_RTS;
+	case IB_QPS_SQD:
+		return C4IW_QP_STATE_CLOSING;
+	case IB_QPS_SQE:
+		return C4IW_QP_STATE_TERMINATE;
+	case IB_QPS_ERR:
+		return C4IW_QP_STATE_ERROR;
+	default:
+		return -1;
+	}
+}
+
+static inline int to_ib_qp_state(int c4iw_qp_state)
+{
+	switch (c4iw_qp_state) {
+	case C4IW_QP_STATE_IDLE:
+		return IB_QPS_INIT;
+	case C4IW_QP_STATE_RTS:
+		return IB_QPS_RTS;
+	case C4IW_QP_STATE_CLOSING:
+		return IB_QPS_SQD;
+	case C4IW_QP_STATE_TERMINATE:
+		return IB_QPS_SQE;
+	case C4IW_QP_STATE_ERROR:
+		return IB_QPS_ERR;
+	}
+	return IB_QPS_ERR;
+}
+
+static inline u32 c4iw_ib_to_tpt_access(int a)
+{
+	return (a & IB_ACCESS_REMOTE_WRITE ? FW_RI_MEM_ACCESS_REM_WRITE : 0) |
+	       (a & IB_ACCESS_REMOTE_READ ? FW_RI_MEM_ACCESS_REM_READ : 0) |
+	       (a & IB_ACCESS_LOCAL_WRITE ? FW_RI_MEM_ACCESS_LOCAL_WRITE : 0) |
+	       FW_RI_MEM_ACCESS_LOCAL_READ;
+}
+
+static inline u32 c4iw_ib_to_tpt_bind_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_WRITE ? FW_RI_MEM_ACCESS_REM_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ ? FW_RI_MEM_ACCESS_REM_READ : 0);
+}
+
+enum c4iw_mmid_state {
+	C4IW_STAG_STATE_VALID,
+	C4IW_STAG_STATE_INVALID
+};
+
+#define C4IW_NODE_DESC "cxgb4 Chelsio Communications"
+
+#define MPA_KEY_REQ "MPA ID Req Frame"
+#define MPA_KEY_REP "MPA ID Rep Frame"
+
+#define MPA_MAX_PRIVATE_DATA	256
+#define MPA_ENHANCED_RDMA_CONN	0x10
+#define MPA_REJECT		0x20
+#define MPA_CRC			0x40
+#define MPA_MARKERS		0x80
+#define MPA_FLAGS_MASK		0xE0
+
+#define MPA_V2_PEER2PEER_MODEL          0x8000
+#define MPA_V2_ZERO_LEN_FPDU_RTR        0x4000
+#define MPA_V2_RDMA_WRITE_RTR           0x8000
+#define MPA_V2_RDMA_READ_RTR            0x4000
+#define MPA_V2_IRD_ORD_MASK             0x3FFF
+
+#define c4iw_put_ep(ep) { \
+	PDBG("put_ep (via %s:%u) ep %p refcnt %d\n", __func__, __LINE__,  \
+	     ep, atomic_read(&((ep)->kref.refcount))); \
+	WARN_ON(atomic_read(&((ep)->kref.refcount)) < 1); \
+	kref_put(&((ep)->kref), _c4iw_free_ep); \
+}
+
+#define c4iw_get_ep(ep) { \
+	PDBG("get_ep (via %s:%u) ep %p, refcnt %d\n", __func__, __LINE__, \
+	     ep, atomic_read(&((ep)->kref.refcount))); \
+	kref_get(&((ep)->kref));  \
+}
+void _c4iw_free_ep(struct kref *kref);
+
+struct mpa_message {
+	u8 key[16];
+	u8 flags;
+	u8 revision;
+	__be16 private_data_size;
+	u8 private_data[0];
+};
+
+struct mpa_v2_conn_params {
+	__be16 ird;
+	__be16 ord;
+};
+
+struct terminate_message {
+	u8 layer_etype;
+	u8 ecode;
+	__be16 hdrct_rsvd;
+	u8 len_hdrs[0];
+};
+
+#define TERM_MAX_LENGTH (sizeof(struct terminate_message) + 2 + 18 + 28)
+
+enum c4iw_layers_types {
+	LAYER_RDMAP		= 0x00,
+	LAYER_DDP		= 0x10,
+	LAYER_MPA		= 0x20,
+	RDMAP_LOCAL_CATA	= 0x00,
+	RDMAP_REMOTE_PROT	= 0x01,
+	RDMAP_REMOTE_OP		= 0x02,
+	DDP_LOCAL_CATA		= 0x00,
+	DDP_TAGGED_ERR		= 0x01,
+	DDP_UNTAGGED_ERR	= 0x02,
+	DDP_LLP			= 0x03
+};
+
+enum c4iw_rdma_ecodes {
+	RDMAP_INV_STAG		= 0x00,
+	RDMAP_BASE_BOUNDS	= 0x01,
+	RDMAP_ACC_VIOL		= 0x02,
+	RDMAP_STAG_NOT_ASSOC	= 0x03,
+	RDMAP_TO_WRAP		= 0x04,
+	RDMAP_INV_VERS		= 0x05,
+	RDMAP_INV_OPCODE	= 0x06,
+	RDMAP_STREAM_CATA	= 0x07,
+	RDMAP_GLOBAL_CATA	= 0x08,
+	RDMAP_CANT_INV_STAG	= 0x09,
+	RDMAP_UNSPECIFIED	= 0xff
+};
+
+enum c4iw_ddp_ecodes {
+	DDPT_INV_STAG		= 0x00,
+	DDPT_BASE_BOUNDS	= 0x01,
+	DDPT_STAG_NOT_ASSOC	= 0x02,
+	DDPT_TO_WRAP		= 0x03,
+	DDPT_INV_VERS		= 0x04,
+	DDPU_INV_QN		= 0x01,
+	DDPU_INV_MSN_NOBUF	= 0x02,
+	DDPU_INV_MSN_RANGE	= 0x03,
+	DDPU_INV_MO		= 0x04,
+	DDPU_MSG_TOOBIG		= 0x05,
+	DDPU_INV_VERS		= 0x06
+};
+
+enum c4iw_mpa_ecodes {
+	MPA_CRC_ERR		= 0x02,
+	MPA_MARKER_ERR          = 0x03,
+	MPA_LOCAL_CATA          = 0x05,
+	MPA_INSUFF_IRD          = 0x06,
+	MPA_NOMATCH_RTR         = 0x07,
+};
+
+enum c4iw_ep_state {
+	IDLE = 0,
+	LISTEN,
+	CONNECTING,
+	MPA_REQ_WAIT,
+	MPA_REQ_SENT,
+	MPA_REQ_RCVD,
+	MPA_REP_SENT,
+	FPDU_MODE,
+	ABORTING,
+	CLOSING,
+	MORIBUND,
+	DEAD,
+};
+
+enum c4iw_ep_flags {
+	PEER_ABORT_IN_PROGRESS	= 0,
+	ABORT_REQ_IN_PROGRESS	= 1,
+	RELEASE_RESOURCES	= 2,
+	CLOSE_SENT		= 3,
+	TIMEOUT                 = 4,
+	QP_REFERENCED           = 5,
+	RELEASE_MAPINFO		= 6,
+};
+
+enum c4iw_ep_history {
+	ACT_OPEN_REQ            = 0,
+	ACT_OFLD_CONN           = 1,
+	ACT_OPEN_RPL            = 2,
+	ACT_ESTAB               = 3,
+	PASS_ACCEPT_REQ         = 4,
+	PASS_ESTAB              = 5,
+	ABORT_UPCALL            = 6,
+	ESTAB_UPCALL            = 7,
+	CLOSE_UPCALL            = 8,
+	ULP_ACCEPT              = 9,
+	ULP_REJECT              = 10,
+	TIMEDOUT                = 11,
+	PEER_ABORT              = 12,
+	PEER_CLOSE              = 13,
+	CONNREQ_UPCALL          = 14,
+	ABORT_CONN              = 15,
+	DISCONN_UPCALL          = 16,
+	EP_DISC_CLOSE           = 17,
+	EP_DISC_ABORT           = 18,
+	CONN_RPL_UPCALL         = 19,
+	ACT_RETRY_NOMEM         = 20,
+	ACT_RETRY_INUSE         = 21
+};
+
+struct c4iw_ep_common {
+	struct iw_cm_id *cm_id;
+	struct c4iw_qp *qp;
+	struct c4iw_dev *dev;
+	enum c4iw_ep_state state;
+	struct kref kref;
+	struct mutex mutex;
+	struct sockaddr_storage local_addr;
+	struct sockaddr_storage remote_addr;
+	struct sockaddr_storage mapped_local_addr;
+	struct sockaddr_storage mapped_remote_addr;
+	struct c4iw_wr_wait wr_wait;
+	unsigned long flags;
+	unsigned long history;
+};
+
+struct c4iw_listen_ep {
+	struct c4iw_ep_common com;
+	unsigned int stid;
+	int backlog;
+};
+
+struct c4iw_ep {
+	struct c4iw_ep_common com;
+	struct c4iw_ep *parent_ep;
+	struct timer_list timer;
+	struct list_head entry;
+	unsigned int atid;
+	u32 hwtid;
+	u32 snd_seq;
+	u32 rcv_seq;
+	struct l2t_entry *l2t;
+	struct dst_entry *dst;
+	struct sk_buff *mpa_skb;
+	struct c4iw_mpa_attributes mpa_attr;
+	u8 mpa_pkt[sizeof(struct mpa_message) + MPA_MAX_PRIVATE_DATA];
+	unsigned int mpa_pkt_len;
+	u32 ird;
+	u32 ord;
+	u32 smac_idx;
+	u32 tx_chan;
+	u32 mtu;
+	u16 mss;
+	u16 emss;
+	u16 plen;
+	u16 rss_qid;
+	u16 txq_idx;
+	u16 ctrlq_idx;
+	u8 tos;
+	u8 retry_with_mpa_v1;
+	u8 tried_with_mpa_v1;
+	unsigned int retry_count;
+	int snd_win;
+	int rcv_win;
+};
+
+static inline void print_addr(struct c4iw_ep_common *epc, const char *func,
+			      const char *msg)
+{
+
+#define SINA(a) (&(((struct sockaddr_in *)(a))->sin_addr.s_addr))
+#define SINP(a) ntohs(((struct sockaddr_in *)(a))->sin_port)
+#define SIN6A(a) (&(((struct sockaddr_in6 *)(a))->sin6_addr))
+#define SIN6P(a) ntohs(((struct sockaddr_in6 *)(a))->sin6_port)
+
+	if (c4iw_debug) {
+		switch (epc->local_addr.ss_family) {
+		case AF_INET:
+			PDBG("%s %s %pI4:%u/%u <-> %pI4:%u/%u\n",
+			     func, msg, SINA(&epc->local_addr),
+			     SINP(&epc->local_addr),
+			     SINP(&epc->mapped_local_addr),
+			     SINA(&epc->remote_addr),
+			     SINP(&epc->remote_addr),
+			     SINP(&epc->mapped_remote_addr));
+			break;
+		case AF_INET6:
+			PDBG("%s %s %pI6:%u/%u <-> %pI6:%u/%u\n",
+			     func, msg, SIN6A(&epc->local_addr),
+			     SIN6P(&epc->local_addr),
+			     SIN6P(&epc->mapped_local_addr),
+			     SIN6A(&epc->remote_addr),
+			     SIN6P(&epc->remote_addr),
+			     SIN6P(&epc->mapped_remote_addr));
+			break;
+		default:
+			break;
+		}
+	}
+#undef SINA
+#undef SINP
+#undef SIN6A
+#undef SIN6P
+}
+
+static inline struct c4iw_ep *to_ep(struct iw_cm_id *cm_id)
+{
+	return cm_id->provider_data;
+}
+
+static inline struct c4iw_listen_ep *to_listen_ep(struct iw_cm_id *cm_id)
+{
+	return cm_id->provider_data;
+}
+
+static inline int compute_wscale(int win)
+{
+	int wscale = 0;
+
+	while (wscale < 14 && (65535<<wscale) < win)
+		wscale++;
+	return wscale;
+}
+
+static inline int ocqp_supported(const struct cxgb4_lld_info *infop)
+{
+#if defined(__i386__) || defined(__x86_64__) || defined(CONFIG_PPC64)
+	return infop->vr->ocq.size > 0;
+#else
+	return 0;
+#endif
+}
+
+u32 c4iw_id_alloc(struct c4iw_id_table *alloc);
+void c4iw_id_free(struct c4iw_id_table *alloc, u32 obj);
+int c4iw_id_table_alloc(struct c4iw_id_table *alloc, u32 start, u32 num,
+			u32 reserved, u32 flags);
+void c4iw_id_table_free(struct c4iw_id_table *alloc);
+
+typedef int (*c4iw_handler_func)(struct c4iw_dev *dev, struct sk_buff *skb);
+
+int c4iw_ep_redirect(void *ctx, struct dst_entry *old, struct dst_entry *new,
+		     struct l2t_entry *l2t);
+void c4iw_put_qpid(struct c4iw_rdev *rdev, u32 qpid,
+		   struct c4iw_dev_ucontext *uctx);
+u32 c4iw_get_resource(struct c4iw_id_table *id_table);
+void c4iw_put_resource(struct c4iw_id_table *id_table, u32 entry);
+int c4iw_init_resource(struct c4iw_rdev *rdev, u32 nr_tpt, u32 nr_pdid);
+int c4iw_init_ctrl_qp(struct c4iw_rdev *rdev);
+int c4iw_pblpool_create(struct c4iw_rdev *rdev);
+int c4iw_rqtpool_create(struct c4iw_rdev *rdev);
+int c4iw_ocqp_pool_create(struct c4iw_rdev *rdev);
+void c4iw_pblpool_destroy(struct c4iw_rdev *rdev);
+void c4iw_rqtpool_destroy(struct c4iw_rdev *rdev);
+void c4iw_ocqp_pool_destroy(struct c4iw_rdev *rdev);
+void c4iw_destroy_resource(struct c4iw_resource *rscp);
+int c4iw_destroy_ctrl_qp(struct c4iw_rdev *rdev);
+int c4iw_register_device(struct c4iw_dev *dev);
+void c4iw_unregister_device(struct c4iw_dev *dev);
+int __init c4iw_cm_init(void);
+void c4iw_cm_term(void);
+void c4iw_release_dev_ucontext(struct c4iw_rdev *rdev,
+			       struct c4iw_dev_ucontext *uctx);
+void c4iw_init_dev_ucontext(struct c4iw_rdev *rdev,
+			    struct c4iw_dev_ucontext *uctx);
+int c4iw_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr);
+int c4iw_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr);
+int c4iw_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
+		 struct ib_mw_bind *mw_bind);
+int c4iw_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param);
+int c4iw_create_listen(struct iw_cm_id *cm_id, int backlog);
+int c4iw_destroy_listen(struct iw_cm_id *cm_id);
+int c4iw_accept_cr(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param);
+int c4iw_reject_cr(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len);
+void c4iw_qp_add_ref(struct ib_qp *qp);
+void c4iw_qp_rem_ref(struct ib_qp *qp);
+void c4iw_free_fastreg_pbl(struct ib_fast_reg_page_list *page_list);
+struct ib_fast_reg_page_list *c4iw_alloc_fastreg_pbl(
+					struct ib_device *device,
+					int page_list_len);
+struct ib_mr *c4iw_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth);
+int c4iw_dealloc_mw(struct ib_mw *mw);
+struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
+struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start,
+					   u64 length, u64 virt, int acc,
+					   struct ib_udata *udata);
+struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc);
+struct ib_mr *c4iw_register_phys_mem(struct ib_pd *pd,
+					struct ib_phys_buf *buffer_list,
+					int num_phys_buf,
+					int acc,
+					u64 *iova_start);
+int c4iw_reregister_phys_mem(struct ib_mr *mr,
+				     int mr_rereg_mask,
+				     struct ib_pd *pd,
+				     struct ib_phys_buf *buffer_list,
+				     int num_phys_buf,
+				     int acc, u64 *iova_start);
+int c4iw_dereg_mr(struct ib_mr *ib_mr);
+int c4iw_destroy_cq(struct ib_cq *ib_cq);
+struct ib_cq *c4iw_create_cq(struct ib_device *ibdev, int entries,
+					int vector,
+					struct ib_ucontext *ib_context,
+					struct ib_udata *udata);
+int c4iw_resize_cq(struct ib_cq *cq, int cqe, struct ib_udata *udata);
+int c4iw_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+int c4iw_destroy_qp(struct ib_qp *ib_qp);
+struct ib_qp *c4iw_create_qp(struct ib_pd *pd,
+			     struct ib_qp_init_attr *attrs,
+			     struct ib_udata *udata);
+int c4iw_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+				 int attr_mask, struct ib_udata *udata);
+int c4iw_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_qp_init_attr *init_attr);
+struct ib_qp *c4iw_get_qp(struct ib_device *dev, int qpn);
+u32 c4iw_rqtpool_alloc(struct c4iw_rdev *rdev, int size);
+void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size);
+u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size);
+void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size);
+u32 c4iw_ocqp_pool_alloc(struct c4iw_rdev *rdev, int size);
+void c4iw_ocqp_pool_free(struct c4iw_rdev *rdev, u32 addr, int size);
+int c4iw_ofld_send(struct c4iw_rdev *rdev, struct sk_buff *skb);
+void c4iw_flush_hw_cq(struct c4iw_cq *chp);
+void c4iw_count_rcqes(struct t4_cq *cq, struct t4_wq *wq, int *count);
+int c4iw_ep_disconnect(struct c4iw_ep *ep, int abrupt, gfp_t gfp);
+int c4iw_flush_rq(struct t4_wq *wq, struct t4_cq *cq, int count);
+int c4iw_flush_sq(struct c4iw_qp *qhp);
+int c4iw_ev_handler(struct c4iw_dev *rnicp, u32 qid);
+u16 c4iw_rqes_posted(struct c4iw_qp *qhp);
+int c4iw_post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe);
+u32 c4iw_get_cqid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx);
+void c4iw_put_cqid(struct c4iw_rdev *rdev, u32 qid,
+		struct c4iw_dev_ucontext *uctx);
+u32 c4iw_get_qpid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx);
+void c4iw_put_qpid(struct c4iw_rdev *rdev, u32 qid,
+		struct c4iw_dev_ucontext *uctx);
+void c4iw_ev_dispatch(struct c4iw_dev *dev, struct t4_cqe *err_cqe);
+
+extern struct cxgb4_client t4c_client;
+extern c4iw_handler_func c4iw_handlers[NUM_CPL_CMDS];
+extern int c4iw_max_read_depth;
+extern int db_fc_threshold;
+extern int db_coalescing_threshold;
+extern int use_dsgl;
+
+
+#endif
diff --git a/drivers/infiniband/hw/cxgb4/mem.c b/drivers/infiniband/hw/cxgb4/mem.c
new file mode 100644
index 00000000000..ec7a2988a70
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/mem.c
@@ -0,0 +1,955 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <rdma/ib_umem.h>
+#include <linux/atomic.h>
+
+#include "iw_cxgb4.h"
+
+int use_dsgl = 0;
+module_param(use_dsgl, int, 0644);
+MODULE_PARM_DESC(use_dsgl, "Use DSGL for PBL/FastReg (default=0)");
+
+#define T4_ULPTX_MIN_IO 32
+#define C4IW_MAX_INLINE_SIZE 96
+#define T4_ULPTX_MAX_DMA 1024
+#define C4IW_INLINE_THRESHOLD 128
+
+static int inline_threshold = C4IW_INLINE_THRESHOLD;
+module_param(inline_threshold, int, 0644);
+MODULE_PARM_DESC(inline_threshold, "inline vs dsgl threshold (default=128)");
+
+static int _c4iw_write_mem_dma_aligned(struct c4iw_rdev *rdev, u32 addr,
+				       u32 len, dma_addr_t data, int wait)
+{
+	struct sk_buff *skb;
+	struct ulp_mem_io *req;
+	struct ulptx_sgl *sgl;
+	u8 wr_len;
+	int ret = 0;
+	struct c4iw_wr_wait wr_wait;
+
+	addr &= 0x7FFFFFF;
+
+	if (wait)
+		c4iw_init_wr_wait(&wr_wait);
+	wr_len = roundup(sizeof(*req) + sizeof(*sgl), 16);
+
+	skb = alloc_skb(wr_len, GFP_KERNEL | __GFP_NOFAIL);
+	if (!skb)
+		return -ENOMEM;
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+	req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
+	memset(req, 0, wr_len);
+	INIT_ULPTX_WR(req, wr_len, 0, 0);
+	req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR) |
+			(wait ? FW_WR_COMPL(1) : 0));
+	req->wr.wr_lo = wait ? (__force __be64)(unsigned long) &wr_wait : 0L;
+	req->wr.wr_mid = cpu_to_be32(FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
+	req->cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
+	req->cmd |= cpu_to_be32(V_T5_ULP_MEMIO_ORDER(1));
+	req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(len>>5));
+	req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr), 16));
+	req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR(addr));
+
+	sgl = (struct ulptx_sgl *)(req + 1);
+	sgl->cmd_nsge = cpu_to_be32(ULPTX_CMD(ULP_TX_SC_DSGL) |
+				    ULPTX_NSGE(1));
+	sgl->len0 = cpu_to_be32(len);
+	sgl->addr0 = cpu_to_be64(data);
+
+	ret = c4iw_ofld_send(rdev, skb);
+	if (ret)
+		return ret;
+	if (wait)
+		ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
+	return ret;
+}
+
+static int _c4iw_write_mem_inline(struct c4iw_rdev *rdev, u32 addr, u32 len,
+				  void *data)
+{
+	struct sk_buff *skb;
+	struct ulp_mem_io *req;
+	struct ulptx_idata *sc;
+	u8 wr_len, *to_dp, *from_dp;
+	int copy_len, num_wqe, i, ret = 0;
+	struct c4iw_wr_wait wr_wait;
+	__be32 cmd = cpu_to_be32(ULPTX_CMD(ULP_TX_MEM_WRITE));
+
+	if (is_t4(rdev->lldi.adapter_type))
+		cmd |= cpu_to_be32(ULP_MEMIO_ORDER(1));
+	else
+		cmd |= cpu_to_be32(V_T5_ULP_MEMIO_IMM(1));
+
+	addr &= 0x7FFFFFF;
+	PDBG("%s addr 0x%x len %u\n", __func__, addr, len);
+	num_wqe = DIV_ROUND_UP(len, C4IW_MAX_INLINE_SIZE);
+	c4iw_init_wr_wait(&wr_wait);
+	for (i = 0; i < num_wqe; i++) {
+
+		copy_len = len > C4IW_MAX_INLINE_SIZE ? C4IW_MAX_INLINE_SIZE :
+			   len;
+		wr_len = roundup(sizeof *req + sizeof *sc +
+				 roundup(copy_len, T4_ULPTX_MIN_IO), 16);
+
+		skb = alloc_skb(wr_len, GFP_KERNEL);
+		if (!skb)
+			return -ENOMEM;
+		set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+		req = (struct ulp_mem_io *)__skb_put(skb, wr_len);
+		memset(req, 0, wr_len);
+		INIT_ULPTX_WR(req, wr_len, 0, 0);
+
+		if (i == (num_wqe-1)) {
+			req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR) |
+						    FW_WR_COMPL(1));
+			req->wr.wr_lo = (__force __be64)(unsigned long) &wr_wait;
+		} else
+			req->wr.wr_hi = cpu_to_be32(FW_WR_OP(FW_ULPTX_WR));
+		req->wr.wr_mid = cpu_to_be32(
+				       FW_WR_LEN16(DIV_ROUND_UP(wr_len, 16)));
+
+		req->cmd = cmd;
+		req->dlen = cpu_to_be32(ULP_MEMIO_DATA_LEN(
+				DIV_ROUND_UP(copy_len, T4_ULPTX_MIN_IO)));
+		req->len16 = cpu_to_be32(DIV_ROUND_UP(wr_len-sizeof(req->wr),
+						      16));
+		req->lock_addr = cpu_to_be32(ULP_MEMIO_ADDR(addr + i * 3));
+
+		sc = (struct ulptx_idata *)(req + 1);
+		sc->cmd_more = cpu_to_be32(ULPTX_CMD(ULP_TX_SC_IMM));
+		sc->len = cpu_to_be32(roundup(copy_len, T4_ULPTX_MIN_IO));
+
+		to_dp = (u8 *)(sc + 1);
+		from_dp = (u8 *)data + i * C4IW_MAX_INLINE_SIZE;
+		if (data)
+			memcpy(to_dp, from_dp, copy_len);
+		else
+			memset(to_dp, 0, copy_len);
+		if (copy_len % T4_ULPTX_MIN_IO)
+			memset(to_dp + copy_len, 0, T4_ULPTX_MIN_IO -
+			       (copy_len % T4_ULPTX_MIN_IO));
+		ret = c4iw_ofld_send(rdev, skb);
+		if (ret)
+			return ret;
+		len -= C4IW_MAX_INLINE_SIZE;
+	}
+
+	ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, 0, __func__);
+	return ret;
+}
+
+static int _c4iw_write_mem_dma(struct c4iw_rdev *rdev, u32 addr, u32 len, void *data)
+{
+	u32 remain = len;
+	u32 dmalen;
+	int ret = 0;
+	dma_addr_t daddr;
+	dma_addr_t save;
+
+	daddr = dma_map_single(&rdev->lldi.pdev->dev, data, len, DMA_TO_DEVICE);
+	if (dma_mapping_error(&rdev->lldi.pdev->dev, daddr))
+		return -1;
+	save = daddr;
+
+	while (remain > inline_threshold) {
+		if (remain < T4_ULPTX_MAX_DMA) {
+			if (remain & ~T4_ULPTX_MIN_IO)
+				dmalen = remain & ~(T4_ULPTX_MIN_IO-1);
+			else
+				dmalen = remain;
+		} else
+			dmalen = T4_ULPTX_MAX_DMA;
+		remain -= dmalen;
+		ret = _c4iw_write_mem_dma_aligned(rdev, addr, dmalen, daddr,
+						 !remain);
+		if (ret)
+			goto out;
+		addr += dmalen >> 5;
+		data += dmalen;
+		daddr += dmalen;
+	}
+	if (remain)
+		ret = _c4iw_write_mem_inline(rdev, addr, remain, data);
+out:
+	dma_unmap_single(&rdev->lldi.pdev->dev, save, len, DMA_TO_DEVICE);
+	return ret;
+}
+
+/*
+ * write len bytes of data into addr (32B aligned address)
+ * If data is NULL, clear len byte of memory to zero.
+ */
+static int write_adapter_mem(struct c4iw_rdev *rdev, u32 addr, u32 len,
+			     void *data)
+{
+	if (is_t5(rdev->lldi.adapter_type) && use_dsgl) {
+		if (len > inline_threshold) {
+			if (_c4iw_write_mem_dma(rdev, addr, len, data)) {
+				printk_ratelimited(KERN_WARNING
+						   "%s: dma map"
+						   " failure (non fatal)\n",
+						   pci_name(rdev->lldi.pdev));
+				return _c4iw_write_mem_inline(rdev, addr, len,
+							      data);
+			} else
+				return 0;
+		} else
+			return _c4iw_write_mem_inline(rdev, addr, len, data);
+	} else
+		return _c4iw_write_mem_inline(rdev, addr, len, data);
+}
+
+/*
+ * Build and write a TPT entry.
+ * IN: stag key, pdid, perm, bind_enabled, zbva, to, len, page_size,
+ *     pbl_size and pbl_addr
+ * OUT: stag index
+ */
+static int write_tpt_entry(struct c4iw_rdev *rdev, u32 reset_tpt_entry,
+			   u32 *stag, u8 stag_state, u32 pdid,
+			   enum fw_ri_stag_type type, enum fw_ri_mem_perms perm,
+			   int bind_enabled, u32 zbva, u64 to,
+			   u64 len, u8 page_size, u32 pbl_size, u32 pbl_addr)
+{
+	int err;
+	struct fw_ri_tpte tpt;
+	u32 stag_idx;
+	static atomic_t key;
+
+	if (c4iw_fatal_error(rdev))
+		return -EIO;
+
+	stag_state = stag_state > 0;
+	stag_idx = (*stag) >> 8;
+
+	if ((!reset_tpt_entry) && (*stag == T4_STAG_UNSET)) {
+		stag_idx = c4iw_get_resource(&rdev->resource.tpt_table);
+		if (!stag_idx) {
+			mutex_lock(&rdev->stats.lock);
+			rdev->stats.stag.fail++;
+			mutex_unlock(&rdev->stats.lock);
+			return -ENOMEM;
+		}
+		mutex_lock(&rdev->stats.lock);
+		rdev->stats.stag.cur += 32;
+		if (rdev->stats.stag.cur > rdev->stats.stag.max)
+			rdev->stats.stag.max = rdev->stats.stag.cur;
+		mutex_unlock(&rdev->stats.lock);
+		*stag = (stag_idx << 8) | (atomic_inc_return(&key) & 0xff);
+	}
+	PDBG("%s stag_state 0x%0x type 0x%0x pdid 0x%0x, stag_idx 0x%x\n",
+	     __func__, stag_state, type, pdid, stag_idx);
+
+	/* write TPT entry */
+	if (reset_tpt_entry)
+		memset(&tpt, 0, sizeof(tpt));
+	else {
+		tpt.valid_to_pdid = cpu_to_be32(F_FW_RI_TPTE_VALID |
+			V_FW_RI_TPTE_STAGKEY((*stag & M_FW_RI_TPTE_STAGKEY)) |
+			V_FW_RI_TPTE_STAGSTATE(stag_state) |
+			V_FW_RI_TPTE_STAGTYPE(type) | V_FW_RI_TPTE_PDID(pdid));
+		tpt.locread_to_qpid = cpu_to_be32(V_FW_RI_TPTE_PERM(perm) |
+			(bind_enabled ? F_FW_RI_TPTE_MWBINDEN : 0) |
+			V_FW_RI_TPTE_ADDRTYPE((zbva ? FW_RI_ZERO_BASED_TO :
+						      FW_RI_VA_BASED_TO))|
+			V_FW_RI_TPTE_PS(page_size));
+		tpt.nosnoop_pbladdr = !pbl_size ? 0 : cpu_to_be32(
+			V_FW_RI_TPTE_PBLADDR(PBL_OFF(rdev, pbl_addr)>>3));
+		tpt.len_lo = cpu_to_be32((u32)(len & 0xffffffffUL));
+		tpt.va_hi = cpu_to_be32((u32)(to >> 32));
+		tpt.va_lo_fbo = cpu_to_be32((u32)(to & 0xffffffffUL));
+		tpt.dca_mwbcnt_pstag = cpu_to_be32(0);
+		tpt.len_hi = cpu_to_be32((u32)(len >> 32));
+	}
+	err = write_adapter_mem(rdev, stag_idx +
+				(rdev->lldi.vr->stag.start >> 5),
+				sizeof(tpt), &tpt);
+
+	if (reset_tpt_entry) {
+		c4iw_put_resource(&rdev->resource.tpt_table, stag_idx);
+		mutex_lock(&rdev->stats.lock);
+		rdev->stats.stag.cur -= 32;
+		mutex_unlock(&rdev->stats.lock);
+	}
+	return err;
+}
+
+static int write_pbl(struct c4iw_rdev *rdev, __be64 *pbl,
+		     u32 pbl_addr, u32 pbl_size)
+{
+	int err;
+
+	PDBG("%s *pdb_addr 0x%x, pbl_base 0x%x, pbl_size %d\n",
+	     __func__, pbl_addr, rdev->lldi.vr->pbl.start,
+	     pbl_size);
+
+	err = write_adapter_mem(rdev, pbl_addr >> 5, pbl_size << 3, pbl);
+	return err;
+}
+
+static int dereg_mem(struct c4iw_rdev *rdev, u32 stag, u32 pbl_size,
+		     u32 pbl_addr)
+{
+	return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0,
+			       pbl_size, pbl_addr);
+}
+
+static int allocate_window(struct c4iw_rdev *rdev, u32 * stag, u32 pdid)
+{
+	*stag = T4_STAG_UNSET;
+	return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_MW, 0, 0, 0,
+			       0UL, 0, 0, 0, 0);
+}
+
+static int deallocate_window(struct c4iw_rdev *rdev, u32 stag)
+{
+	return write_tpt_entry(rdev, 1, &stag, 0, 0, 0, 0, 0, 0, 0UL, 0, 0, 0,
+			       0);
+}
+
+static int allocate_stag(struct c4iw_rdev *rdev, u32 *stag, u32 pdid,
+			 u32 pbl_size, u32 pbl_addr)
+{
+	*stag = T4_STAG_UNSET;
+	return write_tpt_entry(rdev, 0, stag, 0, pdid, FW_RI_STAG_NSMR, 0, 0, 0,
+			       0UL, 0, 0, pbl_size, pbl_addr);
+}
+
+static int finish_mem_reg(struct c4iw_mr *mhp, u32 stag)
+{
+	u32 mmid;
+
+	mhp->attr.state = 1;
+	mhp->attr.stag = stag;
+	mmid = stag >> 8;
+	mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
+	PDBG("%s mmid 0x%x mhp %p\n", __func__, mmid, mhp);
+	return insert_handle(mhp->rhp, &mhp->rhp->mmidr, mhp, mmid);
+}
+
+static int register_mem(struct c4iw_dev *rhp, struct c4iw_pd *php,
+		      struct c4iw_mr *mhp, int shift)
+{
+	u32 stag = T4_STAG_UNSET;
+	int ret;
+
+	ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid,
+			      FW_RI_STAG_NSMR, mhp->attr.perms,
+			      mhp->attr.mw_bind_enable, mhp->attr.zbva,
+			      mhp->attr.va_fbo, mhp->attr.len, shift - 12,
+			      mhp->attr.pbl_size, mhp->attr.pbl_addr);
+	if (ret)
+		return ret;
+
+	ret = finish_mem_reg(mhp, stag);
+	if (ret)
+		dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+	return ret;
+}
+
+static int reregister_mem(struct c4iw_dev *rhp, struct c4iw_pd *php,
+			  struct c4iw_mr *mhp, int shift, int npages)
+{
+	u32 stag;
+	int ret;
+
+	if (npages > mhp->attr.pbl_size)
+		return -ENOMEM;
+
+	stag = mhp->attr.stag;
+	ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, mhp->attr.pdid,
+			      FW_RI_STAG_NSMR, mhp->attr.perms,
+			      mhp->attr.mw_bind_enable, mhp->attr.zbva,
+			      mhp->attr.va_fbo, mhp->attr.len, shift - 12,
+			      mhp->attr.pbl_size, mhp->attr.pbl_addr);
+	if (ret)
+		return ret;
+
+	ret = finish_mem_reg(mhp, stag);
+	if (ret)
+		dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+
+	return ret;
+}
+
+static int alloc_pbl(struct c4iw_mr *mhp, int npages)
+{
+	mhp->attr.pbl_addr = c4iw_pblpool_alloc(&mhp->rhp->rdev,
+						    npages << 3);
+
+	if (!mhp->attr.pbl_addr)
+		return -ENOMEM;
+
+	mhp->attr.pbl_size = npages;
+
+	return 0;
+}
+
+static int build_phys_page_list(struct ib_phys_buf *buffer_list,
+				int num_phys_buf, u64 *iova_start,
+				u64 *total_size, int *npages,
+				int *shift, __be64 **page_list)
+{
+	u64 mask;
+	int i, j, n;
+
+	mask = 0;
+	*total_size = 0;
+	for (i = 0; i < num_phys_buf; ++i) {
+		if (i != 0 && buffer_list[i].addr & ~PAGE_MASK)
+			return -EINVAL;
+		if (i != 0 && i != num_phys_buf - 1 &&
+		    (buffer_list[i].size & ~PAGE_MASK))
+			return -EINVAL;
+		*total_size += buffer_list[i].size;
+		if (i > 0)
+			mask |= buffer_list[i].addr;
+		else
+			mask |= buffer_list[i].addr & PAGE_MASK;
+		if (i != num_phys_buf - 1)
+			mask |= buffer_list[i].addr + buffer_list[i].size;
+		else
+			mask |= (buffer_list[i].addr + buffer_list[i].size +
+				PAGE_SIZE - 1) & PAGE_MASK;
+	}
+
+	if (*total_size > 0xFFFFFFFFULL)
+		return -ENOMEM;
+
+	/* Find largest page shift we can use to cover buffers */
+	for (*shift = PAGE_SHIFT; *shift < 27; ++(*shift))
+		if ((1ULL << *shift) & mask)
+			break;
+
+	buffer_list[0].size += buffer_list[0].addr & ((1ULL << *shift) - 1);
+	buffer_list[0].addr &= ~0ull << *shift;
+
+	*npages = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		*npages += (buffer_list[i].size +
+			(1ULL << *shift) - 1) >> *shift;
+
+	if (!*npages)
+		return -EINVAL;
+
+	*page_list = kmalloc(sizeof(u64) * *npages, GFP_KERNEL);
+	if (!*page_list)
+		return -ENOMEM;
+
+	n = 0;
+	for (i = 0; i < num_phys_buf; ++i)
+		for (j = 0;
+		     j < (buffer_list[i].size + (1ULL << *shift) - 1) >> *shift;
+		     ++j)
+			(*page_list)[n++] = cpu_to_be64(buffer_list[i].addr +
+			    ((u64) j << *shift));
+
+	PDBG("%s va 0x%llx mask 0x%llx shift %d len %lld pbl_size %d\n",
+	     __func__, (unsigned long long)*iova_start,
+	     (unsigned long long)mask, *shift, (unsigned long long)*total_size,
+	     *npages);
+
+	return 0;
+
+}
+
+int c4iw_reregister_phys_mem(struct ib_mr *mr, int mr_rereg_mask,
+			     struct ib_pd *pd, struct ib_phys_buf *buffer_list,
+			     int num_phys_buf, int acc, u64 *iova_start)
+{
+
+	struct c4iw_mr mh, *mhp;
+	struct c4iw_pd *php;
+	struct c4iw_dev *rhp;
+	__be64 *page_list = NULL;
+	int shift = 0;
+	u64 total_size;
+	int npages;
+	int ret;
+
+	PDBG("%s ib_mr %p ib_pd %p\n", __func__, mr, pd);
+
+	/* There can be no memory windows */
+	if (atomic_read(&mr->usecnt))
+		return -EINVAL;
+
+	mhp = to_c4iw_mr(mr);
+	rhp = mhp->rhp;
+	php = to_c4iw_pd(mr->pd);
+
+	/* make sure we are on the same adapter */
+	if (rhp != php->rhp)
+		return -EINVAL;
+
+	memcpy(&mh, mhp, sizeof *mhp);
+
+	if (mr_rereg_mask & IB_MR_REREG_PD)
+		php = to_c4iw_pd(pd);
+	if (mr_rereg_mask & IB_MR_REREG_ACCESS) {
+		mh.attr.perms = c4iw_ib_to_tpt_access(acc);
+		mh.attr.mw_bind_enable = (acc & IB_ACCESS_MW_BIND) ==
+					 IB_ACCESS_MW_BIND;
+	}
+	if (mr_rereg_mask & IB_MR_REREG_TRANS) {
+		ret = build_phys_page_list(buffer_list, num_phys_buf,
+						iova_start,
+						&total_size, &npages,
+						&shift, &page_list);
+		if (ret)
+			return ret;
+	}
+
+	ret = reregister_mem(rhp, php, &mh, shift, npages);
+	kfree(page_list);
+	if (ret)
+		return ret;
+	if (mr_rereg_mask & IB_MR_REREG_PD)
+		mhp->attr.pdid = php->pdid;
+	if (mr_rereg_mask & IB_MR_REREG_ACCESS)
+		mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
+	if (mr_rereg_mask & IB_MR_REREG_TRANS) {
+		mhp->attr.zbva = 0;
+		mhp->attr.va_fbo = *iova_start;
+		mhp->attr.page_size = shift - 12;
+		mhp->attr.len = (u32) total_size;
+		mhp->attr.pbl_size = npages;
+	}
+
+	return 0;
+}
+
+struct ib_mr *c4iw_register_phys_mem(struct ib_pd *pd,
+				     struct ib_phys_buf *buffer_list,
+				     int num_phys_buf, int acc, u64 *iova_start)
+{
+	__be64 *page_list;
+	int shift;
+	u64 total_size;
+	int npages;
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+	struct c4iw_mr *mhp;
+	int ret;
+
+	PDBG("%s ib_pd %p\n", __func__, pd);
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+
+	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
+	if (!mhp)
+		return ERR_PTR(-ENOMEM);
+
+	mhp->rhp = rhp;
+
+	/* First check that we have enough alignment */
+	if ((*iova_start & ~PAGE_MASK) != (buffer_list[0].addr & ~PAGE_MASK)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	if (num_phys_buf > 1 &&
+	    ((buffer_list[0].addr + buffer_list[0].size) & ~PAGE_MASK)) {
+		ret = -EINVAL;
+		goto err;
+	}
+
+	ret = build_phys_page_list(buffer_list, num_phys_buf, iova_start,
+					&total_size, &npages, &shift,
+					&page_list);
+	if (ret)
+		goto err;
+
+	ret = alloc_pbl(mhp, npages);
+	if (ret) {
+		kfree(page_list);
+		goto err;
+	}
+
+	ret = write_pbl(&mhp->rhp->rdev, page_list, mhp->attr.pbl_addr,
+			     npages);
+	kfree(page_list);
+	if (ret)
+		goto err_pbl;
+
+	mhp->attr.pdid = php->pdid;
+	mhp->attr.zbva = 0;
+
+	mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
+	mhp->attr.va_fbo = *iova_start;
+	mhp->attr.page_size = shift - 12;
+
+	mhp->attr.len = (u32) total_size;
+	mhp->attr.pbl_size = npages;
+	ret = register_mem(rhp, php, mhp, shift);
+	if (ret)
+		goto err_pbl;
+
+	return &mhp->ibmr;
+
+err_pbl:
+	c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
+			      mhp->attr.pbl_size << 3);
+
+err:
+	kfree(mhp);
+	return ERR_PTR(ret);
+
+}
+
+struct ib_mr *c4iw_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+	struct c4iw_mr *mhp;
+	int ret;
+	u32 stag = T4_STAG_UNSET;
+
+	PDBG("%s ib_pd %p\n", __func__, pd);
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+
+	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
+	if (!mhp)
+		return ERR_PTR(-ENOMEM);
+
+	mhp->rhp = rhp;
+	mhp->attr.pdid = php->pdid;
+	mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
+	mhp->attr.mw_bind_enable = (acc&IB_ACCESS_MW_BIND) == IB_ACCESS_MW_BIND;
+	mhp->attr.zbva = 0;
+	mhp->attr.va_fbo = 0;
+	mhp->attr.page_size = 0;
+	mhp->attr.len = ~0UL;
+	mhp->attr.pbl_size = 0;
+
+	ret = write_tpt_entry(&rhp->rdev, 0, &stag, 1, php->pdid,
+			      FW_RI_STAG_NSMR, mhp->attr.perms,
+			      mhp->attr.mw_bind_enable, 0, 0, ~0UL, 0, 0, 0);
+	if (ret)
+		goto err1;
+
+	ret = finish_mem_reg(mhp, stag);
+	if (ret)
+		goto err2;
+	return &mhp->ibmr;
+err2:
+	dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		  mhp->attr.pbl_addr);
+err1:
+	kfree(mhp);
+	return ERR_PTR(ret);
+}
+
+struct ib_mr *c4iw_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			       u64 virt, int acc, struct ib_udata *udata)
+{
+	__be64 *pages;
+	int shift, n, len;
+	int i, k, entry;
+	int err = 0;
+	struct scatterlist *sg;
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+	struct c4iw_mr *mhp;
+
+	PDBG("%s ib_pd %p\n", __func__, pd);
+
+	if (length == ~0ULL)
+		return ERR_PTR(-EINVAL);
+
+	if ((length + start) < start)
+		return ERR_PTR(-EINVAL);
+
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
+	if (!mhp)
+		return ERR_PTR(-ENOMEM);
+
+	mhp->rhp = rhp;
+
+	mhp->umem = ib_umem_get(pd->uobject->context, start, length, acc, 0);
+	if (IS_ERR(mhp->umem)) {
+		err = PTR_ERR(mhp->umem);
+		kfree(mhp);
+		return ERR_PTR(err);
+	}
+
+	shift = ffs(mhp->umem->page_size) - 1;
+
+	n = mhp->umem->nmap;
+	err = alloc_pbl(mhp, n);
+	if (err)
+		goto err;
+
+	pages = (__be64 *) __get_free_page(GFP_KERNEL);
+	if (!pages) {
+		err = -ENOMEM;
+		goto err_pbl;
+	}
+
+	i = n = 0;
+
+	for_each_sg(mhp->umem->sg_head.sgl, sg, mhp->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = cpu_to_be64(sg_dma_address(sg) +
+				mhp->umem->page_size * k);
+			if (i == PAGE_SIZE / sizeof *pages) {
+				err = write_pbl(&mhp->rhp->rdev,
+				      pages,
+				      mhp->attr.pbl_addr + (n << 3), i);
+				if (err)
+					goto pbl_done;
+				n += i;
+				i = 0;
+			}
+		}
+	}
+
+	if (i)
+		err = write_pbl(&mhp->rhp->rdev, pages,
+				     mhp->attr.pbl_addr + (n << 3), i);
+
+pbl_done:
+	free_page((unsigned long) pages);
+	if (err)
+		goto err_pbl;
+
+	mhp->attr.pdid = php->pdid;
+	mhp->attr.zbva = 0;
+	mhp->attr.perms = c4iw_ib_to_tpt_access(acc);
+	mhp->attr.va_fbo = virt;
+	mhp->attr.page_size = shift - 12;
+	mhp->attr.len = length;
+
+	err = register_mem(rhp, php, mhp, shift);
+	if (err)
+		goto err_pbl;
+
+	return &mhp->ibmr;
+
+err_pbl:
+	c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
+			      mhp->attr.pbl_size << 3);
+
+err:
+	ib_umem_release(mhp->umem);
+	kfree(mhp);
+	return ERR_PTR(err);
+}
+
+struct ib_mw *c4iw_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+	struct c4iw_mw *mhp;
+	u32 mmid;
+	u32 stag = 0;
+	int ret;
+
+	if (type != IB_MW_TYPE_1)
+		return ERR_PTR(-EINVAL);
+
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
+	if (!mhp)
+		return ERR_PTR(-ENOMEM);
+	ret = allocate_window(&rhp->rdev, &stag, php->pdid);
+	if (ret) {
+		kfree(mhp);
+		return ERR_PTR(ret);
+	}
+	mhp->rhp = rhp;
+	mhp->attr.pdid = php->pdid;
+	mhp->attr.type = FW_RI_STAG_MW;
+	mhp->attr.stag = stag;
+	mmid = (stag) >> 8;
+	mhp->ibmw.rkey = stag;
+	if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
+		deallocate_window(&rhp->rdev, mhp->attr.stag);
+		kfree(mhp);
+		return ERR_PTR(-ENOMEM);
+	}
+	PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
+	return &(mhp->ibmw);
+}
+
+int c4iw_dealloc_mw(struct ib_mw *mw)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_mw *mhp;
+	u32 mmid;
+
+	mhp = to_c4iw_mw(mw);
+	rhp = mhp->rhp;
+	mmid = (mw->rkey) >> 8;
+	remove_handle(rhp, &rhp->mmidr, mmid);
+	deallocate_window(&rhp->rdev, mhp->attr.stag);
+	kfree(mhp);
+	PDBG("%s ib_mw %p mmid 0x%x ptr %p\n", __func__, mw, mmid, mhp);
+	return 0;
+}
+
+struct ib_mr *c4iw_alloc_fast_reg_mr(struct ib_pd *pd, int pbl_depth)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+	struct c4iw_mr *mhp;
+	u32 mmid;
+	u32 stag = 0;
+	int ret = 0;
+
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+	mhp = kzalloc(sizeof(*mhp), GFP_KERNEL);
+	if (!mhp) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	mhp->rhp = rhp;
+	ret = alloc_pbl(mhp, pbl_depth);
+	if (ret)
+		goto err1;
+	mhp->attr.pbl_size = pbl_depth;
+	ret = allocate_stag(&rhp->rdev, &stag, php->pdid,
+				 mhp->attr.pbl_size, mhp->attr.pbl_addr);
+	if (ret)
+		goto err2;
+	mhp->attr.pdid = php->pdid;
+	mhp->attr.type = FW_RI_STAG_NSMR;
+	mhp->attr.stag = stag;
+	mhp->attr.state = 1;
+	mmid = (stag) >> 8;
+	mhp->ibmr.rkey = mhp->ibmr.lkey = stag;
+	if (insert_handle(rhp, &rhp->mmidr, mhp, mmid)) {
+		ret = -ENOMEM;
+		goto err3;
+	}
+
+	PDBG("%s mmid 0x%x mhp %p stag 0x%x\n", __func__, mmid, mhp, stag);
+	return &(mhp->ibmr);
+err3:
+	dereg_mem(&rhp->rdev, stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+err2:
+	c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
+			      mhp->attr.pbl_size << 3);
+err1:
+	kfree(mhp);
+err:
+	return ERR_PTR(ret);
+}
+
+struct ib_fast_reg_page_list *c4iw_alloc_fastreg_pbl(struct ib_device *device,
+						     int page_list_len)
+{
+	struct c4iw_fr_page_list *c4pl;
+	struct c4iw_dev *dev = to_c4iw_dev(device);
+	dma_addr_t dma_addr;
+	int pll_len = roundup(page_list_len * sizeof(u64), 32);
+
+	c4pl = kmalloc(sizeof(*c4pl), GFP_KERNEL);
+	if (!c4pl)
+		return ERR_PTR(-ENOMEM);
+
+	c4pl->ibpl.page_list = dma_alloc_coherent(&dev->rdev.lldi.pdev->dev,
+						  pll_len, &dma_addr,
+						  GFP_KERNEL);
+	if (!c4pl->ibpl.page_list) {
+		kfree(c4pl);
+		return ERR_PTR(-ENOMEM);
+	}
+	dma_unmap_addr_set(c4pl, mapping, dma_addr);
+	c4pl->dma_addr = dma_addr;
+	c4pl->dev = dev;
+	c4pl->pll_len = pll_len;
+
+	PDBG("%s c4pl %p pll_len %u page_list %p dma_addr %pad\n",
+	     __func__, c4pl, c4pl->pll_len, c4pl->ibpl.page_list,
+	     &c4pl->dma_addr);
+
+	return &c4pl->ibpl;
+}
+
+void c4iw_free_fastreg_pbl(struct ib_fast_reg_page_list *ibpl)
+{
+	struct c4iw_fr_page_list *c4pl = to_c4iw_fr_page_list(ibpl);
+
+	PDBG("%s c4pl %p pll_len %u page_list %p dma_addr %pad\n",
+	     __func__, c4pl, c4pl->pll_len, c4pl->ibpl.page_list,
+	     &c4pl->dma_addr);
+
+	dma_free_coherent(&c4pl->dev->rdev.lldi.pdev->dev,
+			  c4pl->pll_len,
+			  c4pl->ibpl.page_list, dma_unmap_addr(c4pl, mapping));
+	kfree(c4pl);
+}
+
+int c4iw_dereg_mr(struct ib_mr *ib_mr)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_mr *mhp;
+	u32 mmid;
+
+	PDBG("%s ib_mr %p\n", __func__, ib_mr);
+	/* There can be no memory windows */
+	if (atomic_read(&ib_mr->usecnt))
+		return -EINVAL;
+
+	mhp = to_c4iw_mr(ib_mr);
+	rhp = mhp->rhp;
+	mmid = mhp->attr.stag >> 8;
+	remove_handle(rhp, &rhp->mmidr, mmid);
+	dereg_mem(&rhp->rdev, mhp->attr.stag, mhp->attr.pbl_size,
+		       mhp->attr.pbl_addr);
+	if (mhp->attr.pbl_size)
+		c4iw_pblpool_free(&mhp->rhp->rdev, mhp->attr.pbl_addr,
+				  mhp->attr.pbl_size << 3);
+	if (mhp->kva)
+		kfree((void *) (unsigned long) mhp->kva);
+	if (mhp->umem)
+		ib_umem_release(mhp->umem);
+	PDBG("%s mmid 0x%x ptr %p\n", __func__, mmid, mhp);
+	kfree(mhp);
+	return 0;
+}
diff --git a/drivers/infiniband/hw/cxgb4/provider.c b/drivers/infiniband/hw/cxgb4/provider.c
new file mode 100644
index 00000000000..b1d305338de
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/provider.c
@@ -0,0 +1,586 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/module.h>
+#include <linux/moduleparam.h>
+#include <linux/device.h>
+#include <linux/netdevice.h>
+#include <linux/etherdevice.h>
+#include <linux/delay.h>
+#include <linux/errno.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/ethtool.h>
+#include <linux/rtnetlink.h>
+#include <linux/inetdevice.h>
+#include <linux/io.h>
+
+#include <asm/irq.h>
+#include <asm/byteorder.h>
+
+#include <rdma/iw_cm.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "iw_cxgb4.h"
+
+static int fastreg_support = 1;
+module_param(fastreg_support, int, 0644);
+MODULE_PARM_DESC(fastreg_support, "Advertise fastreg support (default=1)");
+
+static struct ib_ah *c4iw_ah_create(struct ib_pd *pd,
+				    struct ib_ah_attr *ah_attr)
+{
+	return ERR_PTR(-ENOSYS);
+}
+
+static int c4iw_ah_destroy(struct ib_ah *ah)
+{
+	return -ENOSYS;
+}
+
+static int c4iw_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	return -ENOSYS;
+}
+
+static int c4iw_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	return -ENOSYS;
+}
+
+static int c4iw_process_mad(struct ib_device *ibdev, int mad_flags,
+			    u8 port_num, struct ib_wc *in_wc,
+			    struct ib_grh *in_grh, struct ib_mad *in_mad,
+			    struct ib_mad *out_mad)
+{
+	return -ENOSYS;
+}
+
+static int c4iw_dealloc_ucontext(struct ib_ucontext *context)
+{
+	struct c4iw_dev *rhp = to_c4iw_dev(context->device);
+	struct c4iw_ucontext *ucontext = to_c4iw_ucontext(context);
+	struct c4iw_mm_entry *mm, *tmp;
+
+	PDBG("%s context %p\n", __func__, context);
+	list_for_each_entry_safe(mm, tmp, &ucontext->mmaps, entry)
+		kfree(mm);
+	c4iw_release_dev_ucontext(&rhp->rdev, &ucontext->uctx);
+	kfree(ucontext);
+	return 0;
+}
+
+static struct ib_ucontext *c4iw_alloc_ucontext(struct ib_device *ibdev,
+					       struct ib_udata *udata)
+{
+	struct c4iw_ucontext *context;
+	struct c4iw_dev *rhp = to_c4iw_dev(ibdev);
+	static int warned;
+	struct c4iw_alloc_ucontext_resp uresp;
+	int ret = 0;
+	struct c4iw_mm_entry *mm = NULL;
+
+	PDBG("%s ibdev %p\n", __func__, ibdev);
+	context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (!context) {
+		ret = -ENOMEM;
+		goto err;
+	}
+
+	c4iw_init_dev_ucontext(&rhp->rdev, &context->uctx);
+	INIT_LIST_HEAD(&context->mmaps);
+	spin_lock_init(&context->mmap_lock);
+
+	if (udata->outlen < sizeof(uresp) - sizeof(uresp.reserved)) {
+		if (!warned++)
+			pr_err(MOD "Warning - downlevel libcxgb4 (non-fatal), device status page disabled.");
+		rhp->rdev.flags |= T4_STATUS_PAGE_DISABLED;
+	} else {
+		mm = kmalloc(sizeof(*mm), GFP_KERNEL);
+		if (!mm) {
+			ret = -ENOMEM;
+			goto err_free;
+		}
+
+		uresp.status_page_size = PAGE_SIZE;
+
+		spin_lock(&context->mmap_lock);
+		uresp.status_page_key = context->key;
+		context->key += PAGE_SIZE;
+		spin_unlock(&context->mmap_lock);
+
+		ret = ib_copy_to_udata(udata, &uresp,
+				       sizeof(uresp) - sizeof(uresp.reserved));
+		if (ret)
+			goto err_mm;
+
+		mm->key = uresp.status_page_key;
+		mm->addr = virt_to_phys(rhp->rdev.status_page);
+		mm->len = PAGE_SIZE;
+		insert_mmap(context, mm);
+	}
+	return &context->ibucontext;
+err_mm:
+	kfree(mm);
+err_free:
+	kfree(context);
+err:
+	return ERR_PTR(ret);
+}
+
+static int c4iw_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+	int len = vma->vm_end - vma->vm_start;
+	u32 key = vma->vm_pgoff << PAGE_SHIFT;
+	struct c4iw_rdev *rdev;
+	int ret = 0;
+	struct c4iw_mm_entry *mm;
+	struct c4iw_ucontext *ucontext;
+	u64 addr;
+
+	PDBG("%s pgoff 0x%lx key 0x%x len %d\n", __func__, vma->vm_pgoff,
+	     key, len);
+
+	if (vma->vm_start & (PAGE_SIZE-1))
+		return -EINVAL;
+
+	rdev = &(to_c4iw_dev(context->device)->rdev);
+	ucontext = to_c4iw_ucontext(context);
+
+	mm = remove_mmap(ucontext, key, len);
+	if (!mm)
+		return -EINVAL;
+	addr = mm->addr;
+	kfree(mm);
+
+	if ((addr >= pci_resource_start(rdev->lldi.pdev, 0)) &&
+	    (addr < (pci_resource_start(rdev->lldi.pdev, 0) +
+		    pci_resource_len(rdev->lldi.pdev, 0)))) {
+
+		/*
+		 * MA_SYNC register...
+		 */
+		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+		ret = io_remap_pfn_range(vma, vma->vm_start,
+					 addr >> PAGE_SHIFT,
+					 len, vma->vm_page_prot);
+	} else if ((addr >= pci_resource_start(rdev->lldi.pdev, 2)) &&
+		   (addr < (pci_resource_start(rdev->lldi.pdev, 2) +
+		    pci_resource_len(rdev->lldi.pdev, 2)))) {
+
+		/*
+		 * Map user DB or OCQP memory...
+		 */
+		if (addr >= rdev->oc_mw_pa)
+			vma->vm_page_prot = t4_pgprot_wc(vma->vm_page_prot);
+		else {
+			if (is_t5(rdev->lldi.adapter_type))
+				vma->vm_page_prot =
+					t4_pgprot_wc(vma->vm_page_prot);
+			else
+				vma->vm_page_prot =
+					pgprot_noncached(vma->vm_page_prot);
+		}
+		ret = io_remap_pfn_range(vma, vma->vm_start,
+					 addr >> PAGE_SHIFT,
+					 len, vma->vm_page_prot);
+	} else {
+
+		/*
+		 * Map WQ or CQ contig dma memory...
+		 */
+		ret = remap_pfn_range(vma, vma->vm_start,
+				      addr >> PAGE_SHIFT,
+				      len, vma->vm_page_prot);
+	}
+
+	return ret;
+}
+
+static int c4iw_deallocate_pd(struct ib_pd *pd)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_pd *php;
+
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+	PDBG("%s ibpd %p pdid 0x%x\n", __func__, pd, php->pdid);
+	c4iw_put_resource(&rhp->rdev.resource.pdid_table, php->pdid);
+	mutex_lock(&rhp->rdev.stats.lock);
+	rhp->rdev.stats.pd.cur--;
+	mutex_unlock(&rhp->rdev.stats.lock);
+	kfree(php);
+	return 0;
+}
+
+static struct ib_pd *c4iw_allocate_pd(struct ib_device *ibdev,
+				      struct ib_ucontext *context,
+				      struct ib_udata *udata)
+{
+	struct c4iw_pd *php;
+	u32 pdid;
+	struct c4iw_dev *rhp;
+
+	PDBG("%s ibdev %p\n", __func__, ibdev);
+	rhp = (struct c4iw_dev *) ibdev;
+	pdid =  c4iw_get_resource(&rhp->rdev.resource.pdid_table);
+	if (!pdid)
+		return ERR_PTR(-EINVAL);
+	php = kzalloc(sizeof(*php), GFP_KERNEL);
+	if (!php) {
+		c4iw_put_resource(&rhp->rdev.resource.pdid_table, pdid);
+		return ERR_PTR(-ENOMEM);
+	}
+	php->pdid = pdid;
+	php->rhp = rhp;
+	if (context) {
+		if (ib_copy_to_udata(udata, &php->pdid, sizeof(u32))) {
+			c4iw_deallocate_pd(&php->ibpd);
+			return ERR_PTR(-EFAULT);
+		}
+	}
+	mutex_lock(&rhp->rdev.stats.lock);
+	rhp->rdev.stats.pd.cur++;
+	if (rhp->rdev.stats.pd.cur > rhp->rdev.stats.pd.max)
+		rhp->rdev.stats.pd.max = rhp->rdev.stats.pd.cur;
+	mutex_unlock(&rhp->rdev.stats.lock);
+	PDBG("%s pdid 0x%0x ptr 0x%p\n", __func__, pdid, php);
+	return &php->ibpd;
+}
+
+static int c4iw_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			   u16 *pkey)
+{
+	PDBG("%s ibdev %p\n", __func__, ibdev);
+	*pkey = 0;
+	return 0;
+}
+
+static int c4iw_query_gid(struct ib_device *ibdev, u8 port, int index,
+			  union ib_gid *gid)
+{
+	struct c4iw_dev *dev;
+
+	PDBG("%s ibdev %p, port %d, index %d, gid %p\n",
+	       __func__, ibdev, port, index, gid);
+	dev = to_c4iw_dev(ibdev);
+	BUG_ON(port == 0);
+	memset(&(gid->raw[0]), 0, sizeof(gid->raw));
+	memcpy(&(gid->raw[0]), dev->rdev.lldi.ports[port-1]->dev_addr, 6);
+	return 0;
+}
+
+static int c4iw_query_device(struct ib_device *ibdev,
+			     struct ib_device_attr *props)
+{
+
+	struct c4iw_dev *dev;
+	PDBG("%s ibdev %p\n", __func__, ibdev);
+
+	dev = to_c4iw_dev(ibdev);
+	memset(props, 0, sizeof *props);
+	memcpy(&props->sys_image_guid, dev->rdev.lldi.ports[0]->dev_addr, 6);
+	props->hw_ver = CHELSIO_CHIP_RELEASE(dev->rdev.lldi.adapter_type);
+	props->fw_ver = dev->rdev.lldi.fw_vers;
+	props->device_cap_flags = dev->device_cap_flags;
+	props->page_size_cap = T4_PAGESIZE_MASK;
+	props->vendor_id = (u32)dev->rdev.lldi.pdev->vendor;
+	props->vendor_part_id = (u32)dev->rdev.lldi.pdev->device;
+	props->max_mr_size = T4_MAX_MR_SIZE;
+	props->max_qp = T4_MAX_NUM_QP;
+	props->max_qp_wr = T4_MAX_QP_DEPTH;
+	props->max_sge = T4_MAX_RECV_SGE;
+	props->max_sge_rd = 1;
+	props->max_qp_rd_atom = c4iw_max_read_depth;
+	props->max_qp_init_rd_atom = c4iw_max_read_depth;
+	props->max_cq = T4_MAX_NUM_CQ;
+	props->max_cqe = T4_MAX_CQ_DEPTH;
+	props->max_mr = c4iw_num_stags(&dev->rdev);
+	props->max_pd = T4_MAX_NUM_PD;
+	props->local_ca_ack_delay = 0;
+	props->max_fast_reg_page_list_len = t4_max_fr_depth(use_dsgl);
+
+	return 0;
+}
+
+static int c4iw_query_port(struct ib_device *ibdev, u8 port,
+			   struct ib_port_attr *props)
+{
+	struct c4iw_dev *dev;
+	struct net_device *netdev;
+	struct in_device *inetdev;
+
+	PDBG("%s ibdev %p\n", __func__, ibdev);
+
+	dev = to_c4iw_dev(ibdev);
+	netdev = dev->rdev.lldi.ports[port-1];
+
+	memset(props, 0, sizeof(struct ib_port_attr));
+	props->max_mtu = IB_MTU_4096;
+	if (netdev->mtu >= 4096)
+		props->active_mtu = IB_MTU_4096;
+	else if (netdev->mtu >= 2048)
+		props->active_mtu = IB_MTU_2048;
+	else if (netdev->mtu >= 1024)
+		props->active_mtu = IB_MTU_1024;
+	else if (netdev->mtu >= 512)
+		props->active_mtu = IB_MTU_512;
+	else
+		props->active_mtu = IB_MTU_256;
+
+	if (!netif_carrier_ok(netdev))
+		props->state = IB_PORT_DOWN;
+	else {
+		inetdev = in_dev_get(netdev);
+		if (inetdev) {
+			if (inetdev->ifa_list)
+				props->state = IB_PORT_ACTIVE;
+			else
+				props->state = IB_PORT_INIT;
+			in_dev_put(inetdev);
+		} else
+			props->state = IB_PORT_INIT;
+	}
+
+	props->port_cap_flags =
+	    IB_PORT_CM_SUP |
+	    IB_PORT_SNMP_TUNNEL_SUP |
+	    IB_PORT_REINIT_SUP |
+	    IB_PORT_DEVICE_MGMT_SUP |
+	    IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
+	props->gid_tbl_len = 1;
+	props->pkey_tbl_len = 1;
+	props->active_width = 2;
+	props->active_speed = IB_SPEED_DDR;
+	props->max_msg_sz = -1;
+
+	return 0;
+}
+
+static ssize_t show_rev(struct device *dev, struct device_attribute *attr,
+			char *buf)
+{
+	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
+						 ibdev.dev);
+	PDBG("%s dev 0x%p\n", __func__, dev);
+	return sprintf(buf, "%d\n",
+		       CHELSIO_CHIP_RELEASE(c4iw_dev->rdev.lldi.adapter_type));
+}
+
+static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr,
+			   char *buf)
+{
+	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
+						 ibdev.dev);
+	PDBG("%s dev 0x%p\n", __func__, dev);
+
+	return sprintf(buf, "%u.%u.%u.%u\n",
+			FW_HDR_FW_VER_MAJOR_GET(c4iw_dev->rdev.lldi.fw_vers),
+			FW_HDR_FW_VER_MINOR_GET(c4iw_dev->rdev.lldi.fw_vers),
+			FW_HDR_FW_VER_MICRO_GET(c4iw_dev->rdev.lldi.fw_vers),
+			FW_HDR_FW_VER_BUILD_GET(c4iw_dev->rdev.lldi.fw_vers));
+}
+
+static ssize_t show_hca(struct device *dev, struct device_attribute *attr,
+			char *buf)
+{
+	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
+						 ibdev.dev);
+	struct ethtool_drvinfo info;
+	struct net_device *lldev = c4iw_dev->rdev.lldi.ports[0];
+
+	PDBG("%s dev 0x%p\n", __func__, dev);
+	lldev->ethtool_ops->get_drvinfo(lldev, &info);
+	return sprintf(buf, "%s\n", info.driver);
+}
+
+static ssize_t show_board(struct device *dev, struct device_attribute *attr,
+			  char *buf)
+{
+	struct c4iw_dev *c4iw_dev = container_of(dev, struct c4iw_dev,
+						 ibdev.dev);
+	PDBG("%s dev 0x%p\n", __func__, dev);
+	return sprintf(buf, "%x.%x\n", c4iw_dev->rdev.lldi.pdev->vendor,
+		       c4iw_dev->rdev.lldi.pdev->device);
+}
+
+static int c4iw_get_mib(struct ib_device *ibdev,
+			union rdma_protocol_stats *stats)
+{
+	struct tp_tcp_stats v4, v6;
+	struct c4iw_dev *c4iw_dev = to_c4iw_dev(ibdev);
+
+	cxgb4_get_tcp_stats(c4iw_dev->rdev.lldi.pdev, &v4, &v6);
+	memset(stats, 0, sizeof *stats);
+	stats->iw.tcpInSegs = v4.tcpInSegs + v6.tcpInSegs;
+	stats->iw.tcpOutSegs = v4.tcpOutSegs + v6.tcpOutSegs;
+	stats->iw.tcpRetransSegs = v4.tcpRetransSegs + v6.tcpRetransSegs;
+	stats->iw.tcpOutRsts = v4.tcpOutRsts + v6.tcpOutSegs;
+
+	return 0;
+}
+
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_board, NULL);
+
+static struct device_attribute *c4iw_class_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_fw_ver,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+};
+
+int c4iw_register_device(struct c4iw_dev *dev)
+{
+	int ret;
+	int i;
+
+	PDBG("%s c4iw_dev %p\n", __func__, dev);
+	BUG_ON(!dev->rdev.lldi.ports[0]);
+	strlcpy(dev->ibdev.name, "cxgb4_%d", IB_DEVICE_NAME_MAX);
+	memset(&dev->ibdev.node_guid, 0, sizeof(dev->ibdev.node_guid));
+	memcpy(&dev->ibdev.node_guid, dev->rdev.lldi.ports[0]->dev_addr, 6);
+	dev->ibdev.owner = THIS_MODULE;
+	dev->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_WINDOW;
+	if (fastreg_support)
+		dev->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	dev->ibdev.local_dma_lkey = 0;
+	dev->ibdev.uverbs_cmd_mask =
+	    (1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
+	    (1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
+	    (1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
+	    (1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
+	    (1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
+	    (1ull << IB_USER_VERBS_CMD_REG_MR) |
+	    (1ull << IB_USER_VERBS_CMD_DEREG_MR) |
+	    (1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
+	    (1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
+	    (1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
+	    (1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ) |
+	    (1ull << IB_USER_VERBS_CMD_CREATE_QP) |
+	    (1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
+	    (1ull << IB_USER_VERBS_CMD_QUERY_QP) |
+	    (1ull << IB_USER_VERBS_CMD_POLL_CQ) |
+	    (1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
+	    (1ull << IB_USER_VERBS_CMD_POST_SEND) |
+	    (1ull << IB_USER_VERBS_CMD_POST_RECV);
+	dev->ibdev.node_type = RDMA_NODE_RNIC;
+	memcpy(dev->ibdev.node_desc, C4IW_NODE_DESC, sizeof(C4IW_NODE_DESC));
+	dev->ibdev.phys_port_cnt = dev->rdev.lldi.nports;
+	dev->ibdev.num_comp_vectors =  dev->rdev.lldi.nciq;
+	dev->ibdev.dma_device = &(dev->rdev.lldi.pdev->dev);
+	dev->ibdev.query_device = c4iw_query_device;
+	dev->ibdev.query_port = c4iw_query_port;
+	dev->ibdev.query_pkey = c4iw_query_pkey;
+	dev->ibdev.query_gid = c4iw_query_gid;
+	dev->ibdev.alloc_ucontext = c4iw_alloc_ucontext;
+	dev->ibdev.dealloc_ucontext = c4iw_dealloc_ucontext;
+	dev->ibdev.mmap = c4iw_mmap;
+	dev->ibdev.alloc_pd = c4iw_allocate_pd;
+	dev->ibdev.dealloc_pd = c4iw_deallocate_pd;
+	dev->ibdev.create_ah = c4iw_ah_create;
+	dev->ibdev.destroy_ah = c4iw_ah_destroy;
+	dev->ibdev.create_qp = c4iw_create_qp;
+	dev->ibdev.modify_qp = c4iw_ib_modify_qp;
+	dev->ibdev.query_qp = c4iw_ib_query_qp;
+	dev->ibdev.destroy_qp = c4iw_destroy_qp;
+	dev->ibdev.create_cq = c4iw_create_cq;
+	dev->ibdev.destroy_cq = c4iw_destroy_cq;
+	dev->ibdev.resize_cq = c4iw_resize_cq;
+	dev->ibdev.poll_cq = c4iw_poll_cq;
+	dev->ibdev.get_dma_mr = c4iw_get_dma_mr;
+	dev->ibdev.reg_phys_mr = c4iw_register_phys_mem;
+	dev->ibdev.rereg_phys_mr = c4iw_reregister_phys_mem;
+	dev->ibdev.reg_user_mr = c4iw_reg_user_mr;
+	dev->ibdev.dereg_mr = c4iw_dereg_mr;
+	dev->ibdev.alloc_mw = c4iw_alloc_mw;
+	dev->ibdev.bind_mw = c4iw_bind_mw;
+	dev->ibdev.dealloc_mw = c4iw_dealloc_mw;
+	dev->ibdev.alloc_fast_reg_mr = c4iw_alloc_fast_reg_mr;
+	dev->ibdev.alloc_fast_reg_page_list = c4iw_alloc_fastreg_pbl;
+	dev->ibdev.free_fast_reg_page_list = c4iw_free_fastreg_pbl;
+	dev->ibdev.attach_mcast = c4iw_multicast_attach;
+	dev->ibdev.detach_mcast = c4iw_multicast_detach;
+	dev->ibdev.process_mad = c4iw_process_mad;
+	dev->ibdev.req_notify_cq = c4iw_arm_cq;
+	dev->ibdev.post_send = c4iw_post_send;
+	dev->ibdev.post_recv = c4iw_post_receive;
+	dev->ibdev.get_protocol_stats = c4iw_get_mib;
+	dev->ibdev.uverbs_abi_ver = C4IW_UVERBS_ABI_VERSION;
+
+	dev->ibdev.iwcm = kmalloc(sizeof(struct iw_cm_verbs), GFP_KERNEL);
+	if (!dev->ibdev.iwcm)
+		return -ENOMEM;
+
+	dev->ibdev.iwcm->connect = c4iw_connect;
+	dev->ibdev.iwcm->accept = c4iw_accept_cr;
+	dev->ibdev.iwcm->reject = c4iw_reject_cr;
+	dev->ibdev.iwcm->create_listen = c4iw_create_listen;
+	dev->ibdev.iwcm->destroy_listen = c4iw_destroy_listen;
+	dev->ibdev.iwcm->add_ref = c4iw_qp_add_ref;
+	dev->ibdev.iwcm->rem_ref = c4iw_qp_rem_ref;
+	dev->ibdev.iwcm->get_qp = c4iw_get_qp;
+
+	ret = ib_register_device(&dev->ibdev, NULL);
+	if (ret)
+		goto bail1;
+
+	for (i = 0; i < ARRAY_SIZE(c4iw_class_attributes); ++i) {
+		ret = device_create_file(&dev->ibdev.dev,
+					 c4iw_class_attributes[i]);
+		if (ret)
+			goto bail2;
+	}
+	return 0;
+bail2:
+	ib_unregister_device(&dev->ibdev);
+bail1:
+	kfree(dev->ibdev.iwcm);
+	return ret;
+}
+
+void c4iw_unregister_device(struct c4iw_dev *dev)
+{
+	int i;
+
+	PDBG("%s c4iw_dev %p\n", __func__, dev);
+	for (i = 0; i < ARRAY_SIZE(c4iw_class_attributes); ++i)
+		device_remove_file(&dev->ibdev.dev,
+				   c4iw_class_attributes[i]);
+	ib_unregister_device(&dev->ibdev);
+	kfree(dev->ibdev.iwcm);
+	return;
+}
diff --git a/drivers/infiniband/hw/cxgb4/qp.c b/drivers/infiniband/hw/cxgb4/qp.c
new file mode 100644
index 00000000000..086f62f5dc9
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/qp.c
@@ -0,0 +1,1808 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+
+#include "iw_cxgb4.h"
+
+static int db_delay_usecs = 1;
+module_param(db_delay_usecs, int, 0644);
+MODULE_PARM_DESC(db_delay_usecs, "Usecs to delay awaiting db fifo to drain");
+
+static int ocqp_support = 1;
+module_param(ocqp_support, int, 0644);
+MODULE_PARM_DESC(ocqp_support, "Support on-chip SQs (default=1)");
+
+int db_fc_threshold = 1000;
+module_param(db_fc_threshold, int, 0644);
+MODULE_PARM_DESC(db_fc_threshold,
+		 "QP count/threshold that triggers"
+		 " automatic db flow control mode (default = 1000)");
+
+int db_coalescing_threshold;
+module_param(db_coalescing_threshold, int, 0644);
+MODULE_PARM_DESC(db_coalescing_threshold,
+		 "QP count/threshold that triggers"
+		 " disabling db coalescing (default = 0)");
+
+static int max_fr_immd = T4_MAX_FR_IMMD;
+module_param(max_fr_immd, int, 0644);
+MODULE_PARM_DESC(max_fr_immd, "fastreg threshold for using DSGL instead of immedate");
+
+static void set_state(struct c4iw_qp *qhp, enum c4iw_qp_state state)
+{
+	unsigned long flag;
+	spin_lock_irqsave(&qhp->lock, flag);
+	qhp->attr.state = state;
+	spin_unlock_irqrestore(&qhp->lock, flag);
+}
+
+static void dealloc_oc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
+{
+	c4iw_ocqp_pool_free(rdev, sq->dma_addr, sq->memsize);
+}
+
+static void dealloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
+{
+	dma_free_coherent(&(rdev->lldi.pdev->dev), sq->memsize, sq->queue,
+			  pci_unmap_addr(sq, mapping));
+}
+
+static void dealloc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
+{
+	if (t4_sq_onchip(sq))
+		dealloc_oc_sq(rdev, sq);
+	else
+		dealloc_host_sq(rdev, sq);
+}
+
+static int alloc_oc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
+{
+	if (!ocqp_support || !ocqp_supported(&rdev->lldi))
+		return -ENOSYS;
+	sq->dma_addr = c4iw_ocqp_pool_alloc(rdev, sq->memsize);
+	if (!sq->dma_addr)
+		return -ENOMEM;
+	sq->phys_addr = rdev->oc_mw_pa + sq->dma_addr -
+			rdev->lldi.vr->ocq.start;
+	sq->queue = (__force union t4_wr *)(rdev->oc_mw_kva + sq->dma_addr -
+					    rdev->lldi.vr->ocq.start);
+	sq->flags |= T4_SQ_ONCHIP;
+	return 0;
+}
+
+static int alloc_host_sq(struct c4iw_rdev *rdev, struct t4_sq *sq)
+{
+	sq->queue = dma_alloc_coherent(&(rdev->lldi.pdev->dev), sq->memsize,
+				       &(sq->dma_addr), GFP_KERNEL);
+	if (!sq->queue)
+		return -ENOMEM;
+	sq->phys_addr = virt_to_phys(sq->queue);
+	pci_unmap_addr_set(sq, mapping, sq->dma_addr);
+	return 0;
+}
+
+static int alloc_sq(struct c4iw_rdev *rdev, struct t4_sq *sq, int user)
+{
+	int ret = -ENOSYS;
+	if (user)
+		ret = alloc_oc_sq(rdev, sq);
+	if (ret)
+		ret = alloc_host_sq(rdev, sq);
+	return ret;
+}
+
+static int destroy_qp(struct c4iw_rdev *rdev, struct t4_wq *wq,
+		      struct c4iw_dev_ucontext *uctx)
+{
+	/*
+	 * uP clears EQ contexts when the connection exits rdma mode,
+	 * so no need to post a RESET WR for these EQs.
+	 */
+	dma_free_coherent(&(rdev->lldi.pdev->dev),
+			  wq->rq.memsize, wq->rq.queue,
+			  dma_unmap_addr(&wq->rq, mapping));
+	dealloc_sq(rdev, &wq->sq);
+	c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size);
+	kfree(wq->rq.sw_rq);
+	kfree(wq->sq.sw_sq);
+	c4iw_put_qpid(rdev, wq->rq.qid, uctx);
+	c4iw_put_qpid(rdev, wq->sq.qid, uctx);
+	return 0;
+}
+
+static int create_qp(struct c4iw_rdev *rdev, struct t4_wq *wq,
+		     struct t4_cq *rcq, struct t4_cq *scq,
+		     struct c4iw_dev_ucontext *uctx)
+{
+	int user = (uctx != &rdev->uctx);
+	struct fw_ri_res_wr *res_wr;
+	struct fw_ri_res *res;
+	int wr_len;
+	struct c4iw_wr_wait wr_wait;
+	struct sk_buff *skb;
+	int ret = 0;
+	int eqsize;
+
+	wq->sq.qid = c4iw_get_qpid(rdev, uctx);
+	if (!wq->sq.qid)
+		return -ENOMEM;
+
+	wq->rq.qid = c4iw_get_qpid(rdev, uctx);
+	if (!wq->rq.qid) {
+		ret = -ENOMEM;
+		goto free_sq_qid;
+	}
+
+	if (!user) {
+		wq->sq.sw_sq = kzalloc(wq->sq.size * sizeof *wq->sq.sw_sq,
+				 GFP_KERNEL);
+		if (!wq->sq.sw_sq) {
+			ret = -ENOMEM;
+			goto free_rq_qid;
+		}
+
+		wq->rq.sw_rq = kzalloc(wq->rq.size * sizeof *wq->rq.sw_rq,
+				 GFP_KERNEL);
+		if (!wq->rq.sw_rq) {
+			ret = -ENOMEM;
+			goto free_sw_sq;
+		}
+	}
+
+	/*
+	 * RQT must be a power of 2.
+	 */
+	wq->rq.rqt_size = roundup_pow_of_two(wq->rq.size);
+	wq->rq.rqt_hwaddr = c4iw_rqtpool_alloc(rdev, wq->rq.rqt_size);
+	if (!wq->rq.rqt_hwaddr) {
+		ret = -ENOMEM;
+		goto free_sw_rq;
+	}
+
+	ret = alloc_sq(rdev, &wq->sq, user);
+	if (ret)
+		goto free_hwaddr;
+	memset(wq->sq.queue, 0, wq->sq.memsize);
+	dma_unmap_addr_set(&wq->sq, mapping, wq->sq.dma_addr);
+
+	wq->rq.queue = dma_alloc_coherent(&(rdev->lldi.pdev->dev),
+					  wq->rq.memsize, &(wq->rq.dma_addr),
+					  GFP_KERNEL);
+	if (!wq->rq.queue) {
+		ret = -ENOMEM;
+		goto free_sq;
+	}
+	PDBG("%s sq base va 0x%p pa 0x%llx rq base va 0x%p pa 0x%llx\n",
+		__func__, wq->sq.queue,
+		(unsigned long long)virt_to_phys(wq->sq.queue),
+		wq->rq.queue,
+		(unsigned long long)virt_to_phys(wq->rq.queue));
+	memset(wq->rq.queue, 0, wq->rq.memsize);
+	dma_unmap_addr_set(&wq->rq, mapping, wq->rq.dma_addr);
+
+	wq->db = rdev->lldi.db_reg;
+	wq->gts = rdev->lldi.gts_reg;
+	if (user || is_t5(rdev->lldi.adapter_type)) {
+		u32 off;
+
+		off = (wq->sq.qid << rdev->qpshift) & PAGE_MASK;
+		if (user) {
+			wq->sq.udb = (u64 __iomem *)(rdev->bar2_pa + off);
+		} else {
+			off += 128 * (wq->sq.qid & rdev->qpmask) + 8;
+			wq->sq.udb = (u64 __iomem *)(rdev->bar2_kva + off);
+		}
+		off = (wq->rq.qid << rdev->qpshift) & PAGE_MASK;
+		if (user) {
+			wq->rq.udb = (u64 __iomem *)(rdev->bar2_pa + off);
+		} else {
+			off += 128 * (wq->rq.qid & rdev->qpmask) + 8;
+			wq->rq.udb = (u64 __iomem *)(rdev->bar2_kva + off);
+		}
+	}
+	wq->rdev = rdev;
+	wq->rq.msn = 1;
+
+	/* build fw_ri_res_wr */
+	wr_len = sizeof *res_wr + 2 * sizeof *res;
+
+	skb = alloc_skb(wr_len, GFP_KERNEL);
+	if (!skb) {
+		ret = -ENOMEM;
+		goto free_dma;
+	}
+	set_wr_txq(skb, CPL_PRIORITY_CONTROL, 0);
+
+	res_wr = (struct fw_ri_res_wr *)__skb_put(skb, wr_len);
+	memset(res_wr, 0, wr_len);
+	res_wr->op_nres = cpu_to_be32(
+			FW_WR_OP(FW_RI_RES_WR) |
+			V_FW_RI_RES_WR_NRES(2) |
+			FW_WR_COMPL(1));
+	res_wr->len16_pkd = cpu_to_be32(DIV_ROUND_UP(wr_len, 16));
+	res_wr->cookie = (unsigned long) &wr_wait;
+	res = res_wr->res;
+	res->u.sqrq.restype = FW_RI_RES_TYPE_SQ;
+	res->u.sqrq.op = FW_RI_RES_OP_WRITE;
+
+	/*
+	 * eqsize is the number of 64B entries plus the status page size.
+	 */
+	eqsize = wq->sq.size * T4_SQ_NUM_SLOTS + T4_EQ_STATUS_ENTRIES;
+
+	res->u.sqrq.fetchszm_to_iqid = cpu_to_be32(
+		V_FW_RI_RES_WR_HOSTFCMODE(0) |	/* no host cidx updates */
+		V_FW_RI_RES_WR_CPRIO(0) |	/* don't keep in chip cache */
+		V_FW_RI_RES_WR_PCIECHN(0) |	/* set by uP at ri_init time */
+		(t4_sq_onchip(&wq->sq) ? F_FW_RI_RES_WR_ONCHIP : 0) |
+		V_FW_RI_RES_WR_IQID(scq->cqid));
+	res->u.sqrq.dcaen_to_eqsize = cpu_to_be32(
+		V_FW_RI_RES_WR_DCAEN(0) |
+		V_FW_RI_RES_WR_DCACPU(0) |
+		V_FW_RI_RES_WR_FBMIN(2) |
+		V_FW_RI_RES_WR_FBMAX(2) |
+		V_FW_RI_RES_WR_CIDXFTHRESHO(0) |
+		V_FW_RI_RES_WR_CIDXFTHRESH(0) |
+		V_FW_RI_RES_WR_EQSIZE(eqsize));
+	res->u.sqrq.eqid = cpu_to_be32(wq->sq.qid);
+	res->u.sqrq.eqaddr = cpu_to_be64(wq->sq.dma_addr);
+	res++;
+	res->u.sqrq.restype = FW_RI_RES_TYPE_RQ;
+	res->u.sqrq.op = FW_RI_RES_OP_WRITE;
+
+	/*
+	 * eqsize is the number of 64B entries plus the status page size.
+	 */
+	eqsize = wq->rq.size * T4_RQ_NUM_SLOTS + T4_EQ_STATUS_ENTRIES;
+	res->u.sqrq.fetchszm_to_iqid = cpu_to_be32(
+		V_FW_RI_RES_WR_HOSTFCMODE(0) |	/* no host cidx updates */
+		V_FW_RI_RES_WR_CPRIO(0) |	/* don't keep in chip cache */
+		V_FW_RI_RES_WR_PCIECHN(0) |	/* set by uP at ri_init time */
+		V_FW_RI_RES_WR_IQID(rcq->cqid));
+	res->u.sqrq.dcaen_to_eqsize = cpu_to_be32(
+		V_FW_RI_RES_WR_DCAEN(0) |
+		V_FW_RI_RES_WR_DCACPU(0) |
+		V_FW_RI_RES_WR_FBMIN(2) |
+		V_FW_RI_RES_WR_FBMAX(2) |
+		V_FW_RI_RES_WR_CIDXFTHRESHO(0) |
+		V_FW_RI_RES_WR_CIDXFTHRESH(0) |
+		V_FW_RI_RES_WR_EQSIZE(eqsize));
+	res->u.sqrq.eqid = cpu_to_be32(wq->rq.qid);
+	res->u.sqrq.eqaddr = cpu_to_be64(wq->rq.dma_addr);
+
+	c4iw_init_wr_wait(&wr_wait);
+
+	ret = c4iw_ofld_send(rdev, skb);
+	if (ret)
+		goto free_dma;
+	ret = c4iw_wait_for_reply(rdev, &wr_wait, 0, wq->sq.qid, __func__);
+	if (ret)
+		goto free_dma;
+
+	PDBG("%s sqid 0x%x rqid 0x%x kdb 0x%p squdb 0x%lx rqudb 0x%lx\n",
+	     __func__, wq->sq.qid, wq->rq.qid, wq->db,
+	     (__force unsigned long) wq->sq.udb,
+	     (__force unsigned long) wq->rq.udb);
+
+	return 0;
+free_dma:
+	dma_free_coherent(&(rdev->lldi.pdev->dev),
+			  wq->rq.memsize, wq->rq.queue,
+			  dma_unmap_addr(&wq->rq, mapping));
+free_sq:
+	dealloc_sq(rdev, &wq->sq);
+free_hwaddr:
+	c4iw_rqtpool_free(rdev, wq->rq.rqt_hwaddr, wq->rq.rqt_size);
+free_sw_rq:
+	kfree(wq->rq.sw_rq);
+free_sw_sq:
+	kfree(wq->sq.sw_sq);
+free_rq_qid:
+	c4iw_put_qpid(rdev, wq->rq.qid, uctx);
+free_sq_qid:
+	c4iw_put_qpid(rdev, wq->sq.qid, uctx);
+	return ret;
+}
+
+static int build_immd(struct t4_sq *sq, struct fw_ri_immd *immdp,
+		      struct ib_send_wr *wr, int max, u32 *plenp)
+{
+	u8 *dstp, *srcp;
+	u32 plen = 0;
+	int i;
+	int rem, len;
+
+	dstp = (u8 *)immdp->data;
+	for (i = 0; i < wr->num_sge; i++) {
+		if ((plen + wr->sg_list[i].length) > max)
+			return -EMSGSIZE;
+		srcp = (u8 *)(unsigned long)wr->sg_list[i].addr;
+		plen += wr->sg_list[i].length;
+		rem = wr->sg_list[i].length;
+		while (rem) {
+			if (dstp == (u8 *)&sq->queue[sq->size])
+				dstp = (u8 *)sq->queue;
+			if (rem <= (u8 *)&sq->queue[sq->size] - dstp)
+				len = rem;
+			else
+				len = (u8 *)&sq->queue[sq->size] - dstp;
+			memcpy(dstp, srcp, len);
+			dstp += len;
+			srcp += len;
+			rem -= len;
+		}
+	}
+	len = roundup(plen + sizeof *immdp, 16) - (plen + sizeof *immdp);
+	if (len)
+		memset(dstp, 0, len);
+	immdp->op = FW_RI_DATA_IMMD;
+	immdp->r1 = 0;
+	immdp->r2 = 0;
+	immdp->immdlen = cpu_to_be32(plen);
+	*plenp = plen;
+	return 0;
+}
+
+static int build_isgl(__be64 *queue_start, __be64 *queue_end,
+		      struct fw_ri_isgl *isglp, struct ib_sge *sg_list,
+		      int num_sge, u32 *plenp)
+
+{
+	int i;
+	u32 plen = 0;
+	__be64 *flitp = (__be64 *)isglp->sge;
+
+	for (i = 0; i < num_sge; i++) {
+		if ((plen + sg_list[i].length) < plen)
+			return -EMSGSIZE;
+		plen += sg_list[i].length;
+		*flitp = cpu_to_be64(((u64)sg_list[i].lkey << 32) |
+				     sg_list[i].length);
+		if (++flitp == queue_end)
+			flitp = queue_start;
+		*flitp = cpu_to_be64(sg_list[i].addr);
+		if (++flitp == queue_end)
+			flitp = queue_start;
+	}
+	*flitp = (__force __be64)0;
+	isglp->op = FW_RI_DATA_ISGL;
+	isglp->r1 = 0;
+	isglp->nsge = cpu_to_be16(num_sge);
+	isglp->r2 = 0;
+	if (plenp)
+		*plenp = plen;
+	return 0;
+}
+
+static int build_rdma_send(struct t4_sq *sq, union t4_wr *wqe,
+			   struct ib_send_wr *wr, u8 *len16)
+{
+	u32 plen;
+	int size;
+	int ret;
+
+	if (wr->num_sge > T4_MAX_SEND_SGE)
+		return -EINVAL;
+	switch (wr->opcode) {
+	case IB_WR_SEND:
+		if (wr->send_flags & IB_SEND_SOLICITED)
+			wqe->send.sendop_pkd = cpu_to_be32(
+				V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE));
+		else
+			wqe->send.sendop_pkd = cpu_to_be32(
+				V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND));
+		wqe->send.stag_inv = 0;
+		break;
+	case IB_WR_SEND_WITH_INV:
+		if (wr->send_flags & IB_SEND_SOLICITED)
+			wqe->send.sendop_pkd = cpu_to_be32(
+				V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_SE_INV));
+		else
+			wqe->send.sendop_pkd = cpu_to_be32(
+				V_FW_RI_SEND_WR_SENDOP(FW_RI_SEND_WITH_INV));
+		wqe->send.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+	wqe->send.r3 = 0;
+	wqe->send.r4 = 0;
+
+	plen = 0;
+	if (wr->num_sge) {
+		if (wr->send_flags & IB_SEND_INLINE) {
+			ret = build_immd(sq, wqe->send.u.immd_src, wr,
+					 T4_MAX_SEND_INLINE, &plen);
+			if (ret)
+				return ret;
+			size = sizeof wqe->send + sizeof(struct fw_ri_immd) +
+			       plen;
+		} else {
+			ret = build_isgl((__be64 *)sq->queue,
+					 (__be64 *)&sq->queue[sq->size],
+					 wqe->send.u.isgl_src,
+					 wr->sg_list, wr->num_sge, &plen);
+			if (ret)
+				return ret;
+			size = sizeof wqe->send + sizeof(struct fw_ri_isgl) +
+			       wr->num_sge * sizeof(struct fw_ri_sge);
+		}
+	} else {
+		wqe->send.u.immd_src[0].op = FW_RI_DATA_IMMD;
+		wqe->send.u.immd_src[0].r1 = 0;
+		wqe->send.u.immd_src[0].r2 = 0;
+		wqe->send.u.immd_src[0].immdlen = 0;
+		size = sizeof wqe->send + sizeof(struct fw_ri_immd);
+		plen = 0;
+	}
+	*len16 = DIV_ROUND_UP(size, 16);
+	wqe->send.plen = cpu_to_be32(plen);
+	return 0;
+}
+
+static int build_rdma_write(struct t4_sq *sq, union t4_wr *wqe,
+			    struct ib_send_wr *wr, u8 *len16)
+{
+	u32 plen;
+	int size;
+	int ret;
+
+	if (wr->num_sge > T4_MAX_SEND_SGE)
+		return -EINVAL;
+	wqe->write.r2 = 0;
+	wqe->write.stag_sink = cpu_to_be32(wr->wr.rdma.rkey);
+	wqe->write.to_sink = cpu_to_be64(wr->wr.rdma.remote_addr);
+	if (wr->num_sge) {
+		if (wr->send_flags & IB_SEND_INLINE) {
+			ret = build_immd(sq, wqe->write.u.immd_src, wr,
+					 T4_MAX_WRITE_INLINE, &plen);
+			if (ret)
+				return ret;
+			size = sizeof wqe->write + sizeof(struct fw_ri_immd) +
+			       plen;
+		} else {
+			ret = build_isgl((__be64 *)sq->queue,
+					 (__be64 *)&sq->queue[sq->size],
+					 wqe->write.u.isgl_src,
+					 wr->sg_list, wr->num_sge, &plen);
+			if (ret)
+				return ret;
+			size = sizeof wqe->write + sizeof(struct fw_ri_isgl) +
+			       wr->num_sge * sizeof(struct fw_ri_sge);
+		}
+	} else {
+		wqe->write.u.immd_src[0].op = FW_RI_DATA_IMMD;
+		wqe->write.u.immd_src[0].r1 = 0;
+		wqe->write.u.immd_src[0].r2 = 0;
+		wqe->write.u.immd_src[0].immdlen = 0;
+		size = sizeof wqe->write + sizeof(struct fw_ri_immd);
+		plen = 0;
+	}
+	*len16 = DIV_ROUND_UP(size, 16);
+	wqe->write.plen = cpu_to_be32(plen);
+	return 0;
+}
+
+static int build_rdma_read(union t4_wr *wqe, struct ib_send_wr *wr, u8 *len16)
+{
+	if (wr->num_sge > 1)
+		return -EINVAL;
+	if (wr->num_sge) {
+		wqe->read.stag_src = cpu_to_be32(wr->wr.rdma.rkey);
+		wqe->read.to_src_hi = cpu_to_be32((u32)(wr->wr.rdma.remote_addr
+							>> 32));
+		wqe->read.to_src_lo = cpu_to_be32((u32)wr->wr.rdma.remote_addr);
+		wqe->read.stag_sink = cpu_to_be32(wr->sg_list[0].lkey);
+		wqe->read.plen = cpu_to_be32(wr->sg_list[0].length);
+		wqe->read.to_sink_hi = cpu_to_be32((u32)(wr->sg_list[0].addr
+							 >> 32));
+		wqe->read.to_sink_lo = cpu_to_be32((u32)(wr->sg_list[0].addr));
+	} else {
+		wqe->read.stag_src = cpu_to_be32(2);
+		wqe->read.to_src_hi = 0;
+		wqe->read.to_src_lo = 0;
+		wqe->read.stag_sink = cpu_to_be32(2);
+		wqe->read.plen = 0;
+		wqe->read.to_sink_hi = 0;
+		wqe->read.to_sink_lo = 0;
+	}
+	wqe->read.r2 = 0;
+	wqe->read.r5 = 0;
+	*len16 = DIV_ROUND_UP(sizeof wqe->read, 16);
+	return 0;
+}
+
+static int build_rdma_recv(struct c4iw_qp *qhp, union t4_recv_wr *wqe,
+			   struct ib_recv_wr *wr, u8 *len16)
+{
+	int ret;
+
+	ret = build_isgl((__be64 *)qhp->wq.rq.queue,
+			 (__be64 *)&qhp->wq.rq.queue[qhp->wq.rq.size],
+			 &wqe->recv.isgl, wr->sg_list, wr->num_sge, NULL);
+	if (ret)
+		return ret;
+	*len16 = DIV_ROUND_UP(sizeof wqe->recv +
+			      wr->num_sge * sizeof(struct fw_ri_sge), 16);
+	return 0;
+}
+
+static int build_fastreg(struct t4_sq *sq, union t4_wr *wqe,
+			 struct ib_send_wr *wr, u8 *len16, u8 t5dev)
+{
+
+	struct fw_ri_immd *imdp;
+	__be64 *p;
+	int i;
+	int pbllen = roundup(wr->wr.fast_reg.page_list_len * sizeof(u64), 32);
+	int rem;
+
+	if (wr->wr.fast_reg.page_list_len >
+	    t4_max_fr_depth(use_dsgl))
+		return -EINVAL;
+
+	wqe->fr.qpbinde_to_dcacpu = 0;
+	wqe->fr.pgsz_shift = wr->wr.fast_reg.page_shift - 12;
+	wqe->fr.addr_type = FW_RI_VA_BASED_TO;
+	wqe->fr.mem_perms = c4iw_ib_to_tpt_access(wr->wr.fast_reg.access_flags);
+	wqe->fr.len_hi = 0;
+	wqe->fr.len_lo = cpu_to_be32(wr->wr.fast_reg.length);
+	wqe->fr.stag = cpu_to_be32(wr->wr.fast_reg.rkey);
+	wqe->fr.va_hi = cpu_to_be32(wr->wr.fast_reg.iova_start >> 32);
+	wqe->fr.va_lo_fbo = cpu_to_be32(wr->wr.fast_reg.iova_start &
+					0xffffffff);
+
+	if (t5dev && use_dsgl && (pbllen > max_fr_immd)) {
+		struct c4iw_fr_page_list *c4pl =
+			to_c4iw_fr_page_list(wr->wr.fast_reg.page_list);
+		struct fw_ri_dsgl *sglp;
+
+		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+			wr->wr.fast_reg.page_list->page_list[i] = (__force u64)
+				cpu_to_be64((u64)
+				wr->wr.fast_reg.page_list->page_list[i]);
+		}
+
+		sglp = (struct fw_ri_dsgl *)(&wqe->fr + 1);
+		sglp->op = FW_RI_DATA_DSGL;
+		sglp->r1 = 0;
+		sglp->nsge = cpu_to_be16(1);
+		sglp->addr0 = cpu_to_be64(c4pl->dma_addr);
+		sglp->len0 = cpu_to_be32(pbllen);
+
+		*len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*sglp), 16);
+	} else {
+		imdp = (struct fw_ri_immd *)(&wqe->fr + 1);
+		imdp->op = FW_RI_DATA_IMMD;
+		imdp->r1 = 0;
+		imdp->r2 = 0;
+		imdp->immdlen = cpu_to_be32(pbllen);
+		p = (__be64 *)(imdp + 1);
+		rem = pbllen;
+		for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+			*p = cpu_to_be64(
+				(u64)wr->wr.fast_reg.page_list->page_list[i]);
+			rem -= sizeof(*p);
+			if (++p == (__be64 *)&sq->queue[sq->size])
+				p = (__be64 *)sq->queue;
+		}
+		BUG_ON(rem < 0);
+		while (rem) {
+			*p = 0;
+			rem -= sizeof(*p);
+			if (++p == (__be64 *)&sq->queue[sq->size])
+				p = (__be64 *)sq->queue;
+		}
+		*len16 = DIV_ROUND_UP(sizeof(wqe->fr) + sizeof(*imdp)
+				      + pbllen, 16);
+	}
+	return 0;
+}
+
+static int build_inv_stag(union t4_wr *wqe, struct ib_send_wr *wr,
+			  u8 *len16)
+{
+	wqe->inv.stag_inv = cpu_to_be32(wr->ex.invalidate_rkey);
+	wqe->inv.r2 = 0;
+	*len16 = DIV_ROUND_UP(sizeof wqe->inv, 16);
+	return 0;
+}
+
+void c4iw_qp_add_ref(struct ib_qp *qp)
+{
+	PDBG("%s ib_qp %p\n", __func__, qp);
+	atomic_inc(&(to_c4iw_qp(qp)->refcnt));
+}
+
+void c4iw_qp_rem_ref(struct ib_qp *qp)
+{
+	PDBG("%s ib_qp %p\n", __func__, qp);
+	if (atomic_dec_and_test(&(to_c4iw_qp(qp)->refcnt)))
+		wake_up(&(to_c4iw_qp(qp)->wait));
+}
+
+static void add_to_fc_list(struct list_head *head, struct list_head *entry)
+{
+	if (list_empty(entry))
+		list_add_tail(entry, head);
+}
+
+static int ring_kernel_sq_db(struct c4iw_qp *qhp, u16 inc)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qhp->rhp->lock, flags);
+	spin_lock(&qhp->lock);
+	if (qhp->rhp->db_state == NORMAL)
+		t4_ring_sq_db(&qhp->wq, inc,
+			      is_t5(qhp->rhp->rdev.lldi.adapter_type), NULL);
+	else {
+		add_to_fc_list(&qhp->rhp->db_fc_list, &qhp->db_fc_entry);
+		qhp->wq.sq.wq_pidx_inc += inc;
+	}
+	spin_unlock(&qhp->lock);
+	spin_unlock_irqrestore(&qhp->rhp->lock, flags);
+	return 0;
+}
+
+static int ring_kernel_rq_db(struct c4iw_qp *qhp, u16 inc)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qhp->rhp->lock, flags);
+	spin_lock(&qhp->lock);
+	if (qhp->rhp->db_state == NORMAL)
+		t4_ring_rq_db(&qhp->wq, inc,
+			      is_t5(qhp->rhp->rdev.lldi.adapter_type), NULL);
+	else {
+		add_to_fc_list(&qhp->rhp->db_fc_list, &qhp->db_fc_entry);
+		qhp->wq.rq.wq_pidx_inc += inc;
+	}
+	spin_unlock(&qhp->lock);
+	spin_unlock_irqrestore(&qhp->rhp->lock, flags);
+	return 0;
+}
+
+int c4iw_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		   struct ib_send_wr **bad_wr)
+{
+	int err = 0;
+	u8 len16 = 0;
+	enum fw_wr_opcodes fw_opcode = 0;
+	enum fw_ri_wr_flags fw_flags;
+	struct c4iw_qp *qhp;
+	union t4_wr *wqe = NULL;
+	u32 num_wrs;
+	struct t4_swsqe *swsqe;
+	unsigned long flag;
+	u16 idx = 0;
+
+	qhp = to_c4iw_qp(ibqp);
+	spin_lock_irqsave(&qhp->lock, flag);
+	if (t4_wq_in_error(&qhp->wq)) {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		return -EINVAL;
+	}
+	num_wrs = t4_sq_avail(&qhp->wq);
+	if (num_wrs == 0) {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		return -ENOMEM;
+	}
+	while (wr) {
+		if (num_wrs == 0) {
+			err = -ENOMEM;
+			*bad_wr = wr;
+			break;
+		}
+		wqe = (union t4_wr *)((u8 *)qhp->wq.sq.queue +
+		      qhp->wq.sq.wq_pidx * T4_EQ_ENTRY_SIZE);
+
+		fw_flags = 0;
+		if (wr->send_flags & IB_SEND_SOLICITED)
+			fw_flags |= FW_RI_SOLICITED_EVENT_FLAG;
+		if (wr->send_flags & IB_SEND_SIGNALED || qhp->sq_sig_all)
+			fw_flags |= FW_RI_COMPLETION_FLAG;
+		swsqe = &qhp->wq.sq.sw_sq[qhp->wq.sq.pidx];
+		switch (wr->opcode) {
+		case IB_WR_SEND_WITH_INV:
+		case IB_WR_SEND:
+			if (wr->send_flags & IB_SEND_FENCE)
+				fw_flags |= FW_RI_READ_FENCE_FLAG;
+			fw_opcode = FW_RI_SEND_WR;
+			if (wr->opcode == IB_WR_SEND)
+				swsqe->opcode = FW_RI_SEND;
+			else
+				swsqe->opcode = FW_RI_SEND_WITH_INV;
+			err = build_rdma_send(&qhp->wq.sq, wqe, wr, &len16);
+			break;
+		case IB_WR_RDMA_WRITE:
+			fw_opcode = FW_RI_RDMA_WRITE_WR;
+			swsqe->opcode = FW_RI_RDMA_WRITE;
+			err = build_rdma_write(&qhp->wq.sq, wqe, wr, &len16);
+			break;
+		case IB_WR_RDMA_READ:
+		case IB_WR_RDMA_READ_WITH_INV:
+			fw_opcode = FW_RI_RDMA_READ_WR;
+			swsqe->opcode = FW_RI_READ_REQ;
+			if (wr->opcode == IB_WR_RDMA_READ_WITH_INV)
+				fw_flags = FW_RI_RDMA_READ_INVALIDATE;
+			else
+				fw_flags = 0;
+			err = build_rdma_read(wqe, wr, &len16);
+			if (err)
+				break;
+			swsqe->read_len = wr->sg_list[0].length;
+			if (!qhp->wq.sq.oldest_read)
+				qhp->wq.sq.oldest_read = swsqe;
+			break;
+		case IB_WR_FAST_REG_MR:
+			fw_opcode = FW_RI_FR_NSMR_WR;
+			swsqe->opcode = FW_RI_FAST_REGISTER;
+			err = build_fastreg(&qhp->wq.sq, wqe, wr, &len16,
+					    is_t5(
+					    qhp->rhp->rdev.lldi.adapter_type) ?
+					    1 : 0);
+			break;
+		case IB_WR_LOCAL_INV:
+			if (wr->send_flags & IB_SEND_FENCE)
+				fw_flags |= FW_RI_LOCAL_FENCE_FLAG;
+			fw_opcode = FW_RI_INV_LSTAG_WR;
+			swsqe->opcode = FW_RI_LOCAL_INV;
+			err = build_inv_stag(wqe, wr, &len16);
+			break;
+		default:
+			PDBG("%s post of type=%d TBD!\n", __func__,
+			     wr->opcode);
+			err = -EINVAL;
+		}
+		if (err) {
+			*bad_wr = wr;
+			break;
+		}
+		swsqe->idx = qhp->wq.sq.pidx;
+		swsqe->complete = 0;
+		swsqe->signaled = (wr->send_flags & IB_SEND_SIGNALED) ||
+				  qhp->sq_sig_all;
+		swsqe->flushed = 0;
+		swsqe->wr_id = wr->wr_id;
+
+		init_wr_hdr(wqe, qhp->wq.sq.pidx, fw_opcode, fw_flags, len16);
+
+		PDBG("%s cookie 0x%llx pidx 0x%x opcode 0x%x read_len %u\n",
+		     __func__, (unsigned long long)wr->wr_id, qhp->wq.sq.pidx,
+		     swsqe->opcode, swsqe->read_len);
+		wr = wr->next;
+		num_wrs--;
+		t4_sq_produce(&qhp->wq, len16);
+		idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+	}
+	if (!qhp->rhp->rdev.status_page->db_off) {
+		t4_ring_sq_db(&qhp->wq, idx,
+			      is_t5(qhp->rhp->rdev.lldi.adapter_type), wqe);
+		spin_unlock_irqrestore(&qhp->lock, flag);
+	} else {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		ring_kernel_sq_db(qhp, idx);
+	}
+	return err;
+}
+
+int c4iw_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr)
+{
+	int err = 0;
+	struct c4iw_qp *qhp;
+	union t4_recv_wr *wqe = NULL;
+	u32 num_wrs;
+	u8 len16 = 0;
+	unsigned long flag;
+	u16 idx = 0;
+
+	qhp = to_c4iw_qp(ibqp);
+	spin_lock_irqsave(&qhp->lock, flag);
+	if (t4_wq_in_error(&qhp->wq)) {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		return -EINVAL;
+	}
+	num_wrs = t4_rq_avail(&qhp->wq);
+	if (num_wrs == 0) {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		return -ENOMEM;
+	}
+	while (wr) {
+		if (wr->num_sge > T4_MAX_RECV_SGE) {
+			err = -EINVAL;
+			*bad_wr = wr;
+			break;
+		}
+		wqe = (union t4_recv_wr *)((u8 *)qhp->wq.rq.queue +
+					   qhp->wq.rq.wq_pidx *
+					   T4_EQ_ENTRY_SIZE);
+		if (num_wrs)
+			err = build_rdma_recv(qhp, wqe, wr, &len16);
+		else
+			err = -ENOMEM;
+		if (err) {
+			*bad_wr = wr;
+			break;
+		}
+
+		qhp->wq.rq.sw_rq[qhp->wq.rq.pidx].wr_id = wr->wr_id;
+
+		wqe->recv.opcode = FW_RI_RECV_WR;
+		wqe->recv.r1 = 0;
+		wqe->recv.wrid = qhp->wq.rq.pidx;
+		wqe->recv.r2[0] = 0;
+		wqe->recv.r2[1] = 0;
+		wqe->recv.r2[2] = 0;
+		wqe->recv.len16 = len16;
+		PDBG("%s cookie 0x%llx pidx %u\n", __func__,
+		     (unsigned long long) wr->wr_id, qhp->wq.rq.pidx);
+		t4_rq_produce(&qhp->wq, len16);
+		idx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+		wr = wr->next;
+		num_wrs--;
+	}
+	if (!qhp->rhp->rdev.status_page->db_off) {
+		t4_ring_rq_db(&qhp->wq, idx,
+			      is_t5(qhp->rhp->rdev.lldi.adapter_type), wqe);
+		spin_unlock_irqrestore(&qhp->lock, flag);
+	} else {
+		spin_unlock_irqrestore(&qhp->lock, flag);
+		ring_kernel_rq_db(qhp, idx);
+	}
+	return err;
+}
+
+int c4iw_bind_mw(struct ib_qp *qp, struct ib_mw *mw, struct ib_mw_bind *mw_bind)
+{
+	return -ENOSYS;
+}
+
+static inline void build_term_codes(struct t4_cqe *err_cqe, u8 *layer_type,
+				    u8 *ecode)
+{
+	int status;
+	int tagged;
+	int opcode;
+	int rqtype;
+	int send_inv;
+
+	if (!err_cqe) {
+		*layer_type = LAYER_RDMAP|DDP_LOCAL_CATA;
+		*ecode = 0;
+		return;
+	}
+
+	status = CQE_STATUS(err_cqe);
+	opcode = CQE_OPCODE(err_cqe);
+	rqtype = RQ_TYPE(err_cqe);
+	send_inv = (opcode == FW_RI_SEND_WITH_INV) ||
+		   (opcode == FW_RI_SEND_WITH_SE_INV);
+	tagged = (opcode == FW_RI_RDMA_WRITE) ||
+		 (rqtype && (opcode == FW_RI_READ_RESP));
+
+	switch (status) {
+	case T4_ERR_STAG:
+		if (send_inv) {
+			*layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
+			*ecode = RDMAP_CANT_INV_STAG;
+		} else {
+			*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+			*ecode = RDMAP_INV_STAG;
+		}
+		break;
+	case T4_ERR_PDID:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+		if ((opcode == FW_RI_SEND_WITH_INV) ||
+		    (opcode == FW_RI_SEND_WITH_SE_INV))
+			*ecode = RDMAP_CANT_INV_STAG;
+		else
+			*ecode = RDMAP_STAG_NOT_ASSOC;
+		break;
+	case T4_ERR_QPID:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+		*ecode = RDMAP_STAG_NOT_ASSOC;
+		break;
+	case T4_ERR_ACCESS:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+		*ecode = RDMAP_ACC_VIOL;
+		break;
+	case T4_ERR_WRAP:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+		*ecode = RDMAP_TO_WRAP;
+		break;
+	case T4_ERR_BOUND:
+		if (tagged) {
+			*layer_type = LAYER_DDP|DDP_TAGGED_ERR;
+			*ecode = DDPT_BASE_BOUNDS;
+		} else {
+			*layer_type = LAYER_RDMAP|RDMAP_REMOTE_PROT;
+			*ecode = RDMAP_BASE_BOUNDS;
+		}
+		break;
+	case T4_ERR_INVALIDATE_SHARED_MR:
+	case T4_ERR_INVALIDATE_MR_WITH_MW_BOUND:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
+		*ecode = RDMAP_CANT_INV_STAG;
+		break;
+	case T4_ERR_ECC:
+	case T4_ERR_ECC_PSTAG:
+	case T4_ERR_INTERNAL_ERR:
+		*layer_type = LAYER_RDMAP|RDMAP_LOCAL_CATA;
+		*ecode = 0;
+		break;
+	case T4_ERR_OUT_OF_RQE:
+		*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+		*ecode = DDPU_INV_MSN_NOBUF;
+		break;
+	case T4_ERR_PBL_ADDR_BOUND:
+		*layer_type = LAYER_DDP|DDP_TAGGED_ERR;
+		*ecode = DDPT_BASE_BOUNDS;
+		break;
+	case T4_ERR_CRC:
+		*layer_type = LAYER_MPA|DDP_LLP;
+		*ecode = MPA_CRC_ERR;
+		break;
+	case T4_ERR_MARKER:
+		*layer_type = LAYER_MPA|DDP_LLP;
+		*ecode = MPA_MARKER_ERR;
+		break;
+	case T4_ERR_PDU_LEN_ERR:
+		*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+		*ecode = DDPU_MSG_TOOBIG;
+		break;
+	case T4_ERR_DDP_VERSION:
+		if (tagged) {
+			*layer_type = LAYER_DDP|DDP_TAGGED_ERR;
+			*ecode = DDPT_INV_VERS;
+		} else {
+			*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+			*ecode = DDPU_INV_VERS;
+		}
+		break;
+	case T4_ERR_RDMA_VERSION:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
+		*ecode = RDMAP_INV_VERS;
+		break;
+	case T4_ERR_OPCODE:
+		*layer_type = LAYER_RDMAP|RDMAP_REMOTE_OP;
+		*ecode = RDMAP_INV_OPCODE;
+		break;
+	case T4_ERR_DDP_QUEUE_NUM:
+		*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+		*ecode = DDPU_INV_QN;
+		break;
+	case T4_ERR_MSN:
+	case T4_ERR_MSN_GAP:
+	case T4_ERR_MSN_RANGE:
+	case T4_ERR_IRD_OVERFLOW:
+		*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+		*ecode = DDPU_INV_MSN_RANGE;
+		break;
+	case T4_ERR_TBIT:
+		*layer_type = LAYER_DDP|DDP_LOCAL_CATA;
+		*ecode = 0;
+		break;
+	case T4_ERR_MO:
+		*layer_type = LAYER_DDP|DDP_UNTAGGED_ERR;
+		*ecode = DDPU_INV_MO;
+		break;
+	default:
+		*layer_type = LAYER_RDMAP|DDP_LOCAL_CATA;
+		*ecode = 0;
+		break;
+	}
+}
+
+static void post_terminate(struct c4iw_qp *qhp, struct t4_cqe *err_cqe,
+			   gfp_t gfp)
+{
+	struct fw_ri_wr *wqe;
+	struct sk_buff *skb;
+	struct terminate_message *term;
+
+	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
+	     qhp->ep->hwtid);
+
+	skb = alloc_skb(sizeof *wqe, gfp);
+	if (!skb)
+		return;
+	set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx);
+
+	wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe));
+	memset(wqe, 0, sizeof *wqe);
+	wqe->op_compl = cpu_to_be32(FW_WR_OP(FW_RI_INIT_WR));
+	wqe->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(qhp->ep->hwtid) |
+		FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16)));
+
+	wqe->u.terminate.type = FW_RI_TYPE_TERMINATE;
+	wqe->u.terminate.immdlen = cpu_to_be32(sizeof *term);
+	term = (struct terminate_message *)wqe->u.terminate.termmsg;
+	if (qhp->attr.layer_etype == (LAYER_MPA|DDP_LLP)) {
+		term->layer_etype = qhp->attr.layer_etype;
+		term->ecode = qhp->attr.ecode;
+	} else
+		build_term_codes(err_cqe, &term->layer_etype, &term->ecode);
+	c4iw_ofld_send(&qhp->rhp->rdev, skb);
+}
+
+/*
+ * Assumes qhp lock is held.
+ */
+static void __flush_qp(struct c4iw_qp *qhp, struct c4iw_cq *rchp,
+		       struct c4iw_cq *schp)
+{
+	int count;
+	int flushed;
+	unsigned long flag;
+
+	PDBG("%s qhp %p rchp %p schp %p\n", __func__, qhp, rchp, schp);
+
+	/* locking hierarchy: cq lock first, then qp lock. */
+	spin_lock_irqsave(&rchp->lock, flag);
+	spin_lock(&qhp->lock);
+
+	if (qhp->wq.flushed) {
+		spin_unlock(&qhp->lock);
+		spin_unlock_irqrestore(&rchp->lock, flag);
+		return;
+	}
+	qhp->wq.flushed = 1;
+
+	c4iw_flush_hw_cq(rchp);
+	c4iw_count_rcqes(&rchp->cq, &qhp->wq, &count);
+	flushed = c4iw_flush_rq(&qhp->wq, &rchp->cq, count);
+	spin_unlock(&qhp->lock);
+	spin_unlock_irqrestore(&rchp->lock, flag);
+	if (flushed) {
+		spin_lock_irqsave(&rchp->comp_handler_lock, flag);
+		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
+	}
+
+	/* locking hierarchy: cq lock first, then qp lock. */
+	spin_lock_irqsave(&schp->lock, flag);
+	spin_lock(&qhp->lock);
+	if (schp != rchp)
+		c4iw_flush_hw_cq(schp);
+	flushed = c4iw_flush_sq(qhp);
+	spin_unlock(&qhp->lock);
+	spin_unlock_irqrestore(&schp->lock, flag);
+	if (flushed) {
+		spin_lock_irqsave(&schp->comp_handler_lock, flag);
+		(*schp->ibcq.comp_handler)(&schp->ibcq, schp->ibcq.cq_context);
+		spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
+	}
+}
+
+static void flush_qp(struct c4iw_qp *qhp)
+{
+	struct c4iw_cq *rchp, *schp;
+	unsigned long flag;
+
+	rchp = to_c4iw_cq(qhp->ibqp.recv_cq);
+	schp = to_c4iw_cq(qhp->ibqp.send_cq);
+
+	t4_set_wq_in_error(&qhp->wq);
+	if (qhp->ibqp.uobject) {
+		t4_set_cq_in_error(&rchp->cq);
+		spin_lock_irqsave(&rchp->comp_handler_lock, flag);
+		(*rchp->ibcq.comp_handler)(&rchp->ibcq, rchp->ibcq.cq_context);
+		spin_unlock_irqrestore(&rchp->comp_handler_lock, flag);
+		if (schp != rchp) {
+			t4_set_cq_in_error(&schp->cq);
+			spin_lock_irqsave(&schp->comp_handler_lock, flag);
+			(*schp->ibcq.comp_handler)(&schp->ibcq,
+					schp->ibcq.cq_context);
+			spin_unlock_irqrestore(&schp->comp_handler_lock, flag);
+		}
+		return;
+	}
+	__flush_qp(qhp, rchp, schp);
+}
+
+static int rdma_fini(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
+		     struct c4iw_ep *ep)
+{
+	struct fw_ri_wr *wqe;
+	int ret;
+	struct sk_buff *skb;
+
+	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
+	     ep->hwtid);
+
+	skb = alloc_skb(sizeof *wqe, GFP_KERNEL);
+	if (!skb)
+		return -ENOMEM;
+	set_wr_txq(skb, CPL_PRIORITY_DATA, ep->txq_idx);
+
+	wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe));
+	memset(wqe, 0, sizeof *wqe);
+	wqe->op_compl = cpu_to_be32(
+		FW_WR_OP(FW_RI_INIT_WR) |
+		FW_WR_COMPL(1));
+	wqe->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(ep->hwtid) |
+		FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16)));
+	wqe->cookie = (unsigned long) &ep->com.wr_wait;
+
+	wqe->u.fini.type = FW_RI_TYPE_FINI;
+	ret = c4iw_ofld_send(&rhp->rdev, skb);
+	if (ret)
+		goto out;
+
+	ret = c4iw_wait_for_reply(&rhp->rdev, &ep->com.wr_wait, qhp->ep->hwtid,
+			     qhp->wq.sq.qid, __func__);
+out:
+	PDBG("%s ret %d\n", __func__, ret);
+	return ret;
+}
+
+static void build_rtr_msg(u8 p2p_type, struct fw_ri_init *init)
+{
+	PDBG("%s p2p_type = %d\n", __func__, p2p_type);
+	memset(&init->u, 0, sizeof init->u);
+	switch (p2p_type) {
+	case FW_RI_INIT_P2PTYPE_RDMA_WRITE:
+		init->u.write.opcode = FW_RI_RDMA_WRITE_WR;
+		init->u.write.stag_sink = cpu_to_be32(1);
+		init->u.write.to_sink = cpu_to_be64(1);
+		init->u.write.u.immd_src[0].op = FW_RI_DATA_IMMD;
+		init->u.write.len16 = DIV_ROUND_UP(sizeof init->u.write +
+						   sizeof(struct fw_ri_immd),
+						   16);
+		break;
+	case FW_RI_INIT_P2PTYPE_READ_REQ:
+		init->u.write.opcode = FW_RI_RDMA_READ_WR;
+		init->u.read.stag_src = cpu_to_be32(1);
+		init->u.read.to_src_lo = cpu_to_be32(1);
+		init->u.read.stag_sink = cpu_to_be32(1);
+		init->u.read.to_sink_lo = cpu_to_be32(1);
+		init->u.read.len16 = DIV_ROUND_UP(sizeof init->u.read, 16);
+		break;
+	}
+}
+
+static int rdma_init(struct c4iw_dev *rhp, struct c4iw_qp *qhp)
+{
+	struct fw_ri_wr *wqe;
+	int ret;
+	struct sk_buff *skb;
+
+	PDBG("%s qhp %p qid 0x%x tid %u\n", __func__, qhp, qhp->wq.sq.qid,
+	     qhp->ep->hwtid);
+
+	skb = alloc_skb(sizeof *wqe, GFP_KERNEL);
+	if (!skb)
+		return -ENOMEM;
+	set_wr_txq(skb, CPL_PRIORITY_DATA, qhp->ep->txq_idx);
+
+	wqe = (struct fw_ri_wr *)__skb_put(skb, sizeof(*wqe));
+	memset(wqe, 0, sizeof *wqe);
+	wqe->op_compl = cpu_to_be32(
+		FW_WR_OP(FW_RI_INIT_WR) |
+		FW_WR_COMPL(1));
+	wqe->flowid_len16 = cpu_to_be32(
+		FW_WR_FLOWID(qhp->ep->hwtid) |
+		FW_WR_LEN16(DIV_ROUND_UP(sizeof *wqe, 16)));
+
+	wqe->cookie = (unsigned long) &qhp->ep->com.wr_wait;
+
+	wqe->u.init.type = FW_RI_TYPE_INIT;
+	wqe->u.init.mpareqbit_p2ptype =
+		V_FW_RI_WR_MPAREQBIT(qhp->attr.mpa_attr.initiator) |
+		V_FW_RI_WR_P2PTYPE(qhp->attr.mpa_attr.p2p_type);
+	wqe->u.init.mpa_attrs = FW_RI_MPA_IETF_ENABLE;
+	if (qhp->attr.mpa_attr.recv_marker_enabled)
+		wqe->u.init.mpa_attrs |= FW_RI_MPA_RX_MARKER_ENABLE;
+	if (qhp->attr.mpa_attr.xmit_marker_enabled)
+		wqe->u.init.mpa_attrs |= FW_RI_MPA_TX_MARKER_ENABLE;
+	if (qhp->attr.mpa_attr.crc_enabled)
+		wqe->u.init.mpa_attrs |= FW_RI_MPA_CRC_ENABLE;
+
+	wqe->u.init.qp_caps = FW_RI_QP_RDMA_READ_ENABLE |
+			    FW_RI_QP_RDMA_WRITE_ENABLE |
+			    FW_RI_QP_BIND_ENABLE;
+	if (!qhp->ibqp.uobject)
+		wqe->u.init.qp_caps |= FW_RI_QP_FAST_REGISTER_ENABLE |
+				     FW_RI_QP_STAG0_ENABLE;
+	wqe->u.init.nrqe = cpu_to_be16(t4_rqes_posted(&qhp->wq));
+	wqe->u.init.pdid = cpu_to_be32(qhp->attr.pd);
+	wqe->u.init.qpid = cpu_to_be32(qhp->wq.sq.qid);
+	wqe->u.init.sq_eqid = cpu_to_be32(qhp->wq.sq.qid);
+	wqe->u.init.rq_eqid = cpu_to_be32(qhp->wq.rq.qid);
+	wqe->u.init.scqid = cpu_to_be32(qhp->attr.scq);
+	wqe->u.init.rcqid = cpu_to_be32(qhp->attr.rcq);
+	wqe->u.init.ord_max = cpu_to_be32(qhp->attr.max_ord);
+	wqe->u.init.ird_max = cpu_to_be32(qhp->attr.max_ird);
+	wqe->u.init.iss = cpu_to_be32(qhp->ep->snd_seq);
+	wqe->u.init.irs = cpu_to_be32(qhp->ep->rcv_seq);
+	wqe->u.init.hwrqsize = cpu_to_be32(qhp->wq.rq.rqt_size);
+	wqe->u.init.hwrqaddr = cpu_to_be32(qhp->wq.rq.rqt_hwaddr -
+					 rhp->rdev.lldi.vr->rq.start);
+	if (qhp->attr.mpa_attr.initiator)
+		build_rtr_msg(qhp->attr.mpa_attr.p2p_type, &wqe->u.init);
+
+	ret = c4iw_ofld_send(&rhp->rdev, skb);
+	if (ret)
+		goto out;
+
+	ret = c4iw_wait_for_reply(&rhp->rdev, &qhp->ep->com.wr_wait,
+				  qhp->ep->hwtid, qhp->wq.sq.qid, __func__);
+out:
+	PDBG("%s ret %d\n", __func__, ret);
+	return ret;
+}
+
+int c4iw_modify_qp(struct c4iw_dev *rhp, struct c4iw_qp *qhp,
+		   enum c4iw_qp_attr_mask mask,
+		   struct c4iw_qp_attributes *attrs,
+		   int internal)
+{
+	int ret = 0;
+	struct c4iw_qp_attributes newattr = qhp->attr;
+	int disconnect = 0;
+	int terminate = 0;
+	int abort = 0;
+	int free = 0;
+	struct c4iw_ep *ep = NULL;
+
+	PDBG("%s qhp %p sqid 0x%x rqid 0x%x ep %p state %d -> %d\n", __func__,
+	     qhp, qhp->wq.sq.qid, qhp->wq.rq.qid, qhp->ep, qhp->attr.state,
+	     (mask & C4IW_QP_ATTR_NEXT_STATE) ? attrs->next_state : -1);
+
+	mutex_lock(&qhp->mutex);
+
+	/* Process attr changes if in IDLE */
+	if (mask & C4IW_QP_ATTR_VALID_MODIFY) {
+		if (qhp->attr.state != C4IW_QP_STATE_IDLE) {
+			ret = -EIO;
+			goto out;
+		}
+		if (mask & C4IW_QP_ATTR_ENABLE_RDMA_READ)
+			newattr.enable_rdma_read = attrs->enable_rdma_read;
+		if (mask & C4IW_QP_ATTR_ENABLE_RDMA_WRITE)
+			newattr.enable_rdma_write = attrs->enable_rdma_write;
+		if (mask & C4IW_QP_ATTR_ENABLE_RDMA_BIND)
+			newattr.enable_bind = attrs->enable_bind;
+		if (mask & C4IW_QP_ATTR_MAX_ORD) {
+			if (attrs->max_ord > c4iw_max_read_depth) {
+				ret = -EINVAL;
+				goto out;
+			}
+			newattr.max_ord = attrs->max_ord;
+		}
+		if (mask & C4IW_QP_ATTR_MAX_IRD) {
+			if (attrs->max_ird > c4iw_max_read_depth) {
+				ret = -EINVAL;
+				goto out;
+			}
+			newattr.max_ird = attrs->max_ird;
+		}
+		qhp->attr = newattr;
+	}
+
+	if (mask & C4IW_QP_ATTR_SQ_DB) {
+		ret = ring_kernel_sq_db(qhp, attrs->sq_db_inc);
+		goto out;
+	}
+	if (mask & C4IW_QP_ATTR_RQ_DB) {
+		ret = ring_kernel_rq_db(qhp, attrs->rq_db_inc);
+		goto out;
+	}
+
+	if (!(mask & C4IW_QP_ATTR_NEXT_STATE))
+		goto out;
+	if (qhp->attr.state == attrs->next_state)
+		goto out;
+
+	switch (qhp->attr.state) {
+	case C4IW_QP_STATE_IDLE:
+		switch (attrs->next_state) {
+		case C4IW_QP_STATE_RTS:
+			if (!(mask & C4IW_QP_ATTR_LLP_STREAM_HANDLE)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			if (!(mask & C4IW_QP_ATTR_MPA_ATTR)) {
+				ret = -EINVAL;
+				goto out;
+			}
+			qhp->attr.mpa_attr = attrs->mpa_attr;
+			qhp->attr.llp_stream_handle = attrs->llp_stream_handle;
+			qhp->ep = qhp->attr.llp_stream_handle;
+			set_state(qhp, C4IW_QP_STATE_RTS);
+
+			/*
+			 * Ref the endpoint here and deref when we
+			 * disassociate the endpoint from the QP.  This
+			 * happens in CLOSING->IDLE transition or *->ERROR
+			 * transition.
+			 */
+			c4iw_get_ep(&qhp->ep->com);
+			ret = rdma_init(rhp, qhp);
+			if (ret)
+				goto err;
+			break;
+		case C4IW_QP_STATE_ERROR:
+			set_state(qhp, C4IW_QP_STATE_ERROR);
+			flush_qp(qhp);
+			break;
+		default:
+			ret = -EINVAL;
+			goto out;
+		}
+		break;
+	case C4IW_QP_STATE_RTS:
+		switch (attrs->next_state) {
+		case C4IW_QP_STATE_CLOSING:
+			BUG_ON(atomic_read(&qhp->ep->com.kref.refcount) < 2);
+			t4_set_wq_in_error(&qhp->wq);
+			set_state(qhp, C4IW_QP_STATE_CLOSING);
+			ep = qhp->ep;
+			if (!internal) {
+				abort = 0;
+				disconnect = 1;
+				c4iw_get_ep(&qhp->ep->com);
+			}
+			ret = rdma_fini(rhp, qhp, ep);
+			if (ret)
+				goto err;
+			break;
+		case C4IW_QP_STATE_TERMINATE:
+			t4_set_wq_in_error(&qhp->wq);
+			set_state(qhp, C4IW_QP_STATE_TERMINATE);
+			qhp->attr.layer_etype = attrs->layer_etype;
+			qhp->attr.ecode = attrs->ecode;
+			ep = qhp->ep;
+			if (!internal) {
+				c4iw_get_ep(&qhp->ep->com);
+				terminate = 1;
+				disconnect = 1;
+			} else {
+				terminate = qhp->attr.send_term;
+				ret = rdma_fini(rhp, qhp, ep);
+				if (ret)
+					goto err;
+			}
+			break;
+		case C4IW_QP_STATE_ERROR:
+			t4_set_wq_in_error(&qhp->wq);
+			set_state(qhp, C4IW_QP_STATE_ERROR);
+			if (!internal) {
+				abort = 1;
+				disconnect = 1;
+				ep = qhp->ep;
+				c4iw_get_ep(&qhp->ep->com);
+			}
+			goto err;
+			break;
+		default:
+			ret = -EINVAL;
+			goto out;
+		}
+		break;
+	case C4IW_QP_STATE_CLOSING:
+		if (!internal) {
+			ret = -EINVAL;
+			goto out;
+		}
+		switch (attrs->next_state) {
+		case C4IW_QP_STATE_IDLE:
+			flush_qp(qhp);
+			set_state(qhp, C4IW_QP_STATE_IDLE);
+			qhp->attr.llp_stream_handle = NULL;
+			c4iw_put_ep(&qhp->ep->com);
+			qhp->ep = NULL;
+			wake_up(&qhp->wait);
+			break;
+		case C4IW_QP_STATE_ERROR:
+			goto err;
+		default:
+			ret = -EINVAL;
+			goto err;
+		}
+		break;
+	case C4IW_QP_STATE_ERROR:
+		if (attrs->next_state != C4IW_QP_STATE_IDLE) {
+			ret = -EINVAL;
+			goto out;
+		}
+		if (!t4_sq_empty(&qhp->wq) || !t4_rq_empty(&qhp->wq)) {
+			ret = -EINVAL;
+			goto out;
+		}
+		set_state(qhp, C4IW_QP_STATE_IDLE);
+		break;
+	case C4IW_QP_STATE_TERMINATE:
+		if (!internal) {
+			ret = -EINVAL;
+			goto out;
+		}
+		goto err;
+		break;
+	default:
+		printk(KERN_ERR "%s in a bad state %d\n",
+		       __func__, qhp->attr.state);
+		ret = -EINVAL;
+		goto err;
+		break;
+	}
+	goto out;
+err:
+	PDBG("%s disassociating ep %p qpid 0x%x\n", __func__, qhp->ep,
+	     qhp->wq.sq.qid);
+
+	/* disassociate the LLP connection */
+	qhp->attr.llp_stream_handle = NULL;
+	if (!ep)
+		ep = qhp->ep;
+	qhp->ep = NULL;
+	set_state(qhp, C4IW_QP_STATE_ERROR);
+	free = 1;
+	abort = 1;
+	wake_up(&qhp->wait);
+	BUG_ON(!ep);
+	flush_qp(qhp);
+out:
+	mutex_unlock(&qhp->mutex);
+
+	if (terminate)
+		post_terminate(qhp, NULL, internal ? GFP_ATOMIC : GFP_KERNEL);
+
+	/*
+	 * If disconnect is 1, then we need to initiate a disconnect
+	 * on the EP.  This can be a normal close (RTS->CLOSING) or
+	 * an abnormal close (RTS/CLOSING->ERROR).
+	 */
+	if (disconnect) {
+		c4iw_ep_disconnect(ep, abort, internal ? GFP_ATOMIC :
+							 GFP_KERNEL);
+		c4iw_put_ep(&ep->com);
+	}
+
+	/*
+	 * If free is 1, then we've disassociated the EP from the QP
+	 * and we need to dereference the EP.
+	 */
+	if (free)
+		c4iw_put_ep(&ep->com);
+	PDBG("%s exit state %d\n", __func__, qhp->attr.state);
+	return ret;
+}
+
+int c4iw_destroy_qp(struct ib_qp *ib_qp)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_qp *qhp;
+	struct c4iw_qp_attributes attrs;
+	struct c4iw_ucontext *ucontext;
+
+	qhp = to_c4iw_qp(ib_qp);
+	rhp = qhp->rhp;
+
+	attrs.next_state = C4IW_QP_STATE_ERROR;
+	if (qhp->attr.state == C4IW_QP_STATE_TERMINATE)
+		c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 1);
+	else
+		c4iw_modify_qp(rhp, qhp, C4IW_QP_ATTR_NEXT_STATE, &attrs, 0);
+	wait_event(qhp->wait, !qhp->ep);
+
+	remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid);
+	atomic_dec(&qhp->refcnt);
+	wait_event(qhp->wait, !atomic_read(&qhp->refcnt));
+
+	spin_lock_irq(&rhp->lock);
+	if (!list_empty(&qhp->db_fc_entry))
+		list_del_init(&qhp->db_fc_entry);
+	spin_unlock_irq(&rhp->lock);
+
+	ucontext = ib_qp->uobject ?
+		   to_c4iw_ucontext(ib_qp->uobject->context) : NULL;
+	destroy_qp(&rhp->rdev, &qhp->wq,
+		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+
+	PDBG("%s ib_qp %p qpid 0x%0x\n", __func__, ib_qp, qhp->wq.sq.qid);
+	kfree(qhp);
+	return 0;
+}
+
+struct ib_qp *c4iw_create_qp(struct ib_pd *pd, struct ib_qp_init_attr *attrs,
+			     struct ib_udata *udata)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_qp *qhp;
+	struct c4iw_pd *php;
+	struct c4iw_cq *schp;
+	struct c4iw_cq *rchp;
+	struct c4iw_create_qp_resp uresp;
+	unsigned int sqsize, rqsize;
+	struct c4iw_ucontext *ucontext;
+	int ret;
+	struct c4iw_mm_entry *mm1, *mm2, *mm3, *mm4, *mm5 = NULL;
+
+	PDBG("%s ib_pd %p\n", __func__, pd);
+
+	if (attrs->qp_type != IB_QPT_RC)
+		return ERR_PTR(-EINVAL);
+
+	php = to_c4iw_pd(pd);
+	rhp = php->rhp;
+	schp = get_chp(rhp, ((struct c4iw_cq *)attrs->send_cq)->cq.cqid);
+	rchp = get_chp(rhp, ((struct c4iw_cq *)attrs->recv_cq)->cq.cqid);
+	if (!schp || !rchp)
+		return ERR_PTR(-EINVAL);
+
+	if (attrs->cap.max_inline_data > T4_MAX_SEND_INLINE)
+		return ERR_PTR(-EINVAL);
+
+	rqsize = roundup(attrs->cap.max_recv_wr + 1, 16);
+	if (rqsize > T4_MAX_RQ_SIZE)
+		return ERR_PTR(-E2BIG);
+
+	sqsize = roundup(attrs->cap.max_send_wr + 1, 16);
+	if (sqsize > T4_MAX_SQ_SIZE)
+		return ERR_PTR(-E2BIG);
+
+	ucontext = pd->uobject ? to_c4iw_ucontext(pd->uobject->context) : NULL;
+
+	qhp = kzalloc(sizeof(*qhp), GFP_KERNEL);
+	if (!qhp)
+		return ERR_PTR(-ENOMEM);
+	qhp->wq.sq.size = sqsize;
+	qhp->wq.sq.memsize = (sqsize + 1) * sizeof *qhp->wq.sq.queue;
+	qhp->wq.sq.flush_cidx = -1;
+	qhp->wq.rq.size = rqsize;
+	qhp->wq.rq.memsize = (rqsize + 1) * sizeof *qhp->wq.rq.queue;
+
+	if (ucontext) {
+		qhp->wq.sq.memsize = roundup(qhp->wq.sq.memsize, PAGE_SIZE);
+		qhp->wq.rq.memsize = roundup(qhp->wq.rq.memsize, PAGE_SIZE);
+	}
+
+	PDBG("%s sqsize %u sqmemsize %zu rqsize %u rqmemsize %zu\n",
+	     __func__, sqsize, qhp->wq.sq.memsize, rqsize, qhp->wq.rq.memsize);
+
+	ret = create_qp(&rhp->rdev, &qhp->wq, &schp->cq, &rchp->cq,
+			ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+	if (ret)
+		goto err1;
+
+	attrs->cap.max_recv_wr = rqsize - 1;
+	attrs->cap.max_send_wr = sqsize - 1;
+	attrs->cap.max_inline_data = T4_MAX_SEND_INLINE;
+
+	qhp->rhp = rhp;
+	qhp->attr.pd = php->pdid;
+	qhp->attr.scq = ((struct c4iw_cq *) attrs->send_cq)->cq.cqid;
+	qhp->attr.rcq = ((struct c4iw_cq *) attrs->recv_cq)->cq.cqid;
+	qhp->attr.sq_num_entries = attrs->cap.max_send_wr;
+	qhp->attr.rq_num_entries = attrs->cap.max_recv_wr;
+	qhp->attr.sq_max_sges = attrs->cap.max_send_sge;
+	qhp->attr.sq_max_sges_rdma_write = attrs->cap.max_send_sge;
+	qhp->attr.rq_max_sges = attrs->cap.max_recv_sge;
+	qhp->attr.state = C4IW_QP_STATE_IDLE;
+	qhp->attr.next_state = C4IW_QP_STATE_IDLE;
+	qhp->attr.enable_rdma_read = 1;
+	qhp->attr.enable_rdma_write = 1;
+	qhp->attr.enable_bind = 1;
+	qhp->attr.max_ord = 1;
+	qhp->attr.max_ird = 1;
+	qhp->sq_sig_all = attrs->sq_sig_type == IB_SIGNAL_ALL_WR;
+	spin_lock_init(&qhp->lock);
+	mutex_init(&qhp->mutex);
+	init_waitqueue_head(&qhp->wait);
+	atomic_set(&qhp->refcnt, 1);
+
+	ret = insert_handle(rhp, &rhp->qpidr, qhp, qhp->wq.sq.qid);
+	if (ret)
+		goto err2;
+
+	if (udata) {
+		mm1 = kmalloc(sizeof *mm1, GFP_KERNEL);
+		if (!mm1) {
+			ret = -ENOMEM;
+			goto err3;
+		}
+		mm2 = kmalloc(sizeof *mm2, GFP_KERNEL);
+		if (!mm2) {
+			ret = -ENOMEM;
+			goto err4;
+		}
+		mm3 = kmalloc(sizeof *mm3, GFP_KERNEL);
+		if (!mm3) {
+			ret = -ENOMEM;
+			goto err5;
+		}
+		mm4 = kmalloc(sizeof *mm4, GFP_KERNEL);
+		if (!mm4) {
+			ret = -ENOMEM;
+			goto err6;
+		}
+		if (t4_sq_onchip(&qhp->wq.sq)) {
+			mm5 = kmalloc(sizeof *mm5, GFP_KERNEL);
+			if (!mm5) {
+				ret = -ENOMEM;
+				goto err7;
+			}
+			uresp.flags = C4IW_QPF_ONCHIP;
+		} else
+			uresp.flags = 0;
+		uresp.qid_mask = rhp->rdev.qpmask;
+		uresp.sqid = qhp->wq.sq.qid;
+		uresp.sq_size = qhp->wq.sq.size;
+		uresp.sq_memsize = qhp->wq.sq.memsize;
+		uresp.rqid = qhp->wq.rq.qid;
+		uresp.rq_size = qhp->wq.rq.size;
+		uresp.rq_memsize = qhp->wq.rq.memsize;
+		spin_lock(&ucontext->mmap_lock);
+		if (mm5) {
+			uresp.ma_sync_key = ucontext->key;
+			ucontext->key += PAGE_SIZE;
+		} else {
+			uresp.ma_sync_key =  0;
+		}
+		uresp.sq_key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		uresp.rq_key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		uresp.sq_db_gts_key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		uresp.rq_db_gts_key = ucontext->key;
+		ucontext->key += PAGE_SIZE;
+		spin_unlock(&ucontext->mmap_lock);
+		ret = ib_copy_to_udata(udata, &uresp, sizeof uresp);
+		if (ret)
+			goto err8;
+		mm1->key = uresp.sq_key;
+		mm1->addr = qhp->wq.sq.phys_addr;
+		mm1->len = PAGE_ALIGN(qhp->wq.sq.memsize);
+		insert_mmap(ucontext, mm1);
+		mm2->key = uresp.rq_key;
+		mm2->addr = virt_to_phys(qhp->wq.rq.queue);
+		mm2->len = PAGE_ALIGN(qhp->wq.rq.memsize);
+		insert_mmap(ucontext, mm2);
+		mm3->key = uresp.sq_db_gts_key;
+		mm3->addr = (__force unsigned long) qhp->wq.sq.udb;
+		mm3->len = PAGE_SIZE;
+		insert_mmap(ucontext, mm3);
+		mm4->key = uresp.rq_db_gts_key;
+		mm4->addr = (__force unsigned long) qhp->wq.rq.udb;
+		mm4->len = PAGE_SIZE;
+		insert_mmap(ucontext, mm4);
+		if (mm5) {
+			mm5->key = uresp.ma_sync_key;
+			mm5->addr = (pci_resource_start(rhp->rdev.lldi.pdev, 0)
+				    + A_PCIE_MA_SYNC) & PAGE_MASK;
+			mm5->len = PAGE_SIZE;
+			insert_mmap(ucontext, mm5);
+		}
+	}
+	qhp->ibqp.qp_num = qhp->wq.sq.qid;
+	init_timer(&(qhp->timer));
+	INIT_LIST_HEAD(&qhp->db_fc_entry);
+	PDBG("%s qhp %p sq_num_entries %d, rq_num_entries %d qpid 0x%0x\n",
+	     __func__, qhp, qhp->attr.sq_num_entries, qhp->attr.rq_num_entries,
+	     qhp->wq.sq.qid);
+	return &qhp->ibqp;
+err8:
+	kfree(mm5);
+err7:
+	kfree(mm4);
+err6:
+	kfree(mm3);
+err5:
+	kfree(mm2);
+err4:
+	kfree(mm1);
+err3:
+	remove_handle(rhp, &rhp->qpidr, qhp->wq.sq.qid);
+err2:
+	destroy_qp(&rhp->rdev, &qhp->wq,
+		   ucontext ? &ucontext->uctx : &rhp->rdev.uctx);
+err1:
+	kfree(qhp);
+	return ERR_PTR(ret);
+}
+
+int c4iw_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask, struct ib_udata *udata)
+{
+	struct c4iw_dev *rhp;
+	struct c4iw_qp *qhp;
+	enum c4iw_qp_attr_mask mask = 0;
+	struct c4iw_qp_attributes attrs;
+
+	PDBG("%s ib_qp %p\n", __func__, ibqp);
+
+	/* iwarp does not support the RTR state */
+	if ((attr_mask & IB_QP_STATE) && (attr->qp_state == IB_QPS_RTR))
+		attr_mask &= ~IB_QP_STATE;
+
+	/* Make sure we still have something left to do */
+	if (!attr_mask)
+		return 0;
+
+	memset(&attrs, 0, sizeof attrs);
+	qhp = to_c4iw_qp(ibqp);
+	rhp = qhp->rhp;
+
+	attrs.next_state = c4iw_convert_state(attr->qp_state);
+	attrs.enable_rdma_read = (attr->qp_access_flags &
+			       IB_ACCESS_REMOTE_READ) ?  1 : 0;
+	attrs.enable_rdma_write = (attr->qp_access_flags &
+				IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+	attrs.enable_bind = (attr->qp_access_flags & IB_ACCESS_MW_BIND) ? 1 : 0;
+
+
+	mask |= (attr_mask & IB_QP_STATE) ? C4IW_QP_ATTR_NEXT_STATE : 0;
+	mask |= (attr_mask & IB_QP_ACCESS_FLAGS) ?
+			(C4IW_QP_ATTR_ENABLE_RDMA_READ |
+			 C4IW_QP_ATTR_ENABLE_RDMA_WRITE |
+			 C4IW_QP_ATTR_ENABLE_RDMA_BIND) : 0;
+
+	/*
+	 * Use SQ_PSN and RQ_PSN to pass in IDX_INC values for
+	 * ringing the queue db when we're in DB_FULL mode.
+	 * Only allow this on T4 devices.
+	 */
+	attrs.sq_db_inc = attr->sq_psn;
+	attrs.rq_db_inc = attr->rq_psn;
+	mask |= (attr_mask & IB_QP_SQ_PSN) ? C4IW_QP_ATTR_SQ_DB : 0;
+	mask |= (attr_mask & IB_QP_RQ_PSN) ? C4IW_QP_ATTR_RQ_DB : 0;
+	if (is_t5(to_c4iw_qp(ibqp)->rhp->rdev.lldi.adapter_type) &&
+	    (mask & (C4IW_QP_ATTR_SQ_DB|C4IW_QP_ATTR_RQ_DB)))
+		return -EINVAL;
+
+	return c4iw_modify_qp(rhp, qhp, mask, &attrs, 0);
+}
+
+struct ib_qp *c4iw_get_qp(struct ib_device *dev, int qpn)
+{
+	PDBG("%s ib_dev %p qpn 0x%x\n", __func__, dev, qpn);
+	return (struct ib_qp *)get_qhp(to_c4iw_dev(dev), qpn);
+}
+
+int c4iw_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+	struct c4iw_qp *qhp = to_c4iw_qp(ibqp);
+
+	memset(attr, 0, sizeof *attr);
+	memset(init_attr, 0, sizeof *init_attr);
+	attr->qp_state = to_ib_qp_state(qhp->attr.state);
+	return 0;
+}
diff --git a/drivers/infiniband/hw/cxgb4/resource.c b/drivers/infiniband/hw/cxgb4/resource.c
new file mode 100644
index 00000000000..67df71a7012
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/resource.c
@@ -0,0 +1,453 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/* Crude resource management */
+#include <linux/spinlock.h>
+#include <linux/genalloc.h>
+#include <linux/ratelimit.h>
+#include "iw_cxgb4.h"
+
+static int c4iw_init_qid_table(struct c4iw_rdev *rdev)
+{
+	u32 i;
+
+	if (c4iw_id_table_alloc(&rdev->resource.qid_table,
+				rdev->lldi.vr->qp.start,
+				rdev->lldi.vr->qp.size,
+				rdev->lldi.vr->qp.size, 0))
+		return -ENOMEM;
+
+	for (i = rdev->lldi.vr->qp.start;
+		i < rdev->lldi.vr->qp.start + rdev->lldi.vr->qp.size; i++)
+		if (!(i & rdev->qpmask))
+			c4iw_id_free(&rdev->resource.qid_table, i);
+	return 0;
+}
+
+/* nr_* must be power of 2 */
+int c4iw_init_resource(struct c4iw_rdev *rdev, u32 nr_tpt, u32 nr_pdid)
+{
+	int err = 0;
+	err = c4iw_id_table_alloc(&rdev->resource.tpt_table, 0, nr_tpt, 1,
+					C4IW_ID_TABLE_F_RANDOM);
+	if (err)
+		goto tpt_err;
+	err = c4iw_init_qid_table(rdev);
+	if (err)
+		goto qid_err;
+	err = c4iw_id_table_alloc(&rdev->resource.pdid_table, 0,
+					nr_pdid, 1, 0);
+	if (err)
+		goto pdid_err;
+	return 0;
+ pdid_err:
+	c4iw_id_table_free(&rdev->resource.qid_table);
+ qid_err:
+	c4iw_id_table_free(&rdev->resource.tpt_table);
+ tpt_err:
+	return -ENOMEM;
+}
+
+/*
+ * returns 0 if no resource available
+ */
+u32 c4iw_get_resource(struct c4iw_id_table *id_table)
+{
+	u32 entry;
+	entry = c4iw_id_alloc(id_table);
+	if (entry == (u32)(-1))
+		return 0;
+	return entry;
+}
+
+void c4iw_put_resource(struct c4iw_id_table *id_table, u32 entry)
+{
+	PDBG("%s entry 0x%x\n", __func__, entry);
+	c4iw_id_free(id_table, entry);
+}
+
+u32 c4iw_get_cqid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
+{
+	struct c4iw_qid_list *entry;
+	u32 qid;
+	int i;
+
+	mutex_lock(&uctx->lock);
+	if (!list_empty(&uctx->cqids)) {
+		entry = list_entry(uctx->cqids.next, struct c4iw_qid_list,
+				   entry);
+		list_del(&entry->entry);
+		qid = entry->qid;
+		kfree(entry);
+	} else {
+		qid = c4iw_get_resource(&rdev->resource.qid_table);
+		if (!qid)
+			goto out;
+		mutex_lock(&rdev->stats.lock);
+		rdev->stats.qid.cur += rdev->qpmask + 1;
+		mutex_unlock(&rdev->stats.lock);
+		for (i = qid+1; i & rdev->qpmask; i++) {
+			entry = kmalloc(sizeof *entry, GFP_KERNEL);
+			if (!entry)
+				goto out;
+			entry->qid = i;
+			list_add_tail(&entry->entry, &uctx->cqids);
+		}
+
+		/*
+		 * now put the same ids on the qp list since they all
+		 * map to the same db/gts page.
+		 */
+		entry = kmalloc(sizeof *entry, GFP_KERNEL);
+		if (!entry)
+			goto out;
+		entry->qid = qid;
+		list_add_tail(&entry->entry, &uctx->qpids);
+		for (i = qid+1; i & rdev->qpmask; i++) {
+			entry = kmalloc(sizeof *entry, GFP_KERNEL);
+			if (!entry)
+				goto out;
+			entry->qid = i;
+			list_add_tail(&entry->entry, &uctx->qpids);
+		}
+	}
+out:
+	mutex_unlock(&uctx->lock);
+	PDBG("%s qid 0x%x\n", __func__, qid);
+	mutex_lock(&rdev->stats.lock);
+	if (rdev->stats.qid.cur > rdev->stats.qid.max)
+		rdev->stats.qid.max = rdev->stats.qid.cur;
+	mutex_unlock(&rdev->stats.lock);
+	return qid;
+}
+
+void c4iw_put_cqid(struct c4iw_rdev *rdev, u32 qid,
+		   struct c4iw_dev_ucontext *uctx)
+{
+	struct c4iw_qid_list *entry;
+
+	entry = kmalloc(sizeof *entry, GFP_KERNEL);
+	if (!entry)
+		return;
+	PDBG("%s qid 0x%x\n", __func__, qid);
+	entry->qid = qid;
+	mutex_lock(&uctx->lock);
+	list_add_tail(&entry->entry, &uctx->cqids);
+	mutex_unlock(&uctx->lock);
+}
+
+u32 c4iw_get_qpid(struct c4iw_rdev *rdev, struct c4iw_dev_ucontext *uctx)
+{
+	struct c4iw_qid_list *entry;
+	u32 qid;
+	int i;
+
+	mutex_lock(&uctx->lock);
+	if (!list_empty(&uctx->qpids)) {
+		entry = list_entry(uctx->qpids.next, struct c4iw_qid_list,
+				   entry);
+		list_del(&entry->entry);
+		qid = entry->qid;
+		kfree(entry);
+	} else {
+		qid = c4iw_get_resource(&rdev->resource.qid_table);
+		if (!qid) {
+			mutex_lock(&rdev->stats.lock);
+			rdev->stats.qid.fail++;
+			mutex_unlock(&rdev->stats.lock);
+			goto out;
+		}
+		mutex_lock(&rdev->stats.lock);
+		rdev->stats.qid.cur += rdev->qpmask + 1;
+		mutex_unlock(&rdev->stats.lock);
+		for (i = qid+1; i & rdev->qpmask; i++) {
+			entry = kmalloc(sizeof *entry, GFP_KERNEL);
+			if (!entry)
+				goto out;
+			entry->qid = i;
+			list_add_tail(&entry->entry, &uctx->qpids);
+		}
+
+		/*
+		 * now put the same ids on the cq list since they all
+		 * map to the same db/gts page.
+		 */
+		entry = kmalloc(sizeof *entry, GFP_KERNEL);
+		if (!entry)
+			goto out;
+		entry->qid = qid;
+		list_add_tail(&entry->entry, &uctx->cqids);
+		for (i = qid; i & rdev->qpmask; i++) {
+			entry = kmalloc(sizeof *entry, GFP_KERNEL);
+			if (!entry)
+				goto out;
+			entry->qid = i;
+			list_add_tail(&entry->entry, &uctx->cqids);
+		}
+	}
+out:
+	mutex_unlock(&uctx->lock);
+	PDBG("%s qid 0x%x\n", __func__, qid);
+	mutex_lock(&rdev->stats.lock);
+	if (rdev->stats.qid.cur > rdev->stats.qid.max)
+		rdev->stats.qid.max = rdev->stats.qid.cur;
+	mutex_unlock(&rdev->stats.lock);
+	return qid;
+}
+
+void c4iw_put_qpid(struct c4iw_rdev *rdev, u32 qid,
+		   struct c4iw_dev_ucontext *uctx)
+{
+	struct c4iw_qid_list *entry;
+
+	entry = kmalloc(sizeof *entry, GFP_KERNEL);
+	if (!entry)
+		return;
+	PDBG("%s qid 0x%x\n", __func__, qid);
+	entry->qid = qid;
+	mutex_lock(&uctx->lock);
+	list_add_tail(&entry->entry, &uctx->qpids);
+	mutex_unlock(&uctx->lock);
+}
+
+void c4iw_destroy_resource(struct c4iw_resource *rscp)
+{
+	c4iw_id_table_free(&rscp->tpt_table);
+	c4iw_id_table_free(&rscp->qid_table);
+	c4iw_id_table_free(&rscp->pdid_table);
+}
+
+/*
+ * PBL Memory Manager.  Uses Linux generic allocator.
+ */
+
+#define MIN_PBL_SHIFT 8			/* 256B == min PBL size (32 entries) */
+
+u32 c4iw_pblpool_alloc(struct c4iw_rdev *rdev, int size)
+{
+	unsigned long addr = gen_pool_alloc(rdev->pbl_pool, size);
+	PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size);
+	mutex_lock(&rdev->stats.lock);
+	if (addr) {
+		rdev->stats.pbl.cur += roundup(size, 1 << MIN_PBL_SHIFT);
+		if (rdev->stats.pbl.cur > rdev->stats.pbl.max)
+			rdev->stats.pbl.max = rdev->stats.pbl.cur;
+	} else
+		rdev->stats.pbl.fail++;
+	mutex_unlock(&rdev->stats.lock);
+	return (u32)addr;
+}
+
+void c4iw_pblpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+	PDBG("%s addr 0x%x size %d\n", __func__, addr, size);
+	mutex_lock(&rdev->stats.lock);
+	rdev->stats.pbl.cur -= roundup(size, 1 << MIN_PBL_SHIFT);
+	mutex_unlock(&rdev->stats.lock);
+	gen_pool_free(rdev->pbl_pool, (unsigned long)addr, size);
+}
+
+int c4iw_pblpool_create(struct c4iw_rdev *rdev)
+{
+	unsigned pbl_start, pbl_chunk, pbl_top;
+
+	rdev->pbl_pool = gen_pool_create(MIN_PBL_SHIFT, -1);
+	if (!rdev->pbl_pool)
+		return -ENOMEM;
+
+	pbl_start = rdev->lldi.vr->pbl.start;
+	pbl_chunk = rdev->lldi.vr->pbl.size;
+	pbl_top = pbl_start + pbl_chunk;
+
+	while (pbl_start < pbl_top) {
+		pbl_chunk = min(pbl_top - pbl_start + 1, pbl_chunk);
+		if (gen_pool_add(rdev->pbl_pool, pbl_start, pbl_chunk, -1)) {
+			PDBG("%s failed to add PBL chunk (%x/%x)\n",
+			     __func__, pbl_start, pbl_chunk);
+			if (pbl_chunk <= 1024 << MIN_PBL_SHIFT) {
+				printk(KERN_WARNING MOD
+				       "Failed to add all PBL chunks (%x/%x)\n",
+				       pbl_start,
+				       pbl_top - pbl_start);
+				return 0;
+			}
+			pbl_chunk >>= 1;
+		} else {
+			PDBG("%s added PBL chunk (%x/%x)\n",
+			     __func__, pbl_start, pbl_chunk);
+			pbl_start += pbl_chunk;
+		}
+	}
+
+	return 0;
+}
+
+void c4iw_pblpool_destroy(struct c4iw_rdev *rdev)
+{
+	gen_pool_destroy(rdev->pbl_pool);
+}
+
+/*
+ * RQT Memory Manager.  Uses Linux generic allocator.
+ */
+
+#define MIN_RQT_SHIFT 10	/* 1KB == min RQT size (16 entries) */
+
+u32 c4iw_rqtpool_alloc(struct c4iw_rdev *rdev, int size)
+{
+	unsigned long addr = gen_pool_alloc(rdev->rqt_pool, size << 6);
+	PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size << 6);
+	if (!addr)
+		pr_warn_ratelimited(MOD "%s: Out of RQT memory\n",
+				    pci_name(rdev->lldi.pdev));
+	mutex_lock(&rdev->stats.lock);
+	if (addr) {
+		rdev->stats.rqt.cur += roundup(size << 6, 1 << MIN_RQT_SHIFT);
+		if (rdev->stats.rqt.cur > rdev->stats.rqt.max)
+			rdev->stats.rqt.max = rdev->stats.rqt.cur;
+	} else
+		rdev->stats.rqt.fail++;
+	mutex_unlock(&rdev->stats.lock);
+	return (u32)addr;
+}
+
+void c4iw_rqtpool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+	PDBG("%s addr 0x%x size %d\n", __func__, addr, size << 6);
+	mutex_lock(&rdev->stats.lock);
+	rdev->stats.rqt.cur -= roundup(size << 6, 1 << MIN_RQT_SHIFT);
+	mutex_unlock(&rdev->stats.lock);
+	gen_pool_free(rdev->rqt_pool, (unsigned long)addr, size << 6);
+}
+
+int c4iw_rqtpool_create(struct c4iw_rdev *rdev)
+{
+	unsigned rqt_start, rqt_chunk, rqt_top;
+
+	rdev->rqt_pool = gen_pool_create(MIN_RQT_SHIFT, -1);
+	if (!rdev->rqt_pool)
+		return -ENOMEM;
+
+	rqt_start = rdev->lldi.vr->rq.start;
+	rqt_chunk = rdev->lldi.vr->rq.size;
+	rqt_top = rqt_start + rqt_chunk;
+
+	while (rqt_start < rqt_top) {
+		rqt_chunk = min(rqt_top - rqt_start + 1, rqt_chunk);
+		if (gen_pool_add(rdev->rqt_pool, rqt_start, rqt_chunk, -1)) {
+			PDBG("%s failed to add RQT chunk (%x/%x)\n",
+			     __func__, rqt_start, rqt_chunk);
+			if (rqt_chunk <= 1024 << MIN_RQT_SHIFT) {
+				printk(KERN_WARNING MOD
+				       "Failed to add all RQT chunks (%x/%x)\n",
+				       rqt_start, rqt_top - rqt_start);
+				return 0;
+			}
+			rqt_chunk >>= 1;
+		} else {
+			PDBG("%s added RQT chunk (%x/%x)\n",
+			     __func__, rqt_start, rqt_chunk);
+			rqt_start += rqt_chunk;
+		}
+	}
+	return 0;
+}
+
+void c4iw_rqtpool_destroy(struct c4iw_rdev *rdev)
+{
+	gen_pool_destroy(rdev->rqt_pool);
+}
+
+/*
+ * On-Chip QP Memory.
+ */
+#define MIN_OCQP_SHIFT 12	/* 4KB == min ocqp size */
+
+u32 c4iw_ocqp_pool_alloc(struct c4iw_rdev *rdev, int size)
+{
+	unsigned long addr = gen_pool_alloc(rdev->ocqp_pool, size);
+	PDBG("%s addr 0x%x size %d\n", __func__, (u32)addr, size);
+	if (addr) {
+		mutex_lock(&rdev->stats.lock);
+		rdev->stats.ocqp.cur += roundup(size, 1 << MIN_OCQP_SHIFT);
+		if (rdev->stats.ocqp.cur > rdev->stats.ocqp.max)
+			rdev->stats.ocqp.max = rdev->stats.ocqp.cur;
+		mutex_unlock(&rdev->stats.lock);
+	}
+	return (u32)addr;
+}
+
+void c4iw_ocqp_pool_free(struct c4iw_rdev *rdev, u32 addr, int size)
+{
+	PDBG("%s addr 0x%x size %d\n", __func__, addr, size);
+	mutex_lock(&rdev->stats.lock);
+	rdev->stats.ocqp.cur -= roundup(size, 1 << MIN_OCQP_SHIFT);
+	mutex_unlock(&rdev->stats.lock);
+	gen_pool_free(rdev->ocqp_pool, (unsigned long)addr, size);
+}
+
+int c4iw_ocqp_pool_create(struct c4iw_rdev *rdev)
+{
+	unsigned start, chunk, top;
+
+	rdev->ocqp_pool = gen_pool_create(MIN_OCQP_SHIFT, -1);
+	if (!rdev->ocqp_pool)
+		return -ENOMEM;
+
+	start = rdev->lldi.vr->ocq.start;
+	chunk = rdev->lldi.vr->ocq.size;
+	top = start + chunk;
+
+	while (start < top) {
+		chunk = min(top - start + 1, chunk);
+		if (gen_pool_add(rdev->ocqp_pool, start, chunk, -1)) {
+			PDBG("%s failed to add OCQP chunk (%x/%x)\n",
+			     __func__, start, chunk);
+			if (chunk <= 1024 << MIN_OCQP_SHIFT) {
+				printk(KERN_WARNING MOD
+				       "Failed to add all OCQP chunks (%x/%x)\n",
+				       start, top - start);
+				return 0;
+			}
+			chunk >>= 1;
+		} else {
+			PDBG("%s added OCQP chunk (%x/%x)\n",
+			     __func__, start, chunk);
+			start += chunk;
+		}
+	}
+	return 0;
+}
+
+void c4iw_ocqp_pool_destroy(struct c4iw_rdev *rdev)
+{
+	gen_pool_destroy(rdev->ocqp_pool);
+}
diff --git a/drivers/infiniband/hw/cxgb4/t4.h b/drivers/infiniband/hw/cxgb4/t4.h
new file mode 100644
index 00000000000..68b0a6bf4eb
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/t4.h
@@ -0,0 +1,680 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __T4_H__
+#define __T4_H__
+
+#include "t4_hw.h"
+#include "t4_regs.h"
+#include "t4_msg.h"
+#include "t4fw_ri_api.h"
+
+#define T4_MAX_NUM_QP 65536
+#define T4_MAX_NUM_CQ 65536
+#define T4_MAX_NUM_PD 65536
+#define T4_EQ_STATUS_ENTRIES (L1_CACHE_BYTES > 64 ? 2 : 1)
+#define T4_MAX_EQ_SIZE (65520 - T4_EQ_STATUS_ENTRIES)
+#define T4_MAX_IQ_SIZE (65520 - 1)
+#define T4_MAX_RQ_SIZE (8192 - T4_EQ_STATUS_ENTRIES)
+#define T4_MAX_SQ_SIZE (T4_MAX_EQ_SIZE - 1)
+#define T4_MAX_QP_DEPTH (T4_MAX_RQ_SIZE - 1)
+#define T4_MAX_CQ_DEPTH (T4_MAX_IQ_SIZE - 1)
+#define T4_MAX_NUM_STAG (1<<15)
+#define T4_MAX_MR_SIZE (~0ULL)
+#define T4_PAGESIZE_MASK 0xffff000  /* 4KB-128MB */
+#define T4_STAG_UNSET 0xffffffff
+#define T4_FW_MAJ 0
+#define T4_EQ_STATUS_ENTRIES (L1_CACHE_BYTES > 64 ? 2 : 1)
+#define A_PCIE_MA_SYNC 0x30b4
+
+struct t4_status_page {
+	__be32 rsvd1;	/* flit 0 - hw owns */
+	__be16 rsvd2;
+	__be16 qid;
+	__be16 cidx;
+	__be16 pidx;
+	u8 qp_err;	/* flit 1 - sw owns */
+	u8 db_off;
+	u8 pad;
+	u16 host_wq_pidx;
+	u16 host_cidx;
+	u16 host_pidx;
+};
+
+#define T4_EQ_ENTRY_SIZE 64
+
+#define T4_SQ_NUM_SLOTS 5
+#define T4_SQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_SQ_NUM_SLOTS)
+#define T4_MAX_SEND_SGE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
+			sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
+#define T4_MAX_SEND_INLINE ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_send_wr) - \
+			sizeof(struct fw_ri_immd)))
+#define T4_MAX_WRITE_INLINE ((T4_SQ_NUM_BYTES - \
+			sizeof(struct fw_ri_rdma_write_wr) - \
+			sizeof(struct fw_ri_immd)))
+#define T4_MAX_WRITE_SGE ((T4_SQ_NUM_BYTES - \
+			sizeof(struct fw_ri_rdma_write_wr) - \
+			sizeof(struct fw_ri_isgl)) / sizeof(struct fw_ri_sge))
+#define T4_MAX_FR_IMMD ((T4_SQ_NUM_BYTES - sizeof(struct fw_ri_fr_nsmr_wr) - \
+			sizeof(struct fw_ri_immd)) & ~31UL)
+#define T4_MAX_FR_IMMD_DEPTH (T4_MAX_FR_IMMD / sizeof(u64))
+#define T4_MAX_FR_DSGL 1024
+#define T4_MAX_FR_DSGL_DEPTH (T4_MAX_FR_DSGL / sizeof(u64))
+
+static inline int t4_max_fr_depth(int use_dsgl)
+{
+	return use_dsgl ? T4_MAX_FR_DSGL_DEPTH : T4_MAX_FR_IMMD_DEPTH;
+}
+
+#define T4_RQ_NUM_SLOTS 2
+#define T4_RQ_NUM_BYTES (T4_EQ_ENTRY_SIZE * T4_RQ_NUM_SLOTS)
+#define T4_MAX_RECV_SGE 4
+
+union t4_wr {
+	struct fw_ri_res_wr res;
+	struct fw_ri_wr ri;
+	struct fw_ri_rdma_write_wr write;
+	struct fw_ri_send_wr send;
+	struct fw_ri_rdma_read_wr read;
+	struct fw_ri_bind_mw_wr bind;
+	struct fw_ri_fr_nsmr_wr fr;
+	struct fw_ri_inv_lstag_wr inv;
+	struct t4_status_page status;
+	__be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_SQ_NUM_SLOTS];
+};
+
+union t4_recv_wr {
+	struct fw_ri_recv_wr recv;
+	struct t4_status_page status;
+	__be64 flits[T4_EQ_ENTRY_SIZE / sizeof(__be64) * T4_RQ_NUM_SLOTS];
+};
+
+static inline void init_wr_hdr(union t4_wr *wqe, u16 wrid,
+			       enum fw_wr_opcodes opcode, u8 flags, u8 len16)
+{
+	wqe->send.opcode = (u8)opcode;
+	wqe->send.flags = flags;
+	wqe->send.wrid = wrid;
+	wqe->send.r1[0] = 0;
+	wqe->send.r1[1] = 0;
+	wqe->send.r1[2] = 0;
+	wqe->send.len16 = len16;
+}
+
+/* CQE/AE status codes */
+#define T4_ERR_SUCCESS                     0x0
+#define T4_ERR_STAG                        0x1	/* STAG invalid: either the */
+						/* STAG is offlimt, being 0, */
+						/* or STAG_key mismatch */
+#define T4_ERR_PDID                        0x2	/* PDID mismatch */
+#define T4_ERR_QPID                        0x3	/* QPID mismatch */
+#define T4_ERR_ACCESS                      0x4	/* Invalid access right */
+#define T4_ERR_WRAP                        0x5	/* Wrap error */
+#define T4_ERR_BOUND                       0x6	/* base and bounds voilation */
+#define T4_ERR_INVALIDATE_SHARED_MR        0x7	/* attempt to invalidate a  */
+						/* shared memory region */
+#define T4_ERR_INVALIDATE_MR_WITH_MW_BOUND 0x8	/* attempt to invalidate a  */
+						/* shared memory region */
+#define T4_ERR_ECC                         0x9	/* ECC error detected */
+#define T4_ERR_ECC_PSTAG                   0xA	/* ECC error detected when  */
+						/* reading PSTAG for a MW  */
+						/* Invalidate */
+#define T4_ERR_PBL_ADDR_BOUND              0xB	/* pbl addr out of bounds:  */
+						/* software error */
+#define T4_ERR_SWFLUSH			   0xC	/* SW FLUSHED */
+#define T4_ERR_CRC                         0x10 /* CRC error */
+#define T4_ERR_MARKER                      0x11 /* Marker error */
+#define T4_ERR_PDU_LEN_ERR                 0x12 /* invalid PDU length */
+#define T4_ERR_OUT_OF_RQE                  0x13 /* out of RQE */
+#define T4_ERR_DDP_VERSION                 0x14 /* wrong DDP version */
+#define T4_ERR_RDMA_VERSION                0x15 /* wrong RDMA version */
+#define T4_ERR_OPCODE                      0x16 /* invalid rdma opcode */
+#define T4_ERR_DDP_QUEUE_NUM               0x17 /* invalid ddp queue number */
+#define T4_ERR_MSN                         0x18 /* MSN error */
+#define T4_ERR_TBIT                        0x19 /* tag bit not set correctly */
+#define T4_ERR_MO                          0x1A /* MO not 0 for TERMINATE  */
+						/* or READ_REQ */
+#define T4_ERR_MSN_GAP                     0x1B
+#define T4_ERR_MSN_RANGE                   0x1C
+#define T4_ERR_IRD_OVERFLOW                0x1D
+#define T4_ERR_RQE_ADDR_BOUND              0x1E /* RQE addr out of bounds:  */
+						/* software error */
+#define T4_ERR_INTERNAL_ERR                0x1F /* internal error (opcode  */
+						/* mismatch) */
+/*
+ * CQE defs
+ */
+struct t4_cqe {
+	__be32 header;
+	__be32 len;
+	union {
+		struct {
+			__be32 stag;
+			__be32 msn;
+		} rcqe;
+		struct {
+			u32 nada1;
+			u16 nada2;
+			u16 cidx;
+		} scqe;
+		struct {
+			__be32 wrid_hi;
+			__be32 wrid_low;
+		} gen;
+	} u;
+	__be64 reserved;
+	__be64 bits_type_ts;
+};
+
+/* macros for flit 0 of the cqe */
+
+#define S_CQE_QPID        12
+#define M_CQE_QPID        0xFFFFF
+#define G_CQE_QPID(x)     ((((x) >> S_CQE_QPID)) & M_CQE_QPID)
+#define V_CQE_QPID(x)	  ((x)<<S_CQE_QPID)
+
+#define S_CQE_SWCQE       11
+#define M_CQE_SWCQE       0x1
+#define G_CQE_SWCQE(x)    ((((x) >> S_CQE_SWCQE)) & M_CQE_SWCQE)
+#define V_CQE_SWCQE(x)	  ((x)<<S_CQE_SWCQE)
+
+#define S_CQE_STATUS      5
+#define M_CQE_STATUS      0x1F
+#define G_CQE_STATUS(x)   ((((x) >> S_CQE_STATUS)) & M_CQE_STATUS)
+#define V_CQE_STATUS(x)   ((x)<<S_CQE_STATUS)
+
+#define S_CQE_TYPE        4
+#define M_CQE_TYPE        0x1
+#define G_CQE_TYPE(x)     ((((x) >> S_CQE_TYPE)) & M_CQE_TYPE)
+#define V_CQE_TYPE(x)     ((x)<<S_CQE_TYPE)
+
+#define S_CQE_OPCODE      0
+#define M_CQE_OPCODE      0xF
+#define G_CQE_OPCODE(x)   ((((x) >> S_CQE_OPCODE)) & M_CQE_OPCODE)
+#define V_CQE_OPCODE(x)   ((x)<<S_CQE_OPCODE)
+
+#define SW_CQE(x)         (G_CQE_SWCQE(be32_to_cpu((x)->header)))
+#define CQE_QPID(x)       (G_CQE_QPID(be32_to_cpu((x)->header)))
+#define CQE_TYPE(x)       (G_CQE_TYPE(be32_to_cpu((x)->header)))
+#define SQ_TYPE(x)	  (CQE_TYPE((x)))
+#define RQ_TYPE(x)	  (!CQE_TYPE((x)))
+#define CQE_STATUS(x)     (G_CQE_STATUS(be32_to_cpu((x)->header)))
+#define CQE_OPCODE(x)     (G_CQE_OPCODE(be32_to_cpu((x)->header)))
+
+#define CQE_SEND_OPCODE(x)( \
+	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND) || \
+	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE) || \
+	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_INV) || \
+	(G_CQE_OPCODE(be32_to_cpu((x)->header)) == FW_RI_SEND_WITH_SE_INV))
+
+#define CQE_LEN(x)        (be32_to_cpu((x)->len))
+
+/* used for RQ completion processing */
+#define CQE_WRID_STAG(x)  (be32_to_cpu((x)->u.rcqe.stag))
+#define CQE_WRID_MSN(x)   (be32_to_cpu((x)->u.rcqe.msn))
+
+/* used for SQ completion processing */
+#define CQE_WRID_SQ_IDX(x)	((x)->u.scqe.cidx)
+
+/* generic accessor macros */
+#define CQE_WRID_HI(x)		((x)->u.gen.wrid_hi)
+#define CQE_WRID_LOW(x)		((x)->u.gen.wrid_low)
+
+/* macros for flit 3 of the cqe */
+#define S_CQE_GENBIT	63
+#define M_CQE_GENBIT	0x1
+#define G_CQE_GENBIT(x)	(((x) >> S_CQE_GENBIT) & M_CQE_GENBIT)
+#define V_CQE_GENBIT(x) ((x)<<S_CQE_GENBIT)
+
+#define S_CQE_OVFBIT	62
+#define M_CQE_OVFBIT	0x1
+#define G_CQE_OVFBIT(x)	((((x) >> S_CQE_OVFBIT)) & M_CQE_OVFBIT)
+
+#define S_CQE_IQTYPE	60
+#define M_CQE_IQTYPE	0x3
+#define G_CQE_IQTYPE(x)	((((x) >> S_CQE_IQTYPE)) & M_CQE_IQTYPE)
+
+#define M_CQE_TS	0x0fffffffffffffffULL
+#define G_CQE_TS(x)	((x) & M_CQE_TS)
+
+#define CQE_OVFBIT(x)	((unsigned)G_CQE_OVFBIT(be64_to_cpu((x)->bits_type_ts)))
+#define CQE_GENBIT(x)	((unsigned)G_CQE_GENBIT(be64_to_cpu((x)->bits_type_ts)))
+#define CQE_TS(x)	(G_CQE_TS(be64_to_cpu((x)->bits_type_ts)))
+
+struct t4_swsqe {
+	u64			wr_id;
+	struct t4_cqe		cqe;
+	int			read_len;
+	int			opcode;
+	int			complete;
+	int			signaled;
+	u16			idx;
+	int                     flushed;
+};
+
+static inline pgprot_t t4_pgprot_wc(pgprot_t prot)
+{
+#if defined(__i386__) || defined(__x86_64__) || defined(CONFIG_PPC64)
+	return pgprot_writecombine(prot);
+#else
+	return pgprot_noncached(prot);
+#endif
+}
+
+enum {
+	T4_SQ_ONCHIP = (1<<0),
+};
+
+struct t4_sq {
+	union t4_wr *queue;
+	dma_addr_t dma_addr;
+	DEFINE_DMA_UNMAP_ADDR(mapping);
+	unsigned long phys_addr;
+	struct t4_swsqe *sw_sq;
+	struct t4_swsqe *oldest_read;
+	u64 __iomem *udb;
+	size_t memsize;
+	u32 qid;
+	u16 in_use;
+	u16 size;
+	u16 cidx;
+	u16 pidx;
+	u16 wq_pidx;
+	u16 wq_pidx_inc;
+	u16 flags;
+	short flush_cidx;
+};
+
+struct t4_swrqe {
+	u64 wr_id;
+};
+
+struct t4_rq {
+	union  t4_recv_wr *queue;
+	dma_addr_t dma_addr;
+	DEFINE_DMA_UNMAP_ADDR(mapping);
+	struct t4_swrqe *sw_rq;
+	u64 __iomem *udb;
+	size_t memsize;
+	u32 qid;
+	u32 msn;
+	u32 rqt_hwaddr;
+	u16 rqt_size;
+	u16 in_use;
+	u16 size;
+	u16 cidx;
+	u16 pidx;
+	u16 wq_pidx;
+	u16 wq_pidx_inc;
+};
+
+struct t4_wq {
+	struct t4_sq sq;
+	struct t4_rq rq;
+	void __iomem *db;
+	void __iomem *gts;
+	struct c4iw_rdev *rdev;
+	int flushed;
+};
+
+static inline int t4_rqes_posted(struct t4_wq *wq)
+{
+	return wq->rq.in_use;
+}
+
+static inline int t4_rq_empty(struct t4_wq *wq)
+{
+	return wq->rq.in_use == 0;
+}
+
+static inline int t4_rq_full(struct t4_wq *wq)
+{
+	return wq->rq.in_use == (wq->rq.size - 1);
+}
+
+static inline u32 t4_rq_avail(struct t4_wq *wq)
+{
+	return wq->rq.size - 1 - wq->rq.in_use;
+}
+
+static inline void t4_rq_produce(struct t4_wq *wq, u8 len16)
+{
+	wq->rq.in_use++;
+	if (++wq->rq.pidx == wq->rq.size)
+		wq->rq.pidx = 0;
+	wq->rq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+	if (wq->rq.wq_pidx >= wq->rq.size * T4_RQ_NUM_SLOTS)
+		wq->rq.wq_pidx %= wq->rq.size * T4_RQ_NUM_SLOTS;
+}
+
+static inline void t4_rq_consume(struct t4_wq *wq)
+{
+	wq->rq.in_use--;
+	wq->rq.msn++;
+	if (++wq->rq.cidx == wq->rq.size)
+		wq->rq.cidx = 0;
+}
+
+static inline u16 t4_rq_host_wq_pidx(struct t4_wq *wq)
+{
+	return wq->rq.queue[wq->rq.size].status.host_wq_pidx;
+}
+
+static inline u16 t4_rq_wq_size(struct t4_wq *wq)
+{
+		return wq->rq.size * T4_RQ_NUM_SLOTS;
+}
+
+static inline int t4_sq_onchip(struct t4_sq *sq)
+{
+	return sq->flags & T4_SQ_ONCHIP;
+}
+
+static inline int t4_sq_empty(struct t4_wq *wq)
+{
+	return wq->sq.in_use == 0;
+}
+
+static inline int t4_sq_full(struct t4_wq *wq)
+{
+	return wq->sq.in_use == (wq->sq.size - 1);
+}
+
+static inline u32 t4_sq_avail(struct t4_wq *wq)
+{
+	return wq->sq.size - 1 - wq->sq.in_use;
+}
+
+static inline void t4_sq_produce(struct t4_wq *wq, u8 len16)
+{
+	wq->sq.in_use++;
+	if (++wq->sq.pidx == wq->sq.size)
+		wq->sq.pidx = 0;
+	wq->sq.wq_pidx += DIV_ROUND_UP(len16*16, T4_EQ_ENTRY_SIZE);
+	if (wq->sq.wq_pidx >= wq->sq.size * T4_SQ_NUM_SLOTS)
+		wq->sq.wq_pidx %= wq->sq.size * T4_SQ_NUM_SLOTS;
+}
+
+static inline void t4_sq_consume(struct t4_wq *wq)
+{
+	BUG_ON(wq->sq.in_use < 1);
+	if (wq->sq.cidx == wq->sq.flush_cidx)
+		wq->sq.flush_cidx = -1;
+	wq->sq.in_use--;
+	if (++wq->sq.cidx == wq->sq.size)
+		wq->sq.cidx = 0;
+}
+
+static inline u16 t4_sq_host_wq_pidx(struct t4_wq *wq)
+{
+	return wq->sq.queue[wq->sq.size].status.host_wq_pidx;
+}
+
+static inline u16 t4_sq_wq_size(struct t4_wq *wq)
+{
+		return wq->sq.size * T4_SQ_NUM_SLOTS;
+}
+
+/* This function copies 64 byte coalesced work request to memory
+ * mapped BAR2 space. For coalesced WRs, the SGE fetches data
+ * from the FIFO instead of from Host.
+ */
+static inline void pio_copy(u64 __iomem *dst, u64 *src)
+{
+	int count = 8;
+
+	while (count) {
+		writeq(*src, dst);
+		src++;
+		dst++;
+		count--;
+	}
+}
+
+static inline void t4_ring_sq_db(struct t4_wq *wq, u16 inc, u8 t5,
+				 union t4_wr *wqe)
+{
+
+	/* Flush host queue memory writes. */
+	wmb();
+	if (t5) {
+		if (inc == 1 && wqe) {
+			PDBG("%s: WC wq->sq.pidx = %d\n",
+			     __func__, wq->sq.pidx);
+			pio_copy(wq->sq.udb + 7, (void *)wqe);
+		} else {
+			PDBG("%s: DB wq->sq.pidx = %d\n",
+			     __func__, wq->sq.pidx);
+			writel(PIDX_T5(inc), wq->sq.udb);
+		}
+
+		/* Flush user doorbell area writes. */
+		wmb();
+		return;
+	}
+	writel(QID(wq->sq.qid) | PIDX(inc), wq->db);
+}
+
+static inline void t4_ring_rq_db(struct t4_wq *wq, u16 inc, u8 t5,
+				 union t4_recv_wr *wqe)
+{
+
+	/* Flush host queue memory writes. */
+	wmb();
+	if (t5) {
+		if (inc == 1 && wqe) {
+			PDBG("%s: WC wq->rq.pidx = %d\n",
+			     __func__, wq->rq.pidx);
+			pio_copy(wq->rq.udb + 7, (void *)wqe);
+		} else {
+			PDBG("%s: DB wq->rq.pidx = %d\n",
+			     __func__, wq->rq.pidx);
+			writel(PIDX_T5(inc), wq->rq.udb);
+		}
+
+		/* Flush user doorbell area writes. */
+		wmb();
+		return;
+	}
+	writel(QID(wq->rq.qid) | PIDX(inc), wq->db);
+}
+
+static inline int t4_wq_in_error(struct t4_wq *wq)
+{
+	return wq->rq.queue[wq->rq.size].status.qp_err;
+}
+
+static inline void t4_set_wq_in_error(struct t4_wq *wq)
+{
+	wq->rq.queue[wq->rq.size].status.qp_err = 1;
+}
+
+static inline void t4_disable_wq_db(struct t4_wq *wq)
+{
+	wq->rq.queue[wq->rq.size].status.db_off = 1;
+}
+
+static inline void t4_enable_wq_db(struct t4_wq *wq)
+{
+	wq->rq.queue[wq->rq.size].status.db_off = 0;
+}
+
+static inline int t4_wq_db_enabled(struct t4_wq *wq)
+{
+	return !wq->rq.queue[wq->rq.size].status.db_off;
+}
+
+struct t4_cq {
+	struct t4_cqe *queue;
+	dma_addr_t dma_addr;
+	DEFINE_DMA_UNMAP_ADDR(mapping);
+	struct t4_cqe *sw_queue;
+	void __iomem *gts;
+	struct c4iw_rdev *rdev;
+	u64 ugts;
+	size_t memsize;
+	__be64 bits_type_ts;
+	u32 cqid;
+	int vector;
+	u16 size; /* including status page */
+	u16 cidx;
+	u16 sw_pidx;
+	u16 sw_cidx;
+	u16 sw_in_use;
+	u16 cidx_inc;
+	u8 gen;
+	u8 error;
+};
+
+static inline int t4_arm_cq(struct t4_cq *cq, int se)
+{
+	u32 val;
+
+	while (cq->cidx_inc > CIDXINC_MASK) {
+		val = SEINTARM(0) | CIDXINC(CIDXINC_MASK) | TIMERREG(7) |
+		      INGRESSQID(cq->cqid);
+		writel(val, cq->gts);
+		cq->cidx_inc -= CIDXINC_MASK;
+	}
+	val = SEINTARM(se) | CIDXINC(cq->cidx_inc) | TIMERREG(6) |
+	      INGRESSQID(cq->cqid);
+	writel(val, cq->gts);
+	cq->cidx_inc = 0;
+	return 0;
+}
+
+static inline void t4_swcq_produce(struct t4_cq *cq)
+{
+	cq->sw_in_use++;
+	if (cq->sw_in_use == cq->size) {
+		PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__, cq->cqid);
+		cq->error = 1;
+		BUG_ON(1);
+	}
+	if (++cq->sw_pidx == cq->size)
+		cq->sw_pidx = 0;
+}
+
+static inline void t4_swcq_consume(struct t4_cq *cq)
+{
+	BUG_ON(cq->sw_in_use < 1);
+	cq->sw_in_use--;
+	if (++cq->sw_cidx == cq->size)
+		cq->sw_cidx = 0;
+}
+
+static inline void t4_hwcq_consume(struct t4_cq *cq)
+{
+	cq->bits_type_ts = cq->queue[cq->cidx].bits_type_ts;
+	if (++cq->cidx_inc == (cq->size >> 4) || cq->cidx_inc == CIDXINC_MASK) {
+		u32 val;
+
+		val = SEINTARM(0) | CIDXINC(cq->cidx_inc) | TIMERREG(7) |
+		      INGRESSQID(cq->cqid);
+		writel(val, cq->gts);
+		cq->cidx_inc = 0;
+	}
+	if (++cq->cidx == cq->size) {
+		cq->cidx = 0;
+		cq->gen ^= 1;
+	}
+}
+
+static inline int t4_valid_cqe(struct t4_cq *cq, struct t4_cqe *cqe)
+{
+	return (CQE_GENBIT(cqe) == cq->gen);
+}
+
+static inline int t4_next_hw_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
+{
+	int ret;
+	u16 prev_cidx;
+
+	if (cq->cidx == 0)
+		prev_cidx = cq->size - 1;
+	else
+		prev_cidx = cq->cidx - 1;
+
+	if (cq->queue[prev_cidx].bits_type_ts != cq->bits_type_ts) {
+		ret = -EOVERFLOW;
+		cq->error = 1;
+		printk(KERN_ERR MOD "cq overflow cqid %u\n", cq->cqid);
+		BUG_ON(1);
+	} else if (t4_valid_cqe(cq, &cq->queue[cq->cidx])) {
+
+		/* Ensure CQE is flushed to memory */
+		rmb();
+		*cqe = &cq->queue[cq->cidx];
+		ret = 0;
+	} else
+		ret = -ENODATA;
+	return ret;
+}
+
+static inline struct t4_cqe *t4_next_sw_cqe(struct t4_cq *cq)
+{
+	if (cq->sw_in_use == cq->size) {
+		PDBG("%s cxgb4 sw cq overflow cqid %u\n", __func__, cq->cqid);
+		cq->error = 1;
+		BUG_ON(1);
+		return NULL;
+	}
+	if (cq->sw_in_use)
+		return &cq->sw_queue[cq->sw_cidx];
+	return NULL;
+}
+
+static inline int t4_next_cqe(struct t4_cq *cq, struct t4_cqe **cqe)
+{
+	int ret = 0;
+
+	if (cq->error)
+		ret = -ENODATA;
+	else if (cq->sw_in_use)
+		*cqe = &cq->sw_queue[cq->sw_cidx];
+	else
+		ret = t4_next_hw_cqe(cq, cqe);
+	return ret;
+}
+
+static inline int t4_cq_in_error(struct t4_cq *cq)
+{
+	return ((struct t4_status_page *)&cq->queue[cq->size])->qp_err;
+}
+
+static inline void t4_set_cq_in_error(struct t4_cq *cq)
+{
+	((struct t4_status_page *)&cq->queue[cq->size])->qp_err = 1;
+}
+#endif
+
+struct t4_dev_status_page {
+	u8 db_off;
+};
diff --git a/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
new file mode 100644
index 00000000000..91289a051af
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/t4fw_ri_api.h
@@ -0,0 +1,854 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _T4FW_RI_API_H_
+#define _T4FW_RI_API_H_
+
+#include "t4fw_api.h"
+
+enum fw_ri_wr_opcode {
+	FW_RI_RDMA_WRITE		= 0x0,	/* IETF RDMAP v1.0 ... */
+	FW_RI_READ_REQ			= 0x1,
+	FW_RI_READ_RESP			= 0x2,
+	FW_RI_SEND			= 0x3,
+	FW_RI_SEND_WITH_INV		= 0x4,
+	FW_RI_SEND_WITH_SE		= 0x5,
+	FW_RI_SEND_WITH_SE_INV		= 0x6,
+	FW_RI_TERMINATE			= 0x7,
+	FW_RI_RDMA_INIT			= 0x8,	/* CHELSIO RI specific ... */
+	FW_RI_BIND_MW			= 0x9,
+	FW_RI_FAST_REGISTER		= 0xa,
+	FW_RI_LOCAL_INV			= 0xb,
+	FW_RI_QP_MODIFY			= 0xc,
+	FW_RI_BYPASS			= 0xd,
+	FW_RI_RECEIVE			= 0xe,
+
+	FW_RI_SGE_EC_CR_RETURN		= 0xf
+};
+
+enum fw_ri_wr_flags {
+	FW_RI_COMPLETION_FLAG		= 0x01,
+	FW_RI_NOTIFICATION_FLAG		= 0x02,
+	FW_RI_SOLICITED_EVENT_FLAG	= 0x04,
+	FW_RI_READ_FENCE_FLAG		= 0x08,
+	FW_RI_LOCAL_FENCE_FLAG		= 0x10,
+	FW_RI_RDMA_READ_INVALIDATE	= 0x20
+};
+
+enum fw_ri_mpa_attrs {
+	FW_RI_MPA_RX_MARKER_ENABLE	= 0x01,
+	FW_RI_MPA_TX_MARKER_ENABLE	= 0x02,
+	FW_RI_MPA_CRC_ENABLE		= 0x04,
+	FW_RI_MPA_IETF_ENABLE		= 0x08
+};
+
+enum fw_ri_qp_caps {
+	FW_RI_QP_RDMA_READ_ENABLE	= 0x01,
+	FW_RI_QP_RDMA_WRITE_ENABLE	= 0x02,
+	FW_RI_QP_BIND_ENABLE		= 0x04,
+	FW_RI_QP_FAST_REGISTER_ENABLE	= 0x08,
+	FW_RI_QP_STAG0_ENABLE		= 0x10
+};
+
+enum fw_ri_addr_type {
+	FW_RI_ZERO_BASED_TO		= 0x00,
+	FW_RI_VA_BASED_TO		= 0x01
+};
+
+enum fw_ri_mem_perms {
+	FW_RI_MEM_ACCESS_REM_WRITE	= 0x01,
+	FW_RI_MEM_ACCESS_REM_READ	= 0x02,
+	FW_RI_MEM_ACCESS_REM		= 0x03,
+	FW_RI_MEM_ACCESS_LOCAL_WRITE	= 0x04,
+	FW_RI_MEM_ACCESS_LOCAL_READ	= 0x08,
+	FW_RI_MEM_ACCESS_LOCAL		= 0x0C
+};
+
+enum fw_ri_stag_type {
+	FW_RI_STAG_NSMR			= 0x00,
+	FW_RI_STAG_SMR			= 0x01,
+	FW_RI_STAG_MW			= 0x02,
+	FW_RI_STAG_MW_RELAXED		= 0x03
+};
+
+enum fw_ri_data_op {
+	FW_RI_DATA_IMMD			= 0x81,
+	FW_RI_DATA_DSGL			= 0x82,
+	FW_RI_DATA_ISGL			= 0x83
+};
+
+enum fw_ri_sgl_depth {
+	FW_RI_SGL_DEPTH_MAX_SQ		= 16,
+	FW_RI_SGL_DEPTH_MAX_RQ		= 4
+};
+
+struct fw_ri_dsge_pair {
+	__be32	len[2];
+	__be64	addr[2];
+};
+
+struct fw_ri_dsgl {
+	__u8	op;
+	__u8	r1;
+	__be16	nsge;
+	__be32	len0;
+	__be64	addr0;
+#ifndef C99_NOT_SUPPORTED
+	struct fw_ri_dsge_pair sge[0];
+#endif
+};
+
+struct fw_ri_sge {
+	__be32 stag;
+	__be32 len;
+	__be64 to;
+};
+
+struct fw_ri_isgl {
+	__u8	op;
+	__u8	r1;
+	__be16	nsge;
+	__be32	r2;
+#ifndef C99_NOT_SUPPORTED
+	struct fw_ri_sge sge[0];
+#endif
+};
+
+struct fw_ri_immd {
+	__u8	op;
+	__u8	r1;
+	__be16	r2;
+	__be32	immdlen;
+#ifndef C99_NOT_SUPPORTED
+	__u8	data[0];
+#endif
+};
+
+struct fw_ri_tpte {
+	__be32 valid_to_pdid;
+	__be32 locread_to_qpid;
+	__be32 nosnoop_pbladdr;
+	__be32 len_lo;
+	__be32 va_hi;
+	__be32 va_lo_fbo;
+	__be32 dca_mwbcnt_pstag;
+	__be32 len_hi;
+};
+
+#define S_FW_RI_TPTE_VALID		31
+#define M_FW_RI_TPTE_VALID		0x1
+#define V_FW_RI_TPTE_VALID(x)		((x) << S_FW_RI_TPTE_VALID)
+#define G_FW_RI_TPTE_VALID(x)		\
+    (((x) >> S_FW_RI_TPTE_VALID) & M_FW_RI_TPTE_VALID)
+#define F_FW_RI_TPTE_VALID		V_FW_RI_TPTE_VALID(1U)
+
+#define S_FW_RI_TPTE_STAGKEY		23
+#define M_FW_RI_TPTE_STAGKEY		0xff
+#define V_FW_RI_TPTE_STAGKEY(x)		((x) << S_FW_RI_TPTE_STAGKEY)
+#define G_FW_RI_TPTE_STAGKEY(x)		\
+    (((x) >> S_FW_RI_TPTE_STAGKEY) & M_FW_RI_TPTE_STAGKEY)
+
+#define S_FW_RI_TPTE_STAGSTATE		22
+#define M_FW_RI_TPTE_STAGSTATE		0x1
+#define V_FW_RI_TPTE_STAGSTATE(x)	((x) << S_FW_RI_TPTE_STAGSTATE)
+#define G_FW_RI_TPTE_STAGSTATE(x)	\
+    (((x) >> S_FW_RI_TPTE_STAGSTATE) & M_FW_RI_TPTE_STAGSTATE)
+#define F_FW_RI_TPTE_STAGSTATE		V_FW_RI_TPTE_STAGSTATE(1U)
+
+#define S_FW_RI_TPTE_STAGTYPE		20
+#define M_FW_RI_TPTE_STAGTYPE		0x3
+#define V_FW_RI_TPTE_STAGTYPE(x)	((x) << S_FW_RI_TPTE_STAGTYPE)
+#define G_FW_RI_TPTE_STAGTYPE(x)	\
+    (((x) >> S_FW_RI_TPTE_STAGTYPE) & M_FW_RI_TPTE_STAGTYPE)
+
+#define S_FW_RI_TPTE_PDID		0
+#define M_FW_RI_TPTE_PDID		0xfffff
+#define V_FW_RI_TPTE_PDID(x)		((x) << S_FW_RI_TPTE_PDID)
+#define G_FW_RI_TPTE_PDID(x)		\
+    (((x) >> S_FW_RI_TPTE_PDID) & M_FW_RI_TPTE_PDID)
+
+#define S_FW_RI_TPTE_PERM		28
+#define M_FW_RI_TPTE_PERM		0xf
+#define V_FW_RI_TPTE_PERM(x)		((x) << S_FW_RI_TPTE_PERM)
+#define G_FW_RI_TPTE_PERM(x)		\
+    (((x) >> S_FW_RI_TPTE_PERM) & M_FW_RI_TPTE_PERM)
+
+#define S_FW_RI_TPTE_REMINVDIS		27
+#define M_FW_RI_TPTE_REMINVDIS		0x1
+#define V_FW_RI_TPTE_REMINVDIS(x)	((x) << S_FW_RI_TPTE_REMINVDIS)
+#define G_FW_RI_TPTE_REMINVDIS(x)	\
+    (((x) >> S_FW_RI_TPTE_REMINVDIS) & M_FW_RI_TPTE_REMINVDIS)
+#define F_FW_RI_TPTE_REMINVDIS		V_FW_RI_TPTE_REMINVDIS(1U)
+
+#define S_FW_RI_TPTE_ADDRTYPE		26
+#define M_FW_RI_TPTE_ADDRTYPE		1
+#define V_FW_RI_TPTE_ADDRTYPE(x)	((x) << S_FW_RI_TPTE_ADDRTYPE)
+#define G_FW_RI_TPTE_ADDRTYPE(x)	\
+    (((x) >> S_FW_RI_TPTE_ADDRTYPE) & M_FW_RI_TPTE_ADDRTYPE)
+#define F_FW_RI_TPTE_ADDRTYPE		V_FW_RI_TPTE_ADDRTYPE(1U)
+
+#define S_FW_RI_TPTE_MWBINDEN		25
+#define M_FW_RI_TPTE_MWBINDEN		0x1
+#define V_FW_RI_TPTE_MWBINDEN(x)	((x) << S_FW_RI_TPTE_MWBINDEN)
+#define G_FW_RI_TPTE_MWBINDEN(x)	\
+    (((x) >> S_FW_RI_TPTE_MWBINDEN) & M_FW_RI_TPTE_MWBINDEN)
+#define F_FW_RI_TPTE_MWBINDEN		V_FW_RI_TPTE_MWBINDEN(1U)
+
+#define S_FW_RI_TPTE_PS			20
+#define M_FW_RI_TPTE_PS			0x1f
+#define V_FW_RI_TPTE_PS(x)		((x) << S_FW_RI_TPTE_PS)
+#define G_FW_RI_TPTE_PS(x)		\
+    (((x) >> S_FW_RI_TPTE_PS) & M_FW_RI_TPTE_PS)
+
+#define S_FW_RI_TPTE_QPID		0
+#define M_FW_RI_TPTE_QPID		0xfffff
+#define V_FW_RI_TPTE_QPID(x)		((x) << S_FW_RI_TPTE_QPID)
+#define G_FW_RI_TPTE_QPID(x)		\
+    (((x) >> S_FW_RI_TPTE_QPID) & M_FW_RI_TPTE_QPID)
+
+#define S_FW_RI_TPTE_NOSNOOP		30
+#define M_FW_RI_TPTE_NOSNOOP		0x1
+#define V_FW_RI_TPTE_NOSNOOP(x)		((x) << S_FW_RI_TPTE_NOSNOOP)
+#define G_FW_RI_TPTE_NOSNOOP(x)		\
+    (((x) >> S_FW_RI_TPTE_NOSNOOP) & M_FW_RI_TPTE_NOSNOOP)
+#define F_FW_RI_TPTE_NOSNOOP		V_FW_RI_TPTE_NOSNOOP(1U)
+
+#define S_FW_RI_TPTE_PBLADDR		0
+#define M_FW_RI_TPTE_PBLADDR		0x1fffffff
+#define V_FW_RI_TPTE_PBLADDR(x)		((x) << S_FW_RI_TPTE_PBLADDR)
+#define G_FW_RI_TPTE_PBLADDR(x)		\
+    (((x) >> S_FW_RI_TPTE_PBLADDR) & M_FW_RI_TPTE_PBLADDR)
+
+#define S_FW_RI_TPTE_DCA		24
+#define M_FW_RI_TPTE_DCA		0x1f
+#define V_FW_RI_TPTE_DCA(x)		((x) << S_FW_RI_TPTE_DCA)
+#define G_FW_RI_TPTE_DCA(x)		\
+    (((x) >> S_FW_RI_TPTE_DCA) & M_FW_RI_TPTE_DCA)
+
+#define S_FW_RI_TPTE_MWBCNT_PSTAG	0
+#define M_FW_RI_TPTE_MWBCNT_PSTAG	0xffffff
+#define V_FW_RI_TPTE_MWBCNT_PSTAT(x)	\
+    ((x) << S_FW_RI_TPTE_MWBCNT_PSTAG)
+#define G_FW_RI_TPTE_MWBCNT_PSTAG(x)	\
+    (((x) >> S_FW_RI_TPTE_MWBCNT_PSTAG) & M_FW_RI_TPTE_MWBCNT_PSTAG)
+
+enum fw_ri_res_type {
+	FW_RI_RES_TYPE_SQ,
+	FW_RI_RES_TYPE_RQ,
+	FW_RI_RES_TYPE_CQ,
+};
+
+enum fw_ri_res_op {
+	FW_RI_RES_OP_WRITE,
+	FW_RI_RES_OP_RESET,
+};
+
+struct fw_ri_res {
+	union fw_ri_restype {
+		struct fw_ri_res_sqrq {
+			__u8   restype;
+			__u8   op;
+			__be16 r3;
+			__be32 eqid;
+			__be32 r4[2];
+			__be32 fetchszm_to_iqid;
+			__be32 dcaen_to_eqsize;
+			__be64 eqaddr;
+		} sqrq;
+		struct fw_ri_res_cq {
+			__u8   restype;
+			__u8   op;
+			__be16 r3;
+			__be32 iqid;
+			__be32 r4[2];
+			__be32 iqandst_to_iqandstindex;
+			__be16 iqdroprss_to_iqesize;
+			__be16 iqsize;
+			__be64 iqaddr;
+			__be32 iqns_iqro;
+			__be32 r6_lo;
+			__be64 r7;
+		} cq;
+	} u;
+};
+
+struct fw_ri_res_wr {
+	__be32 op_nres;
+	__be32 len16_pkd;
+	__u64  cookie;
+#ifndef C99_NOT_SUPPORTED
+	struct fw_ri_res res[0];
+#endif
+};
+
+#define S_FW_RI_RES_WR_NRES	0
+#define M_FW_RI_RES_WR_NRES	0xff
+#define V_FW_RI_RES_WR_NRES(x)	((x) << S_FW_RI_RES_WR_NRES)
+#define G_FW_RI_RES_WR_NRES(x)	\
+    (((x) >> S_FW_RI_RES_WR_NRES) & M_FW_RI_RES_WR_NRES)
+
+#define S_FW_RI_RES_WR_FETCHSZM		26
+#define M_FW_RI_RES_WR_FETCHSZM		0x1
+#define V_FW_RI_RES_WR_FETCHSZM(x)	((x) << S_FW_RI_RES_WR_FETCHSZM)
+#define G_FW_RI_RES_WR_FETCHSZM(x)	\
+    (((x) >> S_FW_RI_RES_WR_FETCHSZM) & M_FW_RI_RES_WR_FETCHSZM)
+#define F_FW_RI_RES_WR_FETCHSZM	V_FW_RI_RES_WR_FETCHSZM(1U)
+
+#define S_FW_RI_RES_WR_STATUSPGNS	25
+#define M_FW_RI_RES_WR_STATUSPGNS	0x1
+#define V_FW_RI_RES_WR_STATUSPGNS(x)	((x) << S_FW_RI_RES_WR_STATUSPGNS)
+#define G_FW_RI_RES_WR_STATUSPGNS(x)	\
+    (((x) >> S_FW_RI_RES_WR_STATUSPGNS) & M_FW_RI_RES_WR_STATUSPGNS)
+#define F_FW_RI_RES_WR_STATUSPGNS	V_FW_RI_RES_WR_STATUSPGNS(1U)
+
+#define S_FW_RI_RES_WR_STATUSPGRO	24
+#define M_FW_RI_RES_WR_STATUSPGRO	0x1
+#define V_FW_RI_RES_WR_STATUSPGRO(x)	((x) << S_FW_RI_RES_WR_STATUSPGRO)
+#define G_FW_RI_RES_WR_STATUSPGRO(x)	\
+    (((x) >> S_FW_RI_RES_WR_STATUSPGRO) & M_FW_RI_RES_WR_STATUSPGRO)
+#define F_FW_RI_RES_WR_STATUSPGRO	V_FW_RI_RES_WR_STATUSPGRO(1U)
+
+#define S_FW_RI_RES_WR_FETCHNS		23
+#define M_FW_RI_RES_WR_FETCHNS		0x1
+#define V_FW_RI_RES_WR_FETCHNS(x)	((x) << S_FW_RI_RES_WR_FETCHNS)
+#define G_FW_RI_RES_WR_FETCHNS(x)	\
+    (((x) >> S_FW_RI_RES_WR_FETCHNS) & M_FW_RI_RES_WR_FETCHNS)
+#define F_FW_RI_RES_WR_FETCHNS	V_FW_RI_RES_WR_FETCHNS(1U)
+
+#define S_FW_RI_RES_WR_FETCHRO		22
+#define M_FW_RI_RES_WR_FETCHRO		0x1
+#define V_FW_RI_RES_WR_FETCHRO(x)	((x) << S_FW_RI_RES_WR_FETCHRO)
+#define G_FW_RI_RES_WR_FETCHRO(x)	\
+    (((x) >> S_FW_RI_RES_WR_FETCHRO) & M_FW_RI_RES_WR_FETCHRO)
+#define F_FW_RI_RES_WR_FETCHRO	V_FW_RI_RES_WR_FETCHRO(1U)
+
+#define S_FW_RI_RES_WR_HOSTFCMODE	20
+#define M_FW_RI_RES_WR_HOSTFCMODE	0x3
+#define V_FW_RI_RES_WR_HOSTFCMODE(x)	((x) << S_FW_RI_RES_WR_HOSTFCMODE)
+#define G_FW_RI_RES_WR_HOSTFCMODE(x)	\
+    (((x) >> S_FW_RI_RES_WR_HOSTFCMODE) & M_FW_RI_RES_WR_HOSTFCMODE)
+
+#define S_FW_RI_RES_WR_CPRIO	19
+#define M_FW_RI_RES_WR_CPRIO	0x1
+#define V_FW_RI_RES_WR_CPRIO(x)	((x) << S_FW_RI_RES_WR_CPRIO)
+#define G_FW_RI_RES_WR_CPRIO(x)	\
+    (((x) >> S_FW_RI_RES_WR_CPRIO) & M_FW_RI_RES_WR_CPRIO)
+#define F_FW_RI_RES_WR_CPRIO	V_FW_RI_RES_WR_CPRIO(1U)
+
+#define S_FW_RI_RES_WR_ONCHIP		18
+#define M_FW_RI_RES_WR_ONCHIP		0x1
+#define V_FW_RI_RES_WR_ONCHIP(x)	((x) << S_FW_RI_RES_WR_ONCHIP)
+#define G_FW_RI_RES_WR_ONCHIP(x)	\
+    (((x) >> S_FW_RI_RES_WR_ONCHIP) & M_FW_RI_RES_WR_ONCHIP)
+#define F_FW_RI_RES_WR_ONCHIP	V_FW_RI_RES_WR_ONCHIP(1U)
+
+#define S_FW_RI_RES_WR_PCIECHN		16
+#define M_FW_RI_RES_WR_PCIECHN		0x3
+#define V_FW_RI_RES_WR_PCIECHN(x)	((x) << S_FW_RI_RES_WR_PCIECHN)
+#define G_FW_RI_RES_WR_PCIECHN(x)	\
+    (((x) >> S_FW_RI_RES_WR_PCIECHN) & M_FW_RI_RES_WR_PCIECHN)
+
+#define S_FW_RI_RES_WR_IQID	0
+#define M_FW_RI_RES_WR_IQID	0xffff
+#define V_FW_RI_RES_WR_IQID(x)	((x) << S_FW_RI_RES_WR_IQID)
+#define G_FW_RI_RES_WR_IQID(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQID) & M_FW_RI_RES_WR_IQID)
+
+#define S_FW_RI_RES_WR_DCAEN	31
+#define M_FW_RI_RES_WR_DCAEN	0x1
+#define V_FW_RI_RES_WR_DCAEN(x)	((x) << S_FW_RI_RES_WR_DCAEN)
+#define G_FW_RI_RES_WR_DCAEN(x)	\
+    (((x) >> S_FW_RI_RES_WR_DCAEN) & M_FW_RI_RES_WR_DCAEN)
+#define F_FW_RI_RES_WR_DCAEN	V_FW_RI_RES_WR_DCAEN(1U)
+
+#define S_FW_RI_RES_WR_DCACPU		26
+#define M_FW_RI_RES_WR_DCACPU		0x1f
+#define V_FW_RI_RES_WR_DCACPU(x)	((x) << S_FW_RI_RES_WR_DCACPU)
+#define G_FW_RI_RES_WR_DCACPU(x)	\
+    (((x) >> S_FW_RI_RES_WR_DCACPU) & M_FW_RI_RES_WR_DCACPU)
+
+#define S_FW_RI_RES_WR_FBMIN	23
+#define M_FW_RI_RES_WR_FBMIN	0x7
+#define V_FW_RI_RES_WR_FBMIN(x)	((x) << S_FW_RI_RES_WR_FBMIN)
+#define G_FW_RI_RES_WR_FBMIN(x)	\
+    (((x) >> S_FW_RI_RES_WR_FBMIN) & M_FW_RI_RES_WR_FBMIN)
+
+#define S_FW_RI_RES_WR_FBMAX	20
+#define M_FW_RI_RES_WR_FBMAX	0x7
+#define V_FW_RI_RES_WR_FBMAX(x)	((x) << S_FW_RI_RES_WR_FBMAX)
+#define G_FW_RI_RES_WR_FBMAX(x)	\
+    (((x) >> S_FW_RI_RES_WR_FBMAX) & M_FW_RI_RES_WR_FBMAX)
+
+#define S_FW_RI_RES_WR_CIDXFTHRESHO	19
+#define M_FW_RI_RES_WR_CIDXFTHRESHO	0x1
+#define V_FW_RI_RES_WR_CIDXFTHRESHO(x)	((x) << S_FW_RI_RES_WR_CIDXFTHRESHO)
+#define G_FW_RI_RES_WR_CIDXFTHRESHO(x)	\
+    (((x) >> S_FW_RI_RES_WR_CIDXFTHRESHO) & M_FW_RI_RES_WR_CIDXFTHRESHO)
+#define F_FW_RI_RES_WR_CIDXFTHRESHO	V_FW_RI_RES_WR_CIDXFTHRESHO(1U)
+
+#define S_FW_RI_RES_WR_CIDXFTHRESH	16
+#define M_FW_RI_RES_WR_CIDXFTHRESH	0x7
+#define V_FW_RI_RES_WR_CIDXFTHRESH(x)	((x) << S_FW_RI_RES_WR_CIDXFTHRESH)
+#define G_FW_RI_RES_WR_CIDXFTHRESH(x)	\
+    (((x) >> S_FW_RI_RES_WR_CIDXFTHRESH) & M_FW_RI_RES_WR_CIDXFTHRESH)
+
+#define S_FW_RI_RES_WR_EQSIZE		0
+#define M_FW_RI_RES_WR_EQSIZE		0xffff
+#define V_FW_RI_RES_WR_EQSIZE(x)	((x) << S_FW_RI_RES_WR_EQSIZE)
+#define G_FW_RI_RES_WR_EQSIZE(x)	\
+    (((x) >> S_FW_RI_RES_WR_EQSIZE) & M_FW_RI_RES_WR_EQSIZE)
+
+#define S_FW_RI_RES_WR_IQANDST		15
+#define M_FW_RI_RES_WR_IQANDST		0x1
+#define V_FW_RI_RES_WR_IQANDST(x)	((x) << S_FW_RI_RES_WR_IQANDST)
+#define G_FW_RI_RES_WR_IQANDST(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQANDST) & M_FW_RI_RES_WR_IQANDST)
+#define F_FW_RI_RES_WR_IQANDST	V_FW_RI_RES_WR_IQANDST(1U)
+
+#define S_FW_RI_RES_WR_IQANUS		14
+#define M_FW_RI_RES_WR_IQANUS		0x1
+#define V_FW_RI_RES_WR_IQANUS(x)	((x) << S_FW_RI_RES_WR_IQANUS)
+#define G_FW_RI_RES_WR_IQANUS(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQANUS) & M_FW_RI_RES_WR_IQANUS)
+#define F_FW_RI_RES_WR_IQANUS	V_FW_RI_RES_WR_IQANUS(1U)
+
+#define S_FW_RI_RES_WR_IQANUD		12
+#define M_FW_RI_RES_WR_IQANUD		0x3
+#define V_FW_RI_RES_WR_IQANUD(x)	((x) << S_FW_RI_RES_WR_IQANUD)
+#define G_FW_RI_RES_WR_IQANUD(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQANUD) & M_FW_RI_RES_WR_IQANUD)
+
+#define S_FW_RI_RES_WR_IQANDSTINDEX	0
+#define M_FW_RI_RES_WR_IQANDSTINDEX	0xfff
+#define V_FW_RI_RES_WR_IQANDSTINDEX(x)	((x) << S_FW_RI_RES_WR_IQANDSTINDEX)
+#define G_FW_RI_RES_WR_IQANDSTINDEX(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQANDSTINDEX) & M_FW_RI_RES_WR_IQANDSTINDEX)
+
+#define S_FW_RI_RES_WR_IQDROPRSS	15
+#define M_FW_RI_RES_WR_IQDROPRSS	0x1
+#define V_FW_RI_RES_WR_IQDROPRSS(x)	((x) << S_FW_RI_RES_WR_IQDROPRSS)
+#define G_FW_RI_RES_WR_IQDROPRSS(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQDROPRSS) & M_FW_RI_RES_WR_IQDROPRSS)
+#define F_FW_RI_RES_WR_IQDROPRSS	V_FW_RI_RES_WR_IQDROPRSS(1U)
+
+#define S_FW_RI_RES_WR_IQGTSMODE	14
+#define M_FW_RI_RES_WR_IQGTSMODE	0x1
+#define V_FW_RI_RES_WR_IQGTSMODE(x)	((x) << S_FW_RI_RES_WR_IQGTSMODE)
+#define G_FW_RI_RES_WR_IQGTSMODE(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQGTSMODE) & M_FW_RI_RES_WR_IQGTSMODE)
+#define F_FW_RI_RES_WR_IQGTSMODE	V_FW_RI_RES_WR_IQGTSMODE(1U)
+
+#define S_FW_RI_RES_WR_IQPCIECH		12
+#define M_FW_RI_RES_WR_IQPCIECH		0x3
+#define V_FW_RI_RES_WR_IQPCIECH(x)	((x) << S_FW_RI_RES_WR_IQPCIECH)
+#define G_FW_RI_RES_WR_IQPCIECH(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQPCIECH) & M_FW_RI_RES_WR_IQPCIECH)
+
+#define S_FW_RI_RES_WR_IQDCAEN		11
+#define M_FW_RI_RES_WR_IQDCAEN		0x1
+#define V_FW_RI_RES_WR_IQDCAEN(x)	((x) << S_FW_RI_RES_WR_IQDCAEN)
+#define G_FW_RI_RES_WR_IQDCAEN(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQDCAEN) & M_FW_RI_RES_WR_IQDCAEN)
+#define F_FW_RI_RES_WR_IQDCAEN	V_FW_RI_RES_WR_IQDCAEN(1U)
+
+#define S_FW_RI_RES_WR_IQDCACPU		6
+#define M_FW_RI_RES_WR_IQDCACPU		0x1f
+#define V_FW_RI_RES_WR_IQDCACPU(x)	((x) << S_FW_RI_RES_WR_IQDCACPU)
+#define G_FW_RI_RES_WR_IQDCACPU(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQDCACPU) & M_FW_RI_RES_WR_IQDCACPU)
+
+#define S_FW_RI_RES_WR_IQINTCNTTHRESH		4
+#define M_FW_RI_RES_WR_IQINTCNTTHRESH		0x3
+#define V_FW_RI_RES_WR_IQINTCNTTHRESH(x)	\
+    ((x) << S_FW_RI_RES_WR_IQINTCNTTHRESH)
+#define G_FW_RI_RES_WR_IQINTCNTTHRESH(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQINTCNTTHRESH) & M_FW_RI_RES_WR_IQINTCNTTHRESH)
+
+#define S_FW_RI_RES_WR_IQO	3
+#define M_FW_RI_RES_WR_IQO	0x1
+#define V_FW_RI_RES_WR_IQO(x)	((x) << S_FW_RI_RES_WR_IQO)
+#define G_FW_RI_RES_WR_IQO(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQO) & M_FW_RI_RES_WR_IQO)
+#define F_FW_RI_RES_WR_IQO	V_FW_RI_RES_WR_IQO(1U)
+
+#define S_FW_RI_RES_WR_IQCPRIO		2
+#define M_FW_RI_RES_WR_IQCPRIO		0x1
+#define V_FW_RI_RES_WR_IQCPRIO(x)	((x) << S_FW_RI_RES_WR_IQCPRIO)
+#define G_FW_RI_RES_WR_IQCPRIO(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQCPRIO) & M_FW_RI_RES_WR_IQCPRIO)
+#define F_FW_RI_RES_WR_IQCPRIO	V_FW_RI_RES_WR_IQCPRIO(1U)
+
+#define S_FW_RI_RES_WR_IQESIZE		0
+#define M_FW_RI_RES_WR_IQESIZE		0x3
+#define V_FW_RI_RES_WR_IQESIZE(x)	((x) << S_FW_RI_RES_WR_IQESIZE)
+#define G_FW_RI_RES_WR_IQESIZE(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQESIZE) & M_FW_RI_RES_WR_IQESIZE)
+
+#define S_FW_RI_RES_WR_IQNS	31
+#define M_FW_RI_RES_WR_IQNS	0x1
+#define V_FW_RI_RES_WR_IQNS(x)	((x) << S_FW_RI_RES_WR_IQNS)
+#define G_FW_RI_RES_WR_IQNS(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQNS) & M_FW_RI_RES_WR_IQNS)
+#define F_FW_RI_RES_WR_IQNS	V_FW_RI_RES_WR_IQNS(1U)
+
+#define S_FW_RI_RES_WR_IQRO	30
+#define M_FW_RI_RES_WR_IQRO	0x1
+#define V_FW_RI_RES_WR_IQRO(x)	((x) << S_FW_RI_RES_WR_IQRO)
+#define G_FW_RI_RES_WR_IQRO(x)	\
+    (((x) >> S_FW_RI_RES_WR_IQRO) & M_FW_RI_RES_WR_IQRO)
+#define F_FW_RI_RES_WR_IQRO	V_FW_RI_RES_WR_IQRO(1U)
+
+struct fw_ri_rdma_write_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__be64 r2;
+	__be32 plen;
+	__be32 stag_sink;
+	__be64 to_sink;
+#ifndef C99_NOT_SUPPORTED
+	union {
+		struct fw_ri_immd immd_src[0];
+		struct fw_ri_isgl isgl_src[0];
+	} u;
+#endif
+};
+
+struct fw_ri_send_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__be32 sendop_pkd;
+	__be32 stag_inv;
+	__be32 plen;
+	__be32 r3;
+	__be64 r4;
+#ifndef C99_NOT_SUPPORTED
+	union {
+		struct fw_ri_immd immd_src[0];
+		struct fw_ri_isgl isgl_src[0];
+	} u;
+#endif
+};
+
+#define S_FW_RI_SEND_WR_SENDOP		0
+#define M_FW_RI_SEND_WR_SENDOP		0xf
+#define V_FW_RI_SEND_WR_SENDOP(x)	((x) << S_FW_RI_SEND_WR_SENDOP)
+#define G_FW_RI_SEND_WR_SENDOP(x)	\
+    (((x) >> S_FW_RI_SEND_WR_SENDOP) & M_FW_RI_SEND_WR_SENDOP)
+
+struct fw_ri_rdma_read_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__be64 r2;
+	__be32 stag_sink;
+	__be32 to_sink_hi;
+	__be32 to_sink_lo;
+	__be32 plen;
+	__be32 stag_src;
+	__be32 to_src_hi;
+	__be32 to_src_lo;
+	__be32 r5;
+};
+
+struct fw_ri_recv_wr {
+	__u8   opcode;
+	__u8   r1;
+	__u16  wrid;
+	__u8   r2[3];
+	__u8   len16;
+	struct fw_ri_isgl isgl;
+};
+
+struct fw_ri_bind_mw_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__u8   qpbinde_to_dcacpu;
+	__u8   pgsz_shift;
+	__u8   addr_type;
+	__u8   mem_perms;
+	__be32 stag_mr;
+	__be32 stag_mw;
+	__be32 r3;
+	__be64 len_mw;
+	__be64 va_fbo;
+	__be64 r4;
+};
+
+#define S_FW_RI_BIND_MW_WR_QPBINDE	6
+#define M_FW_RI_BIND_MW_WR_QPBINDE	0x1
+#define V_FW_RI_BIND_MW_WR_QPBINDE(x)	((x) << S_FW_RI_BIND_MW_WR_QPBINDE)
+#define G_FW_RI_BIND_MW_WR_QPBINDE(x)	\
+    (((x) >> S_FW_RI_BIND_MW_WR_QPBINDE) & M_FW_RI_BIND_MW_WR_QPBINDE)
+#define F_FW_RI_BIND_MW_WR_QPBINDE	V_FW_RI_BIND_MW_WR_QPBINDE(1U)
+
+#define S_FW_RI_BIND_MW_WR_NS		5
+#define M_FW_RI_BIND_MW_WR_NS		0x1
+#define V_FW_RI_BIND_MW_WR_NS(x)	((x) << S_FW_RI_BIND_MW_WR_NS)
+#define G_FW_RI_BIND_MW_WR_NS(x)	\
+    (((x) >> S_FW_RI_BIND_MW_WR_NS) & M_FW_RI_BIND_MW_WR_NS)
+#define F_FW_RI_BIND_MW_WR_NS	V_FW_RI_BIND_MW_WR_NS(1U)
+
+#define S_FW_RI_BIND_MW_WR_DCACPU	0
+#define M_FW_RI_BIND_MW_WR_DCACPU	0x1f
+#define V_FW_RI_BIND_MW_WR_DCACPU(x)	((x) << S_FW_RI_BIND_MW_WR_DCACPU)
+#define G_FW_RI_BIND_MW_WR_DCACPU(x)	\
+    (((x) >> S_FW_RI_BIND_MW_WR_DCACPU) & M_FW_RI_BIND_MW_WR_DCACPU)
+
+struct fw_ri_fr_nsmr_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__u8   qpbinde_to_dcacpu;
+	__u8   pgsz_shift;
+	__u8   addr_type;
+	__u8   mem_perms;
+	__be32 stag;
+	__be32 len_hi;
+	__be32 len_lo;
+	__be32 va_hi;
+	__be32 va_lo_fbo;
+};
+
+#define S_FW_RI_FR_NSMR_WR_QPBINDE	6
+#define M_FW_RI_FR_NSMR_WR_QPBINDE	0x1
+#define V_FW_RI_FR_NSMR_WR_QPBINDE(x)	((x) << S_FW_RI_FR_NSMR_WR_QPBINDE)
+#define G_FW_RI_FR_NSMR_WR_QPBINDE(x)	\
+    (((x) >> S_FW_RI_FR_NSMR_WR_QPBINDE) & M_FW_RI_FR_NSMR_WR_QPBINDE)
+#define F_FW_RI_FR_NSMR_WR_QPBINDE	V_FW_RI_FR_NSMR_WR_QPBINDE(1U)
+
+#define S_FW_RI_FR_NSMR_WR_NS		5
+#define M_FW_RI_FR_NSMR_WR_NS		0x1
+#define V_FW_RI_FR_NSMR_WR_NS(x)	((x) << S_FW_RI_FR_NSMR_WR_NS)
+#define G_FW_RI_FR_NSMR_WR_NS(x)	\
+    (((x) >> S_FW_RI_FR_NSMR_WR_NS) & M_FW_RI_FR_NSMR_WR_NS)
+#define F_FW_RI_FR_NSMR_WR_NS	V_FW_RI_FR_NSMR_WR_NS(1U)
+
+#define S_FW_RI_FR_NSMR_WR_DCACPU	0
+#define M_FW_RI_FR_NSMR_WR_DCACPU	0x1f
+#define V_FW_RI_FR_NSMR_WR_DCACPU(x)	((x) << S_FW_RI_FR_NSMR_WR_DCACPU)
+#define G_FW_RI_FR_NSMR_WR_DCACPU(x)	\
+    (((x) >> S_FW_RI_FR_NSMR_WR_DCACPU) & M_FW_RI_FR_NSMR_WR_DCACPU)
+
+struct fw_ri_inv_lstag_wr {
+	__u8   opcode;
+	__u8   flags;
+	__u16  wrid;
+	__u8   r1[3];
+	__u8   len16;
+	__be32 r2;
+	__be32 stag_inv;
+};
+
+enum fw_ri_type {
+	FW_RI_TYPE_INIT,
+	FW_RI_TYPE_FINI,
+	FW_RI_TYPE_TERMINATE
+};
+
+enum fw_ri_init_p2ptype {
+	FW_RI_INIT_P2PTYPE_RDMA_WRITE		= FW_RI_RDMA_WRITE,
+	FW_RI_INIT_P2PTYPE_READ_REQ		= FW_RI_READ_REQ,
+	FW_RI_INIT_P2PTYPE_SEND			= FW_RI_SEND,
+	FW_RI_INIT_P2PTYPE_SEND_WITH_INV	= FW_RI_SEND_WITH_INV,
+	FW_RI_INIT_P2PTYPE_SEND_WITH_SE		= FW_RI_SEND_WITH_SE,
+	FW_RI_INIT_P2PTYPE_SEND_WITH_SE_INV	= FW_RI_SEND_WITH_SE_INV,
+	FW_RI_INIT_P2PTYPE_DISABLED		= 0xf,
+};
+
+struct fw_ri_wr {
+	__be32 op_compl;
+	__be32 flowid_len16;
+	__u64  cookie;
+	union fw_ri {
+		struct fw_ri_init {
+			__u8   type;
+			__u8   mpareqbit_p2ptype;
+			__u8   r4[2];
+			__u8   mpa_attrs;
+			__u8   qp_caps;
+			__be16 nrqe;
+			__be32 pdid;
+			__be32 qpid;
+			__be32 sq_eqid;
+			__be32 rq_eqid;
+			__be32 scqid;
+			__be32 rcqid;
+			__be32 ord_max;
+			__be32 ird_max;
+			__be32 iss;
+			__be32 irs;
+			__be32 hwrqsize;
+			__be32 hwrqaddr;
+			__be64 r5;
+			union fw_ri_init_p2p {
+				struct fw_ri_rdma_write_wr write;
+				struct fw_ri_rdma_read_wr read;
+				struct fw_ri_send_wr send;
+			} u;
+		} init;
+		struct fw_ri_fini {
+			__u8   type;
+			__u8   r3[7];
+			__be64 r4;
+		} fini;
+		struct fw_ri_terminate {
+			__u8   type;
+			__u8   r3[3];
+			__be32 immdlen;
+			__u8   termmsg[40];
+		} terminate;
+	} u;
+};
+
+#define S_FW_RI_WR_MPAREQBIT	7
+#define M_FW_RI_WR_MPAREQBIT	0x1
+#define V_FW_RI_WR_MPAREQBIT(x)	((x) << S_FW_RI_WR_MPAREQBIT)
+#define G_FW_RI_WR_MPAREQBIT(x)	\
+    (((x) >> S_FW_RI_WR_MPAREQBIT) & M_FW_RI_WR_MPAREQBIT)
+#define F_FW_RI_WR_MPAREQBIT	V_FW_RI_WR_MPAREQBIT(1U)
+
+#define S_FW_RI_WR_P2PTYPE	0
+#define M_FW_RI_WR_P2PTYPE	0xf
+#define V_FW_RI_WR_P2PTYPE(x)	((x) << S_FW_RI_WR_P2PTYPE)
+#define G_FW_RI_WR_P2PTYPE(x)	\
+    (((x) >> S_FW_RI_WR_P2PTYPE) & M_FW_RI_WR_P2PTYPE)
+
+struct tcp_options {
+	__be16 mss;
+	__u8 wsf;
+#if defined(__LITTLE_ENDIAN_BITFIELD)
+	__u8:4;
+	__u8 unknown:1;
+	__u8:1;
+	__u8 sack:1;
+	__u8 tstamp:1;
+#else
+	__u8 tstamp:1;
+	__u8 sack:1;
+	__u8:1;
+	__u8 unknown:1;
+	__u8:4;
+#endif
+};
+
+struct cpl_pass_accept_req {
+	union opcode_tid ot;
+	__be16 rsvd;
+	__be16 len;
+	__be32 hdr_len;
+	__be16 vlan;
+	__be16 l2info;
+	__be32 tos_stid;
+	struct tcp_options tcpopt;
+};
+
+/* cpl_pass_accept_req.hdr_len fields */
+#define S_SYN_RX_CHAN    0
+#define M_SYN_RX_CHAN    0xF
+#define V_SYN_RX_CHAN(x) ((x) << S_SYN_RX_CHAN)
+#define G_SYN_RX_CHAN(x) (((x) >> S_SYN_RX_CHAN) & M_SYN_RX_CHAN)
+
+#define S_TCP_HDR_LEN    10
+#define M_TCP_HDR_LEN    0x3F
+#define V_TCP_HDR_LEN(x) ((x) << S_TCP_HDR_LEN)
+#define G_TCP_HDR_LEN(x) (((x) >> S_TCP_HDR_LEN) & M_TCP_HDR_LEN)
+
+#define S_IP_HDR_LEN    16
+#define M_IP_HDR_LEN    0x3FF
+#define V_IP_HDR_LEN(x) ((x) << S_IP_HDR_LEN)
+#define G_IP_HDR_LEN(x) (((x) >> S_IP_HDR_LEN) & M_IP_HDR_LEN)
+
+#define S_ETH_HDR_LEN    26
+#define M_ETH_HDR_LEN    0x1F
+#define V_ETH_HDR_LEN(x) ((x) << S_ETH_HDR_LEN)
+#define G_ETH_HDR_LEN(x) (((x) >> S_ETH_HDR_LEN) & M_ETH_HDR_LEN)
+
+/* cpl_pass_accept_req.l2info fields */
+#define S_SYN_MAC_IDX    0
+#define M_SYN_MAC_IDX    0x1FF
+#define V_SYN_MAC_IDX(x) ((x) << S_SYN_MAC_IDX)
+#define G_SYN_MAC_IDX(x) (((x) >> S_SYN_MAC_IDX) & M_SYN_MAC_IDX)
+
+#define S_SYN_XACT_MATCH    9
+#define V_SYN_XACT_MATCH(x) ((x) << S_SYN_XACT_MATCH)
+#define F_SYN_XACT_MATCH    V_SYN_XACT_MATCH(1U)
+
+#define S_SYN_INTF    12
+#define M_SYN_INTF    0xF
+#define V_SYN_INTF(x) ((x) << S_SYN_INTF)
+#define G_SYN_INTF(x) (((x) >> S_SYN_INTF) & M_SYN_INTF)
+
+struct ulptx_idata {
+	__be32 cmd_more;
+	__be32 len;
+};
+
+#define S_ULPTX_NSGE    0
+#define M_ULPTX_NSGE    0xFFFF
+#define V_ULPTX_NSGE(x) ((x) << S_ULPTX_NSGE)
+
+#define S_RX_DACK_MODE    29
+#define M_RX_DACK_MODE    0x3
+#define V_RX_DACK_MODE(x) ((x) << S_RX_DACK_MODE)
+#define G_RX_DACK_MODE(x) (((x) >> S_RX_DACK_MODE) & M_RX_DACK_MODE)
+
+#define S_RX_DACK_CHANGE    31
+#define V_RX_DACK_CHANGE(x) ((x) << S_RX_DACK_CHANGE)
+#define F_RX_DACK_CHANGE    V_RX_DACK_CHANGE(1U)
+
+enum {                     /* TCP congestion control algorithms */
+	CONG_ALG_RENO,
+	CONG_ALG_TAHOE,
+	CONG_ALG_NEWRENO,
+	CONG_ALG_HIGHSPEED
+};
+
+#define S_CONG_CNTRL    14
+#define M_CONG_CNTRL    0x3
+#define V_CONG_CNTRL(x) ((x) << S_CONG_CNTRL)
+#define G_CONG_CNTRL(x) (((x) >> S_CONG_CNTRL) & M_CONG_CNTRL)
+
+#define CONG_CNTRL_VALID   (1 << 18)
+#define T5_OPT_2_VALID       (1 << 31)
+
+#endif /* _T4FW_RI_API_H_ */
diff --git a/drivers/infiniband/hw/cxgb4/user.h b/drivers/infiniband/hw/cxgb4/user.h
new file mode 100644
index 00000000000..cbd0ce17072
--- /dev/null
+++ b/drivers/infiniband/hw/cxgb4/user.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2009-2010 Chelsio, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef __C4IW_USER_H__
+#define __C4IW_USER_H__
+
+#define C4IW_UVERBS_ABI_VERSION	2
+
+/*
+ * Make sure that all structs defined in this file remain laid out so
+ * that they pack the same way on 32-bit and 64-bit architectures (to
+ * avoid incompatibility between 32-bit userspace and 64-bit kernels).
+ * In particular do not use pointer types -- pass pointers in __u64
+ * instead.
+ */
+struct c4iw_create_cq_resp {
+	__u64 key;
+	__u64 gts_key;
+	__u64 memsize;
+	__u32 cqid;
+	__u32 size;
+	__u32 qid_mask;
+	__u32 reserved; /* explicit padding (optional for i386) */
+};
+
+
+enum {
+	C4IW_QPF_ONCHIP = (1<<0)
+};
+
+struct c4iw_create_qp_resp {
+	__u64 ma_sync_key;
+	__u64 sq_key;
+	__u64 rq_key;
+	__u64 sq_db_gts_key;
+	__u64 rq_db_gts_key;
+	__u64 sq_memsize;
+	__u64 rq_memsize;
+	__u32 sqid;
+	__u32 rqid;
+	__u32 sq_size;
+	__u32 rq_size;
+	__u32 qid_mask;
+	__u32 flags;
+};
+
+struct c4iw_alloc_ucontext_resp {
+	__u64 status_page_key;
+	__u32 status_page_size;
+	__u32 reserved; /* explicit padding (optional for i386) */
+};
+#endif
diff --git a/drivers/infiniband/hw/ehca/ehca_av.c b/drivers/infiniband/hw/ehca/ehca_av.c
index 56735ea2fc5..465926319f3 100644
--- a/drivers/infiniband/hw/ehca/ehca_av.c
+++ b/drivers/infiniband/hw/ehca/ehca_av.c
@@ -41,6 +41,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_tools.h"
 #include "ehca_iverbs.h"
 #include "hcp_if.h"
diff --git a/drivers/infiniband/hw/ehca/ehca_classes.h b/drivers/infiniband/hw/ehca/ehca_classes.h
index 1e9e99a1393..bd45e0f3923 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes.h
@@ -112,7 +112,7 @@ struct ehca_sport {
 
 struct ehca_shca {
 	struct ib_device ib_device;
-	struct of_device *ofdev;
+	struct platform_device *ofdev;
 	u8 num_ports;
 	int hw_level;
 	struct list_head shca_list;
@@ -128,6 +128,8 @@ struct ehca_shca {
 	/* MR pgsize: bit 0-3 means 4K, 64K, 1M, 16M respectively */
 	u32 hca_cap_mr_pgsize;
 	int max_mtu;
+	int max_num_qps;
+	int max_num_cqs;
 	atomic_t num_cqs;
 	atomic_t num_qps;
 };
@@ -156,6 +158,30 @@ struct ehca_mod_qp_parm {
 
 #define EHCA_MOD_QP_PARM_MAX 4
 
+#define QMAP_IDX_MASK 0xFFFFULL
+
+/* struct for tracking if cqes have been reported to the application */
+struct ehca_qmap_entry {
+	u16 app_wr_id;
+	u8 reported;
+	u8 cqe_req;
+};
+
+struct ehca_queue_map {
+	struct ehca_qmap_entry *map;
+	unsigned int entries;
+	unsigned int tail;
+	unsigned int left_to_poll;
+	unsigned int next_wqe_idx;   /* Idx to first wqe to be flushed */
+};
+
+/* function to calculate the next index for the qmap */
+static inline unsigned int next_index(unsigned int cur_index, unsigned int limit)
+{
+	unsigned int temp = cur_index + 1;
+	return (temp == limit) ? 0 : temp;
+}
+
 struct ehca_qp {
 	union {
 		struct ib_qp ib_qp;
@@ -165,7 +191,9 @@ struct ehca_qp {
 	enum ehca_ext_qp_type ext_type;
 	enum ib_qp_state state;
 	struct ipz_queue ipz_squeue;
+	struct ehca_queue_map sq_map;
 	struct ipz_queue ipz_rqueue;
+	struct ehca_queue_map rq_map;
 	struct h_galpas galpas;
 	u32 qkey;
 	u32 real_qp_num;
@@ -194,6 +222,9 @@ struct ehca_qp {
 	u32 packet_count;
 	atomic_t nr_events; /* events seen */
 	wait_queue_head_t wait_completion;
+	int mig_armed;
+	struct list_head sq_err_node;
+	struct list_head rq_err_node;
 };
 
 #define IS_SRQ(qp) (qp->ext_type == EQPT_SRQ)
@@ -223,6 +254,8 @@ struct ehca_cq {
 	/* mmap counter for resources mapped into user space */
 	u32 mm_count_queue;
 	u32 mm_count_galpa;
+	struct list_head sqp_err_list;
+	struct list_head rqp_err_list;
 };
 
 enum ehca_mr_flag {
@@ -289,7 +322,7 @@ struct ehca_mr_pginfo {
 		} phy;
 		struct { /* type EHCA_MR_PGI_USER section */
 			struct ib_umem *region;
-			struct ib_umem_chunk *next_chunk;
+			struct scatterlist *next_sg;
 			u64 next_nmap;
 		} usr;
 		struct { /* type EHCA_MR_PGI_FMR section */
@@ -342,11 +375,12 @@ extern rwlock_t ehca_qp_idr_lock;
 extern rwlock_t ehca_cq_idr_lock;
 extern struct idr ehca_qp_idr;
 extern struct idr ehca_cq_idr;
+extern spinlock_t shca_list_lock;
 
 extern int ehca_static_rate;
 extern int ehca_port_act_time;
-extern int ehca_use_hp_mr;
-extern int ehca_scaling_code;
+extern bool ehca_use_hp_mr;
+extern bool ehca_scaling_code;
 extern int ehca_lock_hcalls;
 extern int ehca_nr_ports;
 extern int ehca_max_cq;
diff --git a/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h b/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
index 1798e6466bd..689c35786dd 100644
--- a/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
+++ b/drivers/infiniband/hw/ehca/ehca_classes_pSeries.h
@@ -165,7 +165,6 @@ struct hcp_modify_qp_control_block {
 #define MQPCB_MASK_ALT_P_KEY_IDX                EHCA_BMASK_IBM( 7,  7)
 #define MQPCB_MASK_RDMA_ATOMIC_CTRL             EHCA_BMASK_IBM( 8,  8)
 #define MQPCB_MASK_QP_STATE                     EHCA_BMASK_IBM( 9,  9)
-#define MQPCB_QP_STATE                          EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_RDMA_NR_ATOMIC_RESP_RES      EHCA_BMASK_IBM(11, 11)
 #define MQPCB_MASK_PATH_MIGRATION_STATE         EHCA_BMASK_IBM(12, 12)
 #define MQPCB_MASK_RDMA_ATOMIC_OUTST_DEST_QP    EHCA_BMASK_IBM(13, 13)
@@ -176,60 +175,33 @@ struct hcp_modify_qp_control_block {
 #define MQPCB_MASK_RETRY_COUNT                  EHCA_BMASK_IBM(18, 18)
 #define MQPCB_MASK_TIMEOUT                      EHCA_BMASK_IBM(19, 19)
 #define MQPCB_MASK_PATH_MTU                     EHCA_BMASK_IBM(20, 20)
-#define MQPCB_PATH_MTU                          EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_MAX_STATIC_RATE              EHCA_BMASK_IBM(21, 21)
-#define MQPCB_MAX_STATIC_RATE                   EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_DLID                         EHCA_BMASK_IBM(22, 22)
-#define MQPCB_DLID                              EHCA_BMASK_IBM(16, 31)
 #define MQPCB_MASK_RNR_RETRY_COUNT              EHCA_BMASK_IBM(23, 23)
-#define MQPCB_RNR_RETRY_COUNT                   EHCA_BMASK_IBM(29, 31)
 #define MQPCB_MASK_SOURCE_PATH_BITS             EHCA_BMASK_IBM(24, 24)
-#define MQPCB_SOURCE_PATH_BITS                  EHCA_BMASK_IBM(25, 31)
 #define MQPCB_MASK_TRAFFIC_CLASS                EHCA_BMASK_IBM(25, 25)
-#define MQPCB_TRAFFIC_CLASS                     EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_HOP_LIMIT                    EHCA_BMASK_IBM(26, 26)
-#define MQPCB_HOP_LIMIT                         EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_SOURCE_GID_IDX               EHCA_BMASK_IBM(27, 27)
-#define MQPCB_SOURCE_GID_IDX                    EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_FLOW_LABEL                   EHCA_BMASK_IBM(28, 28)
-#define MQPCB_FLOW_LABEL                        EHCA_BMASK_IBM(12, 31)
 #define MQPCB_MASK_DEST_GID                     EHCA_BMASK_IBM(30, 30)
 #define MQPCB_MASK_SERVICE_LEVEL_AL             EHCA_BMASK_IBM(31, 31)
-#define MQPCB_SERVICE_LEVEL_AL                  EHCA_BMASK_IBM(28, 31)
 #define MQPCB_MASK_SEND_GRH_FLAG_AL             EHCA_BMASK_IBM(32, 32)
-#define MQPCB_SEND_GRH_FLAG_AL                  EHCA_BMASK_IBM(31, 31)
 #define MQPCB_MASK_RETRY_COUNT_AL               EHCA_BMASK_IBM(33, 33)
-#define MQPCB_RETRY_COUNT_AL                    EHCA_BMASK_IBM(29, 31)
 #define MQPCB_MASK_TIMEOUT_AL                   EHCA_BMASK_IBM(34, 34)
-#define MQPCB_TIMEOUT_AL                        EHCA_BMASK_IBM(27, 31)
 #define MQPCB_MASK_MAX_STATIC_RATE_AL           EHCA_BMASK_IBM(35, 35)
-#define MQPCB_MAX_STATIC_RATE_AL                EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_DLID_AL                      EHCA_BMASK_IBM(36, 36)
-#define MQPCB_DLID_AL                           EHCA_BMASK_IBM(16, 31)
 #define MQPCB_MASK_RNR_RETRY_COUNT_AL           EHCA_BMASK_IBM(37, 37)
-#define MQPCB_RNR_RETRY_COUNT_AL                EHCA_BMASK_IBM(29, 31)
 #define MQPCB_MASK_SOURCE_PATH_BITS_AL          EHCA_BMASK_IBM(38, 38)
-#define MQPCB_SOURCE_PATH_BITS_AL               EHCA_BMASK_IBM(25, 31)
 #define MQPCB_MASK_TRAFFIC_CLASS_AL             EHCA_BMASK_IBM(39, 39)
-#define MQPCB_TRAFFIC_CLASS_AL                  EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_HOP_LIMIT_AL                 EHCA_BMASK_IBM(40, 40)
-#define MQPCB_HOP_LIMIT_AL                      EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_SOURCE_GID_IDX_AL            EHCA_BMASK_IBM(41, 41)
-#define MQPCB_SOURCE_GID_IDX_AL                 EHCA_BMASK_IBM(24, 31)
 #define MQPCB_MASK_FLOW_LABEL_AL                EHCA_BMASK_IBM(42, 42)
-#define MQPCB_FLOW_LABEL_AL                     EHCA_BMASK_IBM(12, 31)
 #define MQPCB_MASK_DEST_GID_AL                  EHCA_BMASK_IBM(44, 44)
 #define MQPCB_MASK_MAX_NR_OUTST_SEND_WR         EHCA_BMASK_IBM(45, 45)
-#define MQPCB_MAX_NR_OUTST_SEND_WR              EHCA_BMASK_IBM(16, 31)
 #define MQPCB_MASK_MAX_NR_OUTST_RECV_WR         EHCA_BMASK_IBM(46, 46)
-#define MQPCB_MAX_NR_OUTST_RECV_WR              EHCA_BMASK_IBM(16, 31)
 #define MQPCB_MASK_DISABLE_ETE_CREDIT_CHECK     EHCA_BMASK_IBM(47, 47)
-#define MQPCB_DISABLE_ETE_CREDIT_CHECK          EHCA_BMASK_IBM(31, 31)
-#define MQPCB_QP_NUMBER                         EHCA_BMASK_IBM( 8, 31)
 #define MQPCB_MASK_QP_ENABLE                    EHCA_BMASK_IBM(48, 48)
-#define MQPCB_QP_ENABLE                         EHCA_BMASK_IBM(31, 31)
 #define MQPCB_MASK_CURR_SRQ_LIMIT               EHCA_BMASK_IBM(49, 49)
-#define MQPCB_CURR_SRQ_LIMIT                    EHCA_BMASK_IBM(16, 31)
 #define MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG       EHCA_BMASK_IBM(50, 50)
 #define MQPCB_MASK_SHARED_RQ_HNDL               EHCA_BMASK_IBM(51, 51)
 
diff --git a/drivers/infiniband/hw/ehca/ehca_cq.c b/drivers/infiniband/hw/ehca/ehca_cq.c
index 5540b276a33..8cc83753776 100644
--- a/drivers/infiniband/hw/ehca/ehca_cq.c
+++ b/drivers/infiniband/hw/ehca/ehca_cq.c
@@ -43,6 +43,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_iverbs.h"
 #include "ehca_classes.h"
 #include "ehca_irq.h"
@@ -126,15 +128,15 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 	void *vpage;
 	u32 counter;
 	u64 rpage, cqx_fec, h_ret;
-	int ipz_rc, ret, i;
+	int ipz_rc, i;
 	unsigned long flags;
 
 	if (cqe >= 0xFFFFFFFF - 64 - additional_cqe)
 		return ERR_PTR(-EINVAL);
 
-	if (!atomic_add_unless(&shca->num_cqs, 1, ehca_max_cq)) {
+	if (!atomic_add_unless(&shca->num_cqs, 1, shca->max_num_cqs)) {
 		ehca_err(device, "Unable to create CQ, max number of %i "
-			"CQs reached.", ehca_max_cq);
+			"CQs reached.", shca->max_num_cqs);
 		ehca_err(device, "To increase the maximum number of CQs "
 			"use the number_of_cqs module parameter.\n");
 		return ERR_PTR(-ENOSPC);
@@ -161,32 +163,19 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 	adapter_handle = shca->ipz_hca_handle;
 	param.eq_handle = shca->eq.ipz_eq_handle;
 
-	do {
-		if (!idr_pre_get(&ehca_cq_idr, GFP_KERNEL)) {
-			cq = ERR_PTR(-ENOMEM);
-			ehca_err(device, "Can't reserve idr nr. device=%p",
-				 device);
-			goto create_cq_exit1;
-		}
-
-		write_lock_irqsave(&ehca_cq_idr_lock, flags);
-		ret = idr_get_new(&ehca_cq_idr, my_cq, &my_cq->token);
-		write_unlock_irqrestore(&ehca_cq_idr_lock, flags);
-	} while (ret == -EAGAIN);
+	idr_preload(GFP_KERNEL);
+	write_lock_irqsave(&ehca_cq_idr_lock, flags);
+	my_cq->token = idr_alloc(&ehca_cq_idr, my_cq, 0, 0x2000000, GFP_NOWAIT);
+	write_unlock_irqrestore(&ehca_cq_idr_lock, flags);
+	idr_preload_end();
 
-	if (ret) {
+	if (my_cq->token < 0) {
 		cq = ERR_PTR(-ENOMEM);
 		ehca_err(device, "Can't allocate new idr entry. device=%p",
 			 device);
 		goto create_cq_exit1;
 	}
 
-	if (my_cq->token > 0x1FFFFFF) {
-		cq = ERR_PTR(-ENOMEM);
-		ehca_err(device, "Invalid number of cq. device=%p", device);
-		goto create_cq_exit2;
-	}
-
 	/*
 	 * CQs maximum depth is 4GB-64, but we need additional 20 as buffer
 	 * for receiving errors CQEs.
@@ -196,7 +185,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 
 	if (h_ret != H_SUCCESS) {
 		ehca_err(device, "hipz_h_alloc_resource_cq() failed "
-			 "h_ret=%li device=%p", h_ret, device);
+			 "h_ret=%lli device=%p", h_ret, device);
 		cq = ERR_PTR(ehca2ib_return_code(h_ret));
 		goto create_cq_exit2;
 	}
@@ -218,7 +207,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 			cq = ERR_PTR(-EAGAIN);
 			goto create_cq_exit4;
 		}
-		rpage = virt_to_abs(vpage);
+		rpage = __pa(vpage);
 
 		h_ret = hipz_h_register_rpage_cq(adapter_handle,
 						 my_cq->ipz_cq_handle,
@@ -232,7 +221,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 
 		if (h_ret < H_SUCCESS) {
 			ehca_err(device, "hipz_h_register_rpage_cq() failed "
-				 "ehca_cq=%p cq_num=%x h_ret=%li counter=%i "
+				 "ehca_cq=%p cq_num=%x h_ret=%lli counter=%i "
 				 "act_pages=%i", my_cq, my_cq->cq_number,
 				 h_ret, counter, param.act_pages);
 			cq = ERR_PTR(-EINVAL);
@@ -244,7 +233,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 			if ((h_ret != H_SUCCESS) || vpage) {
 				ehca_err(device, "Registration of pages not "
 					 "complete ehca_cq=%p cq_num=%x "
-					 "h_ret=%li", my_cq, my_cq->cq_number,
+					 "h_ret=%lli", my_cq, my_cq->cq_number,
 					 h_ret);
 				cq = ERR_PTR(-EAGAIN);
 				goto create_cq_exit4;
@@ -252,7 +241,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 		} else {
 			if (h_ret != H_PAGE_REGISTERED) {
 				ehca_err(device, "Registration of page failed "
-					 "ehca_cq=%p cq_num=%x h_ret=%li "
+					 "ehca_cq=%p cq_num=%x h_ret=%lli "
 					 "counter=%i act_pages=%i",
 					 my_cq, my_cq->cq_number,
 					 h_ret, counter, param.act_pages);
@@ -266,7 +255,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 
 	gal = my_cq->galpas.kernel;
 	cqx_fec = hipz_galpa_load(gal, CQTEMM_OFFSET(cqx_fec));
-	ehca_dbg(device, "ehca_cq=%p cq_num=%x CQX_FEC=%lx",
+	ehca_dbg(device, "ehca_cq=%p cq_num=%x CQX_FEC=%llx",
 		 my_cq, my_cq->cq_number, cqx_fec);
 
 	my_cq->ib_cq.cqe = my_cq->nr_of_entries =
@@ -276,6 +265,9 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 	for (i = 0; i < QP_HASHTAB_LEN; i++)
 		INIT_HLIST_HEAD(&my_cq->qp_hashtab[i]);
 
+	INIT_LIST_HEAD(&my_cq->sqp_err_list);
+	INIT_LIST_HEAD(&my_cq->rqp_err_list);
+
 	if (context) {
 		struct ipz_queue *ipz_queue = &my_cq->ipz_queue;
 		struct ehca_create_cq_resp resp;
@@ -291,6 +283,7 @@ struct ib_cq *ehca_create_cq(struct ib_device *device, int cqe, int comp_vector,
 			(my_cq->galpas.user.fw_handle & (PAGE_SIZE - 1));
 		if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
 			ehca_err(device, "Copy to udata failed.");
+			cq = ERR_PTR(-EFAULT);
 			goto create_cq_exit4;
 		}
 	}
@@ -304,7 +297,7 @@ create_cq_exit3:
 	h_ret = hipz_h_destroy_cq(adapter_handle, my_cq, 1);
 	if (h_ret != H_SUCCESS)
 		ehca_err(device, "hipz_h_destroy_cq() failed ehca_cq=%p "
-			 "cq_num=%x h_ret=%li", my_cq, my_cq->cq_number, h_ret);
+			 "cq_num=%x h_ret=%lli", my_cq, my_cq->cq_number, h_ret);
 
 create_cq_exit2:
 	write_lock_irqsave(&ehca_cq_idr_lock, flags);
@@ -352,7 +345,7 @@ int ehca_destroy_cq(struct ib_cq *cq)
 	h_ret = hipz_h_destroy_cq(adapter_handle, my_cq, 0);
 	if (h_ret == H_R_STATE) {
 		/* cq in err: read err data and destroy it forcibly */
-		ehca_dbg(device, "ehca_cq=%p cq_num=%x ressource=%lx in err "
+		ehca_dbg(device, "ehca_cq=%p cq_num=%x resource=%llx in err "
 			 "state. Try to delete it forcibly.",
 			 my_cq, cq_num, my_cq->ipz_cq_handle.handle);
 		ehca_error_data(shca, my_cq, my_cq->ipz_cq_handle.handle);
@@ -362,7 +355,7 @@ int ehca_destroy_cq(struct ib_cq *cq)
 				 cq_num);
 	}
 	if (h_ret != H_SUCCESS) {
-		ehca_err(device, "hipz_h_destroy_cq() failed h_ret=%li "
+		ehca_err(device, "hipz_h_destroy_cq() failed h_ret=%lli "
 			 "ehca_cq=%p cq_num=%x", h_ret, my_cq, cq_num);
 		return ehca2ib_return_code(h_ret);
 	}
diff --git a/drivers/infiniband/hw/ehca/ehca_eq.c b/drivers/infiniband/hw/ehca/ehca_eq.c
index 49660dfa186..90da6747d39 100644
--- a/drivers/infiniband/hw/ehca/ehca_eq.c
+++ b/drivers/infiniband/hw/ehca/ehca_eq.c
@@ -101,7 +101,7 @@ int ehca_create_eq(struct ehca_shca *shca,
 		if (!vpage)
 			goto create_eq_exit2;
 
-		rpage = virt_to_abs(vpage);
+		rpage = __pa(vpage);
 		h_ret = hipz_h_register_rpage_eq(shca->ipz_hca_handle,
 						 eq->ipz_eq_handle,
 						 &eq->pf,
@@ -113,7 +113,7 @@ int ehca_create_eq(struct ehca_shca *shca,
 			if (h_ret != H_SUCCESS || vpage)
 				goto create_eq_exit2;
 		} else {
-			if (h_ret != H_PAGE_REGISTERED || !vpage)
+			if (h_ret != H_PAGE_REGISTERED)
 				goto create_eq_exit2;
 		}
 	}
@@ -122,21 +122,21 @@ int ehca_create_eq(struct ehca_shca *shca,
 
 	/* register interrupt handlers and initialize work queues */
 	if (type == EHCA_EQ) {
+		tasklet_init(&eq->interrupt_task, ehca_tasklet_eq, (long)shca);
+
 		ret = ibmebus_request_irq(eq->ist, ehca_interrupt_eq,
-					  IRQF_DISABLED, "ehca_eq",
+					  0, "ehca_eq",
 					  (void *)shca);
 		if (ret < 0)
 			ehca_err(ib_dev, "Can't map interrupt handler.");
-
-		tasklet_init(&eq->interrupt_task, ehca_tasklet_eq, (long)shca);
 	} else if (type == EHCA_NEQ) {
+		tasklet_init(&eq->interrupt_task, ehca_tasklet_neq, (long)shca);
+
 		ret = ibmebus_request_irq(eq->ist, ehca_interrupt_neq,
-					  IRQF_DISABLED, "ehca_neq",
+					  0, "ehca_neq",
 					  (void *)shca);
 		if (ret < 0)
 			ehca_err(ib_dev, "Can't map interrupt handler.");
-
-		tasklet_init(&eq->interrupt_task, ehca_tasklet_neq, (long)shca);
 	}
 
 	eq->is_initialized = 1;
@@ -169,12 +169,15 @@ int ehca_destroy_eq(struct ehca_shca *shca, struct ehca_eq *eq)
 	unsigned long flags;
 	u64 h_ret;
 
-	spin_lock_irqsave(&eq->spinlock, flags);
 	ibmebus_free_irq(eq->ist, (void *)shca);
 
-	h_ret = hipz_h_destroy_eq(shca->ipz_hca_handle, eq);
+	spin_lock_irqsave(&shca_list_lock, flags);
+	eq->is_initialized = 0;
+	spin_unlock_irqrestore(&shca_list_lock, flags);
 
-	spin_unlock_irqrestore(&eq->spinlock, flags);
+	tasklet_kill(&eq->interrupt_task);
+
+	h_ret = hipz_h_destroy_eq(shca->ipz_hca_handle, eq);
 
 	if (h_ret != H_SUCCESS) {
 		ehca_err(&shca->ib_device, "Can't free EQ resources.");
diff --git a/drivers/infiniband/hw/ehca/ehca_hca.c b/drivers/infiniband/hw/ehca/ehca_hca.c
index bc3b37d2070..9ed4d258830 100644
--- a/drivers/infiniband/hw/ehca/ehca_hca.c
+++ b/drivers/infiniband/hw/ehca/ehca_hca.c
@@ -39,6 +39,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/gfp.h>
+
 #include "ehca_tools.h"
 #include "ehca_iverbs.h"
 #include "hcp_if.h"
@@ -114,7 +116,9 @@ int ehca_query_device(struct ib_device *ibdev, struct ib_device_attr *props)
 	}
 
 	props->max_pkeys           = 16;
-	props->local_ca_ack_delay  = min_t(u8, rblock->local_ca_ack_delay, 255);
+	/* Some FW versions say 0 here; insert sensible value in that case */
+	props->local_ca_ack_delay  = rblock->local_ca_ack_delay ?
+		min_t(u8, rblock->local_ca_ack_delay, 255) : 12;
 	props->max_raw_ipv6_qp     = limit_uint(rblock->max_raw_ipv6_qp);
 	props->max_raw_ethy_qp     = limit_uint(rblock->max_raw_ethy_qp);
 	props->max_mcast_grp       = limit_uint(rblock->max_mcast_grp);
@@ -229,7 +233,7 @@ int ehca_query_port(struct ib_device *ibdev,
 		props->phys_state      = 5;
 		props->state           = rblock->state;
 		props->active_width    = IB_WIDTH_12X;
-		props->active_speed    = 0x1;
+		props->active_speed    = IB_SPEED_SDR;
 	}
 
 query_port1:
@@ -317,7 +321,7 @@ int ehca_query_gid(struct ib_device *ibdev, u8 port,
 					      ib_device);
 	struct hipz_query_port *rblock;
 
-	if (index > 255) {
+	if (index < 0 || index > 255) {
 		ehca_err(&shca->ib_device, "Invalid index: %x.", index);
 		return -EINVAL;
 	}
@@ -391,7 +395,7 @@ int ehca_modify_port(struct ib_device *ibdev,
 	hret = hipz_h_modify_port(shca->ipz_hca_handle, port,
 				  cap, props->init_type, port_modify_mask);
 	if (hret != H_SUCCESS) {
-		ehca_err(&shca->ib_device, "Modify port failed  h_ret=%li",
+		ehca_err(&shca->ib_device, "Modify port failed  h_ret=%lli",
 			 hret);
 		ret = -EINVAL;
 	}
diff --git a/drivers/infiniband/hw/ehca/ehca_irq.c b/drivers/infiniband/hw/ehca/ehca_irq.c
index 0792d930c48..8615d7cf7e0 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.c
+++ b/drivers/infiniband/hw/ehca/ehca_irq.c
@@ -41,6 +41,9 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+#include <linux/smpboot.h>
+
 #include "ehca_classes.h"
 #include "ehca_irq.h"
 #include "ehca_iverbs.h"
@@ -99,7 +102,7 @@ static void print_error_data(struct ehca_shca *shca, void *data,
 			return;
 
 		ehca_err(&shca->ib_device,
-			 "QP 0x%x (resource=%lx) has errors.",
+			 "QP 0x%x (resource=%llx) has errors.",
 			 qp->ib_qp.qp_num, resource);
 		break;
 	}
@@ -108,21 +111,21 @@ static void print_error_data(struct ehca_shca *shca, void *data,
 		struct ehca_cq *cq = (struct ehca_cq *)data;
 
 		ehca_err(&shca->ib_device,
-			 "CQ 0x%x (resource=%lx) has errors.",
+			 "CQ 0x%x (resource=%llx) has errors.",
 			 cq->cq_number, resource);
 		break;
 	}
 	default:
 		ehca_err(&shca->ib_device,
-			 "Unknown error type: %lx on %s.",
+			 "Unknown error type: %llx on %s.",
 			 type, shca->ib_device.name);
 		break;
 	}
 
-	ehca_err(&shca->ib_device, "Error data is available: %lx.", resource);
+	ehca_err(&shca->ib_device, "Error data is available: %llx.", resource);
 	ehca_err(&shca->ib_device, "EHCA ----- error data begin "
 		 "---------------------------------------------------");
-	ehca_dmp(rblock, length, "resource=%lx", resource);
+	ehca_dmp(rblock, length, "resource=%llx", resource);
 	ehca_err(&shca->ib_device, "EHCA ----- error data end "
 		 "----------------------------------------------------");
 
@@ -152,7 +155,7 @@ int ehca_error_data(struct ehca_shca *shca, void *data,
 
 	if (ret == H_R_STATE)
 		ehca_err(&shca->ib_device,
-			 "No error data is available: %lx.", resource);
+			 "No error data is available: %llx.", resource);
 	else if (ret == H_SUCCESS) {
 		int length;
 
@@ -164,7 +167,7 @@ int ehca_error_data(struct ehca_shca *shca, void *data,
 		print_error_data(shca, data, rblock, length);
 	} else
 		ehca_err(&shca->ib_device,
-			 "Error data could not be fetched: %lx", resource);
+			 "Error data could not be fetched: %llx", resource);
 
 	ehca_free_fw_ctrlblock(rblock);
 
@@ -178,6 +181,10 @@ static void dispatch_qp_event(struct ehca_shca *shca, struct ehca_qp *qp,
 {
 	struct ib_event event;
 
+	/* PATH_MIG without the QP ever having been armed is false alarm */
+	if (event_type == IB_EVENT_PATH_MIG && !qp->mig_armed)
+		return;
+
 	event.device = &shca->ib_device;
 	event.event = event_type;
 
@@ -355,36 +362,48 @@ static void notify_port_conf_change(struct ehca_shca *shca, int port_num)
 	*old_attr = new_attr;
 }
 
+/* replay modify_qp for sqps -- return 0 if all is well, 1 if AQP1 destroyed */
+static int replay_modify_qp(struct ehca_sport *sport)
+{
+	int aqp1_destroyed;
+	unsigned long flags;
+
+	spin_lock_irqsave(&sport->mod_sqp_lock, flags);
+
+	aqp1_destroyed = !sport->ibqp_sqp[IB_QPT_GSI];
+
+	if (sport->ibqp_sqp[IB_QPT_SMI])
+		ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_SMI]);
+	if (!aqp1_destroyed)
+		ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_GSI]);
+
+	spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
+
+	return aqp1_destroyed;
+}
+
 static void parse_ec(struct ehca_shca *shca, u64 eqe)
 {
 	u8 ec   = EHCA_BMASK_GET(NEQE_EVENT_CODE, eqe);
 	u8 port = EHCA_BMASK_GET(NEQE_PORT_NUMBER, eqe);
 	u8 spec_event;
 	struct ehca_sport *sport = &shca->sport[port - 1];
-	unsigned long flags;
 
 	switch (ec) {
 	case 0x30: /* port availability change */
 		if (EHCA_BMASK_GET(NEQE_PORT_AVAILABILITY, eqe)) {
-			int suppress_event;
-			/* replay modify_qp for sqps */
-			spin_lock_irqsave(&sport->mod_sqp_lock, flags);
-			suppress_event = !sport->ibqp_sqp[IB_QPT_GSI];
-			if (sport->ibqp_sqp[IB_QPT_SMI])
-				ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_SMI]);
-			if (!suppress_event)
-				ehca_recover_sqp(sport->ibqp_sqp[IB_QPT_GSI]);
-			spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
-
-			/* AQP1 was destroyed, ignore this event */
-			if (suppress_event)
-				break;
+			/* only replay modify_qp calls in autodetect mode;
+			 * if AQP1 was destroyed, the port is already down
+			 * again and we can drop the event.
+			 */
+			if (ehca_nr_ports < 0)
+				if (replay_modify_qp(sport))
+					break;
 
 			sport->port_state = IB_PORT_ACTIVE;
 			dispatch_port_event(shca, port, IB_EVENT_PORT_ACTIVE,
 					    "is active");
-			ehca_query_sma_attr(shca, port,
-					    &sport->saved_attr);
+			ehca_query_sma_attr(shca, port, &sport->saved_attr);
 		} else {
 			sport->port_state = IB_PORT_DOWN;
 			dispatch_port_event(shca, port, IB_EVENT_PORT_ERR,
@@ -463,13 +482,13 @@ void ehca_tasklet_neq(unsigned long data)
 	struct ehca_eqe *eqe;
 	u64 ret;
 
-	eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
+	eqe = ehca_poll_eq(shca, &shca->neq);
 
 	while (eqe) {
 		if (!EHCA_BMASK_GET(NEQE_COMPLETION_EVENT, eqe->entry))
 			parse_ec(shca, eqe->entry);
 
-		eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->neq);
+		eqe = ehca_poll_eq(shca, &shca->neq);
 	}
 
 	ret = hipz_h_reset_event(shca->ipz_hca_handle,
@@ -498,7 +517,7 @@ static inline void process_eqe(struct ehca_shca *shca, struct ehca_eqe *eqe)
 	struct ehca_cq *cq;
 
 	eqe_value = eqe->entry;
-	ehca_dbg(&shca->ib_device, "eqe_value=%lx", eqe_value);
+	ehca_dbg(&shca->ib_device, "eqe_value=%llx", eqe_value);
 	if (EHCA_BMASK_GET(EQE_COMPLETION_EVENT, eqe_value)) {
 		ehca_dbg(&shca->ib_device, "Got completion event");
 		token = EHCA_BMASK_GET(EQE_CQ_TOKEN, eqe_value);
@@ -532,11 +551,10 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 	struct ehca_eq *eq = &shca->eq;
 	struct ehca_eqe_cache_entry *eqe_cache = eq->eqe_cache;
 	u64 eqe_value, ret;
-	unsigned long flags;
 	int eqe_cnt, i;
 	int eq_empty = 0;
 
-	spin_lock_irqsave(&eq->irq_spinlock, flags);
+	spin_lock(&eq->irq_spinlock);
 	if (is_irq) {
 		const int max_query_cnt = 100;
 		int query_cnt = 0;
@@ -556,8 +574,7 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 	eqe_cnt = 0;
 	do {
 		u32 token;
-		eqe_cache[eqe_cnt].eqe =
-			(struct ehca_eqe *)ehca_poll_eq(shca, eq);
+		eqe_cache[eqe_cnt].eqe = ehca_poll_eq(shca, eq);
 		if (!eqe_cache[eqe_cnt].eqe)
 			break;
 		eqe_value = eqe_cache[eqe_cnt].eqe->entry;
@@ -587,7 +604,7 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 		ret = hipz_h_eoi(eq->ist);
 		if (ret != H_SUCCESS)
 			ehca_err(&shca->ib_device,
-				 "bad return code EOI -rc = %ld\n", ret);
+				 "bad return code EOI -rc = %lld\n", ret);
 		ehca_dbg(&shca->ib_device, "deadman found %x eqe", eqe_cnt);
 	}
 	if (unlikely(eqe_cnt == EHCA_EQE_CACHE_SIZE))
@@ -621,14 +638,14 @@ void ehca_process_eq(struct ehca_shca *shca, int is_irq)
 		goto unlock_irq_spinlock;
 	do {
 		struct ehca_eqe *eqe;
-		eqe = (struct ehca_eqe *)ehca_poll_eq(shca, &shca->eq);
+		eqe = ehca_poll_eq(shca, &shca->eq);
 		if (!eqe)
 			break;
 		process_eqe(shca, eqe);
 	} while (1);
 
 unlock_irq_spinlock:
-	spin_unlock_irqrestore(&eq->irq_spinlock, flags);
+	spin_unlock(&eq->irq_spinlock);
 }
 
 void ehca_tasklet_eq(unsigned long data)
@@ -636,27 +653,30 @@ void ehca_tasklet_eq(unsigned long data)
 	ehca_process_eq((struct ehca_shca*)data, 1);
 }
 
-static inline int find_next_online_cpu(struct ehca_comp_pool *pool)
+static int find_next_online_cpu(struct ehca_comp_pool *pool)
 {
 	int cpu;
 	unsigned long flags;
 
 	WARN_ON_ONCE(!in_interrupt());
 	if (ehca_debug_level >= 3)
-		ehca_dmp(&cpu_online_map, sizeof(cpumask_t), "");
+		ehca_dmp(cpu_online_mask, cpumask_size(), "");
 
 	spin_lock_irqsave(&pool->last_cpu_lock, flags);
-	cpu = next_cpu(pool->last_cpu, cpu_online_map);
-	if (cpu == NR_CPUS)
-		cpu = first_cpu(cpu_online_map);
-	pool->last_cpu = cpu;
+	do {
+		cpu = cpumask_next(pool->last_cpu, cpu_online_mask);
+		if (cpu >= nr_cpu_ids)
+			cpu = cpumask_first(cpu_online_mask);
+		pool->last_cpu = cpu;
+	} while (!per_cpu_ptr(pool->cpu_comp_tasks, cpu)->active);
 	spin_unlock_irqrestore(&pool->last_cpu_lock, flags);
 
 	return cpu;
 }
 
 static void __queue_comp_task(struct ehca_cq *__cq,
-			      struct ehca_cpu_comp_task *cct)
+			      struct ehca_cpu_comp_task *cct,
+			      struct task_struct *thread)
 {
 	unsigned long flags;
 
@@ -667,7 +687,7 @@ static void __queue_comp_task(struct ehca_cq *__cq,
 		__cq->nr_callbacks++;
 		list_add_tail(&__cq->entry, &cct->cq_list);
 		cct->cq_jobs++;
-		wake_up(&cct->wait_queue);
+		wake_up_process(thread);
 	} else
 		__cq->nr_callbacks++;
 
@@ -679,6 +699,7 @@ static void queue_comp_task(struct ehca_cq *__cq)
 {
 	int cpu_id;
 	struct ehca_cpu_comp_task *cct;
+	struct task_struct *thread;
 	int cq_jobs;
 	unsigned long flags;
 
@@ -686,7 +707,8 @@ static void queue_comp_task(struct ehca_cq *__cq)
 	BUG_ON(!cpu_online(cpu_id));
 
 	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
-	BUG_ON(!cct);
+	thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu_id);
+	BUG_ON(!cct || !thread);
 
 	spin_lock_irqsave(&cct->task_lock, flags);
 	cq_jobs = cct->cq_jobs;
@@ -694,28 +716,25 @@ static void queue_comp_task(struct ehca_cq *__cq)
 	if (cq_jobs > 0) {
 		cpu_id = find_next_online_cpu(pool);
 		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu_id);
-		BUG_ON(!cct);
+		thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu_id);
+		BUG_ON(!cct || !thread);
 	}
-
-	__queue_comp_task(__cq, cct);
+	__queue_comp_task(__cq, cct, thread);
 }
 
 static void run_comp_task(struct ehca_cpu_comp_task *cct)
 {
 	struct ehca_cq *cq;
-	unsigned long flags;
-
-	spin_lock_irqsave(&cct->task_lock, flags);
 
 	while (!list_empty(&cct->cq_list)) {
 		cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
-		spin_unlock_irqrestore(&cct->task_lock, flags);
+		spin_unlock_irq(&cct->task_lock);
 
 		comp_event_callback(cq);
 		if (atomic_dec_and_test(&cq->nr_events))
 			wake_up(&cq->wait_completion);
 
-		spin_lock_irqsave(&cct->task_lock, flags);
+		spin_lock_irq(&cct->task_lock);
 		spin_lock(&cq->task_lock);
 		cq->nr_callbacks--;
 		if (!cq->nr_callbacks) {
@@ -724,159 +743,76 @@ static void run_comp_task(struct ehca_cpu_comp_task *cct)
 		}
 		spin_unlock(&cq->task_lock);
 	}
-
-	spin_unlock_irqrestore(&cct->task_lock, flags);
 }
 
-static int comp_task(void *__cct)
+static void comp_task_park(unsigned int cpu)
 {
-	struct ehca_cpu_comp_task *cct = __cct;
-	int cql_empty;
-	DECLARE_WAITQUEUE(wait, current);
-
-	set_current_state(TASK_INTERRUPTIBLE);
-	while (!kthread_should_stop()) {
-		add_wait_queue(&cct->wait_queue, &wait);
-
-		spin_lock_irq(&cct->task_lock);
-		cql_empty = list_empty(&cct->cq_list);
-		spin_unlock_irq(&cct->task_lock);
-		if (cql_empty)
-			schedule();
-		else
-			__set_current_state(TASK_RUNNING);
-
-		remove_wait_queue(&cct->wait_queue, &wait);
+	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+	struct ehca_cpu_comp_task *target;
+	struct task_struct *thread;
+	struct ehca_cq *cq, *tmp;
+	LIST_HEAD(list);
 
-		spin_lock_irq(&cct->task_lock);
-		cql_empty = list_empty(&cct->cq_list);
-		spin_unlock_irq(&cct->task_lock);
-		if (!cql_empty)
-			run_comp_task(__cct);
+	spin_lock_irq(&cct->task_lock);
+	cct->cq_jobs = 0;
+	cct->active = 0;
+	list_splice_init(&cct->cq_list, &list);
+	spin_unlock_irq(&cct->task_lock);
 
-		set_current_state(TASK_INTERRUPTIBLE);
+	cpu = find_next_online_cpu(pool);
+	target = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+	thread = *per_cpu_ptr(pool->cpu_comp_threads, cpu);
+	spin_lock_irq(&target->task_lock);
+	list_for_each_entry_safe(cq, tmp, &list, entry) {
+		list_del(&cq->entry);
+		__queue_comp_task(cq, target, thread);
 	}
-	__set_current_state(TASK_RUNNING);
-
-	return 0;
+	spin_unlock_irq(&target->task_lock);
 }
 
-static struct task_struct *create_comp_task(struct ehca_comp_pool *pool,
-					    int cpu)
+static void comp_task_stop(unsigned int cpu, bool online)
 {
-	struct ehca_cpu_comp_task *cct;
-
-	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-	spin_lock_init(&cct->task_lock);
-	INIT_LIST_HEAD(&cct->cq_list);
-	init_waitqueue_head(&cct->wait_queue);
-	cct->task = kthread_create(comp_task, cct, "ehca_comp/%d", cpu);
-
-	return cct->task;
-}
-
-static void destroy_comp_task(struct ehca_comp_pool *pool,
-			      int cpu)
-{
-	struct ehca_cpu_comp_task *cct;
-	struct task_struct *task;
-	unsigned long flags_cct;
-
-	cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
+	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
 
-	task = cct->task;
-	cct->task = NULL;
+	spin_lock_irq(&cct->task_lock);
 	cct->cq_jobs = 0;
-
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
-
-	if (task)
-		kthread_stop(task);
+	cct->active = 0;
+	WARN_ON(!list_empty(&cct->cq_list));
+	spin_unlock_irq(&cct->task_lock);
 }
 
-static void __cpuinit take_over_work(struct ehca_comp_pool *pool, int cpu)
+static int comp_task_should_run(unsigned int cpu)
 {
 	struct ehca_cpu_comp_task *cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-	LIST_HEAD(list);
-	struct ehca_cq *cq;
-	unsigned long flags_cct;
-
-	spin_lock_irqsave(&cct->task_lock, flags_cct);
-
-	list_splice_init(&cct->cq_list, &list);
-
-	while (!list_empty(&list)) {
-		cq = list_entry(cct->cq_list.next, struct ehca_cq, entry);
-
-		list_del(&cq->entry);
-		__queue_comp_task(cq, per_cpu_ptr(pool->cpu_comp_tasks,
-						  smp_processor_id()));
-	}
-
-	spin_unlock_irqrestore(&cct->task_lock, flags_cct);
 
+	return cct->cq_jobs;
 }
 
-static int __cpuinit comp_pool_callback(struct notifier_block *nfb,
-					unsigned long action,
-					void *hcpu)
+static void comp_task(unsigned int cpu)
 {
-	unsigned int cpu = (unsigned long)hcpu;
-	struct ehca_cpu_comp_task *cct;
+	struct ehca_cpu_comp_task *cct = this_cpu_ptr(pool->cpu_comp_tasks);
+	int cql_empty;
 
-	switch (action) {
-	case CPU_UP_PREPARE:
-	case CPU_UP_PREPARE_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_PREPARE)", cpu);
-		if (!create_comp_task(pool, cpu)) {
-			ehca_gen_err("Can't create comp_task for cpu: %x", cpu);
-			return NOTIFY_BAD;
-		}
-		break;
-	case CPU_UP_CANCELED:
-	case CPU_UP_CANCELED_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_CANCELED)", cpu);
-		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-		kthread_bind(cct->task, any_online_cpu(cpu_online_map));
-		destroy_comp_task(pool, cpu);
-		break;
-	case CPU_ONLINE:
-	case CPU_ONLINE_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_ONLINE)", cpu);
-		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
-		kthread_bind(cct->task, cpu);
-		wake_up_process(cct->task);
-		break;
-	case CPU_DOWN_PREPARE:
-	case CPU_DOWN_PREPARE_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_DOWN_PREPARE)", cpu);
-		break;
-	case CPU_DOWN_FAILED:
-	case CPU_DOWN_FAILED_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_DOWN_FAILED)", cpu);
-		break;
-	case CPU_DEAD:
-	case CPU_DEAD_FROZEN:
-		ehca_gen_dbg("CPU: %x (CPU_DEAD)", cpu);
-		destroy_comp_task(pool, cpu);
-		take_over_work(pool, cpu);
-		break;
+	spin_lock_irq(&cct->task_lock);
+	cql_empty = list_empty(&cct->cq_list);
+	if (!cql_empty) {
+		__set_current_state(TASK_RUNNING);
+		run_comp_task(cct);
 	}
-
-	return NOTIFY_OK;
+	spin_unlock_irq(&cct->task_lock);
 }
 
-static struct notifier_block comp_pool_callback_nb __cpuinitdata = {
-	.notifier_call	= comp_pool_callback,
-	.priority	= 0,
+static struct smp_hotplug_thread comp_pool_threads = {
+	.thread_should_run	= comp_task_should_run,
+	.thread_fn		= comp_task,
+	.thread_comm		= "ehca_comp/%u",
+	.cleanup		= comp_task_stop,
+	.park			= comp_task_park,
 };
 
 int ehca_create_comp_pool(void)
 {
-	int cpu;
-	struct task_struct *task;
+	int cpu, ret = -ENOMEM;
 
 	if (!ehca_scaling_code)
 		return 0;
@@ -886,42 +822,49 @@ int ehca_create_comp_pool(void)
 		return -ENOMEM;
 
 	spin_lock_init(&pool->last_cpu_lock);
-	pool->last_cpu = any_online_cpu(cpu_online_map);
+	pool->last_cpu = cpumask_any(cpu_online_mask);
 
 	pool->cpu_comp_tasks = alloc_percpu(struct ehca_cpu_comp_task);
-	if (pool->cpu_comp_tasks == NULL) {
-		kfree(pool);
-		return -EINVAL;
-	}
+	if (!pool->cpu_comp_tasks)
+		goto out_pool;
 
-	for_each_online_cpu(cpu) {
-		task = create_comp_task(pool, cpu);
-		if (task) {
-			kthread_bind(task, cpu);
-			wake_up_process(task);
-		}
+	pool->cpu_comp_threads = alloc_percpu(struct task_struct *);
+	if (!pool->cpu_comp_threads)
+		goto out_tasks;
+
+	for_each_present_cpu(cpu) {
+		struct ehca_cpu_comp_task *cct;
+
+		cct = per_cpu_ptr(pool->cpu_comp_tasks, cpu);
+		spin_lock_init(&cct->task_lock);
+		INIT_LIST_HEAD(&cct->cq_list);
 	}
 
-	register_hotcpu_notifier(&comp_pool_callback_nb);
+	comp_pool_threads.store = pool->cpu_comp_threads;
+	ret = smpboot_register_percpu_thread(&comp_pool_threads);
+	if (ret)
+		goto out_threads;
 
-	printk(KERN_INFO "eHCA scaling code enabled\n");
+	pr_info("eHCA scaling code enabled\n");
+	return ret;
 
-	return 0;
+out_threads:
+	free_percpu(pool->cpu_comp_threads);
+out_tasks:
+	free_percpu(pool->cpu_comp_tasks);
+out_pool:
+	kfree(pool);
+	return ret;
 }
 
 void ehca_destroy_comp_pool(void)
 {
-	int i;
-
 	if (!ehca_scaling_code)
 		return;
 
-	unregister_hotcpu_notifier(&comp_pool_callback_nb);
+	smpboot_unregister_percpu_thread(&comp_pool_threads);
 
-	for (i = 0; i < NR_CPUS; i++) {
-		if (cpu_online(i))
-			destroy_comp_task(pool, i);
-	}
+	free_percpu(pool->cpu_comp_threads);
 	free_percpu(pool->cpu_comp_tasks);
 	kfree(pool);
 }
diff --git a/drivers/infiniband/hw/ehca/ehca_irq.h b/drivers/infiniband/hw/ehca/ehca_irq.h
index 3346cb06cea..5370199f08c 100644
--- a/drivers/infiniband/hw/ehca/ehca_irq.h
+++ b/drivers/infiniband/hw/ehca/ehca_irq.h
@@ -58,15 +58,15 @@ void ehca_tasklet_eq(unsigned long data);
 void ehca_process_eq(struct ehca_shca *shca, int is_irq);
 
 struct ehca_cpu_comp_task {
-	wait_queue_head_t wait_queue;
 	struct list_head cq_list;
-	struct task_struct *task;
 	spinlock_t task_lock;
 	int cq_jobs;
+	int active;
 };
 
 struct ehca_comp_pool {
-	struct ehca_cpu_comp_task *cpu_comp_tasks;
+	struct ehca_cpu_comp_task __percpu *cpu_comp_tasks;
+	struct task_struct * __percpu *cpu_comp_threads;
 	int last_cpu;
 	spinlock_t last_cpu_lock;
 };
diff --git a/drivers/infiniband/hw/ehca/ehca_iverbs.h b/drivers/infiniband/hw/ehca/ehca_iverbs.h
index a8a2ea585d2..22f79afa7fc 100644
--- a/drivers/infiniband/hw/ehca/ehca_iverbs.h
+++ b/drivers/infiniband/hw/ehca/ehca_iverbs.h
@@ -95,7 +95,7 @@ int ehca_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr);
 
 int ehca_dereg_mr(struct ib_mr *mr);
 
-struct ib_mw *ehca_alloc_mw(struct ib_pd *pd);
+struct ib_mw *ehca_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
 
 int ehca_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
 		 struct ib_mw_bind *mw_bind);
@@ -197,6 +197,8 @@ void ehca_poll_eqs(unsigned long data);
 int ehca_calc_ipd(struct ehca_shca *shca, int port,
 		  enum ib_rate path_rate, u32 *ipd);
 
+void ehca_add_to_err_list(struct ehca_qp *qp, int on_sq);
+
 #ifdef CONFIG_PPC_64K_PAGES
 void *ehca_alloc_fw_ctrlblock(gfp_t flags);
 void ehca_free_fw_ctrlblock(void *ptr);
diff --git a/drivers/infiniband/hw/ehca/ehca_main.c b/drivers/infiniband/hw/ehca/ehca_main.c
index 598844d2edc..cd8d290a09f 100644
--- a/drivers/infiniband/hw/ehca/ehca_main.c
+++ b/drivers/infiniband/hw/ehca/ehca_main.c
@@ -44,29 +44,31 @@
 #include <linux/slab.h>
 #endif
 
+#include <linux/notifier.h>
+#include <linux/memory.h>
 #include "ehca_classes.h"
 #include "ehca_iverbs.h"
 #include "ehca_mrmw.h"
 #include "ehca_tools.h"
 #include "hcp_if.h"
 
-#define HCAD_VERSION "0026"
+#define HCAD_VERSION "0029"
 
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_AUTHOR("Christoph Raisch <raisch@de.ibm.com>");
 MODULE_DESCRIPTION("IBM eServer HCA InfiniBand Device Driver");
 MODULE_VERSION(HCAD_VERSION);
 
-static int ehca_open_aqp1     = 0;
+static bool ehca_open_aqp1    = 0;
 static int ehca_hw_level      = 0;
-static int ehca_poll_all_eqs  = 1;
+static bool ehca_poll_all_eqs = 1;
 
 int ehca_debug_level   = 0;
-int ehca_nr_ports      = 2;
-int ehca_use_hp_mr     = 0;
+int ehca_nr_ports      = -1;
+bool ehca_use_hp_mr    = 0;
 int ehca_port_act_time = 30;
 int ehca_static_rate   = -1;
-int ehca_scaling_code  = 0;
+bool ehca_scaling_code = 0;
 int ehca_lock_hcalls   = -1;
 int ehca_max_cq        = -1;
 int ehca_max_qp        = -1;
@@ -80,7 +82,7 @@ module_param_named(port_act_time, ehca_port_act_time, int,  S_IRUGO);
 module_param_named(poll_all_eqs,  ehca_poll_all_eqs,  bool, S_IRUGO);
 module_param_named(static_rate,   ehca_static_rate,   int,  S_IRUGO);
 module_param_named(scaling_code,  ehca_scaling_code,  bool, S_IRUGO);
-module_param_named(lock_hcalls,   ehca_lock_hcalls,   bool, S_IRUGO);
+module_param_named(lock_hcalls,   ehca_lock_hcalls,   bint, S_IRUGO);
 module_param_named(number_of_cqs, ehca_max_cq,        int,  S_IRUGO);
 module_param_named(number_of_qps, ehca_max_qp,        int,  S_IRUGO);
 
@@ -93,8 +95,8 @@ MODULE_PARM_DESC(hw_level,
 		 "Hardware level (0: autosensing (default), "
 		 "0x10..0x14: eHCA, 0x20..0x23: eHCA2)");
 MODULE_PARM_DESC(nr_ports,
-		 "number of connected ports (-1: autodetect, 1: port one only, "
-		 "2: two ports (default)");
+		 "number of connected ports (-1: autodetect (default), "
+		 "1: port one only, 2: two ports)");
 MODULE_PARM_DESC(use_hp_mr,
 		 "Use high performance MRs (default: no)");
 MODULE_PARM_DESC(port_act_time,
@@ -121,7 +123,7 @@ DEFINE_IDR(ehca_qp_idr);
 DEFINE_IDR(ehca_cq_idr);
 
 static LIST_HEAD(shca_list); /* list of all registered ehcas */
-static DEFINE_SPINLOCK(shca_list_lock);
+DEFINE_SPINLOCK(shca_list_lock);
 
 static struct timer_list poll_eqs_timer;
 
@@ -209,6 +211,7 @@ static int ehca_create_slab_caches(void)
 	if (!ctblk_cache) {
 		ehca_gen_err("Cannot create ctblk SLAB cache.");
 		ehca_cleanup_small_qp_cache();
+		ret = -ENOMEM;
 		goto create_slab_caches6;
 	}
 #endif
@@ -289,8 +292,9 @@ static int ehca_sense_attributes(struct ehca_shca *shca)
 	};
 
 	ehca_gen_dbg("Probing adapter %s...",
-		     shca->ofdev->node->full_name);
-	loc_code = of_get_property(shca->ofdev->node, "ibm,loc-code", NULL);
+		     shca->ofdev->dev.of_node->full_name);
+	loc_code = of_get_property(shca->ofdev->dev.of_node, "ibm,loc-code",
+				   NULL);
 	if (loc_code)
 		ehca_gen_dbg(" ... location lode=%s", loc_code);
 
@@ -302,7 +306,7 @@ static int ehca_sense_attributes(struct ehca_shca *shca)
 
 	h_ret = hipz_h_query_hca(shca->ipz_hca_handle, rblock);
 	if (h_ret != H_SUCCESS) {
-		ehca_gen_err("Cannot query device properties. h_ret=%li",
+		ehca_gen_err("Cannot query device properties. h_ret=%lli",
 			     h_ret);
 		ret = -EPERM;
 		goto sense_attributes1;
@@ -357,7 +361,8 @@ static int ehca_sense_attributes(struct ehca_shca *shca)
 	 * a firmware property, so it's valid across all adapters
 	 */
 	if (ehca_lock_hcalls == -1)
-		ehca_lock_hcalls = !(shca->hca_cap & HCA_CAP_H_ALLOC_RES_SYNC);
+		ehca_lock_hcalls = !EHCA_BMASK_GET(HCA_CAP_H_ALLOC_RES_SYNC,
+					shca->hca_cap);
 
 	/* translate supported MR page sizes; always support 4K */
 	shca->hca_cap_mr_pgsize = EHCA_PAGESIZE;
@@ -366,29 +371,30 @@ static int ehca_sense_attributes(struct ehca_shca *shca)
 			shca->hca_cap_mr_pgsize |= pgsize_map[i + 1];
 
 	/* Set maximum number of CQs and QPs to calculate EQ size */
-	if (ehca_max_qp == -1)
-		ehca_max_qp = min_t(int, rblock->max_qp, EHCA_MAX_NUM_QUEUES);
-	else if (ehca_max_qp < 1 || ehca_max_qp > rblock->max_qp) {
-		ehca_gen_err("Requested number of QPs is out of range (1 - %i) "
-			"specified by HW", rblock->max_qp);
-		ret = -EINVAL;
-		goto sense_attributes1;
+	if (shca->max_num_qps == -1)
+		shca->max_num_qps = min_t(int, rblock->max_qp,
+					  EHCA_MAX_NUM_QUEUES);
+	else if (shca->max_num_qps < 1 || shca->max_num_qps > rblock->max_qp) {
+		ehca_gen_warn("The requested number of QPs is out of range "
+			      "(1 - %i) specified by HW. Value is set to %i",
+			      rblock->max_qp, rblock->max_qp);
+		shca->max_num_qps = rblock->max_qp;
 	}
 
-	if (ehca_max_cq == -1)
-		ehca_max_cq = min_t(int, rblock->max_cq, EHCA_MAX_NUM_QUEUES);
-	else if (ehca_max_cq < 1 || ehca_max_cq > rblock->max_cq) {
-		ehca_gen_err("Requested number of CQs is out of range (1 - %i) "
-			"specified by HW", rblock->max_cq);
-		ret = -EINVAL;
-		goto sense_attributes1;
+	if (shca->max_num_cqs == -1)
+		shca->max_num_cqs = min_t(int, rblock->max_cq,
+					  EHCA_MAX_NUM_QUEUES);
+	else if (shca->max_num_cqs < 1 || shca->max_num_cqs > rblock->max_cq) {
+		ehca_gen_warn("The requested number of CQs is out of range "
+			      "(1 - %i) specified by HW. Value is set to %i",
+			      rblock->max_cq, rblock->max_cq);
 	}
 
 	/* query max MTU from first port -- it's the same for all ports */
 	port = (struct hipz_query_port *)rblock;
 	h_ret = hipz_h_query_port(shca->ipz_hca_handle, 1, port);
 	if (h_ret != H_SUCCESS) {
-		ehca_gen_err("Cannot query port properties. h_ret=%li",
+		ehca_gen_err("Cannot query port properties. h_ret=%lli",
 			     h_ret);
 		ret = -EPERM;
 		goto sense_attributes1;
@@ -503,6 +509,7 @@ static int ehca_init_device(struct ehca_shca *shca)
 	shca->ib_device.detach_mcast	    = ehca_detach_mcast;
 	shca->ib_device.process_mad	    = ehca_process_mad;
 	shca->ib_device.mmap		    = ehca_mmap;
+	shca->ib_device.dma_ops		    = &ehca_dma_mapping_ops;
 
 	if (EHCA_BMASK_GET(HCA_CAP_SRQ, shca->hca_cap)) {
 		shca->ib_device.uverbs_cmd_mask |=
@@ -619,7 +626,7 @@ static struct attribute_group ehca_drv_attr_grp = {
 	.attrs = ehca_drv_attrs
 };
 
-static struct attribute_group *ehca_drv_attr_groups[] = {
+static const struct attribute_group *ehca_drv_attr_groups[] = {
 	&ehca_drv_attr_grp,
 	NULL,
 };
@@ -633,7 +640,7 @@ static ssize_t  ehca_show_##name(struct device *dev,                       \
 	struct hipz_query_hca *rblock;				           \
 	int data;                                                          \
 									   \
-	shca = dev->driver_data;					   \
+	shca = dev_get_drvdata(dev);					   \
 									   \
 	rblock = ehca_alloc_fw_ctrlblock(GFP_KERNEL);			   \
 	if (!rblock) {						           \
@@ -677,9 +684,9 @@ static ssize_t ehca_show_adapter_handle(struct device *dev,
 					struct device_attribute *attr,
 					char *buf)
 {
-	struct ehca_shca *shca = dev->driver_data;
+	struct ehca_shca *shca = dev_get_drvdata(dev);
 
-	return sprintf(buf, "%lx\n", shca->ipz_hca_handle.handle);
+	return sprintf(buf, "%llx\n", shca->ipz_hca_handle.handle);
 
 }
 static DEVICE_ATTR(adapter_handle, S_IRUGO, ehca_show_adapter_handle, NULL);
@@ -707,24 +714,24 @@ static struct attribute_group ehca_dev_attr_grp = {
 	.attrs = ehca_dev_attrs
 };
 
-static int __devinit ehca_probe(struct of_device *dev,
-				const struct of_device_id *id)
+static int ehca_probe(struct platform_device *dev)
 {
 	struct ehca_shca *shca;
 	const u64 *handle;
 	struct ib_pd *ibpd;
 	int ret, i, eq_size;
+	unsigned long flags;
 
-	handle = of_get_property(dev->node, "ibm,hca-handle", NULL);
+	handle = of_get_property(dev->dev.of_node, "ibm,hca-handle", NULL);
 	if (!handle) {
 		ehca_gen_err("Cannot get eHCA handle for adapter: %s.",
-			     dev->node->full_name);
+			     dev->dev.of_node->full_name);
 		return -ENODEV;
 	}
 
 	if (!(*handle)) {
 		ehca_gen_err("Wrong eHCA handle for adapter: %s.",
-			     dev->node->full_name);
+			     dev->dev.of_node->full_name);
 		return -ENODEV;
 	}
 
@@ -733,15 +740,19 @@ static int __devinit ehca_probe(struct of_device *dev,
 		ehca_gen_err("Cannot allocate shca memory.");
 		return -ENOMEM;
 	}
+
 	mutex_init(&shca->modify_mutex);
 	atomic_set(&shca->num_cqs, 0);
 	atomic_set(&shca->num_qps, 0);
+	shca->max_num_qps = ehca_max_qp;
+	shca->max_num_cqs = ehca_max_cq;
+
 	for (i = 0; i < ARRAY_SIZE(shca->sport); i++)
 		spin_lock_init(&shca->sport[i].mod_sqp_lock);
 
 	shca->ofdev = dev;
 	shca->ipz_hca_handle.handle = *handle;
-	dev->dev.driver_data = shca;
+	dev_set_drvdata(&dev->dev, shca);
 
 	ret = ehca_sense_attributes(shca);
 	if (ret < 0) {
@@ -755,7 +766,7 @@ static int __devinit ehca_probe(struct of_device *dev,
 		goto probe1;
 	}
 
-	eq_size = 2 * ehca_max_cq + 4 * ehca_max_qp;
+	eq_size = 2 * shca->max_num_cqs + 4 * shca->max_num_qps;
 	/* create event queues */
 	ret = ehca_create_eq(shca, &shca->eq, EHCA_EQ, eq_size);
 	if (ret) {
@@ -789,7 +800,7 @@ static int __devinit ehca_probe(struct of_device *dev,
 		goto probe5;
 	}
 
-	ret = ib_register_device(&shca->ib_device);
+	ret = ib_register_device(&shca->ib_device, NULL);
 	if (ret) {
 		ehca_err(&shca->ib_device,
 			 "ib_register_device() failed ret=%i", ret);
@@ -823,9 +834,9 @@ static int __devinit ehca_probe(struct of_device *dev,
 		ehca_err(&shca->ib_device,
 			 "Cannot create device attributes  ret=%d", ret);
 
-	spin_lock(&shca_list_lock);
+	spin_lock_irqsave(&shca_list_lock, flags);
 	list_add(&shca->shca_list, &shca_list);
-	spin_unlock(&shca_list_lock);
+	spin_unlock_irqrestore(&shca_list_lock, flags);
 
 	return 0;
 
@@ -868,9 +879,10 @@ probe1:
 	return -EINVAL;
 }
 
-static int __devexit ehca_remove(struct of_device *dev)
+static int ehca_remove(struct platform_device *dev)
 {
-	struct ehca_shca *shca = dev->dev.driver_data;
+	struct ehca_shca *shca = dev_get_drvdata(&dev->dev);
+	unsigned long flags;
 	int ret;
 
 	sysfs_remove_group(&dev->dev.kobj, &ehca_dev_attr_grp);
@@ -908,9 +920,9 @@ static int __devexit ehca_remove(struct of_device *dev)
 
 	ib_dealloc_device(&shca->ib_device);
 
-	spin_lock(&shca_list_lock);
+	spin_lock_irqsave(&shca_list_lock, flags);
 	list_del(&shca->shca_list);
-	spin_unlock(&shca_list_lock);
+	spin_unlock_irqrestore(&shca_list_lock, flags);
 
 	return ret;
 }
@@ -925,13 +937,14 @@ static struct of_device_id ehca_device_table[] =
 };
 MODULE_DEVICE_TABLE(of, ehca_device_table);
 
-static struct of_platform_driver ehca_driver = {
-	.name        = "ehca",
-	.match_table = ehca_device_table,
+static struct platform_driver ehca_driver = {
 	.probe       = ehca_probe,
 	.remove      = ehca_remove,
-	.driver	     = {
+	.driver = {
+		.name = "ehca",
+		.owner = THIS_MODULE,
 		.groups = ehca_drv_attr_groups,
+		.of_match_table = ehca_device_table,
 	},
 };
 
@@ -946,7 +959,7 @@ void ehca_poll_eqs(unsigned long data)
 			struct ehca_eq *eq = &shca->eq;
 			int max = 3;
 			volatile u64 q_ofs, q_ofs2;
-			u64 flags;
+			unsigned long flags;
 			spin_lock_irqsave(&eq->spinlock, flags);
 			q_ofs = eq->ipz_queue.current_q_offset;
 			spin_unlock_irqrestore(&eq->spinlock, flags);
@@ -964,6 +977,41 @@ void ehca_poll_eqs(unsigned long data)
 	spin_unlock(&shca_list_lock);
 }
 
+static int ehca_mem_notifier(struct notifier_block *nb,
+			     unsigned long action, void *data)
+{
+	static unsigned long ehca_dmem_warn_time;
+	unsigned long flags;
+
+	switch (action) {
+	case MEM_CANCEL_OFFLINE:
+	case MEM_CANCEL_ONLINE:
+	case MEM_ONLINE:
+	case MEM_OFFLINE:
+		return NOTIFY_OK;
+	case MEM_GOING_ONLINE:
+	case MEM_GOING_OFFLINE:
+		/* only ok if no hca is attached to the lpar */
+		spin_lock_irqsave(&shca_list_lock, flags);
+		if (list_empty(&shca_list)) {
+			spin_unlock_irqrestore(&shca_list_lock, flags);
+			return NOTIFY_OK;
+		} else {
+			spin_unlock_irqrestore(&shca_list_lock, flags);
+			if (printk_timed_ratelimit(&ehca_dmem_warn_time,
+						   30 * 1000))
+				ehca_gen_err("DMEM operations are not allowed"
+					     "in conjunction with eHCA");
+			return NOTIFY_BAD;
+		}
+	}
+	return NOTIFY_OK;
+}
+
+static struct notifier_block ehca_mem_nb = {
+	.notifier_call = ehca_mem_notifier,
+};
+
 static int __init ehca_module_init(void)
 {
 	int ret;
@@ -984,11 +1032,23 @@ static int __init ehca_module_init(void)
 		goto module_init1;
 	}
 
+	ret = ehca_create_busmap();
+	if (ret) {
+		ehca_gen_err("Cannot create busmap.");
+		goto module_init2;
+	}
+
 	ret = ibmebus_register_driver(&ehca_driver);
 	if (ret) {
 		ehca_gen_err("Cannot register eHCA device driver");
 		ret = -EINVAL;
-		goto module_init2;
+		goto module_init3;
+	}
+
+	ret = register_memory_notifier(&ehca_mem_nb);
+	if (ret) {
+		ehca_gen_err("Failed registering memory add/remove notifier");
+		goto module_init4;
 	}
 
 	if (ehca_poll_all_eqs != 1) {
@@ -1003,6 +1063,12 @@ static int __init ehca_module_init(void)
 
 	return 0;
 
+module_init4:
+	ibmebus_unregister_driver(&ehca_driver);
+
+module_init3:
+	ehca_destroy_busmap();
+
 module_init2:
 	ehca_destroy_slab_caches();
 
@@ -1018,6 +1084,10 @@ static void __exit ehca_module_exit(void)
 
 	ibmebus_unregister_driver(&ehca_driver);
 
+	unregister_memory_notifier(&ehca_mem_nb);
+
+	ehca_destroy_busmap();
+
 	ehca_destroy_slab_caches();
 
 	ehca_destroy_comp_pool();
diff --git a/drivers/infiniband/hw/ehca/ehca_mcast.c b/drivers/infiniband/hw/ehca/ehca_mcast.c
index e3ef0264ccc..120aedf9f98 100644
--- a/drivers/infiniband/hw/ehca/ehca_mcast.c
+++ b/drivers/infiniband/hw/ehca/ehca_mcast.c
@@ -88,7 +88,7 @@ int ehca_attach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	if (h_ret != H_SUCCESS)
 		ehca_err(ibqp->device,
 			 "ehca_qp=%p qp_num=%x hipz_h_attach_mcqp() failed "
-			 "h_ret=%li", my_qp, ibqp->qp_num, h_ret);
+			 "h_ret=%lli", my_qp, ibqp->qp_num, h_ret);
 
 	return ehca2ib_return_code(h_ret);
 }
@@ -125,7 +125,7 @@ int ehca_detach_mcast(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	if (h_ret != H_SUCCESS)
 		ehca_err(ibqp->device,
 			 "ehca_qp=%p qp_num=%x hipz_h_detach_mcqp() failed "
-			 "h_ret=%li", my_qp, ibqp->qp_num, h_ret);
+			 "h_ret=%lli", my_qp, ibqp->qp_num, h_ret);
 
 	return ehca2ib_return_code(h_ret);
 }
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.c b/drivers/infiniband/hw/ehca/ehca_mrmw.c
index f974367cad4..3488e8c9fcb 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.c
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.c
@@ -40,6 +40,7 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
 #include <rdma/ib_umem.h>
 
 #include "ehca_iverbs.h"
@@ -53,6 +54,38 @@
 /* max number of rpages (per hcall register_rpages) */
 #define MAX_RPAGES 512
 
+/* DMEM toleration management */
+#define EHCA_SECTSHIFT        SECTION_SIZE_BITS
+#define EHCA_SECTSIZE          (1UL << EHCA_SECTSHIFT)
+#define EHCA_HUGEPAGESHIFT     34
+#define EHCA_HUGEPAGE_SIZE     (1UL << EHCA_HUGEPAGESHIFT)
+#define EHCA_HUGEPAGE_PFN_MASK ((EHCA_HUGEPAGE_SIZE - 1) >> PAGE_SHIFT)
+#define EHCA_INVAL_ADDR        0xFFFFFFFFFFFFFFFFULL
+#define EHCA_DIR_INDEX_SHIFT 13                   /* 8k Entries in 64k block */
+#define EHCA_TOP_INDEX_SHIFT (EHCA_DIR_INDEX_SHIFT * 2)
+#define EHCA_MAP_ENTRIES (1 << EHCA_DIR_INDEX_SHIFT)
+#define EHCA_TOP_MAP_SIZE (0x10000)               /* currently fixed map size */
+#define EHCA_DIR_MAP_SIZE (0x10000)
+#define EHCA_ENT_MAP_SIZE (0x10000)
+#define EHCA_INDEX_MASK (EHCA_MAP_ENTRIES - 1)
+
+static unsigned long ehca_mr_len;
+
+/*
+ * Memory map data structures
+ */
+struct ehca_dir_bmap {
+	u64 ent[EHCA_MAP_ENTRIES];
+};
+struct ehca_top_bmap {
+	struct ehca_dir_bmap *dir[EHCA_MAP_ENTRIES];
+};
+struct ehca_bmap {
+	struct ehca_top_bmap *top[EHCA_MAP_ENTRIES];
+};
+
+static struct ehca_bmap *ehca_bmap;
+
 static struct kmem_cache *mr_cache;
 static struct kmem_cache *mw_cache;
 
@@ -68,6 +101,8 @@ enum ehca_mr_pgsize {
 #define EHCA_MR_PGSHIFT1M  20
 #define EHCA_MR_PGSHIFT16M 24
 
+static u64 ehca_map_vaddr(void *caddr);
+
 static u32 ehca_encode_hwpage_size(u32 pgsize)
 {
 	int log = ilog2(pgsize);
@@ -77,7 +112,7 @@ static u32 ehca_encode_hwpage_size(u32 pgsize)
 
 static u64 ehca_get_max_hwpage_size(struct ehca_shca *shca)
 {
-	return 1UL << ilog2(shca->hca_cap_mr_pgsize);
+	return rounddown_pow_of_two(shca->hca_cap_mr_pgsize);
 }
 
 static struct ehca_mr *ehca_mr_new(void)
@@ -135,7 +170,8 @@ struct ib_mr *ehca_get_dma_mr(struct ib_pd *pd, int mr_access_flags)
 			goto get_dma_mr_exit0;
 		}
 
-		ret = ehca_reg_maxmr(shca, e_maxmr, (u64 *)KERNELBASE,
+		ret = ehca_reg_maxmr(shca, e_maxmr,
+				     (void *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START)),
 				     mr_access_flags, e_pd,
 				     &e_maxmr->ib.ib_mr.lkey,
 				     &e_maxmr->ib.ib_mr.rkey);
@@ -204,7 +240,7 @@ struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd,
 	}
 	if ((size == 0) ||
 	    (((u64)iova_start + size) < (u64)iova_start)) {
-		ehca_err(pd->device, "bad input values: size=%lx iova_start=%p",
+		ehca_err(pd->device, "bad input values: size=%llx iova_start=%p",
 			 size, iova_start);
 		ib_mr = ERR_PTR(-EINVAL);
 		goto reg_phys_mr_exit0;
@@ -251,7 +287,7 @@ struct ib_mr *ehca_reg_phys_mr(struct ib_pd *pd,
 
 		ret = ehca_reg_mr(shca, e_mr, iova_start, size, mr_access_flags,
 				  e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
-				  &e_mr->ib.ib_mr.rkey);
+				  &e_mr->ib.ib_mr.rkey, EHCA_REG_MR);
 		if (ret) {
 			ib_mr = ERR_PTR(ret);
 			goto reg_phys_mr_exit1;
@@ -309,8 +345,8 @@ struct ib_mr *ehca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	if (length == 0 || virt + length < virt) {
-		ehca_err(pd->device, "bad input values: length=%lx "
-			 "virt_base=%lx", length, virt);
+		ehca_err(pd->device, "bad input values: length=%llx "
+			 "virt_base=%llx", length, virt);
 		ib_mr = ERR_PTR(-EINVAL);
 		goto reg_user_mr_exit0;
 	}
@@ -364,16 +400,13 @@ reg_user_mr_fallback:
 	pginfo.num_hwpages = num_hwpages;
 	pginfo.u.usr.region = e_mr->umem;
 	pginfo.next_hwpage = e_mr->umem->offset / hwpage_size;
-	pginfo.u.usr.next_chunk = list_prepare_entry(pginfo.u.usr.next_chunk,
-						     (&e_mr->umem->chunk_list),
-						     list);
-
+	pginfo.u.usr.next_sg = pginfo.u.usr.region->sg_head.sgl;
 	ret = ehca_reg_mr(shca, e_mr, (u64 *)virt, length, mr_access_flags,
 			  e_pd, &pginfo, &e_mr->ib.ib_mr.lkey,
-			  &e_mr->ib.ib_mr.rkey);
+			  &e_mr->ib.ib_mr.rkey, EHCA_REG_MR);
 	if (ret == -EINVAL && pginfo.hwpage_size > PAGE_SIZE) {
 		ehca_warn(pd->device, "failed to register mr "
-			  "with hwpage_size=%lx", hwpage_size);
+			  "with hwpage_size=%llx", hwpage_size);
 		ehca_info(pd->device, "try to register mr with "
 			  "kpage_size=%lx", PAGE_SIZE);
 		/*
@@ -509,7 +542,7 @@ int ehca_rereg_phys_mr(struct ib_mr *mr,
 			goto rereg_phys_mr_exit1;
 		if ((new_size == 0) ||
 		    (((u64)iova_start + new_size) < (u64)iova_start)) {
-			ehca_err(mr->device, "bad input values: new_size=%lx "
+			ehca_err(mr->device, "bad input values: new_size=%llx "
 				 "iova_start=%p", new_size, iova_start);
 			ret = -EINVAL;
 			goto rereg_phys_mr_exit1;
@@ -580,8 +613,8 @@ int ehca_query_mr(struct ib_mr *mr, struct ib_mr_attr *mr_attr)
 
 	h_ret = hipz_h_query_mr(shca->ipz_hca_handle, e_mr, &hipzout);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(mr->device, "hipz_mr_query failed, h_ret=%li mr=%p "
-			 "hca_hndl=%lx mr_hndl=%lx lkey=%x",
+		ehca_err(mr->device, "hipz_mr_query failed, h_ret=%lli mr=%p "
+			 "hca_hndl=%llx mr_hndl=%llx lkey=%x",
 			 h_ret, mr, shca->ipz_hca_handle.handle,
 			 e_mr->ipz_mr_handle.handle, mr->lkey);
 		ret = ehca2ib_return_code(h_ret);
@@ -630,8 +663,8 @@ int ehca_dereg_mr(struct ib_mr *mr)
 	/* TODO: BUSY: MR still has bound window(s) */
 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(mr->device, "hipz_free_mr failed, h_ret=%li shca=%p "
-			 "e_mr=%p hca_hndl=%lx mr_hndl=%lx mr->lkey=%x",
+		ehca_err(mr->device, "hipz_free_mr failed, h_ret=%lli shca=%p "
+			 "e_mr=%p hca_hndl=%llx mr_hndl=%llx mr->lkey=%x",
 			 h_ret, shca, e_mr, shca->ipz_hca_handle.handle,
 			 e_mr->ipz_mr_handle.handle, mr->lkey);
 		ret = ehca2ib_return_code(h_ret);
@@ -652,7 +685,7 @@ dereg_mr_exit0:
 
 /*----------------------------------------------------------------------*/
 
-struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
+struct ib_mw *ehca_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
 {
 	struct ib_mw *ib_mw;
 	u64 h_ret;
@@ -662,6 +695,9 @@ struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
 		container_of(pd->device, struct ehca_shca, ib_device);
 	struct ehca_mw_hipzout_parms hipzout;
 
+	if (type != IB_MW_TYPE_1)
+		return ERR_PTR(-EINVAL);
+
 	e_mw = ehca_mw_new();
 	if (!e_mw) {
 		ib_mw = ERR_PTR(-ENOMEM);
@@ -671,8 +707,8 @@ struct ib_mw *ehca_alloc_mw(struct ib_pd *pd)
 	h_ret = hipz_h_alloc_resource_mw(shca->ipz_hca_handle, e_mw,
 					 e_pd->fw_pd, &hipzout);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(pd->device, "hipz_mw_allocate failed, h_ret=%li "
-			 "shca=%p hca_hndl=%lx mw=%p",
+		ehca_err(pd->device, "hipz_mw_allocate failed, h_ret=%lli "
+			 "shca=%p hca_hndl=%llx mw=%p",
 			 h_ret, shca, shca->ipz_hca_handle.handle, e_mw);
 		ib_mw = ERR_PTR(ehca2ib_return_code(h_ret));
 		goto alloc_mw_exit1;
@@ -713,8 +749,8 @@ int ehca_dealloc_mw(struct ib_mw *mw)
 
 	h_ret = hipz_h_free_resource_mw(shca->ipz_hca_handle, e_mw);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(mw->device, "hipz_free_mw failed, h_ret=%li shca=%p "
-			 "mw=%p rkey=%x hca_hndl=%lx mw_hndl=%lx",
+		ehca_err(mw->device, "hipz_free_mw failed, h_ret=%lli shca=%p "
+			 "mw=%p rkey=%x hca_hndl=%llx mw_hndl=%llx",
 			 h_ret, shca, mw, mw->rkey, shca->ipz_hca_handle.handle,
 			 e_mw->ipz_mw_handle.handle);
 		return ehca2ib_return_code(h_ret);
@@ -794,7 +830,7 @@ struct ib_fmr *ehca_alloc_fmr(struct ib_pd *pd,
 	ret = ehca_reg_mr(shca, e_fmr, NULL,
 			  fmr_attr->max_pages * (1 << fmr_attr->page_shift),
 			  mr_access_flags, e_pd, &pginfo,
-			  &tmp_lkey, &tmp_rkey);
+			  &tmp_lkey, &tmp_rkey, EHCA_REG_MR);
 	if (ret) {
 		ib_fmr = ERR_PTR(ret);
 		goto alloc_fmr_exit1;
@@ -840,7 +876,7 @@ int ehca_map_phys_fmr(struct ib_fmr *fmr,
 		goto map_phys_fmr_exit0;
 	if (iova % e_fmr->fmr_page_size) {
 		/* only whole-numbered pages */
-		ehca_err(fmr->device, "bad iova, iova=%lx fmr_page_size=%x",
+		ehca_err(fmr->device, "bad iova, iova=%llx fmr_page_size=%x",
 			 iova, e_fmr->fmr_page_size);
 		ret = -EINVAL;
 		goto map_phys_fmr_exit0;
@@ -878,7 +914,7 @@ int ehca_map_phys_fmr(struct ib_fmr *fmr,
 map_phys_fmr_exit0:
 	if (ret)
 		ehca_err(fmr->device, "ret=%i fmr=%p page_list=%p list_len=%x "
-			 "iova=%lx", ret, fmr, page_list, list_len, iova);
+			 "iova=%llx", ret, fmr, page_list, list_len, iova);
 	return ret;
 } /* end ehca_map_phys_fmr() */
 
@@ -897,11 +933,6 @@ int ehca_unmap_fmr(struct list_head *fmr_list)
 	/* check all FMR belong to same SHCA, and check internal flag */
 	list_for_each_entry(ib_fmr, fmr_list, list) {
 		prev_shca = shca;
-		if (!ib_fmr) {
-			ehca_gen_err("bad fmr=%p in list", ib_fmr);
-			ret = -EINVAL;
-			goto unmap_fmr_exit0;
-		}
 		shca = container_of(ib_fmr->device, struct ehca_shca,
 				    ib_device);
 		e_fmr = container_of(ib_fmr, struct ehca_mr, ib.ib_fmr);
@@ -964,8 +995,8 @@ int ehca_dealloc_fmr(struct ib_fmr *fmr)
 
 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(fmr->device, "hipz_free_mr failed, h_ret=%li e_fmr=%p "
-			 "hca_hndl=%lx fmr_hndl=%lx fmr->lkey=%x",
+		ehca_err(fmr->device, "hipz_free_mr failed, h_ret=%lli e_fmr=%p "
+			 "hca_hndl=%llx fmr_hndl=%llx fmr->lkey=%x",
 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
 			 e_fmr->ipz_mr_handle.handle, fmr->lkey);
 		ret = ehca2ib_return_code(h_ret);
@@ -983,6 +1014,10 @@ free_fmr_exit0:
 
 /*----------------------------------------------------------------------*/
 
+static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca,
+				   struct ehca_mr *e_mr,
+				   struct ehca_mr_pginfo *pginfo);
+
 int ehca_reg_mr(struct ehca_shca *shca,
 		struct ehca_mr *e_mr,
 		u64 *iova_start,
@@ -991,7 +1026,8 @@ int ehca_reg_mr(struct ehca_shca *shca,
 		struct ehca_pd *e_pd,
 		struct ehca_mr_pginfo *pginfo,
 		u32 *lkey, /*OUT*/
-		u32 *rkey) /*OUT*/
+		u32 *rkey, /*OUT*/
+		enum ehca_reg_type reg_type)
 {
 	int ret;
 	u64 h_ret;
@@ -1007,15 +1043,21 @@ int ehca_reg_mr(struct ehca_shca *shca,
 					 (u64)iova_start, size, hipz_acl,
 					 e_pd->fw_pd, &hipzout);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(&shca->ib_device, "hipz_alloc_mr failed, h_ret=%li "
-			 "hca_hndl=%lx", h_ret, shca->ipz_hca_handle.handle);
+		ehca_err(&shca->ib_device, "hipz_alloc_mr failed, h_ret=%lli "
+			 "hca_hndl=%llx", h_ret, shca->ipz_hca_handle.handle);
 		ret = ehca2ib_return_code(h_ret);
 		goto ehca_reg_mr_exit0;
 	}
 
 	e_mr->ipz_mr_handle = hipzout.handle;
 
-	ret = ehca_reg_mr_rpages(shca, e_mr, pginfo);
+	if (reg_type == EHCA_REG_BUSMAP_MR)
+		ret = ehca_reg_bmap_mr_rpages(shca, e_mr, pginfo);
+	else if (reg_type == EHCA_REG_MR)
+		ret = ehca_reg_mr_rpages(shca, e_mr, pginfo);
+	else
+		ret = -EINVAL;
+
 	if (ret)
 		goto ehca_reg_mr_exit1;
 
@@ -1033,9 +1075,9 @@ int ehca_reg_mr(struct ehca_shca *shca,
 ehca_reg_mr_exit1:
 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(&shca->ib_device, "h_ret=%li shca=%p e_mr=%p "
-			 "iova_start=%p size=%lx acl=%x e_pd=%p lkey=%x "
-			 "pginfo=%p num_kpages=%lx num_hwpages=%lx ret=%i",
+		ehca_err(&shca->ib_device, "h_ret=%lli shca=%p e_mr=%p "
+			 "iova_start=%p size=%llx acl=%x e_pd=%p lkey=%x "
+			 "pginfo=%p num_kpages=%llx num_hwpages=%llx ret=%i",
 			 h_ret, shca, e_mr, iova_start, size, acl, e_pd,
 			 hipzout.lkey, pginfo, pginfo->num_kpages,
 			 pginfo->num_hwpages, ret);
@@ -1045,8 +1087,8 @@ ehca_reg_mr_exit1:
 ehca_reg_mr_exit0:
 	if (ret)
 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p "
-			 "iova_start=%p size=%lx acl=%x e_pd=%p pginfo=%p "
-			 "num_kpages=%lx num_hwpages=%lx",
+			 "iova_start=%p size=%llx acl=%x e_pd=%p pginfo=%p "
+			 "num_kpages=%llx num_hwpages=%llx",
 			 ret, shca, e_mr, iova_start, size, acl, e_pd, pginfo,
 			 pginfo->num_kpages, pginfo->num_hwpages);
 	return ret;
@@ -1094,7 +1136,7 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
 		}
 
 		if (rnum > 1) {
-			rpage = virt_to_abs(kpage);
+			rpage = __pa(kpage);
 			if (!rpage) {
 				ehca_err(&shca->ib_device, "kpage=%p i=%x",
 					 kpage, i);
@@ -1116,8 +1158,8 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
 			 */
 			if (h_ret != H_SUCCESS) {
 				ehca_err(&shca->ib_device, "last "
-					 "hipz_reg_rpage_mr failed, h_ret=%li "
-					 "e_mr=%p i=%x hca_hndl=%lx mr_hndl=%lx"
+					 "hipz_reg_rpage_mr failed, h_ret=%lli "
+					 "e_mr=%p i=%x hca_hndl=%llx mr_hndl=%llx"
 					 " lkey=%x", h_ret, e_mr, i,
 					 shca->ipz_hca_handle.handle,
 					 e_mr->ipz_mr_handle.handle,
@@ -1128,8 +1170,8 @@ int ehca_reg_mr_rpages(struct ehca_shca *shca,
 				ret = 0;
 		} else if (h_ret != H_PAGE_REGISTERED) {
 			ehca_err(&shca->ib_device, "hipz_reg_rpage_mr failed, "
-				 "h_ret=%li e_mr=%p i=%x lkey=%x hca_hndl=%lx "
-				 "mr_hndl=%lx", h_ret, e_mr, i,
+				 "h_ret=%lli e_mr=%p i=%x lkey=%x hca_hndl=%llx "
+				 "mr_hndl=%llx", h_ret, e_mr, i,
 				 e_mr->ib.ib_mr.lkey,
 				 shca->ipz_hca_handle.handle,
 				 e_mr->ipz_mr_handle.handle);
@@ -1145,7 +1187,7 @@ ehca_reg_mr_rpages_exit1:
 ehca_reg_mr_rpages_exit0:
 	if (ret)
 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p pginfo=%p "
-			 "num_kpages=%lx num_hwpages=%lx", ret, shca, e_mr,
+			 "num_kpages=%llx num_hwpages=%llx", ret, shca, e_mr,
 			 pginfo, pginfo->num_kpages, pginfo->num_hwpages);
 	return ret;
 } /* end ehca_reg_mr_rpages() */
@@ -1184,12 +1226,12 @@ inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
 	ret = ehca_set_pagebuf(pginfo, pginfo->num_hwpages, kpage);
 	if (ret) {
 		ehca_err(&shca->ib_device, "set pagebuf failed, e_mr=%p "
-			 "pginfo=%p type=%x num_kpages=%lx num_hwpages=%lx "
+			 "pginfo=%p type=%x num_kpages=%llx num_hwpages=%llx "
 			 "kpage=%p", e_mr, pginfo, pginfo->type,
 			 pginfo->num_kpages, pginfo->num_hwpages, kpage);
 		goto ehca_rereg_mr_rereg1_exit1;
 	}
-	rpage = virt_to_abs(kpage);
+	rpage = __pa(kpage);
 	if (!rpage) {
 		ehca_err(&shca->ib_device, "kpage=%p", kpage);
 		ret = -EFAULT;
@@ -1205,13 +1247,13 @@ inline int ehca_rereg_mr_rereg1(struct ehca_shca *shca,
 		 * (MW bound or MR is shared)
 		 */
 		ehca_warn(&shca->ib_device, "hipz_h_reregister_pmr failed "
-			  "(Rereg1), h_ret=%li e_mr=%p", h_ret, e_mr);
+			  "(Rereg1), h_ret=%lli e_mr=%p", h_ret, e_mr);
 		*pginfo = pginfo_save;
 		ret = -EAGAIN;
 	} else if ((u64 *)hipzout.vaddr != iova_start) {
 		ehca_err(&shca->ib_device, "PHYP changed iova_start in "
-			 "rereg_pmr, iova_start=%p iova_start_out=%lx e_mr=%p "
-			 "mr_handle=%lx lkey=%x lkey_out=%x", iova_start,
+			 "rereg_pmr, iova_start=%p iova_start_out=%llx e_mr=%p "
+			 "mr_handle=%llx lkey=%x lkey_out=%x", iova_start,
 			 hipzout.vaddr, e_mr, e_mr->ipz_mr_handle.handle,
 			 e_mr->ib.ib_mr.lkey, hipzout.lkey);
 		ret = -EFAULT;
@@ -1235,7 +1277,7 @@ ehca_rereg_mr_rereg1_exit1:
 ehca_rereg_mr_rereg1_exit0:
 	if ( ret && (ret != -EAGAIN) )
 		ehca_err(&shca->ib_device, "ret=%i lkey=%x rkey=%x "
-			 "pginfo=%p num_kpages=%lx num_hwpages=%lx",
+			 "pginfo=%p num_kpages=%llx num_hwpages=%llx",
 			 ret, *lkey, *rkey, pginfo, pginfo->num_kpages,
 			 pginfo->num_hwpages);
 	return ret;
@@ -1263,7 +1305,7 @@ int ehca_rereg_mr(struct ehca_shca *shca,
 	    (e_mr->num_hwpages > MAX_RPAGES) ||
 	    (pginfo->num_hwpages > e_mr->num_hwpages)) {
 		ehca_dbg(&shca->ib_device, "Rereg3 case, "
-			 "pginfo->num_hwpages=%lx e_mr->num_hwpages=%x",
+			 "pginfo->num_hwpages=%llx e_mr->num_hwpages=%x",
 			 pginfo->num_hwpages, e_mr->num_hwpages);
 		rereg_1_hcall = 0;
 		rereg_3_hcall = 1;
@@ -1295,7 +1337,7 @@ int ehca_rereg_mr(struct ehca_shca *shca,
 		h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_mr);
 		if (h_ret != H_SUCCESS) {
 			ehca_err(&shca->ib_device, "hipz_free_mr failed, "
-				 "h_ret=%li e_mr=%p hca_hndl=%lx mr_hndl=%lx "
+				 "h_ret=%lli e_mr=%p hca_hndl=%llx mr_hndl=%llx "
 				 "mr->lkey=%x",
 				 h_ret, e_mr, shca->ipz_hca_handle.handle,
 				 e_mr->ipz_mr_handle.handle,
@@ -1316,7 +1358,7 @@ int ehca_rereg_mr(struct ehca_shca *shca,
 		e_mr->fmr_map_cnt = save_mr.fmr_map_cnt;
 
 		ret = ehca_reg_mr(shca, e_mr, iova_start, size, acl,
-				  e_pd, pginfo, lkey, rkey);
+				  e_pd, pginfo, lkey, rkey, EHCA_REG_MR);
 		if (ret) {
 			u32 offset = (u64)(&e_mr->flags) - (u64)e_mr;
 			memcpy(&e_mr->flags, &(save_mr.flags),
@@ -1328,8 +1370,8 @@ int ehca_rereg_mr(struct ehca_shca *shca,
 ehca_rereg_mr_exit0:
 	if (ret)
 		ehca_err(&shca->ib_device, "ret=%i shca=%p e_mr=%p "
-			 "iova_start=%p size=%lx acl=%x e_pd=%p pginfo=%p "
-			 "num_kpages=%lx lkey=%x rkey=%x rereg_1_hcall=%x "
+			 "iova_start=%p size=%llx acl=%x e_pd=%p pginfo=%p "
+			 "num_kpages=%llx lkey=%x rkey=%x rereg_1_hcall=%x "
 			 "rereg_3_hcall=%x", ret, shca, e_mr, iova_start, size,
 			 acl, e_pd, pginfo, pginfo->num_kpages, *lkey, *rkey,
 			 rereg_1_hcall, rereg_3_hcall);
@@ -1371,8 +1413,8 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
 		 * FMRs are not shared and no MW bound to FMRs
 		 */
 		ehca_err(&shca->ib_device, "hipz_reregister_pmr failed "
-			 "(Rereg1), h_ret=%li e_fmr=%p hca_hndl=%lx "
-			 "mr_hndl=%lx lkey=%x lkey_out=%x",
+			 "(Rereg1), h_ret=%lli e_fmr=%p hca_hndl=%llx "
+			 "mr_hndl=%llx lkey=%x lkey_out=%x",
 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
 			 e_fmr->ipz_mr_handle.handle,
 			 e_fmr->ib.ib_fmr.lkey, hipzout.lkey);
@@ -1383,7 +1425,7 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
 	h_ret = hipz_h_free_resource_mr(shca->ipz_hca_handle, e_fmr);
 	if (h_ret != H_SUCCESS) {
 		ehca_err(&shca->ib_device, "hipz_free_mr failed, "
-			 "h_ret=%li e_fmr=%p hca_hndl=%lx mr_hndl=%lx "
+			 "h_ret=%lli e_fmr=%p hca_hndl=%llx mr_hndl=%llx "
 			 "lkey=%x",
 			 h_ret, e_fmr, shca->ipz_hca_handle.handle,
 			 e_fmr->ipz_mr_handle.handle,
@@ -1409,7 +1451,7 @@ int ehca_unmap_one_fmr(struct ehca_shca *shca,
 	ret = ehca_reg_mr(shca, e_fmr, NULL,
 			  (e_fmr->fmr_max_pages * e_fmr->fmr_page_size),
 			  e_fmr->acl, e_pd, &pginfo, &tmp_lkey,
-			  &tmp_rkey);
+			  &tmp_rkey, EHCA_REG_MR);
 	if (ret) {
 		u32 offset = (u64)(&e_fmr->flags) - (u64)e_fmr;
 		memcpy(&e_fmr->flags, &(save_mr.flags),
@@ -1447,9 +1489,9 @@ int ehca_reg_smr(struct ehca_shca *shca,
 				    (u64)iova_start, hipz_acl, e_pd->fw_pd,
 				    &hipzout);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%li "
+		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lli "
 			 "shca=%p e_origmr=%p e_newmr=%p iova_start=%p acl=%x "
-			 "e_pd=%p hca_hndl=%lx mr_hndl=%lx lkey=%x",
+			 "e_pd=%p hca_hndl=%llx mr_hndl=%llx lkey=%x",
 			 h_ret, shca, e_origmr, e_newmr, iova_start, acl, e_pd,
 			 shca->ipz_hca_handle.handle,
 			 e_origmr->ipz_mr_handle.handle,
@@ -1478,6 +1520,90 @@ ehca_reg_smr_exit0:
 } /* end ehca_reg_smr() */
 
 /*----------------------------------------------------------------------*/
+static inline void *ehca_calc_sectbase(int top, int dir, int idx)
+{
+	unsigned long ret = idx;
+	ret |= dir << EHCA_DIR_INDEX_SHIFT;
+	ret |= top << EHCA_TOP_INDEX_SHIFT;
+	return __va(ret << SECTION_SIZE_BITS);
+}
+
+#define ehca_bmap_valid(entry) \
+	((u64)entry != (u64)EHCA_INVAL_ADDR)
+
+static u64 ehca_reg_mr_section(int top, int dir, int idx, u64 *kpage,
+			       struct ehca_shca *shca, struct ehca_mr *mr,
+			       struct ehca_mr_pginfo *pginfo)
+{
+	u64 h_ret = 0;
+	unsigned long page = 0;
+	u64 rpage = __pa(kpage);
+	int page_count;
+
+	void *sectbase = ehca_calc_sectbase(top, dir, idx);
+	if ((unsigned long)sectbase & (pginfo->hwpage_size - 1)) {
+		ehca_err(&shca->ib_device, "reg_mr_section will probably fail:"
+					   "hwpage_size does not fit to "
+					   "section start address");
+	}
+	page_count = EHCA_SECTSIZE / pginfo->hwpage_size;
+
+	while (page < page_count) {
+		u64 rnum;
+		for (rnum = 0; (rnum < MAX_RPAGES) && (page < page_count);
+		     rnum++) {
+			void *pg = sectbase + ((page++) * pginfo->hwpage_size);
+			kpage[rnum] = __pa(pg);
+		}
+
+		h_ret = hipz_h_register_rpage_mr(shca->ipz_hca_handle, mr,
+			ehca_encode_hwpage_size(pginfo->hwpage_size),
+			0, rpage, rnum);
+
+		if ((h_ret != H_SUCCESS) && (h_ret != H_PAGE_REGISTERED)) {
+			ehca_err(&shca->ib_device, "register_rpage_mr failed");
+			return h_ret;
+		}
+	}
+	return h_ret;
+}
+
+static u64 ehca_reg_mr_sections(int top, int dir, u64 *kpage,
+				struct ehca_shca *shca, struct ehca_mr *mr,
+				struct ehca_mr_pginfo *pginfo)
+{
+	u64 hret = H_SUCCESS;
+	int idx;
+
+	for (idx = 0; idx < EHCA_MAP_ENTRIES; idx++) {
+		if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]->ent[idx]))
+			continue;
+
+		hret = ehca_reg_mr_section(top, dir, idx, kpage, shca, mr,
+					   pginfo);
+		if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED))
+				return hret;
+	}
+	return hret;
+}
+
+static u64 ehca_reg_mr_dir_sections(int top, u64 *kpage, struct ehca_shca *shca,
+				    struct ehca_mr *mr,
+				    struct ehca_mr_pginfo *pginfo)
+{
+	u64 hret = H_SUCCESS;
+	int dir;
+
+	for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) {
+		if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
+			continue;
+
+		hret = ehca_reg_mr_sections(top, dir, kpage, shca, mr, pginfo);
+		if ((hret != H_SUCCESS) && (hret != H_PAGE_REGISTERED))
+				return hret;
+	}
+	return hret;
+}
 
 /* register internal max-MR to internal SHCA */
 int ehca_reg_internal_maxmr(
@@ -1495,6 +1621,11 @@ int ehca_reg_internal_maxmr(
 	u32 num_hwpages;
 	u64 hw_pgsize;
 
+	if (!ehca_bmap) {
+		ret = -EFAULT;
+		goto ehca_reg_internal_maxmr_exit0;
+	}
+
 	e_mr = ehca_mr_new();
 	if (!e_mr) {
 		ehca_err(&shca->ib_device, "out of memory");
@@ -1504,8 +1635,8 @@ int ehca_reg_internal_maxmr(
 	e_mr->flags |= EHCA_MR_FLAG_MAXMR;
 
 	/* register internal max-MR on HCA */
-	size_maxmr = (u64)high_memory - PAGE_OFFSET;
-	iova_start = (u64 *)KERNELBASE;
+	size_maxmr = ehca_mr_len;
+	iova_start = (u64 *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START));
 	ib_pbuf.addr = 0;
 	ib_pbuf.size = size_maxmr;
 	num_kpages = NUM_CHUNKS(((u64)iova_start % PAGE_SIZE) + size_maxmr,
@@ -1524,10 +1655,10 @@ int ehca_reg_internal_maxmr(
 
 	ret = ehca_reg_mr(shca, e_mr, iova_start, size_maxmr, 0, e_pd,
 			  &pginfo, &e_mr->ib.ib_mr.lkey,
-			  &e_mr->ib.ib_mr.rkey);
+			  &e_mr->ib.ib_mr.rkey, EHCA_REG_BUSMAP_MR);
 	if (ret) {
 		ehca_err(&shca->ib_device, "reg of internal max MR failed, "
-			 "e_mr=%p iova_start=%p size_maxmr=%lx num_kpages=%x "
+			 "e_mr=%p iova_start=%p size_maxmr=%llx num_kpages=%x "
 			 "num_hwpages=%x", e_mr, iova_start, size_maxmr,
 			 num_kpages, num_hwpages);
 		goto ehca_reg_internal_maxmr_exit1;
@@ -1573,8 +1704,8 @@ int ehca_reg_maxmr(struct ehca_shca *shca,
 				    (u64)iova_start, hipz_acl, e_pd->fw_pd,
 				    &hipzout);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%li "
-			 "e_origmr=%p hca_hndl=%lx mr_hndl=%lx lkey=%x",
+		ehca_err(&shca->ib_device, "hipz_reg_smr failed, h_ret=%lli "
+			 "e_origmr=%p hca_hndl=%llx mr_hndl=%llx lkey=%x",
 			 h_ret, e_origmr, shca->ipz_hca_handle.handle,
 			 e_origmr->ipz_mr_handle.handle,
 			 e_origmr->ib.ib_mr.lkey);
@@ -1651,28 +1782,28 @@ int ehca_mr_chk_buf_and_calc_size(struct ib_phys_buf *phys_buf_array,
 	/* check first buffer */
 	if (((u64)iova_start & ~PAGE_MASK) != (pbuf->addr & ~PAGE_MASK)) {
 		ehca_gen_err("iova_start/addr mismatch, iova_start=%p "
-			     "pbuf->addr=%lx pbuf->size=%lx",
+			     "pbuf->addr=%llx pbuf->size=%llx",
 			     iova_start, pbuf->addr, pbuf->size);
 		return -EINVAL;
 	}
 	if (((pbuf->addr + pbuf->size) % PAGE_SIZE) &&
 	    (num_phys_buf > 1)) {
-		ehca_gen_err("addr/size mismatch in 1st buf, pbuf->addr=%lx "
-			     "pbuf->size=%lx", pbuf->addr, pbuf->size);
+		ehca_gen_err("addr/size mismatch in 1st buf, pbuf->addr=%llx "
+			     "pbuf->size=%llx", pbuf->addr, pbuf->size);
 		return -EINVAL;
 	}
 
 	for (i = 0; i < num_phys_buf; i++) {
 		if ((i > 0) && (pbuf->addr % PAGE_SIZE)) {
-			ehca_gen_err("bad address, i=%x pbuf->addr=%lx "
-				     "pbuf->size=%lx",
+			ehca_gen_err("bad address, i=%x pbuf->addr=%llx "
+				     "pbuf->size=%llx",
 				     i, pbuf->addr, pbuf->size);
 			return -EINVAL;
 		}
 		if (((i > 0) &&	/* not 1st */
 		     (i < (num_phys_buf - 1)) &&	/* not last */
 		     (pbuf->size % PAGE_SIZE)) || (pbuf->size == 0)) {
-			ehca_gen_err("bad size, i=%x pbuf->size=%lx",
+			ehca_gen_err("bad size, i=%x pbuf->size=%llx",
 				     i, pbuf->size);
 			return -EINVAL;
 		}
@@ -1705,7 +1836,7 @@ int ehca_fmr_check_page_list(struct ehca_mr *e_fmr,
 	page = page_list;
 	for (i = 0; i < list_len; i++) {
 		if (*page % e_fmr->fmr_page_size) {
-			ehca_gen_err("bad page, i=%x *page=%lx page=%p fmr=%p "
+			ehca_gen_err("bad page, i=%x *page=%llx page=%p fmr=%p "
 				     "fmr_page_size=%x", i, *page, page, e_fmr,
 				     e_fmr->fmr_page_size);
 			return -EINVAL;
@@ -1724,62 +1855,39 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
 				  u64 *kpage)
 {
 	int ret = 0;
-	struct ib_umem_chunk *prev_chunk;
-	struct ib_umem_chunk *chunk;
 	u64 pgaddr;
-	u32 i = 0;
 	u32 j = 0;
 	int hwpages_per_kpage = PAGE_SIZE / pginfo->hwpage_size;
-
-	/* loop over desired chunk entries */
-	chunk      = pginfo->u.usr.next_chunk;
-	prev_chunk = pginfo->u.usr.next_chunk;
-	list_for_each_entry_continue(
-		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
-		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
-			pgaddr = page_to_pfn(sg_page(&chunk->page_list[i]))
-				<< PAGE_SHIFT ;
-			*kpage = phys_to_abs(pgaddr +
-					     (pginfo->next_hwpage *
-					      pginfo->hwpage_size));
-			if ( !(*kpage) ) {
-				ehca_gen_err("pgaddr=%lx "
-					     "chunk->page_list[i]=%lx "
-					     "i=%x next_hwpage=%lx",
-					     pgaddr, (u64)sg_dma_address(
-						     &chunk->page_list[i]),
-					     i, pginfo->next_hwpage);
-				return -EFAULT;
-			}
-			(pginfo->hwpage_cnt)++;
-			(pginfo->next_hwpage)++;
-			kpage++;
-			if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
-				(pginfo->kpage_cnt)++;
-				(pginfo->u.usr.next_nmap)++;
-				pginfo->next_hwpage = 0;
-				i++;
-			}
-			j++;
-			if (j >= number) break;
+	struct scatterlist **sg = &pginfo->u.usr.next_sg;
+
+	while (*sg != NULL) {
+		pgaddr = page_to_pfn(sg_page(*sg))
+			<< PAGE_SHIFT;
+		*kpage = pgaddr + (pginfo->next_hwpage *
+				   pginfo->hwpage_size);
+		if (!(*kpage)) {
+			ehca_gen_err("pgaddr=%llx "
+				     "sg_dma_address=%llx "
+				     "entry=%llx next_hwpage=%llx",
+				     pgaddr, (u64)sg_dma_address(*sg),
+				     pginfo->u.usr.next_nmap,
+				     pginfo->next_hwpage);
+			return -EFAULT;
 		}
-		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
-		    (j >= number)) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-			break;
-		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-		} else if (j >= number)
+		(pginfo->hwpage_cnt)++;
+		(pginfo->next_hwpage)++;
+		kpage++;
+		if (pginfo->next_hwpage % hwpages_per_kpage == 0) {
+			(pginfo->kpage_cnt)++;
+			(pginfo->u.usr.next_nmap)++;
+			pginfo->next_hwpage = 0;
+			*sg = sg_next(*sg);
+		}
+		j++;
+		if (j >= number)
 			break;
-		else
-			prev_chunk = chunk;
 	}
-	pginfo->u.usr.next_chunk =
-		list_prepare_entry(prev_chunk,
-				   (&(pginfo->u.usr.region->chunk_list)),
-				   list);
+
 	return ret;
 }
 
@@ -1787,20 +1895,19 @@ static int ehca_set_pagebuf_user1(struct ehca_mr_pginfo *pginfo,
  * check given pages for contiguous layout
  * last page addr is returned in prev_pgaddr for further check
  */
-static int ehca_check_kpages_per_ate(struct scatterlist *page_list,
-				     int start_idx, int end_idx,
+static int ehca_check_kpages_per_ate(struct scatterlist **sg,
+				     int num_pages,
 				     u64 *prev_pgaddr)
 {
-	int t;
-	for (t = start_idx; t <= end_idx; t++) {
-		u64 pgaddr = page_to_pfn(sg_page(&page_list[t])) << PAGE_SHIFT;
+	for (; *sg && num_pages > 0; *sg = sg_next(*sg), num_pages--) {
+		u64 pgaddr = page_to_pfn(sg_page(*sg)) << PAGE_SHIFT;
 		if (ehca_debug_level >= 3)
-			ehca_gen_dbg("chunk_page=%lx value=%016lx", pgaddr,
-				     *(u64 *)abs_to_virt(phys_to_abs(pgaddr)));
+			ehca_gen_dbg("chunk_page=%llx value=%016llx", pgaddr,
+				     *(u64 *)__va(pgaddr));
 		if (pgaddr - PAGE_SIZE != *prev_pgaddr) {
-			ehca_gen_err("uncontiguous page found pgaddr=%lx "
-				     "prev_pgaddr=%lx page_list_i=%x",
-				     pgaddr, *prev_pgaddr, t);
+			ehca_gen_err("uncontiguous page found pgaddr=%llx "
+				     "prev_pgaddr=%llx entries_left_in_hwpage=%x",
+				     pgaddr, *prev_pgaddr, num_pages);
 			return -EINVAL;
 		}
 		*prev_pgaddr = pgaddr;
@@ -1814,113 +1921,80 @@ static int ehca_set_pagebuf_user2(struct ehca_mr_pginfo *pginfo,
 				  u64 *kpage)
 {
 	int ret = 0;
-	struct ib_umem_chunk *prev_chunk;
-	struct ib_umem_chunk *chunk;
 	u64 pgaddr, prev_pgaddr;
-	u32 i = 0;
 	u32 j = 0;
 	int kpages_per_hwpage = pginfo->hwpage_size / PAGE_SIZE;
 	int nr_kpages = kpages_per_hwpage;
+	struct scatterlist **sg = &pginfo->u.usr.next_sg;
+
+	while (*sg != NULL) {
 
-	/* loop over desired chunk entries */
-	chunk      = pginfo->u.usr.next_chunk;
-	prev_chunk = pginfo->u.usr.next_chunk;
-	list_for_each_entry_continue(
-		chunk, (&(pginfo->u.usr.region->chunk_list)), list) {
-		for (i = pginfo->u.usr.next_nmap; i < chunk->nmap; ) {
-			if (nr_kpages == kpages_per_hwpage) {
-				pgaddr = ( page_to_pfn(sg_page(&chunk->page_list[i]))
-					   << PAGE_SHIFT );
-				*kpage = phys_to_abs(pgaddr);
-				if ( !(*kpage) ) {
-					ehca_gen_err("pgaddr=%lx i=%x",
-						     pgaddr, i);
+		if (nr_kpages == kpages_per_hwpage) {
+			pgaddr = (page_to_pfn(sg_page(*sg))
+				   << PAGE_SHIFT);
+			*kpage = pgaddr;
+			if (!(*kpage)) {
+				ehca_gen_err("pgaddr=%llx entry=%llx",
+					     pgaddr, pginfo->u.usr.next_nmap);
+				ret = -EFAULT;
+				return ret;
+			}
+			/*
+			 * The first page in a hwpage must be aligned;
+			 * the first MR page is exempt from this rule.
+			 */
+			if (pgaddr & (pginfo->hwpage_size - 1)) {
+				if (pginfo->hwpage_cnt) {
+					ehca_gen_err(
+						"invalid alignment "
+						"pgaddr=%llx entry=%llx "
+						"mr_pgsize=%llx",
+						pgaddr, pginfo->u.usr.next_nmap,
+						pginfo->hwpage_size);
 					ret = -EFAULT;
 					return ret;
 				}
-				/*
-				 * The first page in a hwpage must be aligned;
-				 * the first MR page is exempt from this rule.
-				 */
-				if (pgaddr & (pginfo->hwpage_size - 1)) {
-					if (pginfo->hwpage_cnt) {
-						ehca_gen_err(
-							"invalid alignment "
-							"pgaddr=%lx i=%x "
-							"mr_pgsize=%lx",
-							pgaddr, i,
-							pginfo->hwpage_size);
-						ret = -EFAULT;
-						return ret;
-					}
-					/* first MR page */
-					pginfo->kpage_cnt =
-						(pgaddr &
-						 (pginfo->hwpage_size - 1)) >>
-						PAGE_SHIFT;
-					nr_kpages -= pginfo->kpage_cnt;
-					*kpage = phys_to_abs(
-						pgaddr &
-						~(pginfo->hwpage_size - 1));
-				}
-				if (ehca_debug_level >= 3) {
-					u64 val = *(u64 *)abs_to_virt(
-						phys_to_abs(pgaddr));
-					ehca_gen_dbg("kpage=%lx chunk_page=%lx "
-						     "value=%016lx",
-						     *kpage, pgaddr, val);
-				}
-				prev_pgaddr = pgaddr;
-				i++;
-				pginfo->kpage_cnt++;
-				pginfo->u.usr.next_nmap++;
-				nr_kpages--;
-				if (!nr_kpages)
-					goto next_kpage;
-				continue;
+				/* first MR page */
+				pginfo->kpage_cnt =
+					(pgaddr &
+					 (pginfo->hwpage_size - 1)) >>
+					PAGE_SHIFT;
+				nr_kpages -= pginfo->kpage_cnt;
+				*kpage = pgaddr &
+					 ~(pginfo->hwpage_size - 1);
 			}
-			if (i + nr_kpages > chunk->nmap) {
-				ret = ehca_check_kpages_per_ate(
-					chunk->page_list, i,
-					chunk->nmap - 1, &prev_pgaddr);
-				if (ret) return ret;
-				pginfo->kpage_cnt += chunk->nmap - i;
-				pginfo->u.usr.next_nmap += chunk->nmap - i;
-				nr_kpages -= chunk->nmap - i;
-				break;
+			if (ehca_debug_level >= 3) {
+				u64 val = *(u64 *)__va(pgaddr);
+				ehca_gen_dbg("kpage=%llx page=%llx "
+					     "value=%016llx",
+					     *kpage, pgaddr, val);
 			}
+			prev_pgaddr = pgaddr;
+			*sg = sg_next(*sg);
+			pginfo->kpage_cnt++;
+			pginfo->u.usr.next_nmap++;
+			nr_kpages--;
+			if (!nr_kpages)
+				goto next_kpage;
+			continue;
+		}
+
+		ret = ehca_check_kpages_per_ate(sg, nr_kpages,
+						&prev_pgaddr);
+		if (ret)
+			return ret;
+		pginfo->kpage_cnt += nr_kpages;
+		pginfo->u.usr.next_nmap += nr_kpages;
 
-			ret = ehca_check_kpages_per_ate(chunk->page_list, i,
-							i + nr_kpages - 1,
-							&prev_pgaddr);
-			if (ret) return ret;
-			i += nr_kpages;
-			pginfo->kpage_cnt += nr_kpages;
-			pginfo->u.usr.next_nmap += nr_kpages;
 next_kpage:
-			nr_kpages = kpages_per_hwpage;
-			(pginfo->hwpage_cnt)++;
-			kpage++;
-			j++;
-			if (j >= number) break;
-		}
-		if ((pginfo->u.usr.next_nmap >= chunk->nmap) &&
-		    (j >= number)) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-			break;
-		} else if (pginfo->u.usr.next_nmap >= chunk->nmap) {
-			pginfo->u.usr.next_nmap = 0;
-			prev_chunk = chunk;
-		} else if (j >= number)
+		nr_kpages = kpages_per_hwpage;
+		(pginfo->hwpage_cnt)++;
+		kpage++;
+		j++;
+		if (j >= number)
 			break;
-		else
-			prev_chunk = chunk;
 	}
-	pginfo->u.usr.next_chunk =
-		list_prepare_entry(prev_chunk,
-				   (&(pginfo->u.usr.region->chunk_list)),
-				   list);
+
 	return ret;
 }
 
@@ -1944,21 +2018,20 @@ static int ehca_set_pagebuf_phys(struct ehca_mr_pginfo *pginfo,
 			if ((pginfo->kpage_cnt >= pginfo->num_kpages) ||
 			    (pginfo->hwpage_cnt >= pginfo->num_hwpages)) {
 				ehca_gen_err("kpage_cnt >= num_kpages, "
-					     "kpage_cnt=%lx num_kpages=%lx "
-					     "hwpage_cnt=%lx "
-					     "num_hwpages=%lx i=%x",
+					     "kpage_cnt=%llx num_kpages=%llx "
+					     "hwpage_cnt=%llx "
+					     "num_hwpages=%llx i=%x",
 					     pginfo->kpage_cnt,
 					     pginfo->num_kpages,
 					     pginfo->hwpage_cnt,
 					     pginfo->num_hwpages, i);
 				return -EFAULT;
 			}
-			*kpage = phys_to_abs(
-				(pbuf->addr & ~(pginfo->hwpage_size - 1)) +
-				(pginfo->next_hwpage * pginfo->hwpage_size));
+			*kpage = (pbuf->addr & ~(pginfo->hwpage_size - 1)) +
+				 (pginfo->next_hwpage * pginfo->hwpage_size);
 			if ( !(*kpage) && pbuf->addr ) {
-				ehca_gen_err("pbuf->addr=%lx pbuf->size=%lx "
-					     "next_hwpage=%lx", pbuf->addr,
+				ehca_gen_err("pbuf->addr=%llx pbuf->size=%llx "
+					     "next_hwpage=%llx", pbuf->addr,
 					     pbuf->size, pginfo->next_hwpage);
 				return -EFAULT;
 			}
@@ -1993,11 +2066,11 @@ static int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
 	/* loop over desired page_list entries */
 	fmrlist = pginfo->u.fmr.page_list + pginfo->u.fmr.next_listelem;
 	for (i = 0; i < number; i++) {
-		*kpage = phys_to_abs((*fmrlist & ~(pginfo->hwpage_size - 1)) +
-				     pginfo->next_hwpage * pginfo->hwpage_size);
+		*kpage = (*fmrlist & ~(pginfo->hwpage_size - 1)) +
+			   pginfo->next_hwpage * pginfo->hwpage_size;
 		if ( !(*kpage) ) {
-			ehca_gen_err("*fmrlist=%lx fmrlist=%p "
-				     "next_listelem=%lx next_hwpage=%lx",
+			ehca_gen_err("*fmrlist=%llx fmrlist=%p "
+				     "next_listelem=%llx next_hwpage=%llx",
 				     *fmrlist, fmrlist,
 				     pginfo->u.fmr.next_listelem,
 				     pginfo->next_hwpage);
@@ -2021,11 +2094,10 @@ static int ehca_set_pagebuf_fmr(struct ehca_mr_pginfo *pginfo,
 			u64 prev = *kpage;
 			/* check if adrs are contiguous */
 			for (j = 1; j < cnt_per_hwpage; j++) {
-				u64 p = phys_to_abs(fmrlist[j] &
-						    ~(pginfo->hwpage_size - 1));
+				u64 p = fmrlist[j] & ~(pginfo->hwpage_size - 1);
 				if (prev + pginfo->u.fmr.fmr_pgsize != p) {
 					ehca_gen_err("uncontiguous fmr pages "
-						     "found prev=%lx p=%lx "
+						     "found prev=%llx p=%llx "
 						     "idx=%x", prev, p, i + j);
 					return -EINVAL;
 				}
@@ -2077,8 +2149,8 @@ int ehca_mr_is_maxmr(u64 size,
 		     u64 *iova_start)
 {
 	/* a MR is treated as max-MR only if it fits following: */
-	if ((size == ((u64)high_memory - PAGE_OFFSET)) &&
-	    (iova_start == (void *)KERNELBASE)) {
+	if ((size == ehca_mr_len) &&
+	    (iova_start == (void *)ehca_map_vaddr((void *)(KERNELBASE + PHYSICAL_START)))) {
 		ehca_gen_dbg("this is a max-MR");
 		return 1;
 	} else
@@ -2184,3 +2256,338 @@ void ehca_cleanup_mrmw_cache(void)
 	if (mw_cache)
 		kmem_cache_destroy(mw_cache);
 }
+
+static inline int ehca_init_top_bmap(struct ehca_top_bmap *ehca_top_bmap,
+				     int dir)
+{
+	if (!ehca_bmap_valid(ehca_top_bmap->dir[dir])) {
+		ehca_top_bmap->dir[dir] =
+			kmalloc(sizeof(struct ehca_dir_bmap), GFP_KERNEL);
+		if (!ehca_top_bmap->dir[dir])
+			return -ENOMEM;
+		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
+		memset(ehca_top_bmap->dir[dir], 0xFF, EHCA_ENT_MAP_SIZE);
+	}
+	return 0;
+}
+
+static inline int ehca_init_bmap(struct ehca_bmap *ehca_bmap, int top, int dir)
+{
+	if (!ehca_bmap_valid(ehca_bmap->top[top])) {
+		ehca_bmap->top[top] =
+			kmalloc(sizeof(struct ehca_top_bmap), GFP_KERNEL);
+		if (!ehca_bmap->top[top])
+			return -ENOMEM;
+		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
+		memset(ehca_bmap->top[top], 0xFF, EHCA_DIR_MAP_SIZE);
+	}
+	return ehca_init_top_bmap(ehca_bmap->top[top], dir);
+}
+
+static inline int ehca_calc_index(unsigned long i, unsigned long s)
+{
+	return (i >> s) & EHCA_INDEX_MASK;
+}
+
+void ehca_destroy_busmap(void)
+{
+	int top, dir;
+
+	if (!ehca_bmap)
+		return;
+
+	for (top = 0; top < EHCA_MAP_ENTRIES; top++) {
+		if (!ehca_bmap_valid(ehca_bmap->top[top]))
+			continue;
+		for (dir = 0; dir < EHCA_MAP_ENTRIES; dir++) {
+			if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
+				continue;
+
+			kfree(ehca_bmap->top[top]->dir[dir]);
+		}
+
+		kfree(ehca_bmap->top[top]);
+	}
+
+	kfree(ehca_bmap);
+	ehca_bmap = NULL;
+}
+
+static int ehca_update_busmap(unsigned long pfn, unsigned long nr_pages)
+{
+	unsigned long i, start_section, end_section;
+	int top, dir, idx;
+
+	if (!nr_pages)
+		return 0;
+
+	if (!ehca_bmap) {
+		ehca_bmap = kmalloc(sizeof(struct ehca_bmap), GFP_KERNEL);
+		if (!ehca_bmap)
+			return -ENOMEM;
+		/* Set map block to 0xFF according to EHCA_INVAL_ADDR */
+		memset(ehca_bmap, 0xFF, EHCA_TOP_MAP_SIZE);
+	}
+
+	start_section = (pfn * PAGE_SIZE) / EHCA_SECTSIZE;
+	end_section = ((pfn + nr_pages) * PAGE_SIZE) / EHCA_SECTSIZE;
+	for (i = start_section; i < end_section; i++) {
+		int ret;
+		top = ehca_calc_index(i, EHCA_TOP_INDEX_SHIFT);
+		dir = ehca_calc_index(i, EHCA_DIR_INDEX_SHIFT);
+		idx = i & EHCA_INDEX_MASK;
+
+		ret = ehca_init_bmap(ehca_bmap, top, dir);
+		if (ret) {
+			ehca_destroy_busmap();
+			return ret;
+		}
+		ehca_bmap->top[top]->dir[dir]->ent[idx] = ehca_mr_len;
+		ehca_mr_len += EHCA_SECTSIZE;
+	}
+	return 0;
+}
+
+static int ehca_is_hugepage(unsigned long pfn)
+{
+	int page_order;
+
+	if (pfn & EHCA_HUGEPAGE_PFN_MASK)
+		return 0;
+
+	page_order = compound_order(pfn_to_page(pfn));
+	if (page_order + PAGE_SHIFT != EHCA_HUGEPAGESHIFT)
+		return 0;
+
+	return 1;
+}
+
+static int ehca_create_busmap_callback(unsigned long initial_pfn,
+				       unsigned long total_nr_pages, void *arg)
+{
+	int ret;
+	unsigned long pfn, start_pfn, end_pfn, nr_pages;
+
+	if ((total_nr_pages * PAGE_SIZE) < EHCA_HUGEPAGE_SIZE)
+		return ehca_update_busmap(initial_pfn, total_nr_pages);
+
+	/* Given chunk is >= 16GB -> check for hugepages */
+	start_pfn = initial_pfn;
+	end_pfn = initial_pfn + total_nr_pages;
+	pfn = start_pfn;
+
+	while (pfn < end_pfn) {
+		if (ehca_is_hugepage(pfn)) {
+			/* Add mem found in front of the hugepage */
+			nr_pages = pfn - start_pfn;
+			ret = ehca_update_busmap(start_pfn, nr_pages);
+			if (ret)
+				return ret;
+			/* Skip the hugepage */
+			pfn += (EHCA_HUGEPAGE_SIZE / PAGE_SIZE);
+			start_pfn = pfn;
+		} else
+			pfn += (EHCA_SECTSIZE / PAGE_SIZE);
+	}
+
+	/* Add mem found behind the hugepage(s)  */
+	nr_pages = pfn - start_pfn;
+	return ehca_update_busmap(start_pfn, nr_pages);
+}
+
+int ehca_create_busmap(void)
+{
+	int ret;
+
+	ehca_mr_len = 0;
+	ret = walk_system_ram_range(0, 1ULL << MAX_PHYSMEM_BITS, NULL,
+				   ehca_create_busmap_callback);
+	return ret;
+}
+
+static int ehca_reg_bmap_mr_rpages(struct ehca_shca *shca,
+				   struct ehca_mr *e_mr,
+				   struct ehca_mr_pginfo *pginfo)
+{
+	int top;
+	u64 hret, *kpage;
+
+	kpage = ehca_alloc_fw_ctrlblock(GFP_KERNEL);
+	if (!kpage) {
+		ehca_err(&shca->ib_device, "kpage alloc failed");
+		return -ENOMEM;
+	}
+	for (top = 0; top < EHCA_MAP_ENTRIES; top++) {
+		if (!ehca_bmap_valid(ehca_bmap->top[top]))
+			continue;
+		hret = ehca_reg_mr_dir_sections(top, kpage, shca, e_mr, pginfo);
+		if ((hret != H_PAGE_REGISTERED) && (hret != H_SUCCESS))
+			break;
+	}
+
+	ehca_free_fw_ctrlblock(kpage);
+
+	if (hret == H_SUCCESS)
+		return 0; /* Everything is fine */
+	else {
+		ehca_err(&shca->ib_device, "ehca_reg_bmap_mr_rpages failed, "
+				 "h_ret=%lli e_mr=%p top=%x lkey=%x "
+				 "hca_hndl=%llx mr_hndl=%llx", hret, e_mr, top,
+				 e_mr->ib.ib_mr.lkey,
+				 shca->ipz_hca_handle.handle,
+				 e_mr->ipz_mr_handle.handle);
+		return ehca2ib_return_code(hret);
+	}
+}
+
+static u64 ehca_map_vaddr(void *caddr)
+{
+	int top, dir, idx;
+	unsigned long abs_addr, offset;
+	u64 entry;
+
+	if (!ehca_bmap)
+		return EHCA_INVAL_ADDR;
+
+	abs_addr = __pa(caddr);
+	top = ehca_calc_index(abs_addr, EHCA_TOP_INDEX_SHIFT + EHCA_SECTSHIFT);
+	if (!ehca_bmap_valid(ehca_bmap->top[top]))
+		return EHCA_INVAL_ADDR;
+
+	dir = ehca_calc_index(abs_addr, EHCA_DIR_INDEX_SHIFT + EHCA_SECTSHIFT);
+	if (!ehca_bmap_valid(ehca_bmap->top[top]->dir[dir]))
+		return EHCA_INVAL_ADDR;
+
+	idx = ehca_calc_index(abs_addr, EHCA_SECTSHIFT);
+
+	entry = ehca_bmap->top[top]->dir[dir]->ent[idx];
+	if (ehca_bmap_valid(entry)) {
+		offset = (unsigned long)caddr & (EHCA_SECTSIZE - 1);
+		return entry | offset;
+	} else
+		return EHCA_INVAL_ADDR;
+}
+
+static int ehca_dma_mapping_error(struct ib_device *dev, u64 dma_addr)
+{
+	return dma_addr == EHCA_INVAL_ADDR;
+}
+
+static u64 ehca_dma_map_single(struct ib_device *dev, void *cpu_addr,
+			       size_t size, enum dma_data_direction direction)
+{
+	if (cpu_addr)
+		return ehca_map_vaddr(cpu_addr);
+	else
+		return EHCA_INVAL_ADDR;
+}
+
+static void ehca_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size,
+				  enum dma_data_direction direction)
+{
+	/* This is only a stub; nothing to be done here */
+}
+
+static u64 ehca_dma_map_page(struct ib_device *dev, struct page *page,
+			     unsigned long offset, size_t size,
+			     enum dma_data_direction direction)
+{
+	u64 addr;
+
+	if (offset + size > PAGE_SIZE)
+		return EHCA_INVAL_ADDR;
+
+	addr = ehca_map_vaddr(page_address(page));
+	if (!ehca_dma_mapping_error(dev, addr))
+		addr += offset;
+
+	return addr;
+}
+
+static void ehca_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size,
+				enum dma_data_direction direction)
+{
+	/* This is only a stub; nothing to be done here */
+}
+
+static int ehca_dma_map_sg(struct ib_device *dev, struct scatterlist *sgl,
+			   int nents, enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	int i;
+
+	for_each_sg(sgl, sg, nents, i) {
+		u64 addr;
+		addr = ehca_map_vaddr(sg_virt(sg));
+		if (ehca_dma_mapping_error(dev, addr))
+			return 0;
+
+		sg->dma_address = addr;
+		sg->dma_length = sg->length;
+	}
+	return nents;
+}
+
+static void ehca_dma_unmap_sg(struct ib_device *dev, struct scatterlist *sg,
+			      int nents, enum dma_data_direction direction)
+{
+	/* This is only a stub; nothing to be done here */
+}
+
+static void ehca_dma_sync_single_for_cpu(struct ib_device *dev, u64 addr,
+					 size_t size,
+					 enum dma_data_direction dir)
+{
+	dma_sync_single_for_cpu(dev->dma_device, addr, size, dir);
+}
+
+static void ehca_dma_sync_single_for_device(struct ib_device *dev, u64 addr,
+					    size_t size,
+					    enum dma_data_direction dir)
+{
+	dma_sync_single_for_device(dev->dma_device, addr, size, dir);
+}
+
+static void *ehca_dma_alloc_coherent(struct ib_device *dev, size_t size,
+				     u64 *dma_handle, gfp_t flag)
+{
+	struct page *p;
+	void *addr = NULL;
+	u64 dma_addr;
+
+	p = alloc_pages(flag, get_order(size));
+	if (p) {
+		addr = page_address(p);
+		dma_addr = ehca_map_vaddr(addr);
+		if (ehca_dma_mapping_error(dev, dma_addr)) {
+			free_pages((unsigned long)addr,	get_order(size));
+			return NULL;
+		}
+		if (dma_handle)
+			*dma_handle = dma_addr;
+		return addr;
+	}
+	return NULL;
+}
+
+static void ehca_dma_free_coherent(struct ib_device *dev, size_t size,
+				   void *cpu_addr, u64 dma_handle)
+{
+	if (cpu_addr && size)
+		free_pages((unsigned long)cpu_addr, get_order(size));
+}
+
+
+struct ib_dma_mapping_ops ehca_dma_mapping_ops = {
+	.mapping_error          = ehca_dma_mapping_error,
+	.map_single             = ehca_dma_map_single,
+	.unmap_single           = ehca_dma_unmap_single,
+	.map_page               = ehca_dma_map_page,
+	.unmap_page             = ehca_dma_unmap_page,
+	.map_sg                 = ehca_dma_map_sg,
+	.unmap_sg               = ehca_dma_unmap_sg,
+	.sync_single_for_cpu    = ehca_dma_sync_single_for_cpu,
+	.sync_single_for_device = ehca_dma_sync_single_for_device,
+	.alloc_coherent         = ehca_dma_alloc_coherent,
+	.free_coherent          = ehca_dma_free_coherent,
+};
diff --git a/drivers/infiniband/hw/ehca/ehca_mrmw.h b/drivers/infiniband/hw/ehca/ehca_mrmw.h
index bc8f4e31c12..50d8b51306d 100644
--- a/drivers/infiniband/hw/ehca/ehca_mrmw.h
+++ b/drivers/infiniband/hw/ehca/ehca_mrmw.h
@@ -42,6 +42,11 @@
 #ifndef _EHCA_MRMW_H_
 #define _EHCA_MRMW_H_
 
+enum ehca_reg_type {
+	EHCA_REG_MR,
+	EHCA_REG_BUSMAP_MR
+};
+
 int ehca_reg_mr(struct ehca_shca *shca,
 		struct ehca_mr *e_mr,
 		u64 *iova_start,
@@ -50,7 +55,8 @@ int ehca_reg_mr(struct ehca_shca *shca,
 		struct ehca_pd *e_pd,
 		struct ehca_mr_pginfo *pginfo,
 		u32 *lkey,
-		u32 *rkey);
+		u32 *rkey,
+		enum ehca_reg_type reg_type);
 
 int ehca_reg_mr_rpages(struct ehca_shca *shca,
 		       struct ehca_mr *e_mr,
@@ -118,4 +124,9 @@ void ehca_mrmw_reverse_map_acl(const u32 *hipz_acl,
 
 void ehca_mr_deletenew(struct ehca_mr *mr);
 
+int ehca_create_busmap(void);
+
+void ehca_destroy_busmap(void);
+
+extern struct ib_dma_mapping_ops ehca_dma_mapping_ops;
 #endif  /*_EHCA_MRMW_H_*/
diff --git a/drivers/infiniband/hw/ehca/ehca_pd.c b/drivers/infiniband/hw/ehca/ehca_pd.c
index 2fe554855fa..351577a6670 100644
--- a/drivers/infiniband/hw/ehca/ehca_pd.c
+++ b/drivers/infiniband/hw/ehca/ehca_pd.c
@@ -38,6 +38,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_tools.h"
 #include "ehca_iverbs.h"
 
diff --git a/drivers/infiniband/hw/ehca/ehca_qes.h b/drivers/infiniband/hw/ehca/ehca_qes.h
index 818803057eb..90c4efa6758 100644
--- a/drivers/infiniband/hw/ehca/ehca_qes.h
+++ b/drivers/infiniband/hw/ehca/ehca_qes.h
@@ -46,7 +46,7 @@
 
 #include "ehca_tools.h"
 
-/* virtual scatter gather entry to specify remote adresses with length */
+/* virtual scatter gather entry to specify remote addresses with length */
 struct ehca_vsgentry {
 	u64 vaddr;
 	u32 lkey;
@@ -148,7 +148,7 @@ struct ehca_wqe {
 	u32 immediate_data;
 	union {
 		struct {
-			u64 remote_virtual_adress;
+			u64 remote_virtual_address;
 			u32 rkey;
 			u32 reserved;
 			u64 atomic_1st_op_dma_len;
@@ -213,6 +213,7 @@ struct ehca_wqe {
 #define WC_STATUS_ERROR_BIT 0x80000000
 #define WC_STATUS_REMOTE_ERROR_FLAGS 0x0000F800
 #define WC_STATUS_PURGE_BIT 0x10
+#define WC_SEND_RECEIVE_BIT 0x80
 
 struct ehca_cqe {
 	u64 work_request_id;
diff --git a/drivers/infiniband/hw/ehca/ehca_qp.c b/drivers/infiniband/hw/ehca/ehca_qp.c
index 3f59587338e..2e89356c46f 100644
--- a/drivers/infiniband/hw/ehca/ehca_qp.c
+++ b/drivers/infiniband/hw/ehca/ehca_qp.c
@@ -43,6 +43,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_classes.h"
 #include "ehca_tools.h"
 #include "ehca_qes.h"
@@ -55,9 +57,7 @@ static struct kmem_cache *qp_cache;
 /*
  * attributes not supported by query qp
  */
-#define QP_ATTR_QUERY_NOT_SUPPORTED (IB_QP_MAX_DEST_RD_ATOMIC | \
-				     IB_QP_MAX_QP_RD_ATOMIC   | \
-				     IB_QP_ACCESS_FLAGS       | \
+#define QP_ATTR_QUERY_NOT_SUPPORTED (IB_QP_ACCESS_FLAGS       | \
 				     IB_QP_EN_SQD_ASYNC_NOTIFY)
 
 /*
@@ -251,7 +251,7 @@ static inline int ibqptype2servicetype(enum ib_qp_type ibqptype)
 		return ST_UD;
 	case IB_QPT_RAW_IPV6:
 		return -EINVAL;
-	case IB_QPT_RAW_ETY:
+	case IB_QPT_RAW_ETHERTYPE:
 		return -EINVAL;
 	default:
 		ehca_gen_err("Invalid ibqptype=%x", ibqptype);
@@ -321,7 +321,7 @@ static inline int init_qp_queue(struct ehca_shca *shca,
 			ret = -EINVAL;
 			goto init_qp_queue1;
 		}
-		rpage = virt_to_abs(vpage);
+		rpage = __pa(vpage);
 
 		h_ret = hipz_h_register_rpage_qp(ipz_hca_handle,
 						 my_qp->ipz_qp_handle,
@@ -331,7 +331,7 @@ static inline int init_qp_queue(struct ehca_shca *shca,
 		if (cnt == (nr_q_pages - 1)) {	/* last page! */
 			if (h_ret != expected_hret) {
 				ehca_err(ib_dev, "hipz_qp_register_rpage() "
-					 "h_ret=%li", h_ret);
+					 "h_ret=%lli", h_ret);
 				ret = ehca2ib_return_code(h_ret);
 				goto init_qp_queue1;
 			}
@@ -345,7 +345,7 @@ static inline int init_qp_queue(struct ehca_shca *shca,
 		} else {
 			if (h_ret != H_PAGE_REGISTERED) {
 				ehca_err(ib_dev, "hipz_qp_register_rpage() "
-					 "h_ret=%li", h_ret);
+					 "h_ret=%lli", h_ret);
 				ret = ehca2ib_return_code(h_ret);
 				goto init_qp_queue1;
 			}
@@ -396,6 +396,54 @@ static void ehca_determine_small_queue(struct ehca_alloc_queue_parms *queue,
 	queue->is_small = (queue->page_size != 0);
 }
 
+/* needs to be called with cq->spinlock held */
+void ehca_add_to_err_list(struct ehca_qp *qp, int on_sq)
+{
+	struct list_head *list, *node;
+
+	/* TODO: support low latency QPs */
+	if (qp->ext_type == EQPT_LLQP)
+		return;
+
+	if (on_sq) {
+		list = &qp->send_cq->sqp_err_list;
+		node = &qp->sq_err_node;
+	} else {
+		list = &qp->recv_cq->rqp_err_list;
+		node = &qp->rq_err_node;
+	}
+
+	if (list_empty(node))
+		list_add_tail(node, list);
+
+	return;
+}
+
+static void del_from_err_list(struct ehca_cq *cq, struct list_head *node)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&cq->spinlock, flags);
+
+	if (!list_empty(node))
+		list_del_init(node);
+
+	spin_unlock_irqrestore(&cq->spinlock, flags);
+}
+
+static void reset_queue_map(struct ehca_queue_map *qmap)
+{
+	int i;
+
+	qmap->tail = qmap->entries - 1;
+	qmap->left_to_poll = 0;
+	qmap->next_wqe_idx = 0;
+	for (i = 0; i < qmap->entries; i++) {
+		qmap->map[i].reported = 1;
+		qmap->map[i].cqe_req = 0;
+	}
+}
+
 /*
  * Create an ib_qp struct that is either a QP or an SRQ, depending on
  * the value of the is_srq parameter. If init_attr and srq_init_attr share
@@ -407,13 +455,13 @@ static struct ehca_qp *internal_create_qp(
 	struct ib_srq_init_attr *srq_init_attr,
 	struct ib_udata *udata, int is_srq)
 {
-	struct ehca_qp *my_qp;
+	struct ehca_qp *my_qp, *my_srq = NULL;
 	struct ehca_pd *my_pd = container_of(pd, struct ehca_pd, ib_pd);
 	struct ehca_shca *shca = container_of(pd->device, struct ehca_shca,
 					      ib_device);
 	struct ib_ucontext *context = NULL;
 	u64 h_ret;
-	int is_llqp = 0, has_srq = 0;
+	int is_llqp = 0, has_srq = 0, is_user = 0;
 	int qp_type, max_send_sge, max_recv_sge, ret;
 
 	/* h_call's out parameters */
@@ -421,9 +469,9 @@ static struct ehca_qp *internal_create_qp(
 	u32 swqe_size = 0, rwqe_size = 0, ib_qp_num;
 	unsigned long flags;
 
-	if (!atomic_add_unless(&shca->num_qps, 1, ehca_max_qp)) {
+	if (!atomic_add_unless(&shca->num_qps, 1, shca->max_num_qps)) {
 		ehca_err(pd->device, "Unable to create QP, max number of %i "
-			 "QPs reached.", ehca_max_qp);
+			 "QPs reached.", shca->max_num_qps);
 		ehca_err(pd->device, "To increase the maximum number of QPs "
 			 "use the number_of_qps module parameter.\n");
 		return ERR_PTR(-ENOSPC);
@@ -456,8 +504,13 @@ static struct ehca_qp *internal_create_qp(
 
 	/* handle SRQ base QPs */
 	if (init_attr->srq) {
-		struct ehca_qp *my_srq =
-			container_of(init_attr->srq, struct ehca_qp, ib_srq);
+		my_srq = container_of(init_attr->srq, struct ehca_qp, ib_srq);
+
+		if (qp_type == IB_QPT_UC) {
+			ehca_err(pd->device, "UC with SRQ not supported");
+			atomic_dec(&shca->num_qps);
+			return ERR_PTR(-EINVAL);
+		}
 
 		has_srq = 1;
 		parms.ext_type = EQPT_SRQBASE;
@@ -556,9 +609,6 @@ static struct ehca_qp *internal_create_qp(
 		}
 	}
 
-	if (pd->uobject && udata)
-		context = pd->uobject->context;
-
 	my_qp = kmem_cache_zalloc(qp_cache, GFP_KERNEL);
 	if (!my_qp) {
 		ehca_err(pd->device, "pd=%p not enough memory to alloc qp", pd);
@@ -566,6 +616,11 @@ static struct ehca_qp *internal_create_qp(
 		return ERR_PTR(-ENOMEM);
 	}
 
+	if (pd->uobject && udata) {
+		is_user = 1;
+		context = pd->uobject->context;
+	}
+
 	atomic_set(&my_qp->nr_events, 0);
 	init_waitqueue_head(&my_qp->wait_completion);
 	spin_lock_init(&my_qp->spinlock_s);
@@ -581,30 +636,26 @@ static struct ehca_qp *internal_create_qp(
 		my_qp->send_cq =
 			container_of(init_attr->send_cq, struct ehca_cq, ib_cq);
 
-	do {
-		if (!idr_pre_get(&ehca_qp_idr, GFP_KERNEL)) {
-			ret = -ENOMEM;
-			ehca_err(pd->device, "Can't reserve idr resources.");
-			goto create_qp_exit0;
-		}
+	idr_preload(GFP_KERNEL);
+	write_lock_irqsave(&ehca_qp_idr_lock, flags);
 
-		write_lock_irqsave(&ehca_qp_idr_lock, flags);
-		ret = idr_get_new(&ehca_qp_idr, my_qp, &my_qp->token);
-		write_unlock_irqrestore(&ehca_qp_idr_lock, flags);
-	} while (ret == -EAGAIN);
+	ret = idr_alloc(&ehca_qp_idr, my_qp, 0, 0x2000000, GFP_NOWAIT);
+	if (ret >= 0)
+		my_qp->token = ret;
 
-	if (ret) {
-		ret = -ENOMEM;
-		ehca_err(pd->device, "Can't allocate new idr entry.");
+	write_unlock_irqrestore(&ehca_qp_idr_lock, flags);
+	idr_preload_end();
+	if (ret < 0) {
+		if (ret == -ENOSPC) {
+			ret = -EINVAL;
+			ehca_err(pd->device, "Invalid number of qp");
+		} else {
+			ret = -ENOMEM;
+			ehca_err(pd->device, "Can't allocate new idr entry.");
+		}
 		goto create_qp_exit0;
 	}
 
-	if (my_qp->token > 0x1FFFFFF) {
-		ret = -EINVAL;
-		ehca_err(pd->device, "Invalid number of qp");
-		goto create_qp_exit1;
-	}
-
 	if (has_srq)
 		parms.srq_token = my_qp->token;
 
@@ -654,9 +705,9 @@ static struct ehca_qp *internal_create_qp(
 			(parms.squeue.is_small || parms.rqueue.is_small);
 	}
 
-	h_ret = hipz_h_alloc_resource_qp(shca->ipz_hca_handle, &parms);
+	h_ret = hipz_h_alloc_resource_qp(shca->ipz_hca_handle, &parms, is_user);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(pd->device, "h_alloc_resource_qp() failed h_ret=%li",
+		ehca_err(pd->device, "h_alloc_resource_qp() failed h_ret=%lli",
 			 h_ret);
 		ret = ehca2ib_return_code(h_ret);
 		goto create_qp_exit1;
@@ -715,6 +766,21 @@ static struct ehca_qp *internal_create_qp(
 				 "and pages ret=%i", ret);
 			goto create_qp_exit2;
 		}
+
+		if (!is_user) {
+			my_qp->sq_map.entries = my_qp->ipz_squeue.queue_length /
+				my_qp->ipz_squeue.qe_size;
+			my_qp->sq_map.map = vmalloc(my_qp->sq_map.entries *
+						    sizeof(struct ehca_qmap_entry));
+			if (!my_qp->sq_map.map) {
+				ehca_err(pd->device, "Couldn't allocate squeue "
+					 "map ret=%i", ret);
+				goto create_qp_exit3;
+			}
+			INIT_LIST_HEAD(&my_qp->sq_err_node);
+			/* to avoid the generation of bogus flush CQEs */
+			reset_queue_map(&my_qp->sq_map);
+		}
 	}
 
 	if (HAS_RQ(my_qp)) {
@@ -724,8 +790,28 @@ static struct ehca_qp *internal_create_qp(
 		if (ret) {
 			ehca_err(pd->device, "Couldn't initialize rqueue "
 				 "and pages ret=%i", ret);
-			goto create_qp_exit3;
+			goto create_qp_exit4;
+		}
+		if (!is_user) {
+			my_qp->rq_map.entries = my_qp->ipz_rqueue.queue_length /
+				my_qp->ipz_rqueue.qe_size;
+			my_qp->rq_map.map = vmalloc(my_qp->rq_map.entries *
+						    sizeof(struct ehca_qmap_entry));
+			if (!my_qp->rq_map.map) {
+				ehca_err(pd->device, "Couldn't allocate squeue "
+					 "map ret=%i", ret);
+				goto create_qp_exit5;
+			}
+			INIT_LIST_HEAD(&my_qp->rq_err_node);
+			/* to avoid the generation of bogus flush CQEs */
+			reset_queue_map(&my_qp->rq_map);
 		}
+	} else if (init_attr->srq && !is_user) {
+		/* this is a base QP, use the queue map of the SRQ */
+		my_qp->rq_map = my_srq->rq_map;
+		INIT_LIST_HEAD(&my_qp->rq_err_node);
+
+		my_qp->ipz_rqueue = my_srq->ipz_rqueue;
 	}
 
 	if (is_srq) {
@@ -770,7 +856,7 @@ static struct ehca_qp *internal_create_qp(
 			if (!my_qp->mod_qp_parm) {
 				ehca_err(pd->device,
 					 "Could not alloc mod_qp_parm");
-				goto create_qp_exit4;
+				goto create_qp_exit5;
 			}
 		}
 	}
@@ -779,8 +865,13 @@ static struct ehca_qp *internal_create_qp(
 	if (qp_type == IB_QPT_GSI) {
 		h_ret = ehca_define_sqp(shca, my_qp, init_attr);
 		if (h_ret != H_SUCCESS) {
+			kfree(my_qp->mod_qp_parm);
+			my_qp->mod_qp_parm = NULL;
+			/* the QP pointer is no longer valid */
+			shca->sport[init_attr->port_num - 1].ibqp_sqp[qp_type] =
+				NULL;
 			ret = ehca2ib_return_code(h_ret);
-			goto create_qp_exit5;
+			goto create_qp_exit6;
 		}
 	}
 
@@ -789,7 +880,7 @@ static struct ehca_qp *internal_create_qp(
 		if (ret) {
 			ehca_err(pd->device,
 				 "Couldn't assign qp to send_cq ret=%i", ret);
-			goto create_qp_exit5;
+			goto create_qp_exit7;
 		}
 	}
 
@@ -815,22 +906,30 @@ static struct ehca_qp *internal_create_qp(
 		if (ib_copy_to_udata(udata, &resp, sizeof resp)) {
 			ehca_err(pd->device, "Copy to udata failed");
 			ret = -EINVAL;
-			goto create_qp_exit6;
+			goto create_qp_exit8;
 		}
 	}
 
 	return my_qp;
 
-create_qp_exit6:
+create_qp_exit8:
 	ehca_cq_unassign_qp(my_qp->send_cq, my_qp->real_qp_num);
 
-create_qp_exit5:
+create_qp_exit7:
 	kfree(my_qp->mod_qp_parm);
 
-create_qp_exit4:
+create_qp_exit6:
+	if (HAS_RQ(my_qp) && !is_user)
+		vfree(my_qp->rq_map.map);
+
+create_qp_exit5:
 	if (HAS_RQ(my_qp))
 		ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue);
 
+create_qp_exit4:
+	if (HAS_SQ(my_qp) && !is_user)
+		vfree(my_qp->sq_map.map);
+
 create_qp_exit3:
 	if (HAS_SQ(my_qp))
 		ipz_queue_dtor(my_pd, &my_qp->ipz_squeue);
@@ -874,6 +973,9 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
 	struct hcp_modify_qp_control_block *mqpcb;
 	u64 hret, update_mask;
 
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC)
+		return ERR_PTR(-ENOSYS);
+
 	/* For common attributes, internal_create_qp() takes its info
 	 * out of qp_init_attr, so copy all common attrs there.
 	 */
@@ -912,7 +1014,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
 				mqpcb, my_qp->galpas.kernel);
 	if (hret != H_SUCCESS) {
 		ehca_err(pd->device, "Could not modify SRQ to INIT "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, my_qp->real_qp_num, hret);
 		goto create_srq2;
 	}
@@ -926,7 +1028,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
 				mqpcb, my_qp->galpas.kernel);
 	if (hret != H_SUCCESS) {
 		ehca_err(pd->device, "Could not enable SRQ "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, my_qp->real_qp_num, hret);
 		goto create_srq2;
 	}
@@ -940,7 +1042,7 @@ struct ib_srq *ehca_create_srq(struct ib_pd *pd,
 				mqpcb, my_qp->galpas.kernel);
 	if (hret != H_SUCCESS) {
 		ehca_err(pd->device, "Could not modify SRQ to RTR "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, my_qp->real_qp_num, hret);
 		goto create_srq2;
 	}
@@ -980,7 +1082,7 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca,
 					   &bad_send_wqe_p, NULL, 2);
 	if (h_ret != H_SUCCESS) {
 		ehca_err(&shca->ib_device, "hipz_h_disable_and_get_wqe() failed"
-			 " ehca_qp=%p qp_num=%x h_ret=%li",
+			 " ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, qp_num, h_ret);
 		return ehca2ib_return_code(h_ret);
 	}
@@ -988,7 +1090,7 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca,
 	ehca_dbg(&shca->ib_device, "qp_num=%x bad_send_wqe_p=%p",
 		 qp_num, bad_send_wqe_p);
 	/* convert wqe pointer to vadr */
-	bad_send_wqe_v = abs_to_virt((u64)bad_send_wqe_p);
+	bad_send_wqe_v = __va((u64)bad_send_wqe_p);
 	if (ehca_debug_level >= 2)
 		ehca_dmp(bad_send_wqe_v, 32, "qp_num=%x bad_wqe", qp_num);
 	squeue = &my_qp->ipz_squeue;
@@ -1021,6 +1123,111 @@ static int prepare_sqe_rts(struct ehca_qp *my_qp, struct ehca_shca *shca,
 	return 0;
 }
 
+static int calc_left_cqes(u64 wqe_p, struct ipz_queue *ipz_queue,
+			  struct ehca_queue_map *qmap)
+{
+	void *wqe_v;
+	u64 q_ofs;
+	u32 wqe_idx;
+	unsigned int tail_idx;
+
+	/* convert real to abs address */
+	wqe_p = wqe_p & (~(1UL << 63));
+
+	wqe_v = __va(wqe_p);
+
+	if (ipz_queue_abs_to_offset(ipz_queue, wqe_p, &q_ofs)) {
+		ehca_gen_err("Invalid offset for calculating left cqes "
+				"wqe_p=%#llx wqe_v=%p\n", wqe_p, wqe_v);
+		return -EFAULT;
+	}
+
+	tail_idx = next_index(qmap->tail, qmap->entries);
+	wqe_idx = q_ofs / ipz_queue->qe_size;
+
+	/* check all processed wqes, whether a cqe is requested or not */
+	while (tail_idx != wqe_idx) {
+		if (qmap->map[tail_idx].cqe_req)
+			qmap->left_to_poll++;
+		tail_idx = next_index(tail_idx, qmap->entries);
+	}
+	/* save index in queue, where we have to start flushing */
+	qmap->next_wqe_idx = wqe_idx;
+	return 0;
+}
+
+static int check_for_left_cqes(struct ehca_qp *my_qp, struct ehca_shca *shca)
+{
+	u64 h_ret;
+	void *send_wqe_p, *recv_wqe_p;
+	int ret;
+	unsigned long flags;
+	int qp_num = my_qp->ib_qp.qp_num;
+
+	/* this hcall is not supported on base QPs */
+	if (my_qp->ext_type != EQPT_SRQBASE) {
+		/* get send and receive wqe pointer */
+		h_ret = hipz_h_disable_and_get_wqe(shca->ipz_hca_handle,
+				my_qp->ipz_qp_handle, &my_qp->pf,
+				&send_wqe_p, &recv_wqe_p, 4);
+		if (h_ret != H_SUCCESS) {
+			ehca_err(&shca->ib_device, "disable_and_get_wqe() "
+				 "failed ehca_qp=%p qp_num=%x h_ret=%lli",
+				 my_qp, qp_num, h_ret);
+			return ehca2ib_return_code(h_ret);
+		}
+
+		/*
+		 * acquire lock to ensure that nobody is polling the cq which
+		 * could mean that the qmap->tail pointer is in an
+		 * inconsistent state.
+		 */
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		ret = calc_left_cqes((u64)send_wqe_p, &my_qp->ipz_squeue,
+				&my_qp->sq_map);
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+		if (ret)
+			return ret;
+
+
+		spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+		ret = calc_left_cqes((u64)recv_wqe_p, &my_qp->ipz_rqueue,
+				&my_qp->rq_map);
+		spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags);
+		if (ret)
+			return ret;
+	} else {
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		my_qp->sq_map.left_to_poll = 0;
+		my_qp->sq_map.next_wqe_idx = next_index(my_qp->sq_map.tail,
+							my_qp->sq_map.entries);
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+
+		spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+		my_qp->rq_map.left_to_poll = 0;
+		my_qp->rq_map.next_wqe_idx = next_index(my_qp->rq_map.tail,
+							my_qp->rq_map.entries);
+		spin_unlock_irqrestore(&my_qp->recv_cq->spinlock, flags);
+	}
+
+	/* this assures flush cqes being generated only for pending wqes */
+	if ((my_qp->sq_map.left_to_poll == 0) &&
+				(my_qp->rq_map.left_to_poll == 0)) {
+		spin_lock_irqsave(&my_qp->send_cq->spinlock, flags);
+		ehca_add_to_err_list(my_qp, 1);
+		spin_unlock_irqrestore(&my_qp->send_cq->spinlock, flags);
+
+		if (HAS_RQ(my_qp)) {
+			spin_lock_irqsave(&my_qp->recv_cq->spinlock, flags);
+			ehca_add_to_err_list(my_qp, 0);
+			spin_unlock_irqrestore(&my_qp->recv_cq->spinlock,
+					flags);
+		}
+	}
+
+	return 0;
+}
+
 /*
  * internal_modify_qp with circumvention to handle aqp0 properly
  * smi_reset2init indicates if this is an internal reset-to-init-call for
@@ -1041,6 +1248,7 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 	u64 update_mask;
 	u64 h_ret;
 	int bad_wqe_cnt = 0;
+	int is_user = 0;
 	int squeue_locked = 0;
 	unsigned long flags = 0;
 
@@ -1058,11 +1266,13 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 				mqpcb, my_qp->galpas.kernel);
 	if (h_ret != H_SUCCESS) {
 		ehca_err(ibqp->device, "hipz_h_query_qp() failed "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, ibqp->qp_num, h_ret);
 		ret = ehca2ib_return_code(h_ret);
 		goto modify_qp_exit1;
 	}
+	if (ibqp->uobject)
+		is_user = 1;
 
 	qp_cur_state = ehca2ib_qp_state(mqpcb->qp_state);
 
@@ -1119,7 +1329,7 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 	qp_new_state = attr_mask & IB_QP_STATE ? attr->qp_state : qp_cur_state;
 	if (!smi_reset2init &&
 	    !ib_modify_qp_is_ok(qp_cur_state, qp_new_state, ibqp->qp_type,
-				attr_mask)) {
+				attr_mask, IB_LINK_LAYER_UNSPECIFIED)) {
 		ret = -EINVAL;
 		ehca_err(ibqp->device,
 			 "Invalid qp transition new_state=%x cur_state=%x "
@@ -1460,6 +1670,8 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 			goto modify_qp_exit2;
 		}
 		mqpcb->path_migration_state = attr->path_mig_state + 1;
+		if (attr->path_mig_state == IB_MIG_REARM)
+			my_qp->mig_armed = 1;
 		update_mask |=
 			EHCA_BMASK_SET(MQPCB_MASK_PATH_MIGRATION_STATE, 1);
 	}
@@ -1485,7 +1697,7 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 
 	if (h_ret != H_SUCCESS) {
 		ret = ehca2ib_return_code(h_ret);
-		ehca_err(ibqp->device, "hipz_h_modify_qp() failed h_ret=%li "
+		ehca_err(ibqp->device, "hipz_h_modify_qp() failed h_ret=%lli "
 			 "ehca_qp=%p qp_num=%x", h_ret, my_qp, ibqp->qp_num);
 		goto modify_qp_exit2;
 	}
@@ -1518,22 +1730,39 @@ static int internal_modify_qp(struct ib_qp *ibqp,
 			ret = ehca2ib_return_code(h_ret);
 			ehca_err(ibqp->device, "ENABLE in context of "
 				 "RESET_2_INIT failed! Maybe you didn't get "
-				 "a LID h_ret=%li ehca_qp=%p qp_num=%x",
+				 "a LID h_ret=%lli ehca_qp=%p qp_num=%x",
 				 h_ret, my_qp, ibqp->qp_num);
 			goto modify_qp_exit2;
 		}
 	}
+	if ((qp_new_state == IB_QPS_ERR) && (qp_cur_state != IB_QPS_ERR)
+	    && !is_user) {
+		ret = check_for_left_cqes(my_qp, shca);
+		if (ret)
+			goto modify_qp_exit2;
+	}
 
 	if (statetrans == IB_QPST_ANY2RESET) {
 		ipz_qeit_reset(&my_qp->ipz_rqueue);
 		ipz_qeit_reset(&my_qp->ipz_squeue);
+
+		if (qp_cur_state == IB_QPS_ERR && !is_user) {
+			del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node);
+
+			if (HAS_RQ(my_qp))
+				del_from_err_list(my_qp->recv_cq,
+						  &my_qp->rq_err_node);
+		}
+		if (!is_user)
+			reset_queue_map(&my_qp->sq_map);
+
+		if (HAS_RQ(my_qp) && !is_user)
+			reset_queue_map(&my_qp->rq_map);
 	}
 
 	if (attr_mask & IB_QP_QKEY)
 		my_qp->qkey = attr->qkey;
 
-	my_qp->state = qp_new_state;
-
 modify_qp_exit2:
 	if (squeue_locked) { /* this means: sqe -> rts */
 		spin_unlock_irqrestore(&my_qp->spinlock_s, flags);
@@ -1549,6 +1778,8 @@ modify_qp_exit1:
 int ehca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
 		   struct ib_udata *udata)
 {
+	int ret = 0;
+
 	struct ehca_shca *shca = container_of(ibqp->device, struct ehca_shca,
 					      ib_device);
 	struct ehca_qp *my_qp = container_of(ibqp, struct ehca_qp, ib_qp);
@@ -1595,12 +1826,18 @@ int ehca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
 				 attr->qp_state, my_qp->init_attr.port_num,
 				 ibqp->qp_type);
 			spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
-			return 0;
+			goto out;
 		}
 		spin_unlock_irqrestore(&sport->mod_sqp_lock, flags);
 	}
 
-	return internal_modify_qp(ibqp, attr, attr_mask, 0);
+	ret = internal_modify_qp(ibqp, attr, attr_mask, 0);
+
+out:
+	if ((ret == 0) && (attr_mask & IB_QP_STATE))
+		my_qp->state = attr->qp_state;
+
+	return ret;
 }
 
 void ehca_recover_sqp(struct ib_qp *sqp)
@@ -1681,7 +1918,7 @@ int ehca_query_qp(struct ib_qp *qp,
 	if (h_ret != H_SUCCESS) {
 		ret = ehca2ib_return_code(h_ret);
 		ehca_err(qp->device, "hipz_h_query_qp() failed "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, qp->qp_num, h_ret);
 		goto query_qp_exit1;
 	}
@@ -1724,19 +1961,13 @@ int ehca_query_qp(struct ib_qp *qp,
 	qp_attr->cap.max_inline_data = my_qp->sq_max_inline_data_size;
 	qp_attr->dest_qp_num = qpcb->dest_qp_nr;
 
-	qp_attr->pkey_index =
-		EHCA_BMASK_GET(MQPCB_PRIM_P_KEY_IDX, qpcb->prim_p_key_idx);
-
-	qp_attr->port_num =
-		EHCA_BMASK_GET(MQPCB_PRIM_PHYS_PORT, qpcb->prim_phys_port);
-
+	qp_attr->pkey_index = qpcb->prim_p_key_idx;
+	qp_attr->port_num = qpcb->prim_phys_port;
 	qp_attr->timeout = qpcb->timeout;
 	qp_attr->retry_cnt = qpcb->retry_count;
 	qp_attr->rnr_retry = qpcb->rnr_retry_count;
 
-	qp_attr->alt_pkey_index =
-		EHCA_BMASK_GET(MQPCB_PRIM_P_KEY_IDX, qpcb->alt_p_key_idx);
-
+	qp_attr->alt_pkey_index = qpcb->alt_p_key_idx;
 	qp_attr->alt_port_num = qpcb->alt_phys_port;
 	qp_attr->alt_timeout = qpcb->timeout_al;
 
@@ -1823,8 +2054,7 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 		update_mask |=
 			EHCA_BMASK_SET(MQPCB_MASK_CURR_SRQ_LIMIT, 1)
 			| EHCA_BMASK_SET(MQPCB_MASK_QP_AFF_ASYN_EV_LOG_REG, 1);
-		mqpcb->curr_srq_limit =
-			EHCA_BMASK_SET(MQPCB_CURR_SRQ_LIMIT, attr->srq_limit);
+		mqpcb->curr_srq_limit = attr->srq_limit;
 		mqpcb->qp_aff_asyn_ev_log_reg =
 			EHCA_BMASK_SET(QPX_AAELOG_RESET_SRQ_LIMIT, 1);
 	}
@@ -1846,7 +2076,7 @@ int ehca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 
 	if (h_ret != H_SUCCESS) {
 		ret = ehca2ib_return_code(h_ret);
-		ehca_err(ibsrq->device, "hipz_h_modify_qp() failed h_ret=%li "
+		ehca_err(ibsrq->device, "hipz_h_modify_qp() failed h_ret=%lli "
 			 "ehca_qp=%p qp_num=%x",
 			 h_ret, my_qp, my_qp->real_qp_num);
 	}
@@ -1880,15 +2110,14 @@ int ehca_query_srq(struct ib_srq *srq, struct ib_srq_attr *srq_attr)
 	if (h_ret != H_SUCCESS) {
 		ret = ehca2ib_return_code(h_ret);
 		ehca_err(srq->device, "hipz_h_query_qp() failed "
-			 "ehca_qp=%p qp_num=%x h_ret=%li",
+			 "ehca_qp=%p qp_num=%x h_ret=%lli",
 			 my_qp, my_qp->real_qp_num, h_ret);
 		goto query_srq_exit1;
 	}
 
 	srq_attr->max_wr = qpcb->max_nr_outst_recv_wr - 1;
 	srq_attr->max_sge = 3;
-	srq_attr->srq_limit = EHCA_BMASK_GET(
-		MQPCB_CURR_SRQ_LIMIT, qpcb->curr_srq_limit);
+	srq_attr->srq_limit = qpcb->curr_srq_limit;
 
 	if (ehca_debug_level >= 2)
 		ehca_dmp(qpcb, 4*70, "qp_num=%x", my_qp->real_qp_num);
@@ -1910,10 +2139,12 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 	int ret;
 	u64 h_ret;
 	u8 port_num;
+	int is_user = 0;
 	enum ib_qp_type	qp_type;
 	unsigned long flags;
 
 	if (uobject) {
+		is_user = 1;
 		if (my_qp->mm_count_galpa ||
 		    my_qp->mm_count_rqueue || my_qp->mm_count_squeue) {
 			ehca_err(dev, "Resources still referenced in "
@@ -1936,12 +2167,22 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 	idr_remove(&ehca_qp_idr, my_qp->token);
 	write_unlock_irqrestore(&ehca_qp_idr_lock, flags);
 
+	/*
+	 * SRQs will never get into an error list and do not have a recv_cq,
+	 * so we need to skip them here.
+	 */
+	if (HAS_RQ(my_qp) && !IS_SRQ(my_qp) && !is_user)
+		del_from_err_list(my_qp->recv_cq, &my_qp->rq_err_node);
+
+	if (HAS_SQ(my_qp) && !is_user)
+		del_from_err_list(my_qp->send_cq, &my_qp->sq_err_node);
+
 	/* now wait until all pending events have completed */
 	wait_event(my_qp->wait_completion, !atomic_read(&my_qp->nr_events));
 
 	h_ret = hipz_h_destroy_qp(shca->ipz_hca_handle, my_qp);
 	if (h_ret != H_SUCCESS) {
-		ehca_err(dev, "hipz_h_destroy_qp() failed h_ret=%li "
+		ehca_err(dev, "hipz_h_destroy_qp() failed h_ret=%lli "
 			 "ehca_qp=%p qp_num=%x", h_ret, my_qp, qp_num);
 		return ehca2ib_return_code(h_ret);
 	}
@@ -1961,7 +2202,7 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 	if (qp_type == IB_QPT_GSI) {
 		struct ib_event event;
 		ehca_info(dev, "device %s: port %x is inactive.",
-			  shca->ib_device.name, port_num);
+				shca->ib_device.name, port_num);
 		event.device = &shca->ib_device;
 		event.event = IB_EVENT_PORT_ERR;
 		event.element.port_num = port_num;
@@ -1969,10 +2210,16 @@ static int internal_destroy_qp(struct ib_device *dev, struct ehca_qp *my_qp,
 		ib_dispatch_event(&event);
 	}
 
-	if (HAS_RQ(my_qp))
+	if (HAS_RQ(my_qp)) {
 		ipz_queue_dtor(my_pd, &my_qp->ipz_rqueue);
-	if (HAS_SQ(my_qp))
+		if (!is_user)
+			vfree(my_qp->rq_map.map);
+	}
+	if (HAS_SQ(my_qp)) {
 		ipz_queue_dtor(my_pd, &my_qp->ipz_squeue);
+		if (!is_user)
+			vfree(my_qp->sq_map.map);
+	}
 	kmem_cache_free(qp_cache, my_qp);
 	atomic_dec(&shca->num_qps);
 	return 0;
diff --git a/drivers/infiniband/hw/ehca/ehca_reqs.c b/drivers/infiniband/hw/ehca/ehca_reqs.c
index dd9bc68f1c7..47f94984353 100644
--- a/drivers/infiniband/hw/ehca/ehca_reqs.c
+++ b/drivers/infiniband/hw/ehca/ehca_reqs.c
@@ -42,7 +42,6 @@
  */
 
 
-#include <asm-powerpc/system.h>
 #include "ehca_classes.h"
 #include "ehca_tools.h"
 #include "ehca_qes.h"
@@ -53,9 +52,25 @@
 /* in RC traffic, insert an empty RDMA READ every this many packets */
 #define ACK_CIRC_THRESHOLD 2000000
 
+static u64 replace_wr_id(u64 wr_id, u16 idx)
+{
+	u64 ret;
+
+	ret = wr_id & ~QMAP_IDX_MASK;
+	ret |= idx & QMAP_IDX_MASK;
+
+	return ret;
+}
+
+static u16 get_app_wr_id(u64 wr_id)
+{
+	return wr_id & QMAP_IDX_MASK;
+}
+
 static inline int ehca_write_rwqe(struct ipz_queue *ipz_rqueue,
 				  struct ehca_wqe *wqe_p,
-				  struct ib_recv_wr *recv_wr)
+				  struct ib_recv_wr *recv_wr,
+				  u32 rq_map_idx)
 {
 	u8 cnt_ds;
 	if (unlikely((recv_wr->num_sge < 0) ||
@@ -69,7 +84,7 @@ static inline int ehca_write_rwqe(struct ipz_queue *ipz_rqueue,
 	/* clear wqe header until sglist */
 	memset(wqe_p, 0, offsetof(struct ehca_wqe, u.ud_av.sg_list));
 
-	wqe_p->work_request_id = recv_wr->wr_id;
+	wqe_p->work_request_id = replace_wr_id(recv_wr->wr_id, rq_map_idx);
 	wqe_p->nr_of_data_seg = recv_wr->num_sge;
 
 	for (cnt_ds = 0; cnt_ds < recv_wr->num_sge; cnt_ds++) {
@@ -120,7 +135,7 @@ static void trace_send_wr_ud(const struct ib_send_wr *send_wr)
 				     mad_hdr->attr_mod);
 		}
 		for (j = 0; j < send_wr->num_sge; j++) {
-			u8 *data = (u8 *)abs_to_virt(sge->addr);
+			u8 *data = __va(sge->addr);
 			ehca_gen_dbg("send_wr#%x sge#%x addr=%p length=%x "
 				     "lkey=%x",
 				     idx, j, data, sge->length, sge->lkey);
@@ -139,12 +154,14 @@ static void trace_send_wr_ud(const struct ib_send_wr *send_wr)
 static inline int ehca_write_swqe(struct ehca_qp *qp,
 				  struct ehca_wqe *wqe_p,
 				  const struct ib_send_wr *send_wr,
+				  u32 sq_map_idx,
 				  int hidden)
 {
 	u32 idx;
 	u64 dma_length;
 	struct ehca_av *my_av;
 	u32 remote_qkey = send_wr->wr.ud.remote_qkey;
+	struct ehca_qmap_entry *qmap_entry = &qp->sq_map.map[sq_map_idx];
 
 	if (unlikely((send_wr->num_sge < 0) ||
 		     (send_wr->num_sge > qp->ipz_squeue.act_nr_of_sg))) {
@@ -157,7 +174,11 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 	/* clear wqe header until sglist */
 	memset(wqe_p, 0, offsetof(struct ehca_wqe, u.ud_av.sg_list));
 
-	wqe_p->work_request_id = send_wr->wr_id;
+	wqe_p->work_request_id = replace_wr_id(send_wr->wr_id, sq_map_idx);
+
+	qmap_entry->app_wr_id = get_app_wr_id(send_wr->wr_id);
+	qmap_entry->reported = 0;
+	qmap_entry->cqe_req = 0;
 
 	switch (send_wr->opcode) {
 	case IB_WR_SEND:
@@ -182,8 +203,10 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 
 	if ((send_wr->send_flags & IB_SEND_SIGNALED ||
 	    qp->init_attr.sq_sig_type == IB_SIGNAL_ALL_WR)
-	    && !hidden)
+	    && !hidden) {
 		wqe_p->wr_flag |= WQE_WRFLAG_REQ_SIGNAL_COM;
+		qmap_entry->cqe_req = 1;
+	}
 
 	if (send_wr->opcode == IB_WR_SEND_WITH_IMM ||
 	    send_wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM) {
@@ -245,7 +268,7 @@ static inline int ehca_write_swqe(struct ehca_qp *qp,
 		/* no break is intentional here */
 	case IB_QPT_RC:
 		/* TODO: atomic not implemented */
-		wqe_p->u.nud.remote_virtual_adress =
+		wqe_p->u.nud.remote_virtual_address =
 			send_wr->wr.rdma.remote_addr;
 		wqe_p->u.nud.rkey = send_wr->wr.rdma.rkey;
 
@@ -376,33 +399,36 @@ static inline void map_ib_wc_status(u32 cqe_status,
 
 static inline int post_one_send(struct ehca_qp *my_qp,
 			 struct ib_send_wr *cur_send_wr,
-			 struct ib_send_wr **bad_send_wr,
 			 int hidden)
 {
 	struct ehca_wqe *wqe_p;
 	int ret;
+	u32 sq_map_idx;
 	u64 start_offset = my_qp->ipz_squeue.current_q_offset;
 
 	/* get pointer next to free WQE */
 	wqe_p = ipz_qeit_get_inc(&my_qp->ipz_squeue);
 	if (unlikely(!wqe_p)) {
 		/* too many posted work requests: queue overflow */
-		if (bad_send_wr)
-			*bad_send_wr = cur_send_wr;
 		ehca_err(my_qp->ib_qp.device, "Too many posted WQEs "
 			 "qp_num=%x", my_qp->ib_qp.qp_num);
 		return -ENOMEM;
 	}
+
+	/*
+	 * Get the index of the WQE in the send queue. The same index is used
+	 * for writing into the sq_map.
+	 */
+	sq_map_idx = start_offset / my_qp->ipz_squeue.qe_size;
+
 	/* write a SEND WQE into the QUEUE */
-	ret = ehca_write_swqe(my_qp, wqe_p, cur_send_wr, hidden);
+	ret = ehca_write_swqe(my_qp, wqe_p, cur_send_wr, sq_map_idx, hidden);
 	/*
 	 * if something failed,
 	 * reset the free entry pointer to the start value
 	 */
 	if (unlikely(ret)) {
 		my_qp->ipz_squeue.current_q_offset = start_offset;
-		if (bad_send_wr)
-			*bad_send_wr = cur_send_wr;
 		ehca_err(my_qp->ib_qp.device, "Could not write WQE "
 			 "qp_num=%x", my_qp->ib_qp.qp_num);
 		return -EINVAL;
@@ -416,7 +442,6 @@ int ehca_post_send(struct ib_qp *qp,
 		   struct ib_send_wr **bad_send_wr)
 {
 	struct ehca_qp *my_qp = container_of(qp, struct ehca_qp, ib_qp);
-	struct ib_send_wr *cur_send_wr;
 	int wqe_cnt = 0;
 	int ret = 0;
 	unsigned long flags;
@@ -425,7 +450,8 @@ int ehca_post_send(struct ib_qp *qp,
 	if (unlikely(my_qp->state < IB_QPS_RTS)) {
 		ehca_err(qp->device, "Invalid QP state  qp_state=%d qpn=%x",
 			 my_qp->state, qp->qp_num);
-		return -EINVAL;
+		ret = -EINVAL;
+		goto out;
 	}
 
 	/* LOCK the QUEUE */
@@ -444,24 +470,21 @@ int ehca_post_send(struct ib_qp *qp,
 		struct ib_send_wr circ_wr;
 		memset(&circ_wr, 0, sizeof(circ_wr));
 		circ_wr.opcode = IB_WR_RDMA_READ;
-		post_one_send(my_qp, &circ_wr, NULL, 1); /* ignore retcode */
+		post_one_send(my_qp, &circ_wr, 1); /* ignore retcode */
 		wqe_cnt++;
 		ehca_dbg(qp->device, "posted circ wr  qp_num=%x", qp->qp_num);
 		my_qp->message_count = my_qp->packet_count = 0;
 	}
 
 	/* loop processes list of send reqs */
-	for (cur_send_wr = send_wr; cur_send_wr != NULL;
-	     cur_send_wr = cur_send_wr->next) {
-		ret = post_one_send(my_qp, cur_send_wr, bad_send_wr, 0);
+	while (send_wr) {
+		ret = post_one_send(my_qp, send_wr, 0);
 		if (unlikely(ret)) {
-			/* if one or more WQEs were successful, don't fail */
-			if (wqe_cnt)
-				ret = 0;
 			goto post_send_exit0;
 		}
 		wqe_cnt++;
-	} /* eof for cur_send_wr */
+		send_wr = send_wr->next;
+	}
 
 post_send_exit0:
 	iosync(); /* serialize GAL register access */
@@ -471,6 +494,10 @@ post_send_exit0:
 			 my_qp, qp->qp_num, wqe_cnt, ret);
 	my_qp->message_count += wqe_cnt;
 	spin_unlock_irqrestore(&my_qp->spinlock_s, flags);
+
+out:
+	if (ret)
+		*bad_send_wr = send_wr;
 	return ret;
 }
 
@@ -479,56 +506,64 @@ static int internal_post_recv(struct ehca_qp *my_qp,
 			      struct ib_recv_wr *recv_wr,
 			      struct ib_recv_wr **bad_recv_wr)
 {
-	struct ib_recv_wr *cur_recv_wr;
 	struct ehca_wqe *wqe_p;
 	int wqe_cnt = 0;
 	int ret = 0;
+	u32 rq_map_idx;
 	unsigned long flags;
+	struct ehca_qmap_entry *qmap_entry;
 
 	if (unlikely(!HAS_RQ(my_qp))) {
 		ehca_err(dev, "QP has no RQ  ehca_qp=%p qp_num=%x ext_type=%d",
 			 my_qp, my_qp->real_qp_num, my_qp->ext_type);
-		return -ENODEV;
+		ret = -ENODEV;
+		goto out;
 	}
 
 	/* LOCK the QUEUE */
 	spin_lock_irqsave(&my_qp->spinlock_r, flags);
 
-	/* loop processes list of send reqs */
-	for (cur_recv_wr = recv_wr; cur_recv_wr != NULL;
-	     cur_recv_wr = cur_recv_wr->next) {
+	/* loop processes list of recv reqs */
+	while (recv_wr) {
 		u64 start_offset = my_qp->ipz_rqueue.current_q_offset;
 		/* get pointer next to free WQE */
 		wqe_p = ipz_qeit_get_inc(&my_qp->ipz_rqueue);
 		if (unlikely(!wqe_p)) {
 			/* too many posted work requests: queue overflow */
-			if (bad_recv_wr)
-				*bad_recv_wr = cur_recv_wr;
-			if (wqe_cnt == 0) {
-				ret = -ENOMEM;
-				ehca_err(dev, "Too many posted WQEs "
-					 "qp_num=%x", my_qp->real_qp_num);
-			}
+			ret = -ENOMEM;
+			ehca_err(dev, "Too many posted WQEs "
+				"qp_num=%x", my_qp->real_qp_num);
 			goto post_recv_exit0;
 		}
+		/*
+		 * Get the index of the WQE in the recv queue. The same index
+		 * is used for writing into the rq_map.
+		 */
+		rq_map_idx = start_offset / my_qp->ipz_rqueue.qe_size;
+
 		/* write a RECV WQE into the QUEUE */
-		ret = ehca_write_rwqe(&my_qp->ipz_rqueue, wqe_p, cur_recv_wr);
+		ret = ehca_write_rwqe(&my_qp->ipz_rqueue, wqe_p, recv_wr,
+				rq_map_idx);
 		/*
 		 * if something failed,
 		 * reset the free entry pointer to the start value
 		 */
 		if (unlikely(ret)) {
 			my_qp->ipz_rqueue.current_q_offset = start_offset;
-			*bad_recv_wr = cur_recv_wr;
-			if (wqe_cnt == 0) {
-				ret = -EINVAL;
-				ehca_err(dev, "Could not write WQE "
-					 "qp_num=%x", my_qp->real_qp_num);
-			}
+			ret = -EINVAL;
+			ehca_err(dev, "Could not write WQE "
+				"qp_num=%x", my_qp->real_qp_num);
 			goto post_recv_exit0;
 		}
+
+		qmap_entry = &my_qp->rq_map.map[rq_map_idx];
+		qmap_entry->app_wr_id = get_app_wr_id(recv_wr->wr_id);
+		qmap_entry->reported = 0;
+		qmap_entry->cqe_req = 1;
+
 		wqe_cnt++;
-	} /* eof for cur_recv_wr */
+		recv_wr = recv_wr->next;
+	} /* eof for recv_wr */
 
 post_recv_exit0:
 	iosync(); /* serialize GAL register access */
@@ -537,6 +572,11 @@ post_recv_exit0:
 	    ehca_dbg(dev, "ehca_qp=%p qp_num=%x wqe_cnt=%d ret=%i",
 		     my_qp, my_qp->real_qp_num, wqe_cnt, ret);
 	spin_unlock_irqrestore(&my_qp->spinlock_r, flags);
+
+out:
+	if (ret)
+		*bad_recv_wr = recv_wr;
+
 	return ret;
 }
 
@@ -550,6 +590,7 @@ int ehca_post_recv(struct ib_qp *qp,
 	if (unlikely(my_qp->state == IB_QPS_RESET)) {
 		ehca_err(qp->device, "Invalid QP state  qp_state=%d qpn=%x",
 			 my_qp->state, qp->qp_num);
+		*bad_recv_wr = recv_wr;
 		return -EINVAL;
 	}
 
@@ -583,13 +624,15 @@ static const u8 ib_wc_opcode[255] = {
 /* internal function to poll one entry of cq */
 static inline int ehca_poll_cq_one(struct ib_cq *cq, struct ib_wc *wc)
 {
-	int ret = 0;
+	int ret = 0, qmap_tail_idx;
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
 	struct ehca_cqe *cqe;
 	struct ehca_qp *my_qp;
+	struct ehca_qmap_entry *qmap_entry;
+	struct ehca_queue_map *qmap;
 	int cqe_count = 0, is_error;
 
-poll_cq_one_read_cqe:
+repoll:
 	cqe = (struct ehca_cqe *)
 		ipz_qeit_get_inc_valid(&my_cq->ipz_queue);
 	if (!cqe) {
@@ -617,7 +660,7 @@ poll_cq_one_read_cqe:
 			ehca_dmp(cqe, 64, "cq_num=%x qp_num=%x",
 				 my_cq->cq_number, cqe->local_qp_number);
 			/* ignore this purged cqe */
-			goto poll_cq_one_read_cqe;
+			goto repoll;
 		}
 		spin_lock_irqsave(&qp->spinlock_s, flags);
 		purgeflag = qp->sqerr_purgeflag;
@@ -636,7 +679,7 @@ poll_cq_one_read_cqe:
 			 * that caused sqe and turn off purge flag
 			 */
 			qp->sqerr_purgeflag = 0;
-			goto poll_cq_one_read_cqe;
+			goto repoll;
 		}
 	}
 
@@ -654,8 +697,62 @@ poll_cq_one_read_cqe:
 			 my_cq, my_cq->cq_number);
 	}
 
-	/* we got a completion! */
-	wc->wr_id = cqe->work_request_id;
+	read_lock(&ehca_qp_idr_lock);
+	my_qp = idr_find(&ehca_qp_idr, cqe->qp_token);
+	read_unlock(&ehca_qp_idr_lock);
+	if (!my_qp)
+		goto repoll;
+	wc->qp = &my_qp->ib_qp;
+
+	qmap_tail_idx = get_app_wr_id(cqe->work_request_id);
+	if (!(cqe->w_completion_flags & WC_SEND_RECEIVE_BIT))
+		/* We got a send completion. */
+		qmap = &my_qp->sq_map;
+	else
+		/* We got a receive completion. */
+		qmap = &my_qp->rq_map;
+
+	/* advance the tail pointer */
+	qmap->tail = qmap_tail_idx;
+
+	if (is_error) {
+		/*
+		 * set left_to_poll to 0 because in error state, we will not
+		 * get any additional CQEs
+		 */
+		my_qp->sq_map.next_wqe_idx = next_index(my_qp->sq_map.tail,
+							my_qp->sq_map.entries);
+		my_qp->sq_map.left_to_poll = 0;
+		ehca_add_to_err_list(my_qp, 1);
+
+		my_qp->rq_map.next_wqe_idx = next_index(my_qp->rq_map.tail,
+							my_qp->rq_map.entries);
+		my_qp->rq_map.left_to_poll = 0;
+		if (HAS_RQ(my_qp))
+			ehca_add_to_err_list(my_qp, 0);
+	}
+
+	qmap_entry = &qmap->map[qmap_tail_idx];
+	if (qmap_entry->reported) {
+		ehca_warn(cq->device, "Double cqe on qp_num=%#x",
+				my_qp->real_qp_num);
+		/* found a double cqe, discard it and read next one */
+		goto repoll;
+	}
+
+	wc->wr_id = replace_wr_id(cqe->work_request_id, qmap_entry->app_wr_id);
+	qmap_entry->reported = 1;
+
+	/* if left_to_poll is decremented to 0, add the QP to the error list */
+	if (qmap->left_to_poll > 0) {
+		qmap->left_to_poll--;
+		if ((my_qp->sq_map.left_to_poll == 0) &&
+				(my_qp->rq_map.left_to_poll == 0)) {
+			ehca_add_to_err_list(my_qp, 1);
+			if (HAS_RQ(my_qp))
+				ehca_add_to_err_list(my_qp, 0);
+		}
+	}
 
 	/* eval ib_wc_opcode */
 	wc->opcode = ib_wc_opcode[cqe->optype]-1;
@@ -667,7 +764,7 @@ poll_cq_one_read_cqe:
 		ehca_dmp(cqe, 64, "ehca_cq=%p cq_num=%x",
 			 my_cq, my_cq->cq_number);
 		/* update also queue adder to throw away this entry!!! */
-		goto poll_cq_one_exit0;
+		goto repoll;
 	}
 
 	/* eval ib_wc_status */
@@ -678,17 +775,16 @@ poll_cq_one_read_cqe:
 	} else
 		wc->status = IB_WC_SUCCESS;
 
-	read_lock(&ehca_qp_idr_lock);
-	my_qp = idr_find(&ehca_qp_idr, cqe->qp_token);
-	wc->qp = &my_qp->ib_qp;
-	read_unlock(&ehca_qp_idr_lock);
-
 	wc->byte_len = cqe->nr_bytes_transferred;
 	wc->pkey_index = cqe->pkey_index;
 	wc->slid = cqe->rlid;
 	wc->dlid_path_bits = cqe->dlid;
 	wc->src_qp = cqe->remote_qp_number;
-	wc->wc_flags = cqe->w_completion_flags;
+	/*
+	 * HW has "Immed data present" and "GRH present" in bits 6 and 5.
+	 * SW defines those in bits 1 and 0, so we can just shift and mask.
+	 */
+	wc->wc_flags = (cqe->w_completion_flags >> 5) & 3;
 	wc->ex.imm_data = cpu_to_be32(cqe->immediate_data);
 	wc->sl = cqe->service_level;
 
@@ -699,13 +795,89 @@ poll_cq_one_exit0:
 	return ret;
 }
 
+static int generate_flush_cqes(struct ehca_qp *my_qp, struct ib_cq *cq,
+			       struct ib_wc *wc, int num_entries,
+			       struct ipz_queue *ipz_queue, int on_sq)
+{
+	int nr = 0;
+	struct ehca_wqe *wqe;
+	u64 offset;
+	struct ehca_queue_map *qmap;
+	struct ehca_qmap_entry *qmap_entry;
+
+	if (on_sq)
+		qmap = &my_qp->sq_map;
+	else
+		qmap = &my_qp->rq_map;
+
+	qmap_entry = &qmap->map[qmap->next_wqe_idx];
+
+	while ((nr < num_entries) && (qmap_entry->reported == 0)) {
+		/* generate flush CQE */
+
+		memset(wc, 0, sizeof(*wc));
+
+		offset = qmap->next_wqe_idx * ipz_queue->qe_size;
+		wqe = (struct ehca_wqe *)ipz_qeit_calc(ipz_queue, offset);
+		if (!wqe) {
+			ehca_err(cq->device, "Invalid wqe offset=%#llx on "
+				 "qp_num=%#x", offset, my_qp->real_qp_num);
+			return nr;
+		}
+
+		wc->wr_id = replace_wr_id(wqe->work_request_id,
+					  qmap_entry->app_wr_id);
+
+		if (on_sq) {
+			switch (wqe->optype) {
+			case WQE_OPTYPE_SEND:
+				wc->opcode = IB_WC_SEND;
+				break;
+			case WQE_OPTYPE_RDMAWRITE:
+				wc->opcode = IB_WC_RDMA_WRITE;
+				break;
+			case WQE_OPTYPE_RDMAREAD:
+				wc->opcode = IB_WC_RDMA_READ;
+				break;
+			default:
+				ehca_err(cq->device, "Invalid optype=%x",
+						wqe->optype);
+				return nr;
+			}
+		} else
+			wc->opcode = IB_WC_RECV;
+
+		if (wqe->wr_flag & WQE_WRFLAG_IMM_DATA_PRESENT) {
+			wc->ex.imm_data = wqe->immediate_data;
+			wc->wc_flags |= IB_WC_WITH_IMM;
+		}
+
+		wc->status = IB_WC_WR_FLUSH_ERR;
+
+		wc->qp = &my_qp->ib_qp;
+
+		/* mark as reported and advance next_wqe pointer */
+		qmap_entry->reported = 1;
+		qmap->next_wqe_idx = next_index(qmap->next_wqe_idx,
+						qmap->entries);
+		qmap_entry = &qmap->map[qmap->next_wqe_idx];
+
+		wc++; nr++;
+	}
+
+	return nr;
+
+}
+
 int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 {
 	struct ehca_cq *my_cq = container_of(cq, struct ehca_cq, ib_cq);
 	int nr;
+	struct ehca_qp *err_qp;
 	struct ib_wc *current_wc = wc;
 	int ret = 0;
 	unsigned long flags;
+	int entries_left = num_entries;
 
 	if (num_entries < 1) {
 		ehca_err(cq->device, "Invalid num_entries=%d ehca_cq=%p "
@@ -715,15 +887,40 @@ int ehca_poll_cq(struct ib_cq *cq, int num_entries, struct ib_wc *wc)
 	}
 
 	spin_lock_irqsave(&my_cq->spinlock, flags);
-	for (nr = 0; nr < num_entries; nr++) {
+
+	/* generate flush cqes for send queues */
+	list_for_each_entry(err_qp, &my_cq->sqp_err_list, sq_err_node) {
+		nr = generate_flush_cqes(err_qp, cq, current_wc, entries_left,
+				&err_qp->ipz_squeue, 1);
+		entries_left -= nr;
+		current_wc += nr;
+
+		if (entries_left == 0)
+			break;
+	}
+
+	/* generate flush cqes for receive queues */
+	list_for_each_entry(err_qp, &my_cq->rqp_err_list, rq_err_node) {
+		nr = generate_flush_cqes(err_qp, cq, current_wc, entries_left,
+				&err_qp->ipz_rqueue, 0);
+		entries_left -= nr;
+		current_wc += nr;
+
+		if (entries_left == 0)
+			break;
+	}
+
+	for (nr = 0; nr < entries_left; nr++) {
 		ret = ehca_poll_cq_one(cq, current_wc);
 		if (ret)
 			break;
 		current_wc++;
 	} /* eof for nr */
+	entries_left -= nr;
+
 	spin_unlock_irqrestore(&my_cq->spinlock, flags);
 	if (ret == -EAGAIN  || !ret)
-		ret = nr;
+		ret = num_entries - entries_left;
 
 poll_cq_exit0:
 	return ret;
diff --git a/drivers/infiniband/hw/ehca/ehca_sqp.c b/drivers/infiniband/hw/ehca/ehca_sqp.c
index 706d97ad555..dba8f9f8b99 100644
--- a/drivers/infiniband/hw/ehca/ehca_sqp.c
+++ b/drivers/infiniband/hw/ehca/ehca_sqp.c
@@ -46,11 +46,11 @@
 #include "ehca_iverbs.h"
 #include "hcp_if.h"
 
-#define IB_MAD_STATUS_REDIRECT		__constant_htons(0x0002)
-#define IB_MAD_STATUS_UNSUP_VERSION	__constant_htons(0x0004)
-#define IB_MAD_STATUS_UNSUP_METHOD	__constant_htons(0x0008)
+#define IB_MAD_STATUS_REDIRECT		cpu_to_be16(0x0002)
+#define IB_MAD_STATUS_UNSUP_VERSION	cpu_to_be16(0x0004)
+#define IB_MAD_STATUS_UNSUP_METHOD	cpu_to_be16(0x0008)
 
-#define IB_PMA_CLASS_PORT_INFO		__constant_htons(0x0001)
+#define IB_PMA_CLASS_PORT_INFO		cpu_to_be16(0x0001)
 
 /**
  * ehca_define_sqp - Defines special queue pair 1 (GSI QP). When special queue
@@ -85,7 +85,7 @@ u64 ehca_define_sqp(struct ehca_shca *shca,
 
 		if (ret != H_SUCCESS) {
 			ehca_err(&shca->ib_device,
-				 "Can't define AQP1 for port %x. h_ret=%li",
+				 "Can't define AQP1 for port %x. h_ret=%lli",
 				 port, ret);
 			return ret;
 		}
@@ -125,14 +125,30 @@ struct ib_perf {
 	u8 data[192];
 } __attribute__ ((packed));
 
+/* TC/SL/FL packed into 32 bits, as in ClassPortInfo */
+struct tcslfl {
+	u32 tc:8;
+	u32 sl:4;
+	u32 fl:20;
+} __attribute__ ((packed));
+
+/* IP Version/TC/FL packed into 32 bits, as in GRH */
+struct vertcfl {
+	u32 ver:4;
+	u32 tc:8;
+	u32 fl:20;
+} __attribute__ ((packed));
 
 static int ehca_process_perf(struct ib_device *ibdev, u8 port_num,
+			     struct ib_wc *in_wc, struct ib_grh *in_grh,
 			     struct ib_mad *in_mad, struct ib_mad *out_mad)
 {
 	struct ib_perf *in_perf = (struct ib_perf *)in_mad;
 	struct ib_perf *out_perf = (struct ib_perf *)out_mad;
 	struct ib_class_port_info *poi =
 		(struct ib_class_port_info *)out_perf->data;
+	struct tcslfl *tcslfl =
+		(struct tcslfl *)&poi->redirect_tcslfl;
 	struct ehca_shca *shca =
 		container_of(ibdev, struct ehca_shca, ib_device);
 	struct ehca_sport *sport = &shca->sport[port_num - 1];
@@ -158,10 +174,29 @@ static int ehca_process_perf(struct ib_device *ibdev, u8 port_num,
 		poi->base_version = 1;
 		poi->class_version = 1;
 		poi->resp_time_value = 18;
-		poi->redirect_lid = sport->saved_attr.lid;
-		poi->redirect_qp = sport->pma_qp_nr;
+
+		/* copy local routing information from WC where applicable */
+		tcslfl->sl         = in_wc->sl;
+		poi->redirect_lid  =
+			sport->saved_attr.lid | in_wc->dlid_path_bits;
+		poi->redirect_qp   = sport->pma_qp_nr;
 		poi->redirect_qkey = IB_QP1_QKEY;
-		poi->redirect_pkey = IB_DEFAULT_PKEY_FULL;
+
+		ehca_query_pkey(ibdev, port_num, in_wc->pkey_index,
+				&poi->redirect_pkey);
+
+		/* if request was globally routed, copy route info */
+		if (in_grh) {
+			struct vertcfl *vertcfl =
+				(struct vertcfl *)&in_grh->version_tclass_flow;
+			memcpy(poi->redirect_gid, in_grh->dgid.raw,
+			       sizeof(poi->redirect_gid));
+			tcslfl->tc        = vertcfl->tc;
+			tcslfl->fl        = vertcfl->fl;
+		} else
+			/* else only fill in default GID */
+			ehca_query_gid(ibdev, port_num, 0,
+				       (union ib_gid *)&poi->redirect_gid);
 
 		ehca_dbg(ibdev, "ehca_pma_lid=%x ehca_pma_qp=%x",
 			 sport->saved_attr.lid, sport->pma_qp_nr);
@@ -183,12 +218,11 @@ perf_reply:
 
 int ehca_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
 		     struct ib_wc *in_wc, struct ib_grh *in_grh,
-		     struct ib_mad *in_mad,
-		     struct ib_mad *out_mad)
+		     struct ib_mad *in_mad, struct ib_mad *out_mad)
 {
 	int ret;
 
-	if (!port_num || port_num > ibdev->phys_port_cnt)
+	if (!port_num || port_num > ibdev->phys_port_cnt || !in_wc)
 		return IB_MAD_RESULT_FAILURE;
 
 	/* accept only pma request */
@@ -196,7 +230,8 @@ int ehca_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
 		return IB_MAD_RESULT_SUCCESS;
 
 	ehca_dbg(ibdev, "port_num=%x src_qp=%x", port_num, in_wc->src_qp);
-	ret = ehca_process_perf(ibdev, port_num, in_mad, out_mad);
+	ret = ehca_process_perf(ibdev, port_num, in_wc, in_grh,
+				in_mad, out_mad);
 
 	return ret;
 }
diff --git a/drivers/infiniband/hw/ehca/ehca_tools.h b/drivers/infiniband/hw/ehca/ehca_tools.h
index ec950bf8c47..d280b12aae6 100644
--- a/drivers/infiniband/hw/ehca/ehca_tools.h
+++ b/drivers/infiniband/hw/ehca/ehca_tools.h
@@ -54,13 +54,11 @@
 #include <linux/module.h>
 #include <linux/moduleparam.h>
 #include <linux/vmalloc.h>
-#include <linux/version.h>
 #include <linux/notifier.h>
 #include <linux/cpu.h>
 #include <linux/device.h>
 
-#include <asm/atomic.h>
-#include <asm/abs_addr.h>
+#include <linux/atomic.h>
 #include <asm/ibmebus.h>
 #include <asm/io.h>
 #include <asm/pgtable.h>
@@ -117,7 +115,7 @@ extern int ehca_debug_level;
 		unsigned char *deb = (unsigned char *)(adr); \
 		for (x = 0; x < l; x += 16) { \
 			printk(KERN_INFO "EHCA_DMP:%s " format \
-			       " adr=%p ofs=%04x %016lx %016lx\n", \
+			       " adr=%p ofs=%04x %016llx %016llx\n", \
 			       __func__, ##args, deb, x, \
 			       *((u64 *)&deb[0]), *((u64 *)&deb[8])); \
 			deb += 16; \
diff --git a/drivers/infiniband/hw/ehca/ehca_uverbs.c b/drivers/infiniband/hw/ehca/ehca_uverbs.c
index e43ed8f8a0c..1a1d5d99fcf 100644
--- a/drivers/infiniband/hw/ehca/ehca_uverbs.c
+++ b/drivers/infiniband/hw/ehca/ehca_uverbs.c
@@ -40,6 +40,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_classes.h"
 #include "ehca_iverbs.h"
 #include "ehca_mrmw.h"
@@ -95,7 +97,7 @@ static void ehca_mm_close(struct vm_area_struct *vma)
 		     vma->vm_start, vma->vm_end, *count);
 }
 
-static struct vm_operations_struct vm_ops = {
+static const struct vm_operations_struct vm_ops = {
 	.open =	ehca_mm_open,
 	.close = ehca_mm_close,
 };
@@ -114,8 +116,8 @@ static int ehca_mmap_fw(struct vm_area_struct *vma, struct h_galpas *galpas,
 
 	physical = galpas->user.fw_handle;
 	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
-	ehca_gen_dbg("vsize=%lx physical=%lx", vsize, physical);
-	/* VM_IO | VM_RESERVED are set by remap_pfn_range() */
+	ehca_gen_dbg("vsize=%llx physical=%llx", vsize, physical);
+	/* VM_IO | VM_DONTEXPAND | VM_DONTDUMP are set by remap_pfn_range() */
 	ret = remap_4k_pfn(vma, vma->vm_start, physical >> EHCA_PAGESHIFT,
 			   vma->vm_page_prot);
 	if (unlikely(ret)) {
@@ -137,7 +139,7 @@ static int ehca_mmap_queue(struct vm_area_struct *vma, struct ipz_queue *queue,
 	u64 start, ofs;
 	struct page *page;
 
-	vma->vm_flags |= VM_RESERVED;
+	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
 	start = vma->vm_start;
 	for (ofs = 0; ofs < queue->queue_length; ofs += PAGE_SIZE) {
 		u64 virt_addr = (u64)ipz_qeit_calc(queue, ofs);
diff --git a/drivers/infiniband/hw/ehca/hcp_if.c b/drivers/infiniband/hw/ehca/hcp_if.c
index 415d3a465de..89517ffb438 100644
--- a/drivers/infiniband/hw/ehca/hcp_if.c
+++ b/drivers/infiniband/hw/ehca/hcp_if.c
@@ -90,26 +90,6 @@
 
 static DEFINE_SPINLOCK(hcall_lock);
 
-static u32 get_longbusy_msecs(int longbusy_rc)
-{
-	switch (longbusy_rc) {
-	case H_LONG_BUSY_ORDER_1_MSEC:
-		return 1;
-	case H_LONG_BUSY_ORDER_10_MSEC:
-		return 10;
-	case H_LONG_BUSY_ORDER_100_MSEC:
-		return 100;
-	case H_LONG_BUSY_ORDER_1_SEC:
-		return 1000;
-	case H_LONG_BUSY_ORDER_10_SEC:
-		return 10000;
-	case H_LONG_BUSY_ORDER_100_SEC:
-		return 100000;
-	default:
-		return 1;
-	}
-}
-
 static long ehca_plpar_hcall_norets(unsigned long opcode,
 				    unsigned long arg1,
 				    unsigned long arg2,
@@ -226,7 +206,7 @@ u64 hipz_h_alloc_resource_eq(const struct ipz_adapter_handle adapter_handle,
 			     u32 *eq_ist)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 	u64 allocate_controls;
 
 	/* resource type */
@@ -249,7 +229,7 @@ u64 hipz_h_alloc_resource_eq(const struct ipz_adapter_handle adapter_handle,
 	*eq_ist = (u32)outs[5];
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
-		ehca_gen_err("Not enough resource - ret=%li ", ret);
+		ehca_gen_err("Not enough resource - ret=%lli ", ret);
 
 	return ret;
 }
@@ -269,8 +249,9 @@ u64 hipz_h_alloc_resource_cq(const struct ipz_adapter_handle adapter_handle,
 			     struct ehca_cq *cq,
 			     struct ehca_alloc_cq_parms *param)
 {
+	int rc;
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
 				adapter_handle.handle,   /* r4  */
@@ -283,21 +264,33 @@ u64 hipz_h_alloc_resource_cq(const struct ipz_adapter_handle adapter_handle,
 	param->act_nr_of_entries = (u32)outs[3];
 	param->act_pages = (u32)outs[4];
 
-	if (ret == H_SUCCESS)
-		hcp_galpas_ctor(&cq->galpas, outs[5], outs[6]);
+	if (ret == H_SUCCESS) {
+		rc = hcp_galpas_ctor(&cq->galpas, 0, outs[5], outs[6]);
+		if (rc) {
+			ehca_gen_err("Could not establish HW access. rc=%d paddr=%#lx",
+				     rc, outs[5]);
+
+			ehca_plpar_hcall_norets(H_FREE_RESOURCE,
+						adapter_handle.handle,     /* r4 */
+						cq->ipz_cq_handle.handle,  /* r5 */
+						0, 0, 0, 0, 0);
+			ret = H_NO_MEM;
+		}
+	}
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
-		ehca_gen_err("Not enough resources. ret=%li", ret);
+		ehca_gen_err("Not enough resources. ret=%lli", ret);
 
 	return ret;
 }
 
 u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
-			     struct ehca_alloc_qp_parms *parms)
+			     struct ehca_alloc_qp_parms *parms, int is_user)
 {
+	int rc;
 	u64 ret;
 	u64 allocate_controls, max_r10_reg, r11, r12;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	allocate_controls =
 		EHCA_BMASK_SET(H_ALL_RES_QP_ENHANCED_OPS, parms->ext_type)
@@ -358,11 +351,22 @@ u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
 	parms->rqueue.queue_size =
 		(u32)EHCA_BMASK_GET(H_ALL_RES_QP_RQUEUE_SIZE_PAGES, outs[4]);
 
-	if (ret == H_SUCCESS)
-		hcp_galpas_ctor(&parms->galpas, outs[6], outs[6]);
+	if (ret == H_SUCCESS) {
+		rc = hcp_galpas_ctor(&parms->galpas, is_user, outs[6], outs[6]);
+		if (rc) {
+			ehca_gen_err("Could not establish HW access. rc=%d paddr=%#lx",
+				     rc, outs[6]);
+
+			ehca_plpar_hcall_norets(H_FREE_RESOURCE,
+						adapter_handle.handle,     /* r4 */
+						parms->qp_handle.handle,  /* r5 */
+						0, 0, 0, 0, 0);
+			ret = H_NO_MEM;
+		}
+	}
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
-		ehca_gen_err("Not enough resources. ret=%li", ret);
+		ehca_gen_err("Not enough resources. ret=%lli", ret);
 
 	return ret;
 }
@@ -372,7 +376,7 @@ u64 hipz_h_query_port(const struct ipz_adapter_handle adapter_handle,
 		      struct hipz_query_port *query_port_response_block)
 {
 	u64 ret;
-	u64 r_cb = virt_to_abs(query_port_response_block);
+	u64 r_cb = __pa(query_port_response_block);
 
 	if (r_cb & (EHCA_PAGESIZE-1)) {
 		ehca_gen_err("response block not page aligned");
@@ -414,7 +418,7 @@ u64 hipz_h_modify_port(const struct ipz_adapter_handle adapter_handle,
 u64 hipz_h_query_hca(const struct ipz_adapter_handle adapter_handle,
 		     struct hipz_query_hca *query_hca_rblock)
 {
-	u64 r_cb = virt_to_abs(query_hca_rblock);
+	u64 r_cb = __pa(query_hca_rblock);
 
 	if (r_cb & (EHCA_PAGESIZE-1)) {
 		ehca_gen_err("response_block=%p not page aligned",
@@ -454,7 +458,7 @@ u64 hipz_h_register_rpage_eq(const struct ipz_adapter_handle adapter_handle,
 			     const u64 count)
 {
 	if (count != 1) {
-		ehca_gen_err("Ppage counter=%lx", count);
+		ehca_gen_err("Ppage counter=%llx", count);
 		return H_PARAMETER;
 	}
 	return hipz_h_register_rpage(adapter_handle,
@@ -489,7 +493,7 @@ u64 hipz_h_register_rpage_cq(const struct ipz_adapter_handle adapter_handle,
 			     const struct h_galpa gal)
 {
 	if (count != 1) {
-		ehca_gen_err("Page counter=%lx", count);
+		ehca_gen_err("Page counter=%llx", count);
 		return H_PARAMETER;
 	}
 
@@ -508,7 +512,7 @@ u64 hipz_h_register_rpage_qp(const struct ipz_adapter_handle adapter_handle,
 			     const struct h_galpa galpa)
 {
 	if (count > 1) {
-		ehca_gen_err("Page counter=%lx", count);
+		ehca_gen_err("Page counter=%llx", count);
 		return H_PARAMETER;
 	}
 
@@ -525,7 +529,7 @@ u64 hipz_h_disable_and_get_wqe(const struct ipz_adapter_handle adapter_handle,
 			       int dis_and_get_function_code)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_DISABLE_AND_GETC, outs,
 				adapter_handle.handle,     /* r4 */
@@ -548,16 +552,16 @@ u64 hipz_h_modify_qp(const struct ipz_adapter_handle adapter_handle,
 		     struct h_galpa gal)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 	ret = ehca_plpar_hcall9(H_MODIFY_QP, outs,
 				adapter_handle.handle, /* r4 */
 				qp_handle.handle,      /* r5 */
 				update_mask,	       /* r6 */
-				virt_to_abs(mqpcb),    /* r7 */
+				__pa(mqpcb),	       /* r7 */
 				0, 0, 0, 0, 0);
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
-		ehca_gen_err("Insufficient resources ret=%li", ret);
+		ehca_gen_err("Insufficient resources ret=%lli", ret);
 
 	return ret;
 }
@@ -571,7 +575,7 @@ u64 hipz_h_query_qp(const struct ipz_adapter_handle adapter_handle,
 	return ehca_plpar_hcall_norets(H_QUERY_QP,
 				       adapter_handle.handle, /* r4 */
 				       qp_handle.handle,      /* r5 */
-				       virt_to_abs(qqpcb),    /* r6 */
+				       __pa(qqpcb),	      /* r6 */
 				       0, 0, 0, 0);
 }
 
@@ -579,7 +583,7 @@ u64 hipz_h_destroy_qp(const struct ipz_adapter_handle adapter_handle,
 		      struct ehca_qp *qp)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = hcp_galpas_dtor(&qp->galpas);
 	if (ret) {
@@ -593,7 +597,7 @@ u64 hipz_h_destroy_qp(const struct ipz_adapter_handle adapter_handle,
 				qp->ipz_qp_handle.handle,  /* r6 */
 				0, 0, 0, 0, 0, 0);
 	if (ret == H_HARDWARE)
-		ehca_gen_err("HCA not operational. ret=%li", ret);
+		ehca_gen_err("HCA not operational. ret=%lli", ret);
 
 	ret = ehca_plpar_hcall_norets(H_FREE_RESOURCE,
 				      adapter_handle.handle,     /* r4 */
@@ -601,7 +605,7 @@ u64 hipz_h_destroy_qp(const struct ipz_adapter_handle adapter_handle,
 				      0, 0, 0, 0, 0);
 
 	if (ret == H_RESOURCE)
-		ehca_gen_err("Resource still in use. ret=%li", ret);
+		ehca_gen_err("Resource still in use. ret=%lli", ret);
 
 	return ret;
 }
@@ -625,7 +629,7 @@ u64 hipz_h_define_aqp1(const struct ipz_adapter_handle adapter_handle,
 		       u32 * bma_qp_nr)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_DEFINE_AQP1, outs,
 				adapter_handle.handle, /* r4 */
@@ -636,7 +640,7 @@ u64 hipz_h_define_aqp1(const struct ipz_adapter_handle adapter_handle,
 	*bma_qp_nr = (u32)outs[1];
 
 	if (ret == H_ALIAS_EXIST)
-		ehca_gen_err("AQP1 already exists. ret=%li", ret);
+		ehca_gen_err("AQP1 already exists. ret=%lli", ret);
 
 	return ret;
 }
@@ -658,7 +662,7 @@ u64 hipz_h_attach_mcqp(const struct ipz_adapter_handle adapter_handle,
 				      0, 0);
 
 	if (ret == H_NOT_ENOUGH_RESOURCES)
-		ehca_gen_err("Not enough resources. ret=%li", ret);
+		ehca_gen_err("Not enough resources. ret=%lli", ret);
 
 	return ret;
 }
@@ -697,7 +701,7 @@ u64 hipz_h_destroy_cq(const struct ipz_adapter_handle adapter_handle,
 				      0, 0, 0, 0);
 
 	if (ret == H_RESOURCE)
-		ehca_gen_err("H_FREE_RESOURCE failed ret=%li ", ret);
+		ehca_gen_err("H_FREE_RESOURCE failed ret=%lli ", ret);
 
 	return ret;
 }
@@ -719,7 +723,7 @@ u64 hipz_h_destroy_eq(const struct ipz_adapter_handle adapter_handle,
 				      0, 0, 0, 0, 0);
 
 	if (ret == H_RESOURCE)
-		ehca_gen_err("Resource in use. ret=%li ", ret);
+		ehca_gen_err("Resource in use. ret=%lli ", ret);
 
 	return ret;
 }
@@ -733,7 +737,7 @@ u64 hipz_h_alloc_resource_mr(const struct ipz_adapter_handle adapter_handle,
 			     struct ehca_mr_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
 				adapter_handle.handle,            /* r4 */
@@ -763,7 +767,7 @@ u64 hipz_h_register_rpage_mr(const struct ipz_adapter_handle adapter_handle,
 		if (count > 1) {
 			u64 *kpage;
 			int i;
-			kpage = (u64 *)abs_to_virt(logical_address_of_page);
+			kpage = __va(logical_address_of_page);
 			for (i = 0; i < count; i++)
 				ehca_gen_dbg("kpage[%d]=%p",
 					     i, (void *)kpage[i]);
@@ -774,9 +778,9 @@ u64 hipz_h_register_rpage_mr(const struct ipz_adapter_handle adapter_handle,
 
 	if ((count > 1) && (logical_address_of_page & (EHCA_PAGESIZE-1))) {
 		ehca_gen_err("logical_address_of_page not on a 4k boundary "
-			     "adapter_handle=%lx mr=%p mr_handle=%lx "
+			     "adapter_handle=%llx mr=%p mr_handle=%llx "
 			     "pagesize=%x queue_type=%x "
-			     "logical_address_of_page=%lx count=%lx",
+			     "logical_address_of_page=%llx count=%llx",
 			     adapter_handle.handle, mr,
 			     mr->ipz_mr_handle.handle, pagesize, queue_type,
 			     logical_address_of_page, count);
@@ -794,7 +798,7 @@ u64 hipz_h_query_mr(const struct ipz_adapter_handle adapter_handle,
 		    struct ehca_mr_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_QUERY_MR, outs,
 				adapter_handle.handle,     /* r4 */
@@ -828,7 +832,7 @@ u64 hipz_h_reregister_pmr(const struct ipz_adapter_handle adapter_handle,
 			  struct ehca_mr_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_REREGISTER_PMR, outs,
 				adapter_handle.handle,    /* r4 */
@@ -855,7 +859,7 @@ u64 hipz_h_register_smr(const struct ipz_adapter_handle adapter_handle,
 			struct ehca_mr_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_REGISTER_SMR, outs,
 				adapter_handle.handle,            /* r4 */
@@ -877,7 +881,7 @@ u64 hipz_h_alloc_resource_mw(const struct ipz_adapter_handle adapter_handle,
 			     struct ehca_mw_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_ALLOC_RESOURCE, outs,
 				adapter_handle.handle,      /* r4 */
@@ -895,7 +899,7 @@ u64 hipz_h_query_mw(const struct ipz_adapter_handle adapter_handle,
 		    struct ehca_mw_hipzout_parms *outparms)
 {
 	u64 ret;
-	u64 outs[PLPAR_HCALL9_BUFSIZE];
+	unsigned long outs[PLPAR_HCALL9_BUFSIZE];
 
 	ret = ehca_plpar_hcall9(H_QUERY_MW, outs,
 				adapter_handle.handle,    /* r4 */
@@ -920,7 +924,7 @@ u64 hipz_h_error_data(const struct ipz_adapter_handle adapter_handle,
 		      void *rblock,
 		      unsigned long *byte_count)
 {
-	u64 r_cb = virt_to_abs(rblock);
+	u64 r_cb = __pa(rblock);
 
 	if (r_cb & (EHCA_PAGESIZE-1)) {
 		ehca_gen_err("rblock not page aligned.");
diff --git a/drivers/infiniband/hw/ehca/hcp_if.h b/drivers/infiniband/hw/ehca/hcp_if.h
index 2c3c6e0ea5c..a46e514c367 100644
--- a/drivers/infiniband/hw/ehca/hcp_if.h
+++ b/drivers/infiniband/hw/ehca/hcp_if.h
@@ -49,7 +49,7 @@
 #include "hipz_hw.h"
 
 /*
- * hipz_h_alloc_resource_eq allocates EQ resources in HW and FW, initalize
+ * hipz_h_alloc_resource_eq allocates EQ resources in HW and FW, initialize
  * resources, create the empty EQPT (ring).
  */
 u64 hipz_h_alloc_resource_eq(const struct ipz_adapter_handle adapter_handle,
@@ -78,7 +78,7 @@ u64 hipz_h_alloc_resource_cq(const struct ipz_adapter_handle adapter_handle,
  * initialize resources, create empty QPPTs (2 rings).
  */
 u64 hipz_h_alloc_resource_qp(const struct ipz_adapter_handle adapter_handle,
-			     struct ehca_alloc_qp_parms *parms);
+			     struct ehca_alloc_qp_parms *parms, int is_user);
 
 u64 hipz_h_query_port(const struct ipz_adapter_handle adapter_handle,
 		      const u8 port_id,
diff --git a/drivers/infiniband/hw/ehca/hcp_phyp.c b/drivers/infiniband/hw/ehca/hcp_phyp.c
index 214821095cb..077376ff3d2 100644
--- a/drivers/infiniband/hw/ehca/hcp_phyp.c
+++ b/drivers/infiniband/hw/ehca/hcp_phyp.c
@@ -42,10 +42,9 @@
 #include "ehca_classes.h"
 #include "hipz_hw.h"
 
-int hcall_map_page(u64 physaddr, u64 *mapaddr)
+u64 hcall_map_page(u64 physaddr)
 {
-	*mapaddr = (u64)(ioremap(physaddr, EHCA_PAGESIZE));
-	return 0;
+	return (u64)ioremap(physaddr, EHCA_PAGESIZE);
 }
 
 int hcall_unmap_page(u64 mapaddr)
@@ -54,12 +53,15 @@ int hcall_unmap_page(u64 mapaddr)
 	return 0;
 }
 
-int hcp_galpas_ctor(struct h_galpas *galpas,
+int hcp_galpas_ctor(struct h_galpas *galpas, int is_user,
 		    u64 paddr_kernel, u64 paddr_user)
 {
-	int ret = hcall_map_page(paddr_kernel, &galpas->kernel.fw_handle);
-	if (ret)
-		return ret;
+	if (!is_user) {
+		galpas->kernel.fw_handle = hcall_map_page(paddr_kernel);
+		if (!galpas->kernel.fw_handle)
+			return -ENOMEM;
+	} else
+		galpas->kernel.fw_handle = 0;
 
 	galpas->user.fw_handle = paddr_user;
 
diff --git a/drivers/infiniband/hw/ehca/hcp_phyp.h b/drivers/infiniband/hw/ehca/hcp_phyp.h
index 5305c2a3ed9..d1b02991024 100644
--- a/drivers/infiniband/hw/ehca/hcp_phyp.h
+++ b/drivers/infiniband/hw/ehca/hcp_phyp.h
@@ -78,12 +78,12 @@ static inline void hipz_galpa_store(struct h_galpa galpa, u32 offset, u64 value)
 	*(volatile u64 __force *)addr = value;
 }
 
-int hcp_galpas_ctor(struct h_galpas *galpas,
+int hcp_galpas_ctor(struct h_galpas *galpas, int is_user,
 		    u64 paddr_kernel, u64 paddr_user);
 
 int hcp_galpas_dtor(struct h_galpas *galpas);
 
-int hcall_map_page(u64 physaddr, u64 * mapaddr);
+u64 hcall_map_page(u64 physaddr);
 
 int hcall_unmap_page(u64 mapaddr);
 
diff --git a/drivers/infiniband/hw/ehca/ipz_pt_fn.c b/drivers/infiniband/hw/ehca/ipz_pt_fn.c
index 661f8db6270..8d594517cd2 100644
--- a/drivers/infiniband/hw/ehca/ipz_pt_fn.c
+++ b/drivers/infiniband/hw/ehca/ipz_pt_fn.c
@@ -38,6 +38,8 @@
  * POSSIBILITY OF SUCH DAMAGE.
  */
 
+#include <linux/slab.h>
+
 #include "ehca_tools.h"
 #include "ipz_pt_fn.h"
 #include "ehca_classes.h"
@@ -79,7 +81,7 @@ int ipz_queue_abs_to_offset(struct ipz_queue *queue, u64 addr, u64 *q_offset)
 {
 	int i;
 	for (i = 0; i < queue->queue_length / queue->pagesize; i++) {
-		u64 page = (u64)virt_to_abs(queue->queue_pages[i]);
+		u64 page = __pa(queue->queue_pages[i]);
 		if (addr >= page && addr < page + queue->pagesize) {
 			*q_offset = addr - page + i * queue->pagesize;
 			return 0;
@@ -163,6 +165,7 @@ static int alloc_small_queue_page(struct ipz_queue *queue, struct ehca_pd *pd)
 
 out:
 	ehca_err(pd->ib_pd.device, "failed to allocate small queue page");
+	mutex_unlock(&pd->lock);
 	return 0;
 }
 
@@ -219,12 +222,15 @@ int ipz_queue_ctor(struct ehca_pd *pd, struct ipz_queue *queue,
 	queue->small_page = NULL;
 
 	/* allocate queue page pointers */
-	queue->queue_pages = vmalloc(nr_of_pages * sizeof(void *));
+	queue->queue_pages = kzalloc(nr_of_pages * sizeof(void *),
+				     GFP_KERNEL | __GFP_NOWARN);
 	if (!queue->queue_pages) {
-		ehca_gen_err("Couldn't allocate queue page list");
-		return 0;
+		queue->queue_pages = vzalloc(nr_of_pages * sizeof(void *));
+		if (!queue->queue_pages) {
+			ehca_gen_err("Couldn't allocate queue page list");
+			return 0;
+		}
 	}
-	memset(queue->queue_pages, 0, nr_of_pages * sizeof(void *));
 
 	/* allocate actual queue pages */
 	if (is_small) {
@@ -239,7 +245,10 @@ int ipz_queue_ctor(struct ehca_pd *pd, struct ipz_queue *queue,
 ipz_queue_ctor_exit0:
 	ehca_gen_err("Couldn't alloc pages queue=%p "
 		 "nr_of_pages=%x",  queue, nr_of_pages);
-	vfree(queue->queue_pages);
+	if (is_vmalloc_addr(queue->queue_pages))
+		vfree(queue->queue_pages);
+	else
+		kfree(queue->queue_pages);
 
 	return 0;
 }
@@ -261,7 +270,10 @@ int ipz_queue_dtor(struct ehca_pd *pd, struct ipz_queue *queue)
 			free_page((unsigned long)queue->queue_pages[i]);
 	}
 
-	vfree(queue->queue_pages);
+	if (is_vmalloc_addr(queue->queue_pages))
+		vfree(queue->queue_pages);
+	else
+		kfree(queue->queue_pages);
 
 	return 1;
 }
diff --git a/drivers/infiniband/hw/ipath/Kconfig b/drivers/infiniband/hw/ipath/Kconfig
index 3c7968f25ec..1d9bb115cbf 100644
--- a/drivers/infiniband/hw/ipath/Kconfig
+++ b/drivers/infiniband/hw/ipath/Kconfig
@@ -1,9 +1,11 @@
 config INFINIBAND_IPATH
-	tristate "QLogic InfiniPath Driver"
-	depends on 64BIT && NET
+	tristate "QLogic HTX HCA support"
+	depends on 64BIT && NET && HT_IRQ
 	---help---
-	This is a driver for QLogic InfiniPath host channel adapters,
+	This is a driver for the obsolete QLogic Hyper-Transport
+	IB host channel adapter (model QHT7140),
 	including InfiniBand verbs support.  This driver allows these
 	devices to be used with both kernel upper level protocols such
 	as IP-over-InfiniBand as well as with userspace applications
 	(in conjunction with InfiniBand userspace access).
+	For QLogic PCIe QLE based cards, use the QIB driver instead.
diff --git a/drivers/infiniband/hw/ipath/Makefile b/drivers/infiniband/hw/ipath/Makefile
index bf945006198..4496f2820c9 100644
--- a/drivers/infiniband/hw/ipath/Makefile
+++ b/drivers/infiniband/hw/ipath/Makefile
@@ -1,4 +1,4 @@
-EXTRA_CFLAGS += -DIPATH_IDSTR='"QLogic kernel.org driver"' \
+ccflags-y := -DIPATH_IDSTR='"QLogic kernel.org driver"' \
 	-DIPATH_KERN_TYPE=0
 
 obj-$(CONFIG_INFINIBAND_IPATH) += ib_ipath.o
@@ -29,13 +29,9 @@ ib_ipath-y := \
 	ipath_user_pages.o \
 	ipath_user_sdma.o \
 	ipath_verbs_mcast.o \
-	ipath_verbs.o \
-	ipath_iba7220.o \
-	ipath_sd7220.o \
-	ipath_sd7220_img.o
+	ipath_verbs.o
 
 ib_ipath-$(CONFIG_HT_IRQ) += ipath_iba6110.o
-ib_ipath-$(CONFIG_PCI_MSI) += ipath_iba6120.o
 
 ib_ipath-$(CONFIG_X86_64) += ipath_wc_x86_64.o
 ib_ipath-$(CONFIG_PPC64) += ipath_wc_ppc64.o
diff --git a/drivers/infiniband/hw/ipath/ipath_cq.c b/drivers/infiniband/hw/ipath/ipath_cq.c
index d385e4168c9..0416c6c0e12 100644
--- a/drivers/infiniband/hw/ipath/ipath_cq.c
+++ b/drivers/infiniband/hw/ipath/ipath_cq.c
@@ -32,6 +32,7 @@
  */
 
 #include <linux/err.h>
+#include <linux/slab.h>
 #include <linux/vmalloc.h>
 
 #include "ipath_verbs.h"
diff --git a/drivers/infiniband/hw/ipath/ipath_diag.c b/drivers/infiniband/hw/ipath/ipath_diag.c
index d4ce8b63e19..45802e97332 100644
--- a/drivers/infiniband/hw/ipath/ipath_diag.c
+++ b/drivers/infiniband/hw/ipath/ipath_diag.c
@@ -45,6 +45,7 @@
 #include <linux/pci.h>
 #include <linux/vmalloc.h>
 #include <linux/fs.h>
+#include <linux/export.h>
 #include <asm/uaccess.h>
 
 #include "ipath_kernel.h"
@@ -65,7 +66,8 @@ static const struct file_operations diag_file_ops = {
 	.write = ipath_diag_write,
 	.read = ipath_diag_read,
 	.open = ipath_diag_open,
-	.release = ipath_diag_release
+	.release = ipath_diag_release,
+	.llseek = default_llseek,
 };
 
 static ssize_t ipath_diagpkt_write(struct file *fp,
@@ -75,6 +77,7 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 static const struct file_operations diagpkt_file_ops = {
 	.owner = THIS_MODULE,
 	.write = ipath_diagpkt_write,
+	.llseek = noop_llseek,
 };
 
 static atomic_t diagpkt_count = ATOMIC_INIT(0);
@@ -323,7 +326,7 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 				   size_t count, loff_t *off)
 {
 	u32 __iomem *piobuf;
-	u32 plen, clen, pbufn;
+	u32 plen, pbufn, maxlen_reserve;
 	struct ipath_diag_pkt odp;
 	struct ipath_diag_xpkt dp;
 	u32 *tmpbuf = NULL;
@@ -332,42 +335,24 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 	u64 val;
 	u32 l_state, lt_state; /* LinkState, LinkTrainingState */
 
-	if (count < sizeof(odp)) {
-		ret = -EINVAL;
-		goto bail;
-	}
 
 	if (count == sizeof(dp)) {
 		if (copy_from_user(&dp, data, sizeof(dp))) {
 			ret = -EFAULT;
 			goto bail;
 		}
-	} else if (copy_from_user(&odp, data, sizeof(odp))) {
-		ret = -EFAULT;
-		goto bail;
-	}
-
-	/*
-	 * Due to padding/alignment issues (lessened with new struct)
-	 * the old and new structs are the same length. We need to
-	 * disambiguate them, which we can do because odp.len has never
-	 * been less than the total of LRH+BTH+DETH so far, while
-	 * dp.unit (same offset) unit is unlikely to get that high.
-	 * Similarly, dp.data, the pointer to user at the same offset
-	 * as odp.unit, is almost certainly at least one (512byte)page
-	 * "above" NULL. The if-block below can be omitted if compatibility
-	 * between a new driver and older diagnostic code is unimportant.
-	 * compatibility the other direction (new diags, old driver) is
-	 * handled in the diagnostic code, with a warning.
-	 */
-	if (dp.unit >= 20 && dp.data < 512) {
-		/* very probable version mismatch. Fix it up */
-		memcpy(&odp, &dp, sizeof(odp));
-		/* We got a legacy dp, copy elements to dp */
+	} else if (count == sizeof(odp)) {
+		if (copy_from_user(&odp, data, sizeof(odp))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		dp.len = odp.len;
 		dp.unit = odp.unit;
 		dp.data = odp.data;
-		dp.len = odp.len;
-		dp.pbc_wd = 0; /* Indicate we need to compute PBC wd */
+		dp.pbc_wd = 0;
+	} else {
+		ret = -EINVAL;
+		goto bail;
 	}
 
 	/* send count must be an exact number of dwords */
@@ -376,7 +361,7 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 		goto bail;
 	}
 
-	clen = dp.len >> 2;
+	plen = dp.len >> 2;
 
 	dd = ipath_lookup(dp.unit);
 	if (!dd || !(dd->ipath_flags & IPATH_PRESENT) ||
@@ -419,16 +404,22 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 		goto bail;
 	}
 
-	/* need total length before first word written */
-	/* +1 word is for the qword padding */
-	plen = sizeof(u32) + dp.len;
-
-	if ((plen + 4) > dd->ipath_ibmaxlen) {
+	/*
+	 * need total length before first word written, plus 2 Dwords. One Dword
+	 * is for padding so we get the full user data when not aligned on
+	 * a word boundary. The other Dword is to make sure we have room for the
+	 * ICRC which gets tacked on later.
+	 */
+	maxlen_reserve = 2 * sizeof(u32);
+	if (dp.len > dd->ipath_ibmaxlen - maxlen_reserve) {
 		ipath_dbg("Pkt len 0x%x > ibmaxlen %x\n",
-			  plen - 4, dd->ipath_ibmaxlen);
+			  dp.len, dd->ipath_ibmaxlen);
 		ret = -EINVAL;
-		goto bail;	/* before writing pbc */
+		goto bail;
 	}
+
+	plen = sizeof(u32) + dp.len;
+
 	tmpbuf = vmalloc(plen);
 	if (!tmpbuf) {
 		dev_info(&dd->pcidev->dev, "Unable to allocate tmp buffer, "
@@ -470,11 +461,11 @@ static ssize_t ipath_diagpkt_write(struct file *fp,
 	 */
 	if (dd->ipath_flags & IPATH_PIO_FLUSH_WC) {
 		ipath_flush_wc();
-		__iowrite32_copy(piobuf + 2, tmpbuf, clen - 1);
+		__iowrite32_copy(piobuf + 2, tmpbuf, plen - 1);
 		ipath_flush_wc();
-		__raw_writel(tmpbuf[clen - 1], piobuf + clen + 1);
+		__raw_writel(tmpbuf[plen - 1], piobuf + plen + 1);
 	} else
-		__iowrite32_copy(piobuf + 2, tmpbuf, clen);
+		__iowrite32_copy(piobuf + 2, tmpbuf, plen);
 
 	ipath_flush_wc();
 
diff --git a/drivers/infiniband/hw/ipath/ipath_dma.c b/drivers/infiniband/hw/ipath/ipath_dma.c
index e90a0ea538a..123a8c05353 100644
--- a/drivers/infiniband/hw/ipath/ipath_dma.c
+++ b/drivers/infiniband/hw/ipath/ipath_dma.c
@@ -31,6 +31,7 @@
  */
 
 #include <linux/scatterlist.h>
+#include <linux/gfp.h>
 #include <rdma/ib_verbs.h>
 
 #include "ipath_verbs.h"
@@ -114,6 +115,10 @@ static int ipath_map_sg(struct ib_device *dev, struct scatterlist *sgl,
 			ret = 0;
 			break;
 		}
+		sg->dma_address = addr + sg->offset;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+		sg->dma_length = sg->length;
+#endif
 	}
 	return ret;
 }
@@ -125,21 +130,6 @@ static void ipath_unmap_sg(struct ib_device *dev,
 	BUG_ON(!valid_dma_direction(direction));
 }
 
-static u64 ipath_sg_dma_address(struct ib_device *dev, struct scatterlist *sg)
-{
-	u64 addr = (u64) page_address(sg_page(sg));
-
-	if (addr)
-		addr += sg->offset;
-	return addr;
-}
-
-static unsigned int ipath_sg_dma_len(struct ib_device *dev,
-				     struct scatterlist *sg)
-{
-	return sg->length;
-}
-
 static void ipath_sync_single_for_cpu(struct ib_device *dev,
 				      u64 addr,
 				      size_t size,
@@ -175,17 +165,15 @@ static void ipath_dma_free_coherent(struct ib_device *dev, size_t size,
 }
 
 struct ib_dma_mapping_ops ipath_dma_mapping_ops = {
-	ipath_mapping_error,
-	ipath_dma_map_single,
-	ipath_dma_unmap_single,
-	ipath_dma_map_page,
-	ipath_dma_unmap_page,
-	ipath_map_sg,
-	ipath_unmap_sg,
-	ipath_sg_dma_address,
-	ipath_sg_dma_len,
-	ipath_sync_single_for_cpu,
-	ipath_sync_single_for_device,
-	ipath_dma_alloc_coherent,
-	ipath_dma_free_coherent
+	.mapping_error = ipath_mapping_error,
+	.map_single = ipath_dma_map_single,
+	.unmap_single = ipath_dma_unmap_single,
+	.map_page = ipath_dma_map_page,
+	.unmap_page = ipath_dma_unmap_page,
+	.map_sg = ipath_map_sg,
+	.unmap_sg = ipath_unmap_sg,
+	.sync_single_for_cpu = ipath_sync_single_for_cpu,
+	.sync_single_for_device = ipath_sync_single_for_device,
+	.alloc_coherent = ipath_dma_alloc_coherent,
+	.free_coherent = ipath_dma_free_coherent
 };
diff --git a/drivers/infiniband/hw/ipath/ipath_driver.c b/drivers/infiniband/hw/ipath/ipath_driver.c
index daad09a4591..bd0caedafe9 100644
--- a/drivers/infiniband/hw/ipath/ipath_driver.c
+++ b/drivers/infiniband/hw/ipath/ipath_driver.c
@@ -31,6 +31,7 @@
  * SOFTWARE.
  */
 
+#include <linux/sched.h>
 #include <linux/spinlock.h>
 #include <linux/idr.h>
 #include <linux/pci.h>
@@ -38,6 +39,9 @@
 #include <linux/delay.h>
 #include <linux/netdevice.h>
 #include <linux/vmalloc.h>
+#include <linux/bitmap.h>
+#include <linux/slab.h>
+#include <linux/module.h>
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
@@ -123,24 +127,18 @@ const char *ipath_ibcstatus_str[] = {
 	"LTState1C", "LTState1D", "LTState1E", "LTState1F"
 };
 
-static void __devexit ipath_remove_one(struct pci_dev *);
-static int __devinit ipath_init_one(struct pci_dev *,
-				    const struct pci_device_id *);
+static void ipath_remove_one(struct pci_dev *);
+static int ipath_init_one(struct pci_dev *, const struct pci_device_id *);
 
 /* Only needed for registration, nothing else needs this info */
 #define PCI_VENDOR_ID_PATHSCALE 0x1fc1
-#define PCI_VENDOR_ID_QLOGIC 0x1077
 #define PCI_DEVICE_ID_INFINIPATH_HT 0xd
-#define PCI_DEVICE_ID_INFINIPATH_PE800 0x10
-#define PCI_DEVICE_ID_INFINIPATH_7220 0x7220
 
 /* Number of seconds before our card status check...  */
 #define STATUS_TIMEOUT 60
 
 static const struct pci_device_id ipath_pci_tbl[] = {
 	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_HT) },
-	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_INFINIPATH_PE800) },
-	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_INFINIPATH_7220) },
 	{ 0, }
 };
 
@@ -149,7 +147,7 @@ MODULE_DEVICE_TABLE(pci, ipath_pci_tbl);
 static struct pci_driver ipath_driver = {
 	.name = IPATH_DRV_NAME,
 	.probe = ipath_init_one,
-	.remove = __devexit_p(ipath_remove_one),
+	.remove = ipath_remove_one,
 	.id_table = ipath_pci_tbl,
 	.driver = {
 		.groups = ipath_driver_attr_groups,
@@ -196,22 +194,17 @@ static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
 	struct ipath_devdata *dd;
 	int ret;
 
-	if (!idr_pre_get(&unit_table, GFP_KERNEL)) {
-		dd = ERR_PTR(-ENOMEM);
-		goto bail;
-	}
-
-	dd = vmalloc(sizeof(*dd));
+	dd = vzalloc(sizeof(*dd));
 	if (!dd) {
 		dd = ERR_PTR(-ENOMEM);
 		goto bail;
 	}
-	memset(dd, 0, sizeof(*dd));
 	dd->ipath_unit = -1;
 
+	idr_preload(GFP_KERNEL);
 	spin_lock_irqsave(&ipath_devs_lock, flags);
 
-	ret = idr_get_new(&unit_table, dd, &dd->ipath_unit);
+	ret = idr_alloc(&unit_table, dd, 0, 0, GFP_NOWAIT);
 	if (ret < 0) {
 		printk(KERN_ERR IPATH_DRV_NAME
 		       ": Could not allocate unit ID: error %d\n", -ret);
@@ -219,6 +212,7 @@ static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
 		dd = ERR_PTR(ret);
 		goto bail_unlock;
 	}
+	dd->ipath_unit = ret;
 
 	dd->pcidev = pdev;
 	pci_set_drvdata(pdev, dd);
@@ -227,7 +221,7 @@ static struct ipath_devdata *ipath_alloc_devdata(struct pci_dev *pdev)
 
 bail_unlock:
 	spin_unlock_irqrestore(&ipath_devs_lock, flags);
-
+	idr_preload_end();
 bail:
 	return dd;
 }
@@ -392,14 +386,14 @@ done:
 	ipath_enable_armlaunch(dd);
 }
 
-static int __devinit ipath_init_one(struct pci_dev *pdev,
-				    const struct pci_device_id *ent)
+static void cleanup_device(struct ipath_devdata *dd);
+
+static int ipath_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
 {
 	int ret, len, j;
 	struct ipath_devdata *dd;
 	unsigned long long addr;
 	u32 bar0 = 0, bar1 = 0;
-	u8 rev;
 
 	dd = ipath_alloc_devdata(pdev);
 	if (IS_ERR(dd)) {
@@ -470,14 +464,14 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 		goto bail_disable;
 	}
 
-	ret = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
+	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
 	if (ret) {
 		/*
 		 * if the 64 bit setup fails, try 32 bit.  Some systems
 		 * do not setup 64 bit maps on systems with 2GB or less
 		 * memory installed.
 		 */
-		ret = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
 		if (ret) {
 			dev_info(&pdev->dev,
 				"Unable to set DMA mask for unit %u: %d\n",
@@ -486,7 +480,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 		}
 		else {
 			ipath_dbg("No 64bit DMA mask, used 32 bit mask\n");
-			ret = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
+			ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
 			if (ret)
 				dev_info(&pdev->dev,
 					"Unable to set DMA consistent mask "
@@ -496,7 +490,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 		}
 	}
 	else {
-		ret = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
 		if (ret)
 			dev_info(&pdev->dev,
 				"Unable to set DMA consistent mask "
@@ -518,30 +512,9 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 	/* setup the chip-specific functions, as early as possible. */
 	switch (ent->device) {
 	case PCI_DEVICE_ID_INFINIPATH_HT:
-#ifdef CONFIG_HT_IRQ
 		ipath_init_iba6110_funcs(dd);
 		break;
-#else
-		ipath_dev_err(dd, "QLogic HT device 0x%x cannot work if "
-			      "CONFIG_HT_IRQ is not enabled\n", ent->device);
-		return -ENODEV;
-#endif
-	case PCI_DEVICE_ID_INFINIPATH_PE800:
-#ifdef CONFIG_PCI_MSI
-		ipath_init_iba6120_funcs(dd);
-		break;
-#else
-		ipath_dev_err(dd, "QLogic PCIE device 0x%x cannot work if "
-			      "CONFIG_PCI_MSI is not enabled\n", ent->device);
-		return -ENODEV;
-#endif
-	case PCI_DEVICE_ID_INFINIPATH_7220:
-#ifndef CONFIG_PCI_MSI
-		ipath_dbg("CONFIG_PCI_MSI is not enabled, "
-			  "using INTx for unit %u\n", dd->ipath_unit);
-#endif
-		ipath_init_iba7220_funcs(dd);
-		break;
+
 	default:
 		ipath_dev_err(dd, "Found unknown QLogic deviceid 0x%x, "
 			      "failing\n", ent->device);
@@ -551,9 +524,8 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 	for (j = 0; j < 6; j++) {
 		if (!pdev->resource[j].start)
 			continue;
-		ipath_cdbg(VERBOSE, "BAR %d start %llx, end %llx, len %llx\n",
-			   j, (unsigned long long)pdev->resource[j].start,
-			   (unsigned long long)pdev->resource[j].end,
+		ipath_cdbg(VERBOSE, "BAR %d %pR, len %llx\n",
+			   j, &pdev->resource[j],
 			   (unsigned long long)pci_resource_len(pdev, j));
 	}
 
@@ -563,13 +535,7 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 		goto bail_regions;
 	}
 
-	ret = pci_read_config_byte(pdev, PCI_REVISION_ID, &rev);
-	if (ret) {
-		ipath_dev_err(dd, "Failed to read PCI revision ID unit "
-			      "%u: err %d\n", dd->ipath_unit, -ret);
-		goto bail_regions;	/* shouldn't ever happen */
-	}
-	dd->ipath_pcirev = rev;
+	dd->ipath_pcirev = pdev->revision;
 
 #if defined(__powerpc__)
 	/* There isn't a generic way to specify writethrough mappings */
@@ -639,8 +605,13 @@ static int __devinit ipath_init_one(struct pci_dev *pdev,
 	goto bail;
 
 bail_irqsetup:
-	if (pdev->irq)
-		free_irq(pdev->irq, dd);
+	cleanup_device(dd);
+
+	if (dd->ipath_irq)
+		dd->ipath_f_free_irq(dd);
+
+	if (dd->ipath_f_cleanup)
+		dd->ipath_f_cleanup(dd);
 
 bail_iounmap:
 	iounmap((volatile void __iomem *) dd->ipath_kregbase);
@@ -658,9 +629,11 @@ bail:
 	return ret;
 }
 
-static void __devexit cleanup_device(struct ipath_devdata *dd)
+static void cleanup_device(struct ipath_devdata *dd)
 {
 	int port;
+	struct ipath_portdata **tmp;
+	unsigned long flags;
 
 	if (*dd->ipath_statusp & IPATH_STATUS_CHIP_PRESENT) {
 		/* can't do anything more with chip; needs re-init */
@@ -742,23 +715,24 @@ static void __devexit cleanup_device(struct ipath_devdata *dd)
 
 	/*
 	 * free any resources still in use (usually just kernel ports)
-	 * at unload; we do for portcnt, not cfgports, because cfgports
-	 * could have changed while we were loaded.
+	 * at unload; we do for portcnt, because that's what we allocate.
+	 * We acquire lock to be really paranoid that ipath_pd isn't being
+	 * accessed from some interrupt-related code (that should not happen,
+	 * but best to be sure).
 	 */
+	spin_lock_irqsave(&dd->ipath_uctxt_lock, flags);
+	tmp = dd->ipath_pd;
+	dd->ipath_pd = NULL;
+	spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
 	for (port = 0; port < dd->ipath_portcnt; port++) {
-		struct ipath_portdata *pd = dd->ipath_pd[port];
-		dd->ipath_pd[port] = NULL;
+		struct ipath_portdata *pd = tmp[port];
+		tmp[port] = NULL; /* debugging paranoia */
 		ipath_free_pddata(dd, pd);
 	}
-	kfree(dd->ipath_pd);
-	/*
-	 * debuggability, in case some cleanup path tries to use it
-	 * after this
-	 */
-	dd->ipath_pd = NULL;
+	kfree(tmp);
 }
 
-static void __devexit ipath_remove_one(struct pci_dev *pdev)
+static void ipath_remove_one(struct pci_dev *pdev)
 {
 	struct ipath_devdata *dd = pci_get_drvdata(pdev);
 
@@ -770,7 +744,7 @@ static void __devexit ipath_remove_one(struct pci_dev *pdev)
 	 */
 	ipath_shutdown_device(dd);
 
-	flush_scheduled_work();
+	flush_workqueue(ib_wq);
 
 	if (dd->verbs_dev)
 		ipath_unregister_ib_device(dd->verbs_dev);
@@ -1259,7 +1233,7 @@ reloop:
 			 */
 			ipath_cdbg(ERRPKT, "Error Pkt, but no eflags! egrbuf"
 				  " %x, len %x hdrq+%x rhf: %Lx\n",
-				  etail, tlen, l,
+				  etail, tlen, l, (unsigned long long)
 				  le64_to_cpu(*(__le64 *) rhf_addr));
 			if (ipath_debug & __IPATH_ERRPKTDBG) {
 				u32 j, *d, dw = rsize-2;
@@ -1457,7 +1431,8 @@ static void ipath_reset_availshadow(struct ipath_devdata *dd)
 			0xaaaaaaaaaaaaaaaaULL); /* All BUSY bits in qword */
 		if (oldval != dd->ipath_pioavailshadow[i])
 			ipath_dbg("shadow[%d] was %Lx, now %lx\n",
-				i, oldval, dd->ipath_pioavailshadow[i]);
+				i, (unsigned long long) oldval,
+				dd->ipath_pioavailshadow[i]);
 	}
 	spin_unlock_irqrestore(&ipath_pioavail_lock, flags);
 }
@@ -1692,7 +1667,7 @@ void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start,
 			      unsigned len, int avail)
 {
 	unsigned long flags;
-	unsigned end, cnt = 0, next;
+	unsigned end, cnt = 0;
 
 	/* There are two bits per send buffer (busy and generation) */
 	start *= 2;
@@ -1743,12 +1718,7 @@ void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start,
 
 	if (dd->ipath_pioupd_thresh) {
 		end = 2 * (dd->ipath_piobcnt2k + dd->ipath_piobcnt4k);
-		next = find_first_bit(dd->ipath_pioavailkernel, end);
-		while (next < end) {
-			cnt++;
-			next = find_next_bit(dd->ipath_pioavailkernel, end,
-					next + 1);
-		}
+		cnt = bitmap_weight(dd->ipath_pioavailkernel, end);
 	}
 	spin_unlock_irqrestore(&ipath_pioavail_lock, flags);
 
@@ -2411,7 +2381,7 @@ void ipath_shutdown_device(struct ipath_devdata *dd)
 	/*
 	 * clear SerdesEnable and turn the leds off; do this here because
 	 * we are unloading, so don't count on interrupts to move along
-	 * Turn the LEDs off explictly for the same reason.
+	 * Turn the LEDs off explicitly for the same reason.
 	 */
 	dd->ipath_f_quiet_serdes(dd);
 
@@ -2530,11 +2500,6 @@ static int __init infinipath_init(void)
 	 * the PCI subsystem.
 	 */
 	idr_init(&unit_table);
-	if (!idr_pre_get(&unit_table, GFP_KERNEL)) {
-		printk(KERN_ERR IPATH_DRV_NAME ": idr_pre_get() failed\n");
-		ret = -ENOMEM;
-		goto bail;
-	}
 
 	ret = pci_register_driver(&ipath_driver);
 	if (ret < 0) {
@@ -2585,6 +2550,7 @@ int ipath_reset_device(int unit)
 {
 	int ret, i;
 	struct ipath_devdata *dd = ipath_lookup(unit);
+	unsigned long flags;
 
 	if (!dd) {
 		ret = -ENODEV;
@@ -2610,18 +2576,21 @@ int ipath_reset_device(int unit)
 		goto bail;
 	}
 
+	spin_lock_irqsave(&dd->ipath_uctxt_lock, flags);
 	if (dd->ipath_pd)
 		for (i = 1; i < dd->ipath_cfgports; i++) {
-			if (dd->ipath_pd[i] && dd->ipath_pd[i]->port_cnt) {
-				ipath_dbg("unit %u port %d is in use "
-					  "(PID %u cmd %s), can't reset\n",
-					  unit, i,
-					  pid_nr(dd->ipath_pd[i]->port_pid),
-					  dd->ipath_pd[i]->port_comm);
-				ret = -EBUSY;
-				goto bail;
-			}
+			if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt)
+				continue;
+			spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
+			ipath_dbg("unit %u port %d is in use "
+				  "(PID %u cmd %s), can't reset\n",
+				  unit, i,
+				  pid_nr(dd->ipath_pd[i]->port_pid),
+				  dd->ipath_pd[i]->port_comm);
+			ret = -EBUSY;
+			goto bail;
 		}
+	spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
 
 	if (dd->ipath_flags & IPATH_HAS_SEND_DMA)
 		teardown_sdma(dd);
@@ -2655,9 +2624,12 @@ static int ipath_signal_procs(struct ipath_devdata *dd, int sig)
 {
 	int i, sub, any = 0;
 	struct pid *pid;
+	unsigned long flags;
 
 	if (!dd->ipath_pd)
 		return 0;
+
+	spin_lock_irqsave(&dd->ipath_uctxt_lock, flags);
 	for (i = 1; i < dd->ipath_cfgports; i++) {
 		if (!dd->ipath_pd[i] || !dd->ipath_pd[i]->port_cnt)
 			continue;
@@ -2681,6 +2653,7 @@ static int ipath_signal_procs(struct ipath_devdata *dd, int sig)
 			any++;
 		}
 	}
+	spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
 	return any;
 }
 
@@ -2703,7 +2676,7 @@ static void ipath_hol_signal_up(struct ipath_devdata *dd)
  * to prevent HoL blocking, then start the HoL timer that
  * periodically continues, then stop procs, so they can detect
  * link down if they want, and do something about it.
- * Timer may already be running, so use __mod_timer, not add_timer.
+ * Timer may already be running, so use mod_timer, not add_timer.
  */
 void ipath_hol_down(struct ipath_devdata *dd)
 {
@@ -2712,7 +2685,7 @@ void ipath_hol_down(struct ipath_devdata *dd)
 	dd->ipath_hol_next = IPATH_HOL_DOWNCONT;
 	dd->ipath_hol_timer.expires = jiffies +
 		msecs_to_jiffies(ipath_hol_timeout_ms);
-	__mod_timer(&dd->ipath_hol_timer, dd->ipath_hol_timer.expires);
+	mod_timer(&dd->ipath_hol_timer, dd->ipath_hol_timer.expires);
 }
 
 /*
@@ -2751,7 +2724,7 @@ void ipath_hol_event(unsigned long opaque)
 	else {
 		dd->ipath_hol_timer.expires = jiffies +
 			msecs_to_jiffies(ipath_hol_timeout_ms);
-		__mod_timer(&dd->ipath_hol_timer,
+		mod_timer(&dd->ipath_hol_timer,
 			dd->ipath_hol_timer.expires);
 	}
 }
diff --git a/drivers/infiniband/hw/ipath/ipath_eeprom.c b/drivers/infiniband/hw/ipath/ipath_eeprom.c
index dc37277f1c8..fc7181985e8 100644
--- a/drivers/infiniband/hw/ipath/ipath_eeprom.c
+++ b/drivers/infiniband/hw/ipath/ipath_eeprom.c
@@ -772,8 +772,8 @@ void ipath_get_eeprom_info(struct ipath_devdata *dd)
 			 "0x%x, not 0x%x\n", csum, ifp->if_csum);
 		goto done;
 	}
-	if (*(__be64 *) ifp->if_guid == 0ULL ||
-	    *(__be64 *) ifp->if_guid == __constant_cpu_to_be64(-1LL)) {
+	if (*(__be64 *) ifp->if_guid == cpu_to_be64(0) ||
+	    *(__be64 *) ifp->if_guid == ~cpu_to_be64(0)) {
 		ipath_dev_err(dd, "Invalid GUID %llx from flash; "
 			      "ignoring\n",
 			      *(unsigned long long *) ifp->if_guid);
diff --git a/drivers/infiniband/hw/ipath/ipath_file_ops.c b/drivers/infiniband/hw/ipath/ipath_file_ops.c
index 35f301c88b5..6d7f453b4d0 100644
--- a/drivers/infiniband/hw/ipath/ipath_file_ops.c
+++ b/drivers/infiniband/hw/ipath/ipath_file_ops.c
@@ -35,11 +35,14 @@
 #include <linux/poll.h>
 #include <linux/cdev.h>
 #include <linux/swap.h>
+#include <linux/export.h>
 #include <linux/vmalloc.h>
+#include <linux/slab.h>
 #include <linux/highmem.h>
 #include <linux/io.h>
+#include <linux/aio.h>
 #include <linux/jiffies.h>
-#include <linux/smp_lock.h>
+#include <linux/cpu.h>
 #include <asm/pgtable.h>
 
 #include "ipath_kernel.h"
@@ -62,7 +65,8 @@ static const struct file_operations ipath_file_ops = {
 	.open = ipath_open,
 	.release = ipath_close,
 	.poll = ipath_poll,
-	.mmap = ipath_mmap
+	.mmap = ipath_mmap,
+	.llseek = noop_llseek,
 };
 
 /*
@@ -223,8 +227,13 @@ static int ipath_get_base_info(struct file *fp,
 			(unsigned long long) kinfo->spi_subport_rcvhdr_base);
 	}
 
-	kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->ipath_piobufbase) /
-		dd->ipath_palign;
+	/*
+	 * All user buffers are 2KB buffers.  If we ever support
+	 * giving 4KB buffers to user processes, this will need some
+	 * work.
+	 */
+	kinfo->spi_pioindex = (kinfo->spi_piobufbase -
+		(dd->ipath_piobufbase & 0xffffffff)) / dd->ipath_palign;
 	kinfo->spi_pioalign = dd->ipath_palign;
 
 	kinfo->spi_qpair = IPATH_KD_QP;
@@ -1146,7 +1155,7 @@ static int ipath_file_vma_fault(struct vm_area_struct *vma,
 	return 0;
 }
 
-static struct vm_operations_struct ipath_file_vm_ops = {
+static const struct vm_operations_struct ipath_file_vm_ops = {
 	.fault = ipath_file_vma_fault,
 };
 
@@ -1217,7 +1226,7 @@ static int mmap_kvaddr(struct vm_area_struct *vma, u64 pgaddr,
 
 	vma->vm_pgoff = (unsigned long) addr >> PAGE_SHIFT;
 	vma->vm_ops = &ipath_file_vm_ops;
-	vma->vm_flags |= VM_RESERVED | VM_DONTEXPAND;
+	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
 	ret = 1;
 
 bail:
@@ -1524,7 +1533,7 @@ static int init_subports(struct ipath_devdata *dd,
 	}
 
 	num_subports = uinfo->spu_subport_cnt;
-	pd->subport_uregbase = vmalloc(PAGE_SIZE * num_subports);
+	pd->subport_uregbase = vzalloc(PAGE_SIZE * num_subports);
 	if (!pd->subport_uregbase) {
 		ret = -ENOMEM;
 		goto bail;
@@ -1532,13 +1541,13 @@ static int init_subports(struct ipath_devdata *dd,
 	/* Note: pd->port_rcvhdrq_size isn't initialized yet. */
 	size = ALIGN(dd->ipath_rcvhdrcnt * dd->ipath_rcvhdrentsize *
 		     sizeof(u32), PAGE_SIZE) * num_subports;
-	pd->subport_rcvhdr_base = vmalloc(size);
+	pd->subport_rcvhdr_base = vzalloc(size);
 	if (!pd->subport_rcvhdr_base) {
 		ret = -ENOMEM;
 		goto bail_ureg;
 	}
 
-	pd->subport_rcvegrbuf = vmalloc(pd->port_rcvegrbuf_chunks *
+	pd->subport_rcvegrbuf = vzalloc(pd->port_rcvegrbuf_chunks *
 					pd->port_rcvegrbuf_size *
 					num_subports);
 	if (!pd->subport_rcvegrbuf) {
@@ -1550,11 +1559,6 @@ static int init_subports(struct ipath_devdata *dd,
 	pd->port_subport_id = uinfo->spu_subport_id;
 	pd->active_slaves = 1;
 	set_bit(IPATH_PORT_MASTER_UNINIT, &pd->port_flag);
-	memset(pd->subport_uregbase, 0, PAGE_SIZE * num_subports);
-	memset(pd->subport_rcvhdr_base, 0, size);
-	memset(pd->subport_rcvegrbuf, 0, pd->port_rcvegrbuf_chunks *
-				         pd->port_rcvegrbuf_size *
-				         num_subports);
 	goto bail;
 
 bail_rhdr:
@@ -1611,7 +1615,7 @@ static int try_alloc_port(struct ipath_devdata *dd, int port,
 		pd->port_cnt = 1;
 		port_fp(fp) = pd;
 		pd->port_pid = get_pid(task_pid(current));
-		strncpy(pd->port_comm, current->comm, sizeof(pd->port_comm));
+		strlcpy(pd->port_comm, current->comm, sizeof(pd->port_comm));
 		ipath_stats.sps_ports++;
 		ret = 0;
 	} else
@@ -1674,7 +1678,7 @@ static int find_best_unit(struct file *fp,
 	 * InfiniPath chip to that processor (we assume reasonable connectivity,
 	 * for now).  This code assumes that if affinity has been set
 	 * before this point, that at most one cpu is set; for now this
-	 * is reasonable.  I check for both cpus_empty() and cpus_full(),
+	 * is reasonable.  I check for both cpumask_empty() and cpumask_full(),
 	 * in case some kernel variant sets none of the bits when no
 	 * affinity is set.  2.6.11 and 12 kernels have all present
 	 * cpus set.  Some day we'll have to fix it up further to handle
@@ -1683,17 +1687,19 @@ static int find_best_unit(struct file *fp,
 	 * information.  There may be some issues with dual core numbering
 	 * as well.  This needs more work prior to release.
 	 */
-	if (!cpus_empty(current->cpus_allowed) &&
-	    !cpus_full(current->cpus_allowed)) {
+	if (!cpumask_empty(tsk_cpus_allowed(current)) &&
+	    !cpumask_full(tsk_cpus_allowed(current))) {
 		int ncpus = num_online_cpus(), curcpu = -1, nset = 0;
-		for (i = 0; i < ncpus; i++)
-			if (cpu_isset(i, current->cpus_allowed)) {
+		get_online_cpus();
+		for_each_online_cpu(i)
+			if (cpumask_test_cpu(i, tsk_cpus_allowed(current))) {
 				ipath_cdbg(PROC, "%s[%u] affinity set for "
 					   "cpu %d/%d\n", current->comm,
 					   current->pid, i, ncpus);
 				curcpu = i;
 				nset++;
 			}
+		put_online_cpus();
 		if (curcpu != -1 && nset != ncpus) {
 			if (npresent) {
 				prefunit = curcpu / (ncpus / npresent);
@@ -1816,7 +1822,6 @@ done:
 static int ipath_open(struct inode *in, struct file *fp)
 {
 	/* The real work is performed later in ipath_assign_port() */
-	cycle_kernel_lock();
 	fp->private_data = kzalloc(sizeof(struct ipath_filedata), GFP_KERNEL);
 	return fp->private_data ? 0 : -ENOMEM;
 }
@@ -1860,9 +1865,9 @@ static int ipath_assign_port(struct file *fp,
 		goto done_chk_sdma;
 	}
 
-	i_minor = iminor(fp->f_path.dentry->d_inode) - IPATH_USER_MINOR_BASE;
+	i_minor = iminor(file_inode(fp)) - IPATH_USER_MINOR_BASE;
 	ipath_cdbg(VERBOSE, "open on dev %lx (minor %d)\n",
-		   (long)fp->f_path.dentry->d_inode->i_rdev, i_minor);
+		   (long)file_inode(fp)->i_rdev, i_minor);
 
 	if (i_minor)
 		ret = find_free_port(i_minor - 1, fp, uinfo);
@@ -1972,7 +1977,7 @@ static int ipath_do_user_init(struct file *fp,
 	 * 0 to 1.  So for those chips, we turn it off and then back on.
 	 * This will (very briefly) affect any other open ports, but the
 	 * duration is very short, and therefore isn't an issue.  We
-	 * explictly set the in-memory tail copy to 0 beforehand, so we
+	 * explicitly set the in-memory tail copy to 0 beforehand, so we
 	 * don't have to wait to be sure the DMA update has happened
 	 * (chip resets head/tail to 0 on transition to enable).
 	 */
@@ -2041,14 +2046,16 @@ static int ipath_close(struct inode *in, struct file *fp)
 	struct ipath_filedata *fd;
 	struct ipath_portdata *pd;
 	struct ipath_devdata *dd;
+	unsigned long flags;
 	unsigned port;
+	struct pid *pid;
 
 	ipath_cdbg(VERBOSE, "close on dev %lx, private data %p\n",
 		   (long)in->i_rdev, fp->private_data);
 
 	mutex_lock(&ipath_mutex);
 
-	fd = (struct ipath_filedata *) fp->private_data;
+	fd = fp->private_data;
 	fp->private_data = NULL;
 	pd = fd->pd;
 	if (!pd) {
@@ -2074,14 +2081,13 @@ static int ipath_close(struct inode *in, struct file *fp)
 		mutex_unlock(&ipath_mutex);
 		goto bail;
 	}
+	/* early; no interrupt users after this */
+	spin_lock_irqsave(&dd->ipath_uctxt_lock, flags);
 	port = pd->port_port;
-
-	if (pd->port_hdrqfull) {
-		ipath_cdbg(PROC, "%s[%u] had %u rcvhdrqfull errors "
-			   "during run\n", pd->port_comm, pid_nr(pd->port_pid),
-			   pd->port_hdrqfull);
-		pd->port_hdrqfull = 0;
-	}
+	dd->ipath_pd[port] = NULL;
+	pid = pd->port_pid;
+	pd->port_pid = NULL;
+	spin_unlock_irqrestore(&dd->ipath_uctxt_lock, flags);
 
 	if (pd->port_rcvwait_to || pd->port_piowait_to
 	    || pd->port_rcvnowait || pd->port_pionowait) {
@@ -2138,13 +2144,11 @@ static int ipath_close(struct inode *in, struct file *fp)
 			unlock_expected_tids(pd);
 		ipath_stats.sps_ports--;
 		ipath_cdbg(PROC, "%s[%u] closed port %u:%u\n",
-			   pd->port_comm, pid_nr(pd->port_pid),
+			   pd->port_comm, pid_nr(pid),
 			   dd->ipath_unit, port);
 	}
 
-	put_pid(pd->port_pid);
-	pd->port_pid = NULL;
-	dd->ipath_pd[pd->port_port] = NULL; /* before releasing mutex */
+	put_pid(pid);
 	mutex_unlock(&ipath_mutex);
 	ipath_free_pddata(dd, pd); /* after releasing the mutex */
 
@@ -2455,7 +2459,7 @@ static int init_cdev(int minor, char *name, const struct file_operations *fops,
 		goto err_cdev;
 	}
 
-	device = device_create(ipath_class, NULL, dev, name);
+	device = device_create(ipath_class, NULL, dev, NULL, name);
 
 	if (IS_ERR(device)) {
 		ret = PTR_ERR(device);
diff --git a/drivers/infiniband/hw/ipath/ipath_fs.c b/drivers/infiniband/hw/ipath/ipath_fs.c
index 23faba9d21e..e0c404bdc4a 100644
--- a/drivers/infiniband/hw/ipath/ipath_fs.c
+++ b/drivers/infiniband/hw/ipath/ipath_fs.c
@@ -31,13 +31,13 @@
  * SOFTWARE.
  */
 
-#include <linux/version.h>
 #include <linux/module.h>
 #include <linux/fs.h>
 #include <linux/mount.h>
 #include <linux/pagemap.h>
 #include <linux/init.h>
 #include <linux/namei.h>
+#include <linux/slab.h>
 
 #include "ipath_kernel.h"
 
@@ -46,7 +46,7 @@
 static struct super_block *ipath_super;
 
 static int ipathfs_mknod(struct inode *dir, struct dentry *dentry,
-			 int mode, const struct file_operations *fops,
+			 umode_t mode, const struct file_operations *fops,
 			 void *data)
 {
 	int error;
@@ -57,13 +57,11 @@ static int ipathfs_mknod(struct inode *dir, struct dentry *dentry,
 		goto bail;
 	}
 
+	inode->i_ino = get_next_ino();
 	inode->i_mode = mode;
-	inode->i_uid = 0;
-	inode->i_gid = 0;
-	inode->i_blocks = 0;
 	inode->i_atime = inode->i_mtime = inode->i_ctime = CURRENT_TIME;
 	inode->i_private = data;
-	if ((mode & S_IFMT) == S_IFDIR) {
+	if (S_ISDIR(mode)) {
 		inode->i_op = &simple_dir_inode_operations;
 		inc_nlink(inode);
 		inc_nlink(dir);
@@ -78,7 +76,7 @@ bail:
 	return error;
 }
 
-static int create_file(const char *name, mode_t mode,
+static int create_file(const char *name, umode_t mode,
 		       struct dentry *parent, struct dentry **dentry,
 		       const struct file_operations *fops, void *data)
 {
@@ -87,11 +85,11 @@ static int create_file(const char *name, mode_t mode,
 	*dentry = NULL;
 	mutex_lock(&parent->d_inode->i_mutex);
 	*dentry = lookup_one_len(name, parent, strlen(name));
-	if (!IS_ERR(dentry))
+	if (!IS_ERR(*dentry))
 		error = ipathfs_mknod(parent->d_inode, *dentry,
 				      mode, fops, data);
 	else
-		error = PTR_ERR(dentry);
+		error = PTR_ERR(*dentry);
 	mutex_unlock(&parent->d_inode->i_mutex);
 
 	return error;
@@ -106,6 +104,7 @@ static ssize_t atomic_stats_read(struct file *file, char __user *buf,
 
 static const struct file_operations atomic_stats_ops = {
 	.read = atomic_stats_read,
+	.llseek = default_llseek,
 };
 
 static ssize_t atomic_counters_read(struct file *file, char __user *buf,
@@ -114,7 +113,7 @@ static ssize_t atomic_counters_read(struct file *file, char __user *buf,
 	struct infinipath_counters counters;
 	struct ipath_devdata *dd;
 
-	dd = file->f_path.dentry->d_inode->i_private;
+	dd = file_inode(file)->i_private;
 	dd->ipath_f_read_counters(dd, &counters);
 
 	return simple_read_from_buffer(buf, count, ppos, &counters,
@@ -123,6 +122,7 @@ static ssize_t atomic_counters_read(struct file *file, char __user *buf,
 
 static const struct file_operations atomic_counters_ops = {
 	.read = atomic_counters_read,
+	.llseek = default_llseek,
 };
 
 static ssize_t flash_read(struct file *file, char __user *buf,
@@ -154,7 +154,7 @@ static ssize_t flash_read(struct file *file, char __user *buf,
 		goto bail;
 	}
 
-	dd = file->f_path.dentry->d_inode->i_private;
+	dd = file_inode(file)->i_private;
 	if (ipath_eeprom_read(dd, pos, tmp, count)) {
 		ipath_dev_err(dd, "failed to read from flash\n");
 		ret = -ENXIO;
@@ -207,7 +207,7 @@ static ssize_t flash_write(struct file *file, const char __user *buf,
 		goto bail_tmp;
 	}
 
-	dd = file->f_path.dentry->d_inode->i_private;
+	dd = file_inode(file)->i_private;
 	if (ipath_eeprom_write(dd, pos, tmp, count)) {
 		ret = -ENXIO;
 		ipath_dev_err(dd, "failed to write to flash\n");
@@ -227,6 +227,7 @@ bail:
 static const struct file_operations flash_ops = {
 	.read = flash_read,
 	.write = flash_write,
+	.llseek = default_llseek,
 };
 
 static int create_device_files(struct super_block *sb,
@@ -276,18 +277,14 @@ static int remove_file(struct dentry *parent, char *name)
 		goto bail;
 	}
 
-	spin_lock(&dcache_lock);
 	spin_lock(&tmp->d_lock);
 	if (!(d_unhashed(tmp) && tmp->d_inode)) {
-		dget_locked(tmp);
+		dget_dlock(tmp);
 		__d_drop(tmp);
 		spin_unlock(&tmp->d_lock);
-		spin_unlock(&dcache_lock);
 		simple_unlink(parent->d_inode, tmp);
-	} else {
+	} else
 		spin_unlock(&tmp->d_lock);
-		spin_unlock(&dcache_lock);
-	}
 
 	ret = 0;
 bail:
@@ -350,10 +347,8 @@ static int ipathfs_fill_super(struct super_block *sb, void *data,
 	list_for_each_entry_safe(dd, tmp, &ipath_dev_list, ipath_list) {
 		spin_unlock_irqrestore(&ipath_devs_lock, flags);
 		ret = create_device_files(sb, dd);
-		if (ret) {
-			deactivate_super(sb);
+		if (ret)
 			goto bail;
-		}
 		spin_lock_irqsave(&ipath_devs_lock, flags);
 	}
 
@@ -363,13 +358,13 @@ bail:
 	return ret;
 }
 
-static int ipathfs_get_sb(struct file_system_type *fs_type, int flags,
-			const char *dev_name, void *data, struct vfsmount *mnt)
+static struct dentry *ipathfs_mount(struct file_system_type *fs_type,
+			int flags, const char *dev_name, void *data)
 {
-	int ret = get_sb_single(fs_type, flags, data,
-				    ipathfs_fill_super, mnt);
-	if (ret >= 0)
-		ipath_super = mnt->mnt_sb;
+	struct dentry *ret;
+	ret = mount_single(fs_type, flags, data, ipathfs_fill_super);
+	if (!IS_ERR(ret))
+		ipath_super = ret->d_sb;
 	return ret;
 }
 
@@ -412,9 +407,10 @@ bail:
 static struct file_system_type ipathfs_fs_type = {
 	.owner =	THIS_MODULE,
 	.name =		"ipathfs",
-	.get_sb =	ipathfs_get_sb,
+	.mount =	ipathfs_mount,
 	.kill_sb =	ipathfs_kill_super,
 };
+MODULE_ALIAS_FS("ipathfs");
 
 int __init ipath_init_ipathfs(void)
 {
diff --git a/drivers/infiniband/hw/ipath/ipath_iba6110.c b/drivers/infiniband/hw/ipath/ipath_iba6110.c
index 02831ad070b..7cc305488a3 100644
--- a/drivers/infiniband/hw/ipath/ipath_iba6110.c
+++ b/drivers/infiniband/hw/ipath/ipath_iba6110.c
@@ -381,7 +381,7 @@ static const ipath_err_t infinipath_hwe_htclnkbbyte1crcerr =
 #define IPATH_GPIO_SCL \
 	(1ULL << (_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
 
-/* keep the code below somewhat more readonable; not used elsewhere */
+/* keep the code below somewhat more readable; not used elsewhere */
 #define _IPATH_HTLINK0_CRCBITS (infinipath_hwe_htclnkabyte0crcerr |	\
 				infinipath_hwe_htclnkabyte1crcerr)
 #define _IPATH_HTLINK1_CRCBITS (infinipath_hwe_htclnkbbyte0crcerr |	\
@@ -596,8 +596,7 @@ static void ipath_ht_handle_hwerrors(struct ipath_devdata *dd, char *msg,
 
 	ipath_format_hwerrors(hwerrs,
 			      ipath_6110_hwerror_msgs,
-			      sizeof(ipath_6110_hwerror_msgs) /
-			      sizeof(ipath_6110_hwerror_msgs[0]),
+			      ARRAY_SIZE(ipath_6110_hwerror_msgs),
 			      msg, msgl);
 
 	if (hwerrs & (_IPATH_HTLINK0_CRCBITS | _IPATH_HTLINK1_CRCBITS))
@@ -809,7 +808,7 @@ static int ipath_setup_ht_reset(struct ipath_devdata *dd)
  * errors.  We only bother to do this at load time, because it's OK if
  * it happened before we were loaded (first time after boot/reset),
  * but any time after that, it's fatal anyway.  Also need to not check
- * for for upper byte errors if we are in 8 bit mode, so figure out
+ * for upper byte errors if we are in 8 bit mode, so figure out
  * our width.  For now, at least, also complain if it's 8 bit.
  */
 static void slave_or_pri_blk(struct ipath_devdata *dd, struct pci_dev *pdev,
@@ -1474,7 +1473,7 @@ static void ipath_ht_quiet_serdes(struct ipath_devdata *dd)
 /**
  * ipath_pe_put_tid - write a TID in chip
  * @dd: the infinipath device
- * @tidptr: pointer to the expected TID (in chip) to udpate
+ * @tidptr: pointer to the expected TID (in chip) to update
  * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
  * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
  *
diff --git a/drivers/infiniband/hw/ipath/ipath_iba6120.c b/drivers/infiniband/hw/ipath/ipath_iba6120.c
deleted file mode 100644
index 421cc2af891..00000000000
--- a/drivers/infiniband/hw/ipath/ipath_iba6120.c
+++ /dev/null
@@ -1,1801 +0,0 @@
-/*
- * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
- * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
- *
- * This software is available to you under a choice of one of two
- * licenses.  You may choose to be licensed under the terms of the GNU
- * General Public License (GPL) Version 2, available from the file
- * COPYING in the main directory of this source tree, or the
- * OpenIB.org BSD license below:
- *
- *     Redistribution and use in source and binary forms, with or
- *     without modification, are permitted provided that the following
- *     conditions are met:
- *
- *      - Redistributions of source code must retain the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer.
- *
- *      - Redistributions in binary form must reproduce the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer in the documentation and/or other materials
- *        provided with the distribution.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
- * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-/*
- * This file contains all of the code that is specific to the
- * InfiniPath PCIe chip.
- */
-
-#include <linux/interrupt.h>
-#include <linux/pci.h>
-#include <linux/delay.h>
-#include <rdma/ib_verbs.h>
-
-#include "ipath_kernel.h"
-#include "ipath_registers.h"
-
-static void ipath_setup_pe_setextled(struct ipath_devdata *, u64, u64);
-
-/*
- * This file contains all the chip-specific register information and
- * access functions for the QLogic InfiniPath PCI-Express chip.
- *
- * This lists the InfiniPath registers, in the actual chip layout.
- * This structure should never be directly accessed.
- */
-struct _infinipath_do_not_use_kernel_regs {
-	unsigned long long Revision;
-	unsigned long long Control;
-	unsigned long long PageAlign;
-	unsigned long long PortCnt;
-	unsigned long long DebugPortSelect;
-	unsigned long long Reserved0;
-	unsigned long long SendRegBase;
-	unsigned long long UserRegBase;
-	unsigned long long CounterRegBase;
-	unsigned long long Scratch;
-	unsigned long long Reserved1;
-	unsigned long long Reserved2;
-	unsigned long long IntBlocked;
-	unsigned long long IntMask;
-	unsigned long long IntStatus;
-	unsigned long long IntClear;
-	unsigned long long ErrorMask;
-	unsigned long long ErrorStatus;
-	unsigned long long ErrorClear;
-	unsigned long long HwErrMask;
-	unsigned long long HwErrStatus;
-	unsigned long long HwErrClear;
-	unsigned long long HwDiagCtrl;
-	unsigned long long MDIO;
-	unsigned long long IBCStatus;
-	unsigned long long IBCCtrl;
-	unsigned long long ExtStatus;
-	unsigned long long ExtCtrl;
-	unsigned long long GPIOOut;
-	unsigned long long GPIOMask;
-	unsigned long long GPIOStatus;
-	unsigned long long GPIOClear;
-	unsigned long long RcvCtrl;
-	unsigned long long RcvBTHQP;
-	unsigned long long RcvHdrSize;
-	unsigned long long RcvHdrCnt;
-	unsigned long long RcvHdrEntSize;
-	unsigned long long RcvTIDBase;
-	unsigned long long RcvTIDCnt;
-	unsigned long long RcvEgrBase;
-	unsigned long long RcvEgrCnt;
-	unsigned long long RcvBufBase;
-	unsigned long long RcvBufSize;
-	unsigned long long RxIntMemBase;
-	unsigned long long RxIntMemSize;
-	unsigned long long RcvPartitionKey;
-	unsigned long long Reserved3;
-	unsigned long long RcvPktLEDCnt;
-	unsigned long long Reserved4[8];
-	unsigned long long SendCtrl;
-	unsigned long long SendPIOBufBase;
-	unsigned long long SendPIOSize;
-	unsigned long long SendPIOBufCnt;
-	unsigned long long SendPIOAvailAddr;
-	unsigned long long TxIntMemBase;
-	unsigned long long TxIntMemSize;
-	unsigned long long Reserved5;
-	unsigned long long PCIeRBufTestReg0;
-	unsigned long long PCIeRBufTestReg1;
-	unsigned long long Reserved51[6];
-	unsigned long long SendBufferError;
-	unsigned long long SendBufferErrorCONT1;
-	unsigned long long Reserved6SBE[6];
-	unsigned long long RcvHdrAddr0;
-	unsigned long long RcvHdrAddr1;
-	unsigned long long RcvHdrAddr2;
-	unsigned long long RcvHdrAddr3;
-	unsigned long long RcvHdrAddr4;
-	unsigned long long Reserved7RHA[11];
-	unsigned long long RcvHdrTailAddr0;
-	unsigned long long RcvHdrTailAddr1;
-	unsigned long long RcvHdrTailAddr2;
-	unsigned long long RcvHdrTailAddr3;
-	unsigned long long RcvHdrTailAddr4;
-	unsigned long long Reserved8RHTA[11];
-	unsigned long long Reserved9SW[8];
-	unsigned long long SerdesConfig0;
-	unsigned long long SerdesConfig1;
-	unsigned long long SerdesStatus;
-	unsigned long long XGXSConfig;
-	unsigned long long IBPLLCfg;
-	unsigned long long Reserved10SW2[3];
-	unsigned long long PCIEQ0SerdesConfig0;
-	unsigned long long PCIEQ0SerdesConfig1;
-	unsigned long long PCIEQ0SerdesStatus;
-	unsigned long long Reserved11;
-	unsigned long long PCIEQ1SerdesConfig0;
-	unsigned long long PCIEQ1SerdesConfig1;
-	unsigned long long PCIEQ1SerdesStatus;
-	unsigned long long Reserved12;
-};
-
-struct _infinipath_do_not_use_counters {
-	__u64 LBIntCnt;
-	__u64 LBFlowStallCnt;
-	__u64 Reserved1;
-	__u64 TxUnsupVLErrCnt;
-	__u64 TxDataPktCnt;
-	__u64 TxFlowPktCnt;
-	__u64 TxDwordCnt;
-	__u64 TxLenErrCnt;
-	__u64 TxMaxMinLenErrCnt;
-	__u64 TxUnderrunCnt;
-	__u64 TxFlowStallCnt;
-	__u64 TxDroppedPktCnt;
-	__u64 RxDroppedPktCnt;
-	__u64 RxDataPktCnt;
-	__u64 RxFlowPktCnt;
-	__u64 RxDwordCnt;
-	__u64 RxLenErrCnt;
-	__u64 RxMaxMinLenErrCnt;
-	__u64 RxICRCErrCnt;
-	__u64 RxVCRCErrCnt;
-	__u64 RxFlowCtrlErrCnt;
-	__u64 RxBadFormatCnt;
-	__u64 RxLinkProblemCnt;
-	__u64 RxEBPCnt;
-	__u64 RxLPCRCErrCnt;
-	__u64 RxBufOvflCnt;
-	__u64 RxTIDFullErrCnt;
-	__u64 RxTIDValidErrCnt;
-	__u64 RxPKeyMismatchCnt;
-	__u64 RxP0HdrEgrOvflCnt;
-	__u64 RxP1HdrEgrOvflCnt;
-	__u64 RxP2HdrEgrOvflCnt;
-	__u64 RxP3HdrEgrOvflCnt;
-	__u64 RxP4HdrEgrOvflCnt;
-	__u64 RxP5HdrEgrOvflCnt;
-	__u64 RxP6HdrEgrOvflCnt;
-	__u64 RxP7HdrEgrOvflCnt;
-	__u64 RxP8HdrEgrOvflCnt;
-	__u64 Reserved6;
-	__u64 Reserved7;
-	__u64 IBStatusChangeCnt;
-	__u64 IBLinkErrRecoveryCnt;
-	__u64 IBLinkDownedCnt;
-	__u64 IBSymbolErrCnt;
-};
-
-#define IPATH_KREG_OFFSET(field) (offsetof( \
-	struct _infinipath_do_not_use_kernel_regs, field) / sizeof(u64))
-#define IPATH_CREG_OFFSET(field) (offsetof( \
-	struct _infinipath_do_not_use_counters, field) / sizeof(u64))
-
-static const struct ipath_kregs ipath_pe_kregs = {
-	.kr_control = IPATH_KREG_OFFSET(Control),
-	.kr_counterregbase = IPATH_KREG_OFFSET(CounterRegBase),
-	.kr_debugportselect = IPATH_KREG_OFFSET(DebugPortSelect),
-	.kr_errorclear = IPATH_KREG_OFFSET(ErrorClear),
-	.kr_errormask = IPATH_KREG_OFFSET(ErrorMask),
-	.kr_errorstatus = IPATH_KREG_OFFSET(ErrorStatus),
-	.kr_extctrl = IPATH_KREG_OFFSET(ExtCtrl),
-	.kr_extstatus = IPATH_KREG_OFFSET(ExtStatus),
-	.kr_gpio_clear = IPATH_KREG_OFFSET(GPIOClear),
-	.kr_gpio_mask = IPATH_KREG_OFFSET(GPIOMask),
-	.kr_gpio_out = IPATH_KREG_OFFSET(GPIOOut),
-	.kr_gpio_status = IPATH_KREG_OFFSET(GPIOStatus),
-	.kr_hwdiagctrl = IPATH_KREG_OFFSET(HwDiagCtrl),
-	.kr_hwerrclear = IPATH_KREG_OFFSET(HwErrClear),
-	.kr_hwerrmask = IPATH_KREG_OFFSET(HwErrMask),
-	.kr_hwerrstatus = IPATH_KREG_OFFSET(HwErrStatus),
-	.kr_ibcctrl = IPATH_KREG_OFFSET(IBCCtrl),
-	.kr_ibcstatus = IPATH_KREG_OFFSET(IBCStatus),
-	.kr_intblocked = IPATH_KREG_OFFSET(IntBlocked),
-	.kr_intclear = IPATH_KREG_OFFSET(IntClear),
-	.kr_intmask = IPATH_KREG_OFFSET(IntMask),
-	.kr_intstatus = IPATH_KREG_OFFSET(IntStatus),
-	.kr_mdio = IPATH_KREG_OFFSET(MDIO),
-	.kr_pagealign = IPATH_KREG_OFFSET(PageAlign),
-	.kr_partitionkey = IPATH_KREG_OFFSET(RcvPartitionKey),
-	.kr_portcnt = IPATH_KREG_OFFSET(PortCnt),
-	.kr_rcvbthqp = IPATH_KREG_OFFSET(RcvBTHQP),
-	.kr_rcvbufbase = IPATH_KREG_OFFSET(RcvBufBase),
-	.kr_rcvbufsize = IPATH_KREG_OFFSET(RcvBufSize),
-	.kr_rcvctrl = IPATH_KREG_OFFSET(RcvCtrl),
-	.kr_rcvegrbase = IPATH_KREG_OFFSET(RcvEgrBase),
-	.kr_rcvegrcnt = IPATH_KREG_OFFSET(RcvEgrCnt),
-	.kr_rcvhdrcnt = IPATH_KREG_OFFSET(RcvHdrCnt),
-	.kr_rcvhdrentsize = IPATH_KREG_OFFSET(RcvHdrEntSize),
-	.kr_rcvhdrsize = IPATH_KREG_OFFSET(RcvHdrSize),
-	.kr_rcvintmembase = IPATH_KREG_OFFSET(RxIntMemBase),
-	.kr_rcvintmemsize = IPATH_KREG_OFFSET(RxIntMemSize),
-	.kr_rcvtidbase = IPATH_KREG_OFFSET(RcvTIDBase),
-	.kr_rcvtidcnt = IPATH_KREG_OFFSET(RcvTIDCnt),
-	.kr_revision = IPATH_KREG_OFFSET(Revision),
-	.kr_scratch = IPATH_KREG_OFFSET(Scratch),
-	.kr_sendbuffererror = IPATH_KREG_OFFSET(SendBufferError),
-	.kr_sendctrl = IPATH_KREG_OFFSET(SendCtrl),
-	.kr_sendpioavailaddr = IPATH_KREG_OFFSET(SendPIOAvailAddr),
-	.kr_sendpiobufbase = IPATH_KREG_OFFSET(SendPIOBufBase),
-	.kr_sendpiobufcnt = IPATH_KREG_OFFSET(SendPIOBufCnt),
-	.kr_sendpiosize = IPATH_KREG_OFFSET(SendPIOSize),
-	.kr_sendregbase = IPATH_KREG_OFFSET(SendRegBase),
-	.kr_txintmembase = IPATH_KREG_OFFSET(TxIntMemBase),
-	.kr_txintmemsize = IPATH_KREG_OFFSET(TxIntMemSize),
-	.kr_userregbase = IPATH_KREG_OFFSET(UserRegBase),
-	.kr_serdesconfig0 = IPATH_KREG_OFFSET(SerdesConfig0),
-	.kr_serdesconfig1 = IPATH_KREG_OFFSET(SerdesConfig1),
-	.kr_serdesstatus = IPATH_KREG_OFFSET(SerdesStatus),
-	.kr_xgxsconfig = IPATH_KREG_OFFSET(XGXSConfig),
-	.kr_ibpllcfg = IPATH_KREG_OFFSET(IBPLLCfg),
-
-	/*
-	 * These should not be used directly via ipath_write_kreg64(),
-	 * use them with ipath_write_kreg64_port(),
-	 */
-	.kr_rcvhdraddr = IPATH_KREG_OFFSET(RcvHdrAddr0),
-	.kr_rcvhdrtailaddr = IPATH_KREG_OFFSET(RcvHdrTailAddr0),
-
-	/* The rcvpktled register controls one of the debug port signals, so
-	 * a packet activity LED can be connected to it. */
-	.kr_rcvpktledcnt = IPATH_KREG_OFFSET(RcvPktLEDCnt),
-	.kr_pcierbuftestreg0 = IPATH_KREG_OFFSET(PCIeRBufTestReg0),
-	.kr_pcierbuftestreg1 = IPATH_KREG_OFFSET(PCIeRBufTestReg1),
-	.kr_pcieq0serdesconfig0 = IPATH_KREG_OFFSET(PCIEQ0SerdesConfig0),
-	.kr_pcieq0serdesconfig1 = IPATH_KREG_OFFSET(PCIEQ0SerdesConfig1),
-	.kr_pcieq0serdesstatus = IPATH_KREG_OFFSET(PCIEQ0SerdesStatus),
-	.kr_pcieq1serdesconfig0 = IPATH_KREG_OFFSET(PCIEQ1SerdesConfig0),
-	.kr_pcieq1serdesconfig1 = IPATH_KREG_OFFSET(PCIEQ1SerdesConfig1),
-	.kr_pcieq1serdesstatus = IPATH_KREG_OFFSET(PCIEQ1SerdesStatus)
-};
-
-static const struct ipath_cregs ipath_pe_cregs = {
-	.cr_badformatcnt = IPATH_CREG_OFFSET(RxBadFormatCnt),
-	.cr_erricrccnt = IPATH_CREG_OFFSET(RxICRCErrCnt),
-	.cr_errlinkcnt = IPATH_CREG_OFFSET(RxLinkProblemCnt),
-	.cr_errlpcrccnt = IPATH_CREG_OFFSET(RxLPCRCErrCnt),
-	.cr_errpkey = IPATH_CREG_OFFSET(RxPKeyMismatchCnt),
-	.cr_errrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowCtrlErrCnt),
-	.cr_err_rlencnt = IPATH_CREG_OFFSET(RxLenErrCnt),
-	.cr_errslencnt = IPATH_CREG_OFFSET(TxLenErrCnt),
-	.cr_errtidfull = IPATH_CREG_OFFSET(RxTIDFullErrCnt),
-	.cr_errtidvalid = IPATH_CREG_OFFSET(RxTIDValidErrCnt),
-	.cr_errvcrccnt = IPATH_CREG_OFFSET(RxVCRCErrCnt),
-	.cr_ibstatuschange = IPATH_CREG_OFFSET(IBStatusChangeCnt),
-	.cr_intcnt = IPATH_CREG_OFFSET(LBIntCnt),
-	.cr_invalidrlencnt = IPATH_CREG_OFFSET(RxMaxMinLenErrCnt),
-	.cr_invalidslencnt = IPATH_CREG_OFFSET(TxMaxMinLenErrCnt),
-	.cr_lbflowstallcnt = IPATH_CREG_OFFSET(LBFlowStallCnt),
-	.cr_pktrcvcnt = IPATH_CREG_OFFSET(RxDataPktCnt),
-	.cr_pktrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowPktCnt),
-	.cr_pktsendcnt = IPATH_CREG_OFFSET(TxDataPktCnt),
-	.cr_pktsendflowcnt = IPATH_CREG_OFFSET(TxFlowPktCnt),
-	.cr_portovflcnt = IPATH_CREG_OFFSET(RxP0HdrEgrOvflCnt),
-	.cr_rcvebpcnt = IPATH_CREG_OFFSET(RxEBPCnt),
-	.cr_rcvovflcnt = IPATH_CREG_OFFSET(RxBufOvflCnt),
-	.cr_senddropped = IPATH_CREG_OFFSET(TxDroppedPktCnt),
-	.cr_sendstallcnt = IPATH_CREG_OFFSET(TxFlowStallCnt),
-	.cr_sendunderruncnt = IPATH_CREG_OFFSET(TxUnderrunCnt),
-	.cr_wordrcvcnt = IPATH_CREG_OFFSET(RxDwordCnt),
-	.cr_wordsendcnt = IPATH_CREG_OFFSET(TxDwordCnt),
-	.cr_unsupvlcnt = IPATH_CREG_OFFSET(TxUnsupVLErrCnt),
-	.cr_rxdroppktcnt = IPATH_CREG_OFFSET(RxDroppedPktCnt),
-	.cr_iblinkerrrecovcnt = IPATH_CREG_OFFSET(IBLinkErrRecoveryCnt),
-	.cr_iblinkdowncnt = IPATH_CREG_OFFSET(IBLinkDownedCnt),
-	.cr_ibsymbolerrcnt = IPATH_CREG_OFFSET(IBSymbolErrCnt)
-};
-
-/* kr_control bits */
-#define INFINIPATH_C_RESET 1U
-
-/* kr_intstatus, kr_intclear, kr_intmask bits */
-#define INFINIPATH_I_RCVURG_MASK ((1U<<5)-1)
-#define INFINIPATH_I_RCVURG_SHIFT 0
-#define INFINIPATH_I_RCVAVAIL_MASK ((1U<<5)-1)
-#define INFINIPATH_I_RCVAVAIL_SHIFT 12
-
-/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
-#define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK  0x000000000000003fULL
-#define INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT 0
-#define INFINIPATH_HWE_PCIEPOISONEDTLP      0x0000000010000000ULL
-#define INFINIPATH_HWE_PCIECPLTIMEOUT       0x0000000020000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYXTLH    0x0000000040000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYXADM    0x0000000080000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYRADM    0x0000000100000000ULL
-#define INFINIPATH_HWE_COREPLL_FBSLIP       0x0080000000000000ULL
-#define INFINIPATH_HWE_COREPLL_RFSLIP       0x0100000000000000ULL
-#define INFINIPATH_HWE_PCIE1PLLFAILED       0x0400000000000000ULL
-#define INFINIPATH_HWE_PCIE0PLLFAILED       0x0800000000000000ULL
-#define INFINIPATH_HWE_SERDESPLLFAILED      0x1000000000000000ULL
-
-#define IBA6120_IBCS_LINKTRAININGSTATE_MASK 0xf
-#define IBA6120_IBCS_LINKSTATE_SHIFT 4
-
-/* kr_extstatus bits */
-#define INFINIPATH_EXTS_FREQSEL 0x2
-#define INFINIPATH_EXTS_SERDESSEL 0x4
-#define INFINIPATH_EXTS_MEMBIST_ENDTEST     0x0000000000004000
-#define INFINIPATH_EXTS_MEMBIST_FOUND       0x0000000000008000
-
-/* kr_xgxsconfig bits */
-#define INFINIPATH_XGXS_RESET          0x5ULL
-
-#define _IPATH_GPIO_SDA_NUM 1
-#define _IPATH_GPIO_SCL_NUM 0
-
-#define IPATH_GPIO_SDA (1ULL << \
-	(_IPATH_GPIO_SDA_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
-#define IPATH_GPIO_SCL (1ULL << \
-	(_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
-
-#define INFINIPATH_RT_BUFSIZE_MASK 0xe0000000ULL
-#define INFINIPATH_RT_BUFSIZE_SHIFTVAL(tid) \
-	((((tid) & INFINIPATH_RT_BUFSIZE_MASK) >> 29) + 11 - 1)
-#define INFINIPATH_RT_BUFSIZE(tid) (1 << INFINIPATH_RT_BUFSIZE_SHIFTVAL(tid))
-#define INFINIPATH_RT_IS_VALID(tid) \
-	(((tid) & INFINIPATH_RT_BUFSIZE_MASK) && \
-	 ((((tid) & INFINIPATH_RT_BUFSIZE_MASK) != INFINIPATH_RT_BUFSIZE_MASK)))
-#define INFINIPATH_RT_ADDR_MASK 0x1FFFFFFFULL /* 29 bits valid */
-#define INFINIPATH_RT_ADDR_SHIFT 10
-
-#define INFINIPATH_R_INTRAVAIL_SHIFT 16
-#define INFINIPATH_R_TAILUPD_SHIFT 31
-
-/* 6120 specific hardware errors... */
-static const struct ipath_hwerror_msgs ipath_6120_hwerror_msgs[] = {
-	INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"),
-	INFINIPATH_HWE_MSG(PCIECPLTIMEOUT, "PCIe completion timeout"),
-	/*
-	 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
-	 * parity or memory parity error failures, because most likely we
-	 * won't be able to talk to the core of the chip.  Nonetheless, we
-	 * might see them, if they are in parts of the PCIe core that aren't
-	 * essential.
-	 */
-	INFINIPATH_HWE_MSG(PCIE1PLLFAILED, "PCIePLL1"),
-	INFINIPATH_HWE_MSG(PCIE0PLLFAILED, "PCIePLL0"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYXTLH, "PCIe XTLH core parity"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYXADM, "PCIe ADM TX core parity"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYRADM, "PCIe ADM RX core parity"),
-	INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"),
-	INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"),
-};
-
-#define TXE_PIO_PARITY ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF | \
-		        INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC) \
-		        << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)
-#define RXE_EAGER_PARITY (INFINIPATH_HWE_RXEMEMPARITYERR_EAGERTID \
-			  << INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT)
-
-static void ipath_pe_put_tid_2(struct ipath_devdata *, u64 __iomem *,
-			       u32, unsigned long);
-
-/*
- * On platforms using this chip, and not having ordered WC stores, we
- * can get TXE parity errors due to speculative reads to the PIO buffers,
- * and this, due to a chip bug can result in (many) false parity error
- * reports.  So it's a debug print on those, and an info print on systems
- * where the speculative reads don't occur.
- */
-static void ipath_pe_txe_recover(struct ipath_devdata *dd)
-{
-	if (ipath_unordered_wc())
-		ipath_dbg("Recovering from TXE PIO parity error\n");
-	else {
-		++ipath_stats.sps_txeparity;
-		dev_info(&dd->pcidev->dev,
-			"Recovering from TXE PIO parity error\n");
-	}
-}
-
-
-/**
- * ipath_pe_handle_hwerrors - display hardware errors.
- * @dd: the infinipath device
- * @msg: the output buffer
- * @msgl: the size of the output buffer
- *
- * Use same msg buffer as regular errors to avoid excessive stack
- * use.  Most hardware errors are catastrophic, but for right now,
- * we'll print them and continue.  We reuse the same message buffer as
- * ipath_handle_errors() to avoid excessive stack usage.
- */
-static void ipath_pe_handle_hwerrors(struct ipath_devdata *dd, char *msg,
-				     size_t msgl)
-{
-	ipath_err_t hwerrs;
-	u32 bits, ctrl;
-	int isfatal = 0;
-	char bitsmsg[64];
-	int log_idx;
-
-	hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
-	if (!hwerrs) {
-		/*
-		 * better than printing cofusing messages
-		 * This seems to be related to clearing the crc error, or
-		 * the pll error during init.
-		 */
-		ipath_cdbg(VERBOSE, "Called but no hardware errors set\n");
-		return;
-	} else if (hwerrs == ~0ULL) {
-		ipath_dev_err(dd, "Read of hardware error status failed "
-			      "(all bits set); ignoring\n");
-		return;
-	}
-	ipath_stats.sps_hwerrs++;
-
-	/* Always clear the error status register, except MEMBISTFAIL,
-	 * regardless of whether we continue or stop using the chip.
-	 * We want that set so we know it failed, even across driver reload.
-	 * We'll still ignore it in the hwerrmask.  We do this partly for
-	 * diagnostics, but also for support */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
-			 hwerrs&~INFINIPATH_HWE_MEMBISTFAILED);
-
-	hwerrs &= dd->ipath_hwerrmask;
-
-	/* We log some errors to EEPROM, check if we have any of those. */
-	for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
-		if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
-			ipath_inc_eeprom_err(dd, log_idx, 1);
-
-	/*
-	 * make sure we get this much out, unless told to be quiet,
-	 * or it's occurred within the last 5 seconds
-	 */
-	if ((hwerrs & ~(dd->ipath_lasthwerror | TXE_PIO_PARITY |
-			RXE_EAGER_PARITY)) ||
-	    (ipath_debug & __IPATH_VERBDBG))
-		dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
-			 "(cleared)\n", (unsigned long long) hwerrs);
-	dd->ipath_lasthwerror |= hwerrs;
-
-	if (hwerrs & ~dd->ipath_hwe_bitsextant)
-		ipath_dev_err(dd, "hwerror interrupt with unknown errors "
-			      "%llx set\n", (unsigned long long)
-			      (hwerrs & ~dd->ipath_hwe_bitsextant));
-
-	ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
-	if ((ctrl & INFINIPATH_C_FREEZEMODE) && !ipath_diag_inuse) {
-		/*
-		 * parity errors in send memory are recoverable,
-		 * just cancel the send (if indicated in * sendbuffererror),
-		 * count the occurrence, unfreeze (if no other handled
-		 * hardware error bits are set), and continue. They can
-		 * occur if a processor speculative read is done to the PIO
-		 * buffer while we are sending a packet, for example.
-		 */
-		if (hwerrs & TXE_PIO_PARITY) {
-			ipath_pe_txe_recover(dd);
-			hwerrs &= ~TXE_PIO_PARITY;
-		}
-		if (!hwerrs) {
-			static u32 freeze_cnt;
-
-			freeze_cnt++;
-			ipath_dbg("Clearing freezemode on ignored or recovered "
-				  "hardware error (%u)\n", freeze_cnt);
-			ipath_clear_freeze(dd);
-		}
-	}
-
-	*msg = '\0';
-
-	if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) {
-		strlcat(msg, "[Memory BIST test failed, InfiniPath hardware unusable]",
-			msgl);
-		/* ignore from now on, so disable until driver reloaded */
-		*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
-		dd->ipath_hwerrmask &= ~INFINIPATH_HWE_MEMBISTFAILED;
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	ipath_format_hwerrors(hwerrs,
-			      ipath_6120_hwerror_msgs,
-			      sizeof(ipath_6120_hwerror_msgs)/
-			      sizeof(ipath_6120_hwerror_msgs[0]),
-			      msg, msgl);
-
-	if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK
-		      << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) {
-		bits = (u32) ((hwerrs >>
-			       INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) &
-			      INFINIPATH_HWE_PCIEMEMPARITYERR_MASK);
-		snprintf(bitsmsg, sizeof bitsmsg,
-			 "[PCIe Mem Parity Errs %x] ", bits);
-		strlcat(msg, bitsmsg, msgl);
-	}
-
-#define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP |	\
-			 INFINIPATH_HWE_COREPLL_RFSLIP )
-
-	if (hwerrs & _IPATH_PLL_FAIL) {
-		snprintf(bitsmsg, sizeof bitsmsg,
-			 "[PLL failed (%llx), InfiniPath hardware unusable]",
-			 (unsigned long long) hwerrs & _IPATH_PLL_FAIL);
-		strlcat(msg, bitsmsg, msgl);
-		/* ignore from now on, so disable until driver reloaded */
-		dd->ipath_hwerrmask &= ~(hwerrs & _IPATH_PLL_FAIL);
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED) {
-		/*
-		 * If it occurs, it is left masked since the external
-		 * interface is unused
-		 */
-		dd->ipath_hwerrmask &= ~INFINIPATH_HWE_SERDESPLLFAILED;
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	if (hwerrs) {
-		/*
-		 * if any set that we aren't ignoring; only
-		 * make the complaint once, in case it's stuck
-		 * or recurring, and we get here multiple
-		 * times.
-		 */
-		ipath_dev_err(dd, "%s hardware error\n", msg);
-		if (dd->ipath_flags & IPATH_INITTED) {
-			ipath_set_linkstate(dd, IPATH_IB_LINKDOWN);
-			ipath_setup_pe_setextled(dd,
-				INFINIPATH_IBCS_L_STATE_DOWN,
-				INFINIPATH_IBCS_LT_STATE_DISABLED);
-			ipath_dev_err(dd, "Fatal Hardware Error (freeze "
-					  "mode), no longer usable, SN %.16s\n",
-					  dd->ipath_serial);
-			isfatal = 1;
-		}
-		*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
-		/* mark as having had error */
-		*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
-		/*
-		 * mark as not usable, at a minimum until driver
-		 * is reloaded, probably until reboot, since no
-		 * other reset is possible.
-		 */
-		dd->ipath_flags &= ~IPATH_INITTED;
-	} else
-		*msg = 0; /* recovered from all of them */
-
-	if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg && msg) {
-		/*
-		 * for /sys status file ; if no trailing brace is copied,
-		 * we'll know it was truncated.
-		 */
-		snprintf(dd->ipath_freezemsg, dd->ipath_freezelen,
-			 "{%s}", msg);
-	}
-}
-
-/**
- * ipath_pe_boardname - fill in the board name
- * @dd: the infinipath device
- * @name: the output buffer
- * @namelen: the size of the output buffer
- *
- * info is based on the board revision register
- */
-static int ipath_pe_boardname(struct ipath_devdata *dd, char *name,
-			      size_t namelen)
-{
-	char *n = NULL;
-	u8 boardrev = dd->ipath_boardrev;
-	int ret;
-
-	switch (boardrev) {
-	case 0:
-		n = "InfiniPath_Emulation";
-		break;
-	case 1:
-		n = "InfiniPath_QLE7140-Bringup";
-		break;
-	case 2:
-		n = "InfiniPath_QLE7140";
-		break;
-	case 3:
-		n = "InfiniPath_QMI7140";
-		break;
-	case 4:
-		n = "InfiniPath_QEM7140";
-		break;
-	case 5:
-		n = "InfiniPath_QMH7140";
-		break;
-	case 6:
-		n = "InfiniPath_QLE7142";
-		break;
-	default:
-		ipath_dev_err(dd,
-			      "Don't yet know about board with ID %u\n",
-			      boardrev);
-		snprintf(name, namelen, "Unknown_InfiniPath_PCIe_%u",
-			 boardrev);
-		break;
-	}
-	if (n)
-		snprintf(name, namelen, "%s", n);
-
-	if (dd->ipath_majrev != 4 || !dd->ipath_minrev || dd->ipath_minrev>2) {
-		ipath_dev_err(dd, "Unsupported InfiniPath hardware revision %u.%u!\n",
-			      dd->ipath_majrev, dd->ipath_minrev);
-		ret = 1;
-	} else {
-		ret = 0;
-		if (dd->ipath_minrev >= 2)
-			dd->ipath_f_put_tid = ipath_pe_put_tid_2;
-	}
-
-	/*
-	 * set here, not in ipath_init_*_funcs because we have to do
-	 * it after we can read chip registers.
-	 */
-	dd->ipath_ureg_align =
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_pagealign);
-
-	return ret;
-}
-
-/**
- * ipath_pe_init_hwerrors - enable hardware errors
- * @dd: the infinipath device
- *
- * now that we have finished initializing everything that might reasonably
- * cause a hardware error, and cleared those errors bits as they occur,
- * we can enable hardware errors in the mask (potentially enabling
- * freeze mode), and enable hardware errors as errors (along with
- * everything else) in errormask
- */
-static void ipath_pe_init_hwerrors(struct ipath_devdata *dd)
-{
-	ipath_err_t val;
-	u64 extsval;
-
-	extsval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
-
-	if (!(extsval & INFINIPATH_EXTS_MEMBIST_ENDTEST))
-		ipath_dev_err(dd, "MemBIST did not complete!\n");
-	if (extsval & INFINIPATH_EXTS_MEMBIST_FOUND)
-		ipath_dbg("MemBIST corrected\n");
-
-	val = ~0ULL;	/* barring bugs, all hwerrors become interrupts, */
-
-	if (!dd->ipath_boardrev)	// no PLL for Emulator
-		val &= ~INFINIPATH_HWE_SERDESPLLFAILED;
-
-	if (dd->ipath_minrev < 2) {
-		/* workaround bug 9460 in internal interface bus parity
-		 * checking. Fixed (HW bug 9490) in Rev2.
-		 */
-		val &= ~INFINIPATH_HWE_PCIEBUSPARITYRADM;
-	}
-	dd->ipath_hwerrmask = val;
-}
-
-/**
- * ipath_pe_bringup_serdes - bring up the serdes
- * @dd: the infinipath device
- */
-static int ipath_pe_bringup_serdes(struct ipath_devdata *dd)
-{
-	u64 val, config1, prev_val;
-	int ret = 0;
-
-	ipath_dbg("Trying to bringup serdes\n");
-
-	if (ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus) &
-	    INFINIPATH_HWE_SERDESPLLFAILED) {
-		ipath_dbg("At start, serdes PLL failed bit set "
-			  "in hwerrstatus, clearing and continuing\n");
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
-				 INFINIPATH_HWE_SERDESPLLFAILED);
-	}
-
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
-	config1 = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig1);
-
-	ipath_cdbg(VERBOSE, "SerDes status config0=%llx config1=%llx, "
-		   "xgxsconfig %llx\n", (unsigned long long) val,
-		   (unsigned long long) config1, (unsigned long long)
-		   ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
-
-	/*
-	 * Force reset on, also set rxdetect enable.  Must do before reading
-	 * serdesstatus at least for simulation, or some of the bits in
-	 * serdes status will come back as undefined and cause simulation
-	 * failures
-	 */
-	val |= INFINIPATH_SERDC0_RESET_PLL | INFINIPATH_SERDC0_RXDETECT_EN
-		| INFINIPATH_SERDC0_L1PWR_DN;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
-	/* be sure chip saw it */
-	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
-	udelay(5);		/* need pll reset set at least for a bit */
-	/*
-	 * after PLL is reset, set the per-lane Resets and TxIdle and
-	 * clear the PLL reset and rxdetect (to get falling edge).
-	 * Leave L1PWR bits set (permanently)
-	 */
-	val &= ~(INFINIPATH_SERDC0_RXDETECT_EN | INFINIPATH_SERDC0_RESET_PLL
-		 | INFINIPATH_SERDC0_L1PWR_DN);
-	val |= INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE;
-	ipath_cdbg(VERBOSE, "Clearing pll reset and setting lane resets "
-		   "and txidle (%llx)\n", (unsigned long long) val);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
-	/* be sure chip saw it */
-	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
-	/* need PLL reset clear for at least 11 usec before lane
-	 * resets cleared; give it a few more to be sure */
-	udelay(15);
-	val &= ~(INFINIPATH_SERDC0_RESET_MASK | INFINIPATH_SERDC0_TXIDLE);
-
-	ipath_cdbg(VERBOSE, "Clearing lane resets and txidle "
-		   "(writing %llx)\n", (unsigned long long) val);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
-	/* be sure chip saw it */
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
-
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
-	prev_val = val;
-	if (val & INFINIPATH_XGXS_RESET)
-		val &= ~INFINIPATH_XGXS_RESET;
-	if (((val >> INFINIPATH_XGXS_RX_POL_SHIFT) &
-	     INFINIPATH_XGXS_RX_POL_MASK) != dd->ipath_rx_pol_inv ) {
-		/* need to compensate for Tx inversion in partner */
-		val &= ~(INFINIPATH_XGXS_RX_POL_MASK <<
-		         INFINIPATH_XGXS_RX_POL_SHIFT);
-		val |= dd->ipath_rx_pol_inv <<
-			INFINIPATH_XGXS_RX_POL_SHIFT;
-	}
-	if (val != prev_val)
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
-
-	/* clear current and de-emphasis bits */
-	config1 &= ~0x0ffffffff00ULL;
-	/* set current to 20ma */
-	config1 |= 0x00000000000ULL;
-	/* set de-emphasis to -5.68dB */
-	config1 |= 0x0cccc000000ULL;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig1, config1);
-
-	ipath_cdbg(VERBOSE, "done: SerDes status config0=%llx "
-		   "config1=%llx, sstatus=%llx xgxs=%llx\n",
-		   (unsigned long long) val, (unsigned long long) config1,
-		   (unsigned long long)
-		   ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesstatus),
-		   (unsigned long long)
-		   ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
-
-	return ret;
-}
-
-/**
- * ipath_pe_quiet_serdes - set serdes to txidle
- * @dd: the infinipath device
- * Called when driver is being unloaded
- */
-static void ipath_pe_quiet_serdes(struct ipath_devdata *dd)
-{
-	u64 val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_serdesconfig0);
-
-	val |= INFINIPATH_SERDC0_TXIDLE;
-	ipath_dbg("Setting TxIdleEn on serdes (config0 = %llx)\n",
-		  (unsigned long long) val);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_serdesconfig0, val);
-}
-
-static int ipath_pe_intconfig(struct ipath_devdata *dd)
-{
-	u32 chiprev;
-
-	/*
-	 * If the chip supports added error indication via GPIO pins,
-	 * enable interrupts on those bits so the interrupt routine
-	 * can count the events. Also set flag so interrupt routine
-	 * can know they are expected.
-	 */
-	chiprev = dd->ipath_revision >> INFINIPATH_R_CHIPREVMINOR_SHIFT;
-	if ((chiprev & INFINIPATH_R_CHIPREVMINOR_MASK) > 1) {
-		/* Rev2+ reports extra errors via internal GPIO pins */
-		dd->ipath_flags |= IPATH_GPIO_ERRINTRS;
-		dd->ipath_gpio_mask |= IPATH_GPIO_ERRINTR_MASK;
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_gpio_mask,
-				 dd->ipath_gpio_mask);
-	}
-	return 0;
-}
-
-/**
- * ipath_setup_pe_setextled - set the state of the two external LEDs
- * @dd: the infinipath device
- * @lst: the L state
- * @ltst: the LT state
-
- * These LEDs indicate the physical and logical state of IB link.
- * For this chip (at least with recommended board pinouts), LED1
- * is Yellow (logical state) and LED2 is Green (physical state),
- *
- * Note:  We try to match the Mellanox HCA LED behavior as best
- * we can.  Green indicates physical link state is OK (something is
- * plugged in, and we can train).
- * Amber indicates the link is logically up (ACTIVE).
- * Mellanox further blinks the amber LED to indicate data packet
- * activity, but we have no hardware support for that, so it would
- * require waking up every 10-20 msecs and checking the counters
- * on the chip, and then turning the LED off if appropriate.  That's
- * visible overhead, so not something we will do.
- *
- */
-static void ipath_setup_pe_setextled(struct ipath_devdata *dd, u64 lst,
-				     u64 ltst)
-{
-	u64 extctl;
-	unsigned long flags = 0;
-
-	/* the diags use the LED to indicate diag info, so we leave
-	 * the external LED alone when the diags are running */
-	if (ipath_diag_inuse)
-		return;
-
-	/* Allow override of LED display for, e.g. Locating system in rack */
-	if (dd->ipath_led_override) {
-		ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
-			? INFINIPATH_IBCS_LT_STATE_LINKUP
-			: INFINIPATH_IBCS_LT_STATE_DISABLED;
-		lst = (dd->ipath_led_override & IPATH_LED_LOG)
-			? INFINIPATH_IBCS_L_STATE_ACTIVE
-			: INFINIPATH_IBCS_L_STATE_DOWN;
-	}
-
-	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
-	extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON |
-				       INFINIPATH_EXTC_LED2PRIPORT_ON);
-
-	if (ltst == INFINIPATH_IBCS_LT_STATE_LINKUP)
-		extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON;
-	if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE)
-		extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON;
-	dd->ipath_extctrl = extctl;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
-	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
-}
-
-/**
- * ipath_setup_pe_cleanup - clean up any per-chip chip-specific stuff
- * @dd: the infinipath device
- *
- * This is called during driver unload.
- * We do the pci_disable_msi here, not in generic code, because it
- * isn't used for the HT chips. If we do end up needing pci_enable_msi
- * at some point in the future for HT, we'll move the call back
- * into the main init_one code.
- */
-static void ipath_setup_pe_cleanup(struct ipath_devdata *dd)
-{
-	dd->ipath_msi_lo = 0;	/* just in case unload fails */
-	pci_disable_msi(dd->pcidev);
-}
-
-static void ipath_6120_pcie_params(struct ipath_devdata *dd)
-{
-	u16 linkstat, speed;
-	int pos;
-
-	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
-	if (!pos) {
-		ipath_dev_err(dd, "Can't find PCI Express capability!\n");
-		goto bail;
-	}
-
-	pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA,
-			     &linkstat);
-	/*
-	 * speed is bits 0-4, linkwidth is bits 4-8
-	 * no defines for them in headers
-	 */
-	speed = linkstat & 0xf;
-	linkstat >>= 4;
-	linkstat &= 0x1f;
-	dd->ipath_lbus_width = linkstat;
-
-	switch (speed) {
-	case 1:
-		dd->ipath_lbus_speed = 2500; /* Gen1, 2.5GHz */
-		break;
-	case 2:
-		dd->ipath_lbus_speed = 5000; /* Gen1, 5GHz */
-		break;
-	default: /* not defined, assume gen1 */
-		dd->ipath_lbus_speed = 2500;
-		break;
-	}
-
-	if (linkstat < 8)
-		ipath_dev_err(dd,
-			"PCIe width %u (x8 HCA), performance reduced\n",
-			linkstat);
-	else
-		ipath_cdbg(VERBOSE, "PCIe speed %u width %u (x8 HCA)\n",
-			dd->ipath_lbus_speed, linkstat);
-
-	if (speed != 1)
-		ipath_dev_err(dd,
-			"PCIe linkspeed %u is incorrect; "
-			"should be 1 (2500)!\n", speed);
-bail:
-	/* fill in string, even on errors */
-	snprintf(dd->ipath_lbus_info, sizeof(dd->ipath_lbus_info),
-		"PCIe,%uMHz,x%u\n",
-		dd->ipath_lbus_speed,
-		dd->ipath_lbus_width);
-
-	return;
-}
-
-/**
- * ipath_setup_pe_config - setup PCIe config related stuff
- * @dd: the infinipath device
- * @pdev: the PCI device
- *
- * The pci_enable_msi() call will fail on systems with MSI quirks
- * such as those with AMD8131, even if the device of interest is not
- * attached to that device, (in the 2.6.13 - 2.6.15 kernels, at least, fixed
- * late in 2.6.16).
- * All that can be done is to edit the kernel source to remove the quirk
- * check until that is fixed.
- * We do not need to call enable_msi() for our HyperTransport chip,
- * even though it uses MSI, and we want to avoid the quirk warning, so
- * So we call enable_msi only for PCIe.  If we do end up needing
- * pci_enable_msi at some point in the future for HT, we'll move the
- * call back into the main init_one code.
- * We save the msi lo and hi values, so we can restore them after
- * chip reset (the kernel PCI infrastructure doesn't yet handle that
- * correctly).
- */
-static int ipath_setup_pe_config(struct ipath_devdata *dd,
-				 struct pci_dev *pdev)
-{
-	int pos, ret;
-
-	dd->ipath_msi_lo = 0;	/* used as a flag during reset processing */
-	ret = pci_enable_msi(dd->pcidev);
-	if (ret)
-		ipath_dev_err(dd, "pci_enable_msi failed: %d, "
-			      "interrupts may not work\n", ret);
-	/* continue even if it fails, we may still be OK... */
-	dd->ipath_irq = pdev->irq;
-
-	if ((pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
-		u16 control;
-		pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
-				      &dd->ipath_msi_lo);
-		pci_read_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
-				      &dd->ipath_msi_hi);
-		pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
-				     &control);
-		/* now save the data (vector) info */
-		pci_read_config_word(dd->pcidev,
-				     pos + ((control & PCI_MSI_FLAGS_64BIT)
-					    ? 12 : 8),
-				     &dd->ipath_msi_data);
-		ipath_cdbg(VERBOSE, "Read msi data 0x%x from config offset "
-			   "0x%x, control=0x%x\n", dd->ipath_msi_data,
-			   pos + ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
-			   control);
-		/* we save the cachelinesize also, although it doesn't
-		 * really matter */
-		pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
-				     &dd->ipath_pci_cacheline);
-	} else
-		ipath_dev_err(dd, "Can't find MSI capability, "
-			      "can't save MSI settings for reset\n");
-
-	ipath_6120_pcie_params(dd);
-
-	dd->ipath_link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
-	dd->ipath_link_speed_supported = IPATH_IB_SDR;
-	dd->ipath_link_width_enabled = IB_WIDTH_4X;
-	dd->ipath_link_speed_enabled = dd->ipath_link_speed_supported;
-	/* these can't change for this chip, so set once */
-	dd->ipath_link_width_active = dd->ipath_link_width_enabled;
-	dd->ipath_link_speed_active = dd->ipath_link_speed_enabled;
-	return 0;
-}
-
-static void ipath_init_pe_variables(struct ipath_devdata *dd)
-{
-	/*
-	 * setup the register offsets, since they are different for each
-	 * chip
-	 */
-	dd->ipath_kregs = &ipath_pe_kregs;
-	dd->ipath_cregs = &ipath_pe_cregs;
-
-	/*
-	 * bits for selecting i2c direction and values,
-	 * used for I2C serial flash
-	 */
-	dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
-	dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
-	dd->ipath_gpio_sda = IPATH_GPIO_SDA;
-	dd->ipath_gpio_scl = IPATH_GPIO_SCL;
-
-	/*
-	 * Fill in data for field-values that change in newer chips.
-	 * We dynamically specify only the mask for LINKTRAININGSTATE
-	 * and only the shift for LINKSTATE, as they are the only ones
-	 * that change.  Also precalculate the 3 link states of interest
-	 * and the combined mask.
-	 */
-	dd->ibcs_ls_shift = IBA6120_IBCS_LINKSTATE_SHIFT;
-	dd->ibcs_lts_mask = IBA6120_IBCS_LINKTRAININGSTATE_MASK;
-	dd->ibcs_mask = (INFINIPATH_IBCS_LINKSTATE_MASK <<
-		dd->ibcs_ls_shift) | dd->ibcs_lts_mask;
-	dd->ib_init = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_INIT << dd->ibcs_ls_shift);
-	dd->ib_arm = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_ARM << dd->ibcs_ls_shift);
-	dd->ib_active = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_ACTIVE << dd->ibcs_ls_shift);
-
-	/*
-	 * Fill in data for ibcc field-values that change in newer chips.
-	 * We dynamically specify only the mask for LINKINITCMD
-	 * and only the shift for LINKCMD and MAXPKTLEN, as they are
-	 * the only ones that change.
-	 */
-	dd->ibcc_lic_mask = INFINIPATH_IBCC_LINKINITCMD_MASK;
-	dd->ibcc_lc_shift = INFINIPATH_IBCC_LINKCMD_SHIFT;
-	dd->ibcc_mpl_shift = INFINIPATH_IBCC_MAXPKTLEN_SHIFT;
-
-	/* Fill in shifts for RcvCtrl. */
-	dd->ipath_r_portenable_shift = INFINIPATH_R_PORTENABLE_SHIFT;
-	dd->ipath_r_intravail_shift = INFINIPATH_R_INTRAVAIL_SHIFT;
-	dd->ipath_r_tailupd_shift = INFINIPATH_R_TAILUPD_SHIFT;
-	dd->ipath_r_portcfg_shift = 0; /* Not on IBA6120 */
-
-	/* variables for sanity checking interrupt and errors */
-	dd->ipath_hwe_bitsextant =
-		(INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) |
-		(INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) |
-		(INFINIPATH_HWE_PCIEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) |
-		INFINIPATH_HWE_PCIE1PLLFAILED |
-		INFINIPATH_HWE_PCIE0PLLFAILED |
-		INFINIPATH_HWE_PCIEPOISONEDTLP |
-		INFINIPATH_HWE_PCIECPLTIMEOUT |
-		INFINIPATH_HWE_PCIEBUSPARITYXTLH |
-		INFINIPATH_HWE_PCIEBUSPARITYXADM |
-		INFINIPATH_HWE_PCIEBUSPARITYRADM |
-		INFINIPATH_HWE_MEMBISTFAILED |
-		INFINIPATH_HWE_COREPLL_FBSLIP |
-		INFINIPATH_HWE_COREPLL_RFSLIP |
-		INFINIPATH_HWE_SERDESPLLFAILED |
-		INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
-		INFINIPATH_HWE_IBCBUSFRSPCPARITYERR;
-	dd->ipath_i_bitsextant =
-		(INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
-		(INFINIPATH_I_RCVAVAIL_MASK <<
-		 INFINIPATH_I_RCVAVAIL_SHIFT) |
-		INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
-		INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO;
-	dd->ipath_e_bitsextant =
-		INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
-		INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
-		INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
-		INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RUNEXPCHAR |
-		INFINIPATH_E_RUNSUPVL | INFINIPATH_E_REBP |
-		INFINIPATH_E_RIBFLOW | INFINIPATH_E_RBADVERSION |
-		INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL |
-		INFINIPATH_E_RBADTID | INFINIPATH_E_RHDRLEN |
-		INFINIPATH_E_RHDR | INFINIPATH_E_RIBLOSTLINK |
-		INFINIPATH_E_SMINPKTLEN | INFINIPATH_E_SMAXPKTLEN |
-		INFINIPATH_E_SUNDERRUN | INFINIPATH_E_SPKTLEN |
-		INFINIPATH_E_SDROPPEDSMPPKT | INFINIPATH_E_SDROPPEDDATAPKT |
-		INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM |
-		INFINIPATH_E_SUNSUPVL | INFINIPATH_E_IBSTATUSCHANGED |
-		INFINIPATH_E_INVALIDADDR | INFINIPATH_E_RESET |
-		INFINIPATH_E_HARDWARE;
-
-	dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
-	dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
-	dd->ipath_i_rcvavail_shift = INFINIPATH_I_RCVAVAIL_SHIFT;
-	dd->ipath_i_rcvurg_shift = INFINIPATH_I_RCVURG_SHIFT;
-
-	/*
-	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
-	 * 2 is Some Misc, 3 is reserved for future.
-	 */
-	dd->ipath_eep_st_masks[0].hwerrs_to_log =
-		INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
-		INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
-
-	/* Ignore errors in PIO/PBC on systems with unordered write-combining */
-	if (ipath_unordered_wc())
-		dd->ipath_eep_st_masks[0].hwerrs_to_log &= ~TXE_PIO_PARITY;
-
-	dd->ipath_eep_st_masks[1].hwerrs_to_log =
-		INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
-		INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
-
-	dd->ipath_eep_st_masks[2].errs_to_log = INFINIPATH_E_RESET;
-	dd->delay_mult = 2; /* SDR, 4X, can't change */
-}
-
-/* setup the MSI stuff again after a reset.  I'd like to just call
- * pci_enable_msi() and request_irq() again, but when I do that,
- * the MSI enable bit doesn't get set in the command word, and
- * we switch to to a different interrupt vector, which is confusing,
- * so I instead just do it all inline.  Perhaps somehow can tie this
- * into the PCIe hotplug support at some point
- * Note, because I'm doing it all here, I don't call pci_disable_msi()
- * or free_irq() at the start of ipath_setup_pe_reset().
- */
-static int ipath_reinit_msi(struct ipath_devdata *dd)
-{
-	int pos;
-	u16 control;
-	int ret;
-
-	if (!dd->ipath_msi_lo) {
-		dev_info(&dd->pcidev->dev, "Can't restore MSI config, "
-			 "initial setup failed?\n");
-		ret = 0;
-		goto bail;
-	}
-
-	if (!(pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI))) {
-		ipath_dev_err(dd, "Can't find MSI capability, "
-			      "can't restore MSI settings\n");
-		ret = 0;
-		goto bail;
-	}
-	ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
-		   dd->ipath_msi_lo, pos + PCI_MSI_ADDRESS_LO);
-	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
-			       dd->ipath_msi_lo);
-	ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
-		   dd->ipath_msi_hi, pos + PCI_MSI_ADDRESS_HI);
-	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
-			       dd->ipath_msi_hi);
-	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
-	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
-		ipath_cdbg(VERBOSE, "MSI control at off %x was %x, "
-			   "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS,
-			   control, control | PCI_MSI_FLAGS_ENABLE);
-		control |= PCI_MSI_FLAGS_ENABLE;
-		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
-				      control);
-	}
-	/* now rewrite the data (vector) info */
-	pci_write_config_word(dd->pcidev, pos +
-			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
-			      dd->ipath_msi_data);
-	/* we restore the cachelinesize also, although it doesn't really
-	 * matter */
-	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
-			      dd->ipath_pci_cacheline);
-	/* and now set the pci master bit again */
-	pci_set_master(dd->pcidev);
-	ret = 1;
-
-bail:
-	return ret;
-}
-
-/* This routine sleeps, so it can only be called from user context, not
- * from interrupt context.  If we need interrupt context, we can split
- * it into two routines.
-*/
-static int ipath_setup_pe_reset(struct ipath_devdata *dd)
-{
-	u64 val;
-	int i;
-	int ret;
-	u16 cmdval;
-
-	pci_read_config_word(dd->pcidev, PCI_COMMAND, &cmdval);
-
-	/* Use ERROR so it shows up in logs, etc. */
-	ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit);
-	/* keep chip from being accessed in a few places */
-	dd->ipath_flags &= ~(IPATH_INITTED|IPATH_PRESENT);
-	val = dd->ipath_control | INFINIPATH_C_RESET;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val);
-	mb();
-
-	for (i = 1; i <= 5; i++) {
-		int r;
-		/* allow MBIST, etc. to complete; longer on each retry.
-		 * We sometimes get machine checks from bus timeout if no
-		 * response, so for now, make it *really* long.
-		 */
-		msleep(1000 + (1 + i) * 2000);
-		if ((r =
-		     pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
-					    dd->ipath_pcibar0)))
-			ipath_dev_err(dd, "rewrite of BAR0 failed: %d\n",
-				      r);
-		if ((r =
-		     pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
-					    dd->ipath_pcibar1)))
-			ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n",
-				      r);
-		/* now re-enable memory access */
-		pci_write_config_word(dd->pcidev, PCI_COMMAND, cmdval);
-		if ((r = pci_enable_device(dd->pcidev)))
-			ipath_dev_err(dd, "pci_enable_device failed after "
-				      "reset: %d\n", r);
-		/*
-		 * whether it fully enabled or not, mark as present,
-		 * again (but not INITTED)
-		 */
-		dd->ipath_flags |= IPATH_PRESENT;
-		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision);
-		if (val == dd->ipath_revision) {
-			ipath_cdbg(VERBOSE, "Got matching revision "
-				   "register %llx on try %d\n",
-				   (unsigned long long) val, i);
-			ret = ipath_reinit_msi(dd);
-			goto bail;
-		}
-		/* Probably getting -1 back */
-		ipath_dbg("Didn't get expected revision register, "
-			  "got %llx, try %d\n", (unsigned long long) val,
-			  i + 1);
-	}
-	ret = 0; /* failed */
-
-bail:
-	if (ret)
-		ipath_6120_pcie_params(dd);
-	return ret;
-}
-
-/**
- * ipath_pe_put_tid - write a TID in chip
- * @dd: the infinipath device
- * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
- * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
- *
- * This exists as a separate routine to allow for special locking etc.
- * It's used for both the full cleanup on exit, as well as the normal
- * setup and teardown.
- */
-static void ipath_pe_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
-			     u32 type, unsigned long pa)
-{
-	u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
-	unsigned long flags = 0; /* keep gcc quiet */
-	int tidx;
-	spinlock_t *tidlockp;
-
-	if (!dd->ipath_kregbase)
-		return;
-
-	if (pa != dd->ipath_tidinvalid) {
-		if (pa & ((1U << 11) - 1)) {
-			dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
-				 "not 2KB aligned!\n", pa);
-			return;
-		}
-		pa >>= 11;
-		/* paranoia check */
-		if (pa & ~INFINIPATH_RT_ADDR_MASK)
-			ipath_dev_err(dd,
-				      "BUG: Physical page address 0x%lx "
-				      "has bits set in 31-29\n", pa);
-
-		if (type == RCVHQ_RCV_TYPE_EAGER)
-			pa |= dd->ipath_tidtemplate;
-		else /* for now, always full 4KB page */
-			pa |= 2 << 29;
-	}
-
-	/*
-	 * Workaround chip bug 9437 by writing the scratch register
-	 * before and after the TID, and with an io write barrier.
-	 * We use a spinlock around the writes, so they can't intermix
-	 * with other TID (eager or expected) writes (the chip bug
-	 * is triggered by back to back TID writes). Unfortunately, this
-	 * call can be done from interrupt level for the port 0 eager TIDs,
-	 * so we have to use irqsave locks.
-	 */
-	/*
-	 * Assumes tidptr always > ipath_egrtidbase
-	 * if type == RCVHQ_RCV_TYPE_EAGER.
-	 */
-	tidx = tidptr - dd->ipath_egrtidbase;
-
-	tidlockp = (type == RCVHQ_RCV_TYPE_EAGER && tidx < dd->ipath_rcvegrcnt)
-		? &dd->ipath_kernel_tid_lock : &dd->ipath_user_tid_lock;
-	spin_lock_irqsave(tidlockp, flags);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xfeeddeaf);
-	writel(pa, tidp32);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_scratch, 0xdeadbeef);
-	mmiowb();
-	spin_unlock_irqrestore(tidlockp, flags);
-}
-
-/**
- * ipath_pe_put_tid_2 - write a TID in chip, Revision 2 or higher
- * @dd: the infinipath device
- * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0) for expected
- * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
- *
- * This exists as a separate routine to allow for selection of the
- * appropriate "flavor". The static calls in cleanup just use the
- * revision-agnostic form, as they are not performance critical.
- */
-static void ipath_pe_put_tid_2(struct ipath_devdata *dd, u64 __iomem *tidptr,
-			     u32 type, unsigned long pa)
-{
-	u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
-	u32 tidx;
-
-	if (!dd->ipath_kregbase)
-		return;
-
-	if (pa != dd->ipath_tidinvalid) {
-		if (pa & ((1U << 11) - 1)) {
-			dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
-				 "not 2KB aligned!\n", pa);
-			return;
-		}
-		pa >>= 11;
-		/* paranoia check */
-		if (pa & ~INFINIPATH_RT_ADDR_MASK)
-			ipath_dev_err(dd,
-				      "BUG: Physical page address 0x%lx "
-				      "has bits set in 31-29\n", pa);
-
-		if (type == RCVHQ_RCV_TYPE_EAGER)
-			pa |= dd->ipath_tidtemplate;
-		else /* for now, always full 4KB page */
-			pa |= 2 << 29;
-	}
-	tidx = tidptr - dd->ipath_egrtidbase;
-	writel(pa, tidp32);
-	mmiowb();
-}
-
-
-/**
- * ipath_pe_clear_tid - clear all TID entries for a port, expected and eager
- * @dd: the infinipath device
- * @port: the port
- *
- * clear all TID entries for a port, expected and eager.
- * Used from ipath_close().  On this chip, TIDs are only 32 bits,
- * not 64, but they are still on 64 bit boundaries, so tidbase
- * is declared as u64 * for the pointer math, even though we write 32 bits
- */
-static void ipath_pe_clear_tids(struct ipath_devdata *dd, unsigned port)
-{
-	u64 __iomem *tidbase;
-	unsigned long tidinv;
-	int i;
-
-	if (!dd->ipath_kregbase)
-		return;
-
-	ipath_cdbg(VERBOSE, "Invalidate TIDs for port %u\n", port);
-
-	tidinv = dd->ipath_tidinvalid;
-	tidbase = (u64 __iomem *)
-		((char __iomem *)(dd->ipath_kregbase) +
-		 dd->ipath_rcvtidbase +
-		 port * dd->ipath_rcvtidcnt * sizeof(*tidbase));
-
-	for (i = 0; i < dd->ipath_rcvtidcnt; i++)
-		dd->ipath_f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
-				 tidinv);
-
-	tidbase = (u64 __iomem *)
-		((char __iomem *)(dd->ipath_kregbase) +
-		 dd->ipath_rcvegrbase +
-		 port * dd->ipath_rcvegrcnt * sizeof(*tidbase));
-
-	for (i = 0; i < dd->ipath_rcvegrcnt; i++)
-		dd->ipath_f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
-				 tidinv);
-}
-
-/**
- * ipath_pe_tidtemplate - setup constants for TID updates
- * @dd: the infinipath device
- *
- * We setup stuff that we use a lot, to avoid calculating each time
- */
-static void ipath_pe_tidtemplate(struct ipath_devdata *dd)
-{
-	u32 egrsize = dd->ipath_rcvegrbufsize;
-
-	/* For now, we always allocate 4KB buffers (at init) so we can
-	 * receive max size packets.  We may want a module parameter to
-	 * specify 2KB or 4KB and/or make be per port instead of per device
-	 * for those who want to reduce memory footprint.  Note that the
-	 * ipath_rcvhdrentsize size must be large enough to hold the largest
-	 * IB header (currently 96 bytes) that we expect to handle (plus of
-	 * course the 2 dwords of RHF).
-	 */
-	if (egrsize == 2048)
-		dd->ipath_tidtemplate = 1U << 29;
-	else if (egrsize == 4096)
-		dd->ipath_tidtemplate = 2U << 29;
-	else {
-		egrsize = 4096;
-		dev_info(&dd->pcidev->dev, "BUG: unsupported egrbufsize "
-			 "%u, using %u\n", dd->ipath_rcvegrbufsize,
-			 egrsize);
-		dd->ipath_tidtemplate = 2U << 29;
-	}
-	dd->ipath_tidinvalid = 0;
-}
-
-static int ipath_pe_early_init(struct ipath_devdata *dd)
-{
-	dd->ipath_flags |= IPATH_4BYTE_TID;
-	if (ipath_unordered_wc())
-		dd->ipath_flags |= IPATH_PIO_FLUSH_WC;
-
-	/*
-	 * For openfabrics, we need to be able to handle an IB header of
-	 * 24 dwords.  HT chip has arbitrary sized receive buffers, so we
-	 * made them the same size as the PIO buffers.  This chip does not
-	 * handle arbitrary size buffers, so we need the header large enough
-	 * to handle largest IB header, but still have room for a 2KB MTU
-	 * standard IB packet.
-	 */
-	dd->ipath_rcvhdrentsize = 24;
-	dd->ipath_rcvhdrsize = IPATH_DFLT_RCVHDRSIZE;
-	dd->ipath_rhf_offset = 0;
-	dd->ipath_egrtidbase = (u64 __iomem *)
-		((char __iomem *) dd->ipath_kregbase + dd->ipath_rcvegrbase);
-
-	dd->ipath_rcvegrbufsize = ipath_mtu4096 ? 4096 : 2048;
-	/*
-	 * the min() check here is currently a nop, but it may not always
-	 * be, depending on just how we do ipath_rcvegrbufsize
-	 */
-	dd->ipath_ibmaxlen = min(ipath_mtu4096 ? dd->ipath_piosize4k :
-				 dd->ipath_piosize2k,
-				 dd->ipath_rcvegrbufsize +
-				 (dd->ipath_rcvhdrentsize << 2));
-	dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen;
-
-	/*
-	 * We can request a receive interrupt for 1 or
-	 * more packets from current offset.  For now, we set this
-	 * up for a single packet.
-	 */
-	dd->ipath_rhdrhead_intr_off = 1ULL<<32;
-
-	ipath_get_eeprom_info(dd);
-
-	return 0;
-}
-
-int __attribute__((weak)) ipath_unordered_wc(void)
-{
-	return 0;
-}
-
-/**
- * ipath_init_pe_get_base_info - set chip-specific flags for user code
- * @pd: the infinipath port
- * @kbase: ipath_base_info pointer
- *
- * We set the PCIE flag because the lower bandwidth on PCIe vs
- * HyperTransport can affect some user packet algorithms.
- */
-static int ipath_pe_get_base_info(struct ipath_portdata *pd, void *kbase)
-{
-	struct ipath_base_info *kinfo = kbase;
-	struct ipath_devdata *dd;
-
-	if (ipath_unordered_wc()) {
-		kinfo->spi_runtime_flags |= IPATH_RUNTIME_FORCE_WC_ORDER;
-		ipath_cdbg(PROC, "Intel processor, forcing WC order\n");
-	}
-	else
-		ipath_cdbg(PROC, "Not Intel processor, WC ordered\n");
-
-	if (pd == NULL)
-		goto done;
-
-	dd = pd->port_dd;
-
-done:
-	kinfo->spi_runtime_flags |= IPATH_RUNTIME_PCIE |
-		IPATH_RUNTIME_FORCE_PIOAVAIL | IPATH_RUNTIME_PIO_REGSWAPPED;
-	return 0;
-}
-
-static void ipath_pe_free_irq(struct ipath_devdata *dd)
-{
-	free_irq(dd->ipath_irq, dd);
-	dd->ipath_irq = 0;
-}
-
-
-static struct ipath_message_header *
-ipath_pe_get_msgheader(struct ipath_devdata *dd, __le32 *rhf_addr)
-{
-	return (struct ipath_message_header *)
-		&rhf_addr[sizeof(u64) / sizeof(u32)];
-}
-
-static void ipath_pe_config_ports(struct ipath_devdata *dd, ushort cfgports)
-{
-	dd->ipath_portcnt =
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_portcnt);
-	dd->ipath_p0_rcvegrcnt =
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_rcvegrcnt);
-}
-
-static void ipath_pe_read_counters(struct ipath_devdata *dd,
-				   struct infinipath_counters *cntrs)
-{
-	cntrs->LBIntCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(LBIntCnt));
-	cntrs->LBFlowStallCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(LBFlowStallCnt));
-	cntrs->TxSDmaDescCnt = 0;
-	cntrs->TxUnsupVLErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxUnsupVLErrCnt));
-	cntrs->TxDataPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxDataPktCnt));
-	cntrs->TxFlowPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxFlowPktCnt));
-	cntrs->TxDwordCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxDwordCnt));
-	cntrs->TxLenErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxLenErrCnt));
-	cntrs->TxMaxMinLenErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxMaxMinLenErrCnt));
-	cntrs->TxUnderrunCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxUnderrunCnt));
-	cntrs->TxFlowStallCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxFlowStallCnt));
-	cntrs->TxDroppedPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(TxDroppedPktCnt));
-	cntrs->RxDroppedPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxDroppedPktCnt));
-	cntrs->RxDataPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxDataPktCnt));
-	cntrs->RxFlowPktCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxFlowPktCnt));
-	cntrs->RxDwordCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxDwordCnt));
-	cntrs->RxLenErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxLenErrCnt));
-	cntrs->RxMaxMinLenErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxMaxMinLenErrCnt));
-	cntrs->RxICRCErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxICRCErrCnt));
-	cntrs->RxVCRCErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxVCRCErrCnt));
-	cntrs->RxFlowCtrlErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxFlowCtrlErrCnt));
-	cntrs->RxBadFormatCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxBadFormatCnt));
-	cntrs->RxLinkProblemCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxLinkProblemCnt));
-	cntrs->RxEBPCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxEBPCnt));
-	cntrs->RxLPCRCErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxLPCRCErrCnt));
-	cntrs->RxBufOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxBufOvflCnt));
-	cntrs->RxTIDFullErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxTIDFullErrCnt));
-	cntrs->RxTIDValidErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxTIDValidErrCnt));
-	cntrs->RxPKeyMismatchCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxPKeyMismatchCnt));
-	cntrs->RxP0HdrEgrOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxP0HdrEgrOvflCnt));
-	cntrs->RxP1HdrEgrOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxP1HdrEgrOvflCnt));
-	cntrs->RxP2HdrEgrOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxP2HdrEgrOvflCnt));
-	cntrs->RxP3HdrEgrOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxP3HdrEgrOvflCnt));
-	cntrs->RxP4HdrEgrOvflCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(RxP4HdrEgrOvflCnt));
-	cntrs->RxP5HdrEgrOvflCnt = 0;
-	cntrs->RxP6HdrEgrOvflCnt = 0;
-	cntrs->RxP7HdrEgrOvflCnt = 0;
-	cntrs->RxP8HdrEgrOvflCnt = 0;
-	cntrs->RxP9HdrEgrOvflCnt = 0;
-	cntrs->RxP10HdrEgrOvflCnt = 0;
-	cntrs->RxP11HdrEgrOvflCnt = 0;
-	cntrs->RxP12HdrEgrOvflCnt = 0;
-	cntrs->RxP13HdrEgrOvflCnt = 0;
-	cntrs->RxP14HdrEgrOvflCnt = 0;
-	cntrs->RxP15HdrEgrOvflCnt = 0;
-	cntrs->RxP16HdrEgrOvflCnt = 0;
-	cntrs->IBStatusChangeCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(IBStatusChangeCnt));
-	cntrs->IBLinkErrRecoveryCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(IBLinkErrRecoveryCnt));
-	cntrs->IBLinkDownedCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(IBLinkDownedCnt));
-	cntrs->IBSymbolErrCnt =
-		ipath_snap_cntr(dd, IPATH_CREG_OFFSET(IBSymbolErrCnt));
-	cntrs->RxVL15DroppedPktCnt = 0;
-	cntrs->RxOtherLocalPhyErrCnt = 0;
-	cntrs->PcieRetryBufDiagQwordCnt = 0;
-	cntrs->ExcessBufferOvflCnt = dd->ipath_overrun_thresh_errs;
-	cntrs->LocalLinkIntegrityErrCnt = dd->ipath_lli_errs;
-	cntrs->RxVlErrCnt = 0;
-	cntrs->RxDlidFltrCnt = 0;
-}
-
-
-/* no interrupt fallback for these chips */
-static int ipath_pe_nointr_fallback(struct ipath_devdata *dd)
-{
-	return 0;
-}
-
-
-/*
- * reset the XGXS (between serdes and IBC).  Slightly less intrusive
- * than resetting the IBC or external link state, and useful in some
- * cases to cause some retraining.  To do this right, we reset IBC
- * as well.
- */
-static void ipath_pe_xgxs_reset(struct ipath_devdata *dd)
-{
-	u64 val, prev_val;
-
-	prev_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
-	val = prev_val | INFINIPATH_XGXS_RESET;
-	prev_val &= ~INFINIPATH_XGXS_RESET; /* be sure */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-			 dd->ipath_control & ~INFINIPATH_C_LINKENABLE);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, prev_val);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-			 dd->ipath_control);
-}
-
-
-static int ipath_pe_get_ib_cfg(struct ipath_devdata *dd, int which)
-{
-	int ret;
-
-	switch (which) {
-	case IPATH_IB_CFG_LWID:
-		ret = dd->ipath_link_width_active;
-		break;
-	case IPATH_IB_CFG_SPD:
-		ret = dd->ipath_link_speed_active;
-		break;
-	case IPATH_IB_CFG_LWID_ENB:
-		ret = dd->ipath_link_width_enabled;
-		break;
-	case IPATH_IB_CFG_SPD_ENB:
-		ret = dd->ipath_link_speed_enabled;
-		break;
-	default:
-		ret =  -ENOTSUPP;
-		break;
-	}
-	return ret;
-}
-
-
-/* we assume range checking is already done, if needed */
-static int ipath_pe_set_ib_cfg(struct ipath_devdata *dd, int which, u32 val)
-{
-	int ret = 0;
-
-	if (which == IPATH_IB_CFG_LWID_ENB)
-		dd->ipath_link_width_enabled = val;
-	else if (which == IPATH_IB_CFG_SPD_ENB)
-		dd->ipath_link_speed_enabled = val;
-	else
-		ret = -ENOTSUPP;
-	return ret;
-}
-
-static void ipath_pe_config_jint(struct ipath_devdata *dd, u16 a, u16 b)
-{
-}
-
-
-static int ipath_pe_ib_updown(struct ipath_devdata *dd, int ibup, u64 ibcs)
-{
-	ipath_setup_pe_setextled(dd, ipath_ib_linkstate(dd, ibcs),
-		ipath_ib_linktrstate(dd, ibcs));
-	return 0;
-}
-
-
-/**
- * ipath_init_iba6120_funcs - set up the chip-specific function pointers
- * @dd: the infinipath device
- *
- * This is global, and is called directly at init to set up the
- * chip-specific function pointers for later use.
- */
-void ipath_init_iba6120_funcs(struct ipath_devdata *dd)
-{
-	dd->ipath_f_intrsetup = ipath_pe_intconfig;
-	dd->ipath_f_bus = ipath_setup_pe_config;
-	dd->ipath_f_reset = ipath_setup_pe_reset;
-	dd->ipath_f_get_boardname = ipath_pe_boardname;
-	dd->ipath_f_init_hwerrors = ipath_pe_init_hwerrors;
-	dd->ipath_f_early_init = ipath_pe_early_init;
-	dd->ipath_f_handle_hwerrors = ipath_pe_handle_hwerrors;
-	dd->ipath_f_quiet_serdes = ipath_pe_quiet_serdes;
-	dd->ipath_f_bringup_serdes = ipath_pe_bringup_serdes;
-	dd->ipath_f_clear_tids = ipath_pe_clear_tids;
-	/*
-	 * _f_put_tid may get changed after we read the chip revision,
-	 * but we start with the safe version for all revs
-	 */
-	dd->ipath_f_put_tid = ipath_pe_put_tid;
-	dd->ipath_f_cleanup = ipath_setup_pe_cleanup;
-	dd->ipath_f_setextled = ipath_setup_pe_setextled;
-	dd->ipath_f_get_base_info = ipath_pe_get_base_info;
-	dd->ipath_f_free_irq = ipath_pe_free_irq;
-	dd->ipath_f_tidtemplate = ipath_pe_tidtemplate;
-	dd->ipath_f_intr_fallback = ipath_pe_nointr_fallback;
-	dd->ipath_f_xgxs_reset = ipath_pe_xgxs_reset;
-	dd->ipath_f_get_msgheader = ipath_pe_get_msgheader;
-	dd->ipath_f_config_ports = ipath_pe_config_ports;
-	dd->ipath_f_read_counters = ipath_pe_read_counters;
-	dd->ipath_f_get_ib_cfg = ipath_pe_get_ib_cfg;
-	dd->ipath_f_set_ib_cfg = ipath_pe_set_ib_cfg;
-	dd->ipath_f_config_jint = ipath_pe_config_jint;
-	dd->ipath_f_ib_updown = ipath_pe_ib_updown;
-
-
-	/* initialize chip-specific variables */
-	ipath_init_pe_variables(dd);
-}
-
diff --git a/drivers/infiniband/hw/ipath/ipath_iba7220.c b/drivers/infiniband/hw/ipath/ipath_iba7220.c
deleted file mode 100644
index fb70712ac85..00000000000
--- a/drivers/infiniband/hw/ipath/ipath_iba7220.c
+++ /dev/null
@@ -1,2558 +0,0 @@
-/*
- * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
- * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
- *
- * This software is available to you under a choice of one of two
- * licenses.  You may choose to be licensed under the terms of the GNU
- * General Public License (GPL) Version 2, available from the file
- * COPYING in the main directory of this source tree, or the
- * OpenIB.org BSD license below:
- *
- *     Redistribution and use in source and binary forms, with or
- *     without modification, are permitted provided that the following
- *     conditions are met:
- *
- *      - Redistributions of source code must retain the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer.
- *
- *      - Redistributions in binary form must reproduce the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer in the documentation and/or other materials
- *        provided with the distribution.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
- * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-/*
- * This file contains all of the code that is specific to the
- * InfiniPath 7220 chip (except that specific to the SerDes)
- */
-
-#include <linux/interrupt.h>
-#include <linux/pci.h>
-#include <linux/delay.h>
-#include <linux/io.h>
-#include <rdma/ib_verbs.h>
-
-#include "ipath_kernel.h"
-#include "ipath_registers.h"
-#include "ipath_7220.h"
-
-static void ipath_setup_7220_setextled(struct ipath_devdata *, u64, u64);
-
-static unsigned ipath_compat_ddr_negotiate = 1;
-
-module_param_named(compat_ddr_negotiate, ipath_compat_ddr_negotiate, uint,
-			S_IWUSR | S_IRUGO);
-MODULE_PARM_DESC(compat_ddr_negotiate,
-		"Attempt pre-IBTA 1.2 DDR speed negotiation");
-
-static unsigned ipath_sdma_fetch_arb = 1;
-module_param_named(fetch_arb, ipath_sdma_fetch_arb, uint, S_IRUGO);
-MODULE_PARM_DESC(fetch_arb, "IBA7220: change SDMA descriptor arbitration");
-
-/*
- * This file contains almost all the chip-specific register information and
- * access functions for the QLogic InfiniPath 7220 PCI-Express chip, with the
- * exception of SerDes support, which in in ipath_sd7220.c.
- *
- * This lists the InfiniPath registers, in the actual chip layout.
- * This structure should never be directly accessed.
- */
-struct _infinipath_do_not_use_kernel_regs {
-	unsigned long long Revision;
-	unsigned long long Control;
-	unsigned long long PageAlign;
-	unsigned long long PortCnt;
-	unsigned long long DebugPortSelect;
-	unsigned long long DebugSigsIntSel; /* was Reserved0;*/
-	unsigned long long SendRegBase;
-	unsigned long long UserRegBase;
-	unsigned long long CounterRegBase;
-	unsigned long long Scratch;
-	unsigned long long EEPROMAddrCmd; /* was Reserved1; */
-	unsigned long long EEPROMData; /* was Reserved2; */
-	unsigned long long IntBlocked;
-	unsigned long long IntMask;
-	unsigned long long IntStatus;
-	unsigned long long IntClear;
-	unsigned long long ErrorMask;
-	unsigned long long ErrorStatus;
-	unsigned long long ErrorClear;
-	unsigned long long HwErrMask;
-	unsigned long long HwErrStatus;
-	unsigned long long HwErrClear;
-	unsigned long long HwDiagCtrl;
-	unsigned long long MDIO;
-	unsigned long long IBCStatus;
-	unsigned long long IBCCtrl;
-	unsigned long long ExtStatus;
-	unsigned long long ExtCtrl;
-	unsigned long long GPIOOut;
-	unsigned long long GPIOMask;
-	unsigned long long GPIOStatus;
-	unsigned long long GPIOClear;
-	unsigned long long RcvCtrl;
-	unsigned long long RcvBTHQP;
-	unsigned long long RcvHdrSize;
-	unsigned long long RcvHdrCnt;
-	unsigned long long RcvHdrEntSize;
-	unsigned long long RcvTIDBase;
-	unsigned long long RcvTIDCnt;
-	unsigned long long RcvEgrBase;
-	unsigned long long RcvEgrCnt;
-	unsigned long long RcvBufBase;
-	unsigned long long RcvBufSize;
-	unsigned long long RxIntMemBase;
-	unsigned long long RxIntMemSize;
-	unsigned long long RcvPartitionKey;
-	unsigned long long RcvQPMulticastPort;
-	unsigned long long RcvPktLEDCnt;
-	unsigned long long IBCDDRCtrl;
-	unsigned long long HRTBT_GUID;
-	unsigned long long IB_SDTEST_IF_TX;
-	unsigned long long IB_SDTEST_IF_RX;
-	unsigned long long IBCDDRCtrl2;
-	unsigned long long IBCDDRStatus;
-	unsigned long long JIntReload;
-	unsigned long long IBNCModeCtrl;
-	unsigned long long SendCtrl;
-	unsigned long long SendBufBase;
-	unsigned long long SendBufSize;
-	unsigned long long SendBufCnt;
-	unsigned long long SendAvailAddr;
-	unsigned long long TxIntMemBase;
-	unsigned long long TxIntMemSize;
-	unsigned long long SendDmaBase;
-	unsigned long long SendDmaLenGen;
-	unsigned long long SendDmaTail;
-	unsigned long long SendDmaHead;
-	unsigned long long SendDmaHeadAddr;
-	unsigned long long SendDmaBufMask0;
-	unsigned long long SendDmaBufMask1;
-	unsigned long long SendDmaBufMask2;
-	unsigned long long SendDmaStatus;
-	unsigned long long SendBufferError;
-	unsigned long long SendBufferErrorCONT1;
-	unsigned long long SendBufErr2; /* was Reserved6SBE[0/6] */
-	unsigned long long Reserved6L[2];
-	unsigned long long AvailUpdCount;
-	unsigned long long RcvHdrAddr0;
-	unsigned long long RcvHdrAddrs[16]; /* Why enumerate? */
-	unsigned long long Reserved7hdtl; /* Align next to 300 */
-	unsigned long long RcvHdrTailAddr0; /* 300, like others */
-	unsigned long long RcvHdrTailAddrs[16];
-	unsigned long long Reserved9SW[7]; /* was [8]; we have 17 ports */
-	unsigned long long IbsdEpbAccCtl; /* IB Serdes EPB access control */
-	unsigned long long IbsdEpbTransReg; /* IB Serdes EPB Transaction */
-	unsigned long long Reserved10sds; /* was SerdesStatus on */
-	unsigned long long XGXSConfig;
-	unsigned long long IBSerDesCtrl; /* Was IBPLLCfg on Monty */
-	unsigned long long EEPCtlStat; /* for "boot" EEPROM/FLASH */
-	unsigned long long EEPAddrCmd;
-	unsigned long long EEPData;
-	unsigned long long PcieEpbAccCtl;
-	unsigned long long PcieEpbTransCtl;
-	unsigned long long EfuseCtl; /* E-Fuse control */
-	unsigned long long EfuseData[4];
-	unsigned long long ProcMon;
-	/* this chip moves following two from previous 200, 208 */
-	unsigned long long PCIeRBufTestReg0;
-	unsigned long long PCIeRBufTestReg1;
-	/* added for this chip */
-	unsigned long long PCIeRBufTestReg2;
-	unsigned long long PCIeRBufTestReg3;
-	/* added for this chip, debug only */
-	unsigned long long SPC_JTAG_ACCESS_REG;
-	unsigned long long LAControlReg;
-	unsigned long long GPIODebugSelReg;
-	unsigned long long DebugPortValueReg;
-	/* added for this chip, DMA */
-	unsigned long long SendDmaBufUsed[3];
-	unsigned long long SendDmaReqTagUsed;
-	/*
-	 * added for this chip, EFUSE: note that these program 64-bit
-	 * words 2 and 3 */
-	unsigned long long efuse_pgm_data[2];
-	unsigned long long Reserved11LAalign[10]; /* Skip 4B0..4F8 */
-	/* we have 30 regs for DDS and RXEQ in IB SERDES */
-	unsigned long long SerDesDDSRXEQ[30];
-	unsigned long long Reserved12LAalign[2]; /* Skip 5F0, 5F8 */
-	/* added for LA debug support */
-	unsigned long long LAMemory[32];
-};
-
-struct _infinipath_do_not_use_counters {
-	__u64 LBIntCnt;
-	__u64 LBFlowStallCnt;
-	__u64 TxSDmaDescCnt;	/* was Reserved1 */
-	__u64 TxUnsupVLErrCnt;
-	__u64 TxDataPktCnt;
-	__u64 TxFlowPktCnt;
-	__u64 TxDwordCnt;
-	__u64 TxLenErrCnt;
-	__u64 TxMaxMinLenErrCnt;
-	__u64 TxUnderrunCnt;
-	__u64 TxFlowStallCnt;
-	__u64 TxDroppedPktCnt;
-	__u64 RxDroppedPktCnt;
-	__u64 RxDataPktCnt;
-	__u64 RxFlowPktCnt;
-	__u64 RxDwordCnt;
-	__u64 RxLenErrCnt;
-	__u64 RxMaxMinLenErrCnt;
-	__u64 RxICRCErrCnt;
-	__u64 RxVCRCErrCnt;
-	__u64 RxFlowCtrlErrCnt;
-	__u64 RxBadFormatCnt;
-	__u64 RxLinkProblemCnt;
-	__u64 RxEBPCnt;
-	__u64 RxLPCRCErrCnt;
-	__u64 RxBufOvflCnt;
-	__u64 RxTIDFullErrCnt;
-	__u64 RxTIDValidErrCnt;
-	__u64 RxPKeyMismatchCnt;
-	__u64 RxP0HdrEgrOvflCnt;
-	__u64 RxP1HdrEgrOvflCnt;
-	__u64 RxP2HdrEgrOvflCnt;
-	__u64 RxP3HdrEgrOvflCnt;
-	__u64 RxP4HdrEgrOvflCnt;
-	__u64 RxP5HdrEgrOvflCnt;
-	__u64 RxP6HdrEgrOvflCnt;
-	__u64 RxP7HdrEgrOvflCnt;
-	__u64 RxP8HdrEgrOvflCnt;
-	__u64 RxP9HdrEgrOvflCnt;	/* was Reserved6 */
-	__u64 RxP10HdrEgrOvflCnt;	/* was Reserved7 */
-	__u64 RxP11HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 RxP12HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 RxP13HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 RxP14HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 RxP15HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 RxP16HdrEgrOvflCnt;	/* new for IBA7220 */
-	__u64 IBStatusChangeCnt;
-	__u64 IBLinkErrRecoveryCnt;
-	__u64 IBLinkDownedCnt;
-	__u64 IBSymbolErrCnt;
-	/* The following are new for IBA7220 */
-	__u64 RxVL15DroppedPktCnt;
-	__u64 RxOtherLocalPhyErrCnt;
-	__u64 PcieRetryBufDiagQwordCnt;
-	__u64 ExcessBufferOvflCnt;
-	__u64 LocalLinkIntegrityErrCnt;
-	__u64 RxVlErrCnt;
-	__u64 RxDlidFltrCnt;
-	__u64 Reserved8[7];
-	__u64 PSStat;
-	__u64 PSStart;
-	__u64 PSInterval;
-	__u64 PSRcvDataCount;
-	__u64 PSRcvPktsCount;
-	__u64 PSXmitDataCount;
-	__u64 PSXmitPktsCount;
-	__u64 PSXmitWaitCount;
-};
-
-#define IPATH_KREG_OFFSET(field) (offsetof( \
-	struct _infinipath_do_not_use_kernel_regs, field) / sizeof(u64))
-#define IPATH_CREG_OFFSET(field) (offsetof( \
-	struct _infinipath_do_not_use_counters, field) / sizeof(u64))
-
-static const struct ipath_kregs ipath_7220_kregs = {
-	.kr_control = IPATH_KREG_OFFSET(Control),
-	.kr_counterregbase = IPATH_KREG_OFFSET(CounterRegBase),
-	.kr_debugportselect = IPATH_KREG_OFFSET(DebugPortSelect),
-	.kr_errorclear = IPATH_KREG_OFFSET(ErrorClear),
-	.kr_errormask = IPATH_KREG_OFFSET(ErrorMask),
-	.kr_errorstatus = IPATH_KREG_OFFSET(ErrorStatus),
-	.kr_extctrl = IPATH_KREG_OFFSET(ExtCtrl),
-	.kr_extstatus = IPATH_KREG_OFFSET(ExtStatus),
-	.kr_gpio_clear = IPATH_KREG_OFFSET(GPIOClear),
-	.kr_gpio_mask = IPATH_KREG_OFFSET(GPIOMask),
-	.kr_gpio_out = IPATH_KREG_OFFSET(GPIOOut),
-	.kr_gpio_status = IPATH_KREG_OFFSET(GPIOStatus),
-	.kr_hwdiagctrl = IPATH_KREG_OFFSET(HwDiagCtrl),
-	.kr_hwerrclear = IPATH_KREG_OFFSET(HwErrClear),
-	.kr_hwerrmask = IPATH_KREG_OFFSET(HwErrMask),
-	.kr_hwerrstatus = IPATH_KREG_OFFSET(HwErrStatus),
-	.kr_ibcctrl = IPATH_KREG_OFFSET(IBCCtrl),
-	.kr_ibcstatus = IPATH_KREG_OFFSET(IBCStatus),
-	.kr_intblocked = IPATH_KREG_OFFSET(IntBlocked),
-	.kr_intclear = IPATH_KREG_OFFSET(IntClear),
-	.kr_intmask = IPATH_KREG_OFFSET(IntMask),
-	.kr_intstatus = IPATH_KREG_OFFSET(IntStatus),
-	.kr_mdio = IPATH_KREG_OFFSET(MDIO),
-	.kr_pagealign = IPATH_KREG_OFFSET(PageAlign),
-	.kr_partitionkey = IPATH_KREG_OFFSET(RcvPartitionKey),
-	.kr_portcnt = IPATH_KREG_OFFSET(PortCnt),
-	.kr_rcvbthqp = IPATH_KREG_OFFSET(RcvBTHQP),
-	.kr_rcvbufbase = IPATH_KREG_OFFSET(RcvBufBase),
-	.kr_rcvbufsize = IPATH_KREG_OFFSET(RcvBufSize),
-	.kr_rcvctrl = IPATH_KREG_OFFSET(RcvCtrl),
-	.kr_rcvegrbase = IPATH_KREG_OFFSET(RcvEgrBase),
-	.kr_rcvegrcnt = IPATH_KREG_OFFSET(RcvEgrCnt),
-	.kr_rcvhdrcnt = IPATH_KREG_OFFSET(RcvHdrCnt),
-	.kr_rcvhdrentsize = IPATH_KREG_OFFSET(RcvHdrEntSize),
-	.kr_rcvhdrsize = IPATH_KREG_OFFSET(RcvHdrSize),
-	.kr_rcvintmembase = IPATH_KREG_OFFSET(RxIntMemBase),
-	.kr_rcvintmemsize = IPATH_KREG_OFFSET(RxIntMemSize),
-	.kr_rcvtidbase = IPATH_KREG_OFFSET(RcvTIDBase),
-	.kr_rcvtidcnt = IPATH_KREG_OFFSET(RcvTIDCnt),
-	.kr_revision = IPATH_KREG_OFFSET(Revision),
-	.kr_scratch = IPATH_KREG_OFFSET(Scratch),
-	.kr_sendbuffererror = IPATH_KREG_OFFSET(SendBufferError),
-	.kr_sendctrl = IPATH_KREG_OFFSET(SendCtrl),
-	.kr_sendpioavailaddr = IPATH_KREG_OFFSET(SendAvailAddr),
-	.kr_sendpiobufbase = IPATH_KREG_OFFSET(SendBufBase),
-	.kr_sendpiobufcnt = IPATH_KREG_OFFSET(SendBufCnt),
-	.kr_sendpiosize = IPATH_KREG_OFFSET(SendBufSize),
-	.kr_sendregbase = IPATH_KREG_OFFSET(SendRegBase),
-	.kr_txintmembase = IPATH_KREG_OFFSET(TxIntMemBase),
-	.kr_txintmemsize = IPATH_KREG_OFFSET(TxIntMemSize),
-	.kr_userregbase = IPATH_KREG_OFFSET(UserRegBase),
-
-	.kr_xgxsconfig = IPATH_KREG_OFFSET(XGXSConfig),
-
-	/* send dma related regs */
-	.kr_senddmabase = IPATH_KREG_OFFSET(SendDmaBase),
-	.kr_senddmalengen = IPATH_KREG_OFFSET(SendDmaLenGen),
-	.kr_senddmatail = IPATH_KREG_OFFSET(SendDmaTail),
-	.kr_senddmahead = IPATH_KREG_OFFSET(SendDmaHead),
-	.kr_senddmaheadaddr = IPATH_KREG_OFFSET(SendDmaHeadAddr),
-	.kr_senddmabufmask0 = IPATH_KREG_OFFSET(SendDmaBufMask0),
-	.kr_senddmabufmask1 = IPATH_KREG_OFFSET(SendDmaBufMask1),
-	.kr_senddmabufmask2 = IPATH_KREG_OFFSET(SendDmaBufMask2),
-	.kr_senddmastatus = IPATH_KREG_OFFSET(SendDmaStatus),
-
-	/* SerDes related regs */
-	.kr_ibserdesctrl = IPATH_KREG_OFFSET(IBSerDesCtrl),
-	.kr_ib_epbacc = IPATH_KREG_OFFSET(IbsdEpbAccCtl),
-	.kr_ib_epbtrans = IPATH_KREG_OFFSET(IbsdEpbTransReg),
-	.kr_pcie_epbacc = IPATH_KREG_OFFSET(PcieEpbAccCtl),
-	.kr_pcie_epbtrans = IPATH_KREG_OFFSET(PcieEpbTransCtl),
-	.kr_ib_ddsrxeq = IPATH_KREG_OFFSET(SerDesDDSRXEQ),
-
-	/*
-	 * These should not be used directly via ipath_read_kreg64(),
-	 * use them with ipath_read_kreg64_port()
-	 */
-	.kr_rcvhdraddr = IPATH_KREG_OFFSET(RcvHdrAddr0),
-	.kr_rcvhdrtailaddr = IPATH_KREG_OFFSET(RcvHdrTailAddr0),
-
-	/*
-	 * The rcvpktled register controls one of the debug port signals, so
-	 * a packet activity LED can be connected to it.
-	 */
-	.kr_rcvpktledcnt = IPATH_KREG_OFFSET(RcvPktLEDCnt),
-	.kr_pcierbuftestreg0 = IPATH_KREG_OFFSET(PCIeRBufTestReg0),
-	.kr_pcierbuftestreg1 = IPATH_KREG_OFFSET(PCIeRBufTestReg1),
-
-	.kr_hrtbt_guid = IPATH_KREG_OFFSET(HRTBT_GUID),
-	.kr_ibcddrctrl = IPATH_KREG_OFFSET(IBCDDRCtrl),
-	.kr_ibcddrstatus = IPATH_KREG_OFFSET(IBCDDRStatus),
-	.kr_jintreload = IPATH_KREG_OFFSET(JIntReload)
-};
-
-static const struct ipath_cregs ipath_7220_cregs = {
-	.cr_badformatcnt = IPATH_CREG_OFFSET(RxBadFormatCnt),
-	.cr_erricrccnt = IPATH_CREG_OFFSET(RxICRCErrCnt),
-	.cr_errlinkcnt = IPATH_CREG_OFFSET(RxLinkProblemCnt),
-	.cr_errlpcrccnt = IPATH_CREG_OFFSET(RxLPCRCErrCnt),
-	.cr_errpkey = IPATH_CREG_OFFSET(RxPKeyMismatchCnt),
-	.cr_errrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowCtrlErrCnt),
-	.cr_err_rlencnt = IPATH_CREG_OFFSET(RxLenErrCnt),
-	.cr_errslencnt = IPATH_CREG_OFFSET(TxLenErrCnt),
-	.cr_errtidfull = IPATH_CREG_OFFSET(RxTIDFullErrCnt),
-	.cr_errtidvalid = IPATH_CREG_OFFSET(RxTIDValidErrCnt),
-	.cr_errvcrccnt = IPATH_CREG_OFFSET(RxVCRCErrCnt),
-	.cr_ibstatuschange = IPATH_CREG_OFFSET(IBStatusChangeCnt),
-	.cr_intcnt = IPATH_CREG_OFFSET(LBIntCnt),
-	.cr_invalidrlencnt = IPATH_CREG_OFFSET(RxMaxMinLenErrCnt),
-	.cr_invalidslencnt = IPATH_CREG_OFFSET(TxMaxMinLenErrCnt),
-	.cr_lbflowstallcnt = IPATH_CREG_OFFSET(LBFlowStallCnt),
-	.cr_pktrcvcnt = IPATH_CREG_OFFSET(RxDataPktCnt),
-	.cr_pktrcvflowctrlcnt = IPATH_CREG_OFFSET(RxFlowPktCnt),
-	.cr_pktsendcnt = IPATH_CREG_OFFSET(TxDataPktCnt),
-	.cr_pktsendflowcnt = IPATH_CREG_OFFSET(TxFlowPktCnt),
-	.cr_portovflcnt = IPATH_CREG_OFFSET(RxP0HdrEgrOvflCnt),
-	.cr_rcvebpcnt = IPATH_CREG_OFFSET(RxEBPCnt),
-	.cr_rcvovflcnt = IPATH_CREG_OFFSET(RxBufOvflCnt),
-	.cr_senddropped = IPATH_CREG_OFFSET(TxDroppedPktCnt),
-	.cr_sendstallcnt = IPATH_CREG_OFFSET(TxFlowStallCnt),
-	.cr_sendunderruncnt = IPATH_CREG_OFFSET(TxUnderrunCnt),
-	.cr_wordrcvcnt = IPATH_CREG_OFFSET(RxDwordCnt),
-	.cr_wordsendcnt = IPATH_CREG_OFFSET(TxDwordCnt),
-	.cr_unsupvlcnt = IPATH_CREG_OFFSET(TxUnsupVLErrCnt),
-	.cr_rxdroppktcnt = IPATH_CREG_OFFSET(RxDroppedPktCnt),
-	.cr_iblinkerrrecovcnt = IPATH_CREG_OFFSET(IBLinkErrRecoveryCnt),
-	.cr_iblinkdowncnt = IPATH_CREG_OFFSET(IBLinkDownedCnt),
-	.cr_ibsymbolerrcnt = IPATH_CREG_OFFSET(IBSymbolErrCnt),
-	.cr_vl15droppedpktcnt = IPATH_CREG_OFFSET(RxVL15DroppedPktCnt),
-	.cr_rxotherlocalphyerrcnt =
-		IPATH_CREG_OFFSET(RxOtherLocalPhyErrCnt),
-	.cr_excessbufferovflcnt = IPATH_CREG_OFFSET(ExcessBufferOvflCnt),
-	.cr_locallinkintegrityerrcnt =
-		IPATH_CREG_OFFSET(LocalLinkIntegrityErrCnt),
-	.cr_rxvlerrcnt = IPATH_CREG_OFFSET(RxVlErrCnt),
-	.cr_rxdlidfltrcnt = IPATH_CREG_OFFSET(RxDlidFltrCnt),
-	.cr_psstat = IPATH_CREG_OFFSET(PSStat),
-	.cr_psstart = IPATH_CREG_OFFSET(PSStart),
-	.cr_psinterval = IPATH_CREG_OFFSET(PSInterval),
-	.cr_psrcvdatacount = IPATH_CREG_OFFSET(PSRcvDataCount),
-	.cr_psrcvpktscount = IPATH_CREG_OFFSET(PSRcvPktsCount),
-	.cr_psxmitdatacount = IPATH_CREG_OFFSET(PSXmitDataCount),
-	.cr_psxmitpktscount = IPATH_CREG_OFFSET(PSXmitPktsCount),
-	.cr_psxmitwaitcount = IPATH_CREG_OFFSET(PSXmitWaitCount),
-};
-
-/* kr_control bits */
-#define INFINIPATH_C_RESET (1U<<7)
-
-/* kr_intstatus, kr_intclear, kr_intmask bits */
-#define INFINIPATH_I_RCVURG_MASK ((1ULL<<17)-1)
-#define INFINIPATH_I_RCVURG_SHIFT 32
-#define INFINIPATH_I_RCVAVAIL_MASK ((1ULL<<17)-1)
-#define INFINIPATH_I_RCVAVAIL_SHIFT 0
-#define INFINIPATH_I_SERDESTRIMDONE (1ULL<<27)
-
-/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
-#define INFINIPATH_HWE_PCIEMEMPARITYERR_MASK  0x00000000000000ffULL
-#define INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT 0
-#define INFINIPATH_HWE_PCIEPOISONEDTLP      0x0000000010000000ULL
-#define INFINIPATH_HWE_PCIECPLTIMEOUT       0x0000000020000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYXTLH    0x0000000040000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYXADM    0x0000000080000000ULL
-#define INFINIPATH_HWE_PCIEBUSPARITYRADM    0x0000000100000000ULL
-#define INFINIPATH_HWE_COREPLL_FBSLIP       0x0080000000000000ULL
-#define INFINIPATH_HWE_COREPLL_RFSLIP       0x0100000000000000ULL
-#define INFINIPATH_HWE_PCIE1PLLFAILED       0x0400000000000000ULL
-#define INFINIPATH_HWE_PCIE0PLLFAILED       0x0800000000000000ULL
-#define INFINIPATH_HWE_SERDESPLLFAILED      0x1000000000000000ULL
-/* specific to this chip */
-#define INFINIPATH_HWE_PCIECPLDATAQUEUEERR         0x0000000000000040ULL
-#define INFINIPATH_HWE_PCIECPLHDRQUEUEERR          0x0000000000000080ULL
-#define INFINIPATH_HWE_SDMAMEMREADERR              0x0000000010000000ULL
-#define INFINIPATH_HWE_CLK_UC_PLLNOTLOCKED	   0x2000000000000000ULL
-#define INFINIPATH_HWE_PCIESERDESQ0PCLKNOTDETECT   0x0100000000000000ULL
-#define INFINIPATH_HWE_PCIESERDESQ1PCLKNOTDETECT   0x0200000000000000ULL
-#define INFINIPATH_HWE_PCIESERDESQ2PCLKNOTDETECT   0x0400000000000000ULL
-#define INFINIPATH_HWE_PCIESERDESQ3PCLKNOTDETECT   0x0800000000000000ULL
-#define INFINIPATH_HWE_DDSRXEQMEMORYPARITYERR	   0x0000008000000000ULL
-#define INFINIPATH_HWE_IB_UC_MEMORYPARITYERR	   0x0000004000000000ULL
-#define INFINIPATH_HWE_PCIE_UC_OCT0MEMORYPARITYERR 0x0000001000000000ULL
-#define INFINIPATH_HWE_PCIE_UC_OCT1MEMORYPARITYERR 0x0000002000000000ULL
-
-#define IBA7220_IBCS_LINKTRAININGSTATE_MASK 0x1F
-#define IBA7220_IBCS_LINKSTATE_SHIFT 5
-#define IBA7220_IBCS_LINKSPEED_SHIFT 8
-#define IBA7220_IBCS_LINKWIDTH_SHIFT 9
-
-#define IBA7220_IBCC_LINKINITCMD_MASK 0x7ULL
-#define IBA7220_IBCC_LINKCMD_SHIFT 19
-#define IBA7220_IBCC_MAXPKTLEN_SHIFT 21
-
-/* kr_ibcddrctrl bits */
-#define IBA7220_IBC_DLIDLMC_MASK	0xFFFFFFFFUL
-#define IBA7220_IBC_DLIDLMC_SHIFT	32
-#define IBA7220_IBC_HRTBT_MASK	3
-#define IBA7220_IBC_HRTBT_SHIFT	16
-#define IBA7220_IBC_HRTBT_ENB	0x10000UL
-#define IBA7220_IBC_LANE_REV_SUPPORTED (1<<8)
-#define IBA7220_IBC_LREV_MASK	1
-#define IBA7220_IBC_LREV_SHIFT	8
-#define IBA7220_IBC_RXPOL_MASK	1
-#define IBA7220_IBC_RXPOL_SHIFT	7
-#define IBA7220_IBC_WIDTH_SHIFT	5
-#define IBA7220_IBC_WIDTH_MASK	0x3
-#define IBA7220_IBC_WIDTH_1X_ONLY	(0<<IBA7220_IBC_WIDTH_SHIFT)
-#define IBA7220_IBC_WIDTH_4X_ONLY	(1<<IBA7220_IBC_WIDTH_SHIFT)
-#define IBA7220_IBC_WIDTH_AUTONEG	(2<<IBA7220_IBC_WIDTH_SHIFT)
-#define IBA7220_IBC_SPEED_AUTONEG	(1<<1)
-#define IBA7220_IBC_SPEED_SDR		(1<<2)
-#define IBA7220_IBC_SPEED_DDR		(1<<3)
-#define IBA7220_IBC_SPEED_AUTONEG_MASK  (0x7<<1)
-#define IBA7220_IBC_IBTA_1_2_MASK	(1)
-
-/* kr_ibcddrstatus */
-/* link latency shift is 0, don't bother defining */
-#define IBA7220_DDRSTAT_LINKLAT_MASK    0x3ffffff
-
-/* kr_extstatus bits */
-#define INFINIPATH_EXTS_FREQSEL 0x2
-#define INFINIPATH_EXTS_SERDESSEL 0x4
-#define INFINIPATH_EXTS_MEMBIST_ENDTEST     0x0000000000004000
-#define INFINIPATH_EXTS_MEMBIST_DISABLED    0x0000000000008000
-
-/* kr_xgxsconfig bits */
-#define INFINIPATH_XGXS_RESET          0x5ULL
-#define INFINIPATH_XGXS_FC_SAFE        (1ULL<<63)
-
-/* kr_rcvpktledcnt */
-#define IBA7220_LEDBLINK_ON_SHIFT 32 /* 4ns period on after packet */
-#define IBA7220_LEDBLINK_OFF_SHIFT 0 /* 4ns period off before next on */
-
-#define _IPATH_GPIO_SDA_NUM 1
-#define _IPATH_GPIO_SCL_NUM 0
-
-#define IPATH_GPIO_SDA (1ULL << \
-	(_IPATH_GPIO_SDA_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
-#define IPATH_GPIO_SCL (1ULL << \
-	(_IPATH_GPIO_SCL_NUM+INFINIPATH_EXTC_GPIOOE_SHIFT))
-
-#define IBA7220_R_INTRAVAIL_SHIFT 17
-#define IBA7220_R_TAILUPD_SHIFT 35
-#define IBA7220_R_PORTCFG_SHIFT 36
-
-#define INFINIPATH_JINT_PACKETSHIFT 16
-#define INFINIPATH_JINT_DEFAULT_IDLE_TICKS  0
-#define INFINIPATH_JINT_DEFAULT_MAX_PACKETS 0
-
-#define IBA7220_HDRHEAD_PKTINT_SHIFT 32 /* interrupt cnt in upper 32 bits */
-
-/*
- * the size bits give us 2^N, in KB units.  0 marks as invalid,
- * and 7 is reserved.  We currently use only 2KB and 4KB
- */
-#define IBA7220_TID_SZ_SHIFT 37 /* shift to 3bit size selector */
-#define IBA7220_TID_SZ_2K (1UL<<IBA7220_TID_SZ_SHIFT) /* 2KB */
-#define IBA7220_TID_SZ_4K (2UL<<IBA7220_TID_SZ_SHIFT) /* 4KB */
-#define IBA7220_TID_PA_SHIFT 11U /* TID addr in chip stored w/o low bits */
-
-#define IPATH_AUTONEG_TRIES 5 /* sequential retries to negotiate DDR */
-
-static char int_type[16] = "auto";
-module_param_string(interrupt_type, int_type, sizeof(int_type), 0444);
-MODULE_PARM_DESC(int_type, " interrupt_type=auto|force_msi|force_intx\n");
-
-/* packet rate matching delay; chip has support */
-static u8 rate_to_delay[2][2] = {
-	/* 1x, 4x */
-	{   8, 2 }, /* SDR */
-	{   4, 1 }  /* DDR */
-};
-
-/* 7220 specific hardware errors... */
-static const struct ipath_hwerror_msgs ipath_7220_hwerror_msgs[] = {
-	INFINIPATH_HWE_MSG(PCIEPOISONEDTLP, "PCIe Poisoned TLP"),
-	INFINIPATH_HWE_MSG(PCIECPLTIMEOUT, "PCIe completion timeout"),
-	/*
-	 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
-	 * parity or memory parity error failures, because most likely we
-	 * won't be able to talk to the core of the chip.  Nonetheless, we
-	 * might see them, if they are in parts of the PCIe core that aren't
-	 * essential.
-	 */
-	INFINIPATH_HWE_MSG(PCIE1PLLFAILED, "PCIePLL1"),
-	INFINIPATH_HWE_MSG(PCIE0PLLFAILED, "PCIePLL0"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYXTLH, "PCIe XTLH core parity"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYXADM, "PCIe ADM TX core parity"),
-	INFINIPATH_HWE_MSG(PCIEBUSPARITYRADM, "PCIe ADM RX core parity"),
-	INFINIPATH_HWE_MSG(RXDSYNCMEMPARITYERR, "Rx Dsync"),
-	INFINIPATH_HWE_MSG(SERDESPLLFAILED, "SerDes PLL"),
-	INFINIPATH_HWE_MSG(PCIECPLDATAQUEUEERR, "PCIe cpl header queue"),
-	INFINIPATH_HWE_MSG(PCIECPLHDRQUEUEERR, "PCIe cpl data queue"),
-	INFINIPATH_HWE_MSG(SDMAMEMREADERR, "Send DMA memory read"),
-	INFINIPATH_HWE_MSG(CLK_UC_PLLNOTLOCKED, "uC PLL clock not locked"),
-	INFINIPATH_HWE_MSG(PCIESERDESQ0PCLKNOTDETECT,
-		"PCIe serdes Q0 no clock"),
-	INFINIPATH_HWE_MSG(PCIESERDESQ1PCLKNOTDETECT,
-		"PCIe serdes Q1 no clock"),
-	INFINIPATH_HWE_MSG(PCIESERDESQ2PCLKNOTDETECT,
-		"PCIe serdes Q2 no clock"),
-	INFINIPATH_HWE_MSG(PCIESERDESQ3PCLKNOTDETECT,
-		"PCIe serdes Q3 no clock"),
-	INFINIPATH_HWE_MSG(DDSRXEQMEMORYPARITYERR,
-		"DDS RXEQ memory parity"),
-	INFINIPATH_HWE_MSG(IB_UC_MEMORYPARITYERR, "IB uC memory parity"),
-	INFINIPATH_HWE_MSG(PCIE_UC_OCT0MEMORYPARITYERR,
-		"PCIe uC oct0 memory parity"),
-	INFINIPATH_HWE_MSG(PCIE_UC_OCT1MEMORYPARITYERR,
-		"PCIe uC oct1 memory parity"),
-};
-
-static void autoneg_work(struct work_struct *);
-
-/*
- * the offset is different for different configured port numbers, since
- * port0 is fixed in size, but others can vary.   Make it a function to
- * make the issue more obvious.
-*/
-static inline u32 port_egrtid_idx(struct ipath_devdata *dd, unsigned port)
-{
-	 return port ? dd->ipath_p0_rcvegrcnt +
-		 (port-1) * dd->ipath_rcvegrcnt : 0;
-}
-
-static void ipath_7220_txe_recover(struct ipath_devdata *dd)
-{
-	++ipath_stats.sps_txeparity;
-
-	dev_info(&dd->pcidev->dev,
-		"Recovering from TXE PIO parity error\n");
-	ipath_disarm_senderrbufs(dd);
-}
-
-
-/**
- * ipath_7220_handle_hwerrors - display hardware errors.
- * @dd: the infinipath device
- * @msg: the output buffer
- * @msgl: the size of the output buffer
- *
- * Use same msg buffer as regular errors to avoid excessive stack
- * use.  Most hardware errors are catastrophic, but for right now,
- * we'll print them and continue.  We reuse the same message buffer as
- * ipath_handle_errors() to avoid excessive stack usage.
- */
-static void ipath_7220_handle_hwerrors(struct ipath_devdata *dd, char *msg,
-				       size_t msgl)
-{
-	ipath_err_t hwerrs;
-	u32 bits, ctrl;
-	int isfatal = 0;
-	char bitsmsg[64];
-	int log_idx;
-
-	hwerrs = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
-	if (!hwerrs) {
-		/*
-		 * better than printing cofusing messages
-		 * This seems to be related to clearing the crc error, or
-		 * the pll error during init.
-		 */
-		ipath_cdbg(VERBOSE, "Called but no hardware errors set\n");
-		goto bail;
-	} else if (hwerrs == ~0ULL) {
-		ipath_dev_err(dd, "Read of hardware error status failed "
-			      "(all bits set); ignoring\n");
-		goto bail;
-	}
-	ipath_stats.sps_hwerrs++;
-
-	/*
-	 * Always clear the error status register, except MEMBISTFAIL,
-	 * regardless of whether we continue or stop using the chip.
-	 * We want that set so we know it failed, even across driver reload.
-	 * We'll still ignore it in the hwerrmask.  We do this partly for
-	 * diagnostics, but also for support.
-	 */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
-			 hwerrs&~INFINIPATH_HWE_MEMBISTFAILED);
-
-	hwerrs &= dd->ipath_hwerrmask;
-
-	/* We log some errors to EEPROM, check if we have any of those. */
-	for (log_idx = 0; log_idx < IPATH_EEP_LOG_CNT; ++log_idx)
-		if (hwerrs & dd->ipath_eep_st_masks[log_idx].hwerrs_to_log)
-			ipath_inc_eeprom_err(dd, log_idx, 1);
-	/*
-	 * Make sure we get this much out, unless told to be quiet,
-	 * or it's occurred within the last 5 seconds.
-	 */
-	if ((hwerrs & ~(dd->ipath_lasthwerror |
-			((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
-			  INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
-			 << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT))) ||
-	    (ipath_debug & __IPATH_VERBDBG))
-		dev_info(&dd->pcidev->dev, "Hardware error: hwerr=0x%llx "
-			 "(cleared)\n", (unsigned long long) hwerrs);
-	dd->ipath_lasthwerror |= hwerrs;
-
-	if (hwerrs & ~dd->ipath_hwe_bitsextant)
-		ipath_dev_err(dd, "hwerror interrupt with unknown errors "
-			      "%llx set\n", (unsigned long long)
-			      (hwerrs & ~dd->ipath_hwe_bitsextant));
-
-	if (hwerrs & INFINIPATH_HWE_IB_UC_MEMORYPARITYERR)
-		ipath_sd7220_clr_ibpar(dd);
-
-	ctrl = ipath_read_kreg32(dd, dd->ipath_kregs->kr_control);
-	if ((ctrl & INFINIPATH_C_FREEZEMODE) && !ipath_diag_inuse) {
-		/*
-		 * Parity errors in send memory are recoverable by h/w
-		 * just do housekeeping, exit freeze mode and continue.
-		 */
-		if (hwerrs & ((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
-			       INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
-			      << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT)) {
-			ipath_7220_txe_recover(dd);
-			hwerrs &= ~((INFINIPATH_HWE_TXEMEMPARITYERR_PIOBUF |
-				     INFINIPATH_HWE_TXEMEMPARITYERR_PIOPBC)
-				    << INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT);
-		}
-		if (hwerrs) {
-			/*
-			 * If any set that we aren't ignoring only make the
-			 * complaint once, in case it's stuck or recurring,
-			 * and we get here multiple times
-			 * Force link down, so switch knows, and
-			 * LEDs are turned off.
-			 */
-			if (dd->ipath_flags & IPATH_INITTED) {
-				ipath_set_linkstate(dd, IPATH_IB_LINKDOWN);
-				ipath_setup_7220_setextled(dd,
-					INFINIPATH_IBCS_L_STATE_DOWN,
-					INFINIPATH_IBCS_LT_STATE_DISABLED);
-				ipath_dev_err(dd, "Fatal Hardware Error "
-					      "(freeze mode), no longer"
-					      " usable, SN %.16s\n",
-						  dd->ipath_serial);
-				isfatal = 1;
-			}
-			/*
-			 * Mark as having had an error for driver, and also
-			 * for /sys and status word mapped to user programs.
-			 * This marks unit as not usable, until reset.
-			 */
-			*dd->ipath_statusp &= ~IPATH_STATUS_IB_READY;
-			*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
-			dd->ipath_flags &= ~IPATH_INITTED;
-		} else {
-			ipath_dbg("Clearing freezemode on ignored or "
-				"recovered hardware error\n");
-			ipath_clear_freeze(dd);
-		}
-	}
-
-	*msg = '\0';
-
-	if (hwerrs & INFINIPATH_HWE_MEMBISTFAILED) {
-		strlcat(msg, "[Memory BIST test failed, "
-			"InfiniPath hardware unusable]", msgl);
-		/* ignore from now on, so disable until driver reloaded */
-		*dd->ipath_statusp |= IPATH_STATUS_HWERROR;
-		dd->ipath_hwerrmask &= ~INFINIPATH_HWE_MEMBISTFAILED;
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	ipath_format_hwerrors(hwerrs,
-			      ipath_7220_hwerror_msgs,
-			      ARRAY_SIZE(ipath_7220_hwerror_msgs),
-			      msg, msgl);
-
-	if (hwerrs & (INFINIPATH_HWE_PCIEMEMPARITYERR_MASK
-		      << INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT)) {
-		bits = (u32) ((hwerrs >>
-			       INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) &
-			      INFINIPATH_HWE_PCIEMEMPARITYERR_MASK);
-		snprintf(bitsmsg, sizeof bitsmsg,
-			 "[PCIe Mem Parity Errs %x] ", bits);
-		strlcat(msg, bitsmsg, msgl);
-	}
-
-#define _IPATH_PLL_FAIL (INFINIPATH_HWE_COREPLL_FBSLIP |	\
-			 INFINIPATH_HWE_COREPLL_RFSLIP)
-
-	if (hwerrs & _IPATH_PLL_FAIL) {
-		snprintf(bitsmsg, sizeof bitsmsg,
-			 "[PLL failed (%llx), InfiniPath hardware unusable]",
-			 (unsigned long long) hwerrs & _IPATH_PLL_FAIL);
-		strlcat(msg, bitsmsg, msgl);
-		/* ignore from now on, so disable until driver reloaded */
-		dd->ipath_hwerrmask &= ~(hwerrs & _IPATH_PLL_FAIL);
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	if (hwerrs & INFINIPATH_HWE_SERDESPLLFAILED) {
-		/*
-		 * If it occurs, it is left masked since the eternal
-		 * interface is unused.
-		 */
-		dd->ipath_hwerrmask &= ~INFINIPATH_HWE_SERDESPLLFAILED;
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-				 dd->ipath_hwerrmask);
-	}
-
-	ipath_dev_err(dd, "%s hardware error\n", msg);
-	/*
-	 * For /sys status file. if no trailing } is copied, we'll
-	 * know it was truncated.
-	 */
-	if (isfatal && !ipath_diag_inuse && dd->ipath_freezemsg)
-		snprintf(dd->ipath_freezemsg, dd->ipath_freezelen,
-			 "{%s}", msg);
-bail:;
-}
-
-/**
- * ipath_7220_boardname - fill in the board name
- * @dd: the infinipath device
- * @name: the output buffer
- * @namelen: the size of the output buffer
- *
- * info is based on the board revision register
- */
-static int ipath_7220_boardname(struct ipath_devdata *dd, char *name,
-	size_t namelen)
-{
-	char *n = NULL;
-	u8 boardrev = dd->ipath_boardrev;
-	int ret;
-
-	if (boardrev == 15) {
-		/*
-		 * Emulator sometimes comes up all-ones, rather than zero.
-		 */
-		boardrev = 0;
-		dd->ipath_boardrev = boardrev;
-	}
-	switch (boardrev) {
-	case 0:
-		n = "InfiniPath_7220_Emulation";
-		break;
-	case 1:
-		n = "InfiniPath_QLE7240";
-		break;
-	case 2:
-		n = "InfiniPath_QLE7280";
-		break;
-	case 3:
-		n = "InfiniPath_QLE7242";
-		break;
-	case 4:
-		n = "InfiniPath_QEM7240";
-		break;
-	case 5:
-		n = "InfiniPath_QMI7240";
-		break;
-	case 6:
-		n = "InfiniPath_QMI7264";
-		break;
-	case 7:
-		n = "InfiniPath_QMH7240";
-		break;
-	case 8:
-		n = "InfiniPath_QME7240";
-		break;
-	case 9:
-		n = "InfiniPath_QLE7250";
-		break;
-	case 10:
-		n = "InfiniPath_QLE7290";
-		break;
-	case 11:
-		n = "InfiniPath_QEM7250";
-		break;
-	case 12:
-		n = "InfiniPath_QLE-Bringup";
-		break;
-	default:
-		ipath_dev_err(dd,
-			      "Don't yet know about board with ID %u\n",
-			      boardrev);
-		snprintf(name, namelen, "Unknown_InfiniPath_PCIe_%u",
-			 boardrev);
-		break;
-	}
-	if (n)
-		snprintf(name, namelen, "%s", n);
-
-	if (dd->ipath_majrev != 5 || !dd->ipath_minrev ||
-		dd->ipath_minrev > 2) {
-		ipath_dev_err(dd, "Unsupported InfiniPath hardware "
-			      "revision %u.%u!\n",
-			      dd->ipath_majrev, dd->ipath_minrev);
-		ret = 1;
-	} else if (dd->ipath_minrev == 1 &&
-		!(dd->ipath_flags & IPATH_INITTED)) {
-		/* Rev1 chips are prototype. Complain at init, but allow use */
-		ipath_dev_err(dd, "Unsupported hardware "
-			      "revision %u.%u, Contact support@qlogic.com\n",
-			      dd->ipath_majrev, dd->ipath_minrev);
-		ret = 0;
-	} else
-		ret = 0;
-
-	/*
-	 * Set here not in ipath_init_*_funcs because we have to do
-	 * it after we can read chip registers.
-	 */
-	dd->ipath_ureg_align = 0x10000;  /* 64KB alignment */
-
-	return ret;
-}
-
-/**
- * ipath_7220_init_hwerrors - enable hardware errors
- * @dd: the infinipath device
- *
- * now that we have finished initializing everything that might reasonably
- * cause a hardware error, and cleared those errors bits as they occur,
- * we can enable hardware errors in the mask (potentially enabling
- * freeze mode), and enable hardware errors as errors (along with
- * everything else) in errormask
- */
-static void ipath_7220_init_hwerrors(struct ipath_devdata *dd)
-{
-	ipath_err_t val;
-	u64 extsval;
-
-	extsval = ipath_read_kreg64(dd, dd->ipath_kregs->kr_extstatus);
-
-	if (!(extsval & (INFINIPATH_EXTS_MEMBIST_ENDTEST |
-			INFINIPATH_EXTS_MEMBIST_DISABLED)))
-		ipath_dev_err(dd, "MemBIST did not complete!\n");
-	if (extsval & INFINIPATH_EXTS_MEMBIST_DISABLED)
-		dev_info(&dd->pcidev->dev, "MemBIST is disabled.\n");
-
-	val = ~0ULL;	/* barring bugs, all hwerrors become interrupts, */
-
-	if (!dd->ipath_boardrev)	/* no PLL for Emulator */
-		val &= ~INFINIPATH_HWE_SERDESPLLFAILED;
-
-	if (dd->ipath_minrev == 1)
-		val &= ~(1ULL << 42); /* TXE LaunchFIFO Parity rev1 issue */
-
-	val &= ~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR;
-	dd->ipath_hwerrmask = val;
-
-	/*
-	 * special trigger "error" is for debugging purposes. It
-	 * works around a processor/chipset problem.  The error
-	 * interrupt allows us to count occurrences, but we don't
-	 * want to pay the overhead for normal use.  Emulation only
-	 */
-	if (!dd->ipath_boardrev)
-		dd->ipath_maskederrs = INFINIPATH_E_SENDSPECIALTRIGGER;
-}
-
-/*
- * All detailed interaction with the SerDes has been moved to ipath_sd7220.c
- *
- * The portion of IBA7220-specific bringup_serdes() that actually deals with
- * registers and memory within the SerDes itself is ipath_sd7220_init().
- */
-
-/**
- * ipath_7220_bringup_serdes - bring up the serdes
- * @dd: the infinipath device
- */
-static int ipath_7220_bringup_serdes(struct ipath_devdata *dd)
-{
-	int ret = 0;
-	u64 val, prev_val, guid;
-	int was_reset;		/* Note whether uC was reset */
-
-	ipath_dbg("Trying to bringup serdes\n");
-
-	if (ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus) &
-	    INFINIPATH_HWE_SERDESPLLFAILED) {
-		ipath_dbg("At start, serdes PLL failed bit set "
-			  "in hwerrstatus, clearing and continuing\n");
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
-				 INFINIPATH_HWE_SERDESPLLFAILED);
-	}
-
-	if (!dd->ipath_ibcddrctrl) {
-		/* not on re-init after reset */
-		dd->ipath_ibcddrctrl =
-			ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcddrctrl);
-
-		if (dd->ipath_link_speed_enabled ==
-			(IPATH_IB_SDR | IPATH_IB_DDR))
-			dd->ipath_ibcddrctrl |=
-				IBA7220_IBC_SPEED_AUTONEG_MASK |
-				IBA7220_IBC_IBTA_1_2_MASK;
-		else
-			dd->ipath_ibcddrctrl |=
-				dd->ipath_link_speed_enabled == IPATH_IB_DDR
-				?  IBA7220_IBC_SPEED_DDR :
-				IBA7220_IBC_SPEED_SDR;
-		if ((dd->ipath_link_width_enabled & (IB_WIDTH_1X |
-			IB_WIDTH_4X)) == (IB_WIDTH_1X | IB_WIDTH_4X))
-			dd->ipath_ibcddrctrl |= IBA7220_IBC_WIDTH_AUTONEG;
-		else
-			dd->ipath_ibcddrctrl |=
-				dd->ipath_link_width_enabled == IB_WIDTH_4X
-				? IBA7220_IBC_WIDTH_4X_ONLY :
-				IBA7220_IBC_WIDTH_1X_ONLY;
-
-		/* always enable these on driver reload, not sticky */
-		dd->ipath_ibcddrctrl |=
-			IBA7220_IBC_RXPOL_MASK << IBA7220_IBC_RXPOL_SHIFT;
-		dd->ipath_ibcddrctrl |=
-			IBA7220_IBC_HRTBT_MASK << IBA7220_IBC_HRTBT_SHIFT;
-		/*
-		 * automatic lane reversal detection for receive
-		 * doesn't work correctly in rev 1, so disable it
-		 * on that rev, otherwise enable (disabling not
-		 * sticky across reload for >rev1)
-		 */
-		if (dd->ipath_minrev == 1)
-			dd->ipath_ibcddrctrl &=
-			~IBA7220_IBC_LANE_REV_SUPPORTED;
-		else
-			dd->ipath_ibcddrctrl |=
-				IBA7220_IBC_LANE_REV_SUPPORTED;
-	}
-
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl,
-			dd->ipath_ibcddrctrl);
-
-	ipath_write_kreg(dd, IPATH_KREG_OFFSET(IBNCModeCtrl), 0Ull);
-
-	/* IBA7220 has SERDES MPU reset in D0 of what _was_ IBPLLCfg */
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl);
-	/* remember if uC was in Reset or not, for dactrim */
-	was_reset = (val & 1);
-	ipath_cdbg(VERBOSE, "IBReset %s xgxsconfig %llx\n",
-		   was_reset ? "Asserted" : "Negated", (unsigned long long)
-		   ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig));
-
-	if (dd->ipath_boardrev) {
-		/*
-		 * Hardware is not emulator, and may have been reset. Init it.
-		 * Below will release reset, but needs to know if chip was
-		 * originally in reset, to only trim DACs on first time
-		 * after chip reset or powercycle (not driver reload)
-		 */
-		ret = ipath_sd7220_init(dd, was_reset);
-	}
-
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
-	prev_val = val;
-	val |= INFINIPATH_XGXS_FC_SAFE;
-	if (val != prev_val) {
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	}
-	if (val & INFINIPATH_XGXS_RESET)
-		val &= ~INFINIPATH_XGXS_RESET;
-	if (val != prev_val)
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-
-	ipath_cdbg(VERBOSE, "done: xgxs=%llx from %llx\n",
-		   (unsigned long long)
-		   ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig),
-		   prev_val);
-
-	guid = be64_to_cpu(dd->ipath_guid);
-
-	if (!guid) {
-		/* have to have something, so use likely unique tsc */
-		guid = get_cycles();
-		ipath_dbg("No GUID for heartbeat, faking %llx\n",
-			(unsigned long long)guid);
-	} else
-		ipath_cdbg(VERBOSE, "Wrote %llX to HRTBT_GUID\n", guid);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_hrtbt_guid, guid);
-	return ret;
-}
-
-static void ipath_7220_config_jint(struct ipath_devdata *dd,
-				   u16 idle_ticks, u16 max_packets)
-{
-
-	/*
-	 * We can request a receive interrupt for 1 or more packets
-	 * from current offset.
-	 */
-	if (idle_ticks == 0 || max_packets == 0)
-		/* interrupt after one packet if no mitigation */
-		dd->ipath_rhdrhead_intr_off =
-			1ULL << IBA7220_HDRHEAD_PKTINT_SHIFT;
-	else
-		/* Turn off RcvHdrHead interrupts if using mitigation */
-		dd->ipath_rhdrhead_intr_off = 0ULL;
-
-	/* refresh kernel RcvHdrHead registers... */
-	ipath_write_ureg(dd, ur_rcvhdrhead,
-			 dd->ipath_rhdrhead_intr_off |
-			 dd->ipath_pd[0]->port_head, 0);
-
-	dd->ipath_jint_max_packets = max_packets;
-	dd->ipath_jint_idle_ticks = idle_ticks;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_jintreload,
-			 ((u64) max_packets << INFINIPATH_JINT_PACKETSHIFT) |
-			 idle_ticks);
-}
-
-/**
- * ipath_7220_quiet_serdes - set serdes to txidle
- * @dd: the infinipath device
- * Called when driver is being unloaded
- */
-static void ipath_7220_quiet_serdes(struct ipath_devdata *dd)
-{
-	u64 val;
-	dd->ipath_flags &= ~IPATH_IB_AUTONEG_INPROG;
-	wake_up(&dd->ipath_autoneg_wait);
-	cancel_delayed_work(&dd->ipath_autoneg_work);
-	flush_scheduled_work();
-	ipath_shutdown_relock_poll(dd);
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
-	val |= INFINIPATH_XGXS_RESET;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-}
-
-static int ipath_7220_intconfig(struct ipath_devdata *dd)
-{
-	ipath_7220_config_jint(dd, dd->ipath_jint_idle_ticks,
-			       dd->ipath_jint_max_packets);
-	return 0;
-}
-
-/**
- * ipath_setup_7220_setextled - set the state of the two external LEDs
- * @dd: the infinipath device
- * @lst: the L state
- * @ltst: the LT state
- *
- * These LEDs indicate the physical and logical state of IB link.
- * For this chip (at least with recommended board pinouts), LED1
- * is Yellow (logical state) and LED2 is Green (physical state),
- *
- * Note:  We try to match the Mellanox HCA LED behavior as best
- * we can.  Green indicates physical link state is OK (something is
- * plugged in, and we can train).
- * Amber indicates the link is logically up (ACTIVE).
- * Mellanox further blinks the amber LED to indicate data packet
- * activity, but we have no hardware support for that, so it would
- * require waking up every 10-20 msecs and checking the counters
- * on the chip, and then turning the LED off if appropriate.  That's
- * visible overhead, so not something we will do.
- *
- */
-static void ipath_setup_7220_setextled(struct ipath_devdata *dd, u64 lst,
-				       u64 ltst)
-{
-	u64 extctl, ledblink = 0;
-	unsigned long flags = 0;
-
-	/* the diags use the LED to indicate diag info, so we leave
-	 * the external LED alone when the diags are running */
-	if (ipath_diag_inuse)
-		return;
-
-	/* Allow override of LED display for, e.g. Locating system in rack */
-	if (dd->ipath_led_override) {
-		ltst = (dd->ipath_led_override & IPATH_LED_PHYS)
-			? INFINIPATH_IBCS_LT_STATE_LINKUP
-			: INFINIPATH_IBCS_LT_STATE_DISABLED;
-		lst = (dd->ipath_led_override & IPATH_LED_LOG)
-			? INFINIPATH_IBCS_L_STATE_ACTIVE
-			: INFINIPATH_IBCS_L_STATE_DOWN;
-	}
-
-	spin_lock_irqsave(&dd->ipath_gpio_lock, flags);
-	extctl = dd->ipath_extctrl & ~(INFINIPATH_EXTC_LED1PRIPORT_ON |
-				       INFINIPATH_EXTC_LED2PRIPORT_ON);
-	if (ltst == INFINIPATH_IBCS_LT_STATE_LINKUP) {
-		extctl |= INFINIPATH_EXTC_LED1PRIPORT_ON;
-		/*
-		 * counts are in chip clock (4ns) periods.
-		 * This is 1/16 sec (66.6ms) on,
-		 * 3/16 sec (187.5 ms) off, with packets rcvd
-		 */
-		ledblink = ((66600*1000UL/4) << IBA7220_LEDBLINK_ON_SHIFT)
-			| ((187500*1000UL/4) << IBA7220_LEDBLINK_OFF_SHIFT);
-	}
-	if (lst == INFINIPATH_IBCS_L_STATE_ACTIVE)
-		extctl |= INFINIPATH_EXTC_LED2PRIPORT_ON;
-	dd->ipath_extctrl = extctl;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_extctrl, extctl);
-	spin_unlock_irqrestore(&dd->ipath_gpio_lock, flags);
-
-	if (ledblink) /* blink the LED on packet receive */
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvpktledcnt,
-			ledblink);
-}
-
-/*
- * Similar to pci_intx(pdev, 1), except that we make sure
- * msi is off...
- */
-static void ipath_enable_intx(struct pci_dev *pdev)
-{
-	u16 cw, new;
-	int pos;
-
-	/* first, turn on INTx */
-	pci_read_config_word(pdev, PCI_COMMAND, &cw);
-	new = cw & ~PCI_COMMAND_INTX_DISABLE;
-	if (new != cw)
-		pci_write_config_word(pdev, PCI_COMMAND, new);
-
-	/* then turn off MSI */
-	pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
-	if (pos) {
-		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
-		new = cw & ~PCI_MSI_FLAGS_ENABLE;
-		if (new != cw)
-			pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
-	}
-}
-
-static int ipath_msi_enabled(struct pci_dev *pdev)
-{
-	int pos, ret = 0;
-
-	pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
-	if (pos) {
-		u16 cw;
-
-		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
-		ret = !!(cw & PCI_MSI_FLAGS_ENABLE);
-	}
-	return ret;
-}
-
-/*
- * disable msi interrupt if enabled, and clear the flag.
- * flag is used primarily for the fallback to INTx, but
- * is also used in reinit after reset as a flag.
- */
-static void ipath_7220_nomsi(struct ipath_devdata *dd)
-{
-	dd->ipath_msi_lo = 0;
-
-	if (ipath_msi_enabled(dd->pcidev)) {
-		/*
-		 * free, but don't zero; later kernels require
-		 * it be freed before disable_msi, so the intx
-		 * setup has to request it again.
-		 */
-		 if (dd->ipath_irq)
-			free_irq(dd->ipath_irq, dd);
-		pci_disable_msi(dd->pcidev);
-	}
-}
-
-/*
- * ipath_setup_7220_cleanup - clean up any per-chip chip-specific stuff
- * @dd: the infinipath device
- *
- * Nothing but msi interrupt cleanup for now.
- *
- * This is called during driver unload.
- */
-static void ipath_setup_7220_cleanup(struct ipath_devdata *dd)
-{
-	ipath_7220_nomsi(dd);
-}
-
-
-static void ipath_7220_pcie_params(struct ipath_devdata *dd, u32 boardrev)
-{
-	u16 linkstat, minwidth, speed;
-	int pos;
-
-	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_EXP);
-	if (!pos) {
-		ipath_dev_err(dd, "Can't find PCI Express capability!\n");
-		goto bail;
-	}
-
-	pci_read_config_word(dd->pcidev, pos + PCI_EXP_LNKSTA,
-			     &linkstat);
-	/*
-	 * speed is bits 0-4, linkwidth is bits 4-8
-	 * no defines for them in headers
-	 */
-	speed = linkstat & 0xf;
-	linkstat >>= 4;
-	linkstat &= 0x1f;
-	dd->ipath_lbus_width = linkstat;
-	switch (boardrev) {
-	case 0:
-	case 2:
-	case 10:
-	case 12:
-		minwidth = 16; /* x16 capable boards */
-		break;
-	default:
-		minwidth = 8; /* x8 capable boards */
-		break;
-	}
-
-	switch (speed) {
-	case 1:
-		dd->ipath_lbus_speed = 2500; /* Gen1, 2.5GHz */
-		break;
-	case 2:
-		dd->ipath_lbus_speed = 5000; /* Gen1, 5GHz */
-		break;
-	default: /* not defined, assume gen1 */
-		dd->ipath_lbus_speed = 2500;
-		break;
-	}
-
-	if (linkstat < minwidth)
-		ipath_dev_err(dd,
-			"PCIe width %u (x%u HCA), performance "
-			"reduced\n", linkstat, minwidth);
-	else
-		ipath_cdbg(VERBOSE, "PCIe speed %u width %u (x%u HCA)\n",
-			dd->ipath_lbus_speed, linkstat, minwidth);
-
-	if (speed != 1)
-		ipath_dev_err(dd,
-			"PCIe linkspeed %u is incorrect; "
-			"should be 1 (2500)!\n", speed);
-
-bail:
-	/* fill in string, even on errors */
-	snprintf(dd->ipath_lbus_info, sizeof(dd->ipath_lbus_info),
-		"PCIe,%uMHz,x%u\n",
-		dd->ipath_lbus_speed,
-		dd->ipath_lbus_width);
-	return;
-}
-
-
-/**
- * ipath_setup_7220_config - setup PCIe config related stuff
- * @dd: the infinipath device
- * @pdev: the PCI device
- *
- * The pci_enable_msi() call will fail on systems with MSI quirks
- * such as those with AMD8131, even if the device of interest is not
- * attached to that device, (in the 2.6.13 - 2.6.15 kernels, at least, fixed
- * late in 2.6.16).
- * All that can be done is to edit the kernel source to remove the quirk
- * check until that is fixed.
- * We do not need to call enable_msi() for our HyperTransport chip,
- * even though it uses MSI, and we want to avoid the quirk warning, so
- * So we call enable_msi only for PCIe.  If we do end up needing
- * pci_enable_msi at some point in the future for HT, we'll move the
- * call back into the main init_one code.
- * We save the msi lo and hi values, so we can restore them after
- * chip reset (the kernel PCI infrastructure doesn't yet handle that
- * correctly).
- */
-static int ipath_setup_7220_config(struct ipath_devdata *dd,
-				   struct pci_dev *pdev)
-{
-	int pos, ret = -1;
-	u32 boardrev;
-
-	dd->ipath_msi_lo = 0;	/* used as a flag during reset processing */
-
-	pos = pci_find_capability(pdev, PCI_CAP_ID_MSI);
-	if (!strcmp(int_type, "force_msi") || !strcmp(int_type, "auto"))
-		ret = pci_enable_msi(pdev);
-	if (ret) {
-		if (!strcmp(int_type, "force_msi")) {
-			ipath_dev_err(dd, "pci_enable_msi failed: %d, "
-				      "force_msi is on, so not continuing.\n",
-				      ret);
-			return ret;
-		}
-
-		ipath_enable_intx(pdev);
-		if (!strcmp(int_type, "auto"))
-			ipath_dev_err(dd, "pci_enable_msi failed: %d, "
-				      "falling back to INTx\n", ret);
-	} else if (pos) {
-		u16 control;
-		pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
-				      &dd->ipath_msi_lo);
-		pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
-				      &dd->ipath_msi_hi);
-		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS,
-				     &control);
-		/* now save the data (vector) info */
-		pci_read_config_word(pdev,
-				     pos + ((control & PCI_MSI_FLAGS_64BIT)
-					    ? PCI_MSI_DATA_64 :
-					    PCI_MSI_DATA_32),
-				     &dd->ipath_msi_data);
-	} else
-		ipath_dev_err(dd, "Can't find MSI capability, "
-			      "can't save MSI settings for reset\n");
-
-	dd->ipath_irq = pdev->irq;
-
-	/*
-	 * We save the cachelinesize also, although it doesn't
-	 * really matter.
-	 */
-	pci_read_config_byte(pdev, PCI_CACHE_LINE_SIZE,
-			     &dd->ipath_pci_cacheline);
-
-	/*
-	 * this function called early, ipath_boardrev not set yet.  Can't
-	 * use ipath_read_kreg64() yet, too early in init, so use readq()
-	 */
-	boardrev = (readq(&dd->ipath_kregbase[dd->ipath_kregs->kr_revision])
-		 >> INFINIPATH_R_BOARDID_SHIFT) & INFINIPATH_R_BOARDID_MASK;
-
-	ipath_7220_pcie_params(dd, boardrev);
-
-	dd->ipath_flags |= IPATH_NODMA_RTAIL | IPATH_HAS_SEND_DMA |
-		IPATH_HAS_PBC_CNT | IPATH_HAS_THRESH_UPDATE;
-	dd->ipath_pioupd_thresh = 4U; /* set default update threshold */
-	return 0;
-}
-
-static void ipath_init_7220_variables(struct ipath_devdata *dd)
-{
-	/*
-	 * setup the register offsets, since they are different for each
-	 * chip
-	 */
-	dd->ipath_kregs = &ipath_7220_kregs;
-	dd->ipath_cregs = &ipath_7220_cregs;
-
-	/*
-	 * bits for selecting i2c direction and values,
-	 * used for I2C serial flash
-	 */
-	dd->ipath_gpio_sda_num = _IPATH_GPIO_SDA_NUM;
-	dd->ipath_gpio_scl_num = _IPATH_GPIO_SCL_NUM;
-	dd->ipath_gpio_sda = IPATH_GPIO_SDA;
-	dd->ipath_gpio_scl = IPATH_GPIO_SCL;
-
-	/*
-	 * Fill in data for field-values that change in IBA7220.
-	 * We dynamically specify only the mask for LINKTRAININGSTATE
-	 * and only the shift for LINKSTATE, as they are the only ones
-	 * that change.  Also precalculate the 3 link states of interest
-	 * and the combined mask.
-	 */
-	dd->ibcs_ls_shift = IBA7220_IBCS_LINKSTATE_SHIFT;
-	dd->ibcs_lts_mask = IBA7220_IBCS_LINKTRAININGSTATE_MASK;
-	dd->ibcs_mask = (INFINIPATH_IBCS_LINKSTATE_MASK <<
-		dd->ibcs_ls_shift) | dd->ibcs_lts_mask;
-	dd->ib_init = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_INIT << dd->ibcs_ls_shift);
-	dd->ib_arm = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_ARM << dd->ibcs_ls_shift);
-	dd->ib_active = (INFINIPATH_IBCS_LT_STATE_LINKUP <<
-		INFINIPATH_IBCS_LINKTRAININGSTATE_SHIFT) |
-		(INFINIPATH_IBCS_L_STATE_ACTIVE << dd->ibcs_ls_shift);
-
-	/*
-	 * Fill in data for ibcc field-values that change in IBA7220.
-	 * We dynamically specify only the mask for LINKINITCMD
-	 * and only the shift for LINKCMD and MAXPKTLEN, as they are
-	 * the only ones that change.
-	 */
-	dd->ibcc_lic_mask = IBA7220_IBCC_LINKINITCMD_MASK;
-	dd->ibcc_lc_shift = IBA7220_IBCC_LINKCMD_SHIFT;
-	dd->ibcc_mpl_shift = IBA7220_IBCC_MAXPKTLEN_SHIFT;
-
-	/* Fill in shifts for RcvCtrl. */
-	dd->ipath_r_portenable_shift = INFINIPATH_R_PORTENABLE_SHIFT;
-	dd->ipath_r_intravail_shift = IBA7220_R_INTRAVAIL_SHIFT;
-	dd->ipath_r_tailupd_shift = IBA7220_R_TAILUPD_SHIFT;
-	dd->ipath_r_portcfg_shift = IBA7220_R_PORTCFG_SHIFT;
-
-	/* variables for sanity checking interrupt and errors */
-	dd->ipath_hwe_bitsextant =
-		(INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT) |
-		(INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT) |
-		(INFINIPATH_HWE_PCIEMEMPARITYERR_MASK <<
-		 INFINIPATH_HWE_PCIEMEMPARITYERR_SHIFT) |
-		INFINIPATH_HWE_PCIE1PLLFAILED |
-		INFINIPATH_HWE_PCIE0PLLFAILED |
-		INFINIPATH_HWE_PCIEPOISONEDTLP |
-		INFINIPATH_HWE_PCIECPLTIMEOUT |
-		INFINIPATH_HWE_PCIEBUSPARITYXTLH |
-		INFINIPATH_HWE_PCIEBUSPARITYXADM |
-		INFINIPATH_HWE_PCIEBUSPARITYRADM |
-		INFINIPATH_HWE_MEMBISTFAILED |
-		INFINIPATH_HWE_COREPLL_FBSLIP |
-		INFINIPATH_HWE_COREPLL_RFSLIP |
-		INFINIPATH_HWE_SERDESPLLFAILED |
-		INFINIPATH_HWE_IBCBUSTOSPCPARITYERR |
-		INFINIPATH_HWE_IBCBUSFRSPCPARITYERR |
-		INFINIPATH_HWE_PCIECPLDATAQUEUEERR |
-		INFINIPATH_HWE_PCIECPLHDRQUEUEERR |
-		INFINIPATH_HWE_SDMAMEMREADERR |
-		INFINIPATH_HWE_CLK_UC_PLLNOTLOCKED |
-		INFINIPATH_HWE_PCIESERDESQ0PCLKNOTDETECT |
-		INFINIPATH_HWE_PCIESERDESQ1PCLKNOTDETECT |
-		INFINIPATH_HWE_PCIESERDESQ2PCLKNOTDETECT |
-		INFINIPATH_HWE_PCIESERDESQ3PCLKNOTDETECT |
-		INFINIPATH_HWE_DDSRXEQMEMORYPARITYERR |
-		INFINIPATH_HWE_IB_UC_MEMORYPARITYERR |
-		INFINIPATH_HWE_PCIE_UC_OCT0MEMORYPARITYERR |
-		INFINIPATH_HWE_PCIE_UC_OCT1MEMORYPARITYERR;
-	dd->ipath_i_bitsextant =
-		INFINIPATH_I_SDMAINT | INFINIPATH_I_SDMADISABLED |
-		(INFINIPATH_I_RCVURG_MASK << INFINIPATH_I_RCVURG_SHIFT) |
-		(INFINIPATH_I_RCVAVAIL_MASK <<
-		 INFINIPATH_I_RCVAVAIL_SHIFT) |
-		INFINIPATH_I_ERROR | INFINIPATH_I_SPIOSENT |
-		INFINIPATH_I_SPIOBUFAVAIL | INFINIPATH_I_GPIO |
-		INFINIPATH_I_JINT | INFINIPATH_I_SERDESTRIMDONE;
-	dd->ipath_e_bitsextant =
-		INFINIPATH_E_RFORMATERR | INFINIPATH_E_RVCRC |
-		INFINIPATH_E_RICRC | INFINIPATH_E_RMINPKTLEN |
-		INFINIPATH_E_RMAXPKTLEN | INFINIPATH_E_RLONGPKTLEN |
-		INFINIPATH_E_RSHORTPKTLEN | INFINIPATH_E_RUNEXPCHAR |
-		INFINIPATH_E_RUNSUPVL | INFINIPATH_E_REBP |
-		INFINIPATH_E_RIBFLOW | INFINIPATH_E_RBADVERSION |
-		INFINIPATH_E_RRCVEGRFULL | INFINIPATH_E_RRCVHDRFULL |
-		INFINIPATH_E_RBADTID | INFINIPATH_E_RHDRLEN |
-		INFINIPATH_E_RHDR | INFINIPATH_E_RIBLOSTLINK |
-		INFINIPATH_E_SENDSPECIALTRIGGER |
-		INFINIPATH_E_SDMADISABLED | INFINIPATH_E_SMINPKTLEN |
-		INFINIPATH_E_SMAXPKTLEN | INFINIPATH_E_SUNDERRUN |
-		INFINIPATH_E_SPKTLEN | INFINIPATH_E_SDROPPEDSMPPKT |
-		INFINIPATH_E_SDROPPEDDATAPKT |
-		INFINIPATH_E_SPIOARMLAUNCH | INFINIPATH_E_SUNEXPERRPKTNUM |
-		INFINIPATH_E_SUNSUPVL | INFINIPATH_E_SENDBUFMISUSE |
-		INFINIPATH_E_SDMAGENMISMATCH | INFINIPATH_E_SDMAOUTOFBOUND |
-		INFINIPATH_E_SDMATAILOUTOFBOUND | INFINIPATH_E_SDMABASE |
-		INFINIPATH_E_SDMA1STDESC | INFINIPATH_E_SDMARPYTAG |
-		INFINIPATH_E_SDMADWEN | INFINIPATH_E_SDMAMISSINGDW |
-		INFINIPATH_E_SDMAUNEXPDATA |
-		INFINIPATH_E_IBSTATUSCHANGED | INFINIPATH_E_INVALIDADDR |
-		INFINIPATH_E_RESET | INFINIPATH_E_HARDWARE |
-		INFINIPATH_E_SDMADESCADDRMISALIGN |
-		INFINIPATH_E_INVALIDEEPCMD;
-
-	dd->ipath_i_rcvavail_mask = INFINIPATH_I_RCVAVAIL_MASK;
-	dd->ipath_i_rcvurg_mask = INFINIPATH_I_RCVURG_MASK;
-	dd->ipath_i_rcvavail_shift = INFINIPATH_I_RCVAVAIL_SHIFT;
-	dd->ipath_i_rcvurg_shift = INFINIPATH_I_RCVURG_SHIFT;
-	dd->ipath_flags |= IPATH_INTREG_64 | IPATH_HAS_MULT_IB_SPEED
-		| IPATH_HAS_LINK_LATENCY;
-
-	/*
-	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
-	 * 2 is Some Misc, 3 is reserved for future.
-	 */
-	dd->ipath_eep_st_masks[0].hwerrs_to_log =
-		INFINIPATH_HWE_TXEMEMPARITYERR_MASK <<
-		INFINIPATH_HWE_TXEMEMPARITYERR_SHIFT;
-
-	dd->ipath_eep_st_masks[1].hwerrs_to_log =
-		INFINIPATH_HWE_RXEMEMPARITYERR_MASK <<
-		INFINIPATH_HWE_RXEMEMPARITYERR_SHIFT;
-
-	dd->ipath_eep_st_masks[2].errs_to_log = INFINIPATH_E_RESET;
-
-	ipath_linkrecovery = 0;
-
-	init_waitqueue_head(&dd->ipath_autoneg_wait);
-	INIT_DELAYED_WORK(&dd->ipath_autoneg_work,  autoneg_work);
-
-	dd->ipath_link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
-	dd->ipath_link_speed_supported = IPATH_IB_SDR | IPATH_IB_DDR;
-
-	dd->ipath_link_width_enabled = dd->ipath_link_width_supported;
-	dd->ipath_link_speed_enabled = dd->ipath_link_speed_supported;
-	/*
-	 * set the initial values to reasonable default, will be set
-	 * for real when link is up.
-	 */
-	dd->ipath_link_width_active = IB_WIDTH_4X;
-	dd->ipath_link_speed_active = IPATH_IB_SDR;
-	dd->delay_mult = rate_to_delay[0][1];
-}
-
-
-/*
- * Setup the MSI stuff again after a reset.  I'd like to just call
- * pci_enable_msi() and request_irq() again, but when I do that,
- * the MSI enable bit doesn't get set in the command word, and
- * we switch to to a different interrupt vector, which is confusing,
- * so I instead just do it all inline.  Perhaps somehow can tie this
- * into the PCIe hotplug support at some point
- * Note, because I'm doing it all here, I don't call pci_disable_msi()
- * or free_irq() at the start of ipath_setup_7220_reset().
- */
-static int ipath_reinit_msi(struct ipath_devdata *dd)
-{
-	int ret = 0;
-
-	int pos;
-	u16 control;
-	if (!dd->ipath_msi_lo) /* Using intX, or init problem */
-		goto bail;
-
-	pos = pci_find_capability(dd->pcidev, PCI_CAP_ID_MSI);
-	if (!pos) {
-		ipath_dev_err(dd, "Can't find MSI capability, "
-			      "can't restore MSI settings\n");
-		goto bail;
-	}
-	ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
-		   dd->ipath_msi_lo, pos + PCI_MSI_ADDRESS_LO);
-	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
-			       dd->ipath_msi_lo);
-	ipath_cdbg(VERBOSE, "Writing msi_lo 0x%x to config offset 0x%x\n",
-		   dd->ipath_msi_hi, pos + PCI_MSI_ADDRESS_HI);
-	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
-			       dd->ipath_msi_hi);
-	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
-	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
-		ipath_cdbg(VERBOSE, "MSI control at off %x was %x, "
-			   "setting MSI enable (%x)\n", pos + PCI_MSI_FLAGS,
-			   control, control | PCI_MSI_FLAGS_ENABLE);
-		control |= PCI_MSI_FLAGS_ENABLE;
-		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
-				      control);
-	}
-	/* now rewrite the data (vector) info */
-	pci_write_config_word(dd->pcidev, pos +
-			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
-			      dd->ipath_msi_data);
-	ret = 1;
-
-bail:
-	if (!ret) {
-		ipath_dbg("Using INTx, MSI disabled or not configured\n");
-		ipath_enable_intx(dd->pcidev);
-		ret = 1;
-	}
-	/*
-	 * We restore the cachelinesize also, although it doesn't really
-	 * matter.
-	 */
-	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE,
-			      dd->ipath_pci_cacheline);
-	/* and now set the pci master bit again */
-	pci_set_master(dd->pcidev);
-
-	return ret;
-}
-
-/*
- * This routine sleeps, so it can only be called from user context, not
- * from interrupt context.  If we need interrupt context, we can split
- * it into two routines.
- */
-static int ipath_setup_7220_reset(struct ipath_devdata *dd)
-{
-	u64 val;
-	int i;
-	int ret;
-	u16 cmdval;
-
-	pci_read_config_word(dd->pcidev, PCI_COMMAND, &cmdval);
-
-	/* Use dev_err so it shows up in logs, etc. */
-	ipath_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->ipath_unit);
-
-	/* keep chip from being accessed in a few places */
-	dd->ipath_flags &= ~(IPATH_INITTED | IPATH_PRESENT);
-	val = dd->ipath_control | INFINIPATH_C_RESET;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control, val);
-	mb();
-
-	for (i = 1; i <= 5; i++) {
-		int r;
-
-		/*
-		 * Allow MBIST, etc. to complete; longer on each retry.
-		 * We sometimes get machine checks from bus timeout if no
-		 * response, so for now, make it *really* long.
-		 */
-		msleep(1000 + (1 + i) * 2000);
-		r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
-					   dd->ipath_pcibar0);
-		if (r)
-			ipath_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
-		r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
-					   dd->ipath_pcibar1);
-		if (r)
-			ipath_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
-		/* now re-enable memory access */
-		pci_write_config_word(dd->pcidev, PCI_COMMAND, cmdval);
-		r = pci_enable_device(dd->pcidev);
-		if (r)
-			ipath_dev_err(dd, "pci_enable_device failed after "
-				      "reset: %d\n", r);
-		/*
-		 * whether it fully enabled or not, mark as present,
-		 * again (but not INITTED)
-		 */
-		dd->ipath_flags |= IPATH_PRESENT;
-		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_revision);
-		if (val == dd->ipath_revision) {
-			ipath_cdbg(VERBOSE, "Got matching revision "
-				   "register %llx on try %d\n",
-				   (unsigned long long) val, i);
-			ret = ipath_reinit_msi(dd);
-			goto bail;
-		}
-		/* Probably getting -1 back */
-		ipath_dbg("Didn't get expected revision register, "
-			  "got %llx, try %d\n", (unsigned long long) val,
-			  i + 1);
-	}
-	ret = 0; /* failed */
-
-bail:
-	if (ret)
-		ipath_7220_pcie_params(dd, dd->ipath_boardrev);
-
-	return ret;
-}
-
-/**
- * ipath_7220_put_tid - write a TID to the chip
- * @dd: the infinipath device
- * @tidptr: pointer to the expected TID (in chip) to udpate
- * @tidtype: 0 for eager, 1 for expected
- * @pa: physical address of in memory buffer; ipath_tidinvalid if freeing
- *
- * This exists as a separate routine to allow for selection of the
- * appropriate "flavor". The static calls in cleanup just use the
- * revision-agnostic form, as they are not performance critical.
- */
-static void ipath_7220_put_tid(struct ipath_devdata *dd, u64 __iomem *tidptr,
-			     u32 type, unsigned long pa)
-{
-	if (pa != dd->ipath_tidinvalid) {
-		u64 chippa = pa >> IBA7220_TID_PA_SHIFT;
-
-		/* paranoia checks */
-		if (pa != (chippa << IBA7220_TID_PA_SHIFT)) {
-			dev_info(&dd->pcidev->dev, "BUG: physaddr %lx "
-				 "not 2KB aligned!\n", pa);
-			return;
-		}
-		if (pa >= (1UL << IBA7220_TID_SZ_SHIFT)) {
-			ipath_dev_err(dd,
-				      "BUG: Physical page address 0x%lx "
-				      "larger than supported\n", pa);
-			return;
-		}
-
-		if (type == RCVHQ_RCV_TYPE_EAGER)
-			chippa |= dd->ipath_tidtemplate;
-		else /* for now, always full 4KB page */
-			chippa |= IBA7220_TID_SZ_4K;
-		writeq(chippa, tidptr);
-	} else
-		writeq(pa, tidptr);
-	mmiowb();
-}
-
-/**
- * ipath_7220_clear_tid - clear all TID entries for a port, expected and eager
- * @dd: the infinipath device
- * @port: the port
- *
- * clear all TID entries for a port, expected and eager.
- * Used from ipath_close().  On this chip, TIDs are only 32 bits,
- * not 64, but they are still on 64 bit boundaries, so tidbase
- * is declared as u64 * for the pointer math, even though we write 32 bits
- */
-static void ipath_7220_clear_tids(struct ipath_devdata *dd, unsigned port)
-{
-	u64 __iomem *tidbase;
-	unsigned long tidinv;
-	int i;
-
-	if (!dd->ipath_kregbase)
-		return;
-
-	ipath_cdbg(VERBOSE, "Invalidate TIDs for port %u\n", port);
-
-	tidinv = dd->ipath_tidinvalid;
-	tidbase = (u64 __iomem *)
-		((char __iomem *)(dd->ipath_kregbase) +
-		 dd->ipath_rcvtidbase +
-		 port * dd->ipath_rcvtidcnt * sizeof(*tidbase));
-
-	for (i = 0; i < dd->ipath_rcvtidcnt; i++)
-		ipath_7220_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
-				   tidinv);
-
-	tidbase = (u64 __iomem *)
-		((char __iomem *)(dd->ipath_kregbase) +
-		 dd->ipath_rcvegrbase + port_egrtid_idx(dd, port)
-		 * sizeof(*tidbase));
-
-	for (i = port ? dd->ipath_rcvegrcnt : dd->ipath_p0_rcvegrcnt; i; i--)
-		ipath_7220_put_tid(dd, &tidbase[i-1], RCVHQ_RCV_TYPE_EAGER,
-			tidinv);
-}
-
-/**
- * ipath_7220_tidtemplate - setup constants for TID updates
- * @dd: the infinipath device
- *
- * We setup stuff that we use a lot, to avoid calculating each time
- */
-static void ipath_7220_tidtemplate(struct ipath_devdata *dd)
-{
-	/* For now, we always allocate 4KB buffers (at init) so we can
-	 * receive max size packets.  We may want a module parameter to
-	 * specify 2KB or 4KB and/or make be per port instead of per device
-	 * for those who want to reduce memory footprint.  Note that the
-	 * ipath_rcvhdrentsize size must be large enough to hold the largest
-	 * IB header (currently 96 bytes) that we expect to handle (plus of
-	 * course the 2 dwords of RHF).
-	 */
-	if (dd->ipath_rcvegrbufsize == 2048)
-		dd->ipath_tidtemplate = IBA7220_TID_SZ_2K;
-	else if (dd->ipath_rcvegrbufsize == 4096)
-		dd->ipath_tidtemplate = IBA7220_TID_SZ_4K;
-	else {
-		dev_info(&dd->pcidev->dev, "BUG: unsupported egrbufsize "
-			 "%u, using %u\n", dd->ipath_rcvegrbufsize,
-			 4096);
-		dd->ipath_tidtemplate = IBA7220_TID_SZ_4K;
-	}
-	dd->ipath_tidinvalid = 0;
-}
-
-static int ipath_7220_early_init(struct ipath_devdata *dd)
-{
-	u32 i, s;
-
-	if (strcmp(int_type, "auto") &&
-	    strcmp(int_type, "force_msi") &&
-	    strcmp(int_type, "force_intx")) {
-		ipath_dev_err(dd, "Invalid interrupt_type: '%s', expecting "
-			      "auto, force_msi or force_intx\n", int_type);
-		return -EINVAL;
-	}
-
-	/*
-	 * Control[4] has been added to change the arbitration within
-	 * the SDMA engine between favoring data fetches over descriptor
-	 * fetches.  ipath_sdma_fetch_arb==0 gives data fetches priority.
-	 */
-	if (ipath_sdma_fetch_arb && (dd->ipath_minrev > 1))
-		dd->ipath_control |= 1<<4;
-
-	dd->ipath_flags |= IPATH_4BYTE_TID;
-
-	/*
-	 * For openfabrics, we need to be able to handle an IB header of
-	 * 24 dwords.  HT chip has arbitrary sized receive buffers, so we
-	 * made them the same size as the PIO buffers.  This chip does not
-	 * handle arbitrary size buffers, so we need the header large enough
-	 * to handle largest IB header, but still have room for a 2KB MTU
-	 * standard IB packet.
-	 */
-	dd->ipath_rcvhdrentsize = 24;
-	dd->ipath_rcvhdrsize = IPATH_DFLT_RCVHDRSIZE;
-	dd->ipath_rhf_offset =
-		dd->ipath_rcvhdrentsize - sizeof(u64) / sizeof(u32);
-
-	dd->ipath_rcvegrbufsize = ipath_mtu4096 ? 4096 : 2048;
-	/*
-	 * the min() check here is currently a nop, but it may not always
-	 * be, depending on just how we do ipath_rcvegrbufsize
-	 */
-	dd->ipath_ibmaxlen = min(ipath_mtu4096 ? dd->ipath_piosize4k :
-				 dd->ipath_piosize2k,
-				 dd->ipath_rcvegrbufsize +
-				 (dd->ipath_rcvhdrentsize << 2));
-	dd->ipath_init_ibmaxlen = dd->ipath_ibmaxlen;
-
-	ipath_7220_config_jint(dd, INFINIPATH_JINT_DEFAULT_IDLE_TICKS,
-			       INFINIPATH_JINT_DEFAULT_MAX_PACKETS);
-
-	if (dd->ipath_boardrev) /* no eeprom on emulator */
-		ipath_get_eeprom_info(dd);
-
-	/* start of code to check and print procmon */
-	s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon));
-	s &= ~(1U<<31); /* clear done bit */
-	s |= 1U<<14; /* clear counter (write 1 to clear) */
-	ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s);
-	/* make sure clear_counter low long enough before start */
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-
-	s &= ~(1U<<14); /* allow counter to count (before starting) */
-	ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon));
-
-	s |= 1U<<15; /* start the counter */
-	s &= ~(1U<<31); /* clear done bit */
-	s &= ~0x7ffU; /* clear frequency bits */
-	s |= 0xe29; /* set frequency bits, in case cleared */
-	ipath_write_kreg(dd, IPATH_KREG_OFFSET(ProcMon), s);
-
-	s = 0;
-	for (i = 500; i > 0 && !(s&(1ULL<<31)); i--) {
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-		s = ipath_read_kreg32(dd, IPATH_KREG_OFFSET(ProcMon));
-	}
-	if (!(s&(1U<<31)))
-		ipath_dev_err(dd, "ProcMon register not valid: 0x%x\n", s);
-	else
-		ipath_dbg("ProcMon=0x%x, count=0x%x\n", s, (s>>16)&0x1ff);
-
-	return 0;
-}
-
-/**
- * ipath_init_7220_get_base_info - set chip-specific flags for user code
- * @pd: the infinipath port
- * @kbase: ipath_base_info pointer
- *
- * We set the PCIE flag because the lower bandwidth on PCIe vs
- * HyperTransport can affect some user packet algorithims.
- */
-static int ipath_7220_get_base_info(struct ipath_portdata *pd, void *kbase)
-{
-	struct ipath_base_info *kinfo = kbase;
-
-	kinfo->spi_runtime_flags |=
-		IPATH_RUNTIME_PCIE | IPATH_RUNTIME_NODMA_RTAIL |
-		IPATH_RUNTIME_SDMA;
-
-	return 0;
-}
-
-static void ipath_7220_free_irq(struct ipath_devdata *dd)
-{
-	free_irq(dd->ipath_irq, dd);
-	dd->ipath_irq = 0;
-}
-
-static struct ipath_message_header *
-ipath_7220_get_msgheader(struct ipath_devdata *dd, __le32 *rhf_addr)
-{
-	u32 offset = ipath_hdrget_offset(rhf_addr);
-
-	return (struct ipath_message_header *)
-		(rhf_addr - dd->ipath_rhf_offset + offset);
-}
-
-static void ipath_7220_config_ports(struct ipath_devdata *dd, ushort cfgports)
-{
-	u32 nchipports;
-
-	nchipports = ipath_read_kreg32(dd, dd->ipath_kregs->kr_portcnt);
-	if (!cfgports) {
-		int ncpus = num_online_cpus();
-
-		if (ncpus <= 4)
-			dd->ipath_portcnt = 5;
-		else if (ncpus <= 8)
-			dd->ipath_portcnt = 9;
-		if (dd->ipath_portcnt)
-			ipath_dbg("Auto-configured for %u ports, %d cpus "
-				"online\n", dd->ipath_portcnt, ncpus);
-	} else if (cfgports <= nchipports)
-		dd->ipath_portcnt = cfgports;
-	if (!dd->ipath_portcnt) /* none of the above, set to max */
-		dd->ipath_portcnt = nchipports;
-	/*
-	 * chip can be configured for 5, 9, or 17 ports, and choice
-	 * affects number of eager TIDs per port (1K, 2K, 4K).
-	 */
-	if (dd->ipath_portcnt > 9)
-		dd->ipath_rcvctrl |= 2ULL << IBA7220_R_PORTCFG_SHIFT;
-	else if (dd->ipath_portcnt > 5)
-		dd->ipath_rcvctrl |= 1ULL << IBA7220_R_PORTCFG_SHIFT;
-	/* else configure for default 5 receive ports */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_rcvctrl,
-			 dd->ipath_rcvctrl);
-	dd->ipath_p0_rcvegrcnt = 2048; /* always */
-	if (dd->ipath_flags & IPATH_HAS_SEND_DMA)
-		dd->ipath_pioreserved = 3; /* kpiobufs used for PIO */
-}
-
-
-static int ipath_7220_get_ib_cfg(struct ipath_devdata *dd, int which)
-{
-	int lsb, ret = 0;
-	u64 maskr; /* right-justified mask */
-
-	switch (which) {
-	case IPATH_IB_CFG_HRTBT: /* Get Heartbeat off/enable/auto */
-		lsb = IBA7220_IBC_HRTBT_SHIFT;
-		maskr = IBA7220_IBC_HRTBT_MASK;
-		break;
-
-	case IPATH_IB_CFG_LWID_ENB: /* Get allowed Link-width */
-		ret = dd->ipath_link_width_enabled;
-		goto done;
-
-	case IPATH_IB_CFG_LWID: /* Get currently active Link-width */
-		ret = dd->ipath_link_width_active;
-		goto done;
-
-	case IPATH_IB_CFG_SPD_ENB: /* Get allowed Link speeds */
-		ret = dd->ipath_link_speed_enabled;
-		goto done;
-
-	case IPATH_IB_CFG_SPD: /* Get current Link spd */
-		ret = dd->ipath_link_speed_active;
-		goto done;
-
-	case IPATH_IB_CFG_RXPOL_ENB: /* Get Auto-RX-polarity enable */
-		lsb = IBA7220_IBC_RXPOL_SHIFT;
-		maskr = IBA7220_IBC_RXPOL_MASK;
-		break;
-
-	case IPATH_IB_CFG_LREV_ENB: /* Get Auto-Lane-reversal enable */
-		lsb = IBA7220_IBC_LREV_SHIFT;
-		maskr = IBA7220_IBC_LREV_MASK;
-		break;
-
-	case IPATH_IB_CFG_LINKLATENCY:
-		ret = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcddrstatus)
-			& IBA7220_DDRSTAT_LINKLAT_MASK;
-		goto done;
-
-	default:
-		ret = -ENOTSUPP;
-		goto done;
-	}
-	ret = (int)((dd->ipath_ibcddrctrl >> lsb) & maskr);
-done:
-	return ret;
-}
-
-static int ipath_7220_set_ib_cfg(struct ipath_devdata *dd, int which, u32 val)
-{
-	int lsb, ret = 0, setforce = 0;
-	u64 maskr; /* right-justified mask */
-
-	switch (which) {
-	case IPATH_IB_CFG_LIDLMC:
-		/*
-		 * Set LID and LMC. Combined to avoid possible hazard
-		 * caller puts LMC in 16MSbits, DLID in 16LSbits of val
-		 */
-		lsb = IBA7220_IBC_DLIDLMC_SHIFT;
-		maskr = IBA7220_IBC_DLIDLMC_MASK;
-		break;
-
-	case IPATH_IB_CFG_HRTBT: /* set Heartbeat off/enable/auto */
-		if (val & IPATH_IB_HRTBT_ON &&
-			(dd->ipath_flags & IPATH_NO_HRTBT))
-			goto bail;
-		lsb = IBA7220_IBC_HRTBT_SHIFT;
-		maskr = IBA7220_IBC_HRTBT_MASK;
-		break;
-
-	case IPATH_IB_CFG_LWID_ENB: /* set allowed Link-width */
-		/*
-		 * As with speed, only write the actual register if
-		 * the link is currently down, otherwise takes effect
-		 * on next link change.
-		 */
-		dd->ipath_link_width_enabled = val;
-		if ((dd->ipath_flags & (IPATH_LINKDOWN|IPATH_LINKINIT)) !=
-			IPATH_LINKDOWN)
-			goto bail;
-		/*
-		 * We set the IPATH_IB_FORCE_NOTIFY bit so updown
-		 * will get called because we want update
-		 * link_width_active, and the change may not take
-		 * effect for some time (if we are in POLL), so this
-		 * flag will force the updown routine to be called
-		 * on the next ibstatuschange down interrupt, even
-		 * if it's not an down->up transition.
-		 */
-		val--; /* convert from IB to chip */
-		maskr = IBA7220_IBC_WIDTH_MASK;
-		lsb = IBA7220_IBC_WIDTH_SHIFT;
-		setforce = 1;
-		dd->ipath_flags |= IPATH_IB_FORCE_NOTIFY;
-		break;
-
-	case IPATH_IB_CFG_SPD_ENB: /* set allowed Link speeds */
-		/*
-		 * If we turn off IB1.2, need to preset SerDes defaults,
-		 * but not right now. Set a flag for the next time
-		 * we command the link down.  As with width, only write the
-		 * actual register if the link is currently down, otherwise
-		 * takes effect on next link change.  Since setting is being
-		 * explictly requested (via MAD or sysfs), clear autoneg
-		 * failure status if speed autoneg is enabled.
-		 */
-		dd->ipath_link_speed_enabled = val;
-		if (dd->ipath_ibcddrctrl & IBA7220_IBC_IBTA_1_2_MASK &&
-		    !(val & (val - 1)))
-			dd->ipath_presets_needed = 1;
-		if ((dd->ipath_flags & (IPATH_LINKDOWN|IPATH_LINKINIT)) !=
-			IPATH_LINKDOWN)
-			goto bail;
-		/*
-		 * We set the IPATH_IB_FORCE_NOTIFY bit so updown
-		 * will get called because we want update
-		 * link_speed_active, and the change may not take
-		 * effect for some time (if we are in POLL), so this
-		 * flag will force the updown routine to be called
-		 * on the next ibstatuschange down interrupt, even
-		 * if it's not an down->up transition.  When setting
-		 * speed autoneg, clear AUTONEG_FAILED.
-		 */
-		if (val == (IPATH_IB_SDR | IPATH_IB_DDR)) {
-			val = IBA7220_IBC_SPEED_AUTONEG_MASK |
-				IBA7220_IBC_IBTA_1_2_MASK;
-			dd->ipath_flags &= ~IPATH_IB_AUTONEG_FAILED;
-		} else
-			val = val == IPATH_IB_DDR ?  IBA7220_IBC_SPEED_DDR
-				: IBA7220_IBC_SPEED_SDR;
-		maskr = IBA7220_IBC_SPEED_AUTONEG_MASK |
-			IBA7220_IBC_IBTA_1_2_MASK;
-		lsb = 0; /* speed bits are low bits */
-		setforce = 1;
-		break;
-
-	case IPATH_IB_CFG_RXPOL_ENB: /* set Auto-RX-polarity enable */
-		lsb = IBA7220_IBC_RXPOL_SHIFT;
-		maskr = IBA7220_IBC_RXPOL_MASK;
-		break;
-
-	case IPATH_IB_CFG_LREV_ENB: /* set Auto-Lane-reversal enable */
-		lsb = IBA7220_IBC_LREV_SHIFT;
-		maskr = IBA7220_IBC_LREV_MASK;
-		break;
-
-	default:
-		ret = -ENOTSUPP;
-		goto bail;
-	}
-	dd->ipath_ibcddrctrl &= ~(maskr << lsb);
-	dd->ipath_ibcddrctrl |= (((u64) val & maskr) << lsb);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl,
-			 dd->ipath_ibcddrctrl);
-	if (setforce)
-		dd->ipath_flags |= IPATH_IB_FORCE_NOTIFY;
-bail:
-	return ret;
-}
-
-static void ipath_7220_read_counters(struct ipath_devdata *dd,
-				     struct infinipath_counters *cntrs)
-{
-	u64 *counters = (u64 *) cntrs;
-	int i;
-
-	for (i = 0; i < sizeof(*cntrs) / sizeof(u64); i++)
-		counters[i] = ipath_snap_cntr(dd, i);
-}
-
-/* if we are using MSI, try to fallback to INTx */
-static int ipath_7220_intr_fallback(struct ipath_devdata *dd)
-{
-	if (dd->ipath_msi_lo) {
-		dev_info(&dd->pcidev->dev, "MSI interrupt not detected,"
-			" trying INTx interrupts\n");
-		ipath_7220_nomsi(dd);
-		ipath_enable_intx(dd->pcidev);
-		/*
-		 * some newer kernels require free_irq before disable_msi,
-		 * and irq can be changed during disable and intx enable
-		 * and we need to therefore use the pcidev->irq value,
-		 * not our saved MSI value.
-		 */
-		dd->ipath_irq = dd->pcidev->irq;
-		if (request_irq(dd->ipath_irq, ipath_intr, IRQF_SHARED,
-			IPATH_DRV_NAME, dd))
-			ipath_dev_err(dd,
-				"Could not re-request_irq for INTx\n");
-		return 1;
-	}
-	return 0;
-}
-
-/*
- * reset the XGXS (between serdes and IBC).  Slightly less intrusive
- * than resetting the IBC or external link state, and useful in some
- * cases to cause some retraining.  To do this right, we reset IBC
- * as well.
- */
-static void ipath_7220_xgxs_reset(struct ipath_devdata *dd)
-{
-	u64 val, prev_val;
-
-	prev_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_xgxsconfig);
-	val = prev_val | INFINIPATH_XGXS_RESET;
-	prev_val &= ~INFINIPATH_XGXS_RESET; /* be sure */
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-			 dd->ipath_control & ~INFINIPATH_C_LINKENABLE);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, val);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_xgxsconfig, prev_val);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_control,
-			 dd->ipath_control);
-}
-
-
-/* Still needs cleanup, too much hardwired stuff */
-static void autoneg_send(struct ipath_devdata *dd,
-	u32 *hdr, u32 dcnt, u32 *data)
-{
-	int i;
-	u64 cnt;
-	u32 __iomem *piobuf;
-	u32 pnum;
-
-	i = 0;
-	cnt = 7 + dcnt + 1; /* 7 dword header, dword data, icrc */
-	while (!(piobuf = ipath_getpiobuf(dd, cnt, &pnum))) {
-		if (i++ > 15) {
-			ipath_dbg("Couldn't get pio buffer for send\n");
-			return;
-		}
-		udelay(2);
-	}
-	if (dd->ipath_flags&IPATH_HAS_PBC_CNT)
-		cnt |= 0x80000000UL<<32; /* mark as VL15 */
-	writeq(cnt, piobuf);
-	ipath_flush_wc();
-	__iowrite32_copy(piobuf + 2, hdr, 7);
-	__iowrite32_copy(piobuf + 9, data, dcnt);
-	ipath_flush_wc();
-}
-
-/*
- * _start packet gets sent twice at start, _done gets sent twice at end
- */
-static void ipath_autoneg_send(struct ipath_devdata *dd, int which)
-{
-	static u32 swapped;
-	u32 dw, i, hcnt, dcnt, *data;
-	static u32 hdr[7] = { 0xf002ffff, 0x48ffff, 0x6400abba };
-	static u32 madpayload_start[0x40] = {
-		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
-		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
-		0x1, 0x1388, 0x15e, 0x1, /* rest 0's */
-		};
-	static u32 madpayload_done[0x40] = {
-		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
-		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
-		0x40000001, 0x1388, 0x15e, /* rest 0's */
-		};
-	dcnt = ARRAY_SIZE(madpayload_start);
-	hcnt = ARRAY_SIZE(hdr);
-	if (!swapped) {
-		/* for maintainability, do it at runtime */
-		for (i = 0; i < hcnt; i++) {
-			dw = (__force u32) cpu_to_be32(hdr[i]);
-			hdr[i] = dw;
-		}
-		for (i = 0; i < dcnt; i++) {
-			dw = (__force u32) cpu_to_be32(madpayload_start[i]);
-			madpayload_start[i] = dw;
-			dw = (__force u32) cpu_to_be32(madpayload_done[i]);
-			madpayload_done[i] = dw;
-		}
-		swapped = 1;
-	}
-
-	data = which ? madpayload_done : madpayload_start;
-	ipath_cdbg(PKT, "Sending %s special MADs\n", which?"done":"start");
-
-	autoneg_send(dd, hdr, dcnt, data);
-	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
-	udelay(2);
-	autoneg_send(dd, hdr, dcnt, data);
-	ipath_read_kreg64(dd, dd->ipath_kregs->kr_scratch);
-	udelay(2);
-}
-
-
-
-/*
- * Do the absolute minimum to cause an IB speed change, and make it
- * ready, but don't actually trigger the change.   The caller will
- * do that when ready (if link is in Polling training state, it will
- * happen immediately, otherwise when link next goes down)
- *
- * This routine should only be used as part of the DDR autonegotation
- * code for devices that are not compliant with IB 1.2 (or code that
- * fixes things up for same).
- *
- * When link has gone down, and autoneg enabled, or autoneg has
- * failed and we give up until next time we set both speeds, and
- * then we want IBTA enabled as well as "use max enabled speed.
- */
-static void set_speed_fast(struct ipath_devdata *dd, u32 speed)
-{
-	dd->ipath_ibcddrctrl &= ~(IBA7220_IBC_SPEED_AUTONEG_MASK |
-		IBA7220_IBC_IBTA_1_2_MASK |
-		(IBA7220_IBC_WIDTH_MASK << IBA7220_IBC_WIDTH_SHIFT));
-
-	if (speed == (IPATH_IB_SDR | IPATH_IB_DDR))
-		dd->ipath_ibcddrctrl |= IBA7220_IBC_SPEED_AUTONEG_MASK |
-			IBA7220_IBC_IBTA_1_2_MASK;
-	else
-		dd->ipath_ibcddrctrl |= speed == IPATH_IB_DDR ?
-			IBA7220_IBC_SPEED_DDR : IBA7220_IBC_SPEED_SDR;
-
-	/*
-	 * Convert from IB-style 1 = 1x, 2 = 4x, 3 = auto
-	 * to chip-centric       0 = 1x, 1 = 4x, 2 = auto
-	 */
-	dd->ipath_ibcddrctrl |= (u64)(dd->ipath_link_width_enabled - 1) <<
-		IBA7220_IBC_WIDTH_SHIFT;
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibcddrctrl,
-			dd->ipath_ibcddrctrl);
-	ipath_cdbg(VERBOSE, "setup for IB speed (%x) done\n", speed);
-}
-
-
-/*
- * this routine is only used when we are not talking to another
- * IB 1.2-compliant device that we think can do DDR.
- * (This includes all existing switch chips as of Oct 2007.)
- * 1.2-compliant devices go directly to DDR prior to reaching INIT
- */
-static void try_auto_neg(struct ipath_devdata *dd)
-{
-	/*
-	 * required for older non-IB1.2 DDR switches.  Newer
-	 * non-IB-compliant switches don't need it, but so far,
-	 * aren't bothered by it either.  "Magic constant"
-	 */
-	ipath_write_kreg(dd, IPATH_KREG_OFFSET(IBNCModeCtrl),
-		0x3b9dc07);
-	dd->ipath_flags |= IPATH_IB_AUTONEG_INPROG;
-	ipath_autoneg_send(dd, 0);
-	set_speed_fast(dd, IPATH_IB_DDR);
-	ipath_toggle_rclkrls(dd);
-	/* 2 msec is minimum length of a poll cycle */
-	schedule_delayed_work(&dd->ipath_autoneg_work,
-		msecs_to_jiffies(2));
-}
-
-
-static int ipath_7220_ib_updown(struct ipath_devdata *dd, int ibup, u64 ibcs)
-{
-	int ret = 0;
-	u32 ltstate = ipath_ib_linkstate(dd, ibcs);
-
-	dd->ipath_link_width_active =
-		((ibcs >> IBA7220_IBCS_LINKWIDTH_SHIFT) & 1) ?
-		    IB_WIDTH_4X : IB_WIDTH_1X;
-	dd->ipath_link_speed_active =
-		((ibcs >> IBA7220_IBCS_LINKSPEED_SHIFT) & 1) ?
-		    IPATH_IB_DDR : IPATH_IB_SDR;
-
-	if (!ibup) {
-		/*
-		 * when link goes down we don't want aeq running, so it
-		 * won't't interfere with IBC training, etc., and we need
-		 * to go back to the static SerDes preset values
-		 */
-		if (dd->ipath_x1_fix_tries &&
-			 ltstate <= INFINIPATH_IBCS_LT_STATE_SLEEPQUIET &&
-			ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP)
-			dd->ipath_x1_fix_tries = 0;
-		if (!(dd->ipath_flags & (IPATH_IB_AUTONEG_FAILED |
-			IPATH_IB_AUTONEG_INPROG)))
-			set_speed_fast(dd, dd->ipath_link_speed_enabled);
-		if (!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG)) {
-			ipath_cdbg(VERBOSE, "Setting RXEQ defaults\n");
-			ipath_sd7220_presets(dd);
-		}
-		/* this might better in ipath_sd7220_presets() */
-		ipath_set_relock_poll(dd, ibup);
-	} else {
-		if (ipath_compat_ddr_negotiate &&
-		    !(dd->ipath_flags & (IPATH_IB_AUTONEG_FAILED |
-			IPATH_IB_AUTONEG_INPROG)) &&
-			dd->ipath_link_speed_active == IPATH_IB_SDR &&
-			(dd->ipath_link_speed_enabled &
-			    (IPATH_IB_DDR | IPATH_IB_SDR)) ==
-			    (IPATH_IB_DDR | IPATH_IB_SDR) &&
-			dd->ipath_autoneg_tries < IPATH_AUTONEG_TRIES) {
-			/* we are SDR, and DDR auto-negotiation enabled */
-			++dd->ipath_autoneg_tries;
-			ipath_dbg("DDR negotiation try, %u/%u\n",
-				dd->ipath_autoneg_tries,
-				IPATH_AUTONEG_TRIES);
-			try_auto_neg(dd);
-			ret = 1; /* no other IB status change processing */
-		} else if ((dd->ipath_flags & IPATH_IB_AUTONEG_INPROG)
-			&& dd->ipath_link_speed_active == IPATH_IB_SDR) {
-			ipath_autoneg_send(dd, 1);
-			set_speed_fast(dd, IPATH_IB_DDR);
-			udelay(2);
-			ipath_toggle_rclkrls(dd);
-			ret = 1; /* no other IB status change processing */
-		} else {
-			if ((dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) &&
-				(dd->ipath_link_speed_active & IPATH_IB_DDR)) {
-				ipath_dbg("Got to INIT with DDR autoneg\n");
-				dd->ipath_flags &= ~(IPATH_IB_AUTONEG_INPROG
-					| IPATH_IB_AUTONEG_FAILED);
-				dd->ipath_autoneg_tries = 0;
-				/* re-enable SDR, for next link down */
-				set_speed_fast(dd,
-					dd->ipath_link_speed_enabled);
-				wake_up(&dd->ipath_autoneg_wait);
-			} else if (dd->ipath_flags & IPATH_IB_AUTONEG_FAILED) {
-				/*
-				 * clear autoneg failure flag, and do setup
-				 * so we'll try next time link goes down and
-				 * back to INIT (possibly connected to different
-				 * device).
-				 */
-				ipath_dbg("INIT %sDR after autoneg failure\n",
-					(dd->ipath_link_speed_active &
-					  IPATH_IB_DDR) ? "D" : "S");
-				dd->ipath_flags &= ~IPATH_IB_AUTONEG_FAILED;
-				dd->ipath_ibcddrctrl |=
-					IBA7220_IBC_IBTA_1_2_MASK;
-				ipath_write_kreg(dd,
-					IPATH_KREG_OFFSET(IBNCModeCtrl), 0);
-			}
-		}
-		/*
-		 * if we are in 1X, and are in autoneg width, it
-		 * could be due to an xgxs problem, so if we haven't
-		 * already tried, try twice to get to 4X; if we
-		 * tried, and couldn't, report it, since it will
-		 * probably not be what is desired.
-		 */
-		if ((dd->ipath_link_width_enabled & (IB_WIDTH_1X |
-			IB_WIDTH_4X)) == (IB_WIDTH_1X | IB_WIDTH_4X)
-			&& dd->ipath_link_width_active == IB_WIDTH_1X
-			&& dd->ipath_x1_fix_tries < 3) {
-			if (++dd->ipath_x1_fix_tries == 3)
-				dev_info(&dd->pcidev->dev,
-					"IB link is in 1X mode\n");
-			else {
-				ipath_cdbg(VERBOSE, "IB 1X in "
-					"auto-width, try %u to be "
-					"sure it's really 1X; "
-					"ltstate %u\n",
-					 dd->ipath_x1_fix_tries,
-					 ltstate);
-				dd->ipath_f_xgxs_reset(dd);
-				ret = 1; /* skip other processing */
-			}
-		}
-
-		if (!ret) {
-			dd->delay_mult = rate_to_delay
-			    [(ibcs >> IBA7220_IBCS_LINKSPEED_SHIFT) & 1]
-			    [(ibcs >> IBA7220_IBCS_LINKWIDTH_SHIFT) & 1];
-
-			ipath_set_relock_poll(dd, ibup);
-		}
-	}
-
-	if (!ret)
-		ipath_setup_7220_setextled(dd, ipath_ib_linkstate(dd, ibcs),
-			ltstate);
-	return ret;
-}
-
-
-/*
- * Handle the empirically determined mechanism for auto-negotiation
- * of DDR speed with switches.
- */
-static void autoneg_work(struct work_struct *work)
-{
-	struct ipath_devdata *dd;
-	u64 startms;
-	u32 lastlts, i;
-
-	dd = container_of(work, struct ipath_devdata,
-		ipath_autoneg_work.work);
-
-	startms = jiffies_to_msecs(jiffies);
-
-	/*
-	 * busy wait for this first part, it should be at most a
-	 * few hundred usec, since we scheduled ourselves for 2msec.
-	 */
-	for (i = 0; i < 25; i++) {
-		lastlts = ipath_ib_linktrstate(dd, dd->ipath_lastibcstat);
-		if (lastlts == INFINIPATH_IBCS_LT_STATE_POLLQUIET) {
-			ipath_set_linkstate(dd, IPATH_IB_LINKDOWN_DISABLE);
-			break;
-		}
-		udelay(100);
-	}
-
-	if (!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG))
-		goto done; /* we got there early or told to stop */
-
-	/* we expect this to timeout */
-	if (wait_event_timeout(dd->ipath_autoneg_wait,
-		!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG),
-		msecs_to_jiffies(90)))
-		goto done;
-
-	ipath_toggle_rclkrls(dd);
-
-	/* we expect this to timeout */
-	if (wait_event_timeout(dd->ipath_autoneg_wait,
-		!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG),
-		msecs_to_jiffies(1700)))
-		goto done;
-
-	set_speed_fast(dd, IPATH_IB_SDR);
-	ipath_toggle_rclkrls(dd);
-
-	/*
-	 * wait up to 250 msec for link to train and get to INIT at DDR;
-	 * this should terminate early
-	 */
-	wait_event_timeout(dd->ipath_autoneg_wait,
-		!(dd->ipath_flags & IPATH_IB_AUTONEG_INPROG),
-		msecs_to_jiffies(250));
-done:
-	if (dd->ipath_flags & IPATH_IB_AUTONEG_INPROG) {
-		ipath_dbg("Did not get to DDR INIT (%x) after %Lu msecs\n",
-			ipath_ib_state(dd, dd->ipath_lastibcstat),
-			jiffies_to_msecs(jiffies)-startms);
-		dd->ipath_flags &= ~IPATH_IB_AUTONEG_INPROG;
-		if (dd->ipath_autoneg_tries == IPATH_AUTONEG_TRIES) {
-			dd->ipath_flags |= IPATH_IB_AUTONEG_FAILED;
-			ipath_dbg("Giving up on DDR until next IB "
-				"link Down\n");
-			dd->ipath_autoneg_tries = 0;
-		}
-		set_speed_fast(dd, dd->ipath_link_speed_enabled);
-	}
-}
-
-
-/**
- * ipath_init_iba7220_funcs - set up the chip-specific function pointers
- * @dd: the infinipath device
- *
- * This is global, and is called directly at init to set up the
- * chip-specific function pointers for later use.
- */
-void ipath_init_iba7220_funcs(struct ipath_devdata *dd)
-{
-	dd->ipath_f_intrsetup = ipath_7220_intconfig;
-	dd->ipath_f_bus = ipath_setup_7220_config;
-	dd->ipath_f_reset = ipath_setup_7220_reset;
-	dd->ipath_f_get_boardname = ipath_7220_boardname;
-	dd->ipath_f_init_hwerrors = ipath_7220_init_hwerrors;
-	dd->ipath_f_early_init = ipath_7220_early_init;
-	dd->ipath_f_handle_hwerrors = ipath_7220_handle_hwerrors;
-	dd->ipath_f_quiet_serdes = ipath_7220_quiet_serdes;
-	dd->ipath_f_bringup_serdes = ipath_7220_bringup_serdes;
-	dd->ipath_f_clear_tids = ipath_7220_clear_tids;
-	dd->ipath_f_put_tid = ipath_7220_put_tid;
-	dd->ipath_f_cleanup = ipath_setup_7220_cleanup;
-	dd->ipath_f_setextled = ipath_setup_7220_setextled;
-	dd->ipath_f_get_base_info = ipath_7220_get_base_info;
-	dd->ipath_f_free_irq = ipath_7220_free_irq;
-	dd->ipath_f_tidtemplate = ipath_7220_tidtemplate;
-	dd->ipath_f_intr_fallback = ipath_7220_intr_fallback;
-	dd->ipath_f_xgxs_reset = ipath_7220_xgxs_reset;
-	dd->ipath_f_get_ib_cfg = ipath_7220_get_ib_cfg;
-	dd->ipath_f_set_ib_cfg = ipath_7220_set_ib_cfg;
-	dd->ipath_f_config_jint = ipath_7220_config_jint;
-	dd->ipath_f_config_ports = ipath_7220_config_ports;
-	dd->ipath_f_read_counters = ipath_7220_read_counters;
-	dd->ipath_f_get_msgheader = ipath_7220_get_msgheader;
-	dd->ipath_f_ib_updown = ipath_7220_ib_updown;
-
-	/* initialize chip-specific variables */
-	ipath_init_7220_variables(dd);
-}
diff --git a/drivers/infiniband/hw/ipath/ipath_init_chip.c b/drivers/infiniband/hw/ipath/ipath_init_chip.c
index 3e5baa43fc8..be2a60e142b 100644
--- a/drivers/infiniband/hw/ipath/ipath_init_chip.c
+++ b/drivers/infiniband/hw/ipath/ipath_init_chip.c
@@ -33,6 +33,9 @@
 
 #include <linux/pci.h>
 #include <linux/netdevice.h>
+#include <linux/moduleparam.h>
+#include <linux/slab.h>
+#include <linux/stat.h>
 #include <linux/vmalloc.h>
 
 #include "ipath_kernel.h"
@@ -229,6 +232,7 @@ static int init_chip_first(struct ipath_devdata *dd)
 	spin_lock_init(&dd->ipath_kernel_tid_lock);
 	spin_lock_init(&dd->ipath_user_tid_lock);
 	spin_lock_init(&dd->ipath_sendctrl_lock);
+	spin_lock_init(&dd->ipath_uctxt_lock);
 	spin_lock_init(&dd->ipath_sdma_lock);
 	spin_lock_init(&dd->ipath_gpio_lock);
 	spin_lock_init(&dd->ipath_eep_st_lock);
@@ -333,7 +337,7 @@ done:
  * @dd: the infinipath device
  *
  * sanity check at least some of the values after reset, and
- * ensure no receive or transmit (explictly, in case reset
+ * ensure no receive or transmit (explicitly, in case reset
  * failed
  */
 static int init_chip_reset(struct ipath_devdata *dd)
@@ -440,7 +444,7 @@ static void init_shadow_tids(struct ipath_devdata *dd)
 	struct page **pages;
 	dma_addr_t *addrs;
 
-	pages = vmalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
+	pages = vzalloc(dd->ipath_cfgports * dd->ipath_rcvtidcnt *
 			sizeof(struct page *));
 	if (!pages) {
 		ipath_dev_err(dd, "failed to allocate shadow page * "
@@ -454,14 +458,11 @@ static void init_shadow_tids(struct ipath_devdata *dd)
 	if (!addrs) {
 		ipath_dev_err(dd, "failed to allocate shadow dma handle "
 			      "array, no expected sends!\n");
-		vfree(dd->ipath_pageshadow);
+		vfree(pages);
 		dd->ipath_pageshadow = NULL;
 		return;
 	}
 
-	memset(pages, 0, dd->ipath_cfgports * dd->ipath_rcvtidcnt *
-	       sizeof(struct page *));
-
 	dd->ipath_pageshadow = pages;
 	dd->ipath_physshadow = addrs;
 }
@@ -718,16 +719,6 @@ int ipath_init_chip(struct ipath_devdata *dd, int reinit)
 		goto done;
 
 	/*
-	 * we ignore most issues after reporting them, but have to specially
-	 * handle hardware-disabled chips.
-	 */
-	if (ret == 2) {
-		/* unique error, known to ipath_init_one */
-		ret = -EPERM;
-		goto done;
-	}
-
-	/*
 	 * We could bump this to allow for full rcvegrcnt + rcvtidcnt,
 	 * but then it no longer nicely fits power of two, and since
 	 * we now use routines that backend onto __get_free_pages, the
diff --git a/drivers/infiniband/hw/ipath/ipath_intr.c b/drivers/infiniband/hw/ipath/ipath_intr.c
index 26900b3b7a4..01ba792791a 100644
--- a/drivers/infiniband/hw/ipath/ipath_intr.c
+++ b/drivers/infiniband/hw/ipath/ipath_intr.c
@@ -33,6 +33,7 @@
 
 #include <linux/pci.h>
 #include <linux/delay.h>
+#include <linux/sched.h>
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
@@ -69,7 +70,7 @@ void ipath_disarm_senderrbufs(struct ipath_devdata *dd)
 	if (sbuf[0] || sbuf[1] || (piobcnt > 128 && (sbuf[2] || sbuf[3]))) {
 		int i;
 		if (ipath_debug & (__IPATH_PKTDBG|__IPATH_DBG) &&
-			dd->ipath_lastcancel > jiffies) {
+			time_after(dd->ipath_lastcancel, jiffies)) {
 			__IPATH_DBG_WHICH(__IPATH_PKTDBG|__IPATH_DBG,
 					  "SendbufErrs %lx %lx", sbuf[0],
 					  sbuf[1]);
@@ -208,8 +209,7 @@ void ipath_format_hwerrors(u64 hwerrs,
 {
 	int i;
 	const int glen =
-	    sizeof(ipath_generic_hwerror_msgs) /
-	    sizeof(ipath_generic_hwerror_msgs[0]);
+	    ARRAY_SIZE(ipath_generic_hwerror_msgs);
 
 	for (i=0; i<glen; i++) {
 		if (hwerrs & ipath_generic_hwerror_msgs[i].mask) {
@@ -356,9 +356,10 @@ static void handle_e_ibstatuschanged(struct ipath_devdata *dd,
 			dd->ipath_cregs->cr_iblinkerrrecovcnt);
 		if (linkrecov != dd->ipath_lastlinkrecov) {
 			ipath_dbg("IB linkrecov up %Lx (%s %s) recov %Lu\n",
-				ibcs, ib_linkstate(dd, ibcs),
+				(unsigned long long) ibcs,
+				ib_linkstate(dd, ibcs),
 				ipath_ibcstatus_str[ltstate],
-				linkrecov);
+				(unsigned long long) linkrecov);
 			/* and no more until active again */
 			dd->ipath_lastlinkrecov = 0;
 			ipath_set_linkstate(dd, IPATH_IB_LINKDOWN);
@@ -754,7 +755,7 @@ static int handle_errors(struct ipath_devdata *dd, ipath_err_t errs)
 
 	/* likely due to cancel; so suppress message unless verbose */
 	if ((errs & (INFINIPATH_E_SPKTLEN | INFINIPATH_E_SPIOARMLAUNCH)) &&
-		dd->ipath_lastcancel > jiffies) {
+		time_after(dd->ipath_lastcancel, jiffies)) {
 		/* armlaunch takes precedence; it often causes both. */
 		ipath_cdbg(VERBOSE,
 			"Suppressed %s error (%llx) after sendbuf cancel\n",
@@ -1118,9 +1119,11 @@ irqreturn_t ipath_intr(int irq, void *data)
 	if (unlikely(istat & ~dd->ipath_i_bitsextant))
 		ipath_dev_err(dd,
 			      "interrupt with unknown interrupts %Lx set\n",
+			      (unsigned long long)
 			      istat & ~dd->ipath_i_bitsextant);
 	else if (istat & ~INFINIPATH_I_ERROR) /* errors do own printing */
-		ipath_cdbg(VERBOSE, "intr stat=0x%Lx\n", istat);
+		ipath_cdbg(VERBOSE, "intr stat=0x%Lx\n",
+			(unsigned long long) istat);
 
 	if (istat & INFINIPATH_I_ERROR) {
 		ipath_stats.sps_errints++;
@@ -1128,7 +1131,8 @@ irqreturn_t ipath_intr(int irq, void *data)
 					  dd->ipath_kregs->kr_errorstatus);
 		if (!estat)
 			dev_info(&dd->pcidev->dev, "error interrupt (%Lx), "
-				 "but no error bits set!\n", istat);
+				 "but no error bits set!\n",
+				 (unsigned long long) istat);
 		else if (estat == -1LL)
 			/*
 			 * should we try clearing all, or hope next read
diff --git a/drivers/infiniband/hw/ipath/ipath_kernel.h b/drivers/infiniband/hw/ipath/ipath_kernel.h
index 0bd8bcb184a..6559af60bff 100644
--- a/drivers/infiniband/hw/ipath/ipath_kernel.h
+++ b/drivers/infiniband/hw/ipath/ipath_kernel.h
@@ -355,6 +355,19 @@ struct ipath_devdata {
 	/* errors masked because they occur too fast */
 	ipath_err_t ipath_maskederrs;
 	u64 ipath_lastlinkrecov; /* link recoveries at last ACTIVE */
+	/* these 5 fields are used to establish deltas for IB Symbol
+	 * errors and linkrecovery errors. They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+
 	/* time in jiffies at which to re-enable maskederrs */
 	unsigned long ipath_unmasktime;
 	/* count of egrfull errors, combined for all ports */
@@ -464,6 +477,8 @@ struct ipath_devdata {
 	spinlock_t ipath_kernel_tid_lock;
 	spinlock_t ipath_user_tid_lock;
 	spinlock_t ipath_sendctrl_lock;
+	/* around ipath_pd and (user ports) port_cnt use (intr vs free) */
+	spinlock_t ipath_uctxt_lock;
 
 	/*
 	 * IPATH_STATUS_*,
@@ -1015,8 +1030,6 @@ void ipath_free_data(struct ipath_portdata *dd);
 u32 __iomem *ipath_getpiobuf(struct ipath_devdata *, u32, u32 *);
 void ipath_chg_pioavailkernel(struct ipath_devdata *dd, unsigned start,
 				unsigned len, int avail);
-void ipath_init_iba7220_funcs(struct ipath_devdata *);
-void ipath_init_iba6120_funcs(struct ipath_devdata *);
 void ipath_init_iba6110_funcs(struct ipath_devdata *);
 void ipath_get_eeprom_info(struct ipath_devdata *);
 int ipath_update_eeprom_log(struct ipath_devdata *dd);
@@ -1271,7 +1284,7 @@ struct device_driver;
 
 extern const char ib_ipath_version[];
 
-extern struct attribute_group *ipath_driver_attr_groups[];
+extern const struct attribute_group *ipath_driver_attr_groups[];
 
 int ipath_device_create_group(struct device *, struct ipath_devdata *);
 void ipath_device_remove_group(struct device *, struct ipath_devdata *);
diff --git a/drivers/infiniband/hw/ipath/ipath_keys.c b/drivers/infiniband/hw/ipath/ipath_keys.c
index 8f32b17a5ee..c0e933fec21 100644
--- a/drivers/infiniband/hw/ipath/ipath_keys.c
+++ b/drivers/infiniband/hw/ipath/ipath_keys.c
@@ -132,6 +132,7 @@ int ipath_lkey_ok(struct ipath_qp *qp, struct ipath_sge *isge,
 	 * (see ipath_get_dma_mr and ipath_dma.c).
 	 */
 	if (sge->lkey == 0) {
+		/* always a kernel port, no locking needed */
 		struct ipath_pd *pd = to_ipd(qp->ibqp.pd);
 
 		if (pd->user) {
@@ -211,6 +212,7 @@ int ipath_rkey_ok(struct ipath_qp *qp, struct ipath_sge_state *ss,
 	 * (see ipath_get_dma_mr and ipath_dma.c).
 	 */
 	if (rkey == 0) {
+		/* always a kernel port, no locking needed */
 		struct ipath_pd *pd = to_ipd(qp->ibqp.pd);
 
 		if (pd->user) {
diff --git a/drivers/infiniband/hw/ipath/ipath_mad.c b/drivers/infiniband/hw/ipath/ipath_mad.c
index be4fc9ada8e..43f2d0424d4 100644
--- a/drivers/infiniband/hw/ipath/ipath_mad.c
+++ b/drivers/infiniband/hw/ipath/ipath_mad.c
@@ -32,15 +32,16 @@
  */
 
 #include <rdma/ib_smi.h>
+#include <rdma/ib_pma.h>
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
 #include "ipath_common.h"
 
-#define IB_SMP_UNSUP_VERSION	__constant_htons(0x0004)
-#define IB_SMP_UNSUP_METHOD	__constant_htons(0x0008)
-#define IB_SMP_UNSUP_METH_ATTR	__constant_htons(0x000C)
-#define IB_SMP_INVALID_FIELD	__constant_htons(0x001C)
+#define IB_SMP_UNSUP_VERSION	cpu_to_be16(0x0004)
+#define IB_SMP_UNSUP_METHOD	cpu_to_be16(0x0008)
+#define IB_SMP_UNSUP_METH_ATTR	cpu_to_be16(0x000C)
+#define IB_SMP_INVALID_FIELD	cpu_to_be16(0x001C)
 
 static int reply(struct ib_smp *smp)
 {
@@ -60,7 +61,7 @@ static int recv_subn_get_nodedescription(struct ib_smp *smp,
 	if (smp->attr_mod)
 		smp->status |= IB_SMP_INVALID_FIELD;
 
-	strncpy(smp->data, ibdev->node_desc, sizeof(smp->data));
+	memcpy(smp->data, ibdev->node_desc, sizeof(smp->data));
 
 	return reply(smp);
 }
@@ -348,6 +349,7 @@ bail:
  */
 static int get_pkeys(struct ipath_devdata *dd, u16 * pkeys)
 {
+	/* always a kernel port, no locking needed */
 	struct ipath_portdata *pd = dd->ipath_pd[0];
 
 	memcpy(pkeys, pd->port_pkeys, sizeof(pd->port_pkeys));
@@ -730,6 +732,7 @@ static int set_pkeys(struct ipath_devdata *dd, u16 *pkeys)
 	int i;
 	int changed = 0;
 
+	/* always a kernel port, no locking needed */
 	pd = dd->ipath_pd[0];
 
 	for (i = 0; i < ARRAY_SIZE(pd->port_pkeys); i++) {
@@ -787,151 +790,18 @@ static int recv_subn_set_pkeytable(struct ib_smp *smp,
 	return recv_subn_get_pkeytable(smp, ibdev);
 }
 
-#define IB_PMA_CLASS_PORT_INFO		__constant_htons(0x0001)
-#define IB_PMA_PORT_SAMPLES_CONTROL	__constant_htons(0x0010)
-#define IB_PMA_PORT_SAMPLES_RESULT	__constant_htons(0x0011)
-#define IB_PMA_PORT_COUNTERS		__constant_htons(0x0012)
-#define IB_PMA_PORT_COUNTERS_EXT	__constant_htons(0x001D)
-#define IB_PMA_PORT_SAMPLES_RESULT_EXT	__constant_htons(0x001E)
-
-struct ib_perf {
-	u8 base_version;
-	u8 mgmt_class;
-	u8 class_version;
-	u8 method;
-	__be16 status;
-	__be16 unused;
-	__be64 tid;
-	__be16 attr_id;
-	__be16 resv;
-	__be32 attr_mod;
-	u8 reserved[40];
-	u8 data[192];
-} __attribute__ ((packed));
-
-struct ib_pma_classportinfo {
-	u8 base_version;
-	u8 class_version;
-	__be16 cap_mask;
-	u8 reserved[3];
-	u8 resp_time_value;	/* only lower 5 bits */
-	union ib_gid redirect_gid;
-	__be32 redirect_tc_sl_fl;	/* 8, 4, 20 bits respectively */
-	__be16 redirect_lid;
-	__be16 redirect_pkey;
-	__be32 redirect_qp;	/* only lower 24 bits */
-	__be32 redirect_qkey;
-	union ib_gid trap_gid;
-	__be32 trap_tc_sl_fl;	/* 8, 4, 20 bits respectively */
-	__be16 trap_lid;
-	__be16 trap_pkey;
-	__be32 trap_hl_qp;	/* 8, 24 bits respectively */
-	__be32 trap_qkey;
-} __attribute__ ((packed));
-
-struct ib_pma_portsamplescontrol {
-	u8 opcode;
-	u8 port_select;
-	u8 tick;
-	u8 counter_width;	/* only lower 3 bits */
-	__be32 counter_mask0_9;	/* 2, 10 * 3, bits */
-	__be16 counter_mask10_14;	/* 1, 5 * 3, bits */
-	u8 sample_mechanisms;
-	u8 sample_status;	/* only lower 2 bits */
-	__be64 option_mask;
-	__be64 vendor_mask;
-	__be32 sample_start;
-	__be32 sample_interval;
-	__be16 tag;
-	__be16 counter_select[15];
-} __attribute__ ((packed));
-
-struct ib_pma_portsamplesresult {
-	__be16 tag;
-	__be16 sample_status;	/* only lower 2 bits */
-	__be32 counter[15];
-} __attribute__ ((packed));
-
-struct ib_pma_portsamplesresult_ext {
-	__be16 tag;
-	__be16 sample_status;	/* only lower 2 bits */
-	__be32 extended_width;	/* only upper 2 bits */
-	__be64 counter[15];
-} __attribute__ ((packed));
-
-struct ib_pma_portcounters {
-	u8 reserved;
-	u8 port_select;
-	__be16 counter_select;
-	__be16 symbol_error_counter;
-	u8 link_error_recovery_counter;
-	u8 link_downed_counter;
-	__be16 port_rcv_errors;
-	__be16 port_rcv_remphys_errors;
-	__be16 port_rcv_switch_relay_errors;
-	__be16 port_xmit_discards;
-	u8 port_xmit_constraint_errors;
-	u8 port_rcv_constraint_errors;
-	u8 reserved1;
-	u8 lli_ebor_errors;	/* 4, 4, bits */
-	__be16 reserved2;
-	__be16 vl15_dropped;
-	__be32 port_xmit_data;
-	__be32 port_rcv_data;
-	__be32 port_xmit_packets;
-	__be32 port_rcv_packets;
-} __attribute__ ((packed));
-
-#define IB_PMA_SEL_SYMBOL_ERROR			__constant_htons(0x0001)
-#define IB_PMA_SEL_LINK_ERROR_RECOVERY		__constant_htons(0x0002)
-#define IB_PMA_SEL_LINK_DOWNED			__constant_htons(0x0004)
-#define IB_PMA_SEL_PORT_RCV_ERRORS		__constant_htons(0x0008)
-#define IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS	__constant_htons(0x0010)
-#define IB_PMA_SEL_PORT_XMIT_DISCARDS		__constant_htons(0x0040)
-#define IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS	__constant_htons(0x0200)
-#define IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS	__constant_htons(0x0400)
-#define IB_PMA_SEL_PORT_VL15_DROPPED		__constant_htons(0x0800)
-#define IB_PMA_SEL_PORT_XMIT_DATA		__constant_htons(0x1000)
-#define IB_PMA_SEL_PORT_RCV_DATA		__constant_htons(0x2000)
-#define IB_PMA_SEL_PORT_XMIT_PACKETS		__constant_htons(0x4000)
-#define IB_PMA_SEL_PORT_RCV_PACKETS		__constant_htons(0x8000)
-
-struct ib_pma_portcounters_ext {
-	u8 reserved;
-	u8 port_select;
-	__be16 counter_select;
-	__be32 reserved1;
-	__be64 port_xmit_data;
-	__be64 port_rcv_data;
-	__be64 port_xmit_packets;
-	__be64 port_rcv_packets;
-	__be64 port_unicast_xmit_packets;
-	__be64 port_unicast_rcv_packets;
-	__be64 port_multicast_xmit_packets;
-	__be64 port_multicast_rcv_packets;
-} __attribute__ ((packed));
-
-#define IB_PMA_SELX_PORT_XMIT_DATA		__constant_htons(0x0001)
-#define IB_PMA_SELX_PORT_RCV_DATA		__constant_htons(0x0002)
-#define IB_PMA_SELX_PORT_XMIT_PACKETS		__constant_htons(0x0004)
-#define IB_PMA_SELX_PORT_RCV_PACKETS		__constant_htons(0x0008)
-#define IB_PMA_SELX_PORT_UNI_XMIT_PACKETS	__constant_htons(0x0010)
-#define IB_PMA_SELX_PORT_UNI_RCV_PACKETS	__constant_htons(0x0020)
-#define IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS	__constant_htons(0x0040)
-#define IB_PMA_SELX_PORT_MULTI_RCV_PACKETS	__constant_htons(0x0080)
-
-static int recv_pma_get_classportinfo(struct ib_perf *pmp)
+static int recv_pma_get_classportinfo(struct ib_pma_mad *pmp)
 {
-	struct ib_pma_classportinfo *p =
-		(struct ib_pma_classportinfo *)pmp->data;
+	struct ib_class_port_info *p =
+		(struct ib_class_port_info *)pmp->data;
 
 	memset(pmp->data, 0, sizeof(pmp->data));
 
-	if (pmp->attr_mod != 0)
-		pmp->status |= IB_SMP_INVALID_FIELD;
+	if (pmp->mad_hdr.attr_mod != 0)
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 
 	/* Indicate AllPortSelect is valid (only one port anyway) */
-	p->cap_mask = __constant_cpu_to_be16(1 << 8);
+	p->capability_mask = cpu_to_be16(1 << 8);
 	p->base_version = 1;
 	p->class_version = 1;
 	/*
@@ -949,14 +819,13 @@ static int recv_pma_get_classportinfo(struct ib_perf *pmp)
  * We support 5 counters which only count the mandatory quantities.
  */
 #define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
-#define COUNTER_MASK0_9 \
-	__constant_cpu_to_be32(COUNTER_MASK(1, 0) | \
-			       COUNTER_MASK(1, 1) | \
-			       COUNTER_MASK(1, 2) | \
-			       COUNTER_MASK(1, 3) | \
-			       COUNTER_MASK(1, 4))
-
-static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
+#define COUNTER_MASK0_9 cpu_to_be32(COUNTER_MASK(1, 0) | \
+				    COUNTER_MASK(1, 1) | \
+				    COUNTER_MASK(1, 2) | \
+				    COUNTER_MASK(1, 3) | \
+				    COUNTER_MASK(1, 4))
+
+static int recv_pma_get_portsamplescontrol(struct ib_pma_mad *pmp,
 					   struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portsamplescontrol *p =
@@ -969,9 +838,9 @@ static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 	memset(pmp->data, 0, sizeof(pmp->data));
 
 	p->port_select = port_select;
-	if (pmp->attr_mod != 0 ||
+	if (pmp->mad_hdr.attr_mod != 0 ||
 	    (port_select != port && port_select != 0xFF))
-		pmp->status |= IB_SMP_INVALID_FIELD;
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 	/*
 	 * Ticks are 10x the link transfer period which for 2.5Gbs is 4
 	 * nsec.  0 == 4 nsec., 1 == 8 nsec., ..., 255 == 1020 nsec.  Sample
@@ -1005,7 +874,7 @@ static int recv_pma_get_portsamplescontrol(struct ib_perf *pmp,
 	return reply((struct ib_smp *) pmp);
 }
 
-static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
+static int recv_pma_set_portsamplescontrol(struct ib_pma_mad *pmp,
 					   struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portsamplescontrol *p =
@@ -1016,9 +885,9 @@ static int recv_pma_set_portsamplescontrol(struct ib_perf *pmp,
 	u8 status;
 	int ret;
 
-	if (pmp->attr_mod != 0 ||
+	if (pmp->mad_hdr.attr_mod != 0 ||
 	    (p->port_select != port && p->port_select != 0xFF)) {
-		pmp->status |= IB_SMP_INVALID_FIELD;
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 		ret = reply((struct ib_smp *) pmp);
 		goto bail;
 	}
@@ -1092,7 +961,7 @@ static u64 get_counter(struct ipath_ibdev *dev,
 	return ret;
 }
 
-static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
+static int recv_pma_get_portsamplesresult(struct ib_pma_mad *pmp,
 					  struct ib_device *ibdev)
 {
 	struct ib_pma_portsamplesresult *p =
@@ -1117,7 +986,7 @@ static int recv_pma_get_portsamplesresult(struct ib_perf *pmp,
 	return reply((struct ib_smp *) pmp);
 }
 
-static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
+static int recv_pma_get_portsamplesresult_ext(struct ib_pma_mad *pmp,
 					      struct ib_device *ibdev)
 {
 	struct ib_pma_portsamplesresult_ext *p =
@@ -1135,7 +1004,7 @@ static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
 		status = dev->pma_sample_status;
 	p->sample_status = cpu_to_be16(status);
 	/* 64 bits */
-	p->extended_width = __constant_cpu_to_be32(0x80000000);
+	p->extended_width = cpu_to_be32(0x80000000);
 	for (i = 0; i < ARRAY_SIZE(dev->pma_counter_select); i++)
 		p->counter[i] = (status != IB_PMA_SAMPLE_STATUS_DONE) ? 0 :
 		    cpu_to_be64(
@@ -1144,7 +1013,7 @@ static int recv_pma_get_portsamplesresult_ext(struct ib_perf *pmp,
 	return reply((struct ib_smp *) pmp);
 }
 
-static int recv_pma_get_portcounters(struct ib_perf *pmp,
+static int recv_pma_get_portcounters(struct ib_pma_mad *pmp,
 				     struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
@@ -1178,12 +1047,12 @@ static int recv_pma_get_portcounters(struct ib_perf *pmp,
 	memset(pmp->data, 0, sizeof(pmp->data));
 
 	p->port_select = port_select;
-	if (pmp->attr_mod != 0 ||
+	if (pmp->mad_hdr.attr_mod != 0 ||
 	    (port_select != port && port_select != 0xFF))
-		pmp->status |= IB_SMP_INVALID_FIELD;
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 
 	if (cntrs.symbol_error_counter > 0xFFFFUL)
-		p->symbol_error_counter = __constant_cpu_to_be16(0xFFFF);
+		p->symbol_error_counter = cpu_to_be16(0xFFFF);
 	else
 		p->symbol_error_counter =
 			cpu_to_be16((u16)cntrs.symbol_error_counter);
@@ -1197,17 +1066,17 @@ static int recv_pma_get_portcounters(struct ib_perf *pmp,
 	else
 		p->link_downed_counter = (u8)cntrs.link_downed_counter;
 	if (cntrs.port_rcv_errors > 0xFFFFUL)
-		p->port_rcv_errors = __constant_cpu_to_be16(0xFFFF);
+		p->port_rcv_errors = cpu_to_be16(0xFFFF);
 	else
 		p->port_rcv_errors =
 			cpu_to_be16((u16) cntrs.port_rcv_errors);
 	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
-		p->port_rcv_remphys_errors = __constant_cpu_to_be16(0xFFFF);
+		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
 	else
 		p->port_rcv_remphys_errors =
 			cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
 	if (cntrs.port_xmit_discards > 0xFFFFUL)
-		p->port_xmit_discards = __constant_cpu_to_be16(0xFFFF);
+		p->port_xmit_discards = cpu_to_be16(0xFFFF);
 	else
 		p->port_xmit_discards =
 			cpu_to_be16((u16)cntrs.port_xmit_discards);
@@ -1215,27 +1084,27 @@ static int recv_pma_get_portcounters(struct ib_perf *pmp,
 		cntrs.local_link_integrity_errors = 0xFUL;
 	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
 		cntrs.excessive_buffer_overrun_errors = 0xFUL;
-	p->lli_ebor_errors = (cntrs.local_link_integrity_errors << 4) |
+	p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
 		cntrs.excessive_buffer_overrun_errors;
 	if (cntrs.vl15_dropped > 0xFFFFUL)
-		p->vl15_dropped = __constant_cpu_to_be16(0xFFFF);
+		p->vl15_dropped = cpu_to_be16(0xFFFF);
 	else
 		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
 	if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
-		p->port_xmit_data = __constant_cpu_to_be32(0xFFFFFFFF);
+		p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
 	else
 		p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
 	if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
-		p->port_rcv_data = __constant_cpu_to_be32(0xFFFFFFFF);
+		p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
 	else
 		p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
 	if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
-		p->port_xmit_packets = __constant_cpu_to_be32(0xFFFFFFFF);
+		p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
 	else
 		p->port_xmit_packets =
 			cpu_to_be32((u32)cntrs.port_xmit_packets);
 	if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
-		p->port_rcv_packets = __constant_cpu_to_be32(0xFFFFFFFF);
+		p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
 	else
 		p->port_rcv_packets =
 			cpu_to_be32((u32) cntrs.port_rcv_packets);
@@ -1243,7 +1112,7 @@ static int recv_pma_get_portcounters(struct ib_perf *pmp,
 	return reply((struct ib_smp *) pmp);
 }
 
-static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
+static int recv_pma_get_portcounters_ext(struct ib_pma_mad *pmp,
 					 struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portcounters_ext *p =
@@ -1264,9 +1133,9 @@ static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
 	memset(pmp->data, 0, sizeof(pmp->data));
 
 	p->port_select = port_select;
-	if (pmp->attr_mod != 0 ||
+	if (pmp->mad_hdr.attr_mod != 0 ||
 	    (port_select != port && port_select != 0xFF))
-		pmp->status |= IB_SMP_INVALID_FIELD;
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
 
 	p->port_xmit_data = cpu_to_be64(swords);
 	p->port_rcv_data = cpu_to_be64(rwords);
@@ -1280,7 +1149,7 @@ static int recv_pma_get_portcounters_ext(struct ib_perf *pmp,
 	return reply((struct ib_smp *) pmp);
 }
 
-static int recv_pma_set_portcounters(struct ib_perf *pmp,
+static int recv_pma_set_portcounters(struct ib_pma_mad *pmp,
 				     struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
@@ -1343,7 +1212,7 @@ static int recv_pma_set_portcounters(struct ib_perf *pmp,
 	return recv_pma_get_portcounters(pmp, ibdev, port);
 }
 
-static int recv_pma_set_portcounters_ext(struct ib_perf *pmp,
+static int recv_pma_set_portcounters_ext(struct ib_pma_mad *pmp,
 					 struct ib_device *ibdev, u8 port)
 {
 	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
@@ -1517,19 +1386,19 @@ static int process_perf(struct ib_device *ibdev, u8 port_num,
 			struct ib_mad *in_mad,
 			struct ib_mad *out_mad)
 {
-	struct ib_perf *pmp = (struct ib_perf *)out_mad;
+	struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
 	int ret;
 
 	*out_mad = *in_mad;
-	if (pmp->class_version != 1) {
-		pmp->status |= IB_SMP_UNSUP_VERSION;
+	if (pmp->mad_hdr.class_version != 1) {
+		pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
 		ret = reply((struct ib_smp *) pmp);
 		goto bail;
 	}
 
-	switch (pmp->method) {
+	switch (pmp->mad_hdr.method) {
 	case IB_MGMT_METHOD_GET:
-		switch (pmp->attr_id) {
+		switch (pmp->mad_hdr.attr_id) {
 		case IB_PMA_CLASS_PORT_INFO:
 			ret = recv_pma_get_classportinfo(pmp);
 			goto bail;
@@ -1553,13 +1422,13 @@ static int process_perf(struct ib_device *ibdev, u8 port_num,
 							    port_num);
 			goto bail;
 		default:
-			pmp->status |= IB_SMP_UNSUP_METH_ATTR;
+			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
 			ret = reply((struct ib_smp *) pmp);
 			goto bail;
 		}
 
 	case IB_MGMT_METHOD_SET:
-		switch (pmp->attr_id) {
+		switch (pmp->mad_hdr.attr_id) {
 		case IB_PMA_PORT_SAMPLES_CONTROL:
 			ret = recv_pma_set_portsamplescontrol(pmp, ibdev,
 							      port_num);
@@ -1573,7 +1442,7 @@ static int process_perf(struct ib_device *ibdev, u8 port_num,
 							    port_num);
 			goto bail;
 		default:
-			pmp->status |= IB_SMP_UNSUP_METH_ATTR;
+			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
 			ret = reply((struct ib_smp *) pmp);
 			goto bail;
 		}
@@ -1587,7 +1456,7 @@ static int process_perf(struct ib_device *ibdev, u8 port_num,
 		ret = IB_MAD_RESULT_SUCCESS;
 		goto bail;
 	default:
-		pmp->status |= IB_SMP_UNSUP_METHOD;
+		pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
 		ret = reply((struct ib_smp *) pmp);
 	}
 
diff --git a/drivers/infiniband/hw/ipath/ipath_mmap.c b/drivers/infiniband/hw/ipath/ipath_mmap.c
index fa830e22002..e7327422940 100644
--- a/drivers/infiniband/hw/ipath/ipath_mmap.c
+++ b/drivers/infiniband/hw/ipath/ipath_mmap.c
@@ -32,6 +32,7 @@
 
 #include <linux/module.h>
 #include <linux/vmalloc.h>
+#include <linux/slab.h>
 #include <linux/mm.h>
 #include <linux/errno.h>
 #include <asm/pgtable.h>
@@ -74,7 +75,7 @@ static void ipath_vma_close(struct vm_area_struct *vma)
 	kref_put(&ip->ref, ipath_release_mmap_info);
 }
 
-static struct vm_operations_struct ipath_vm_ops = {
+static const struct vm_operations_struct ipath_vm_ops = {
 	.open =     ipath_vma_open,
 	.close =    ipath_vma_close,
 };
diff --git a/drivers/infiniband/hw/ipath/ipath_mr.c b/drivers/infiniband/hw/ipath/ipath_mr.c
index 9d343b7c2f3..5e61e9bff69 100644
--- a/drivers/infiniband/hw/ipath/ipath_mr.c
+++ b/drivers/infiniband/hw/ipath/ipath_mr.c
@@ -31,6 +31,8 @@
  * SOFTWARE.
  */
 
+#include <linux/slab.h>
+
 #include <rdma/ib_umem.h>
 #include <rdma/ib_pack.h>
 #include <rdma/ib_smi.h>
@@ -186,8 +188,8 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 {
 	struct ipath_mr *mr;
 	struct ib_umem *umem;
-	struct ib_umem_chunk *chunk;
-	int n, m, i;
+	int n, m, entry;
+	struct scatterlist *sg;
 	struct ib_mr *ret;
 
 	if (length == 0) {
@@ -200,10 +202,7 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	if (IS_ERR(umem))
 		return (void *) umem;
 
-	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		n += chunk->nents;
-
+	n = umem->nmap;
 	mr = alloc_mr(n, &to_idev(pd->device)->lk_table);
 	if (!mr) {
 		ret = ERR_PTR(-ENOMEM);
@@ -222,22 +221,20 @@ struct ib_mr *ipath_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	m = 0;
 	n = 0;
-	list_for_each_entry(chunk, &umem->chunk_list, list) {
-		for (i = 0; i < chunk->nents; i++) {
-			void *vaddr;
-
-			vaddr = page_address(sg_page(&chunk->page_list[i]));
-			if (!vaddr) {
-				ret = ERR_PTR(-EINVAL);
-				goto bail;
-			}
-			mr->mr.map[m]->segs[n].vaddr = vaddr;
-			mr->mr.map[m]->segs[n].length = umem->page_size;
-			n++;
-			if (n == IPATH_SEGSZ) {
-				m++;
-				n = 0;
-			}
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		void *vaddr;
+
+		vaddr = page_address(sg_page(sg));
+		if (!vaddr) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+		mr->mr.map[m]->segs[n].vaddr = vaddr;
+		mr->mr.map[m]->segs[n].length = umem->page_size;
+		n++;
+		if (n == IPATH_SEGSZ) {
+			m++;
+			n = 0;
 		}
 	}
 	ret = &mr->ibmr;
diff --git a/drivers/infiniband/hw/ipath/ipath_qp.c b/drivers/infiniband/hw/ipath/ipath_qp.c
index 4715911101e..face87602dc 100644
--- a/drivers/infiniband/hw/ipath/ipath_qp.c
+++ b/drivers/infiniband/hw/ipath/ipath_qp.c
@@ -32,6 +32,8 @@
  */
 
 #include <linux/err.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
 #include <linux/vmalloc.h>
 
 #include "ipath_verbs.h"
@@ -461,7 +463,7 @@ int ipath_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
 
 	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
-				attr_mask))
+				attr_mask, IB_LINK_LAYER_UNSPECIFIED))
 		goto inval;
 
 	if (attr_mask & IB_QP_AV) {
@@ -745,6 +747,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 	struct ipath_swqe *swq = NULL;
 	struct ipath_ibdev *dev;
 	size_t sz;
+	size_t sg_list_sz;
 	struct ib_qp *ret;
 
 	if (init_attr->create_flags) {
@@ -789,19 +792,31 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 			goto bail;
 		}
 		sz = sizeof(*qp);
+		sg_list_sz = 0;
 		if (init_attr->srq) {
 			struct ipath_srq *srq = to_isrq(init_attr->srq);
 
-			sz += sizeof(*qp->r_sg_list) *
-				srq->rq.max_sge;
-		} else
-			sz += sizeof(*qp->r_sg_list) *
-				init_attr->cap.max_recv_sge;
-		qp = kmalloc(sz, GFP_KERNEL);
+			if (srq->rq.max_sge > 1)
+				sg_list_sz = sizeof(*qp->r_sg_list) *
+					(srq->rq.max_sge - 1);
+		} else if (init_attr->cap.max_recv_sge > 1)
+			sg_list_sz = sizeof(*qp->r_sg_list) *
+				(init_attr->cap.max_recv_sge - 1);
+		qp = kmalloc(sz + sg_list_sz, GFP_KERNEL);
 		if (!qp) {
 			ret = ERR_PTR(-ENOMEM);
 			goto bail_swq;
 		}
+		if (sg_list_sz && (init_attr->qp_type == IB_QPT_UD ||
+		    init_attr->qp_type == IB_QPT_SMI ||
+		    init_attr->qp_type == IB_QPT_GSI)) {
+			qp->r_ud_sg_list = kmalloc(sg_list_sz, GFP_KERNEL);
+			if (!qp->r_ud_sg_list) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_qp;
+			}
+		} else
+			qp->r_ud_sg_list = NULL;
 		if (init_attr->srq) {
 			sz = 0;
 			qp->r_rq.size = 0;
@@ -818,7 +833,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 					      qp->r_rq.size * sz);
 			if (!qp->r_rq.wq) {
 				ret = ERR_PTR(-ENOMEM);
-				goto bail_qp;
+				goto bail_sg_list;
 			}
 		}
 
@@ -848,7 +863,7 @@ struct ib_qp *ipath_create_qp(struct ib_pd *ibpd,
 		if (err) {
 			ret = ERR_PTR(err);
 			vfree(qp->r_rq.wq);
-			goto bail_qp;
+			goto bail_sg_list;
 		}
 		qp->ip = NULL;
 		qp->s_tx = NULL;
@@ -925,6 +940,8 @@ bail_ip:
 		vfree(qp->r_rq.wq);
 	ipath_free_qp(&dev->qp_table, qp);
 	free_qpn(&dev->qp_table, qp->ibqp.qp_num);
+bail_sg_list:
+	kfree(qp->r_ud_sg_list);
 bail_qp:
 	kfree(qp);
 bail_swq:
@@ -989,6 +1006,7 @@ int ipath_destroy_qp(struct ib_qp *ibqp)
 		kref_put(&qp->ip->ref, ipath_release_mmap_info);
 	else
 		vfree(qp->r_rq.wq);
+	kfree(qp->r_ud_sg_list);
 	vfree(qp->s_wq);
 	kfree(qp);
 	return 0;
diff --git a/drivers/infiniband/hw/ipath/ipath_rc.c b/drivers/infiniband/hw/ipath/ipath_rc.c
index 97710522624..79b3dbc9717 100644
--- a/drivers/infiniband/hw/ipath/ipath_rc.c
+++ b/drivers/infiniband/hw/ipath/ipath_rc.c
@@ -573,9 +573,8 @@ int ipath_make_rc_req(struct ipath_qp *qp)
 		ohdr->u.rc.reth.length = cpu_to_be32(qp->s_len);
 		qp->s_state = OP(RDMA_READ_REQUEST);
 		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
-		bth2 = qp->s_psn++ & IPATH_PSN_MASK;
-		if (ipath_cmp24(qp->s_psn, qp->s_next_psn) > 0)
-			qp->s_next_psn = qp->s_psn;
+		bth2 = qp->s_psn & IPATH_PSN_MASK;
+		qp->s_psn = wqe->lpsn + 1;
 		ss = NULL;
 		len = 0;
 		qp->s_cur++;
@@ -675,7 +674,8 @@ static void send_rc_ack(struct ipath_qp *qp)
 	hdr.lrh[0] = cpu_to_be16(lrh0);
 	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
-	hdr.lrh[3] = cpu_to_be16(dd->ipath_lid);
+	hdr.lrh[3] = cpu_to_be16(dd->ipath_lid |
+				 qp->remote_ah_attr.src_path_bits);
 	ohdr->bth[0] = cpu_to_be32(bth0);
 	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & IPATH_PSN_MASK);
@@ -1744,7 +1744,7 @@ void ipath_rc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		/* Signal completion event if the solicited bit is set. */
 		ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 			       (ohdr->bth[0] &
-				__constant_cpu_to_be32(1 << 23)) != 0);
+				cpu_to_be32(1 << 23)) != 0);
 		break;
 
 	case OP(RDMA_WRITE_FIRST):
diff --git a/drivers/infiniband/hw/ipath/ipath_ruc.c b/drivers/infiniband/hw/ipath/ipath_ruc.c
index af051f75766..1f95bbaf760 100644
--- a/drivers/infiniband/hw/ipath/ipath_ruc.c
+++ b/drivers/infiniband/hw/ipath/ipath_ruc.c
@@ -31,6 +31,7 @@
  * SOFTWARE.
  */
 
+#include <linux/sched.h>
 #include <linux/spinlock.h>
 
 #include "ipath_verbs.h"
@@ -156,7 +157,7 @@ bail:
 /**
  * ipath_get_rwqe - copy the next RWQE into the QP's RWQE
  * @qp: the QP
- * @wr_id_only: update wr_id only, not SGEs
+ * @wr_id_only: update qp->r_wr_id only, not qp->r_sge
  *
  * Return 0 if no RWQE is available, otherwise return 1.
  *
@@ -173,8 +174,6 @@ int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
 	u32 tail;
 	int ret;
 
-	qp->r_sge.sg_list = qp->r_sg_list;
-
 	if (qp->ibqp.srq) {
 		srq = to_isrq(qp->ibqp.srq);
 		handler = srq->ibsrq.event_handler;
@@ -206,8 +205,10 @@ int ipath_get_rwqe(struct ipath_qp *qp, int wr_id_only)
 		wqe = get_rwqe_ptr(rq, tail);
 		if (++tail >= rq->size)
 			tail = 0;
-	} while (!wr_id_only && !ipath_init_sge(qp, wqe, &qp->r_len,
-						&qp->r_sge));
+		if (wr_id_only)
+			break;
+		qp->r_sge.sg_list = qp->r_sg_list;
+	} while (!ipath_init_sge(qp, wqe, &qp->r_len, &qp->r_sge));
 	qp->r_wr_id = wqe->wr_id;
 	wq->tail = tail;
 
@@ -618,7 +619,8 @@ void ipath_make_ruc_header(struct ipath_ibdev *dev, struct ipath_qp *qp,
 	qp->s_hdr.lrh[0] = cpu_to_be16(lrh0);
 	qp->s_hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
 	qp->s_hdr.lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
-	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid);
+	qp->s_hdr.lrh[3] = cpu_to_be16(dev->dd->ipath_lid |
+				       qp->remote_ah_attr.src_path_bits);
 	bth0 |= ipath_get_pkey(dev->dd, qp->s_pkey_index);
 	bth0 |= extra_bytes << 20;
 	ohdr->bth[0] = cpu_to_be32(bth0 | (1 << 22));
diff --git a/drivers/infiniband/hw/ipath/ipath_sd7220_img.c b/drivers/infiniband/hw/ipath/ipath_sd7220_img.c
deleted file mode 100644
index 5ef59da9270..00000000000
--- a/drivers/infiniband/hw/ipath/ipath_sd7220_img.c
+++ /dev/null
@@ -1,1082 +0,0 @@
-/*
- * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved.
- *
- * This software is available to you under a choice of one of two
- * licenses.  You may choose to be licensed under the terms of the GNU
- * General Public License (GPL) Version 2, available from the file
- * COPYING in the main directory of this source tree, or the
- * OpenIB.org BSD license below:
- *
- *     Redistribution and use in source and binary forms, with or
- *     without modification, are permitted provided that the following
- *     conditions are met:
- *
- *      - Redistributions of source code must retain the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer.
- *
- *      - Redistributions in binary form must reproduce the above
- *        copyright notice, this list of conditions and the following
- *        disclaimer in the documentation and/or other materials
- *        provided with the distribution.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
- * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
- * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
- * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
- * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
- * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
- * SOFTWARE.
- */
-
-/*
- * This file contains the memory image from the vendor, to be copied into
- * the IB SERDES of the IBA7220 during initialization.
- * The file also includes the two functions which use this image.
- */
-#include <linux/pci.h>
-#include <linux/delay.h>
-
-#include "ipath_kernel.h"
-#include "ipath_registers.h"
-#include "ipath_7220.h"
-
-static unsigned char ipath_sd7220_ib_img[] = {
-/*0000*/0x02, 0x0A, 0x29, 0x02, 0x0A, 0x87, 0xE5, 0xE6,
-	0x30, 0xE6, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F,
-/*0010*/0x00, 0xE5, 0xE2, 0x30, 0xE4, 0x04, 0x7E, 0x01,
-	0x80, 0x02, 0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x08,
-/*0020*/0x53, 0xF9, 0xF7, 0xE4, 0xF5, 0xFE, 0x80, 0x08,
-	0x7F, 0x0A, 0x12, 0x17, 0x31, 0x12, 0x0E, 0xA2,
-/*0030*/0x75, 0xFC, 0x08, 0xE4, 0xF5, 0xFD, 0xE5, 0xE7,
-	0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0x22, 0x00,
-/*0040*/0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x75,
-	0x51, 0x01, 0xE4, 0xF5, 0x52, 0xF5, 0x53, 0xF5,
-/*0050*/0x52, 0xF5, 0x7E, 0x7F, 0x04, 0x02, 0x04, 0x38,
-	0xC2, 0x36, 0x05, 0x52, 0xE5, 0x52, 0xD3, 0x94,
-/*0060*/0x0C, 0x40, 0x05, 0x75, 0x52, 0x01, 0xD2, 0x36,
-	0x90, 0x07, 0x0C, 0x74, 0x07, 0xF0, 0xA3, 0x74,
-/*0070*/0xFF, 0xF0, 0xE4, 0xF5, 0x0C, 0xA3, 0xF0, 0x90,
-	0x07, 0x14, 0xF0, 0xA3, 0xF0, 0x75, 0x0B, 0x20,
-/*0080*/0xF5, 0x09, 0xE4, 0xF5, 0x08, 0xE5, 0x08, 0xD3,
-	0x94, 0x30, 0x40, 0x03, 0x02, 0x04, 0x04, 0x12,
-/*0090*/0x00, 0x06, 0x15, 0x0B, 0xE5, 0x08, 0x70, 0x04,
-	0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x09,
-/*00A0*/0x70, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00,
-	0xEE, 0x5F, 0x60, 0x05, 0x12, 0x18, 0x71, 0xD2,
-/*00B0*/0x35, 0x53, 0xE1, 0xF7, 0xE5, 0x08, 0x45, 0x09,
-	0xFF, 0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24,
-/*00C0*/0x83, 0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83,
-	0xEF, 0xF0, 0x85, 0xE2, 0x20, 0xE5, 0x52, 0xD3,
-/*00D0*/0x94, 0x01, 0x40, 0x0D, 0x12, 0x19, 0xF3, 0xE0,
-	0x54, 0xA0, 0x64, 0x40, 0x70, 0x03, 0x02, 0x03,
-/*00E0*/0xFB, 0x53, 0xF9, 0xF8, 0x90, 0x94, 0x70, 0xE4,
-	0xF0, 0xE0, 0xF5, 0x10, 0xAF, 0x09, 0x12, 0x1E,
-/*00F0*/0xB3, 0xAF, 0x08, 0xEF, 0x44, 0x08, 0xF5, 0x82,
-	0x75, 0x83, 0x80, 0xE0, 0xF5, 0x29, 0xEF, 0x44,
-/*0100*/0x07, 0x12, 0x1A, 0x3C, 0xF5, 0x22, 0x54, 0x40,
-	0xD3, 0x94, 0x00, 0x40, 0x1E, 0xE5, 0x29, 0x54,
-/*0110*/0xF0, 0x70, 0x21, 0x12, 0x19, 0xF3, 0xE0, 0x44,
-	0x80, 0xF0, 0xE5, 0x22, 0x54, 0x30, 0x65, 0x08,
-/*0120*/0x70, 0x09, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xBF,
-	0xF0, 0x80, 0x09, 0x12, 0x19, 0xF3, 0x74, 0x40,
-/*0130*/0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12, 0x75,
-	0x83, 0xAE, 0x74, 0xFF, 0xF0, 0xAF, 0x08, 0x7E,
-/*0140*/0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0xE0, 0xFD,
-	0xE5, 0x0B, 0x25, 0xE0, 0x25, 0xE0, 0x24, 0x81,
-/*0150*/0xF5, 0x82, 0xE4, 0x34, 0x07, 0xF5, 0x83, 0xED,
-	0xF0, 0x90, 0x07, 0x0E, 0xE0, 0x04, 0xF0, 0xEF,
-/*0160*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0x98, 0xE0,
-	0xF5, 0x28, 0x12, 0x1A, 0x23, 0x40, 0x0C, 0x12,
-/*0170*/0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12, 0x1A, 0x32,
-	0x02, 0x03, 0xF6, 0xAF, 0x08, 0x7E, 0x00, 0x74,
-/*0180*/0x80, 0xCD, 0xEF, 0xCD, 0x8D, 0x82, 0xF5, 0x83,
-	0xE0, 0x30, 0xE0, 0x0A, 0x12, 0x19, 0xF3, 0xE0,
-/*0190*/0x44, 0x20, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x19,
-	0xF3, 0xE0, 0x54, 0xDF, 0xF0, 0xEE, 0x44, 0xAE,
-/*01A0*/0x12, 0x1A, 0x43, 0x30, 0xE4, 0x03, 0x02, 0x03,
-	0xFB, 0x74, 0x9E, 0x12, 0x1A, 0x05, 0x20, 0xE0,
-/*01B0*/0x03, 0x02, 0x03, 0xFB, 0x8F, 0x82, 0x8E, 0x83,
-	0xE0, 0x20, 0xE0, 0x03, 0x02, 0x03, 0xFB, 0x12,
-/*01C0*/0x19, 0xF3, 0xE0, 0x44, 0x10, 0xF0, 0xE5, 0xE3,
-	0x20, 0xE7, 0x08, 0xE5, 0x08, 0x12, 0x1A, 0x3A,
-/*01D0*/0x44, 0x04, 0xF0, 0xAF, 0x08, 0x7E, 0x00, 0xEF,
-	0x12, 0x1A, 0x3A, 0x20, 0xE2, 0x34, 0x12, 0x19,
-/*01E0*/0xF3, 0xE0, 0x44, 0x08, 0xF0, 0xE5, 0xE4, 0x30,
-	0xE6, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00,
-/*01F0*/0xE5, 0x7E, 0xC3, 0x94, 0x04, 0x50, 0x04, 0x7C,
-	0x01, 0x80, 0x02, 0x7C, 0x00, 0xEC, 0x4D, 0x60,
-/*0200*/0x05, 0xC2, 0x35, 0x02, 0x03, 0xFB, 0xEE, 0x44,
-	0xD2, 0x12, 0x1A, 0x43, 0x44, 0x40, 0xF0, 0x02,
-/*0210*/0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xF7,
-	0xF0, 0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0,
-/*0220*/0x54, 0xBF, 0xF0, 0x90, 0x07, 0x14, 0xE0, 0x04,
-	0xF0, 0xE5, 0x7E, 0x70, 0x03, 0x75, 0x7E, 0x01,
-/*0230*/0xAF, 0x08, 0x7E, 0x00, 0x12, 0x1A, 0x23, 0x40,
-	0x12, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x01, 0x12,
-/*0240*/0x19, 0xF2, 0xE0, 0x54, 0x02, 0x12, 0x1A, 0x32,
-	0x02, 0x03, 0xFB, 0x12, 0x19, 0xF3, 0xE0, 0x44,
-/*0250*/0x02, 0x12, 0x19, 0xF2, 0xE0, 0x54, 0xFE, 0xF0,
-	0xC2, 0x35, 0xEE, 0x44, 0x8A, 0x8F, 0x82, 0xF5,
-/*0260*/0x83, 0xE0, 0xF5, 0x17, 0x54, 0x8F, 0x44, 0x40,
-	0xF0, 0x74, 0x90, 0xFC, 0xE5, 0x08, 0x44, 0x07,
-/*0270*/0xFD, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0x54, 0x3F,
-	0x90, 0x07, 0x02, 0xF0, 0xE0, 0x54, 0xC0, 0x8D,
-/*0280*/0x82, 0x8C, 0x83, 0xF0, 0x74, 0x92, 0x12, 0x1A,
-	0x05, 0x90, 0x07, 0x03, 0x12, 0x1A, 0x19, 0x74,
-/*0290*/0x82, 0x12, 0x1A, 0x05, 0x90, 0x07, 0x04, 0x12,
-	0x1A, 0x19, 0x74, 0xB4, 0x12, 0x1A, 0x05, 0x90,
-/*02A0*/0x07, 0x05, 0x12, 0x1A, 0x19, 0x74, 0x94, 0xFE,
-	0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A, 0xF5,
-/*02B0*/0x10, 0x30, 0xE0, 0x04, 0xD2, 0x37, 0x80, 0x02,
-	0xC2, 0x37, 0xE5, 0x10, 0x54, 0x7F, 0x8F, 0x82,
-/*02C0*/0x8E, 0x83, 0xF0, 0x30, 0x44, 0x30, 0x12, 0x1A,
-	0x03, 0x54, 0x80, 0xD3, 0x94, 0x00, 0x40, 0x04,
-/*02D0*/0xD2, 0x39, 0x80, 0x02, 0xC2, 0x39, 0x8F, 0x82,
-	0x8E, 0x83, 0xE0, 0x44, 0x80, 0xF0, 0x12, 0x1A,
-/*02E0*/0x03, 0x54, 0x40, 0xD3, 0x94, 0x00, 0x40, 0x04,
-	0xD2, 0x3A, 0x80, 0x02, 0xC2, 0x3A, 0x8F, 0x82,
-/*02F0*/0x8E, 0x83, 0xE0, 0x44, 0x40, 0xF0, 0x74, 0x92,
-	0xFE, 0xE5, 0x08, 0x44, 0x06, 0x12, 0x1A, 0x0A,
-/*0300*/0x30, 0xE7, 0x04, 0xD2, 0x38, 0x80, 0x02, 0xC2,
-	0x38, 0x8F, 0x82, 0x8E, 0x83, 0xE0, 0x54, 0x7F,
-/*0310*/0xF0, 0x12, 0x1E, 0x46, 0xE4, 0xF5, 0x0A, 0x20,
-	0x03, 0x02, 0x80, 0x03, 0x30, 0x43, 0x03, 0x12,
-/*0320*/0x19, 0x95, 0x20, 0x02, 0x02, 0x80, 0x03, 0x30,
-	0x42, 0x03, 0x12, 0x0C, 0x8F, 0x30, 0x30, 0x06,
-/*0330*/0x12, 0x19, 0x95, 0x12, 0x0C, 0x8F, 0x12, 0x0D,
-	0x47, 0x12, 0x19, 0xF3, 0xE0, 0x54, 0xFB, 0xF0,
-/*0340*/0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40, 0x46, 0x43,
-	0xE1, 0x08, 0x12, 0x19, 0xF3, 0xE0, 0x44, 0x04,
-/*0350*/0xF0, 0xE5, 0xE4, 0x20, 0xE7, 0x2A, 0x12, 0x1A,
-	0x12, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3,
-/*0360*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
-	0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40,
-/*0370*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF,
-	0x5E, 0x60, 0x05, 0x12, 0x1D, 0xD7, 0x80, 0x17,
-/*0380*/0x12, 0x1A, 0x12, 0x75, 0x83, 0xD2, 0xE0, 0x44,
-	0x08, 0xF0, 0x02, 0x03, 0xFB, 0x12, 0x1A, 0x12,
-/*0390*/0x75, 0x83, 0xD2, 0xE0, 0x54, 0xF7, 0xF0, 0x12,
-	0x1E, 0x46, 0x7F, 0x08, 0x12, 0x17, 0x31, 0x74,
-/*03A0*/0x8E, 0xFE, 0x12, 0x1A, 0x12, 0x8E, 0x83, 0xE0,
-	0xF5, 0x10, 0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44,
-/*03B0*/0x01, 0xFF, 0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07,
-	0xF5, 0x82, 0xEF, 0xF0, 0xE5, 0x10, 0x54, 0xFE,
-/*03C0*/0xFF, 0xED, 0x44, 0x07, 0xF5, 0x82, 0xEF, 0x12,
-	0x1A, 0x11, 0x75, 0x83, 0x86, 0xE0, 0x44, 0x10,
-/*03D0*/0x12, 0x1A, 0x11, 0xE0, 0x44, 0x10, 0xF0, 0x12,
-	0x19, 0xF3, 0xE0, 0x54, 0xFD, 0x44, 0x01, 0xFF,
-/*03E0*/0x12, 0x19, 0xF3, 0xEF, 0x12, 0x1A, 0x32, 0x30,
-	0x32, 0x0C, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82,
-/*03F0*/0x75, 0x83, 0x82, 0x74, 0x05, 0xF0, 0xAF, 0x0B,
-	0x12, 0x18, 0xD7, 0x74, 0x10, 0x25, 0x08, 0xF5,
-/*0400*/0x08, 0x02, 0x00, 0x85, 0x05, 0x09, 0xE5, 0x09,
-	0xD3, 0x94, 0x07, 0x50, 0x03, 0x02, 0x00, 0x82,
-/*0410*/0xE5, 0x7E, 0xD3, 0x94, 0x00, 0x40, 0x04, 0x7F,
-	0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x7E, 0xC3,
-/*0420*/0x94, 0xFA, 0x50, 0x04, 0x7E, 0x01, 0x80, 0x02,
-	0x7E, 0x00, 0xEE, 0x5F, 0x60, 0x02, 0x05, 0x7E,
-/*0430*/0x30, 0x35, 0x0B, 0x43, 0xE1, 0x01, 0x7F, 0x09,
-	0x12, 0x17, 0x31, 0x02, 0x00, 0x58, 0x53, 0xE1,
-/*0440*/0xFE, 0x02, 0x00, 0x58, 0x8E, 0x6A, 0x8F, 0x6B,
-	0x8C, 0x6C, 0x8D, 0x6D, 0x75, 0x6E, 0x01, 0x75,
-/*0450*/0x6F, 0x01, 0x75, 0x70, 0x01, 0xE4, 0xF5, 0x73,
-	0xF5, 0x74, 0xF5, 0x75, 0x90, 0x07, 0x2F, 0xF0,
-/*0460*/0xF5, 0x3C, 0xF5, 0x3E, 0xF5, 0x46, 0xF5, 0x47,
-	0xF5, 0x3D, 0xF5, 0x3F, 0xF5, 0x6F, 0xE5, 0x6F,
-/*0470*/0x70, 0x0F, 0xE5, 0x6B, 0x45, 0x6A, 0x12, 0x07,
-	0x2A, 0x75, 0x83, 0x80, 0x74, 0x3A, 0xF0, 0x80,
-/*0480*/0x09, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74,
-	0x1A, 0xF0, 0xE4, 0xF5, 0x6E, 0xC3, 0x74, 0x3F,
-/*0490*/0x95, 0x6E, 0xFF, 0x12, 0x08, 0x65, 0x75, 0x83,
-	0x82, 0xEF, 0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x08,
-/*04A0*/0xC6, 0xE5, 0x33, 0xF0, 0x12, 0x08, 0xFA, 0x12,
-	0x08, 0xB1, 0x40, 0xE1, 0xE5, 0x6F, 0x70, 0x0B,
-/*04B0*/0x12, 0x07, 0x2A, 0x75, 0x83, 0x80, 0x74, 0x36,
-	0xF0, 0x80, 0x09, 0x12, 0x07, 0x2A, 0x75, 0x83,
-/*04C0*/0x80, 0x74, 0x16, 0xF0, 0x75, 0x6E, 0x01, 0x12,
-	0x07, 0x2A, 0x75, 0x83, 0xB4, 0xE5, 0x6E, 0xF0,
-/*04D0*/0x12, 0x1A, 0x4D, 0x74, 0x3F, 0x25, 0x6E, 0xF5,
-	0x82, 0xE4, 0x34, 0x00, 0xF5, 0x83, 0xE5, 0x33,
-/*04E0*/0xF0, 0x74, 0xBF, 0x25, 0x6E, 0xF5, 0x82, 0xE4,
-	0x34, 0x00, 0x12, 0x08, 0xB1, 0x40, 0xD8, 0xE4,
-/*04F0*/0xF5, 0x70, 0xF5, 0x46, 0xF5, 0x47, 0xF5, 0x6E,
-	0x12, 0x08, 0xFA, 0xF5, 0x83, 0xE0, 0xFE, 0x12,
-/*0500*/0x08, 0xC6, 0xE0, 0x7C, 0x00, 0x24, 0x00, 0xFF,
-	0xEC, 0x3E, 0xFE, 0xAD, 0x3B, 0xD3, 0xEF, 0x9D,
-/*0510*/0xEE, 0x9C, 0x50, 0x04, 0x7B, 0x01, 0x80, 0x02,
-	0x7B, 0x00, 0xE5, 0x70, 0x70, 0x04, 0x7A, 0x01,
-/*0520*/0x80, 0x02, 0x7A, 0x00, 0xEB, 0x5A, 0x60, 0x06,
-	0x85, 0x6E, 0x46, 0x75, 0x70, 0x01, 0xD3, 0xEF,
-/*0530*/0x9D, 0xEE, 0x9C, 0x50, 0x04, 0x7F, 0x01, 0x80,
-	0x02, 0x7F, 0x00, 0xE5, 0x70, 0xB4, 0x01, 0x04,
-/*0540*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF, 0x5E,
-	0x60, 0x03, 0x85, 0x6E, 0x47, 0x05, 0x6E, 0xE5,
-/*0550*/0x6E, 0x64, 0x7F, 0x70, 0xA3, 0xE5, 0x46, 0x60,
-	0x05, 0xE5, 0x47, 0xB4, 0x7E, 0x03, 0x85, 0x46,
-/*0560*/0x47, 0xE5, 0x6F, 0x70, 0x08, 0x85, 0x46, 0x76,
-	0x85, 0x47, 0x77, 0x80, 0x0E, 0xC3, 0x74, 0x7F,
-/*0570*/0x95, 0x46, 0xF5, 0x78, 0xC3, 0x74, 0x7F, 0x95,
-	0x47, 0xF5, 0x79, 0xE5, 0x6F, 0x70, 0x37, 0xE5,
-/*0580*/0x46, 0x65, 0x47, 0x70, 0x0C, 0x75, 0x73, 0x01,
-	0x75, 0x74, 0x01, 0xF5, 0x3C, 0xF5, 0x3D, 0x80,
-/*0590*/0x35, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5, 0x47, 0x95,
-	0x46, 0xF5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25,
-/*05A0*/0x46, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05,
-	0xE4, 0xF5, 0x3D, 0x80, 0x40, 0xC3, 0x74, 0x3F,
-/*05B0*/0x95, 0x72, 0xF5, 0x3D, 0x80, 0x37, 0xE5, 0x46,
-	0x65, 0x47, 0x70, 0x0F, 0x75, 0x73, 0x01, 0x75,
-/*05C0*/0x75, 0x01, 0xF5, 0x3E, 0xF5, 0x3F, 0x75, 0x4E,
-	0x01, 0x80, 0x22, 0xE4, 0xF5, 0x4E, 0xC3, 0xE5,
-/*05D0*/0x47, 0x95, 0x46, 0xF5, 0x3E, 0xC3, 0x13, 0xF5,
-	0x71, 0x25, 0x46, 0xF5, 0x72, 0xD3, 0x94, 0x3F,
-/*05E0*/0x50, 0x05, 0xE4, 0xF5, 0x3F, 0x80, 0x06, 0xE5,
-	0x72, 0x24, 0xC1, 0xF5, 0x3F, 0x05, 0x6F, 0xE5,
-/*05F0*/0x6F, 0xC3, 0x94, 0x02, 0x50, 0x03, 0x02, 0x04,
-	0x6E, 0xE5, 0x6D, 0x45, 0x6C, 0x70, 0x02, 0x80,
-/*0600*/0x04, 0xE5, 0x74, 0x45, 0x75, 0x90, 0x07, 0x2F,
-	0xF0, 0x7F, 0x01, 0xE5, 0x3E, 0x60, 0x04, 0xE5,
-/*0610*/0x3C, 0x70, 0x14, 0xE4, 0xF5, 0x3C, 0xF5, 0x3D,
-	0xF5, 0x3E, 0xF5, 0x3F, 0x12, 0x08, 0xD2, 0x70,
-/*0620*/0x04, 0xF0, 0x02, 0x06, 0xA4, 0x80, 0x7A, 0xE5,
-	0x3C, 0xC3, 0x95, 0x3E, 0x40, 0x07, 0xE5, 0x3C,
-/*0630*/0x95, 0x3E, 0xFF, 0x80, 0x06, 0xC3, 0xE5, 0x3E,
-	0x95, 0x3C, 0xFF, 0xE5, 0x76, 0xD3, 0x95, 0x79,
-/*0640*/0x40, 0x05, 0x85, 0x76, 0x7A, 0x80, 0x03, 0x85,
-	0x79, 0x7A, 0xE5, 0x77, 0xC3, 0x95, 0x78, 0x50,
-/*0650*/0x05, 0x85, 0x77, 0x7B, 0x80, 0x03, 0x85, 0x78,
-	0x7B, 0xE5, 0x7B, 0xD3, 0x95, 0x7A, 0x40, 0x30,
-/*0660*/0xE5, 0x7B, 0x95, 0x7A, 0xF5, 0x3C, 0xF5, 0x3E,
-	0xC3, 0xE5, 0x7B, 0x95, 0x7A, 0x90, 0x07, 0x19,
-/*0670*/0xF0, 0xE5, 0x3C, 0xC3, 0x13, 0xF5, 0x71, 0x25,
-	0x7A, 0xF5, 0x72, 0xC3, 0x94, 0x3F, 0x40, 0x05,
-/*0680*/0xE4, 0xF5, 0x3D, 0x80, 0x1F, 0xC3, 0x74, 0x3F,
-	0x95, 0x72, 0xF5, 0x3D, 0xF5, 0x3F, 0x80, 0x14,
-/*0690*/0xE4, 0xF5, 0x3C, 0xF5, 0x3E, 0x90, 0x07, 0x19,
-	0xF0, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80,
-/*06A0*/0x03, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x65, 0x75,
-	0x83, 0xD0, 0xE0, 0x54, 0x0F, 0xFE, 0xAD, 0x3C,
-/*06B0*/0x70, 0x02, 0x7E, 0x07, 0xBE, 0x0F, 0x02, 0x7E,
-	0x80, 0xEE, 0xFB, 0xEF, 0xD3, 0x9B, 0x74, 0x80,
-/*06C0*/0xF8, 0x98, 0x40, 0x1F, 0xE4, 0xF5, 0x3C, 0xF5,
-	0x3E, 0x12, 0x08, 0xD2, 0x70, 0x03, 0xF0, 0x80,
-/*06D0*/0x12, 0x74, 0x01, 0xF0, 0xE5, 0x08, 0xFB, 0xEB,
-	0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xD2, 0xE0,
-/*06E0*/0x44, 0x10, 0xF0, 0xE5, 0x08, 0xFB, 0xEB, 0x44,
-	0x09, 0xF5, 0x82, 0x75, 0x83, 0x9E, 0xED, 0xF0,
-/*06F0*/0xEB, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xCA,
-	0xED, 0xF0, 0x12, 0x08, 0x65, 0x75, 0x83, 0xCC,
-/*0700*/0xEF, 0xF0, 0x22, 0xE5, 0x08, 0x44, 0x07, 0xF5,
-	0x82, 0x75, 0x83, 0xBC, 0xE0, 0x54, 0xF0, 0xF0,
-/*0710*/0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82, 0x75, 0x83,
-	0xBE, 0xE0, 0x54, 0xF0, 0xF0, 0xE5, 0x08, 0x44,
-/*0720*/0x07, 0xF5, 0x82, 0x75, 0x83, 0xC0, 0xE0, 0x54,
-	0xF0, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82,
-/*0730*/0x22, 0xF0, 0x90, 0x07, 0x28, 0xE0, 0xFE, 0xA3,
-	0xE0, 0xF5, 0x82, 0x8E, 0x83, 0x22, 0x85, 0x42,
-/*0740*/0x42, 0x85, 0x41, 0x41, 0x85, 0x40, 0x40, 0x74,
-	0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E, 0xF5,
-/*0750*/0x83, 0xE5, 0x42, 0xF0, 0x74, 0xE0, 0x2F, 0xF5,
-	0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xE5,
-/*0760*/0x42, 0x29, 0xFD, 0xE4, 0x33, 0xFC, 0xE5, 0x3C,
-	0xC3, 0x9D, 0xEC, 0x64, 0x80, 0xF8, 0x74, 0x80,
-/*0770*/0x98, 0x22, 0xF5, 0x83, 0xE0, 0x90, 0x07, 0x22,
-	0x54, 0x1F, 0xFD, 0xE0, 0xFA, 0xA3, 0xE0, 0xF5,
-/*0780*/0x82, 0x8A, 0x83, 0xED, 0xF0, 0x22, 0x90, 0x07,
-	0x22, 0xE0, 0xFC, 0xA3, 0xE0, 0xF5, 0x82, 0x8C,
-/*0790*/0x83, 0x22, 0x90, 0x07, 0x24, 0xFF, 0xED, 0x44,
-	0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22, 0x85,
-/*07A0*/0x38, 0x38, 0x85, 0x39, 0x39, 0x85, 0x3A, 0x3A,
-	0x74, 0xC0, 0x2F, 0xF5, 0x82, 0x74, 0x02, 0x3E,
-/*07B0*/0xF5, 0x83, 0x22, 0x90, 0x07, 0x26, 0xFF, 0xED,
-	0x44, 0x07, 0xCF, 0xF0, 0xA3, 0xEF, 0xF0, 0x22,
-/*07C0*/0xF0, 0x74, 0xA0, 0x2F, 0xF5, 0x82, 0x74, 0x02,
-	0x3E, 0xF5, 0x83, 0x22, 0x74, 0xC0, 0x25, 0x11,
-/*07D0*/0xF5, 0x82, 0xE4, 0x34, 0x01, 0xF5, 0x83, 0x22,
-	0x74, 0x00, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
-/*07E0*/0x02, 0xF5, 0x83, 0x22, 0x74, 0x60, 0x25, 0x11,
-	0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22,
-/*07F0*/0x74, 0x80, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
-	0x03, 0xF5, 0x83, 0x22, 0x74, 0xE0, 0x25, 0x11,
-/*0800*/0xF5, 0x82, 0xE4, 0x34, 0x03, 0xF5, 0x83, 0x22,
-	0x74, 0x40, 0x25, 0x11, 0xF5, 0x82, 0xE4, 0x34,
-/*0810*/0x06, 0xF5, 0x83, 0x22, 0x74, 0x80, 0x2F, 0xF5,
-	0x82, 0x74, 0x02, 0x3E, 0xF5, 0x83, 0x22, 0xAF,
-/*0820*/0x08, 0x7E, 0x00, 0xEF, 0x44, 0x07, 0xF5, 0x82,
-	0x22, 0xF5, 0x83, 0xE5, 0x82, 0x44, 0x07, 0xF5,
-/*0830*/0x82, 0xE5, 0x40, 0xF0, 0x22, 0x74, 0x40, 0x25,
-	0x11, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83,
-/*0840*/0x22, 0x74, 0xC0, 0x25, 0x11, 0xF5, 0x82, 0xE4,
-	0x34, 0x03, 0xF5, 0x83, 0x22, 0x74, 0x00, 0x25,
-/*0850*/0x11, 0xF5, 0x82, 0xE4, 0x34, 0x06, 0xF5, 0x83,
-	0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4,
-/*0860*/0x34, 0x06, 0xF5, 0x83, 0x22, 0xE5, 0x08, 0xFD,
-	0xED, 0x44, 0x07, 0xF5, 0x82, 0x22, 0xE5, 0x41,
-/*0870*/0xF0, 0xE5, 0x65, 0x64, 0x01, 0x45, 0x64, 0x22,
-	0x7E, 0x00, 0xFB, 0x7A, 0x00, 0xFD, 0x7C, 0x00,
-/*0880*/0x22, 0x74, 0x20, 0x25, 0x11, 0xF5, 0x82, 0xE4,
-	0x34, 0x02, 0x22, 0x74, 0xA0, 0x25, 0x11, 0xF5,
-/*0890*/0x82, 0xE4, 0x34, 0x03, 0x22, 0x85, 0x3E, 0x42,
-	0x85, 0x3F, 0x41, 0x8F, 0x40, 0x22, 0x85, 0x3C,
-/*08A0*/0x42, 0x85, 0x3D, 0x41, 0x8F, 0x40, 0x22, 0x75,
-	0x45, 0x3F, 0x90, 0x07, 0x20, 0xE4, 0xF0, 0xA3,
-/*08B0*/0x22, 0xF5, 0x83, 0xE5, 0x32, 0xF0, 0x05, 0x6E,
-	0xE5, 0x6E, 0xC3, 0x94, 0x40, 0x22, 0xF0, 0xE5,
-/*08C0*/0x08, 0x44, 0x06, 0xF5, 0x82, 0x22, 0x74, 0x00,
-	0x25, 0x6E, 0xF5, 0x82, 0xE4, 0x34, 0x00, 0xF5,
-/*08D0*/0x83, 0x22, 0xE5, 0x6D, 0x45, 0x6C, 0x90, 0x07,
-	0x2F, 0x22, 0xE4, 0xF9, 0xE5, 0x3C, 0xD3, 0x95,
-/*08E0*/0x3E, 0x22, 0x74, 0x80, 0x2E, 0xF5, 0x82, 0xE4,
-	0x34, 0x02, 0xF5, 0x83, 0xE0, 0x22, 0x74, 0xA0,
-/*08F0*/0x2E, 0xF5, 0x82, 0xE4, 0x34, 0x02, 0xF5, 0x83,
-	0xE0, 0x22, 0x74, 0x80, 0x25, 0x6E, 0xF5, 0x82,
-/*0900*/0xE4, 0x34, 0x00, 0x22, 0x25, 0x42, 0xFD, 0xE4,
-	0x33, 0xFC, 0x22, 0x85, 0x42, 0x42, 0x85, 0x41,
-/*0910*/0x41, 0x85, 0x40, 0x40, 0x22, 0xED, 0x4C, 0x60,
-	0x03, 0x02, 0x09, 0xE5, 0xEF, 0x4E, 0x70, 0x37,
-/*0920*/0x90, 0x07, 0x26, 0x12, 0x07, 0x89, 0xE0, 0xFD,
-	0x12, 0x07, 0xCC, 0xED, 0xF0, 0x90, 0x07, 0x28,
-/*0930*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xD8,
-	0xED, 0xF0, 0x12, 0x07, 0x86, 0xE0, 0x54, 0x1F,
-/*0940*/0xFD, 0x12, 0x08, 0x81, 0xF5, 0x83, 0xED, 0xF0,
-	0x90, 0x07, 0x24, 0x12, 0x07, 0x89, 0xE0, 0x54,
-/*0950*/0x1F, 0xFD, 0x12, 0x08, 0x35, 0xED, 0xF0, 0xEF,
-	0x64, 0x04, 0x4E, 0x70, 0x37, 0x90, 0x07, 0x26,
-/*0960*/0x12, 0x07, 0x89, 0xE0, 0xFD, 0x12, 0x07, 0xE4,
-	0xED, 0xF0, 0x90, 0x07, 0x28, 0x12, 0x07, 0x89,
-/*0970*/0xE0, 0xFD, 0x12, 0x07, 0xF0, 0xED, 0xF0, 0x12,
-	0x07, 0x86, 0xE0, 0x54, 0x1F, 0xFD, 0x12, 0x08,
-/*0980*/0x8B, 0xF5, 0x83, 0xED, 0xF0, 0x90, 0x07, 0x24,
-	0x12, 0x07, 0x89, 0xE0, 0x54, 0x1F, 0xFD, 0x12,
-/*0990*/0x08, 0x41, 0xED, 0xF0, 0xEF, 0x64, 0x01, 0x4E,
-	0x70, 0x04, 0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00,
-/*09A0*/0xEF, 0x64, 0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01,
-	0x80, 0x02, 0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x78,
-/*09B0*/0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE0, 0xFF,
-	0x12, 0x07, 0xFC, 0xEF, 0x12, 0x07, 0x31, 0xE0,
-/*09C0*/0xFF, 0x12, 0x08, 0x08, 0xEF, 0xF0, 0x90, 0x07,
-	0x22, 0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF,
-/*09D0*/0x12, 0x08, 0x4D, 0xEF, 0xF0, 0x90, 0x07, 0x24,
-	0x12, 0x07, 0x35, 0xE0, 0x54, 0x1F, 0xFF, 0x12,
-/*09E0*/0x08, 0x59, 0xEF, 0xF0, 0x22, 0x12, 0x07, 0xCC,
-	0xE4, 0xF0, 0x12, 0x07, 0xD8, 0xE4, 0xF0, 0x12,
-/*09F0*/0x08, 0x81, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08,
-	0x35, 0x74, 0x14, 0xF0, 0x12, 0x07, 0xE4, 0xE4,
-/*0A00*/0xF0, 0x12, 0x07, 0xF0, 0xE4, 0xF0, 0x12, 0x08,
-	0x8B, 0xF5, 0x83, 0xE4, 0xF0, 0x12, 0x08, 0x41,
-/*0A10*/0x74, 0x14, 0xF0, 0x12, 0x07, 0xFC, 0xE4, 0xF0,
-	0x12, 0x08, 0x08, 0xE4, 0xF0, 0x12, 0x08, 0x4D,
-/*0A20*/0xE4, 0xF0, 0x12, 0x08, 0x59, 0x74, 0x14, 0xF0,
-	0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFC, 0x10, 0xE4,
-/*0A30*/0xF5, 0xFD, 0x75, 0xFE, 0x30, 0xF5, 0xFF, 0xE5,
-	0xE7, 0x20, 0xE7, 0x03, 0x43, 0xF9, 0x08, 0xE5,
-/*0A40*/0xE6, 0x20, 0xE7, 0x0B, 0x78, 0xFF, 0xE4, 0xF6,
-	0xD8, 0xFD, 0x53, 0xE6, 0xFE, 0x80, 0x09, 0x78,
-/*0A50*/0x08, 0xE4, 0xF6, 0xD8, 0xFD, 0x53, 0xE6, 0xFE,
-	0x75, 0x81, 0x80, 0xE4, 0xF5, 0xA8, 0xD2, 0xA8,
-/*0A60*/0xC2, 0xA9, 0xD2, 0xAF, 0xE5, 0xE2, 0x20, 0xE5,
-	0x05, 0x20, 0xE6, 0x02, 0x80, 0x03, 0x43, 0xE1,
-/*0A70*/0x02, 0xE5, 0xE2, 0x20, 0xE0, 0x0E, 0x90, 0x00,
-	0x00, 0x7F, 0x00, 0x7E, 0x08, 0xE4, 0xF0, 0xA3,
-/*0A80*/0xDF, 0xFC, 0xDE, 0xFA, 0x02, 0x0A, 0xDB, 0x43,
-	0xFA, 0x01, 0xC0, 0xE0, 0xC0, 0xF0, 0xC0, 0x83,
-/*0A90*/0xC0, 0x82, 0xC0, 0xD0, 0x12, 0x1C, 0xE7, 0xD0,
-	0xD0, 0xD0, 0x82, 0xD0, 0x83, 0xD0, 0xF0, 0xD0,
-/*0AA0*/0xE0, 0x53, 0xFA, 0xFE, 0x32, 0x02, 0x1B, 0x55,
-	0xE4, 0x93, 0xA3, 0xF8, 0xE4, 0x93, 0xA3, 0xF6,
-/*0AB0*/0x08, 0xDF, 0xF9, 0x80, 0x29, 0xE4, 0x93, 0xA3,
-	0xF8, 0x54, 0x07, 0x24, 0x0C, 0xC8, 0xC3, 0x33,
-/*0AC0*/0xC4, 0x54, 0x0F, 0x44, 0x20, 0xC8, 0x83, 0x40,
-	0x04, 0xF4, 0x56, 0x80, 0x01, 0x46, 0xF6, 0xDF,
-/*0AD0*/0xE4, 0x80, 0x0B, 0x01, 0x02, 0x04, 0x08, 0x10,
-	0x20, 0x40, 0x80, 0x90, 0x00, 0x3F, 0xE4, 0x7E,
-/*0AE0*/0x01, 0x93, 0x60, 0xC1, 0xA3, 0xFF, 0x54, 0x3F,
-	0x30, 0xE5, 0x09, 0x54, 0x1F, 0xFE, 0xE4, 0x93,
-/*0AF0*/0xA3, 0x60, 0x01, 0x0E, 0xCF, 0x54, 0xC0, 0x25,
-	0xE0, 0x60, 0xAD, 0x40, 0xB8, 0x80, 0xFE, 0x8C,
-/*0B00*/0x64, 0x8D, 0x65, 0x8A, 0x66, 0x8B, 0x67, 0xE4,
-	0xF5, 0x69, 0xEF, 0x4E, 0x70, 0x03, 0x02, 0x1D,
-/*0B10*/0x55, 0xE4, 0xF5, 0x68, 0xE5, 0x67, 0x45, 0x66,
-	0x70, 0x32, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90,
-/*0B20*/0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE4,
-	0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4, 0x12,
-/*0B30*/0x08, 0x70, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75,
-	0x83, 0x92, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83,
-/*0B40*/0xC6, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8,
-	0xE4, 0xF0, 0x80, 0x11, 0x90, 0x07, 0x26, 0x12,
-/*0B50*/0x07, 0x35, 0xE4, 0x12, 0x08, 0x70, 0x70, 0x05,
-	0x12, 0x07, 0x32, 0xE4, 0xF0, 0x12, 0x1D, 0x55,
-/*0B60*/0x12, 0x1E, 0xBF, 0xE5, 0x67, 0x45, 0x66, 0x70,
-	0x33, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x90, 0xE5,
-/*0B70*/0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5,
-	0x41, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x12,
-/*0B80*/0x08, 0x6E, 0x70, 0x29, 0x12, 0x07, 0x2A, 0x75,
-	0x83, 0x92, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75,
-/*0B90*/0x83, 0xC6, 0xE5, 0x40, 0x12, 0x07, 0x29, 0x75,
-	0x83, 0xC8, 0x80, 0x0E, 0x90, 0x07, 0x26, 0x12,
-/*0BA0*/0x07, 0x35, 0x12, 0x08, 0x6E, 0x70, 0x06, 0x12,
-	0x07, 0x32, 0xE5, 0x40, 0xF0, 0xAF, 0x69, 0x7E,
-/*0BB0*/0x00, 0xAD, 0x67, 0xAC, 0x66, 0x12, 0x04, 0x44,
-	0x12, 0x07, 0x2A, 0x75, 0x83, 0xCA, 0xE0, 0xD3,
-/*0BC0*/0x94, 0x00, 0x50, 0x0C, 0x05, 0x68, 0xE5, 0x68,
-	0xC3, 0x94, 0x05, 0x50, 0x03, 0x02, 0x0B, 0x14,
-/*0BD0*/0x22, 0x8C, 0x60, 0x8D, 0x61, 0x12, 0x08, 0xDA,
-	0x74, 0x20, 0x40, 0x0D, 0x2F, 0xF5, 0x82, 0x74,
-/*0BE0*/0x03, 0x3E, 0xF5, 0x83, 0xE5, 0x3E, 0xF0, 0x80,
-	0x0B, 0x2F, 0xF5, 0x82, 0x74, 0x03, 0x3E, 0xF5,
-/*0BF0*/0x83, 0xE5, 0x3C, 0xF0, 0xE5, 0x3C, 0xD3, 0x95,
-	0x3E, 0x40, 0x3C, 0xE5, 0x61, 0x45, 0x60, 0x70,
-/*0C00*/0x10, 0xE9, 0x12, 0x09, 0x04, 0xE5, 0x3E, 0x12,
-	0x07, 0x68, 0x40, 0x3B, 0x12, 0x08, 0x95, 0x80,
-/*0C10*/0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40, 0x1D,
-	0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05, 0x85,
-/*0C20*/0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39, 0x8F,
-	0x3A, 0x12, 0x08, 0x14, 0xE5, 0x3E, 0x12, 0x07,
-/*0C30*/0xC0, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x43, 0xE5,
-	0x61, 0x45, 0x60, 0x70, 0x19, 0x12, 0x07, 0x5F,
-/*0C40*/0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x27, 0x12,
-	0x09, 0x0B, 0x12, 0x08, 0x14, 0xE5, 0x42, 0x12,
-/*0C50*/0x07, 0xC0, 0xE5, 0x41, 0xF0, 0x22, 0xE5, 0x3C,
-	0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C, 0x38,
-/*0C60*/0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39, 0x80,
-	0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x08,
-/*0C70*/0x14, 0xE5, 0x3C, 0x12, 0x07, 0xC0, 0xE5, 0x3D,
-	0xF0, 0x22, 0x85, 0x38, 0x38, 0x85, 0x39, 0x39,
-/*0C80*/0x85, 0x3A, 0x3A, 0x12, 0x08, 0x14, 0xE5, 0x38,
-	0x12, 0x07, 0xC0, 0xE5, 0x39, 0xF0, 0x22, 0x7F,
-/*0C90*/0x06, 0x12, 0x17, 0x31, 0x12, 0x1D, 0x23, 0x12,
-	0x0E, 0x04, 0x12, 0x0E, 0x33, 0xE0, 0x44, 0x0A,
-/*0CA0*/0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E, 0x04, 0x12,
-	0x0E, 0x0B, 0xEF, 0xF0, 0xE5, 0x28, 0x30, 0xE5,
-/*0CB0*/0x03, 0xD3, 0x80, 0x01, 0xC3, 0x40, 0x05, 0x75,
-	0x14, 0x20, 0x80, 0x03, 0x75, 0x14, 0x08, 0x12,
-/*0CC0*/0x0E, 0x04, 0x75, 0x83, 0x8A, 0xE5, 0x14, 0xF0,
-	0xB4, 0xFF, 0x05, 0x75, 0x12, 0x80, 0x80, 0x06,
-/*0CD0*/0xE5, 0x14, 0xC3, 0x13, 0xF5, 0x12, 0xE4, 0xF5,
-	0x16, 0xF5, 0x7F, 0x12, 0x19, 0x36, 0x12, 0x13,
-/*0CE0*/0xA3, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x50, 0x09,
-	0x05, 0x16, 0xE5, 0x16, 0xC3, 0x94, 0x14, 0x40,
-/*0CF0*/0xEA, 0xE5, 0xE4, 0x20, 0xE7, 0x28, 0x12, 0x0E,
-	0x04, 0x75, 0x83, 0xD2, 0xE0, 0x54, 0x08, 0xD3,
-/*0D00*/0x94, 0x00, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
-	0x7F, 0x00, 0xE5, 0x0A, 0xC3, 0x94, 0x01, 0x40,
-/*0D10*/0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEF,
-	0x5E, 0x60, 0x03, 0x12, 0x1D, 0xD7, 0xE5, 0x7F,
-/*0D20*/0xC3, 0x94, 0x11, 0x40, 0x14, 0x12, 0x0E, 0x04,
-	0x75, 0x83, 0xD2, 0xE0, 0x44, 0x80, 0xF0, 0xE5,
-/*0D30*/0xE4, 0x20, 0xE7, 0x0F, 0x12, 0x1D, 0xD7, 0x80,
-	0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0xD2, 0xE0,
-/*0D40*/0x54, 0x7F, 0xF0, 0x12, 0x1D, 0x23, 0x22, 0x74,
-	0x8A, 0x85, 0x08, 0x82, 0xF5, 0x83, 0xE5, 0x17,
-/*0D50*/0xF0, 0x12, 0x0E, 0x3A, 0xE4, 0xF0, 0x90, 0x07,
-	0x02, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x90,
-/*0D60*/0xEF, 0xF0, 0x74, 0x92, 0xFE, 0xE5, 0x08, 0x44,
-	0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0, 0x54,
-/*0D70*/0xC0, 0xFD, 0x90, 0x07, 0x03, 0xE0, 0x54, 0x3F,
-	0x4D, 0x8F, 0x82, 0x8E, 0x83, 0xF0, 0x90, 0x07,
-/*0D80*/0x04, 0xE0, 0x12, 0x0E, 0x17, 0x75, 0x83, 0x82,
-	0xEF, 0xF0, 0x90, 0x07, 0x05, 0xE0, 0xFF, 0xED,
-/*0D90*/0x44, 0x07, 0xF5, 0x82, 0x75, 0x83, 0xB4, 0xEF,
-	0x12, 0x0E, 0x03, 0x75, 0x83, 0x80, 0xE0, 0x54,
-/*0DA0*/0xBF, 0xF0, 0x30, 0x37, 0x0A, 0x12, 0x0E, 0x91,
-	0x75, 0x83, 0x94, 0xE0, 0x44, 0x80, 0xF0, 0x30,
-/*0DB0*/0x38, 0x0A, 0x12, 0x0E, 0x91, 0x75, 0x83, 0x92,
-	0xE0, 0x44, 0x80, 0xF0, 0xE5, 0x28, 0x30, 0xE4,
-/*0DC0*/0x1A, 0x20, 0x39, 0x0A, 0x12, 0x0E, 0x04, 0x75,
-	0x83, 0x88, 0xE0, 0x54, 0x7F, 0xF0, 0x20, 0x3A,
-/*0DD0*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x88, 0xE0,
-	0x54, 0xBF, 0xF0, 0x74, 0x8C, 0xFE, 0x12, 0x0E,
-/*0DE0*/0x04, 0x8E, 0x83, 0xE0, 0x54, 0x0F, 0x12, 0x0E,
-	0x03, 0x75, 0x83, 0x86, 0xE0, 0x54, 0xBF, 0xF0,
-/*0DF0*/0xE5, 0x08, 0x44, 0x06, 0x12, 0x0D, 0xFD, 0x75,
-	0x83, 0x8A, 0xE4, 0xF0, 0x22, 0xF5, 0x82, 0x75,
-/*0E00*/0x83, 0x82, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07,
-	0xF5, 0x82, 0x22, 0x8E, 0x83, 0xE0, 0xF5, 0x10,
-/*0E10*/0x54, 0xFE, 0xF0, 0xE5, 0x10, 0x44, 0x01, 0xFF,
-	0xE5, 0x08, 0xFD, 0xED, 0x44, 0x07, 0xF5, 0x82,
-/*0E20*/0x22, 0xE5, 0x15, 0xC4, 0x54, 0x07, 0xFF, 0xE5,
-	0x08, 0xFD, 0xED, 0x44, 0x08, 0xF5, 0x82, 0x75,
-/*0E30*/0x83, 0x82, 0x22, 0x75, 0x83, 0x80, 0xE0, 0x44,
-	0x40, 0xF0, 0xE5, 0x08, 0x44, 0x08, 0xF5, 0x82,
-/*0E40*/0x75, 0x83, 0x8A, 0x22, 0xE5, 0x16, 0x25, 0xE0,
-	0x25, 0xE0, 0x24, 0xAF, 0xF5, 0x82, 0xE4, 0x34,
-/*0E50*/0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0D, 0x22,
-	0x43, 0xE1, 0x10, 0x43, 0xE1, 0x80, 0x53, 0xE1,
-/*0E60*/0xFD, 0x85, 0xE1, 0x10, 0x22, 0xE5, 0x16, 0x25,
-	0xE0, 0x25, 0xE0, 0x24, 0xB2, 0xF5, 0x82, 0xE4,
-/*0E70*/0x34, 0x1A, 0xF5, 0x83, 0xE4, 0x93, 0x22, 0x85,
-	0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15, 0xF0,
-/*0E80*/0x22, 0xE5, 0xE2, 0x54, 0x20, 0xD3, 0x94, 0x00,
-	0x22, 0xE5, 0xE2, 0x54, 0x40, 0xD3, 0x94, 0x00,
-/*0E90*/0x22, 0xE5, 0x08, 0x44, 0x06, 0xF5, 0x82, 0x22,
-	0xFD, 0xE5, 0x08, 0xFB, 0xEB, 0x44, 0x07, 0xF5,
-/*0EA0*/0x82, 0x22, 0x53, 0xF9, 0xF7, 0x75, 0xFE, 0x30,
-	0x22, 0xEF, 0x4E, 0x70, 0x26, 0x12, 0x07, 0xCC,
-/*0EB0*/0xE0, 0xFD, 0x90, 0x07, 0x26, 0x12, 0x07, 0x7B,
-	0x12, 0x07, 0xD8, 0xE0, 0xFD, 0x90, 0x07, 0x28,
-/*0EC0*/0x12, 0x07, 0x7B, 0x12, 0x08, 0x81, 0x12, 0x07,
-	0x72, 0x12, 0x08, 0x35, 0xE0, 0x90, 0x07, 0x24,
-/*0ED0*/0x12, 0x07, 0x78, 0xEF, 0x64, 0x04, 0x4E, 0x70,
-	0x29, 0x12, 0x07, 0xE4, 0xE0, 0xFD, 0x90, 0x07,
-/*0EE0*/0x26, 0x12, 0x07, 0x7B, 0x12, 0x07, 0xF0, 0xE0,
-	0xFD, 0x90, 0x07, 0x28, 0x12, 0x07, 0x7B, 0x12,
-/*0EF0*/0x08, 0x8B, 0x12, 0x07, 0x72, 0x12, 0x08, 0x41,
-	0xE0, 0x54, 0x1F, 0xFD, 0x90, 0x07, 0x24, 0x12,
-/*0F00*/0x07, 0x7B, 0xEF, 0x64, 0x01, 0x4E, 0x70, 0x04,
-	0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0xEF, 0x64,
-/*0F10*/0x02, 0x4E, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02,
-	0x7F, 0x00, 0xEF, 0x4D, 0x60, 0x35, 0x12, 0x07,
-/*0F20*/0xFC, 0xE0, 0xFF, 0x90, 0x07, 0x26, 0x12, 0x07,
-	0x89, 0xEF, 0xF0, 0x12, 0x08, 0x08, 0xE0, 0xFF,
-/*0F30*/0x90, 0x07, 0x28, 0x12, 0x07, 0x89, 0xEF, 0xF0,
-	0x12, 0x08, 0x4D, 0xE0, 0x54, 0x1F, 0xFF, 0x12,
-/*0F40*/0x07, 0x86, 0xEF, 0xF0, 0x12, 0x08, 0x59, 0xE0,
-	0x54, 0x1F, 0xFF, 0x90, 0x07, 0x24, 0x12, 0x07,
-/*0F50*/0x89, 0xEF, 0xF0, 0x22, 0xE4, 0xF5, 0x53, 0x12,
-	0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01, 0x80, 0x02,
-/*0F60*/0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40, 0x04, 0x7E,
-	0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x70,
-/*0F70*/0x03, 0x02, 0x0F, 0xF6, 0x85, 0xE1, 0x10, 0x43,
-	0xE1, 0x02, 0x53, 0xE1, 0x0F, 0x85, 0xE1, 0x10,
-/*0F80*/0xE4, 0xF5, 0x51, 0xE5, 0xE3, 0x54, 0x3F, 0xF5,
-	0x52, 0x12, 0x0E, 0x89, 0x40, 0x1D, 0xAD, 0x52,
-/*0F90*/0xAF, 0x51, 0x12, 0x11, 0x18, 0xEF, 0x60, 0x08,
-	0x85, 0xE1, 0x10, 0x43, 0xE1, 0x40, 0x80, 0x0B,
-/*0FA0*/0x53, 0xE1, 0xBF, 0x12, 0x0E, 0x58, 0x12, 0x00,
-	0x06, 0x80, 0xFB, 0xE5, 0xE3, 0x54, 0x3F, 0xF5,
-/*0FB0*/0x51, 0xE5, 0xE4, 0x54, 0x3F, 0xF5, 0x52, 0x12,
-	0x0E, 0x81, 0x40, 0x1D, 0xAD, 0x52, 0xAF, 0x51,
-/*0FC0*/0x12, 0x11, 0x18, 0xEF, 0x60, 0x08, 0x85, 0xE1,
-	0x10, 0x43, 0xE1, 0x20, 0x80, 0x0B, 0x53, 0xE1,
-/*0FD0*/0xDF, 0x12, 0x0E, 0x58, 0x12, 0x00, 0x06, 0x80,
-	0xFB, 0x12, 0x0E, 0x81, 0x40, 0x04, 0x7F, 0x01,
-/*0FE0*/0x80, 0x02, 0x7F, 0x00, 0x12, 0x0E, 0x89, 0x40,
-	0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE,
-/*0FF0*/0x4F, 0x60, 0x03, 0x12, 0x0E, 0x5B, 0x22, 0x12,
-	0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x22,
-/*1000*/0x02, 0x11, 0x00, 0x02, 0x10, 0x40, 0x02, 0x10,
-	0x90, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1010*/0x01, 0x20, 0x01, 0x20, 0xE4, 0xF5, 0x57, 0x12,
-	0x16, 0xBD, 0x12, 0x16, 0x44, 0xE4, 0x12, 0x10,
-/*1020*/0x56, 0x12, 0x14, 0xB7, 0x90, 0x07, 0x26, 0x12,
-	0x07, 0x35, 0xE4, 0x12, 0x07, 0x31, 0xE4, 0xF0,
-/*1030*/0x12, 0x10, 0x56, 0x12, 0x14, 0xB7, 0x90, 0x07,
-	0x26, 0x12, 0x07, 0x35, 0xE5, 0x41, 0x12, 0x07,
-/*1040*/0x31, 0xE5, 0x40, 0xF0, 0xAF, 0x57, 0x7E, 0x00,
-	0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF,
-/*1050*/0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xFF, 0x90,
-	0x07, 0x20, 0xA3, 0xE0, 0xFD, 0xE4, 0xF5, 0x56,
-/*1060*/0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x12,
-	0x11, 0x51, 0x7F, 0x0F, 0x7D, 0x18, 0xE4, 0xF5,
-/*1070*/0x56, 0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA,
-	0x12, 0x15, 0x41, 0xAF, 0x56, 0x7E, 0x00, 0x12,
-/*1080*/0x1A, 0xFF, 0xE4, 0xFF, 0xF5, 0x56, 0x7D, 0x1F,
-	0xF5, 0x40, 0xFE, 0xFC, 0xAB, 0x56, 0xFA, 0x22,
-/*1090*/0x22, 0xE4, 0xF5, 0x55, 0xE5, 0x08, 0xFD, 0x74,
-	0xA0, 0xF5, 0x56, 0xED, 0x44, 0x07, 0xF5, 0x57,
-/*10A0*/0xE5, 0x28, 0x30, 0xE5, 0x03, 0xD3, 0x80, 0x01,
-	0xC3, 0x40, 0x05, 0x7F, 0x28, 0xEF, 0x80, 0x04,
-/*10B0*/0x7F, 0x14, 0xEF, 0xC3, 0x13, 0xF5, 0x54, 0xE4,
-	0xF9, 0x12, 0x0E, 0x18, 0x75, 0x83, 0x8E, 0xE0,
-/*10C0*/0xF5, 0x10, 0xCE, 0xEF, 0xCE, 0xEE, 0xD3, 0x94,
-	0x00, 0x40, 0x26, 0xE5, 0x10, 0x54, 0xFE, 0x12,
-/*10D0*/0x0E, 0x98, 0x75, 0x83, 0x8E, 0xED, 0xF0, 0xE5,
-	0x10, 0x44, 0x01, 0xFD, 0xEB, 0x44, 0x07, 0xF5,
-/*10E0*/0x82, 0xED, 0xF0, 0x85, 0x57, 0x82, 0x85, 0x56,
-	0x83, 0xE0, 0x30, 0xE3, 0x01, 0x09, 0x1E, 0x80,
-/*10F0*/0xD4, 0xC2, 0x34, 0xE9, 0xC3, 0x95, 0x54, 0x40,
-	0x02, 0xD2, 0x34, 0x22, 0x02, 0x00, 0x06, 0x22,
-/*1100*/0x30, 0x30, 0x11, 0x90, 0x10, 0x00, 0xE4, 0x93,
-	0xF5, 0x10, 0x90, 0x10, 0x10, 0xE4, 0x93, 0xF5,
-/*1110*/0x10, 0x12, 0x10, 0x90, 0x12, 0x11, 0x50, 0x22,
-	0xE4, 0xFC, 0xC3, 0xED, 0x9F, 0xFA, 0xEF, 0xF5,
-/*1120*/0x83, 0x75, 0x82, 0x00, 0x79, 0xFF, 0xE4, 0x93,
-	0xCC, 0x6C, 0xCC, 0xA3, 0xD9, 0xF8, 0xDA, 0xF6,
-/*1130*/0xE5, 0xE2, 0x30, 0xE4, 0x02, 0x8C, 0xE5, 0xED,
-	0x24, 0xFF, 0xFF, 0xEF, 0x75, 0x82, 0xFF, 0xF5,
-/*1140*/0x83, 0xE4, 0x93, 0x6C, 0x70, 0x03, 0x7F, 0x01,
-	0x22, 0x7F, 0x00, 0x22, 0x22, 0x11, 0x00, 0x00,
-/*1150*/0x22, 0x8E, 0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D,
-	0x5B, 0x8A, 0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01,
-/*1160*/0xE4, 0xF5, 0x5F, 0xF5, 0x60, 0xF5, 0x62, 0x12,
-	0x07, 0x2A, 0x75, 0x83, 0xD0, 0xE0, 0xFF, 0xC4,
-/*1170*/0x54, 0x0F, 0xF5, 0x61, 0x12, 0x1E, 0xA5, 0x85,
-	0x59, 0x5E, 0xD3, 0xE5, 0x5E, 0x95, 0x5B, 0xE5,
-/*1180*/0x5A, 0x12, 0x07, 0x6B, 0x50, 0x4B, 0x12, 0x07,
-	0x03, 0x75, 0x83, 0xBC, 0xE0, 0x45, 0x5E, 0x12,
-/*1190*/0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x45, 0x5E,
-	0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0, 0x45,
-/*11A0*/0x5E, 0xF0, 0xAF, 0x5F, 0xE5, 0x60, 0x12, 0x08,
-	0x78, 0x12, 0x0A, 0xFF, 0xAF, 0x62, 0x7E, 0x00,
-/*11B0*/0xAD, 0x5D, 0xAC, 0x5C, 0x12, 0x04, 0x44, 0xE5,
-	0x61, 0xAF, 0x5E, 0x7E, 0x00, 0xB4, 0x03, 0x05,
-/*11C0*/0x12, 0x1E, 0x21, 0x80, 0x07, 0xAD, 0x5D, 0xAC,
-	0x5C, 0x12, 0x13, 0x17, 0x05, 0x5E, 0x02, 0x11,
-/*11D0*/0x7A, 0x12, 0x07, 0x03, 0x75, 0x83, 0xBC, 0xE0,
-	0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBE,
-/*11E0*/0xE0, 0x45, 0x40, 0x12, 0x07, 0x29, 0x75, 0x83,
-	0xC0, 0xE0, 0x45, 0x40, 0xF0, 0x22, 0x8E, 0x58,
-/*11F0*/0x8F, 0x59, 0x75, 0x5A, 0x01, 0x79, 0x01, 0x75,
-	0x5B, 0x01, 0xE4, 0xFB, 0x12, 0x07, 0x2A, 0x75,
-/*1200*/0x83, 0xAE, 0xE0, 0x54, 0x1A, 0xFF, 0x12, 0x08,
-	0x65, 0xE0, 0xC4, 0x13, 0x54, 0x07, 0xFE, 0xEF,
-/*1210*/0x70, 0x0C, 0xEE, 0x65, 0x35, 0x70, 0x07, 0x90,
-	0x07, 0x2F, 0xE0, 0xB4, 0x01, 0x0D, 0xAF, 0x35,
-/*1220*/0x7E, 0x00, 0x12, 0x0E, 0xA9, 0xCF, 0xEB, 0xCF,
-	0x02, 0x1E, 0x60, 0xE5, 0x59, 0x64, 0x02, 0x45,
-/*1230*/0x58, 0x70, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F,
-	0x00, 0xE5, 0x59, 0x45, 0x58, 0x70, 0x04, 0x7E,
-/*1240*/0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60,
-	0x23, 0x85, 0x41, 0x49, 0x85, 0x40, 0x4B, 0xE5,
-/*1250*/0x59, 0x45, 0x58, 0x70, 0x2C, 0xAF, 0x5A, 0xFE,
-	0xCD, 0xE9, 0xCD, 0xFC, 0xAB, 0x59, 0xAA, 0x58,
-/*1260*/0x12, 0x0A, 0xFF, 0xAF, 0x5B, 0x7E, 0x00, 0x12,
-	0x1E, 0x60, 0x80, 0x15, 0xAF, 0x5B, 0x7E, 0x00,
-/*1270*/0x12, 0x1E, 0x60, 0x90, 0x07, 0x26, 0x12, 0x07,
-	0x35, 0xE5, 0x49, 0x12, 0x07, 0x31, 0xE5, 0x4B,
-/*1280*/0xF0, 0xE4, 0xFD, 0xAF, 0x35, 0xFE, 0xFC, 0x12,
-	0x09, 0x15, 0x22, 0x8C, 0x64, 0x8D, 0x65, 0x12,
-/*1290*/0x08, 0xDA, 0x40, 0x3C, 0xE5, 0x65, 0x45, 0x64,
-	0x70, 0x10, 0x12, 0x09, 0x04, 0xC3, 0xE5, 0x3E,
-/*12A0*/0x12, 0x07, 0x69, 0x40, 0x3B, 0x12, 0x08, 0x95,
-	0x80, 0x18, 0xE5, 0x3E, 0xC3, 0x95, 0x38, 0x40,
-/*12B0*/0x1D, 0x85, 0x3E, 0x38, 0xE5, 0x3E, 0x60, 0x05,
-	0x85, 0x3F, 0x39, 0x80, 0x03, 0x85, 0x39, 0x39,
-/*12C0*/0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3E, 0x12,
-	0x07, 0x53, 0xE5, 0x3F, 0xF0, 0x22, 0x80, 0x3B,
-/*12D0*/0xE5, 0x65, 0x45, 0x64, 0x70, 0x11, 0x12, 0x07,
-	0x5F, 0x40, 0x05, 0x12, 0x08, 0x9E, 0x80, 0x1F,
-/*12E0*/0x12, 0x07, 0x3E, 0xE5, 0x41, 0xF0, 0x22, 0xE5,
-	0x3C, 0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3C,
-/*12F0*/0x38, 0xE5, 0x3C, 0x60, 0x05, 0x85, 0x3D, 0x39,
-	0x80, 0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12,
-/*1300*/0x07, 0xA8, 0xE5, 0x3C, 0x12, 0x07, 0x53, 0xE5,
-	0x3D, 0xF0, 0x22, 0x12, 0x07, 0x9F, 0xE5, 0x38,
-/*1310*/0x12, 0x07, 0x53, 0xE5, 0x39, 0xF0, 0x22, 0x8C,
-	0x63, 0x8D, 0x64, 0x12, 0x08, 0xDA, 0x40, 0x3C,
-/*1320*/0xE5, 0x64, 0x45, 0x63, 0x70, 0x10, 0x12, 0x09,
-	0x04, 0xC3, 0xE5, 0x3E, 0x12, 0x07, 0x69, 0x40,
-/*1330*/0x3B, 0x12, 0x08, 0x95, 0x80, 0x18, 0xE5, 0x3E,
-	0xC3, 0x95, 0x38, 0x40, 0x1D, 0x85, 0x3E, 0x38,
-/*1340*/0xE5, 0x3E, 0x60, 0x05, 0x85, 0x3F, 0x39, 0x80,
-	0x03, 0x85, 0x39, 0x39, 0x8F, 0x3A, 0x12, 0x07,
-/*1350*/0xA8, 0xE5, 0x3E, 0x12, 0x07, 0x53, 0xE5, 0x3F,
-	0xF0, 0x22, 0x80, 0x3B, 0xE5, 0x64, 0x45, 0x63,
-/*1360*/0x70, 0x11, 0x12, 0x07, 0x5F, 0x40, 0x05, 0x12,
-	0x08, 0x9E, 0x80, 0x1F, 0x12, 0x07, 0x3E, 0xE5,
-/*1370*/0x41, 0xF0, 0x22, 0xE5, 0x3C, 0xC3, 0x95, 0x38,
-	0x40, 0x1D, 0x85, 0x3C, 0x38, 0xE5, 0x3C, 0x60,
-/*1380*/0x05, 0x85, 0x3D, 0x39, 0x80, 0x03, 0x85, 0x39,
-	0x39, 0x8F, 0x3A, 0x12, 0x07, 0xA8, 0xE5, 0x3C,
-/*1390*/0x12, 0x07, 0x53, 0xE5, 0x3D, 0xF0, 0x22, 0x12,
-	0x07, 0x9F, 0xE5, 0x38, 0x12, 0x07, 0x53, 0xE5,
-/*13A0*/0x39, 0xF0, 0x22, 0xE5, 0x0D, 0xFE, 0xE5, 0x08,
-	0x8E, 0x54, 0x44, 0x05, 0xF5, 0x55, 0x75, 0x15,
-/*13B0*/0x0F, 0xF5, 0x82, 0x12, 0x0E, 0x7A, 0x12, 0x17,
-	0xA3, 0x20, 0x31, 0x05, 0x75, 0x15, 0x03, 0x80,
-/*13C0*/0x03, 0x75, 0x15, 0x0B, 0xE5, 0x0A, 0xC3, 0x94,
-	0x01, 0x50, 0x38, 0x12, 0x14, 0x20, 0x20, 0x31,
-/*13D0*/0x06, 0x05, 0x15, 0x05, 0x15, 0x80, 0x04, 0x15,
-	0x15, 0x15, 0x15, 0xE5, 0x0A, 0xC3, 0x94, 0x01,
-/*13E0*/0x50, 0x21, 0x12, 0x14, 0x20, 0x20, 0x31, 0x04,
-	0x05, 0x15, 0x80, 0x02, 0x15, 0x15, 0xE5, 0x0A,
-/*13F0*/0xC3, 0x94, 0x01, 0x50, 0x0E, 0x12, 0x0E, 0x77,
-	0x12, 0x17, 0xA3, 0x20, 0x31, 0x05, 0x05, 0x15,
-/*1400*/0x12, 0x0E, 0x77, 0xE5, 0x15, 0xB4, 0x08, 0x04,
-	0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x15,
-/*1410*/0xB4, 0x07, 0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E,
-	0x00, 0xEE, 0x4F, 0x60, 0x02, 0x05, 0x7F, 0x22,
-/*1420*/0x85, 0x55, 0x82, 0x85, 0x54, 0x83, 0xE5, 0x15,
-	0xF0, 0x12, 0x17, 0xA3, 0x22, 0x12, 0x07, 0x2A,
-/*1430*/0x75, 0x83, 0xAE, 0x74, 0xFF, 0x12, 0x07, 0x29,
-	0xE0, 0x54, 0x1A, 0xF5, 0x34, 0xE0, 0xC4, 0x13,
-/*1440*/0x54, 0x07, 0xF5, 0x35, 0x24, 0xFE, 0x60, 0x24,
-	0x24, 0xFE, 0x60, 0x3C, 0x24, 0x04, 0x70, 0x63,
-/*1450*/0x75, 0x31, 0x2D, 0xE5, 0x08, 0xFD, 0x74, 0xB6,
-	0x12, 0x07, 0x92, 0x74, 0xBC, 0x90, 0x07, 0x22,
-/*1460*/0x12, 0x07, 0x95, 0x74, 0x90, 0x12, 0x07, 0xB3,
-	0x74, 0x92, 0x80, 0x3C, 0x75, 0x31, 0x3A, 0xE5,
-/*1470*/0x08, 0xFD, 0x74, 0xBA, 0x12, 0x07, 0x92, 0x74,
-	0xC0, 0x90, 0x07, 0x22, 0x12, 0x07, 0xB6, 0x74,
-/*1480*/0xC4, 0x12, 0x07, 0xB3, 0x74, 0xC8, 0x80, 0x20,
-	0x75, 0x31, 0x35, 0xE5, 0x08, 0xFD, 0x74, 0xB8,
-/*1490*/0x12, 0x07, 0x92, 0x74, 0xBE, 0xFF, 0xED, 0x44,
-	0x07, 0x90, 0x07, 0x22, 0xCF, 0xF0, 0xA3, 0xEF,
-/*14A0*/0xF0, 0x74, 0xC2, 0x12, 0x07, 0xB3, 0x74, 0xC6,
-	0xFF, 0xED, 0x44, 0x07, 0xA3, 0xCF, 0xF0, 0xA3,
-/*14B0*/0xEF, 0xF0, 0x22, 0x75, 0x34, 0x01, 0x22, 0x8E,
-	0x58, 0x8F, 0x59, 0x8C, 0x5A, 0x8D, 0x5B, 0x8A,
-/*14C0*/0x5C, 0x8B, 0x5D, 0x75, 0x5E, 0x01, 0xE4, 0xF5,
-	0x5F, 0x12, 0x1E, 0xA5, 0x85, 0x59, 0x5E, 0xD3,
-/*14D0*/0xE5, 0x5E, 0x95, 0x5B, 0xE5, 0x5A, 0x12, 0x07,
-	0x6B, 0x50, 0x57, 0xE5, 0x5D, 0x45, 0x5C, 0x70,
-/*14E0*/0x30, 0x12, 0x07, 0x2A, 0x75, 0x83, 0x92, 0xE5,
-	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC6, 0xE5,
-/*14F0*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC8, 0xE5,
-	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0x90, 0xE5,
-/*1500*/0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2, 0xE5,
-	0x5E, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0x80,
-/*1510*/0x03, 0x12, 0x07, 0x32, 0xE5, 0x5E, 0xF0, 0xAF,
-	0x5F, 0x7E, 0x00, 0xAD, 0x5D, 0xAC, 0x5C, 0x12,
-/*1520*/0x04, 0x44, 0xAF, 0x5E, 0x7E, 0x00, 0xAD, 0x5D,
-	0xAC, 0x5C, 0x12, 0x0B, 0xD1, 0x05, 0x5E, 0x02,
-/*1530*/0x14, 0xCF, 0xAB, 0x5D, 0xAA, 0x5C, 0xAD, 0x5B,
-	0xAC, 0x5A, 0xAF, 0x59, 0xAE, 0x58, 0x02, 0x1B,
-/*1540*/0xFB, 0x8C, 0x5C, 0x8D, 0x5D, 0x8A, 0x5E, 0x8B,
-	0x5F, 0x75, 0x60, 0x01, 0xE4, 0xF5, 0x61, 0xF5,
-/*1550*/0x62, 0xF5, 0x63, 0x12, 0x1E, 0xA5, 0x8F, 0x60,
-	0xD3, 0xE5, 0x60, 0x95, 0x5D, 0xE5, 0x5C, 0x12,
-/*1560*/0x07, 0x6B, 0x50, 0x61, 0xE5, 0x5F, 0x45, 0x5E,
-	0x70, 0x27, 0x12, 0x07, 0x2A, 0x75, 0x83, 0xB6,
-/*1570*/0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xB8,
-	0xE5, 0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0xBA,
-/*1580*/0xE5, 0x60, 0xF0, 0xAF, 0x61, 0x7E, 0x00, 0xE5,
-	0x62, 0x12, 0x08, 0x7A, 0x12, 0x0A, 0xFF, 0x80,
-/*1590*/0x19, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE5,
-	0x60, 0x12, 0x07, 0x29, 0x75, 0x83, 0x8E, 0xE4,
-/*15A0*/0x12, 0x07, 0x29, 0x74, 0x01, 0x12, 0x07, 0x29,
-	0xE4, 0xF0, 0xAF, 0x63, 0x7E, 0x00, 0xAD, 0x5F,
-/*15B0*/0xAC, 0x5E, 0x12, 0x04, 0x44, 0xAF, 0x60, 0x7E,
-	0x00, 0xAD, 0x5F, 0xAC, 0x5E, 0x12, 0x12, 0x8B,
-/*15C0*/0x05, 0x60, 0x02, 0x15, 0x58, 0x22, 0x90, 0x11,
-	0x4D, 0xE4, 0x93, 0x90, 0x07, 0x2E, 0xF0, 0x12,
-/*15D0*/0x08, 0x1F, 0x75, 0x83, 0xAE, 0xE0, 0x54, 0x1A,
-	0xF5, 0x34, 0x70, 0x67, 0xEF, 0x44, 0x07, 0xF5,
-/*15E0*/0x82, 0x75, 0x83, 0xCE, 0xE0, 0xFF, 0x13, 0x13,
-	0x13, 0x54, 0x07, 0xF5, 0x36, 0x54, 0x0F, 0xD3,
-/*15F0*/0x94, 0x00, 0x40, 0x06, 0x12, 0x14, 0x2D, 0x12,
-	0x1B, 0xA9, 0xE5, 0x36, 0x54, 0x0F, 0x24, 0xFE,
-/*1600*/0x60, 0x0C, 0x14, 0x60, 0x0C, 0x14, 0x60, 0x19,
-	0x24, 0x03, 0x70, 0x37, 0x80, 0x10, 0x02, 0x1E,
-/*1610*/0x91, 0x12, 0x1E, 0x91, 0x12, 0x07, 0x2A, 0x75,
-	0x83, 0xCE, 0xE0, 0x54, 0xEF, 0xF0, 0x02, 0x1D,
-/*1620*/0xAE, 0x12, 0x10, 0x14, 0xE4, 0xF5, 0x55, 0x12,
-	0x1D, 0x85, 0x05, 0x55, 0xE5, 0x55, 0xC3, 0x94,
-/*1630*/0x05, 0x40, 0xF4, 0x12, 0x07, 0x2A, 0x75, 0x83,
-	0xCE, 0xE0, 0x54, 0xC7, 0x12, 0x07, 0x29, 0xE0,
-/*1640*/0x44, 0x08, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5,
-	0x59, 0xAF, 0x08, 0xEF, 0x44, 0x07, 0xF5, 0x82,
-/*1650*/0x75, 0x83, 0xD0, 0xE0, 0xFD, 0xC4, 0x54, 0x0F,
-	0xF5, 0x5A, 0xEF, 0x44, 0x07, 0xF5, 0x82, 0x75,
-/*1660*/0x83, 0x80, 0x74, 0x01, 0xF0, 0x12, 0x08, 0x21,
-	0x75, 0x83, 0x82, 0xE5, 0x45, 0xF0, 0xEF, 0x44,
-/*1670*/0x07, 0xF5, 0x82, 0x75, 0x83, 0x8A, 0x74, 0xFF,
-	0xF0, 0x12, 0x1A, 0x4D, 0x12, 0x07, 0x2A, 0x75,
-/*1680*/0x83, 0xBC, 0xE0, 0x54, 0xEF, 0x12, 0x07, 0x29,
-	0x75, 0x83, 0xBE, 0xE0, 0x54, 0xEF, 0x12, 0x07,
-/*1690*/0x29, 0x75, 0x83, 0xC0, 0xE0, 0x54, 0xEF, 0x12,
-	0x07, 0x29, 0x75, 0x83, 0xBC, 0xE0, 0x44, 0x10,
-/*16A0*/0x12, 0x07, 0x29, 0x75, 0x83, 0xBE, 0xE0, 0x44,
-	0x10, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC0, 0xE0,
-/*16B0*/0x44, 0x10, 0xF0, 0xAF, 0x58, 0xE5, 0x59, 0x12,
-	0x08, 0x78, 0x02, 0x0A, 0xFF, 0xE4, 0xF5, 0x58,
-/*16C0*/0x7D, 0x01, 0xF5, 0x59, 0xAF, 0x35, 0xFE, 0xFC,
-	0x12, 0x09, 0x15, 0x12, 0x07, 0x2A, 0x75, 0x83,
-/*16D0*/0xB6, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-	0xB8, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-/*16E0*/0xBA, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-	0xBC, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-/*16F0*/0xBE, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-	0xC0, 0x74, 0x10, 0x12, 0x07, 0x29, 0x75, 0x83,
-/*1700*/0x90, 0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC2,
-	0xE4, 0x12, 0x07, 0x29, 0x75, 0x83, 0xC4, 0xE4,
-/*1710*/0x12, 0x07, 0x29, 0x75, 0x83, 0x92, 0xE4, 0x12,
-	0x07, 0x29, 0x75, 0x83, 0xC6, 0xE4, 0x12, 0x07,
-/*1720*/0x29, 0x75, 0x83, 0xC8, 0xE4, 0xF0, 0xAF, 0x58,
-	0xFE, 0xE5, 0x59, 0x12, 0x08, 0x7A, 0x02, 0x0A,
-/*1730*/0xFF, 0xE5, 0xE2, 0x30, 0xE4, 0x6C, 0xE5, 0xE7,
-	0x54, 0xC0, 0x64, 0x40, 0x70, 0x64, 0xE5, 0x09,
-/*1740*/0xC4, 0x54, 0x30, 0xFE, 0xE5, 0x08, 0x25, 0xE0,
-	0x25, 0xE0, 0x54, 0xC0, 0x4E, 0xFE, 0xEF, 0x54,
-/*1750*/0x3F, 0x4E, 0xFD, 0xE5, 0x2B, 0xAE, 0x2A, 0x78,
-	0x02, 0xC3, 0x33, 0xCE, 0x33, 0xCE, 0xD8, 0xF9,
-/*1760*/0xF5, 0x82, 0x8E, 0x83, 0xED, 0xF0, 0xE5, 0x2B,
-	0xAE, 0x2A, 0x78, 0x02, 0xC3, 0x33, 0xCE, 0x33,
-/*1770*/0xCE, 0xD8, 0xF9, 0xFF, 0xF5, 0x82, 0x8E, 0x83,
-	0xA3, 0xE5, 0xFE, 0xF0, 0x8F, 0x82, 0x8E, 0x83,
-/*1780*/0xA3, 0xA3, 0xE5, 0xFD, 0xF0, 0x8F, 0x82, 0x8E,
-	0x83, 0xA3, 0xA3, 0xA3, 0xE5, 0xFC, 0xF0, 0xC3,
-/*1790*/0xE5, 0x2B, 0x94, 0xFA, 0xE5, 0x2A, 0x94, 0x00,
-	0x50, 0x08, 0x05, 0x2B, 0xE5, 0x2B, 0x70, 0x02,
-/*17A0*/0x05, 0x2A, 0x22, 0xE4, 0xFF, 0xE4, 0xF5, 0x58,
-	0xF5, 0x56, 0xF5, 0x57, 0x74, 0x82, 0xFC, 0x12,
-/*17B0*/0x0E, 0x04, 0x8C, 0x83, 0xE0, 0xF5, 0x10, 0x54,
-	0x7F, 0xF0, 0xE5, 0x10, 0x44, 0x80, 0x12, 0x0E,
-/*17C0*/0x98, 0xED, 0xF0, 0x7E, 0x0A, 0x12, 0x0E, 0x04,
-	0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0, 0x26, 0xDE,
-/*17D0*/0xF4, 0x05, 0x57, 0xE5, 0x57, 0x70, 0x02, 0x05,
-	0x56, 0xE5, 0x14, 0x24, 0x01, 0xFD, 0xE4, 0x33,
-/*17E0*/0xFC, 0xD3, 0xE5, 0x57, 0x9D, 0xE5, 0x56, 0x9C,
-	0x40, 0xD9, 0xE5, 0x0A, 0x94, 0x20, 0x50, 0x02,
-/*17F0*/0x05, 0x0A, 0x43, 0xE1, 0x08, 0xC2, 0x31, 0x12,
-	0x0E, 0x04, 0x75, 0x83, 0xA6, 0xE0, 0x55, 0x12,
-/*1800*/0x65, 0x12, 0x70, 0x03, 0xD2, 0x31, 0x22, 0xC2,
-	0x31, 0x22, 0x90, 0x07, 0x26, 0xE0, 0xFA, 0xA3,
-/*1810*/0xE0, 0xF5, 0x82, 0x8A, 0x83, 0xE0, 0xF5, 0x41,
-	0xE5, 0x39, 0xC3, 0x95, 0x41, 0x40, 0x26, 0xE5,
-/*1820*/0x39, 0x95, 0x41, 0xC3, 0x9F, 0xEE, 0x12, 0x07,
-	0x6B, 0x40, 0x04, 0x7C, 0x01, 0x80, 0x02, 0x7C,
-/*1830*/0x00, 0xE5, 0x41, 0x64, 0x3F, 0x60, 0x04, 0x7B,
-	0x01, 0x80, 0x02, 0x7B, 0x00, 0xEC, 0x5B, 0x60,
-/*1840*/0x29, 0x05, 0x41, 0x80, 0x28, 0xC3, 0xE5, 0x41,
-	0x95, 0x39, 0xC3, 0x9F, 0xEE, 0x12, 0x07, 0x6B,
-/*1850*/0x40, 0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00,
-	0xE5, 0x41, 0x60, 0x04, 0x7E, 0x01, 0x80, 0x02,
-/*1860*/0x7E, 0x00, 0xEF, 0x5E, 0x60, 0x04, 0x15, 0x41,
-	0x80, 0x03, 0x85, 0x39, 0x41, 0x85, 0x3A, 0x40,
-/*1870*/0x22, 0xE5, 0xE2, 0x30, 0xE4, 0x60, 0xE5, 0xE1,
-	0x30, 0xE2, 0x5B, 0xE5, 0x09, 0x70, 0x04, 0x7F,
-/*1880*/0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5, 0x08, 0x70,
-	0x04, 0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE,
-/*1890*/0x5F, 0x60, 0x43, 0x53, 0xF9, 0xF8, 0xE5, 0xE2,
-	0x30, 0xE4, 0x3B, 0xE5, 0xE1, 0x30, 0xE2, 0x2E,
-/*18A0*/0x43, 0xFA, 0x02, 0x53, 0xFA, 0xFB, 0xE4, 0xF5,
-	0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0xE5,
-/*18B0*/0xE1, 0x30, 0xE2, 0xE7, 0x90, 0x94, 0x70, 0xE0,
-	0x65, 0x10, 0x60, 0x03, 0x43, 0xFA, 0x04, 0x05,
-/*18C0*/0x10, 0x90, 0x94, 0x70, 0xE5, 0x10, 0xF0, 0x70,
-	0xE6, 0x12, 0x00, 0x06, 0x80, 0xE1, 0x53, 0xFA,
-/*18D0*/0xFD, 0x53, 0xFA, 0xFB, 0x80, 0xC0, 0x22, 0x8F,
-	0x54, 0x12, 0x00, 0x06, 0xE5, 0xE1, 0x30, 0xE0,
-/*18E0*/0x04, 0x7F, 0x01, 0x80, 0x02, 0x7F, 0x00, 0xE5,
-	0x7E, 0xD3, 0x94, 0x05, 0x40, 0x04, 0x7E, 0x01,
-/*18F0*/0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F, 0x60, 0x3D,
-	0x85, 0x54, 0x11, 0xE5, 0xE2, 0x20, 0xE1, 0x32,
-/*1900*/0x74, 0xCE, 0x12, 0x1A, 0x05, 0x30, 0xE7, 0x04,
-	0x7D, 0x01, 0x80, 0x02, 0x7D, 0x00, 0x8F, 0x82,
-/*1910*/0x8E, 0x83, 0xE0, 0x30, 0xE6, 0x04, 0x7F, 0x01,
-	0x80, 0x02, 0x7F, 0x00, 0xEF, 0x5D, 0x70, 0x15,
-/*1920*/0x12, 0x15, 0xC6, 0x74, 0xCE, 0x12, 0x1A, 0x05,
-	0x30, 0xE6, 0x07, 0xE0, 0x44, 0x80, 0xF0, 0x43,
-/*1930*/0xF9, 0x80, 0x12, 0x18, 0x71, 0x22, 0x12, 0x0E,
-	0x44, 0xE5, 0x16, 0x25, 0xE0, 0x25, 0xE0, 0x24,
-/*1940*/0xB0, 0xF5, 0x82, 0xE4, 0x34, 0x1A, 0xF5, 0x83,
-	0xE4, 0x93, 0xF5, 0x0F, 0xE5, 0x16, 0x25, 0xE0,
-/*1950*/0x25, 0xE0, 0x24, 0xB1, 0xF5, 0x82, 0xE4, 0x34,
-	0x1A, 0xF5, 0x83, 0xE4, 0x93, 0xF5, 0x0E, 0x12,
-/*1960*/0x0E, 0x65, 0xF5, 0x10, 0xE5, 0x0F, 0x54, 0xF0,
-	0x12, 0x0E, 0x17, 0x75, 0x83, 0x8C, 0xEF, 0xF0,
-/*1970*/0xE5, 0x0F, 0x30, 0xE0, 0x0C, 0x12, 0x0E, 0x04,
-	0x75, 0x83, 0x86, 0xE0, 0x44, 0x40, 0xF0, 0x80,
-/*1980*/0x0A, 0x12, 0x0E, 0x04, 0x75, 0x83, 0x86, 0xE0,
-	0x54, 0xBF, 0xF0, 0x12, 0x0E, 0x91, 0x75, 0x83,
-/*1990*/0x82, 0xE5, 0x0E, 0xF0, 0x22, 0x7F, 0x05, 0x12,
-	0x17, 0x31, 0x12, 0x0E, 0x04, 0x12, 0x0E, 0x33,
-/*19A0*/0x74, 0x02, 0xF0, 0x74, 0x8E, 0xFE, 0x12, 0x0E,
-	0x04, 0x12, 0x0E, 0x0B, 0xEF, 0xF0, 0x75, 0x15,
-/*19B0*/0x70, 0x12, 0x0F, 0xF7, 0x20, 0x34, 0x05, 0x75,
-	0x15, 0x10, 0x80, 0x03, 0x75, 0x15, 0x50, 0x12,
-/*19C0*/0x0F, 0xF7, 0x20, 0x34, 0x04, 0x74, 0x10, 0x80,
-	0x02, 0x74, 0xF0, 0x25, 0x15, 0xF5, 0x15, 0x12,
-/*19D0*/0x0E, 0x21, 0xEF, 0xF0, 0x12, 0x10, 0x91, 0x20,
-	0x34, 0x17, 0xE5, 0x15, 0x64, 0x30, 0x60, 0x0C,
-/*19E0*/0x74, 0x10, 0x25, 0x15, 0xF5, 0x15, 0xB4, 0x80,
-	0x03, 0xE4, 0xF5, 0x15, 0x12, 0x0E, 0x21, 0xEF,
-/*19F0*/0xF0, 0x22, 0xF0, 0xE5, 0x0B, 0x25, 0xE0, 0x25,
-	0xE0, 0x24, 0x82, 0xF5, 0x82, 0xE4, 0x34, 0x07,
-/*1A00*/0xF5, 0x83, 0x22, 0x74, 0x88, 0xFE, 0xE5, 0x08,
-	0x44, 0x07, 0xFF, 0xF5, 0x82, 0x8E, 0x83, 0xE0,
-/*1A10*/0x22, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0xF5, 0x82,
-	0x22, 0xF0, 0xE0, 0x54, 0xC0, 0x8F, 0x82, 0x8E,
-/*1A20*/0x83, 0xF0, 0x22, 0xEF, 0x44, 0x07, 0xF5, 0x82,
-	0x75, 0x83, 0x86, 0xE0, 0x54, 0x10, 0xD3, 0x94,
-/*1A30*/0x00, 0x22, 0xF0, 0x90, 0x07, 0x15, 0xE0, 0x04,
-	0xF0, 0x22, 0x44, 0x06, 0xF5, 0x82, 0x75, 0x83,
-/*1A40*/0x9E, 0xE0, 0x22, 0xFE, 0xEF, 0x44, 0x07, 0xF5,
-	0x82, 0x8E, 0x83, 0xE0, 0x22, 0xE4, 0x90, 0x07,
-/*1A50*/0x2A, 0xF0, 0xA3, 0xF0, 0x12, 0x07, 0x2A, 0x75,
-	0x83, 0x82, 0xE0, 0x54, 0x7F, 0x12, 0x07, 0x29,
-/*1A60*/0xE0, 0x44, 0x80, 0xF0, 0x12, 0x10, 0xFC, 0x12,
-	0x08, 0x1F, 0x75, 0x83, 0xA0, 0xE0, 0x20, 0xE0,
-/*1A70*/0x1A, 0x90, 0x07, 0x2B, 0xE0, 0x04, 0xF0, 0x70,
-	0x06, 0x90, 0x07, 0x2A, 0xE0, 0x04, 0xF0, 0x90,
-/*1A80*/0x07, 0x2A, 0xE0, 0xB4, 0x10, 0xE1, 0xA3, 0xE0,
-	0xB4, 0x00, 0xDC, 0xEE, 0x44, 0xA6, 0xFC, 0xEF,
-/*1A90*/0x44, 0x07, 0xF5, 0x82, 0x8C, 0x83, 0xE0, 0xF5,
-	0x32, 0xEE, 0x44, 0xA8, 0xFE, 0xEF, 0x44, 0x07,
-/*1AA0*/0xF5, 0x82, 0x8E, 0x83, 0xE0, 0xF5, 0x33, 0x22,
-	0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x90,
-/*1AB0*/0x00, 0x20, 0x0F, 0x92, 0x00, 0x21, 0x0F, 0x94,
-	0x00, 0x22, 0x0F, 0x96, 0x00, 0x23, 0x0F, 0x98,
-/*1AC0*/0x00, 0x24, 0x0F, 0x9A, 0x00, 0x25, 0x0F, 0x9C,
-	0x00, 0x26, 0x0F, 0x9E, 0x00, 0x27, 0x0F, 0xA0,
-/*1AD0*/0x01, 0x20, 0x01, 0xA2, 0x01, 0x21, 0x01, 0xA4,
-	0x01, 0x22, 0x01, 0xA6, 0x01, 0x23, 0x01, 0xA8,
-/*1AE0*/0x01, 0x24, 0x01, 0xAA, 0x01, 0x25, 0x01, 0xAC,
-	0x01, 0x26, 0x01, 0xAE, 0x01, 0x27, 0x01, 0xB0,
-/*1AF0*/0x01, 0x28, 0x01, 0xB4, 0x00, 0x28, 0x0F, 0xB6,
-	0x40, 0x28, 0x0F, 0xB8, 0x61, 0x28, 0x01, 0xCB,
-/*1B00*/0xEF, 0xCB, 0xCA, 0xEE, 0xCA, 0x7F, 0x01, 0xE4,
-	0xFD, 0xEB, 0x4A, 0x70, 0x24, 0xE5, 0x08, 0xF5,
-/*1B10*/0x82, 0x74, 0xB6, 0x12, 0x08, 0x29, 0xE5, 0x08,
-	0xF5, 0x82, 0x74, 0xB8, 0x12, 0x08, 0x29, 0xE5,
-/*1B20*/0x08, 0xF5, 0x82, 0x74, 0xBA, 0x12, 0x08, 0x29,
-	0x7E, 0x00, 0x7C, 0x00, 0x12, 0x0A, 0xFF, 0x80,
-/*1B30*/0x12, 0x90, 0x07, 0x26, 0x12, 0x07, 0x35, 0xE5,
-	0x41, 0xF0, 0x90, 0x07, 0x24, 0x12, 0x07, 0x35,
-/*1B40*/0xE5, 0x40, 0xF0, 0x12, 0x07, 0x2A, 0x75, 0x83,
-	0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74, 0x01, 0x12,
-/*1B50*/0x07, 0x29, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x26,
-	0xF5, 0x27, 0x53, 0xE1, 0xFE, 0xF5, 0x2A, 0x75,
-/*1B60*/0x2B, 0x01, 0xF5, 0x08, 0x7F, 0x01, 0x12, 0x17,
-	0x31, 0x30, 0x30, 0x1C, 0x90, 0x1A, 0xA9, 0xE4,
-/*1B70*/0x93, 0xF5, 0x10, 0x90, 0x1F, 0xF9, 0xE4, 0x93,
-	0xF5, 0x10, 0x90, 0x00, 0x41, 0xE4, 0x93, 0xF5,
-/*1B80*/0x10, 0x90, 0x1E, 0xCA, 0xE4, 0x93, 0xF5, 0x10,
-	0x7F, 0x02, 0x12, 0x17, 0x31, 0x12, 0x0F, 0x54,
-/*1B90*/0x7F, 0x03, 0x12, 0x17, 0x31, 0x12, 0x00, 0x06,
-	0xE5, 0xE2, 0x30, 0xE7, 0x09, 0x12, 0x10, 0x00,
-/*1BA0*/0x30, 0x30, 0x03, 0x12, 0x11, 0x00, 0x02, 0x00,
-	0x47, 0x12, 0x08, 0x1F, 0x75, 0x83, 0xD0, 0xE0,
-/*1BB0*/0xC4, 0x54, 0x0F, 0xFD, 0x75, 0x43, 0x01, 0x75,
-	0x44, 0xFF, 0x12, 0x08, 0xAA, 0x74, 0x04, 0xF0,
-/*1BC0*/0x75, 0x3B, 0x01, 0xED, 0x14, 0x60, 0x0C, 0x14,
-	0x60, 0x0B, 0x14, 0x60, 0x0F, 0x24, 0x03, 0x70,
-/*1BD0*/0x0B, 0x80, 0x09, 0x80, 0x00, 0x12, 0x08, 0xA7,
-	0x04, 0xF0, 0x80, 0x06, 0x12, 0x08, 0xA7, 0x74,
-/*1BE0*/0x04, 0xF0, 0xEE, 0x44, 0x82, 0xFE, 0xEF, 0x44,
-	0x07, 0xF5, 0x82, 0x8E, 0x83, 0xE5, 0x45, 0x12,
-/*1BF0*/0x08, 0xBE, 0x75, 0x83, 0x82, 0xE5, 0x31, 0xF0,
-	0x02, 0x11, 0x4C, 0x8E, 0x60, 0x8F, 0x61, 0x12,
-/*1C00*/0x1E, 0xA5, 0xE4, 0xFF, 0xCE, 0xED, 0xCE, 0xEE,
-	0xD3, 0x95, 0x61, 0xE5, 0x60, 0x12, 0x07, 0x6B,
-/*1C10*/0x40, 0x39, 0x74, 0x20, 0x2E, 0xF5, 0x82, 0xE4,
-	0x34, 0x03, 0xF5, 0x83, 0xE0, 0x70, 0x03, 0xFF,
-/*1C20*/0x80, 0x26, 0x12, 0x08, 0xE2, 0xFD, 0xC3, 0x9F,
-	0x40, 0x1E, 0xCF, 0xED, 0xCF, 0xEB, 0x4A, 0x70,
-/*1C30*/0x0B, 0x8D, 0x42, 0x12, 0x08, 0xEE, 0xF5, 0x41,
-	0x8E, 0x40, 0x80, 0x0C, 0x12, 0x08, 0xE2, 0xF5,
-/*1C40*/0x38, 0x12, 0x08, 0xEE, 0xF5, 0x39, 0x8E, 0x3A,
-	0x1E, 0x80, 0xBC, 0x22, 0x75, 0x58, 0x01, 0xE5,
-/*1C50*/0x35, 0x70, 0x0C, 0x12, 0x07, 0xCC, 0xE0, 0xF5,
-	0x4A, 0x12, 0x07, 0xD8, 0xE0, 0xF5, 0x4C, 0xE5,
-/*1C60*/0x35, 0xB4, 0x04, 0x0C, 0x12, 0x07, 0xE4, 0xE0,
-	0xF5, 0x4A, 0x12, 0x07, 0xF0, 0xE0, 0xF5, 0x4C,
-/*1C70*/0xE5, 0x35, 0xB4, 0x01, 0x04, 0x7F, 0x01, 0x80,
-	0x02, 0x7F, 0x00, 0xE5, 0x35, 0xB4, 0x02, 0x04,
-/*1C80*/0x7E, 0x01, 0x80, 0x02, 0x7E, 0x00, 0xEE, 0x4F,
-	0x60, 0x0C, 0x12, 0x07, 0xFC, 0xE0, 0xF5, 0x4A,
-/*1C90*/0x12, 0x08, 0x08, 0xE0, 0xF5, 0x4C, 0x85, 0x41,
-	0x49, 0x85, 0x40, 0x4B, 0x22, 0x75, 0x5B, 0x01,
-/*1CA0*/0x90, 0x07, 0x24, 0x12, 0x07, 0x35, 0xE0, 0x54,
-	0x1F, 0xFF, 0xD3, 0x94, 0x02, 0x50, 0x04, 0x8F,
-/*1CB0*/0x58, 0x80, 0x05, 0xEF, 0x24, 0xFE, 0xF5, 0x58,
-	0xEF, 0xC3, 0x94, 0x18, 0x40, 0x05, 0x75, 0x59,
-/*1CC0*/0x18, 0x80, 0x04, 0xEF, 0x04, 0xF5, 0x59, 0x85,
-	0x43, 0x5A, 0xAF, 0x58, 0x7E, 0x00, 0xAD, 0x59,
-/*1CD0*/0x7C, 0x00, 0xAB, 0x5B, 0x7A, 0x00, 0x12, 0x15,
-	0x41, 0xAF, 0x5A, 0x7E, 0x00, 0x12, 0x18, 0x0A,
-/*1CE0*/0xAF, 0x5B, 0x7E, 0x00, 0x02, 0x1A, 0xFF, 0xE5,
-	0xE2, 0x30, 0xE7, 0x0E, 0x12, 0x10, 0x03, 0xC2,
-/*1CF0*/0x30, 0x30, 0x30, 0x03, 0x12, 0x10, 0xFF, 0x20,
-	0x33, 0x28, 0xE5, 0xE7, 0x30, 0xE7, 0x05, 0x12,
-/*1D00*/0x0E, 0xA2, 0x80, 0x0D, 0xE5, 0xFE, 0xC3, 0x94,
-	0x20, 0x50, 0x06, 0x12, 0x0E, 0xA2, 0x43, 0xF9,
-/*1D10*/0x08, 0xE5, 0xF2, 0x30, 0xE7, 0x03, 0x53, 0xF9,
-	0x7F, 0xE5, 0xF1, 0x54, 0x70, 0xD3, 0x94, 0x00,
-/*1D20*/0x50, 0xD8, 0x22, 0x12, 0x0E, 0x04, 0x75, 0x83,
-	0x80, 0xE4, 0xF0, 0xE5, 0x08, 0x44, 0x07, 0x12,
-/*1D30*/0x0D, 0xFD, 0x75, 0x83, 0x84, 0x12, 0x0E, 0x02,
-	0x75, 0x83, 0x86, 0x12, 0x0E, 0x02, 0x75, 0x83,
-/*1D40*/0x8C, 0xE0, 0x54, 0xF3, 0x12, 0x0E, 0x03, 0x75,
-	0x83, 0x8E, 0x12, 0x0E, 0x02, 0x75, 0x83, 0x94,
-/*1D50*/0xE0, 0x54, 0xFB, 0xF0, 0x22, 0x12, 0x07, 0x2A,
-	0x75, 0x83, 0x8E, 0xE4, 0x12, 0x07, 0x29, 0x74,
-/*1D60*/0x01, 0x12, 0x07, 0x29, 0xE4, 0x12, 0x08, 0xBE,
-	0x75, 0x83, 0x8C, 0xE0, 0x44, 0x20, 0x12, 0x08,
-/*1D70*/0xBE, 0xE0, 0x54, 0xDF, 0xF0, 0x74, 0x84, 0x85,
-	0x08, 0x82, 0xF5, 0x83, 0xE0, 0x54, 0x7F, 0xF0,
-/*1D80*/0xE0, 0x44, 0x80, 0xF0, 0x22, 0x75, 0x56, 0x01,
-	0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE, 0xFC,
-/*1D90*/0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12, 0x1E,
-	0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E, 0x00,
-/*1DA0*/0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44, 0xAF,
-	0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0x75, 0x56,
-/*1DB0*/0x01, 0xE4, 0xFD, 0xF5, 0x57, 0xAF, 0x35, 0xFE,
-	0xFC, 0x12, 0x09, 0x15, 0x12, 0x1C, 0x9D, 0x12,
-/*1DC0*/0x1E, 0x7A, 0x12, 0x1C, 0x4C, 0xAF, 0x57, 0x7E,
-	0x00, 0xAD, 0x56, 0x7C, 0x00, 0x12, 0x04, 0x44,
-/*1DD0*/0xAF, 0x56, 0x7E, 0x00, 0x02, 0x11, 0xEE, 0xE4,
-	0xF5, 0x16, 0x12, 0x0E, 0x44, 0xFE, 0xE5, 0x08,
-/*1DE0*/0x44, 0x05, 0xFF, 0x12, 0x0E, 0x65, 0x8F, 0x82,
-	0x8E, 0x83, 0xF0, 0x05, 0x16, 0xE5, 0x16, 0xC3,
-/*1DF0*/0x94, 0x14, 0x40, 0xE6, 0xE5, 0x08, 0x12, 0x0E,
-	0x2B, 0xE4, 0xF0, 0x22, 0xE4, 0xF5, 0x58, 0xF5,
-/*1E00*/0x59, 0xF5, 0x5A, 0xFF, 0xFE, 0xAD, 0x58, 0xFC,
-	0x12, 0x09, 0x15, 0x7F, 0x04, 0x7E, 0x00, 0xAD,
-/*1E10*/0x58, 0x7C, 0x00, 0x12, 0x09, 0x15, 0x7F, 0x02,
-	0x7E, 0x00, 0xAD, 0x58, 0x7C, 0x00, 0x02, 0x09,
-/*1E20*/0x15, 0xE5, 0x3C, 0x25, 0x3E, 0xFC, 0xE5, 0x42,
-	0x24, 0x00, 0xFB, 0xE4, 0x33, 0xFA, 0xEC, 0xC3,
-/*1E30*/0x9B, 0xEA, 0x12, 0x07, 0x6B, 0x40, 0x0B, 0x8C,
-	0x42, 0xE5, 0x3D, 0x25, 0x3F, 0xF5, 0x41, 0x8F,
-/*1E40*/0x40, 0x22, 0x12, 0x09, 0x0B, 0x22, 0x74, 0x84,
-	0xF5, 0x18, 0x85, 0x08, 0x19, 0x85, 0x19, 0x82,
-/*1E50*/0x85, 0x18, 0x83, 0xE0, 0x54, 0x7F, 0xF0, 0xE0,
-	0x44, 0x80, 0xF0, 0xE0, 0x44, 0x80, 0xF0, 0x22,
-/*1E60*/0xEF, 0x4E, 0x70, 0x0B, 0x12, 0x07, 0x2A, 0x75,
-	0x83, 0xD2, 0xE0, 0x54, 0xDF, 0xF0, 0x22, 0x12,
-/*1E70*/0x07, 0x2A, 0x75, 0x83, 0xD2, 0xE0, 0x44, 0x20,
-	0xF0, 0x22, 0x75, 0x58, 0x01, 0x90, 0x07, 0x26,
-/*1E80*/0x12, 0x07, 0x35, 0xE0, 0x54, 0x3F, 0xF5, 0x41,
-	0x12, 0x07, 0x32, 0xE0, 0x54, 0x3F, 0xF5, 0x40,
-/*1E90*/0x22, 0x75, 0x56, 0x02, 0xE4, 0xF5, 0x57, 0x12,
-	0x1D, 0xFC, 0xAF, 0x57, 0x7E, 0x00, 0xAD, 0x56,
-/*1EA0*/0x7C, 0x00, 0x02, 0x04, 0x44, 0xE4, 0xF5, 0x42,
-	0xF5, 0x41, 0xF5, 0x40, 0xF5, 0x38, 0xF5, 0x39,
-/*1EB0*/0xF5, 0x3A, 0x22, 0xEF, 0x54, 0x07, 0xFF, 0xE5,
-	0xF9, 0x54, 0xF8, 0x4F, 0xF5, 0xF9, 0x22, 0x7F,
-/*1EC0*/0x01, 0xE4, 0xFE, 0x0F, 0x0E, 0xBE, 0xFF, 0xFB,
-	0x22, 0x01, 0x20, 0x00, 0x01, 0x04, 0x20, 0x00,
-/*1ED0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1EE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1EF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F00*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F10*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F20*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F30*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F40*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F50*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F60*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F70*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F80*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1F90*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FA0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FB0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FC0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FD0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FE0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-/*1FF0*/0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
-	0x01, 0x20, 0x11, 0x00, 0x04, 0x20, 0x00, 0x81
-};
-
-int ipath_sd7220_ib_load(struct ipath_devdata *dd)
-{
-	return ipath_sd7220_prog_ld(dd, IB_7220_SERDES, ipath_sd7220_ib_img,
-		sizeof(ipath_sd7220_ib_img), 0);
-}
-
-int ipath_sd7220_ib_vfy(struct ipath_devdata *dd)
-{
-	return ipath_sd7220_prog_vfy(dd, IB_7220_SERDES, ipath_sd7220_ib_img,
-		sizeof(ipath_sd7220_ib_img), 0);
-}
diff --git a/drivers/infiniband/hw/ipath/ipath_sdma.c b/drivers/infiniband/hw/ipath/ipath_sdma.c
index eaba03273e4..17a517766ad 100644
--- a/drivers/infiniband/hw/ipath/ipath_sdma.c
+++ b/drivers/infiniband/hw/ipath/ipath_sdma.c
@@ -31,6 +31,7 @@
  */
 
 #include <linux/spinlock.h>
+#include <linux/gfp.h>
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
@@ -246,7 +247,7 @@ static void sdma_abort_task(unsigned long opaque)
 
 	/* ipath_sdma_abort() is done, waiting for interrupt */
 	if (status == IPATH_SDMA_ABORT_DISARMED) {
-		if (jiffies < dd->ipath_sdma_abort_intr_timeout)
+		if (time_before(jiffies, dd->ipath_sdma_abort_intr_timeout))
 			goto resched_noprint;
 		/* give up, intr got lost somewhere */
 		ipath_dbg("give up waiting for SDMADISABLED intr\n");
@@ -340,7 +341,7 @@ resched:
 	 * JAG - this is bad to just have default be a loop without
 	 * state change
 	 */
-	if (jiffies > dd->ipath_sdma_abort_jiffies) {
+	if (time_after(jiffies, dd->ipath_sdma_abort_jiffies)) {
 		ipath_dbg("looping with status 0x%08lx\n",
 			  dd->ipath_sdma_status);
 		dd->ipath_sdma_abort_jiffies = jiffies + 5 * HZ;
@@ -698,10 +699,8 @@ retry:
 
 	addr = dma_map_single(&dd->pcidev->dev, tx->txreq.map_addr,
 			      tx->map_len, DMA_TO_DEVICE);
-	if (dma_mapping_error(addr)) {
-		ret = -EIO;
-		goto unlock;
-	}
+	if (dma_mapping_error(&dd->pcidev->dev, addr))
+		goto ioerr;
 
 	dwoffset = tx->map_len >> 2;
 	make_sdma_desc(dd, sdmadesc, (u64) addr, dwoffset, 0);
@@ -741,6 +740,8 @@ retry:
 		dw = (len + 3) >> 2;
 		addr = dma_map_single(&dd->pcidev->dev, sge->vaddr, dw << 2,
 				      DMA_TO_DEVICE);
+		if (dma_mapping_error(&dd->pcidev->dev, addr))
+			goto unmap;
 		make_sdma_desc(dd, sdmadesc, (u64) addr, dw, dwoffset);
 		/* SDmaUseLargeBuf has to be set in every descriptor */
 		if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_USELARGEBUF)
@@ -781,10 +782,10 @@ retry:
 		descqp = &dd->ipath_sdma_descq[dd->ipath_sdma_descq_cnt].qw[0];
 	descqp -= 2;
 	/* SDmaLastDesc */
-	descqp[0] |= __constant_cpu_to_le64(1ULL << 11);
+	descqp[0] |= cpu_to_le64(1ULL << 11);
 	if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_INTREQ) {
 		/* SDmaIntReq */
-		descqp[0] |= __constant_cpu_to_le64(1ULL << 15);
+		descqp[0] |= cpu_to_le64(1ULL << 15);
 	}
 
 	/* Commit writes to memory and advance the tail on the chip */
@@ -798,7 +799,18 @@ retry:
 	list_add_tail(&tx->txreq.list, &dd->ipath_sdma_activelist);
 	if (tx->txreq.flags & IPATH_SDMA_TXREQ_F_VL15)
 		vl15_watchdog_enq(dd);
-
+	goto unlock;
+
+unmap:
+	while (tail != dd->ipath_sdma_descq_tail) {
+		if (!tail)
+			tail = dd->ipath_sdma_descq_cnt - 1;
+		else
+			tail--;
+		unmap_desc(dd, tail);
+	}
+ioerr:
+	ret = -EIO;
 unlock:
 	spin_unlock_irqrestore(&dd->ipath_sdma_lock, flags);
 fail:
diff --git a/drivers/infiniband/hw/ipath/ipath_srq.c b/drivers/infiniband/hw/ipath/ipath_srq.c
index e3d80ca84c1..26271984b71 100644
--- a/drivers/infiniband/hw/ipath/ipath_srq.c
+++ b/drivers/infiniband/hw/ipath/ipath_srq.c
@@ -32,6 +32,7 @@
  */
 
 #include <linux/err.h>
+#include <linux/slab.h>
 #include <linux/vmalloc.h>
 
 #include "ipath_verbs.h"
@@ -106,6 +107,11 @@ struct ib_srq *ipath_create_srq(struct ib_pd *ibpd,
 	u32 sz;
 	struct ib_srq *ret;
 
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
+		ret = ERR_PTR(-ENOSYS);
+		goto done;
+	}
+
 	if (srq_init_attr->attr.max_wr == 0) {
 		ret = ERR_PTR(-EINVAL);
 		goto done;
diff --git a/drivers/infiniband/hw/ipath/ipath_stats.c b/drivers/infiniband/hw/ipath/ipath_stats.c
index c8e3d65f0de..f63e143e329 100644
--- a/drivers/infiniband/hw/ipath/ipath_stats.c
+++ b/drivers/infiniband/hw/ipath/ipath_stats.c
@@ -112,6 +112,14 @@ u64 ipath_snap_cntr(struct ipath_devdata *dd, ipath_creg creg)
 			dd->ipath_lastrpkts = val;
 		}
 		val64 = dd->ipath_rpkts;
+	} else if (creg == dd->ipath_cregs->cr_ibsymbolerrcnt) {
+		if (dd->ibdeltainprog)
+			val64 -= val64 - dd->ibsymsnap;
+		val64 -= dd->ibsymdelta;
+	} else if (creg == dd->ipath_cregs->cr_iblinkerrrecovcnt) {
+		if (dd->ibdeltainprog)
+			val64 -= val64 - dd->iblnkerrsnap;
+		val64 -= dd->iblnkerrdelta;
 	} else
 		val64 = (u64) val;
 
diff --git a/drivers/infiniband/hw/ipath/ipath_sysfs.c b/drivers/infiniband/hw/ipath/ipath_sysfs.c
index a6c8efbdc0c..75558f33f1c 100644
--- a/drivers/infiniband/hw/ipath/ipath_sysfs.c
+++ b/drivers/infiniband/hw/ipath/ipath_sysfs.c
@@ -32,6 +32,7 @@
  */
 
 #include <linux/ctype.h>
+#include <linux/stat.h>
 
 #include "ipath_kernel.h"
 #include "ipath_verbs.h"
@@ -557,6 +558,7 @@ static ssize_t store_reset(struct device *dev,
 		dev_info(dev,"Unit %d is disabled, can't reset\n",
 			 dd->ipath_unit);
 		ret = -EINVAL;
+		goto bail;
 	}
 	ret = ipath_reset_device(dd->ipath_unit);
 bail:
@@ -1069,7 +1071,7 @@ static ssize_t show_tempsense(struct device *dev,
 	return ret;
 }
 
-struct attribute_group *ipath_driver_attr_groups[] = {
+const struct attribute_group *ipath_driver_attr_groups[] = {
 	&driver_attr_group,
 	NULL,
 };
diff --git a/drivers/infiniband/hw/ipath/ipath_uc.c b/drivers/infiniband/hw/ipath/ipath_uc.c
index 82cc588b8bf..22e60998f1a 100644
--- a/drivers/infiniband/hw/ipath/ipath_uc.c
+++ b/drivers/infiniband/hw/ipath/ipath_uc.c
@@ -419,7 +419,7 @@ void ipath_uc_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 		/* Signal completion event if the solicited bit is set. */
 		ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 			       (ohdr->bth[0] &
-				__constant_cpu_to_be32(1 << 23)) != 0);
+				cpu_to_be32(1 << 23)) != 0);
 		break;
 
 	case OP(RDMA_WRITE_FIRST):
diff --git a/drivers/infiniband/hw/ipath/ipath_ud.c b/drivers/infiniband/hw/ipath/ipath_ud.c
index 36aa242c487..e8a2a915251 100644
--- a/drivers/infiniband/hw/ipath/ipath_ud.c
+++ b/drivers/infiniband/hw/ipath/ipath_ud.c
@@ -31,6 +31,7 @@
  * SOFTWARE.
  */
 
+#include <linux/sched.h>
 #include <rdma/ib_smi.h>
 
 #include "ipath_verbs.h"
@@ -70,8 +71,6 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 		goto done;
 	}
 
-	rsge.sg_list = NULL;
-
 	/*
 	 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
 	 * Qkeys with the high order bit set mean use the
@@ -87,7 +86,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	}
 
 	/*
-	 * A GRH is expected to preceed the data even if not
+	 * A GRH is expected to precede the data even if not
 	 * present on the wire.
 	 */
 	length = swqe->length;
@@ -115,21 +114,6 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 		rq = &qp->r_rq;
 	}
 
-	if (rq->max_sge > 1) {
-		/*
-		 * XXX We could use GFP_KERNEL if ipath_do_send()
-		 * was always called from the tasklet instead of
-		 * from ipath_post_send().
-		 */
-		rsge.sg_list = kmalloc((rq->max_sge - 1) *
-					sizeof(struct ipath_sge),
-				       GFP_ATOMIC);
-		if (!rsge.sg_list) {
-			dev->n_pkt_drops++;
-			goto drop;
-		}
-	}
-
 	/*
 	 * Get the next work request entry to find where to put the data.
 	 * Note that it is safe to drop the lock after changing rq->tail
@@ -147,6 +131,7 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 		goto drop;
 	}
 	wqe = get_rwqe_ptr(rq, tail);
+	rsge.sg_list = qp->r_ud_sg_list;
 	if (!ipath_init_sge(qp, wqe, &rlen, &rsge)) {
 		spin_unlock_irqrestore(&rq->lock, flags);
 		dev->n_pkt_drops++;
@@ -242,7 +227,6 @@ static void ipath_ud_loopback(struct ipath_qp *sqp, struct ipath_swqe *swqe)
 	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 		       swqe->wr.send_flags & IB_SEND_SOLICITED);
 drop:
-	kfree(rsge.sg_list);
 	if (atomic_dec_and_test(&qp->refcount))
 		wake_up(&qp->wait);
 done:;
@@ -267,6 +251,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 	u16 lrh0;
 	u16 lid;
 	int ret = 0;
+	int next_cur;
 
 	spin_lock_irqsave(&qp->s_lock, flags);
 
@@ -290,8 +275,9 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 		goto bail;
 
 	wqe = get_swqe_ptr(qp, qp->s_cur);
-	if (++qp->s_cur >= qp->s_size)
-		qp->s_cur = 0;
+	next_cur = qp->s_cur + 1;
+	if (next_cur >= qp->s_size)
+		next_cur = 0;
 
 	/* Construct the header. */
 	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
@@ -315,6 +301,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 				qp->s_flags |= IPATH_S_WAIT_DMA;
 				goto bail;
 			}
+			qp->s_cur = next_cur;
 			spin_unlock_irqrestore(&qp->s_lock, flags);
 			ipath_ud_loopback(qp, wqe);
 			spin_lock_irqsave(&qp->s_lock, flags);
@@ -323,6 +310,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 		}
 	}
 
+	qp->s_cur = next_cur;
 	extra_bytes = -wqe->length & 3;
 	nwords = (wqe->length + extra_bytes) >> 2;
 
@@ -383,7 +371,7 @@ int ipath_make_ud_req(struct ipath_qp *qp)
 	 */
 	ohdr->bth[1] = ah_attr->dlid >= IPATH_MULTICAST_LID_BASE &&
 		ah_attr->dlid != IPATH_PERMISSIVE_LID ?
-		__constant_cpu_to_be32(IPATH_MULTICAST_QPN) :
+		cpu_to_be32(IPATH_MULTICAST_QPN) :
 		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
 	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & IPATH_PSN_MASK);
 	/*
@@ -527,7 +515,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 	}
 
 	/*
-	 * A GRH is expected to preceed the data even if not
+	 * A GRH is expected to precede the data even if not
 	 * present on the wire.
 	 */
 	wc.byte_len = tlen + sizeof(struct ib_grh);
@@ -586,7 +574,7 @@ void ipath_ud_rcv(struct ipath_ibdev *dev, struct ipath_ib_header *hdr,
 	/* Signal completion event if the solicited bit is set. */
 	ipath_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
 		       (ohdr->bth[0] &
-			__constant_cpu_to_be32(1 << 23)) != 0);
+			cpu_to_be32(1 << 23)) != 0);
 
 bail:;
 }
diff --git a/drivers/infiniband/hw/ipath/ipath_user_pages.c b/drivers/infiniband/hw/ipath/ipath_user_pages.c
index 0190edc8044..dc66c450691 100644
--- a/drivers/infiniband/hw/ipath/ipath_user_pages.c
+++ b/drivers/infiniband/hw/ipath/ipath_user_pages.c
@@ -33,6 +33,8 @@
 
 #include <linux/mm.h>
 #include <linux/device.h>
+#include <linux/slab.h>
+#include <linux/sched.h>
 
 #include "ipath_kernel.h"
 
@@ -51,15 +53,14 @@ static void __ipath_release_user_pages(struct page **p, size_t num_pages,
 }
 
 /* call with current->mm->mmap_sem held */
-static int __get_user_pages(unsigned long start_page, size_t num_pages,
-			struct page **p, struct vm_area_struct **vma)
+static int __ipath_get_user_pages(unsigned long start_page, size_t num_pages,
+				  struct page **p, struct vm_area_struct **vma)
 {
 	unsigned long lock_limit;
 	size_t got;
 	int ret;
 
-	lock_limit = current->signal->rlim[RLIMIT_MEMLOCK].rlim_cur >>
-		PAGE_SHIFT;
+	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
 
 	if (num_pages > lock_limit) {
 		ret = -ENOMEM;
@@ -78,7 +79,7 @@ static int __get_user_pages(unsigned long start_page, size_t num_pages,
 			goto bail_release;
 	}
 
-	current->mm->locked_vm += num_pages;
+	current->mm->pinned_vm += num_pages;
 
 	ret = 0;
 	goto bail;
@@ -164,7 +165,7 @@ int ipath_get_user_pages(unsigned long start_page, size_t num_pages,
 
 	down_write(&current->mm->mmap_sem);
 
-	ret = __get_user_pages(start_page, num_pages, p, NULL);
+	ret = __ipath_get_user_pages(start_page, num_pages, p, NULL);
 
 	up_write(&current->mm->mmap_sem);
 
@@ -177,7 +178,7 @@ void ipath_release_user_pages(struct page **p, size_t num_pages)
 
 	__ipath_release_user_pages(p, num_pages, 1);
 
-	current->mm->locked_vm -= num_pages;
+	current->mm->pinned_vm -= num_pages;
 
 	up_write(&current->mm->mmap_sem);
 }
@@ -194,7 +195,7 @@ static void user_pages_account(struct work_struct *_work)
 		container_of(_work, struct ipath_user_pages_work, work);
 
 	down_write(&work->mm->mmap_sem);
-	work->mm->locked_vm -= work->num_pages;
+	work->mm->pinned_vm -= work->num_pages;
 	up_write(&work->mm->mmap_sem);
 	mmput(work->mm);
 	kfree(work);
@@ -209,20 +210,20 @@ void ipath_release_user_pages_on_close(struct page **p, size_t num_pages)
 
 	mm = get_task_mm(current);
 	if (!mm)
-		goto bail;
+		return;
 
 	work = kmalloc(sizeof(*work), GFP_KERNEL);
 	if (!work)
 		goto bail_mm;
 
-	goto bail;
-
 	INIT_WORK(&work->work, user_pages_account);
 	work->mm = mm;
 	work->num_pages = num_pages;
 
+	queue_work(ib_wq, &work->work);
+	return;
+
 bail_mm:
 	mmput(mm);
-bail:
 	return;
 }
diff --git a/drivers/infiniband/hw/ipath/ipath_user_sdma.c b/drivers/infiniband/hw/ipath/ipath_user_sdma.c
index 86e016916cd..cc04b7ba348 100644
--- a/drivers/infiniband/hw/ipath/ipath_user_sdma.c
+++ b/drivers/infiniband/hw/ipath/ipath_user_sdma.c
@@ -33,6 +33,7 @@
 #include <linux/types.h>
 #include <linux/device.h>
 #include <linux/dmapool.h>
+#include <linux/sched.h>
 #include <linux/slab.h>
 #include <linux/list.h>
 #include <linux/highmem.h>
@@ -206,7 +207,7 @@ static int ipath_user_sdma_coalesce(const struct ipath_devdata *dd,
 
 	dma_addr = dma_map_page(&dd->pcidev->dev, page, 0, len,
 				DMA_TO_DEVICE);
-	if (dma_mapping_error(dma_addr)) {
+	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 		ret = -ENOMEM;
 		goto free_unmap;
 	}
@@ -235,7 +236,7 @@ static int ipath_user_sdma_num_pages(const struct iovec *iov)
 	return 1 + ((epage - spage) >> PAGE_SHIFT);
 }
 
-/* truncate length to page boundry */
+/* truncate length to page boundary */
 static int ipath_user_sdma_page_length(unsigned long addr, unsigned long len)
 {
 	const unsigned long offset = addr & ~PAGE_MASK;
@@ -279,9 +280,7 @@ static int ipath_user_sdma_pin_pages(const struct ipath_devdata *dd,
 	int j;
 	int ret;
 
-	ret = get_user_pages(current, current->mm, addr,
-			     npages, 0, 1, pages, NULL);
-
+	ret = get_user_pages_fast(addr, npages, 0, pages);
 	if (ret != npages) {
 		int i;
 
@@ -301,7 +300,7 @@ static int ipath_user_sdma_pin_pages(const struct ipath_devdata *dd,
 				     pages[j], 0, flen, DMA_TO_DEVICE);
 		unsigned long fofs = addr & ~PAGE_MASK;
 
-		if (dma_mapping_error(dma_addr)) {
+		if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 			ret = -ENOMEM;
 			goto done;
 		}
@@ -508,7 +507,7 @@ static int ipath_user_sdma_queue_pkts(const struct ipath_devdata *dd,
 		if (page) {
 			dma_addr = dma_map_page(&dd->pcidev->dev,
 						page, 0, len, DMA_TO_DEVICE);
-			if (dma_mapping_error(dma_addr)) {
+			if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
 				ret = -ENOMEM;
 				goto free_pbc;
 			}
@@ -667,13 +666,13 @@ static inline __le64 ipath_sdma_make_desc0(struct ipath_devdata *dd,
 
 static inline __le64 ipath_sdma_make_first_desc0(__le64 descq)
 {
-	return descq | __constant_cpu_to_le64(1ULL << 12);
+	return descq | cpu_to_le64(1ULL << 12);
 }
 
 static inline __le64 ipath_sdma_make_last_desc0(__le64 descq)
 {
 					      /* last */  /* dma head */
-	return descq | __constant_cpu_to_le64(1ULL << 11 | 1ULL << 13);
+	return descq | cpu_to_le64(1ULL << 11 | 1ULL << 13);
 }
 
 static inline __le64 ipath_sdma_make_desc1(u64 addr)
@@ -763,7 +762,7 @@ static int ipath_user_sdma_push_pkts(struct ipath_devdata *dd,
 		if (ofs >= IPATH_SMALLBUF_DWORDS) {
 			for (i = 0; i < pkt->naddr; i++) {
 				dd->ipath_sdma_descq[dtail].qw[0] |=
-					__constant_cpu_to_le64(1ULL << 14);
+					cpu_to_le64(1ULL << 14);
 				if (++dtail == dd->ipath_sdma_descq_cnt)
 					dtail = 0;
 			}
@@ -810,10 +809,7 @@ int ipath_user_sdma_writev(struct ipath_devdata *dd,
 	while (dim) {
 		const int mxp = 8;
 
-		down_write(&current->mm->mmap_sem);
 		ret = ipath_user_sdma_queue_pkts(dd, pq, &list, iov, dim, mxp);
-		up_write(&current->mm->mmap_sem);
-
 		if (ret <= 0)
 			goto done_unlock;
 		else {
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.c b/drivers/infiniband/hw/ipath/ipath_verbs.c
index 55c71882882..44ea9390417 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.c
@@ -34,6 +34,8 @@
 #include <rdma/ib_mad.h>
 #include <rdma/ib_user_verbs.h>
 #include <linux/io.h>
+#include <linux/slab.h>
+#include <linux/module.h>
 #include <linux/utsname.h>
 #include <linux/rculist.h>
 
@@ -340,9 +342,16 @@ static int ipath_post_one_send(struct ipath_qp *qp, struct ib_send_wr *wr)
 	int acc;
 	int ret;
 	unsigned long flags;
+	struct ipath_devdata *dd = to_idev(qp->ibqp.device)->dd;
 
 	spin_lock_irqsave(&qp->s_lock, flags);
 
+	if (qp->ibqp.qp_type != IB_QPT_SMI &&
+	    !(dd->ipath_flags & IPATH_LINKACTIVE)) {
+		ret = -ENETDOWN;
+		goto bail;
+	}
+
 	/* Check that state is OK to post send. */
 	if (unlikely(!(ib_ipath_state_ops[qp->state] & IPATH_POST_SEND_OK)))
 		goto bail_inval;
@@ -611,7 +620,7 @@ void ipath_ib_rcv(struct ipath_ibdev *dev, void *rhdr, void *data,
 		goto bail;
 	}
 
-	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0x7f;
 	dev->opstats[opcode].n_bytes += tlen;
 	dev->opstats[opcode].n_packets++;
 
@@ -1021,7 +1030,7 @@ static void sdma_complete(void *cookie, int status)
 	struct ipath_verbs_txreq *tx = cookie;
 	struct ipath_qp *qp = tx->qp;
 	struct ipath_ibdev *dev = to_idev(qp->ibqp.device);
-	unsigned int flags;
+	unsigned long flags;
 	enum ib_wc_status ibs = status == IPATH_SDMA_TXREQ_S_OK ?
 		IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR;
 
@@ -1051,7 +1060,7 @@ static void sdma_complete(void *cookie, int status)
 
 static void decrement_dma_busy(struct ipath_qp *qp)
 {
-	unsigned int flags;
+	unsigned long flags;
 
 	if (atomic_dec_and_test(&qp->s_dma_busy)) {
 		spin_lock_irqsave(&qp->s_lock, flags);
@@ -1221,7 +1230,7 @@ static int ipath_verbs_send_pio(struct ipath_qp *qp,
 	unsigned flush_wc;
 	u32 control;
 	int ret;
-	unsigned int flags;
+	unsigned long flags;
 
 	piobuf = ipath_getpiobuf(dd, plen, NULL);
 	if (unlikely(piobuf == NULL)) {
@@ -1578,7 +1587,7 @@ static int ipath_query_port(struct ib_device *ibdev,
 	u64 ibcstat;
 
 	memset(props, 0, sizeof(*props));
-	props->lid = lid ? lid : __constant_be16_to_cpu(IB_LID_PERMISSIVE);
+	props->lid = lid ? lid : be16_to_cpu(IB_LID_PERMISSIVE);
 	props->lmc = dd->ipath_lmc;
 	props->sm_lid = dev->sm_lid;
 	props->sm_sl = dev->sm_sl;
@@ -1845,7 +1854,7 @@ unsigned ipath_get_npkeys(struct ipath_devdata *dd)
 }
 
 /**
- * ipath_get_pkey - return the indexed PKEY from the port 0 PKEY table
+ * ipath_get_pkey - return the indexed PKEY from the port PKEY table
  * @dd: the infinipath device
  * @index: the PKEY index
  */
@@ -1853,6 +1862,7 @@ unsigned ipath_get_pkey(struct ipath_devdata *dd, unsigned index)
 {
 	unsigned ret;
 
+	/* always a kernel port, no locking needed */
 	if (index >= ARRAY_SIZE(dd->ipath_pd[0]->port_pkeys))
 		ret = 0;
 	else
@@ -2173,11 +2183,12 @@ int ipath_register_ib_device(struct ipath_devdata *dd)
 	snprintf(dev->node_desc, sizeof(dev->node_desc),
 		 IPATH_IDSTR " %s", init_utsname()->nodename);
 
-	ret = ib_register_device(dev);
+	ret = ib_register_device(dev, NULL);
 	if (ret)
 		goto err_reg;
 
-	if (ipath_verbs_register_sysfs(dev))
+	ret = ipath_verbs_register_sysfs(dev);
+	if (ret)
 		goto err_class;
 
 	enable_timer(dd);
@@ -2317,15 +2328,15 @@ static int ipath_verbs_register_sysfs(struct ib_device *dev)
 	int i;
 	int ret;
 
-	for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i)
-		if (device_create_file(&dev->dev,
-				       ipath_class_attributes[i])) {
-			ret = 1;
+	for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i) {
+		ret = device_create_file(&dev->dev,
+				       ipath_class_attributes[i]);
+		if (ret)
 			goto bail;
-		}
-
-	ret = 0;
-
+	}
+	return 0;
 bail:
+	for (i = 0; i < ARRAY_SIZE(ipath_class_attributes); ++i)
+		device_remove_file(&dev->dev, ipath_class_attributes[i]);
 	return ret;
 }
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs.h b/drivers/infiniband/hw/ipath/ipath_verbs.h
index 9d12ae8a778..ae6cff4abff 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs.h
+++ b/drivers/infiniband/hw/ipath/ipath_verbs.h
@@ -86,11 +86,11 @@
 #define IB_PMA_SAMPLE_STATUS_RUNNING	0x02
 
 /* Mandatory IB performance counter select values. */
-#define IB_PMA_PORT_XMIT_DATA	__constant_htons(0x0001)
-#define IB_PMA_PORT_RCV_DATA	__constant_htons(0x0002)
-#define IB_PMA_PORT_XMIT_PKTS	__constant_htons(0x0003)
-#define IB_PMA_PORT_RCV_PKTS	__constant_htons(0x0004)
-#define IB_PMA_PORT_XMIT_WAIT	__constant_htons(0x0005)
+#define IB_PMA_PORT_XMIT_DATA	cpu_to_be16(0x0001)
+#define IB_PMA_PORT_RCV_DATA	cpu_to_be16(0x0002)
+#define IB_PMA_PORT_XMIT_PKTS	cpu_to_be16(0x0003)
+#define IB_PMA_PORT_RCV_PKTS	cpu_to_be16(0x0004)
+#define IB_PMA_PORT_XMIT_WAIT	cpu_to_be16(0x0005)
 
 struct ib_reth {
 	__be64 vaddr;
@@ -431,6 +431,7 @@ struct ipath_qp {
 	u32 s_lsn;		/* limit sequence number (credit) */
 	struct ipath_swqe *s_wq;	/* send work queue */
 	struct ipath_swqe *s_wqe;
+	struct ipath_sge *r_ud_sg_list;
 	struct ipath_rq r_rq;		/* receive work queue */
 	struct ipath_sge r_sg_list[0];	/* verified SGEs */
 };
diff --git a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
index d73e3223287..6216ea92385 100644
--- a/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
+++ b/drivers/infiniband/hw/ipath/ipath_verbs_mcast.c
@@ -32,6 +32,8 @@
  */
 
 #include <linux/rculist.h>
+#include <linux/sched.h>
+#include <linux/slab.h>
 
 #include "ipath_verbs.h"
 
diff --git a/drivers/infiniband/hw/mlx4/Kconfig b/drivers/infiniband/hw/mlx4/Kconfig
index 4175a4bd0c7..fc01deac1d3 100644
--- a/drivers/infiniband/hw/mlx4/Kconfig
+++ b/drivers/infiniband/hw/mlx4/Kconfig
@@ -1,5 +1,7 @@
 config MLX4_INFINIBAND
 	tristate "Mellanox ConnectX HCA support"
+	depends on NETDEVICES && ETHERNET && PCI && INET
+	select NET_VENDOR_MELLANOX
 	select MLX4_CORE
 	---help---
 	  This driver provides low-level InfiniBand support for
diff --git a/drivers/infiniband/hw/mlx4/Makefile b/drivers/infiniband/hw/mlx4/Makefile
index 70f09c7826d..f4213b3a8fe 100644
--- a/drivers/infiniband/hw/mlx4/Makefile
+++ b/drivers/infiniband/hw/mlx4/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_MLX4_INFINIBAND)	+= mlx4_ib.o
 
-mlx4_ib-y :=	ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o
+mlx4_ib-y :=	ah.o cq.o doorbell.o mad.o main.o mr.o qp.o srq.o mcg.o cm.o alias_GUID.o sysfs.o
diff --git a/drivers/infiniband/hw/mlx4/ah.c b/drivers/infiniband/hw/mlx4/ah.c
index c75ac9463e2..2d8c3397774 100644
--- a/drivers/infiniband/hw/mlx4/ah.c
+++ b/drivers/infiniband/hw/mlx4/ah.c
@@ -30,64 +30,141 @@
  * SOFTWARE.
  */
 
+#include <rdma/ib_addr.h>
+#include <rdma/ib_cache.h>
+
+#include <linux/slab.h>
+#include <linux/inet.h>
+#include <linux/string.h>
+
 #include "mlx4_ib.h"
 
-struct ib_ah *mlx4_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
+static struct ib_ah *create_ib_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr,
+				  struct mlx4_ib_ah *ah)
 {
 	struct mlx4_dev *dev = to_mdev(pd->device)->dev;
-	struct mlx4_ib_ah *ah;
 
-	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
-	if (!ah)
-		return ERR_PTR(-ENOMEM);
-
-	memset(&ah->av, 0, sizeof ah->av);
-
-	ah->av.port_pd = cpu_to_be32(to_mpd(pd)->pdn | (ah_attr->port_num << 24));
-	ah->av.g_slid  = ah_attr->src_path_bits;
-	ah->av.dlid    = cpu_to_be16(ah_attr->dlid);
-	if (ah_attr->static_rate) {
-		ah->av.stat_rate = ah_attr->static_rate + MLX4_STAT_RATE_OFFSET;
-		while (ah->av.stat_rate > IB_RATE_2_5_GBPS + MLX4_STAT_RATE_OFFSET &&
-		       !(1 << ah->av.stat_rate & dev->caps.stat_rate_support))
-			--ah->av.stat_rate;
-	}
-	ah->av.sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
+	ah->av.ib.port_pd = cpu_to_be32(to_mpd(pd)->pdn | (ah_attr->port_num << 24));
+	ah->av.ib.g_slid  = ah_attr->src_path_bits;
 	if (ah_attr->ah_flags & IB_AH_GRH) {
-		ah->av.g_slid   |= 0x80;
-		ah->av.gid_index = ah_attr->grh.sgid_index;
-		ah->av.hop_limit = ah_attr->grh.hop_limit;
-		ah->av.sl_tclass_flowlabel |=
+		ah->av.ib.g_slid   |= 0x80;
+		ah->av.ib.gid_index = ah_attr->grh.sgid_index;
+		ah->av.ib.hop_limit = ah_attr->grh.hop_limit;
+		ah->av.ib.sl_tclass_flowlabel |=
 			cpu_to_be32((ah_attr->grh.traffic_class << 20) |
 				    ah_attr->grh.flow_label);
-		memcpy(ah->av.dgid, ah_attr->grh.dgid.raw, 16);
+		memcpy(ah->av.ib.dgid, ah_attr->grh.dgid.raw, 16);
 	}
 
+	ah->av.ib.dlid    = cpu_to_be16(ah_attr->dlid);
+	if (ah_attr->static_rate) {
+		ah->av.ib.stat_rate = ah_attr->static_rate + MLX4_STAT_RATE_OFFSET;
+		while (ah->av.ib.stat_rate > IB_RATE_2_5_GBPS + MLX4_STAT_RATE_OFFSET &&
+		       !(1 << ah->av.ib.stat_rate & dev->caps.stat_rate_support))
+			--ah->av.ib.stat_rate;
+	}
+	ah->av.ib.sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 28);
+
 	return &ah->ibah;
 }
 
+static struct ib_ah *create_iboe_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr,
+				    struct mlx4_ib_ah *ah)
+{
+	struct mlx4_ib_dev *ibdev = to_mdev(pd->device);
+	struct mlx4_dev *dev = ibdev->dev;
+	int is_mcast = 0;
+	struct in6_addr in6;
+	u16 vlan_tag;
+
+	memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6));
+	if (rdma_is_multicast_addr(&in6)) {
+		is_mcast = 1;
+		rdma_get_mcast_mac(&in6, ah->av.eth.mac);
+	} else {
+		memcpy(ah->av.eth.mac, ah_attr->dmac, ETH_ALEN);
+	}
+	vlan_tag = ah_attr->vlan_id;
+	if (vlan_tag < 0x1000)
+		vlan_tag |= (ah_attr->sl & 7) << 13;
+	ah->av.eth.port_pd = cpu_to_be32(to_mpd(pd)->pdn | (ah_attr->port_num << 24));
+	ah->av.eth.gid_index = ah_attr->grh.sgid_index;
+	ah->av.eth.vlan = cpu_to_be16(vlan_tag);
+	if (ah_attr->static_rate) {
+		ah->av.eth.stat_rate = ah_attr->static_rate + MLX4_STAT_RATE_OFFSET;
+		while (ah->av.eth.stat_rate > IB_RATE_2_5_GBPS + MLX4_STAT_RATE_OFFSET &&
+		       !(1 << ah->av.eth.stat_rate & dev->caps.stat_rate_support))
+			--ah->av.eth.stat_rate;
+	}
+
+	/*
+	 * HW requires multicast LID so we just choose one.
+	 */
+	if (is_mcast)
+		ah->av.ib.dlid = cpu_to_be16(0xc000);
+
+	memcpy(ah->av.eth.dgid, ah_attr->grh.dgid.raw, 16);
+	ah->av.eth.sl_tclass_flowlabel = cpu_to_be32(ah_attr->sl << 29);
+
+	return &ah->ibah;
+}
+
+struct ib_ah *mlx4_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
+{
+	struct mlx4_ib_ah *ah;
+	struct ib_ah *ret;
+
+	ah = kzalloc(sizeof *ah, GFP_ATOMIC);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	if (rdma_port_get_link_layer(pd->device, ah_attr->port_num) == IB_LINK_LAYER_ETHERNET) {
+		if (!(ah_attr->ah_flags & IB_AH_GRH)) {
+			ret = ERR_PTR(-EINVAL);
+		} else {
+			/*
+			 * TBD: need to handle the case when we get
+			 * called in an atomic context and there we
+			 * might sleep.  We don't expect this
+			 * currently since we're working with link
+			 * local addresses which we can translate
+			 * without going to sleep.
+			 */
+			ret = create_iboe_ah(pd, ah_attr, ah);
+		}
+
+		if (IS_ERR(ret))
+			kfree(ah);
+
+		return ret;
+	} else
+		return create_ib_ah(pd, ah_attr, ah); /* never fails */
+}
+
 int mlx4_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
 {
 	struct mlx4_ib_ah *ah = to_mah(ibah);
+	enum rdma_link_layer ll;
 
 	memset(ah_attr, 0, sizeof *ah_attr);
-	ah_attr->dlid	       = be16_to_cpu(ah->av.dlid);
-	ah_attr->sl	       = be32_to_cpu(ah->av.sl_tclass_flowlabel) >> 28;
-	ah_attr->port_num      = be32_to_cpu(ah->av.port_pd) >> 24;
-	if (ah->av.stat_rate)
-		ah_attr->static_rate = ah->av.stat_rate - MLX4_STAT_RATE_OFFSET;
-	ah_attr->src_path_bits = ah->av.g_slid & 0x7F;
+	ah_attr->sl = be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
+	ah_attr->port_num = be32_to_cpu(ah->av.ib.port_pd) >> 24;
+	ll = rdma_port_get_link_layer(ibah->device, ah_attr->port_num);
+	ah_attr->dlid = ll == IB_LINK_LAYER_INFINIBAND ? be16_to_cpu(ah->av.ib.dlid) : 0;
+	if (ah->av.ib.stat_rate)
+		ah_attr->static_rate = ah->av.ib.stat_rate - MLX4_STAT_RATE_OFFSET;
+	ah_attr->src_path_bits = ah->av.ib.g_slid & 0x7F;
 
 	if (mlx4_ib_ah_grh_present(ah)) {
 		ah_attr->ah_flags = IB_AH_GRH;
 
 		ah_attr->grh.traffic_class =
-			be32_to_cpu(ah->av.sl_tclass_flowlabel) >> 20;
+			be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 20;
 		ah_attr->grh.flow_label =
-			be32_to_cpu(ah->av.sl_tclass_flowlabel) & 0xfffff;
-		ah_attr->grh.hop_limit  = ah->av.hop_limit;
-		ah_attr->grh.sgid_index = ah->av.gid_index;
-		memcpy(ah_attr->grh.dgid.raw, ah->av.dgid, 16);
+			be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) & 0xfffff;
+		ah_attr->grh.hop_limit  = ah->av.ib.hop_limit;
+		ah_attr->grh.sgid_index = ah->av.ib.gid_index;
+		memcpy(ah_attr->grh.dgid.raw, ah->av.ib.dgid, 16);
 	}
 
 	return 0;
diff --git a/drivers/infiniband/hw/mlx4/alias_GUID.c b/drivers/infiniband/hw/mlx4/alias_GUID.c
new file mode 100644
index 00000000000..0eb141c4141
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/alias_GUID.c
@@ -0,0 +1,688 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+ /***********************************************************/
+/*This file support the handling of the Alias GUID feature. */
+/***********************************************************/
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_cache.h>
+#include <rdma/ib_sa.h>
+#include <rdma/ib_pack.h>
+#include <linux/mlx4/cmd.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <rdma/ib_user_verbs.h>
+#include <linux/delay.h>
+#include "mlx4_ib.h"
+
+/*
+The driver keeps the current state of all guids, as they are in the HW.
+Whenever we receive an smp mad GUIDInfo record, the data will be cached.
+*/
+
+struct mlx4_alias_guid_work_context {
+	u8 port;
+	struct mlx4_ib_dev     *dev ;
+	struct ib_sa_query     *sa_query;
+	struct completion	done;
+	int			query_id;
+	struct list_head	list;
+	int			block_num;
+};
+
+struct mlx4_next_alias_guid_work {
+	u8 port;
+	u8 block_num;
+	struct mlx4_sriov_alias_guid_info_rec_det rec_det;
+};
+
+
+void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev, int block_num,
+					 u8 port_num, u8 *p_data)
+{
+	int i;
+	u64 guid_indexes;
+	int slave_id;
+	int port_index = port_num - 1;
+
+	if (!mlx4_is_master(dev->dev))
+		return;
+
+	guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+				   ports_guid[port_num - 1].
+				   all_rec_per_port[block_num].guid_indexes);
+	pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+
+	for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+		/* The location of the specific index starts from bit number 4
+		 * until bit num 11 */
+		if (test_bit(i + 4, (unsigned long *)&guid_indexes)) {
+			slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+			if (slave_id >= dev->dev->num_slaves) {
+				pr_debug("The last slave: %d\n", slave_id);
+				return;
+			}
+
+			/* cache the guid: */
+			memcpy(&dev->sriov.demux[port_index].guid_cache[slave_id],
+			       &p_data[i * GUID_REC_SIZE],
+			       GUID_REC_SIZE);
+		} else
+			pr_debug("Guid number: %d in block: %d"
+				 " was not updated\n", i, block_num);
+	}
+}
+
+static __be64 get_cached_alias_guid(struct mlx4_ib_dev *dev, int port, int index)
+{
+	if (index >= NUM_ALIAS_GUID_PER_PORT) {
+		pr_err("%s: ERROR: asked for index:%d\n", __func__, index);
+		return (__force __be64) -1;
+	}
+	return *(__be64 *)&dev->sriov.demux[port - 1].guid_cache[index];
+}
+
+
+ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index)
+{
+	return IB_SA_COMP_MASK(4 + index);
+}
+
+/*
+ * Whenever new GUID is set/unset (guid table change) create event and
+ * notify the relevant slave (master also should be notified).
+ * If the GUID value is not as we have in the cache the slave will not be
+ * updated; in this case it waits for the smp_snoop or the port management
+ * event to call the function and to update the slave.
+ * block_number - the index of the block (16 blocks available)
+ * port_number - 1 or 2
+ */
+void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
+					  int block_num, u8 port_num,
+					  u8 *p_data)
+{
+	int i;
+	u64 guid_indexes;
+	int slave_id;
+	enum slave_port_state new_state;
+	enum slave_port_state prev_state;
+	__be64 tmp_cur_ag, form_cache_ag;
+	enum slave_port_gen_event gen_event;
+
+	if (!mlx4_is_master(dev->dev))
+		return;
+
+	guid_indexes = be64_to_cpu((__force __be64) dev->sriov.alias_guid.
+				   ports_guid[port_num - 1].
+				   all_rec_per_port[block_num].guid_indexes);
+	pr_debug("port: %d, guid_indexes: 0x%llx\n", port_num, guid_indexes);
+
+	/*calculate the slaves and notify them*/
+	for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+		/* the location of the specific index runs from bits 4..11 */
+		if (!(test_bit(i + 4, (unsigned long *)&guid_indexes)))
+			continue;
+
+		slave_id = (block_num * NUM_ALIAS_GUID_IN_REC) + i ;
+		if (slave_id >= dev->dev->num_vfs + 1)
+			return;
+		tmp_cur_ag = *(__be64 *)&p_data[i * GUID_REC_SIZE];
+		form_cache_ag = get_cached_alias_guid(dev, port_num,
+					(NUM_ALIAS_GUID_IN_REC * block_num) + i);
+		/*
+		 * Check if guid is not the same as in the cache,
+		 * If it is different, wait for the snoop_smp or the port mgmt
+		 * change event to update the slave on its port state change
+		 */
+		if (tmp_cur_ag != form_cache_ag)
+			continue;
+		mlx4_gen_guid_change_eqe(dev->dev, slave_id, port_num);
+
+		/*2 cases: Valid GUID, and Invalid Guid*/
+
+		if (tmp_cur_ag != MLX4_NOT_SET_GUID) { /*valid GUID*/
+			prev_state = mlx4_get_slave_port_state(dev->dev, slave_id, port_num);
+			new_state = set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+								  MLX4_PORT_STATE_IB_PORT_STATE_EVENT_GID_VALID,
+								  &gen_event);
+			pr_debug("slave: %d, port: %d prev_port_state: %d,"
+				 " new_port_state: %d, gen_event: %d\n",
+				 slave_id, port_num, prev_state, new_state, gen_event);
+			if (gen_event == SLAVE_PORT_GEN_EVENT_UP) {
+				pr_debug("sending PORT_UP event to slave: %d, port: %d\n",
+					 slave_id, port_num);
+				mlx4_gen_port_state_change_eqe(dev->dev, slave_id,
+							       port_num, MLX4_PORT_CHANGE_SUBTYPE_ACTIVE);
+			}
+		} else { /* request to invalidate GUID */
+			set_and_calc_slave_port_state(dev->dev, slave_id, port_num,
+						      MLX4_PORT_STATE_IB_EVENT_GID_INVALID,
+						      &gen_event);
+			pr_debug("sending PORT DOWN event to slave: %d, port: %d\n",
+				 slave_id, port_num);
+			mlx4_gen_port_state_change_eqe(dev->dev, slave_id, port_num,
+						       MLX4_PORT_CHANGE_SUBTYPE_DOWN);
+		}
+	}
+}
+
+static void aliasguid_query_handler(int status,
+				    struct ib_sa_guidinfo_rec *guid_rec,
+				    void *context)
+{
+	struct mlx4_ib_dev *dev;
+	struct mlx4_alias_guid_work_context *cb_ctx = context;
+	u8 port_index ;
+	int i;
+	struct mlx4_sriov_alias_guid_info_rec_det *rec;
+	unsigned long flags, flags1;
+
+	if (!context)
+		return;
+
+	dev = cb_ctx->dev;
+	port_index = cb_ctx->port - 1;
+	rec = &dev->sriov.alias_guid.ports_guid[port_index].
+		all_rec_per_port[cb_ctx->block_num];
+
+	if (status) {
+		rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+		pr_debug("(port: %d) failed: status = %d\n",
+			 cb_ctx->port, status);
+		goto out;
+	}
+
+	if (guid_rec->block_num != cb_ctx->block_num) {
+		pr_err("block num mismatch: %d != %d\n",
+		       cb_ctx->block_num, guid_rec->block_num);
+		goto out;
+	}
+
+	pr_debug("lid/port: %d/%d, block_num: %d\n",
+		 be16_to_cpu(guid_rec->lid), cb_ctx->port,
+		 guid_rec->block_num);
+
+	rec = &dev->sriov.alias_guid.ports_guid[port_index].
+		all_rec_per_port[guid_rec->block_num];
+
+	rec->status = MLX4_GUID_INFO_STATUS_SET;
+	rec->method = MLX4_GUID_INFO_RECORD_SET;
+
+	for (i = 0 ; i < NUM_ALIAS_GUID_IN_REC; i++) {
+		__be64 tmp_cur_ag;
+		tmp_cur_ag = *(__be64 *)&guid_rec->guid_info_list[i * GUID_REC_SIZE];
+		/* check if the SM didn't assign one of the records.
+		 * if it didn't, if it was not sysadmin request:
+		 * ask the SM to give a new GUID, (instead of the driver request).
+		 */
+		if (tmp_cur_ag == MLX4_NOT_SET_GUID) {
+			mlx4_ib_warn(&dev->ib_dev, "%s:Record num %d in "
+				     "block_num: %d was declined by SM, "
+				     "ownership by %d (0 = driver, 1=sysAdmin,"
+				     " 2=None)\n", __func__, i,
+				     guid_rec->block_num, rec->ownership);
+			if (rec->ownership == MLX4_GUID_DRIVER_ASSIGN) {
+				/* if it is driver assign, asks for new GUID from SM*/
+				*(__be64 *)&rec->all_recs[i * GUID_REC_SIZE] =
+					MLX4_NOT_SET_GUID;
+
+				/* Mark the record as not assigned, and let it
+				 * be sent again in the next work sched.*/
+				rec->status = MLX4_GUID_INFO_STATUS_IDLE;
+				rec->guid_indexes |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
+			}
+		} else {
+		       /* properly assigned record. */
+		       /* We save the GUID we just got from the SM in the
+			* admin_guid in order to be persistent, and in the
+			* request from the sm the process will ask for the same GUID */
+			if (rec->ownership == MLX4_GUID_SYSADMIN_ASSIGN &&
+			    tmp_cur_ag != *(__be64 *)&rec->all_recs[i * GUID_REC_SIZE]) {
+				/* the sysadmin assignment failed.*/
+				mlx4_ib_warn(&dev->ib_dev, "%s: Failed to set"
+					     " admin guid after SysAdmin "
+					     "configuration. "
+					     "Record num %d in block_num:%d "
+					     "was declined by SM, "
+					     "new val(0x%llx) was kept\n",
+					      __func__, i,
+					     guid_rec->block_num,
+					     be64_to_cpu(*(__be64 *) &
+							 rec->all_recs[i * GUID_REC_SIZE]));
+			} else {
+				memcpy(&rec->all_recs[i * GUID_REC_SIZE],
+				       &guid_rec->guid_info_list[i * GUID_REC_SIZE],
+				       GUID_REC_SIZE);
+			}
+		}
+	}
+	/*
+	The func is call here to close the cases when the
+	sm doesn't send smp, so in the sa response the driver
+	notifies the slave.
+	*/
+	mlx4_ib_notify_slaves_on_guid_change(dev, guid_rec->block_num,
+					     cb_ctx->port,
+					     guid_rec->guid_info_list);
+out:
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	if (!dev->sriov.is_going_down)
+		queue_delayed_work(dev->sriov.alias_guid.ports_guid[port_index].wq,
+				   &dev->sriov.alias_guid.ports_guid[port_index].
+				   alias_guid_work, 0);
+	if (cb_ctx->sa_query) {
+		list_del(&cb_ctx->list);
+		kfree(cb_ctx);
+	} else
+		complete(&cb_ctx->done);
+	spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+static void invalidate_guid_record(struct mlx4_ib_dev *dev, u8 port, int index)
+{
+	int i;
+	u64 cur_admin_val;
+	ib_sa_comp_mask comp_mask = 0;
+
+	dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].status
+		= MLX4_GUID_INFO_STATUS_IDLE;
+	dev->sriov.alias_guid.ports_guid[port - 1].all_rec_per_port[index].method
+		= MLX4_GUID_INFO_RECORD_SET;
+
+	/* calculate the comp_mask for that record.*/
+	for (i = 0; i < NUM_ALIAS_GUID_IN_REC; i++) {
+		cur_admin_val =
+			*(u64 *)&dev->sriov.alias_guid.ports_guid[port - 1].
+			all_rec_per_port[index].all_recs[GUID_REC_SIZE * i];
+		/*
+		check the admin value: if it's for delete (~00LL) or
+		it is the first guid of the first record (hw guid) or
+		the records is not in ownership of the sysadmin and the sm doesn't
+		need to assign GUIDs, then don't put it up for assignment.
+		*/
+		if (MLX4_GUID_FOR_DELETE_VAL == cur_admin_val ||
+		    (!index && !i) ||
+		    MLX4_GUID_NONE_ASSIGN == dev->sriov.alias_guid.
+		    ports_guid[port - 1].all_rec_per_port[index].ownership)
+			continue;
+		comp_mask |= mlx4_ib_get_aguid_comp_mask_from_ix(i);
+	}
+	dev->sriov.alias_guid.ports_guid[port - 1].
+		all_rec_per_port[index].guid_indexes = comp_mask;
+}
+
+static int set_guid_rec(struct ib_device *ibdev,
+			u8 port, int index,
+			struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+{
+	int err;
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	struct ib_sa_guidinfo_rec guid_info_rec;
+	ib_sa_comp_mask comp_mask;
+	struct ib_port_attr attr;
+	struct mlx4_alias_guid_work_context *callback_context;
+	unsigned long resched_delay, flags, flags1;
+	struct list_head *head =
+		&dev->sriov.alias_guid.ports_guid[port - 1].cb_list;
+
+	err = __mlx4_ib_query_port(ibdev, port, &attr, 1);
+	if (err) {
+		pr_debug("mlx4_ib_query_port failed (err: %d), port: %d\n",
+			 err, port);
+		return err;
+	}
+	/*check the port was configured by the sm, otherwise no need to send */
+	if (attr.state != IB_PORT_ACTIVE) {
+		pr_debug("port %d not active...rescheduling\n", port);
+		resched_delay = 5 * HZ;
+		err = -EAGAIN;
+		goto new_schedule;
+	}
+
+	callback_context = kmalloc(sizeof *callback_context, GFP_KERNEL);
+	if (!callback_context) {
+		err = -ENOMEM;
+		resched_delay = HZ * 5;
+		goto new_schedule;
+	}
+	callback_context->port = port;
+	callback_context->dev = dev;
+	callback_context->block_num = index;
+
+	memset(&guid_info_rec, 0, sizeof (struct ib_sa_guidinfo_rec));
+
+	guid_info_rec.lid = cpu_to_be16(attr.lid);
+	guid_info_rec.block_num = index;
+
+	memcpy(guid_info_rec.guid_info_list, rec_det->all_recs,
+	       GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC);
+	comp_mask = IB_SA_GUIDINFO_REC_LID | IB_SA_GUIDINFO_REC_BLOCK_NUM |
+		rec_det->guid_indexes;
+
+	init_completion(&callback_context->done);
+	spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	list_add_tail(&callback_context->list, head);
+	spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+
+	callback_context->query_id =
+		ib_sa_guid_info_rec_query(dev->sriov.alias_guid.sa_client,
+					  ibdev, port, &guid_info_rec,
+					  comp_mask, rec_det->method, 1000,
+					  GFP_KERNEL, aliasguid_query_handler,
+					  callback_context,
+					  &callback_context->sa_query);
+	if (callback_context->query_id < 0) {
+		pr_debug("ib_sa_guid_info_rec_query failed, query_id: "
+			 "%d. will reschedule to the next 1 sec.\n",
+			 callback_context->query_id);
+		spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+		list_del(&callback_context->list);
+		kfree(callback_context);
+		spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+		resched_delay = 1 * HZ;
+		err = -EAGAIN;
+		goto new_schedule;
+	}
+	err = 0;
+	goto out;
+
+new_schedule:
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	invalidate_guid_record(dev, port, index);
+	if (!dev->sriov.is_going_down) {
+		queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+				   &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+				   resched_delay);
+	}
+	spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+
+out:
+	return err;
+}
+
+void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port)
+{
+	int i;
+	unsigned long flags, flags1;
+
+	pr_debug("port %d\n", port);
+
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	for (i = 0; i < NUM_ALIAS_GUID_REC_IN_PORT; i++)
+		invalidate_guid_record(dev, port, i);
+
+	if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down) {
+		/*
+		make sure no work waits in the queue, if the work is already
+		queued(not on the timer) the cancel will fail. That is not a problem
+		because we just want the work started.
+		*/
+		cancel_delayed_work(&dev->sriov.alias_guid.
+				      ports_guid[port - 1].alias_guid_work);
+		queue_delayed_work(dev->sriov.alias_guid.ports_guid[port - 1].wq,
+				   &dev->sriov.alias_guid.ports_guid[port - 1].alias_guid_work,
+				   0);
+	}
+	spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+/* The function returns the next record that was
+ * not configured (or failed to be configured) */
+static int get_next_record_to_update(struct mlx4_ib_dev *dev, u8 port,
+				     struct mlx4_next_alias_guid_work *rec)
+{
+	int j;
+	unsigned long flags;
+
+	for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+		spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags);
+		if (dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status ==
+		    MLX4_GUID_INFO_STATUS_IDLE) {
+			memcpy(&rec->rec_det,
+			       &dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j],
+			       sizeof (struct mlx4_sriov_alias_guid_info_rec_det));
+			rec->port = port;
+			rec->block_num = j;
+			dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[j].status =
+				MLX4_GUID_INFO_STATUS_PENDING;
+			spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+			return 0;
+		}
+		spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags);
+	}
+	return -ENOENT;
+}
+
+static void set_administratively_guid_record(struct mlx4_ib_dev *dev, int port,
+					     int rec_index,
+					     struct mlx4_sriov_alias_guid_info_rec_det *rec_det)
+{
+	dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].guid_indexes =
+		rec_det->guid_indexes;
+	memcpy(dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].all_recs,
+	       rec_det->all_recs, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+	dev->sriov.alias_guid.ports_guid[port].all_rec_per_port[rec_index].status =
+		rec_det->status;
+}
+
+static void set_all_slaves_guids(struct mlx4_ib_dev *dev, int port)
+{
+	int j;
+	struct mlx4_sriov_alias_guid_info_rec_det rec_det ;
+
+	for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT ; j++) {
+		memset(rec_det.all_recs, 0, NUM_ALIAS_GUID_IN_REC * GUID_REC_SIZE);
+		rec_det.guid_indexes = (!j ? 0 : IB_SA_GUIDINFO_REC_GID0) |
+			IB_SA_GUIDINFO_REC_GID1 | IB_SA_GUIDINFO_REC_GID2 |
+			IB_SA_GUIDINFO_REC_GID3 | IB_SA_GUIDINFO_REC_GID4 |
+			IB_SA_GUIDINFO_REC_GID5 | IB_SA_GUIDINFO_REC_GID6 |
+			IB_SA_GUIDINFO_REC_GID7;
+		rec_det.status = MLX4_GUID_INFO_STATUS_IDLE;
+		set_administratively_guid_record(dev, port, j, &rec_det);
+	}
+}
+
+static void alias_guid_work(struct work_struct *work)
+{
+	struct delayed_work *delay = to_delayed_work(work);
+	int ret = 0;
+	struct mlx4_next_alias_guid_work *rec;
+	struct mlx4_sriov_alias_guid_port_rec_det *sriov_alias_port =
+		container_of(delay, struct mlx4_sriov_alias_guid_port_rec_det,
+			     alias_guid_work);
+	struct mlx4_sriov_alias_guid *sriov_alias_guid = sriov_alias_port->parent;
+	struct mlx4_ib_sriov *ib_sriov = container_of(sriov_alias_guid,
+						struct mlx4_ib_sriov,
+						alias_guid);
+	struct mlx4_ib_dev *dev = container_of(ib_sriov, struct mlx4_ib_dev, sriov);
+
+	rec = kzalloc(sizeof *rec, GFP_KERNEL);
+	if (!rec) {
+		pr_err("alias_guid_work: No Memory\n");
+		return;
+	}
+
+	pr_debug("starting [port: %d]...\n", sriov_alias_port->port + 1);
+	ret = get_next_record_to_update(dev, sriov_alias_port->port, rec);
+	if (ret) {
+		pr_debug("No more records to update.\n");
+		goto out;
+	}
+
+	set_guid_rec(&dev->ib_dev, rec->port + 1, rec->block_num,
+		     &rec->rec_det);
+
+out:
+	kfree(rec);
+}
+
+
+void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port)
+{
+	unsigned long flags, flags1;
+
+	if (!mlx4_is_master(dev->dev))
+		return;
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	spin_lock_irqsave(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	if (!dev->sriov.is_going_down) {
+		queue_delayed_work(dev->sriov.alias_guid.ports_guid[port].wq,
+			   &dev->sriov.alias_guid.ports_guid[port].alias_guid_work, 0);
+	}
+	spin_unlock_irqrestore(&dev->sriov.alias_guid.ag_work_lock, flags1);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev)
+{
+	int i;
+	struct mlx4_ib_sriov *sriov = &dev->sriov;
+	struct mlx4_alias_guid_work_context *cb_ctx;
+	struct mlx4_sriov_alias_guid_port_rec_det *det;
+	struct ib_sa_query *sa_query;
+	unsigned long flags;
+
+	for (i = 0 ; i < dev->num_ports; i++) {
+		cancel_delayed_work(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work);
+		det = &sriov->alias_guid.ports_guid[i];
+		spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+		while (!list_empty(&det->cb_list)) {
+			cb_ctx = list_entry(det->cb_list.next,
+					    struct mlx4_alias_guid_work_context,
+					    list);
+			sa_query = cb_ctx->sa_query;
+			cb_ctx->sa_query = NULL;
+			list_del(&cb_ctx->list);
+			spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+			ib_sa_cancel_query(cb_ctx->query_id, sa_query);
+			wait_for_completion(&cb_ctx->done);
+			kfree(cb_ctx);
+			spin_lock_irqsave(&sriov->alias_guid.ag_work_lock, flags);
+		}
+		spin_unlock_irqrestore(&sriov->alias_guid.ag_work_lock, flags);
+	}
+	for (i = 0 ; i < dev->num_ports; i++) {
+		flush_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+		destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+	}
+	ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+	kfree(dev->sriov.alias_guid.sa_client);
+}
+
+int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev)
+{
+	char alias_wq_name[15];
+	int ret = 0;
+	int i, j, k;
+	union ib_gid gid;
+
+	if (!mlx4_is_master(dev->dev))
+		return 0;
+	dev->sriov.alias_guid.sa_client =
+		kzalloc(sizeof *dev->sriov.alias_guid.sa_client, GFP_KERNEL);
+	if (!dev->sriov.alias_guid.sa_client)
+		return -ENOMEM;
+
+	ib_sa_register_client(dev->sriov.alias_guid.sa_client);
+
+	spin_lock_init(&dev->sriov.alias_guid.ag_work_lock);
+
+	for (i = 1; i <= dev->num_ports; ++i) {
+		if (dev->ib_dev.query_gid(&dev->ib_dev , i, 0, &gid)) {
+			ret = -EFAULT;
+			goto err_unregister;
+		}
+	}
+
+	for (i = 0 ; i < dev->num_ports; i++) {
+		memset(&dev->sriov.alias_guid.ports_guid[i], 0,
+		       sizeof (struct mlx4_sriov_alias_guid_port_rec_det));
+		/*Check if the SM doesn't need to assign the GUIDs*/
+		for (j = 0; j < NUM_ALIAS_GUID_REC_IN_PORT; j++) {
+			if (mlx4_ib_sm_guid_assign) {
+				dev->sriov.alias_guid.ports_guid[i].
+					all_rec_per_port[j].
+					ownership = MLX4_GUID_DRIVER_ASSIGN;
+				continue;
+			}
+			dev->sriov.alias_guid.ports_guid[i].all_rec_per_port[j].
+					ownership = MLX4_GUID_NONE_ASSIGN;
+			/*mark each val as it was deleted,
+			  till the sysAdmin will give it valid val*/
+			for (k = 0; k < NUM_ALIAS_GUID_IN_REC; k++) {
+				*(__be64 *)&dev->sriov.alias_guid.ports_guid[i].
+					all_rec_per_port[j].all_recs[GUID_REC_SIZE * k] =
+						cpu_to_be64(MLX4_GUID_FOR_DELETE_VAL);
+			}
+		}
+		INIT_LIST_HEAD(&dev->sriov.alias_guid.ports_guid[i].cb_list);
+		/*prepare the records, set them to be allocated by sm*/
+		for (j = 0 ; j < NUM_ALIAS_GUID_REC_IN_PORT; j++)
+			invalidate_guid_record(dev, i + 1, j);
+
+		dev->sriov.alias_guid.ports_guid[i].parent = &dev->sriov.alias_guid;
+		dev->sriov.alias_guid.ports_guid[i].port  = i;
+		if (mlx4_ib_sm_guid_assign)
+			set_all_slaves_guids(dev, i);
+
+		snprintf(alias_wq_name, sizeof alias_wq_name, "alias_guid%d", i);
+		dev->sriov.alias_guid.ports_guid[i].wq =
+			create_singlethread_workqueue(alias_wq_name);
+		if (!dev->sriov.alias_guid.ports_guid[i].wq) {
+			ret = -ENOMEM;
+			goto err_thread;
+		}
+		INIT_DELAYED_WORK(&dev->sriov.alias_guid.ports_guid[i].alias_guid_work,
+			  alias_guid_work);
+	}
+	return 0;
+
+err_thread:
+	for (--i; i >= 0; i--) {
+		destroy_workqueue(dev->sriov.alias_guid.ports_guid[i].wq);
+		dev->sriov.alias_guid.ports_guid[i].wq = NULL;
+	}
+
+err_unregister:
+	ib_sa_unregister_client(dev->sriov.alias_guid.sa_client);
+	kfree(dev->sriov.alias_guid.sa_client);
+	dev->sriov.alias_guid.sa_client = NULL;
+	pr_err("init_alias_guid_service: Failed. (ret:%d)\n", ret);
+	return ret;
+}
diff --git a/drivers/infiniband/hw/mlx4/cm.c b/drivers/infiniband/hw/mlx4/cm.c
new file mode 100644
index 00000000000..56a593e0ae5
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/cm.c
@@ -0,0 +1,478 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+
+#include <linux/mlx4/cmd.h>
+#include <linux/rbtree.h>
+#include <linux/idr.h>
+#include <rdma/ib_cm.h>
+
+#include "mlx4_ib.h"
+
+#define CM_CLEANUP_CACHE_TIMEOUT  (5 * HZ)
+
+struct id_map_entry {
+	struct rb_node node;
+
+	u32 sl_cm_id;
+	u32 pv_cm_id;
+	int slave_id;
+	int scheduled_delete;
+	struct mlx4_ib_dev *dev;
+
+	struct list_head list;
+	struct delayed_work timeout;
+};
+
+struct cm_generic_msg {
+	struct ib_mad_hdr hdr;
+
+	__be32 local_comm_id;
+	__be32 remote_comm_id;
+};
+
+struct cm_sidr_generic_msg {
+	struct ib_mad_hdr hdr;
+	__be32 request_id;
+};
+
+struct cm_req_msg {
+	unsigned char unused[0x60];
+	union ib_gid primary_path_sgid;
+};
+
+
+static void set_local_comm_id(struct ib_mad *mad, u32 cm_id)
+{
+	if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		struct cm_sidr_generic_msg *msg =
+			(struct cm_sidr_generic_msg *)mad;
+		msg->request_id = cpu_to_be32(cm_id);
+	} else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
+		pr_err("trying to set local_comm_id in SIDR_REP\n");
+		return;
+	} else {
+		struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+		msg->local_comm_id = cpu_to_be32(cm_id);
+	}
+}
+
+static u32 get_local_comm_id(struct ib_mad *mad)
+{
+	if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		struct cm_sidr_generic_msg *msg =
+			(struct cm_sidr_generic_msg *)mad;
+		return be32_to_cpu(msg->request_id);
+	} else if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
+		pr_err("trying to set local_comm_id in SIDR_REP\n");
+		return -1;
+	} else {
+		struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+		return be32_to_cpu(msg->local_comm_id);
+	}
+}
+
+static void set_remote_comm_id(struct ib_mad *mad, u32 cm_id)
+{
+	if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
+		struct cm_sidr_generic_msg *msg =
+			(struct cm_sidr_generic_msg *)mad;
+		msg->request_id = cpu_to_be32(cm_id);
+	} else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		pr_err("trying to set remote_comm_id in SIDR_REQ\n");
+		return;
+	} else {
+		struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+		msg->remote_comm_id = cpu_to_be32(cm_id);
+	}
+}
+
+static u32 get_remote_comm_id(struct ib_mad *mad)
+{
+	if (mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
+		struct cm_sidr_generic_msg *msg =
+			(struct cm_sidr_generic_msg *)mad;
+		return be32_to_cpu(msg->request_id);
+	} else if (mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		pr_err("trying to set remote_comm_id in SIDR_REQ\n");
+		return -1;
+	} else {
+		struct cm_generic_msg *msg = (struct cm_generic_msg *)mad;
+		return be32_to_cpu(msg->remote_comm_id);
+	}
+}
+
+static union ib_gid gid_from_req_msg(struct ib_device *ibdev, struct ib_mad *mad)
+{
+	struct cm_req_msg *msg = (struct cm_req_msg *)mad;
+
+	return msg->primary_path_sgid;
+}
+
+/* Lock should be taken before called */
+static struct id_map_entry *
+id_map_find_by_sl_id(struct ib_device *ibdev, u32 slave_id, u32 sl_cm_id)
+{
+	struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+	struct rb_node *node = sl_id_map->rb_node;
+
+	while (node) {
+		struct id_map_entry *id_map_entry =
+			rb_entry(node, struct id_map_entry, node);
+
+		if (id_map_entry->sl_cm_id > sl_cm_id)
+			node = node->rb_left;
+		else if (id_map_entry->sl_cm_id < sl_cm_id)
+			node = node->rb_right;
+		else if (id_map_entry->slave_id > slave_id)
+			node = node->rb_left;
+		else if (id_map_entry->slave_id < slave_id)
+			node = node->rb_right;
+		else
+			return id_map_entry;
+	}
+	return NULL;
+}
+
+static void id_map_ent_timeout(struct work_struct *work)
+{
+	struct delayed_work *delay = to_delayed_work(work);
+	struct id_map_entry *ent = container_of(delay, struct id_map_entry, timeout);
+	struct id_map_entry *db_ent, *found_ent;
+	struct mlx4_ib_dev *dev = ent->dev;
+	struct mlx4_ib_sriov *sriov = &dev->sriov;
+	struct rb_root *sl_id_map = &sriov->sl_id_map;
+	int pv_id = (int) ent->pv_cm_id;
+
+	spin_lock(&sriov->id_map_lock);
+	db_ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_id);
+	if (!db_ent)
+		goto out;
+	found_ent = id_map_find_by_sl_id(&dev->ib_dev, ent->slave_id, ent->sl_cm_id);
+	if (found_ent && found_ent == ent)
+		rb_erase(&found_ent->node, sl_id_map);
+	idr_remove(&sriov->pv_id_table, pv_id);
+
+out:
+	list_del(&ent->list);
+	spin_unlock(&sriov->id_map_lock);
+	kfree(ent);
+}
+
+static void id_map_find_del(struct ib_device *ibdev, int pv_cm_id)
+{
+	struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+	struct rb_root *sl_id_map = &sriov->sl_id_map;
+	struct id_map_entry *ent, *found_ent;
+
+	spin_lock(&sriov->id_map_lock);
+	ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, pv_cm_id);
+	if (!ent)
+		goto out;
+	found_ent = id_map_find_by_sl_id(ibdev, ent->slave_id, ent->sl_cm_id);
+	if (found_ent && found_ent == ent)
+		rb_erase(&found_ent->node, sl_id_map);
+	idr_remove(&sriov->pv_id_table, pv_cm_id);
+out:
+	spin_unlock(&sriov->id_map_lock);
+}
+
+static void sl_id_map_add(struct ib_device *ibdev, struct id_map_entry *new)
+{
+	struct rb_root *sl_id_map = &to_mdev(ibdev)->sriov.sl_id_map;
+	struct rb_node **link = &sl_id_map->rb_node, *parent = NULL;
+	struct id_map_entry *ent;
+	int slave_id = new->slave_id;
+	int sl_cm_id = new->sl_cm_id;
+
+	ent = id_map_find_by_sl_id(ibdev, slave_id, sl_cm_id);
+	if (ent) {
+		pr_debug("overriding existing sl_id_map entry (cm_id = %x)\n",
+			 sl_cm_id);
+
+		rb_replace_node(&ent->node, &new->node, sl_id_map);
+		return;
+	}
+
+	/* Go to the bottom of the tree */
+	while (*link) {
+		parent = *link;
+		ent = rb_entry(parent, struct id_map_entry, node);
+
+		if (ent->sl_cm_id > sl_cm_id || (ent->sl_cm_id == sl_cm_id && ent->slave_id > slave_id))
+			link = &(*link)->rb_left;
+		else
+			link = &(*link)->rb_right;
+	}
+
+	rb_link_node(&new->node, parent, link);
+	rb_insert_color(&new->node, sl_id_map);
+}
+
+static struct id_map_entry *
+id_map_alloc(struct ib_device *ibdev, int slave_id, u32 sl_cm_id)
+{
+	int ret;
+	struct id_map_entry *ent;
+	struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+
+	ent = kmalloc(sizeof (struct id_map_entry), GFP_KERNEL);
+	if (!ent) {
+		mlx4_ib_warn(ibdev, "Couldn't allocate id cache entry - out of memory\n");
+		return ERR_PTR(-ENOMEM);
+	}
+
+	ent->sl_cm_id = sl_cm_id;
+	ent->slave_id = slave_id;
+	ent->scheduled_delete = 0;
+	ent->dev = to_mdev(ibdev);
+	INIT_DELAYED_WORK(&ent->timeout, id_map_ent_timeout);
+
+	idr_preload(GFP_KERNEL);
+	spin_lock(&to_mdev(ibdev)->sriov.id_map_lock);
+
+	ret = idr_alloc_cyclic(&sriov->pv_id_table, ent, 0, 0, GFP_NOWAIT);
+	if (ret >= 0) {
+		ent->pv_cm_id = (u32)ret;
+		sl_id_map_add(ibdev, ent);
+		list_add_tail(&ent->list, &sriov->cm_list);
+	}
+
+	spin_unlock(&sriov->id_map_lock);
+	idr_preload_end();
+
+	if (ret >= 0)
+		return ent;
+
+	/*error flow*/
+	kfree(ent);
+	mlx4_ib_warn(ibdev, "No more space in the idr (err:0x%x)\n", ret);
+	return ERR_PTR(-ENOMEM);
+}
+
+static struct id_map_entry *
+id_map_get(struct ib_device *ibdev, int *pv_cm_id, int sl_cm_id, int slave_id)
+{
+	struct id_map_entry *ent;
+	struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+
+	spin_lock(&sriov->id_map_lock);
+	if (*pv_cm_id == -1) {
+		ent = id_map_find_by_sl_id(ibdev, sl_cm_id, slave_id);
+		if (ent)
+			*pv_cm_id = (int) ent->pv_cm_id;
+	} else
+		ent = (struct id_map_entry *)idr_find(&sriov->pv_id_table, *pv_cm_id);
+	spin_unlock(&sriov->id_map_lock);
+
+	return ent;
+}
+
+static void schedule_delayed(struct ib_device *ibdev, struct id_map_entry *id)
+{
+	struct mlx4_ib_sriov *sriov = &to_mdev(ibdev)->sriov;
+	unsigned long flags;
+
+	spin_lock(&sriov->id_map_lock);
+	spin_lock_irqsave(&sriov->going_down_lock, flags);
+	/*make sure that there is no schedule inside the scheduled work.*/
+	if (!sriov->is_going_down) {
+		id->scheduled_delete = 1;
+		schedule_delayed_work(&id->timeout, CM_CLEANUP_CACHE_TIMEOUT);
+	}
+	spin_unlock_irqrestore(&sriov->going_down_lock, flags);
+	spin_unlock(&sriov->id_map_lock);
+}
+
+int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
+		struct ib_mad *mad)
+{
+	struct id_map_entry *id;
+	u32 sl_cm_id;
+	int pv_cm_id = -1;
+
+	if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
+			mad->mad_hdr.attr_id == CM_REP_ATTR_ID ||
+			mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		sl_cm_id = get_local_comm_id(mad);
+		id = id_map_alloc(ibdev, slave_id, sl_cm_id);
+		if (IS_ERR(id)) {
+			mlx4_ib_warn(ibdev, "%s: id{slave: %d, sl_cm_id: 0x%x} Failed to id_map_alloc\n",
+				__func__, slave_id, sl_cm_id);
+			return PTR_ERR(id);
+		}
+	} else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
+		   mad->mad_hdr.attr_id == CM_SIDR_REP_ATTR_ID) {
+		return 0;
+	} else {
+		sl_cm_id = get_local_comm_id(mad);
+		id = id_map_get(ibdev, &pv_cm_id, slave_id, sl_cm_id);
+	}
+
+	if (!id) {
+		pr_debug("id{slave: %d, sl_cm_id: 0x%x} is NULL!\n",
+			 slave_id, sl_cm_id);
+		return -EINVAL;
+	}
+
+	set_local_comm_id(mad, id->pv_cm_id);
+
+	if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
+		schedule_delayed(ibdev, id);
+	else if (mad->mad_hdr.attr_id == CM_DREP_ATTR_ID)
+		id_map_find_del(ibdev, pv_cm_id);
+
+	return 0;
+}
+
+int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
+			     struct ib_mad *mad)
+{
+	u32 pv_cm_id;
+	struct id_map_entry *id;
+
+	if (mad->mad_hdr.attr_id == CM_REQ_ATTR_ID ||
+	    mad->mad_hdr.attr_id == CM_SIDR_REQ_ATTR_ID) {
+		union ib_gid gid;
+
+		if (!slave)
+			return 0;
+
+		gid = gid_from_req_msg(ibdev, mad);
+		*slave = mlx4_ib_find_real_gid(ibdev, port, gid.global.interface_id);
+		if (*slave < 0) {
+			mlx4_ib_warn(ibdev, "failed matching slave_id by gid (0x%llx)\n",
+					gid.global.interface_id);
+			return -ENOENT;
+		}
+		return 0;
+	}
+
+	pv_cm_id = get_remote_comm_id(mad);
+	id = id_map_get(ibdev, (int *)&pv_cm_id, -1, -1);
+
+	if (!id) {
+		pr_debug("Couldn't find an entry for pv_cm_id 0x%x\n", pv_cm_id);
+		return -ENOENT;
+	}
+
+	if (slave)
+		*slave = id->slave_id;
+	set_remote_comm_id(mad, id->sl_cm_id);
+
+	if (mad->mad_hdr.attr_id == CM_DREQ_ATTR_ID)
+		schedule_delayed(ibdev, id);
+	else if (mad->mad_hdr.attr_id == CM_REJ_ATTR_ID ||
+			mad->mad_hdr.attr_id == CM_DREP_ATTR_ID) {
+		id_map_find_del(ibdev, (int) pv_cm_id);
+	}
+
+	return 0;
+}
+
+void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev)
+{
+	spin_lock_init(&dev->sriov.id_map_lock);
+	INIT_LIST_HEAD(&dev->sriov.cm_list);
+	dev->sriov.sl_id_map = RB_ROOT;
+	idr_init(&dev->sriov.pv_id_table);
+}
+
+/* slave = -1 ==> all slaves */
+/* TBD -- call paravirt clean for single slave.  Need for slave RESET event */
+void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave)
+{
+	struct mlx4_ib_sriov *sriov = &dev->sriov;
+	struct rb_root *sl_id_map = &sriov->sl_id_map;
+	struct list_head lh;
+	struct rb_node *nd;
+	int need_flush = 1;
+	struct id_map_entry *map, *tmp_map;
+	/* cancel all delayed work queue entries */
+	INIT_LIST_HEAD(&lh);
+	spin_lock(&sriov->id_map_lock);
+	list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+		if (slave < 0 || slave == map->slave_id) {
+			if (map->scheduled_delete)
+				need_flush &= !!cancel_delayed_work(&map->timeout);
+		}
+	}
+
+	spin_unlock(&sriov->id_map_lock);
+
+	if (!need_flush)
+		flush_scheduled_work(); /* make sure all timers were flushed */
+
+	/* now, remove all leftover entries from databases*/
+	spin_lock(&sriov->id_map_lock);
+	if (slave < 0) {
+		while (rb_first(sl_id_map)) {
+			struct id_map_entry *ent =
+				rb_entry(rb_first(sl_id_map),
+					 struct id_map_entry, node);
+
+			rb_erase(&ent->node, sl_id_map);
+			idr_remove(&sriov->pv_id_table, (int) ent->pv_cm_id);
+		}
+		list_splice_init(&dev->sriov.cm_list, &lh);
+	} else {
+		/* first, move nodes belonging to slave to db remove list */
+		nd = rb_first(sl_id_map);
+		while (nd) {
+			struct id_map_entry *ent =
+				rb_entry(nd, struct id_map_entry, node);
+			nd = rb_next(nd);
+			if (ent->slave_id == slave)
+				list_move_tail(&ent->list, &lh);
+		}
+		/* remove those nodes from databases */
+		list_for_each_entry_safe(map, tmp_map, &lh, list) {
+			rb_erase(&map->node, sl_id_map);
+			idr_remove(&sriov->pv_id_table, (int) map->pv_cm_id);
+		}
+
+		/* add remaining nodes from cm_list */
+		list_for_each_entry_safe(map, tmp_map, &dev->sriov.cm_list, list) {
+			if (slave == map->slave_id)
+				list_move_tail(&map->list, &lh);
+		}
+	}
+
+	spin_unlock(&sriov->id_map_lock);
+
+	/* free any map entries left behind due to cancel_delayed_work above */
+	list_for_each_entry_safe(map, tmp_map, &lh, list) {
+		list_del(&map->list);
+		kfree(map);
+	}
+}
diff --git a/drivers/infiniband/hw/mlx4/cq.c b/drivers/infiniband/hw/mlx4/cq.c
index 299f20832ab..1066eec854a 100644
--- a/drivers/infiniband/hw/mlx4/cq.c
+++ b/drivers/infiniband/hw/mlx4/cq.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -32,6 +33,8 @@
 
 #include <linux/mlx4/cq.h>
 #include <linux/mlx4/qp.h>
+#include <linux/mlx4/srq.h>
+#include <linux/slab.h>
 
 #include "mlx4_ib.h"
 #include "user.h"
@@ -48,7 +51,7 @@ static void mlx4_ib_cq_event(struct mlx4_cq *cq, enum mlx4_event type)
 	struct ib_cq *ibcq;
 
 	if (type != MLX4_EVENT_TYPE_CQ_ERROR) {
-		printk(KERN_WARNING "mlx4_ib: Unexpected event type %d "
+		pr_warn("Unexpected event type %d "
 		       "on CQ %06x\n", type, cq->cqn);
 		return;
 	}
@@ -64,7 +67,7 @@ static void mlx4_ib_cq_event(struct mlx4_cq *cq, enum mlx4_event type)
 
 static void *get_cqe_from_buf(struct mlx4_ib_cq_buf *buf, int n)
 {
-	return mlx4_buf_offset(&buf->buf, n * sizeof (struct mlx4_cqe));
+	return mlx4_buf_offset(&buf->buf, n * buf->entry_size);
 }
 
 static void *get_cqe(struct mlx4_ib_cq *cq, int n)
@@ -75,8 +78,9 @@ static void *get_cqe(struct mlx4_ib_cq *cq, int n)
 static void *get_sw_cqe(struct mlx4_ib_cq *cq, int n)
 {
 	struct mlx4_cqe *cqe = get_cqe(cq, n & cq->ibcq.cqe);
+	struct mlx4_cqe *tcqe = ((cq->buf.entry_size == 64) ? (cqe + 1) : cqe);
 
-	return (!!(cqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK) ^
+	return (!!(tcqe->owner_sr_opcode & MLX4_CQE_OWNER_MASK) ^
 		!!(n & (cq->ibcq.cqe + 1))) ? NULL : cqe;
 }
 
@@ -97,18 +101,19 @@ static int mlx4_ib_alloc_cq_buf(struct mlx4_ib_dev *dev, struct mlx4_ib_cq_buf *
 {
 	int err;
 
-	err = mlx4_buf_alloc(dev->dev, nent * sizeof(struct mlx4_cqe),
-			     PAGE_SIZE * 2, &buf->buf);
+	err = mlx4_buf_alloc(dev->dev, nent * dev->dev->caps.cqe_size,
+			     PAGE_SIZE * 2, &buf->buf, GFP_KERNEL);
 
 	if (err)
 		goto out;
 
+	buf->entry_size = dev->dev->caps.cqe_size;
 	err = mlx4_mtt_init(dev->dev, buf->buf.npages, buf->buf.page_shift,
 				    &buf->mtt);
 	if (err)
 		goto err_buf;
 
-	err = mlx4_buf_write_mtt(dev->dev, &buf->mtt, &buf->buf);
+	err = mlx4_buf_write_mtt(dev->dev, &buf->mtt, &buf->buf, GFP_KERNEL);
 	if (err)
 		goto err_mtt;
 
@@ -118,8 +123,7 @@ err_mtt:
 	mlx4_mtt_cleanup(dev->dev, &buf->mtt);
 
 err_buf:
-	mlx4_buf_free(dev->dev, nent * sizeof(struct mlx4_cqe),
-			      &buf->buf);
+	mlx4_buf_free(dev->dev, nent * buf->entry_size, &buf->buf);
 
 out:
 	return err;
@@ -127,7 +131,7 @@ out:
 
 static void mlx4_ib_free_cq_buf(struct mlx4_ib_dev *dev, struct mlx4_ib_cq_buf *buf, int cqe)
 {
-	mlx4_buf_free(dev->dev, (cqe + 1) * sizeof(struct mlx4_cqe), &buf->buf);
+	mlx4_buf_free(dev->dev, (cqe + 1) * buf->entry_size, &buf->buf);
 }
 
 static int mlx4_ib_get_cq_umem(struct mlx4_ib_dev *dev, struct ib_ucontext *context,
@@ -135,8 +139,9 @@ static int mlx4_ib_get_cq_umem(struct mlx4_ib_dev *dev, struct ib_ucontext *cont
 			       u64 buf_addr, int cqe)
 {
 	int err;
+	int cqe_size = dev->dev->caps.cqe_size;
 
-	*umem = ib_umem_get(context, buf_addr, cqe * sizeof (struct mlx4_cqe),
+	*umem = ib_umem_get(context, buf_addr, cqe * cqe_size,
 			    IB_ACCESS_LOCAL_WRITE, 1);
 	if (IS_ERR(*umem))
 		return PTR_ERR(*umem);
@@ -204,7 +209,7 @@ struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, int entries, int vector
 
 		uar = &to_mucontext(context)->uar;
 	} else {
-		err = mlx4_db_alloc(dev->dev, &cq->db, 1);
+		err = mlx4_db_alloc(dev->dev, &cq->db, 1, GFP_KERNEL);
 		if (err)
 			goto err_cq;
 
@@ -220,8 +225,11 @@ struct ib_cq *mlx4_ib_create_cq(struct ib_device *ibdev, int entries, int vector
 		uar = &dev->priv_uar;
 	}
 
+	if (dev->eq_table)
+		vector = dev->eq_table[vector % ibdev->num_comp_vectors];
+
 	err = mlx4_cq_alloc(dev->dev, entries, &cq->buf.mtt, uar,
-			    cq->db.dma, &cq->mcq, 0);
+			    cq->db.dma, &cq->mcq, vector, 0, 0);
 	if (err)
 		goto err_dbmap;
 
@@ -316,7 +324,7 @@ static int mlx4_ib_get_outstanding_cqes(struct mlx4_ib_cq *cq)
 	u32 i;
 
 	i = cq->mcq.cons_index;
-	while (get_sw_cqe(cq, i & cq->ibcq.cqe))
+	while (get_sw_cqe(cq, i))
 		++i;
 
 	return i - cq->mcq.cons_index;
@@ -324,16 +332,25 @@ static int mlx4_ib_get_outstanding_cqes(struct mlx4_ib_cq *cq)
 
 static void mlx4_ib_cq_resize_copy_cqes(struct mlx4_ib_cq *cq)
 {
-	struct mlx4_cqe *cqe;
+	struct mlx4_cqe *cqe, *new_cqe;
 	int i;
+	int cqe_size = cq->buf.entry_size;
+	int cqe_inc = cqe_size == 64 ? 1 : 0;
 
 	i = cq->mcq.cons_index;
 	cqe = get_cqe(cq, i & cq->ibcq.cqe);
+	cqe += cqe_inc;
+
 	while ((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) != MLX4_CQE_OPCODE_RESIZE) {
-		memcpy(get_cqe_from_buf(&cq->resize_buf->buf,
-					(i + 1) & cq->resize_buf->cqe),
-			get_cqe(cq, i & cq->ibcq.cqe), sizeof(struct mlx4_cqe));
+		new_cqe = get_cqe_from_buf(&cq->resize_buf->buf,
+					   (i + 1) & cq->resize_buf->cqe);
+		memcpy(new_cqe, get_cqe(cq, i & cq->ibcq.cqe), cqe_size);
+		new_cqe += cqe_inc;
+
+		new_cqe->owner_sr_opcode = (cqe->owner_sr_opcode & ~MLX4_CQE_OWNER_MASK) |
+			(((i + 1) & (cq->resize_buf->cqe + 1)) ? MLX4_CQE_OWNER_MASK : 0);
 		cqe = get_cqe(cq, ++i & cq->ibcq.cqe);
+		cqe += cqe_inc;
 	}
 	++cq->mcq.cons_index;
 }
@@ -342,12 +359,13 @@ int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 {
 	struct mlx4_ib_dev *dev = to_mdev(ibcq->device);
 	struct mlx4_ib_cq *cq = to_mcq(ibcq);
+	struct mlx4_mtt mtt;
 	int outst_cqe;
 	int err;
 
 	mutex_lock(&cq->resize_mutex);
 
-	if (entries < 1 || entries > dev->dev->caps.max_cqes) {
+	if (entries < 1) {
 		err = -EINVAL;
 		goto out;
 	}
@@ -358,12 +376,17 @@ int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 		goto out;
 	}
 
+	if (entries > dev->dev->caps.max_cqes) {
+		err = -EINVAL;
+		goto out;
+	}
+
 	if (ibcq->uobject) {
 		err = mlx4_alloc_resize_umem(dev, cq, entries, udata);
 		if (err)
 			goto out;
 	} else {
-		/* Can't be smaller then the number of outstanding CQEs */
+		/* Can't be smaller than the number of outstanding CQEs */
 		outst_cqe = mlx4_ib_get_outstanding_cqes(cq);
 		if (entries < outst_cqe + 1) {
 			err = 0;
@@ -375,10 +398,13 @@ int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 			goto out;
 	}
 
+	mtt = cq->buf.mtt;
+
 	err = mlx4_cq_resize(dev->dev, &cq->mcq, entries, &cq->resize_buf->buf.mtt);
 	if (err)
 		goto err_buf;
 
+	mlx4_mtt_cleanup(dev->dev, &mtt);
 	if (ibcq->uobject) {
 		cq->buf      = cq->resize_buf->buf;
 		cq->ibcq.cqe = cq->resize_buf->cqe;
@@ -389,10 +415,14 @@ int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 		cq->resize_buf = NULL;
 		cq->resize_umem = NULL;
 	} else {
+		struct mlx4_ib_cq_buf tmp_buf;
+		int tmp_cqe = 0;
+
 		spin_lock_irq(&cq->lock);
 		if (cq->resize_buf) {
 			mlx4_ib_cq_resize_copy_cqes(cq);
-			mlx4_ib_free_cq_buf(dev, &cq->buf, cq->ibcq.cqe);
+			tmp_buf = cq->buf;
+			tmp_cqe = cq->ibcq.cqe;
 			cq->buf      = cq->resize_buf->buf;
 			cq->ibcq.cqe = cq->resize_buf->cqe;
 
@@ -400,11 +430,15 @@ int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
 			cq->resize_buf = NULL;
 		}
 		spin_unlock_irq(&cq->lock);
+
+		if (tmp_cqe)
+			mlx4_ib_free_cq_buf(dev, &tmp_buf, tmp_cqe);
 	}
 
 	goto out;
 
 err_buf:
+	mlx4_mtt_cleanup(dev->dev, &cq->resize_buf->buf.mtt);
 	if (!ibcq->uobject)
 		mlx4_ib_free_cq_buf(dev, &cq->resize_buf->buf,
 				    cq->resize_buf->cqe);
@@ -419,6 +453,7 @@ err_buf:
 
 out:
 	mutex_unlock(&cq->resize_mutex);
+
 	return err;
 }
 
@@ -447,7 +482,7 @@ static void dump_cqe(void *cqe)
 {
 	__be32 *buf = cqe;
 
-	printk(KERN_DEBUG "CQE contents %08x %08x %08x %08x %08x %08x %08x %08x\n",
+	pr_debug("CQE contents %08x %08x %08x %08x %08x %08x %08x %08x\n",
 	       be32_to_cpu(buf[0]), be32_to_cpu(buf[1]), be32_to_cpu(buf[2]),
 	       be32_to_cpu(buf[3]), be32_to_cpu(buf[4]), be32_to_cpu(buf[5]),
 	       be32_to_cpu(buf[6]), be32_to_cpu(buf[7]));
@@ -457,7 +492,7 @@ static void mlx4_ib_handle_error_cqe(struct mlx4_err_cqe *cqe,
 				     struct ib_wc *wc)
 {
 	if (cqe->syndrome == MLX4_CQE_SYNDROME_LOCAL_QP_OP_ERR) {
-		printk(KERN_DEBUG "local QP operation err "
+		pr_debug("local QP operation err "
 		       "(QPN %06x, WQE index %x, vendor syndrome %02x, "
 		       "opcode = %02x)\n",
 		       be32_to_cpu(cqe->my_qpn), be16_to_cpu(cqe->wqe_index),
@@ -514,20 +549,48 @@ static void mlx4_ib_handle_error_cqe(struct mlx4_err_cqe *cqe,
 	wc->vendor_err = cqe->vendor_err_syndrome;
 }
 
-static int mlx4_ib_ipoib_csum_ok(__be32 status, __be16 checksum)
+static int mlx4_ib_ipoib_csum_ok(__be16 status, __be16 checksum)
 {
-	return ((status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4	|
-				      MLX4_CQE_IPOIB_STATUS_IPV4F	|
-				      MLX4_CQE_IPOIB_STATUS_IPV4OPT	|
-				      MLX4_CQE_IPOIB_STATUS_IPV6	|
-				      MLX4_CQE_IPOIB_STATUS_IPOK)) ==
-		cpu_to_be32(MLX4_CQE_IPOIB_STATUS_IPV4	|
-			    MLX4_CQE_IPOIB_STATUS_IPOK))		&&
-		(status & cpu_to_be32(MLX4_CQE_IPOIB_STATUS_UDP	|
-				      MLX4_CQE_IPOIB_STATUS_TCP))	&&
+	return ((status & cpu_to_be16(MLX4_CQE_STATUS_IPV4      |
+				      MLX4_CQE_STATUS_IPV4F     |
+				      MLX4_CQE_STATUS_IPV4OPT   |
+				      MLX4_CQE_STATUS_IPV6      |
+				      MLX4_CQE_STATUS_IPOK)) ==
+		cpu_to_be16(MLX4_CQE_STATUS_IPV4        |
+			    MLX4_CQE_STATUS_IPOK))              &&
+		(status & cpu_to_be16(MLX4_CQE_STATUS_UDP       |
+				      MLX4_CQE_STATUS_TCP))     &&
 		checksum == cpu_to_be16(0xffff);
 }
 
+static int use_tunnel_data(struct mlx4_ib_qp *qp, struct mlx4_ib_cq *cq, struct ib_wc *wc,
+			   unsigned tail, struct mlx4_cqe *cqe, int is_eth)
+{
+	struct mlx4_ib_proxy_sqp_hdr *hdr;
+
+	ib_dma_sync_single_for_cpu(qp->ibqp.device,
+				   qp->sqp_proxy_rcv[tail].map,
+				   sizeof (struct mlx4_ib_proxy_sqp_hdr),
+				   DMA_FROM_DEVICE);
+	hdr = (struct mlx4_ib_proxy_sqp_hdr *) (qp->sqp_proxy_rcv[tail].addr);
+	wc->pkey_index	= be16_to_cpu(hdr->tun.pkey_index);
+	wc->src_qp	= be32_to_cpu(hdr->tun.flags_src_qp) & 0xFFFFFF;
+	wc->wc_flags   |= (hdr->tun.g_ml_path & 0x80) ? (IB_WC_GRH) : 0;
+	wc->dlid_path_bits = 0;
+
+	if (is_eth) {
+		wc->vlan_id = be16_to_cpu(hdr->tun.sl_vid);
+		memcpy(&(wc->smac[0]), (char *)&hdr->tun.mac_31_0, 4);
+		memcpy(&(wc->smac[4]), (char *)&hdr->tun.slid_mac_47_32, 2);
+		wc->wc_flags |= (IB_WC_WITH_VLAN | IB_WC_WITH_SMAC);
+	} else {
+		wc->slid        = be16_to_cpu(hdr->tun.slid_mac_47_32);
+		wc->sl          = (u8) (be16_to_cpu(hdr->tun.sl_vid) >> 12);
+	}
+
+	return 0;
+}
+
 static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
 			    struct mlx4_ib_qp **cur_qp,
 			    struct ib_wc *wc)
@@ -536,16 +599,22 @@ static int mlx4_ib_poll_one(struct mlx4_ib_cq *cq,
 	struct mlx4_qp *mqp;
 	struct mlx4_ib_wq *wq;
 	struct mlx4_ib_srq *srq;
+	struct mlx4_srq *msrq = NULL;
 	int is_send;
 	int is_error;
+	int is_eth;
 	u32 g_mlpath_rqpn;
 	u16 wqe_ctr;
+	unsigned tail = 0;
 
 repoll:
 	cqe = next_cqe_sw(cq);
 	if (!cqe)
 		return -EAGAIN;
 
+	if (cq->buf.entry_size == 64)
+		cqe++;
+
 	++cq->mcq.cons_index;
 
 	/*
@@ -560,7 +629,7 @@ repoll:
 
 	if (unlikely((cqe->owner_sr_opcode & MLX4_CQE_OPCODE_MASK) == MLX4_OPCODE_NOP &&
 		     is_send)) {
-		printk(KERN_WARNING "Completion for NOP opcode detected!\n");
+		pr_warn("Completion for NOP opcode detected!\n");
 		return -EINVAL;
 	}
 
@@ -581,17 +650,17 @@ repoll:
 	}
 
 	if (!*cur_qp ||
-	    (be32_to_cpu(cqe->my_qpn) & 0xffffff) != (*cur_qp)->mqp.qpn) {
+	    (be32_to_cpu(cqe->vlan_my_qpn) & MLX4_CQE_QPN_MASK) != (*cur_qp)->mqp.qpn) {
 		/*
 		 * We do not have to take the QP table lock here,
 		 * because CQs will be locked while QPs are removed
 		 * from the table.
 		 */
 		mqp = __mlx4_qp_lookup(to_mdev(cq->ibcq.device)->dev,
-				       be32_to_cpu(cqe->my_qpn));
+				       be32_to_cpu(cqe->vlan_my_qpn));
 		if (unlikely(!mqp)) {
-			printk(KERN_WARNING "CQ %06x with entry for unknown QPN %06x\n",
-			       cq->mcq.cqn, be32_to_cpu(cqe->my_qpn) & 0xffffff);
+			pr_warn("CQ %06x with entry for unknown QPN %06x\n",
+			       cq->mcq.cqn, be32_to_cpu(cqe->vlan_my_qpn) & MLX4_CQE_QPN_MASK);
 			return -EINVAL;
 		}
 
@@ -600,6 +669,20 @@ repoll:
 
 	wc->qp = &(*cur_qp)->ibqp;
 
+	if (wc->qp->qp_type == IB_QPT_XRC_TGT) {
+		u32 srq_num;
+		g_mlpath_rqpn = be32_to_cpu(cqe->g_mlpath_rqpn);
+		srq_num       = g_mlpath_rqpn & 0xffffff;
+		/* SRQ is also in the radix tree */
+		msrq = mlx4_srq_lookup(to_mdev(cq->ibcq.device)->dev,
+				       srq_num);
+		if (unlikely(!msrq)) {
+			pr_warn("CQ %06x with entry for unknown SRQN %06x\n",
+				cq->mcq.cqn, srq_num);
+			return -EINVAL;
+		}
+	}
+
 	if (is_send) {
 		wq = &(*cur_qp)->sq;
 		if (!(*cur_qp)->sq_signal_bits) {
@@ -613,9 +696,15 @@ repoll:
 		wqe_ctr = be16_to_cpu(cqe->wqe_index);
 		wc->wr_id = srq->wrid[wqe_ctr];
 		mlx4_ib_free_srq_wqe(srq, wqe_ctr);
+	} else if (msrq) {
+		srq = to_mibsrq(msrq);
+		wqe_ctr = be16_to_cpu(cqe->wqe_index);
+		wc->wr_id = srq->wrid[wqe_ctr];
+		mlx4_ib_free_srq_wqe(srq, wqe_ctr);
 	} else {
 		wq	  = &(*cur_qp)->rq;
-		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		tail	  = wq->tail & (wq->wqe_cnt - 1);
+		wc->wr_id = wq->wrid[tail];
 		++wq->tail;
 	}
 
@@ -637,6 +726,7 @@ repoll:
 		case MLX4_OPCODE_SEND_IMM:
 			wc->wc_flags |= IB_WC_WITH_IMM;
 		case MLX4_OPCODE_SEND:
+		case MLX4_OPCODE_SEND_INVAL:
 			wc->opcode    = IB_WC_SEND;
 			break;
 		case MLX4_OPCODE_RDMA_READ:
@@ -651,12 +741,26 @@ repoll:
 			wc->opcode    = IB_WC_FETCH_ADD;
 			wc->byte_len  = 8;
 			break;
+		case MLX4_OPCODE_MASKED_ATOMIC_CS:
+			wc->opcode    = IB_WC_MASKED_COMP_SWAP;
+			wc->byte_len  = 8;
+			break;
+		case MLX4_OPCODE_MASKED_ATOMIC_FA:
+			wc->opcode    = IB_WC_MASKED_FETCH_ADD;
+			wc->byte_len  = 8;
+			break;
 		case MLX4_OPCODE_BIND_MW:
 			wc->opcode    = IB_WC_BIND_MW;
 			break;
 		case MLX4_OPCODE_LSO:
 			wc->opcode    = IB_WC_LSO;
 			break;
+		case MLX4_OPCODE_FMR:
+			wc->opcode    = IB_WC_FAST_REG_MR;
+			break;
+		case MLX4_OPCODE_LOCAL_INVAL:
+			wc->opcode    = IB_WC_LOCAL_INV;
+			break;
 		}
 	} else {
 		wc->byte_len = be32_to_cpu(cqe->byte_cnt);
@@ -667,6 +771,11 @@ repoll:
 			wc->wc_flags	= IB_WC_WITH_IMM;
 			wc->ex.imm_data = cqe->immed_rss_invalid;
 			break;
+		case MLX4_RECV_OPCODE_SEND_INVAL:
+			wc->opcode	= IB_WC_RECV;
+			wc->wc_flags	= IB_WC_WITH_INVALIDATE;
+			wc->ex.invalidate_rkey = be32_to_cpu(cqe->immed_rss_invalid);
+			break;
 		case MLX4_RECV_OPCODE_SEND:
 			wc->opcode   = IB_WC_RECV;
 			wc->wc_flags = 0;
@@ -678,15 +787,40 @@ repoll:
 			break;
 		}
 
+		is_eth = (rdma_port_get_link_layer(wc->qp->device,
+						  (*cur_qp)->port) ==
+			  IB_LINK_LAYER_ETHERNET);
+		if (mlx4_is_mfunc(to_mdev(cq->ibcq.device)->dev)) {
+			if ((*cur_qp)->mlx4_ib_qp_type &
+			    (MLX4_IB_QPT_PROXY_SMI_OWNER |
+			     MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI))
+				return use_tunnel_data(*cur_qp, cq, wc, tail,
+						       cqe, is_eth);
+		}
+
 		wc->slid	   = be16_to_cpu(cqe->rlid);
-		wc->sl		   = cqe->sl >> 4;
 		g_mlpath_rqpn	   = be32_to_cpu(cqe->g_mlpath_rqpn);
 		wc->src_qp	   = g_mlpath_rqpn & 0xffffff;
 		wc->dlid_path_bits = (g_mlpath_rqpn >> 24) & 0x7f;
 		wc->wc_flags	  |= g_mlpath_rqpn & 0x80000000 ? IB_WC_GRH : 0;
 		wc->pkey_index     = be32_to_cpu(cqe->immed_rss_invalid) & 0x7f;
-		wc->csum_ok	   = mlx4_ib_ipoib_csum_ok(cqe->ipoib_status,
-							   cqe->checksum);
+		wc->wc_flags	  |= mlx4_ib_ipoib_csum_ok(cqe->status,
+					cqe->checksum) ? IB_WC_IP_CSUM_OK : 0;
+		if (is_eth) {
+			wc->sl  = be16_to_cpu(cqe->sl_vid) >> 13;
+			if (be32_to_cpu(cqe->vlan_my_qpn) &
+					MLX4_CQE_VLAN_PRESENT_MASK) {
+				wc->vlan_id = be16_to_cpu(cqe->sl_vid) &
+					MLX4_CQE_VID_MASK;
+			} else {
+				wc->vlan_id = 0xffff;
+			}
+			memcpy(wc->smac, cqe->smac, ETH_ALEN);
+			wc->wc_flags |= (IB_WC_WITH_VLAN | IB_WC_WITH_SMAC);
+		} else {
+			wc->sl  = be16_to_cpu(cqe->sl_vid) >> 12;
+			wc->vlan_id = 0xffff;
+		}
 	}
 
 	return 0;
@@ -708,8 +842,7 @@ int mlx4_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
 			break;
 	}
 
-	if (npolled)
-		mlx4_cq_set_ci(&cq->mcq);
+	mlx4_cq_set_ci(&cq->mcq);
 
 	spin_unlock_irqrestore(&cq->lock, flags);
 
@@ -736,6 +869,7 @@ void __mlx4_ib_cq_clean(struct mlx4_ib_cq *cq, u32 qpn, struct mlx4_ib_srq *srq)
 	int nfreed = 0;
 	struct mlx4_cqe *cqe, *dest;
 	u8 owner_bit;
+	int cqe_inc = cq->buf.entry_size == 64 ? 1 : 0;
 
 	/*
 	 * First we need to find the current producer index, so we
@@ -754,12 +888,16 @@ void __mlx4_ib_cq_clean(struct mlx4_ib_cq *cq, u32 qpn, struct mlx4_ib_srq *srq)
 	 */
 	while ((int) --prod_index - (int) cq->mcq.cons_index >= 0) {
 		cqe = get_cqe(cq, prod_index & cq->ibcq.cqe);
-		if ((be32_to_cpu(cqe->my_qpn) & 0xffffff) == qpn) {
+		cqe += cqe_inc;
+
+		if ((be32_to_cpu(cqe->vlan_my_qpn) & MLX4_CQE_QPN_MASK) == qpn) {
 			if (srq && !(cqe->owner_sr_opcode & MLX4_CQE_IS_SEND_MASK))
 				mlx4_ib_free_srq_wqe(srq, be16_to_cpu(cqe->wqe_index));
 			++nfreed;
 		} else if (nfreed) {
 			dest = get_cqe(cq, (prod_index + nfreed) & cq->ibcq.cqe);
+			dest += cqe_inc;
+
 			owner_bit = dest->owner_sr_opcode & MLX4_CQE_OWNER_MASK;
 			memcpy(dest, cqe, sizeof *cqe);
 			dest->owner_sr_opcode = owner_bit |
diff --git a/drivers/infiniband/hw/mlx4/doorbell.c b/drivers/infiniband/hw/mlx4/doorbell.c
index 8aee4233b38..c5174098636 100644
--- a/drivers/infiniband/hw/mlx4/doorbell.c
+++ b/drivers/infiniband/hw/mlx4/doorbell.c
@@ -45,7 +45,6 @@ int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
 			struct mlx4_db *db)
 {
 	struct mlx4_ib_user_db_page *page;
-	struct ib_umem_chunk *chunk;
 	int err = 0;
 
 	mutex_lock(&context->db_page_mutex);
@@ -73,8 +72,7 @@ int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
 	list_add(&page->list, &context->db_page_list);
 
 found:
-	chunk = list_entry(page->umem->chunk_list.next, struct ib_umem_chunk, list);
-	db->dma		= sg_dma_address(chunk->page_list) + (virt & ~PAGE_MASK);
+	db->dma = sg_dma_address(page->umem->sg_head.sgl) + (virt & ~PAGE_MASK);
 	db->u.user_page = page;
 	++page->refcnt;
 
diff --git a/drivers/infiniband/hw/mlx4/mad.c b/drivers/infiniband/hw/mlx4/mad.c
index cdca3a511e1..287ad0564ac 100644
--- a/drivers/infiniband/hw/mlx4/mad.c
+++ b/drivers/infiniband/hw/mlx4/mad.c
@@ -32,8 +32,13 @@
 
 #include <rdma/ib_mad.h>
 #include <rdma/ib_smi.h>
+#include <rdma/ib_sa.h>
+#include <rdma/ib_cache.h>
 
+#include <linux/random.h>
 #include <linux/mlx4/cmd.h>
+#include <linux/gfp.h>
+#include <rdma/ib_pma.h>
 
 #include "mlx4_ib.h"
 
@@ -42,7 +47,62 @@ enum {
 	MLX4_IB_VENDOR_CLASS2 = 0xa
 };
 
-int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+#define MLX4_TUN_SEND_WRID_SHIFT 34
+#define MLX4_TUN_QPN_SHIFT 32
+#define MLX4_TUN_WRID_RECV (((u64) 1) << MLX4_TUN_SEND_WRID_SHIFT)
+#define MLX4_TUN_SET_WRID_QPN(a) (((u64) ((a) & 0x3)) << MLX4_TUN_QPN_SHIFT)
+
+#define MLX4_TUN_IS_RECV(a)  (((a) >>  MLX4_TUN_SEND_WRID_SHIFT) & 0x1)
+#define MLX4_TUN_WRID_QPN(a) (((a) >> MLX4_TUN_QPN_SHIFT) & 0x3)
+
+ /* Port mgmt change event handling */
+
+#define GET_BLK_PTR_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.block_ptr)
+#define GET_MASK_FROM_EQE(eqe) be32_to_cpu(eqe->event.port_mgmt_change.params.tbl_change_info.tbl_entries_mask)
+#define NUM_IDX_IN_PKEY_TBL_BLK 32
+#define GUID_TBL_ENTRY_SIZE 8	   /* size in bytes */
+#define GUID_TBL_BLK_NUM_ENTRIES 8
+#define GUID_TBL_BLK_SIZE (GUID_TBL_ENTRY_SIZE * GUID_TBL_BLK_NUM_ENTRIES)
+
+struct mlx4_mad_rcv_buf {
+	struct ib_grh grh;
+	u8 payload[256];
+} __packed;
+
+struct mlx4_mad_snd_buf {
+	u8 payload[256];
+} __packed;
+
+struct mlx4_tunnel_mad {
+	struct ib_grh grh;
+	struct mlx4_ib_tunnel_header hdr;
+	struct ib_mad mad;
+} __packed;
+
+struct mlx4_rcv_tunnel_mad {
+	struct mlx4_rcv_tunnel_hdr hdr;
+	struct ib_grh grh;
+	struct ib_mad mad;
+} __packed;
+
+static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num);
+static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num);
+static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+				int block, u32 change_bitmap);
+
+__be64 mlx4_ib_gen_node_guid(void)
+{
+#define NODE_GUID_HI	((u64) (((u64)IB_OPENIB_OUI) << 40))
+	return cpu_to_be64(NODE_GUID_HI | prandom_u32());
+}
+
+__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx)
+{
+	return cpu_to_be64(atomic_inc_return(&ctx->tid)) |
+		cpu_to_be64(0xff00000000000000LL);
+}
+
+int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags,
 		 int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
 		 void *in_mad, void *response_mad)
 {
@@ -69,10 +129,13 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
 	 * Key check traps can't be generated unless we have in_wc to
 	 * tell us where to send the trap.
 	 */
-	if (ignore_mkey || !in_wc)
+	if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_MKEY) || !in_wc)
 		op_modifier |= 0x1;
-	if (ignore_bkey || !in_wc)
+	if ((mad_ifc_flags & MLX4_MAD_IFC_IGNORE_BKEY) || !in_wc)
 		op_modifier |= 0x2;
+	if (mlx4_is_mfunc(dev->dev) &&
+	    (mad_ifc_flags & MLX4_MAD_IFC_NET_VIEW || in_wc))
+		op_modifier |= 0x8;
 
 	if (in_wc) {
 		struct {
@@ -105,9 +168,10 @@ int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
 		in_modifier |= in_wc->slid << 16;
 	}
 
-	err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma,
-			   in_modifier, op_modifier,
-			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C);
+	err = mlx4_cmd_box(dev->dev, inmailbox->dma, outmailbox->dma, in_modifier,
+			   mlx4_is_master(dev->dev) ? (op_modifier & ~0x8) : op_modifier,
+			   MLX4_CMD_MAD_IFC, MLX4_CMD_TIME_CLASS_C,
+			   (op_modifier & 0x8) ? MLX4_CMD_NATIVE : MLX4_CMD_WRAPPED);
 
 	if (!err)
 		memcpy(response_mad, outmailbox->buf, 256);
@@ -122,6 +186,7 @@ static void update_sm_ah(struct mlx4_ib_dev *dev, u8 port_num, u16 lid, u8 sl)
 {
 	struct ib_ah *new_ah;
 	struct ib_ah_attr ah_attr;
+	unsigned long flags;
 
 	if (!dev->send_agent[port_num - 1][0])
 		return;
@@ -136,48 +201,134 @@ static void update_sm_ah(struct mlx4_ib_dev *dev, u8 port_num, u16 lid, u8 sl)
 	if (IS_ERR(new_ah))
 		return;
 
-	spin_lock(&dev->sm_lock);
+	spin_lock_irqsave(&dev->sm_lock, flags);
 	if (dev->sm_ah[port_num - 1])
 		ib_destroy_ah(dev->sm_ah[port_num - 1]);
 	dev->sm_ah[port_num - 1] = new_ah;
-	spin_unlock(&dev->sm_lock);
+	spin_unlock_irqrestore(&dev->sm_lock, flags);
 }
 
 /*
- * Snoop SM MADs for port info and P_Key table sets, so we can
- * synthesize LID change and P_Key change events.
+ * Snoop SM MADs for port info, GUID info, and  P_Key table sets, so we can
+ * synthesize LID change, Client-Rereg, GID change, and P_Key change events.
  */
-static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad)
+static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad,
+		      u16 prev_lid)
 {
-	struct ib_event event;
+	struct ib_port_info *pinfo;
+	u16 lid;
+	__be16 *base;
+	u32 bn, pkey_change_bitmap;
+	int i;
+
 
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
-	    mad->mad_hdr.method == IB_MGMT_METHOD_SET) {
-		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
-			struct ib_port_info *pinfo =
-				(struct ib_port_info *) ((struct ib_smp *) mad)->data;
+	    mad->mad_hdr.method == IB_MGMT_METHOD_SET)
+		switch (mad->mad_hdr.attr_id) {
+		case IB_SMP_ATTR_PORT_INFO:
+			pinfo = (struct ib_port_info *) ((struct ib_smp *) mad)->data;
+			lid = be16_to_cpu(pinfo->lid);
 
-			update_sm_ah(to_mdev(ibdev), port_num,
+			update_sm_ah(dev, port_num,
 				     be16_to_cpu(pinfo->sm_lid),
 				     pinfo->neighbormtu_mastersmsl & 0xf);
 
-			event.device	       = ibdev;
-			event.element.port_num = port_num;
-
 			if (pinfo->clientrereg_resv_subnetto & 0x80)
-				event.event    = IB_EVENT_CLIENT_REREGISTER;
-			else
-				event.event    = IB_EVENT_LID_CHANGE;
+				handle_client_rereg_event(dev, port_num);
+
+			if (prev_lid != lid)
+				handle_lid_change_event(dev, port_num);
+			break;
+
+		case IB_SMP_ATTR_PKEY_TABLE:
+			if (!mlx4_is_mfunc(dev->dev)) {
+				mlx4_ib_dispatch_event(dev, port_num,
+						       IB_EVENT_PKEY_CHANGE);
+				break;
+			}
+
+			/* at this point, we are running in the master.
+			 * Slaves do not receive SMPs.
+			 */
+			bn  = be32_to_cpu(((struct ib_smp *)mad)->attr_mod) & 0xFFFF;
+			base = (__be16 *) &(((struct ib_smp *)mad)->data[0]);
+			pkey_change_bitmap = 0;
+			for (i = 0; i < 32; i++) {
+				pr_debug("PKEY[%d] = x%x\n",
+					 i + bn*32, be16_to_cpu(base[i]));
+				if (be16_to_cpu(base[i]) !=
+				    dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32]) {
+					pkey_change_bitmap |= (1 << i);
+					dev->pkeys.phys_pkey_cache[port_num - 1][i + bn*32] =
+						be16_to_cpu(base[i]);
+				}
+			}
+			pr_debug("PKEY Change event: port=%d, "
+				 "block=0x%x, change_bitmap=0x%x\n",
+				 port_num, bn, pkey_change_bitmap);
 
-			ib_dispatch_event(&event);
+			if (pkey_change_bitmap) {
+				mlx4_ib_dispatch_event(dev, port_num,
+						       IB_EVENT_PKEY_CHANGE);
+				if (!dev->sriov.is_going_down)
+					__propagate_pkey_ev(dev, port_num, bn,
+							    pkey_change_bitmap);
+			}
+			break;
+
+		case IB_SMP_ATTR_GUID_INFO:
+			/* paravirtualized master's guid is guid 0 -- does not change */
+			if (!mlx4_is_master(dev->dev))
+				mlx4_ib_dispatch_event(dev, port_num,
+						       IB_EVENT_GID_CHANGE);
+			/*if master, notify relevant slaves*/
+			if (mlx4_is_master(dev->dev) &&
+			    !dev->sriov.is_going_down) {
+				bn = be32_to_cpu(((struct ib_smp *)mad)->attr_mod);
+				mlx4_ib_update_cache_on_guid_change(dev, bn, port_num,
+								    (u8 *)(&((struct ib_smp *)mad)->data));
+				mlx4_ib_notify_slaves_on_guid_change(dev, bn, port_num,
+								     (u8 *)(&((struct ib_smp *)mad)->data));
+			}
+			break;
+
+		default:
+			break;
 		}
+}
+
+static void __propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+				int block, u32 change_bitmap)
+{
+	int i, ix, slave, err;
+	int have_event = 0;
 
-		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
-			event.device	       = ibdev;
-			event.event	       = IB_EVENT_PKEY_CHANGE;
-			event.element.port_num = port_num;
-			ib_dispatch_event(&event);
+	for (slave = 0; slave < dev->dev->caps.sqp_demux; slave++) {
+		if (slave == mlx4_master_func_num(dev->dev))
+			continue;
+		if (!mlx4_is_slave_active(dev->dev, slave))
+			continue;
+
+		have_event = 0;
+		for (i = 0; i < 32; i++) {
+			if (!(change_bitmap & (1 << i)))
+				continue;
+			for (ix = 0;
+			     ix < dev->dev->caps.pkey_table_len[port_num]; ix++) {
+				if (dev->pkeys.virt2phys_pkey[slave][port_num - 1]
+				    [ix] == i + 32 * block) {
+					err = mlx4_gen_pkey_eqe(dev->dev, slave, port_num);
+					pr_debug("propagate_pkey_ev: slave %d,"
+						 " port %d, ix %d (%d)\n",
+						 slave, port_num, ix, err);
+					have_event = 1;
+					break;
+				}
+			}
+			if (have_event)
+				break;
 		}
 	}
 }
@@ -185,13 +336,15 @@ static void smp_snoop(struct ib_device *ibdev, u8 port_num, struct ib_mad *mad)
 static void node_desc_override(struct ib_device *dev,
 			       struct ib_mad *mad)
 {
+	unsigned long flags;
+
 	if ((mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
 	     mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
 	    mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP &&
 	    mad->mad_hdr.attr_id == IB_SMP_ATTR_NODE_DESC) {
-		spin_lock(&to_mdev(dev)->sm_lock);
+		spin_lock_irqsave(&to_mdev(dev)->sm_lock, flags);
 		memcpy(((struct ib_smp *) mad)->data, dev->node_desc, 64);
-		spin_unlock(&to_mdev(dev)->sm_lock);
+		spin_unlock_irqrestore(&to_mdev(dev)->sm_lock, flags);
 	}
 }
 
@@ -201,35 +354,391 @@ static void forward_trap(struct mlx4_ib_dev *dev, u8 port_num, struct ib_mad *ma
 	struct ib_mad_send_buf *send_buf;
 	struct ib_mad_agent *agent = dev->send_agent[port_num - 1][qpn];
 	int ret;
+	unsigned long flags;
 
 	if (agent) {
 		send_buf = ib_create_send_mad(agent, qpn, 0, 0, IB_MGMT_MAD_HDR,
 					      IB_MGMT_MAD_DATA, GFP_ATOMIC);
+		if (IS_ERR(send_buf))
+			return;
 		/*
 		 * We rely here on the fact that MLX QPs don't use the
 		 * address handle after the send is posted (this is
 		 * wrong following the IB spec strictly, but we know
 		 * it's OK for our devices).
 		 */
-		spin_lock(&dev->sm_lock);
+		spin_lock_irqsave(&dev->sm_lock, flags);
 		memcpy(send_buf->mad, mad, sizeof *mad);
 		if ((send_buf->ah = dev->sm_ah[port_num - 1]))
 			ret = ib_post_send_mad(send_buf, NULL);
 		else
 			ret = -EINVAL;
-		spin_unlock(&dev->sm_lock);
+		spin_unlock_irqrestore(&dev->sm_lock, flags);
 
 		if (ret)
 			ib_free_send_mad(send_buf);
 	}
 }
 
-int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags,	u8 port_num,
+static int mlx4_ib_demux_sa_handler(struct ib_device *ibdev, int port, int slave,
+							     struct ib_sa_mad *sa_mad)
+{
+	int ret = 0;
+
+	/* dispatch to different sa handlers */
+	switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) {
+	case IB_SA_ATTR_MC_MEMBER_REC:
+		ret = mlx4_ib_mcg_demux_handler(ibdev, port, slave, sa_mad);
+		break;
+	default:
+		break;
+	}
+	return ret;
+}
+
+int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	int i;
+
+	for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+		if (dev->sriov.demux[port - 1].guid_cache[i] == guid)
+			return i;
+	}
+	return -1;
+}
+
+
+static int find_slave_port_pkey_ix(struct mlx4_ib_dev *dev, int slave,
+				   u8 port, u16 pkey, u16 *ix)
+{
+	int i, ret;
+	u8 unassigned_pkey_ix, pkey_ix, partial_ix = 0xFF;
+	u16 slot_pkey;
+
+	if (slave == mlx4_master_func_num(dev->dev))
+		return ib_find_cached_pkey(&dev->ib_dev, port, pkey, ix);
+
+	unassigned_pkey_ix = dev->dev->phys_caps.pkey_phys_table_len[port] - 1;
+
+	for (i = 0; i < dev->dev->caps.pkey_table_len[port]; i++) {
+		if (dev->pkeys.virt2phys_pkey[slave][port - 1][i] == unassigned_pkey_ix)
+			continue;
+
+		pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][i];
+
+		ret = ib_get_cached_pkey(&dev->ib_dev, port, pkey_ix, &slot_pkey);
+		if (ret)
+			continue;
+		if ((slot_pkey & 0x7FFF) == (pkey & 0x7FFF)) {
+			if (slot_pkey & 0x8000) {
+				*ix = (u16) pkey_ix;
+				return 0;
+			} else {
+				/* take first partial pkey index found */
+				if (partial_ix == 0xFF)
+					partial_ix = pkey_ix;
+			}
+		}
+	}
+
+	if (partial_ix < 0xFF) {
+		*ix = (u16) partial_ix;
+		return 0;
+	}
+
+	return -EINVAL;
+}
+
+int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
+			  enum ib_qp_type dest_qpt, struct ib_wc *wc,
+			  struct ib_grh *grh, struct ib_mad *mad)
+{
+	struct ib_sge list;
+	struct ib_send_wr wr, *bad_wr;
+	struct mlx4_ib_demux_pv_ctx *tun_ctx;
+	struct mlx4_ib_demux_pv_qp *tun_qp;
+	struct mlx4_rcv_tunnel_mad *tun_mad;
+	struct ib_ah_attr attr;
+	struct ib_ah *ah;
+	struct ib_qp *src_qp = NULL;
+	unsigned tun_tx_ix = 0;
+	int dqpn;
+	int ret = 0;
+	u16 tun_pkey_ix;
+	u16 cached_pkey;
+	u8 is_eth = dev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
+
+	if (dest_qpt > IB_QPT_GSI)
+		return -EINVAL;
+
+	tun_ctx = dev->sriov.demux[port-1].tun[slave];
+
+	/* check if proxy qp created */
+	if (!tun_ctx || tun_ctx->state != DEMUX_PV_STATE_ACTIVE)
+		return -EAGAIN;
+
+	if (!dest_qpt)
+		tun_qp = &tun_ctx->qp[0];
+	else
+		tun_qp = &tun_ctx->qp[1];
+
+	/* compute P_Key index to put in tunnel header for slave */
+	if (dest_qpt) {
+		u16 pkey_ix;
+		ret = ib_get_cached_pkey(&dev->ib_dev, port, wc->pkey_index, &cached_pkey);
+		if (ret)
+			return -EINVAL;
+
+		ret = find_slave_port_pkey_ix(dev, slave, port, cached_pkey, &pkey_ix);
+		if (ret)
+			return -EINVAL;
+		tun_pkey_ix = pkey_ix;
+	} else
+		tun_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+
+	dqpn = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave + port + (dest_qpt * 2) - 1;
+
+	/* get tunnel tx data buf for slave */
+	src_qp = tun_qp->qp;
+
+	/* create ah. Just need an empty one with the port num for the post send.
+	 * The driver will set the force loopback bit in post_send */
+	memset(&attr, 0, sizeof attr);
+	attr.port_num = port;
+	if (is_eth) {
+		memcpy(&attr.grh.dgid.raw[0], &grh->dgid.raw[0], 16);
+		attr.ah_flags = IB_AH_GRH;
+	}
+	ah = ib_create_ah(tun_ctx->pd, &attr);
+	if (IS_ERR(ah))
+		return -ENOMEM;
+
+	/* allocate tunnel tx buf after pass failure returns */
+	spin_lock(&tun_qp->tx_lock);
+	if (tun_qp->tx_ix_head - tun_qp->tx_ix_tail >=
+	    (MLX4_NUM_TUNNEL_BUFS - 1))
+		ret = -EAGAIN;
+	else
+		tun_tx_ix = (++tun_qp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
+	spin_unlock(&tun_qp->tx_lock);
+	if (ret)
+		goto out;
+
+	tun_mad = (struct mlx4_rcv_tunnel_mad *) (tun_qp->tx_ring[tun_tx_ix].buf.addr);
+	if (tun_qp->tx_ring[tun_tx_ix].ah)
+		ib_destroy_ah(tun_qp->tx_ring[tun_tx_ix].ah);
+	tun_qp->tx_ring[tun_tx_ix].ah = ah;
+	ib_dma_sync_single_for_cpu(&dev->ib_dev,
+				   tun_qp->tx_ring[tun_tx_ix].buf.map,
+				   sizeof (struct mlx4_rcv_tunnel_mad),
+				   DMA_TO_DEVICE);
+
+	/* copy over to tunnel buffer */
+	if (grh)
+		memcpy(&tun_mad->grh, grh, sizeof *grh);
+	memcpy(&tun_mad->mad, mad, sizeof *mad);
+
+	/* adjust tunnel data */
+	tun_mad->hdr.pkey_index = cpu_to_be16(tun_pkey_ix);
+	tun_mad->hdr.flags_src_qp = cpu_to_be32(wc->src_qp & 0xFFFFFF);
+	tun_mad->hdr.g_ml_path = (grh && (wc->wc_flags & IB_WC_GRH)) ? 0x80 : 0;
+
+	if (is_eth) {
+		u16 vlan = 0;
+		if (mlx4_get_slave_default_vlan(dev->dev, port, slave, &vlan,
+						NULL)) {
+			/* VST mode */
+			if (vlan != wc->vlan_id)
+				/* Packet vlan is not the VST-assigned vlan.
+				 * Drop the packet.
+				 */
+				goto out;
+			 else
+				/* Remove the vlan tag before forwarding
+				 * the packet to the VF.
+				 */
+				vlan = 0xffff;
+		} else {
+			vlan = wc->vlan_id;
+		}
+
+		tun_mad->hdr.sl_vid = cpu_to_be16(vlan);
+		memcpy((char *)&tun_mad->hdr.mac_31_0, &(wc->smac[0]), 4);
+		memcpy((char *)&tun_mad->hdr.slid_mac_47_32, &(wc->smac[4]), 2);
+	} else {
+		tun_mad->hdr.sl_vid = cpu_to_be16(((u16)(wc->sl)) << 12);
+		tun_mad->hdr.slid_mac_47_32 = cpu_to_be16(wc->slid);
+	}
+
+	ib_dma_sync_single_for_device(&dev->ib_dev,
+				      tun_qp->tx_ring[tun_tx_ix].buf.map,
+				      sizeof (struct mlx4_rcv_tunnel_mad),
+				      DMA_TO_DEVICE);
+
+	list.addr = tun_qp->tx_ring[tun_tx_ix].buf.map;
+	list.length = sizeof (struct mlx4_rcv_tunnel_mad);
+	list.lkey = tun_ctx->mr->lkey;
+
+	wr.wr.ud.ah = ah;
+	wr.wr.ud.port_num = port;
+	wr.wr.ud.remote_qkey = IB_QP_SET_QKEY;
+	wr.wr.ud.remote_qpn = dqpn;
+	wr.next = NULL;
+	wr.wr_id = ((u64) tun_tx_ix) | MLX4_TUN_SET_WRID_QPN(dest_qpt);
+	wr.sg_list = &list;
+	wr.num_sge = 1;
+	wr.opcode = IB_WR_SEND;
+	wr.send_flags = IB_SEND_SIGNALED;
+
+	ret = ib_post_send(src_qp, &wr, &bad_wr);
+out:
+	if (ret)
+		ib_destroy_ah(ah);
+	return ret;
+}
+
+static int mlx4_ib_demux_mad(struct ib_device *ibdev, u8 port,
+			struct ib_wc *wc, struct ib_grh *grh,
+			struct ib_mad *mad)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	int err;
+	int slave;
+	u8 *slave_id;
+	int is_eth = 0;
+
+	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
+		is_eth = 0;
+	else
+		is_eth = 1;
+
+	if (is_eth) {
+		if (!(wc->wc_flags & IB_WC_GRH)) {
+			mlx4_ib_warn(ibdev, "RoCE grh not present.\n");
+			return -EINVAL;
+		}
+		if (mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_CM) {
+			mlx4_ib_warn(ibdev, "RoCE mgmt class is not CM\n");
+			return -EINVAL;
+		}
+		if (mlx4_get_slave_from_roce_gid(dev->dev, port, grh->dgid.raw, &slave)) {
+			mlx4_ib_warn(ibdev, "failed matching grh\n");
+			return -ENOENT;
+		}
+		if (slave >= dev->dev->caps.sqp_demux) {
+			mlx4_ib_warn(ibdev, "slave id: %d is bigger than allowed:%d\n",
+				     slave, dev->dev->caps.sqp_demux);
+			return -ENOENT;
+		}
+
+		if (mlx4_ib_demux_cm_handler(ibdev, port, NULL, mad))
+			return 0;
+
+		err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad);
+		if (err)
+			pr_debug("failed sending to slave %d via tunnel qp (%d)\n",
+				 slave, err);
+		return 0;
+	}
+
+	/* Initially assume that this mad is for us */
+	slave = mlx4_master_func_num(dev->dev);
+
+	/* See if the slave id is encoded in a response mad */
+	if (mad->mad_hdr.method & 0x80) {
+		slave_id = (u8 *) &mad->mad_hdr.tid;
+		slave = *slave_id;
+		if (slave != 255) /*255 indicates the dom0*/
+			*slave_id = 0; /* remap tid */
+	}
+
+	/* If a grh is present, we demux according to it */
+	if (wc->wc_flags & IB_WC_GRH) {
+		slave = mlx4_ib_find_real_gid(ibdev, port, grh->dgid.global.interface_id);
+		if (slave < 0) {
+			mlx4_ib_warn(ibdev, "failed matching grh\n");
+			return -ENOENT;
+		}
+	}
+	/* Class-specific handling */
+	switch (mad->mad_hdr.mgmt_class) {
+	case IB_MGMT_CLASS_SUBN_LID_ROUTED:
+	case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
+		/* 255 indicates the dom0 */
+		if (slave != 255 && slave != mlx4_master_func_num(dev->dev)) {
+			if (!mlx4_vf_smi_enabled(dev->dev, slave, port))
+				return -EPERM;
+			/* for a VF. drop unsolicited MADs */
+			if (!(mad->mad_hdr.method & IB_MGMT_METHOD_RESP)) {
+				mlx4_ib_warn(ibdev, "demux QP0. rejecting unsolicited mad for slave %d class 0x%x, method 0x%x\n",
+					     slave, mad->mad_hdr.mgmt_class,
+					     mad->mad_hdr.method);
+				return -EINVAL;
+			}
+		}
+		break;
+	case IB_MGMT_CLASS_SUBN_ADM:
+		if (mlx4_ib_demux_sa_handler(ibdev, port, slave,
+					     (struct ib_sa_mad *) mad))
+			return 0;
+		break;
+	case IB_MGMT_CLASS_CM:
+		if (mlx4_ib_demux_cm_handler(ibdev, port, &slave, mad))
+			return 0;
+		break;
+	case IB_MGMT_CLASS_DEVICE_MGMT:
+		if (mad->mad_hdr.method != IB_MGMT_METHOD_GET_RESP)
+			return 0;
+		break;
+	default:
+		/* Drop unsupported classes for slaves in tunnel mode */
+		if (slave != mlx4_master_func_num(dev->dev)) {
+			pr_debug("dropping unsupported ingress mad from class:%d "
+				 "for slave:%d\n", mad->mad_hdr.mgmt_class, slave);
+			return 0;
+		}
+	}
+	/*make sure that no slave==255 was not handled yet.*/
+	if (slave >= dev->dev->caps.sqp_demux) {
+		mlx4_ib_warn(ibdev, "slave id: %d is bigger than allowed:%d\n",
+			     slave, dev->dev->caps.sqp_demux);
+		return -ENOENT;
+	}
+
+	err = mlx4_ib_send_to_slave(dev, slave, port, wc->qp->qp_type, wc, grh, mad);
+	if (err)
+		pr_debug("failed sending to slave %d via tunnel qp (%d)\n",
+			 slave, err);
+	return 0;
+}
+
+static int ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
 			struct ib_wc *in_wc, struct ib_grh *in_grh,
 			struct ib_mad *in_mad, struct ib_mad *out_mad)
 {
-	u16 slid;
+	u16 slid, prev_lid = 0;
 	int err;
+	struct ib_port_attr pattr;
+
+	if (in_wc && in_wc->qp->qp_num) {
+		pr_debug("received MAD: slid:%d sqpn:%d "
+			"dlid_bits:%d dqpn:%d wc_flags:0x%x, cls %x, mtd %x, atr %x\n",
+			in_wc->slid, in_wc->src_qp,
+			in_wc->dlid_path_bits,
+			in_wc->qp->qp_num,
+			in_wc->wc_flags,
+			in_mad->mad_hdr.mgmt_class, in_mad->mad_hdr.method,
+			be16_to_cpu(in_mad->mad_hdr.attr_id));
+		if (in_wc->wc_flags & IB_WC_GRH) {
+			pr_debug("sgid_hi:0x%016llx sgid_lo:0x%016llx\n",
+				 be64_to_cpu(in_grh->sgid.global.subnet_prefix),
+				 be64_to_cpu(in_grh->sgid.global.interface_id));
+			pr_debug("dgid_hi:0x%016llx dgid_lo:0x%016llx\n",
+				 be64_to_cpu(in_grh->dgid.global.subnet_prefix),
+				 be64_to_cpu(in_grh->dgid.global.interface_id));
+		}
+	}
 
 	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
 
@@ -246,12 +755,9 @@ int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags,	u8 port_num,
 			return IB_MAD_RESULT_SUCCESS;
 
 		/*
-		 * Don't process SMInfo queries or vendor-specific
-		 * MADs -- the SMA can't handle them.
+		 * Don't process SMInfo queries -- the SMA can't handle them.
 		 */
-		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO ||
-		    ((in_mad->mad_hdr.attr_id & IB_SMP_ATTR_VENDOR_MASK) ==
-		     IB_SMP_ATTR_VENDOR_MASK))
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO)
 			return IB_MAD_RESULT_SUCCESS;
 	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
 		   in_mad->mad_hdr.mgmt_class == MLX4_IB_VENDOR_CLASS1   ||
@@ -263,16 +769,27 @@ int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags,	u8 port_num,
 	} else
 		return IB_MAD_RESULT_SUCCESS;
 
+	if ((in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	     in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
+	    in_mad->mad_hdr.method == IB_MGMT_METHOD_SET &&
+	    in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
+	    !ib_query_port(ibdev, port_num, &pattr))
+		prev_lid = pattr.lid;
+
 	err = mlx4_MAD_IFC(to_mdev(ibdev),
-			   mad_flags & IB_MAD_IGNORE_MKEY,
-			   mad_flags & IB_MAD_IGNORE_BKEY,
+			   (mad_flags & IB_MAD_IGNORE_MKEY ? MLX4_MAD_IFC_IGNORE_MKEY : 0) |
+			   (mad_flags & IB_MAD_IGNORE_BKEY ? MLX4_MAD_IFC_IGNORE_BKEY : 0) |
+			   MLX4_MAD_IFC_NET_VIEW,
 			   port_num, in_wc, in_grh, in_mad, out_mad);
 	if (err)
 		return IB_MAD_RESULT_FAILURE;
 
 	if (!out_mad->mad_hdr.status) {
-		smp_snoop(ibdev, port_num, in_mad);
-		node_desc_override(ibdev, out_mad);
+		if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_PORT_MNG_CHG_EV))
+			smp_snoop(ibdev, port_num, in_mad, prev_lid);
+		/* slaves get node desc from FW */
+		if (!mlx4_is_slave(to_mdev(ibdev)->dev))
+			node_desc_override(ibdev, out_mad);
 	}
 
 	/* set return bit in status of directed route responses */
@@ -286,9 +803,77 @@ int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags,	u8 port_num,
 	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
 }
 
+static void edit_counter(struct mlx4_counter *cnt,
+					struct ib_pma_portcounters *pma_cnt)
+{
+	pma_cnt->port_xmit_data = cpu_to_be32((be64_to_cpu(cnt->tx_bytes)>>2));
+	pma_cnt->port_rcv_data  = cpu_to_be32((be64_to_cpu(cnt->rx_bytes)>>2));
+	pma_cnt->port_xmit_packets = cpu_to_be32(be64_to_cpu(cnt->tx_frames));
+	pma_cnt->port_rcv_packets  = cpu_to_be32(be64_to_cpu(cnt->rx_frames));
+}
+
+static int iboe_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	struct mlx4_cmd_mailbox *mailbox;
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	int err;
+	u32 inmod = dev->counters[port_num - 1] & 0xffff;
+	u8 mode;
+
+	if (in_mad->mad_hdr.mgmt_class != IB_MGMT_CLASS_PERF_MGMT)
+		return -EINVAL;
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev->dev);
+	if (IS_ERR(mailbox))
+		return IB_MAD_RESULT_FAILURE;
+
+	err = mlx4_cmd_box(dev->dev, 0, mailbox->dma, inmod, 0,
+			   MLX4_CMD_QUERY_IF_STAT, MLX4_CMD_TIME_CLASS_C,
+			   MLX4_CMD_WRAPPED);
+	if (err)
+		err = IB_MAD_RESULT_FAILURE;
+	else {
+		memset(out_mad->data, 0, sizeof out_mad->data);
+		mode = ((struct mlx4_counter *)mailbox->buf)->counter_mode;
+		switch (mode & 0xf) {
+		case 0:
+			edit_counter(mailbox->buf,
+						(void *)(out_mad->data + 40));
+			err = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+			break;
+		default:
+			err = IB_MAD_RESULT_FAILURE;
+		}
+	}
+
+	mlx4_free_cmd_mailbox(dev->dev, mailbox);
+
+	return err;
+}
+
+int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	switch (rdma_port_get_link_layer(ibdev, port_num)) {
+	case IB_LINK_LAYER_INFINIBAND:
+		return ib_process_mad(ibdev, mad_flags, port_num, in_wc,
+				      in_grh, in_mad, out_mad);
+	case IB_LINK_LAYER_ETHERNET:
+		return iboe_process_mad(ibdev, mad_flags, port_num, in_wc,
+					  in_grh, in_mad, out_mad);
+	default:
+		return -EINVAL;
+	}
+}
+
 static void send_handler(struct ib_mad_agent *agent,
 			 struct ib_mad_send_wc *mad_send_wc)
 {
+	if (mad_send_wc->send_buf->context[0])
+		ib_destroy_ah(mad_send_wc->send_buf->context[0]);
 	ib_free_send_mad(mad_send_wc->send_buf);
 }
 
@@ -297,24 +882,30 @@ int mlx4_ib_mad_init(struct mlx4_ib_dev *dev)
 	struct ib_mad_agent *agent;
 	int p, q;
 	int ret;
+	enum rdma_link_layer ll;
 
-	for (p = 0; p < dev->dev->caps.num_ports; ++p)
+	for (p = 0; p < dev->num_ports; ++p) {
+		ll = rdma_port_get_link_layer(&dev->ib_dev, p + 1);
 		for (q = 0; q <= 1; ++q) {
-			agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
-						      q ? IB_QPT_GSI : IB_QPT_SMI,
-						      NULL, 0, send_handler,
-						      NULL, NULL);
-			if (IS_ERR(agent)) {
-				ret = PTR_ERR(agent);
-				goto err;
-			}
-			dev->send_agent[p][q] = agent;
+			if (ll == IB_LINK_LAYER_INFINIBAND) {
+				agent = ib_register_mad_agent(&dev->ib_dev, p + 1,
+							      q ? IB_QPT_GSI : IB_QPT_SMI,
+							      NULL, 0, send_handler,
+							      NULL, NULL);
+				if (IS_ERR(agent)) {
+					ret = PTR_ERR(agent);
+					goto err;
+				}
+				dev->send_agent[p][q] = agent;
+			} else
+				dev->send_agent[p][q] = NULL;
 		}
+	}
 
 	return 0;
 
 err:
-	for (p = 0; p < dev->dev->caps.num_ports; ++p)
+	for (p = 0; p < dev->num_ports; ++p)
 		for (q = 0; q <= 1; ++q)
 			if (dev->send_agent[p][q])
 				ib_unregister_mad_agent(dev->send_agent[p][q]);
@@ -327,14 +918,1246 @@ void mlx4_ib_mad_cleanup(struct mlx4_ib_dev *dev)
 	struct ib_mad_agent *agent;
 	int p, q;
 
-	for (p = 0; p < dev->dev->caps.num_ports; ++p) {
+	for (p = 0; p < dev->num_ports; ++p) {
 		for (q = 0; q <= 1; ++q) {
 			agent = dev->send_agent[p][q];
-			dev->send_agent[p][q] = NULL;
-			ib_unregister_mad_agent(agent);
+			if (agent) {
+				dev->send_agent[p][q] = NULL;
+				ib_unregister_mad_agent(agent);
+			}
 		}
 
 		if (dev->sm_ah[p])
 			ib_destroy_ah(dev->sm_ah[p]);
 	}
 }
+
+static void handle_lid_change_event(struct mlx4_ib_dev *dev, u8 port_num)
+{
+	mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_LID_CHANGE);
+
+	if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+		mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
+					    MLX4_EQ_PORT_INFO_LID_CHANGE_MASK);
+}
+
+static void handle_client_rereg_event(struct mlx4_ib_dev *dev, u8 port_num)
+{
+	/* re-configure the alias-guid and mcg's */
+	if (mlx4_is_master(dev->dev)) {
+		mlx4_ib_invalidate_all_guid_record(dev, port_num);
+
+		if (!dev->sriov.is_going_down) {
+			mlx4_ib_mcg_port_cleanup(&dev->sriov.demux[port_num - 1], 0);
+			mlx4_gen_slaves_port_mgt_ev(dev->dev, port_num,
+						    MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK);
+		}
+	}
+	mlx4_ib_dispatch_event(dev, port_num, IB_EVENT_CLIENT_REREGISTER);
+}
+
+static void propagate_pkey_ev(struct mlx4_ib_dev *dev, int port_num,
+			      struct mlx4_eqe *eqe)
+{
+	__propagate_pkey_ev(dev, port_num, GET_BLK_PTR_FROM_EQE(eqe),
+			    GET_MASK_FROM_EQE(eqe));
+}
+
+static void handle_slaves_guid_change(struct mlx4_ib_dev *dev, u8 port_num,
+				      u32 guid_tbl_blk_num, u32 change_bitmap)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad  = NULL;
+	u16 i;
+
+	if (!mlx4_is_mfunc(dev->dev) || !mlx4_is_master(dev->dev))
+		return;
+
+	in_mad  = kmalloc(sizeof *in_mad, GFP_KERNEL);
+	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
+	if (!in_mad || !out_mad) {
+		mlx4_ib_warn(&dev->ib_dev, "failed to allocate memory for guid info mads\n");
+		goto out;
+	}
+
+	guid_tbl_blk_num  *= 4;
+
+	for (i = 0; i < 4; i++) {
+		if (change_bitmap && (!((change_bitmap >> (8 * i)) & 0xff)))
+			continue;
+		memset(in_mad, 0, sizeof *in_mad);
+		memset(out_mad, 0, sizeof *out_mad);
+
+		in_mad->base_version  = 1;
+		in_mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+		in_mad->class_version = 1;
+		in_mad->method        = IB_MGMT_METHOD_GET;
+		in_mad->attr_id       = IB_SMP_ATTR_GUID_INFO;
+		in_mad->attr_mod      = cpu_to_be32(guid_tbl_blk_num + i);
+
+		if (mlx4_MAD_IFC(dev,
+				 MLX4_MAD_IFC_IGNORE_KEYS | MLX4_MAD_IFC_NET_VIEW,
+				 port_num, NULL, NULL, in_mad, out_mad)) {
+			mlx4_ib_warn(&dev->ib_dev, "Failed in get GUID INFO MAD_IFC\n");
+			goto out;
+		}
+
+		mlx4_ib_update_cache_on_guid_change(dev, guid_tbl_blk_num + i,
+						    port_num,
+						    (u8 *)(&((struct ib_smp *)out_mad)->data));
+		mlx4_ib_notify_slaves_on_guid_change(dev, guid_tbl_blk_num + i,
+						     port_num,
+						     (u8 *)(&((struct ib_smp *)out_mad)->data));
+	}
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return;
+}
+
+void handle_port_mgmt_change_event(struct work_struct *work)
+{
+	struct ib_event_work *ew = container_of(work, struct ib_event_work, work);
+	struct mlx4_ib_dev *dev = ew->ib_dev;
+	struct mlx4_eqe *eqe = &(ew->ib_eqe);
+	u8 port = eqe->event.port_mgmt_change.port;
+	u32 changed_attr;
+	u32 tbl_block;
+	u32 change_bitmap;
+
+	switch (eqe->subtype) {
+	case MLX4_DEV_PMC_SUBTYPE_PORT_INFO:
+		changed_attr = be32_to_cpu(eqe->event.port_mgmt_change.params.port_info.changed_attr);
+
+		/* Update the SM ah - This should be done before handling
+		   the other changed attributes so that MADs can be sent to the SM */
+		if (changed_attr & MSTR_SM_CHANGE_MASK) {
+			u16 lid = be16_to_cpu(eqe->event.port_mgmt_change.params.port_info.mstr_sm_lid);
+			u8 sl = eqe->event.port_mgmt_change.params.port_info.mstr_sm_sl & 0xf;
+			update_sm_ah(dev, port, lid, sl);
+		}
+
+		/* Check if it is a lid change event */
+		if (changed_attr & MLX4_EQ_PORT_INFO_LID_CHANGE_MASK)
+			handle_lid_change_event(dev, port);
+
+		/* Generate GUID changed event */
+		if (changed_attr & MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK) {
+			mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
+			/*if master, notify all slaves*/
+			if (mlx4_is_master(dev->dev))
+				mlx4_gen_slaves_port_mgt_ev(dev->dev, port,
+							    MLX4_EQ_PORT_INFO_GID_PFX_CHANGE_MASK);
+		}
+
+		if (changed_attr & MLX4_EQ_PORT_INFO_CLIENT_REREG_MASK)
+			handle_client_rereg_event(dev, port);
+		break;
+
+	case MLX4_DEV_PMC_SUBTYPE_PKEY_TABLE:
+		mlx4_ib_dispatch_event(dev, port, IB_EVENT_PKEY_CHANGE);
+		if (mlx4_is_master(dev->dev) && !dev->sriov.is_going_down)
+			propagate_pkey_ev(dev, port, eqe);
+		break;
+	case MLX4_DEV_PMC_SUBTYPE_GUID_INFO:
+		/* paravirtualized master's guid is guid 0 -- does not change */
+		if (!mlx4_is_master(dev->dev))
+			mlx4_ib_dispatch_event(dev, port, IB_EVENT_GID_CHANGE);
+		/*if master, notify relevant slaves*/
+		else if (!dev->sriov.is_going_down) {
+			tbl_block = GET_BLK_PTR_FROM_EQE(eqe);
+			change_bitmap = GET_MASK_FROM_EQE(eqe);
+			handle_slaves_guid_change(dev, port, tbl_block, change_bitmap);
+		}
+		break;
+	default:
+		pr_warn("Unsupported subtype 0x%x for "
+			"Port Management Change event\n", eqe->subtype);
+	}
+
+	kfree(ew);
+}
+
+void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
+			    enum ib_event_type type)
+{
+	struct ib_event event;
+
+	event.device		= &dev->ib_dev;
+	event.element.port_num	= port_num;
+	event.event		= type;
+
+	ib_dispatch_event(&event);
+}
+
+static void mlx4_ib_tunnel_comp_handler(struct ib_cq *cq, void *arg)
+{
+	unsigned long flags;
+	struct mlx4_ib_demux_pv_ctx *ctx = cq->cq_context;
+	struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	if (!dev->sriov.is_going_down && ctx->state == DEMUX_PV_STATE_ACTIVE)
+		queue_work(ctx->wq, &ctx->work);
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+}
+
+static int mlx4_ib_post_pv_qp_buf(struct mlx4_ib_demux_pv_ctx *ctx,
+				  struct mlx4_ib_demux_pv_qp *tun_qp,
+				  int index)
+{
+	struct ib_sge sg_list;
+	struct ib_recv_wr recv_wr, *bad_recv_wr;
+	int size;
+
+	size = (tun_qp->qp->qp_type == IB_QPT_UD) ?
+		sizeof (struct mlx4_tunnel_mad) : sizeof (struct mlx4_mad_rcv_buf);
+
+	sg_list.addr = tun_qp->ring[index].map;
+	sg_list.length = size;
+	sg_list.lkey = ctx->mr->lkey;
+
+	recv_wr.next = NULL;
+	recv_wr.sg_list = &sg_list;
+	recv_wr.num_sge = 1;
+	recv_wr.wr_id = (u64) index | MLX4_TUN_WRID_RECV |
+		MLX4_TUN_SET_WRID_QPN(tun_qp->proxy_qpt);
+	ib_dma_sync_single_for_device(ctx->ib_dev, tun_qp->ring[index].map,
+				      size, DMA_FROM_DEVICE);
+	return ib_post_recv(tun_qp->qp, &recv_wr, &bad_recv_wr);
+}
+
+static int mlx4_ib_multiplex_sa_handler(struct ib_device *ibdev, int port,
+		int slave, struct ib_sa_mad *sa_mad)
+{
+	int ret = 0;
+
+	/* dispatch to different sa handlers */
+	switch (be16_to_cpu(sa_mad->mad_hdr.attr_id)) {
+	case IB_SA_ATTR_MC_MEMBER_REC:
+		ret = mlx4_ib_mcg_multiplex_handler(ibdev, port, slave, sa_mad);
+		break;
+	default:
+		break;
+	}
+	return ret;
+}
+
+static int is_proxy_qp0(struct mlx4_ib_dev *dev, int qpn, int slave)
+{
+	int proxy_start = dev->dev->phys_caps.base_proxy_sqpn + 8 * slave;
+
+	return (qpn >= proxy_start && qpn <= proxy_start + 1);
+}
+
+
+int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
+			 enum ib_qp_type dest_qpt, u16 pkey_index,
+			 u32 remote_qpn, u32 qkey, struct ib_ah_attr *attr,
+			 u8 *s_mac, struct ib_mad *mad)
+{
+	struct ib_sge list;
+	struct ib_send_wr wr, *bad_wr;
+	struct mlx4_ib_demux_pv_ctx *sqp_ctx;
+	struct mlx4_ib_demux_pv_qp *sqp;
+	struct mlx4_mad_snd_buf *sqp_mad;
+	struct ib_ah *ah;
+	struct ib_qp *send_qp = NULL;
+	unsigned wire_tx_ix = 0;
+	int ret = 0;
+	u16 wire_pkey_ix;
+	int src_qpnum;
+	u8 sgid_index;
+
+
+	sqp_ctx = dev->sriov.sqps[port-1];
+
+	/* check if proxy qp created */
+	if (!sqp_ctx || sqp_ctx->state != DEMUX_PV_STATE_ACTIVE)
+		return -EAGAIN;
+
+	if (dest_qpt == IB_QPT_SMI) {
+		src_qpnum = 0;
+		sqp = &sqp_ctx->qp[0];
+		wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][0];
+	} else {
+		src_qpnum = 1;
+		sqp = &sqp_ctx->qp[1];
+		wire_pkey_ix = dev->pkeys.virt2phys_pkey[slave][port - 1][pkey_index];
+	}
+
+	send_qp = sqp->qp;
+
+	/* create ah */
+	sgid_index = attr->grh.sgid_index;
+	attr->grh.sgid_index = 0;
+	ah = ib_create_ah(sqp_ctx->pd, attr);
+	if (IS_ERR(ah))
+		return -ENOMEM;
+	attr->grh.sgid_index = sgid_index;
+	to_mah(ah)->av.ib.gid_index = sgid_index;
+	/* get rid of force-loopback bit */
+	to_mah(ah)->av.ib.port_pd &= cpu_to_be32(0x7FFFFFFF);
+	spin_lock(&sqp->tx_lock);
+	if (sqp->tx_ix_head - sqp->tx_ix_tail >=
+	    (MLX4_NUM_TUNNEL_BUFS - 1))
+		ret = -EAGAIN;
+	else
+		wire_tx_ix = (++sqp->tx_ix_head) & (MLX4_NUM_TUNNEL_BUFS - 1);
+	spin_unlock(&sqp->tx_lock);
+	if (ret)
+		goto out;
+
+	sqp_mad = (struct mlx4_mad_snd_buf *) (sqp->tx_ring[wire_tx_ix].buf.addr);
+	if (sqp->tx_ring[wire_tx_ix].ah)
+		ib_destroy_ah(sqp->tx_ring[wire_tx_ix].ah);
+	sqp->tx_ring[wire_tx_ix].ah = ah;
+	ib_dma_sync_single_for_cpu(&dev->ib_dev,
+				   sqp->tx_ring[wire_tx_ix].buf.map,
+				   sizeof (struct mlx4_mad_snd_buf),
+				   DMA_TO_DEVICE);
+
+	memcpy(&sqp_mad->payload, mad, sizeof *mad);
+
+	ib_dma_sync_single_for_device(&dev->ib_dev,
+				      sqp->tx_ring[wire_tx_ix].buf.map,
+				      sizeof (struct mlx4_mad_snd_buf),
+				      DMA_TO_DEVICE);
+
+	list.addr = sqp->tx_ring[wire_tx_ix].buf.map;
+	list.length = sizeof (struct mlx4_mad_snd_buf);
+	list.lkey = sqp_ctx->mr->lkey;
+
+	wr.wr.ud.ah = ah;
+	wr.wr.ud.port_num = port;
+	wr.wr.ud.pkey_index = wire_pkey_ix;
+	wr.wr.ud.remote_qkey = qkey;
+	wr.wr.ud.remote_qpn = remote_qpn;
+	wr.next = NULL;
+	wr.wr_id = ((u64) wire_tx_ix) | MLX4_TUN_SET_WRID_QPN(src_qpnum);
+	wr.sg_list = &list;
+	wr.num_sge = 1;
+	wr.opcode = IB_WR_SEND;
+	wr.send_flags = IB_SEND_SIGNALED;
+	if (s_mac)
+		memcpy(to_mah(ah)->av.eth.s_mac, s_mac, 6);
+
+
+	ret = ib_post_send(send_qp, &wr, &bad_wr);
+out:
+	if (ret)
+		ib_destroy_ah(ah);
+	return ret;
+}
+
+static int get_slave_base_gid_ix(struct mlx4_ib_dev *dev, int slave, int port)
+{
+	if (rdma_port_get_link_layer(&dev->ib_dev, port) == IB_LINK_LAYER_INFINIBAND)
+		return slave;
+	return mlx4_get_base_gid_ix(dev->dev, slave, port);
+}
+
+static void fill_in_real_sgid_index(struct mlx4_ib_dev *dev, int slave, int port,
+				    struct ib_ah_attr *ah_attr)
+{
+	if (rdma_port_get_link_layer(&dev->ib_dev, port) == IB_LINK_LAYER_INFINIBAND)
+		ah_attr->grh.sgid_index = slave;
+	else
+		ah_attr->grh.sgid_index += get_slave_base_gid_ix(dev, slave, port);
+}
+
+static void mlx4_ib_multiplex_mad(struct mlx4_ib_demux_pv_ctx *ctx, struct ib_wc *wc)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+	struct mlx4_ib_demux_pv_qp *tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc->wr_id)];
+	int wr_ix = wc->wr_id & (MLX4_NUM_TUNNEL_BUFS - 1);
+	struct mlx4_tunnel_mad *tunnel = tun_qp->ring[wr_ix].addr;
+	struct mlx4_ib_ah ah;
+	struct ib_ah_attr ah_attr;
+	u8 *slave_id;
+	int slave;
+	int port;
+
+	/* Get slave that sent this packet */
+	if (wc->src_qp < dev->dev->phys_caps.base_proxy_sqpn ||
+	    wc->src_qp >= dev->dev->phys_caps.base_proxy_sqpn + 8 * MLX4_MFUNC_MAX ||
+	    (wc->src_qp & 0x1) != ctx->port - 1 ||
+	    wc->src_qp & 0x4) {
+		mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d\n", wc->src_qp);
+		return;
+	}
+	slave = ((wc->src_qp & ~0x7) - dev->dev->phys_caps.base_proxy_sqpn) / 8;
+	if (slave != ctx->slave) {
+		mlx4_ib_warn(ctx->ib_dev, "can't multiplex bad sqp:%d: "
+			     "belongs to another slave\n", wc->src_qp);
+		return;
+	}
+
+	/* Map transaction ID */
+	ib_dma_sync_single_for_cpu(ctx->ib_dev, tun_qp->ring[wr_ix].map,
+				   sizeof (struct mlx4_tunnel_mad),
+				   DMA_FROM_DEVICE);
+	switch (tunnel->mad.mad_hdr.method) {
+	case IB_MGMT_METHOD_SET:
+	case IB_MGMT_METHOD_GET:
+	case IB_MGMT_METHOD_REPORT:
+	case IB_SA_METHOD_GET_TABLE:
+	case IB_SA_METHOD_DELETE:
+	case IB_SA_METHOD_GET_MULTI:
+	case IB_SA_METHOD_GET_TRACE_TBL:
+		slave_id = (u8 *) &tunnel->mad.mad_hdr.tid;
+		if (*slave_id) {
+			mlx4_ib_warn(ctx->ib_dev, "egress mad has non-null tid msb:%d "
+				     "class:%d slave:%d\n", *slave_id,
+				     tunnel->mad.mad_hdr.mgmt_class, slave);
+			return;
+		} else
+			*slave_id = slave;
+	default:
+		/* nothing */;
+	}
+
+	/* Class-specific handling */
+	switch (tunnel->mad.mad_hdr.mgmt_class) {
+	case IB_MGMT_CLASS_SUBN_LID_ROUTED:
+	case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
+		if (slave != mlx4_master_func_num(dev->dev) &&
+		    !mlx4_vf_smi_enabled(dev->dev, slave, ctx->port))
+			return;
+		break;
+	case IB_MGMT_CLASS_SUBN_ADM:
+		if (mlx4_ib_multiplex_sa_handler(ctx->ib_dev, ctx->port, slave,
+			      (struct ib_sa_mad *) &tunnel->mad))
+			return;
+		break;
+	case IB_MGMT_CLASS_CM:
+		if (mlx4_ib_multiplex_cm_handler(ctx->ib_dev, ctx->port, slave,
+			      (struct ib_mad *) &tunnel->mad))
+			return;
+		break;
+	case IB_MGMT_CLASS_DEVICE_MGMT:
+		if (tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_GET &&
+		    tunnel->mad.mad_hdr.method != IB_MGMT_METHOD_SET)
+			return;
+		break;
+	default:
+		/* Drop unsupported classes for slaves in tunnel mode */
+		if (slave != mlx4_master_func_num(dev->dev)) {
+			mlx4_ib_warn(ctx->ib_dev, "dropping unsupported egress mad from class:%d "
+				     "for slave:%d\n", tunnel->mad.mad_hdr.mgmt_class, slave);
+			return;
+		}
+	}
+
+	/* We are using standard ib_core services to send the mad, so generate a
+	 * stadard address handle by decoding the tunnelled mlx4_ah fields */
+	memcpy(&ah.av, &tunnel->hdr.av, sizeof (struct mlx4_av));
+	ah.ibah.device = ctx->ib_dev;
+	mlx4_ib_query_ah(&ah.ibah, &ah_attr);
+	if (ah_attr.ah_flags & IB_AH_GRH)
+		fill_in_real_sgid_index(dev, slave, ctx->port, &ah_attr);
+
+	port = mlx4_slave_convert_port(dev->dev, slave, ah_attr.port_num);
+	if (port < 0)
+		return;
+	ah_attr.port_num = port;
+	memcpy(ah_attr.dmac, tunnel->hdr.mac, 6);
+	ah_attr.vlan_id = be16_to_cpu(tunnel->hdr.vlan);
+	/* if slave have default vlan use it */
+	mlx4_get_slave_default_vlan(dev->dev, ctx->port, slave,
+				    &ah_attr.vlan_id, &ah_attr.sl);
+
+	mlx4_ib_send_to_wire(dev, slave, ctx->port,
+			     is_proxy_qp0(dev, wc->src_qp, slave) ?
+			     IB_QPT_SMI : IB_QPT_GSI,
+			     be16_to_cpu(tunnel->hdr.pkey_index),
+			     be32_to_cpu(tunnel->hdr.remote_qpn),
+			     be32_to_cpu(tunnel->hdr.qkey),
+			     &ah_attr, wc->smac, &tunnel->mad);
+}
+
+static int mlx4_ib_alloc_pv_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
+				 enum ib_qp_type qp_type, int is_tun)
+{
+	int i;
+	struct mlx4_ib_demux_pv_qp *tun_qp;
+	int rx_buf_size, tx_buf_size;
+
+	if (qp_type > IB_QPT_GSI)
+		return -EINVAL;
+
+	tun_qp = &ctx->qp[qp_type];
+
+	tun_qp->ring = kzalloc(sizeof (struct mlx4_ib_buf) * MLX4_NUM_TUNNEL_BUFS,
+			       GFP_KERNEL);
+	if (!tun_qp->ring)
+		return -ENOMEM;
+
+	tun_qp->tx_ring = kcalloc(MLX4_NUM_TUNNEL_BUFS,
+				  sizeof (struct mlx4_ib_tun_tx_buf),
+				  GFP_KERNEL);
+	if (!tun_qp->tx_ring) {
+		kfree(tun_qp->ring);
+		tun_qp->ring = NULL;
+		return -ENOMEM;
+	}
+
+	if (is_tun) {
+		rx_buf_size = sizeof (struct mlx4_tunnel_mad);
+		tx_buf_size = sizeof (struct mlx4_rcv_tunnel_mad);
+	} else {
+		rx_buf_size = sizeof (struct mlx4_mad_rcv_buf);
+		tx_buf_size = sizeof (struct mlx4_mad_snd_buf);
+	}
+
+	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+		tun_qp->ring[i].addr = kmalloc(rx_buf_size, GFP_KERNEL);
+		if (!tun_qp->ring[i].addr)
+			goto err;
+		tun_qp->ring[i].map = ib_dma_map_single(ctx->ib_dev,
+							tun_qp->ring[i].addr,
+							rx_buf_size,
+							DMA_FROM_DEVICE);
+	}
+
+	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+		tun_qp->tx_ring[i].buf.addr =
+			kmalloc(tx_buf_size, GFP_KERNEL);
+		if (!tun_qp->tx_ring[i].buf.addr)
+			goto tx_err;
+		tun_qp->tx_ring[i].buf.map =
+			ib_dma_map_single(ctx->ib_dev,
+					  tun_qp->tx_ring[i].buf.addr,
+					  tx_buf_size,
+					  DMA_TO_DEVICE);
+		tun_qp->tx_ring[i].ah = NULL;
+	}
+	spin_lock_init(&tun_qp->tx_lock);
+	tun_qp->tx_ix_head = 0;
+	tun_qp->tx_ix_tail = 0;
+	tun_qp->proxy_qpt = qp_type;
+
+	return 0;
+
+tx_err:
+	while (i > 0) {
+		--i;
+		ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map,
+				    tx_buf_size, DMA_TO_DEVICE);
+		kfree(tun_qp->tx_ring[i].buf.addr);
+	}
+	kfree(tun_qp->tx_ring);
+	tun_qp->tx_ring = NULL;
+	i = MLX4_NUM_TUNNEL_BUFS;
+err:
+	while (i > 0) {
+		--i;
+		ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map,
+				    rx_buf_size, DMA_FROM_DEVICE);
+		kfree(tun_qp->ring[i].addr);
+	}
+	kfree(tun_qp->ring);
+	tun_qp->ring = NULL;
+	return -ENOMEM;
+}
+
+static void mlx4_ib_free_pv_qp_bufs(struct mlx4_ib_demux_pv_ctx *ctx,
+				     enum ib_qp_type qp_type, int is_tun)
+{
+	int i;
+	struct mlx4_ib_demux_pv_qp *tun_qp;
+	int rx_buf_size, tx_buf_size;
+
+	if (qp_type > IB_QPT_GSI)
+		return;
+
+	tun_qp = &ctx->qp[qp_type];
+	if (is_tun) {
+		rx_buf_size = sizeof (struct mlx4_tunnel_mad);
+		tx_buf_size = sizeof (struct mlx4_rcv_tunnel_mad);
+	} else {
+		rx_buf_size = sizeof (struct mlx4_mad_rcv_buf);
+		tx_buf_size = sizeof (struct mlx4_mad_snd_buf);
+	}
+
+
+	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+		ib_dma_unmap_single(ctx->ib_dev, tun_qp->ring[i].map,
+				    rx_buf_size, DMA_FROM_DEVICE);
+		kfree(tun_qp->ring[i].addr);
+	}
+
+	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+		ib_dma_unmap_single(ctx->ib_dev, tun_qp->tx_ring[i].buf.map,
+				    tx_buf_size, DMA_TO_DEVICE);
+		kfree(tun_qp->tx_ring[i].buf.addr);
+		if (tun_qp->tx_ring[i].ah)
+			ib_destroy_ah(tun_qp->tx_ring[i].ah);
+	}
+	kfree(tun_qp->tx_ring);
+	kfree(tun_qp->ring);
+}
+
+static void mlx4_ib_tunnel_comp_worker(struct work_struct *work)
+{
+	struct mlx4_ib_demux_pv_ctx *ctx;
+	struct mlx4_ib_demux_pv_qp *tun_qp;
+	struct ib_wc wc;
+	int ret;
+	ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work);
+	ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+
+	while (ib_poll_cq(ctx->cq, 1, &wc) == 1) {
+		tun_qp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)];
+		if (wc.status == IB_WC_SUCCESS) {
+			switch (wc.opcode) {
+			case IB_WC_RECV:
+				mlx4_ib_multiplex_mad(ctx, &wc);
+				ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp,
+							     wc.wr_id &
+							     (MLX4_NUM_TUNNEL_BUFS - 1));
+				if (ret)
+					pr_err("Failed reposting tunnel "
+					       "buf:%lld\n", wc.wr_id);
+				break;
+			case IB_WC_SEND:
+				pr_debug("received tunnel send completion:"
+					 "wrid=0x%llx, status=0x%x\n",
+					 wc.wr_id, wc.status);
+				ib_destroy_ah(tun_qp->tx_ring[wc.wr_id &
+					      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+				tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+					= NULL;
+				spin_lock(&tun_qp->tx_lock);
+				tun_qp->tx_ix_tail++;
+				spin_unlock(&tun_qp->tx_lock);
+
+				break;
+			default:
+				break;
+			}
+		} else  {
+			pr_debug("mlx4_ib: completion error in tunnel: %d."
+				 " status = %d, wrid = 0x%llx\n",
+				 ctx->slave, wc.status, wc.wr_id);
+			if (!MLX4_TUN_IS_RECV(wc.wr_id)) {
+				ib_destroy_ah(tun_qp->tx_ring[wc.wr_id &
+					      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+				tun_qp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+					= NULL;
+				spin_lock(&tun_qp->tx_lock);
+				tun_qp->tx_ix_tail++;
+				spin_unlock(&tun_qp->tx_lock);
+			}
+		}
+	}
+}
+
+static void pv_qp_event_handler(struct ib_event *event, void *qp_context)
+{
+	struct mlx4_ib_demux_pv_ctx *sqp = qp_context;
+
+	/* It's worse than that! He's dead, Jim! */
+	pr_err("Fatal error (%d) on a MAD QP on port %d\n",
+	       event->event, sqp->port);
+}
+
+static int create_pv_sqp(struct mlx4_ib_demux_pv_ctx *ctx,
+			    enum ib_qp_type qp_type, int create_tun)
+{
+	int i, ret;
+	struct mlx4_ib_demux_pv_qp *tun_qp;
+	struct mlx4_ib_qp_tunnel_init_attr qp_init_attr;
+	struct ib_qp_attr attr;
+	int qp_attr_mask_INIT;
+
+	if (qp_type > IB_QPT_GSI)
+		return -EINVAL;
+
+	tun_qp = &ctx->qp[qp_type];
+
+	memset(&qp_init_attr, 0, sizeof qp_init_attr);
+	qp_init_attr.init_attr.send_cq = ctx->cq;
+	qp_init_attr.init_attr.recv_cq = ctx->cq;
+	qp_init_attr.init_attr.sq_sig_type = IB_SIGNAL_ALL_WR;
+	qp_init_attr.init_attr.cap.max_send_wr = MLX4_NUM_TUNNEL_BUFS;
+	qp_init_attr.init_attr.cap.max_recv_wr = MLX4_NUM_TUNNEL_BUFS;
+	qp_init_attr.init_attr.cap.max_send_sge = 1;
+	qp_init_attr.init_attr.cap.max_recv_sge = 1;
+	if (create_tun) {
+		qp_init_attr.init_attr.qp_type = IB_QPT_UD;
+		qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_TUNNEL_QP;
+		qp_init_attr.port = ctx->port;
+		qp_init_attr.slave = ctx->slave;
+		qp_init_attr.proxy_qp_type = qp_type;
+		qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX |
+			   IB_QP_QKEY | IB_QP_PORT;
+	} else {
+		qp_init_attr.init_attr.qp_type = qp_type;
+		qp_init_attr.init_attr.create_flags = MLX4_IB_SRIOV_SQP;
+		qp_attr_mask_INIT = IB_QP_STATE | IB_QP_PKEY_INDEX | IB_QP_QKEY;
+	}
+	qp_init_attr.init_attr.port_num = ctx->port;
+	qp_init_attr.init_attr.qp_context = ctx;
+	qp_init_attr.init_attr.event_handler = pv_qp_event_handler;
+	tun_qp->qp = ib_create_qp(ctx->pd, &qp_init_attr.init_attr);
+	if (IS_ERR(tun_qp->qp)) {
+		ret = PTR_ERR(tun_qp->qp);
+		tun_qp->qp = NULL;
+		pr_err("Couldn't create %s QP (%d)\n",
+		       create_tun ? "tunnel" : "special", ret);
+		return ret;
+	}
+
+	memset(&attr, 0, sizeof attr);
+	attr.qp_state = IB_QPS_INIT;
+	ret = 0;
+	if (create_tun)
+		ret = find_slave_port_pkey_ix(to_mdev(ctx->ib_dev), ctx->slave,
+					      ctx->port, IB_DEFAULT_PKEY_FULL,
+					      &attr.pkey_index);
+	if (ret || !create_tun)
+		attr.pkey_index =
+			to_mdev(ctx->ib_dev)->pkeys.virt2phys_pkey[ctx->slave][ctx->port - 1][0];
+	attr.qkey = IB_QP1_QKEY;
+	attr.port_num = ctx->port;
+	ret = ib_modify_qp(tun_qp->qp, &attr, qp_attr_mask_INIT);
+	if (ret) {
+		pr_err("Couldn't change %s qp state to INIT (%d)\n",
+		       create_tun ? "tunnel" : "special", ret);
+		goto err_qp;
+	}
+	attr.qp_state = IB_QPS_RTR;
+	ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE);
+	if (ret) {
+		pr_err("Couldn't change %s qp state to RTR (%d)\n",
+		       create_tun ? "tunnel" : "special", ret);
+		goto err_qp;
+	}
+	attr.qp_state = IB_QPS_RTS;
+	attr.sq_psn = 0;
+	ret = ib_modify_qp(tun_qp->qp, &attr, IB_QP_STATE | IB_QP_SQ_PSN);
+	if (ret) {
+		pr_err("Couldn't change %s qp state to RTS (%d)\n",
+		       create_tun ? "tunnel" : "special", ret);
+		goto err_qp;
+	}
+
+	for (i = 0; i < MLX4_NUM_TUNNEL_BUFS; i++) {
+		ret = mlx4_ib_post_pv_qp_buf(ctx, tun_qp, i);
+		if (ret) {
+			pr_err(" mlx4_ib_post_pv_buf error"
+			       " (err = %d, i = %d)\n", ret, i);
+			goto err_qp;
+		}
+	}
+	return 0;
+
+err_qp:
+	ib_destroy_qp(tun_qp->qp);
+	tun_qp->qp = NULL;
+	return ret;
+}
+
+/*
+ * IB MAD completion callback for real SQPs
+ */
+static void mlx4_ib_sqp_comp_worker(struct work_struct *work)
+{
+	struct mlx4_ib_demux_pv_ctx *ctx;
+	struct mlx4_ib_demux_pv_qp *sqp;
+	struct ib_wc wc;
+	struct ib_grh *grh;
+	struct ib_mad *mad;
+
+	ctx = container_of(work, struct mlx4_ib_demux_pv_ctx, work);
+	ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+
+	while (mlx4_ib_poll_cq(ctx->cq, 1, &wc) == 1) {
+		sqp = &ctx->qp[MLX4_TUN_WRID_QPN(wc.wr_id)];
+		if (wc.status == IB_WC_SUCCESS) {
+			switch (wc.opcode) {
+			case IB_WC_SEND:
+				ib_destroy_ah(sqp->tx_ring[wc.wr_id &
+					      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+				sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+					= NULL;
+				spin_lock(&sqp->tx_lock);
+				sqp->tx_ix_tail++;
+				spin_unlock(&sqp->tx_lock);
+				break;
+			case IB_WC_RECV:
+				mad = (struct ib_mad *) &(((struct mlx4_mad_rcv_buf *)
+						(sqp->ring[wc.wr_id &
+						(MLX4_NUM_TUNNEL_BUFS - 1)].addr))->payload);
+				grh = &(((struct mlx4_mad_rcv_buf *)
+						(sqp->ring[wc.wr_id &
+						(MLX4_NUM_TUNNEL_BUFS - 1)].addr))->grh);
+				mlx4_ib_demux_mad(ctx->ib_dev, ctx->port, &wc, grh, mad);
+				if (mlx4_ib_post_pv_qp_buf(ctx, sqp, wc.wr_id &
+							   (MLX4_NUM_TUNNEL_BUFS - 1)))
+					pr_err("Failed reposting SQP "
+					       "buf:%lld\n", wc.wr_id);
+				break;
+			default:
+				BUG_ON(1);
+				break;
+			}
+		} else  {
+			pr_debug("mlx4_ib: completion error in tunnel: %d."
+				 " status = %d, wrid = 0x%llx\n",
+				 ctx->slave, wc.status, wc.wr_id);
+			if (!MLX4_TUN_IS_RECV(wc.wr_id)) {
+				ib_destroy_ah(sqp->tx_ring[wc.wr_id &
+					      (MLX4_NUM_TUNNEL_BUFS - 1)].ah);
+				sqp->tx_ring[wc.wr_id & (MLX4_NUM_TUNNEL_BUFS - 1)].ah
+					= NULL;
+				spin_lock(&sqp->tx_lock);
+				sqp->tx_ix_tail++;
+				spin_unlock(&sqp->tx_lock);
+			}
+		}
+	}
+}
+
+static int alloc_pv_object(struct mlx4_ib_dev *dev, int slave, int port,
+			       struct mlx4_ib_demux_pv_ctx **ret_ctx)
+{
+	struct mlx4_ib_demux_pv_ctx *ctx;
+
+	*ret_ctx = NULL;
+	ctx = kzalloc(sizeof (struct mlx4_ib_demux_pv_ctx), GFP_KERNEL);
+	if (!ctx) {
+		pr_err("failed allocating pv resource context "
+		       "for port %d, slave %d\n", port, slave);
+		return -ENOMEM;
+	}
+
+	ctx->ib_dev = &dev->ib_dev;
+	ctx->port = port;
+	ctx->slave = slave;
+	*ret_ctx = ctx;
+	return 0;
+}
+
+static void free_pv_object(struct mlx4_ib_dev *dev, int slave, int port)
+{
+	if (dev->sriov.demux[port - 1].tun[slave]) {
+		kfree(dev->sriov.demux[port - 1].tun[slave]);
+		dev->sriov.demux[port - 1].tun[slave] = NULL;
+	}
+}
+
+static int create_pv_resources(struct ib_device *ibdev, int slave, int port,
+			       int create_tun, struct mlx4_ib_demux_pv_ctx *ctx)
+{
+	int ret, cq_size;
+
+	if (ctx->state != DEMUX_PV_STATE_DOWN)
+		return -EEXIST;
+
+	ctx->state = DEMUX_PV_STATE_STARTING;
+	/* have QP0 only if link layer is IB */
+	if (rdma_port_get_link_layer(ibdev, ctx->port) ==
+	    IB_LINK_LAYER_INFINIBAND)
+		ctx->has_smi = 1;
+
+	if (ctx->has_smi) {
+		ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_SMI, create_tun);
+		if (ret) {
+			pr_err("Failed allocating qp0 tunnel bufs (%d)\n", ret);
+			goto err_out;
+		}
+	}
+
+	ret = mlx4_ib_alloc_pv_bufs(ctx, IB_QPT_GSI, create_tun);
+	if (ret) {
+		pr_err("Failed allocating qp1 tunnel bufs (%d)\n", ret);
+		goto err_out_qp0;
+	}
+
+	cq_size = 2 * MLX4_NUM_TUNNEL_BUFS;
+	if (ctx->has_smi)
+		cq_size *= 2;
+
+	ctx->cq = ib_create_cq(ctx->ib_dev, mlx4_ib_tunnel_comp_handler,
+			       NULL, ctx, cq_size, 0);
+	if (IS_ERR(ctx->cq)) {
+		ret = PTR_ERR(ctx->cq);
+		pr_err("Couldn't create tunnel CQ (%d)\n", ret);
+		goto err_buf;
+	}
+
+	ctx->pd = ib_alloc_pd(ctx->ib_dev);
+	if (IS_ERR(ctx->pd)) {
+		ret = PTR_ERR(ctx->pd);
+		pr_err("Couldn't create tunnel PD (%d)\n", ret);
+		goto err_cq;
+	}
+
+	ctx->mr = ib_get_dma_mr(ctx->pd, IB_ACCESS_LOCAL_WRITE);
+	if (IS_ERR(ctx->mr)) {
+		ret = PTR_ERR(ctx->mr);
+		pr_err("Couldn't get tunnel DMA MR (%d)\n", ret);
+		goto err_pd;
+	}
+
+	if (ctx->has_smi) {
+		ret = create_pv_sqp(ctx, IB_QPT_SMI, create_tun);
+		if (ret) {
+			pr_err("Couldn't create %s QP0 (%d)\n",
+			       create_tun ? "tunnel for" : "",  ret);
+			goto err_mr;
+		}
+	}
+
+	ret = create_pv_sqp(ctx, IB_QPT_GSI, create_tun);
+	if (ret) {
+		pr_err("Couldn't create %s QP1 (%d)\n",
+		       create_tun ? "tunnel for" : "",  ret);
+		goto err_qp0;
+	}
+
+	if (create_tun)
+		INIT_WORK(&ctx->work, mlx4_ib_tunnel_comp_worker);
+	else
+		INIT_WORK(&ctx->work, mlx4_ib_sqp_comp_worker);
+
+	ctx->wq = to_mdev(ibdev)->sriov.demux[port - 1].wq;
+
+	ret = ib_req_notify_cq(ctx->cq, IB_CQ_NEXT_COMP);
+	if (ret) {
+		pr_err("Couldn't arm tunnel cq (%d)\n", ret);
+		goto err_wq;
+	}
+	ctx->state = DEMUX_PV_STATE_ACTIVE;
+	return 0;
+
+err_wq:
+	ctx->wq = NULL;
+	ib_destroy_qp(ctx->qp[1].qp);
+	ctx->qp[1].qp = NULL;
+
+
+err_qp0:
+	if (ctx->has_smi)
+		ib_destroy_qp(ctx->qp[0].qp);
+	ctx->qp[0].qp = NULL;
+
+err_mr:
+	ib_dereg_mr(ctx->mr);
+	ctx->mr = NULL;
+
+err_pd:
+	ib_dealloc_pd(ctx->pd);
+	ctx->pd = NULL;
+
+err_cq:
+	ib_destroy_cq(ctx->cq);
+	ctx->cq = NULL;
+
+err_buf:
+	mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, create_tun);
+
+err_out_qp0:
+	if (ctx->has_smi)
+		mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, create_tun);
+err_out:
+	ctx->state = DEMUX_PV_STATE_DOWN;
+	return ret;
+}
+
+static void destroy_pv_resources(struct mlx4_ib_dev *dev, int slave, int port,
+				 struct mlx4_ib_demux_pv_ctx *ctx, int flush)
+{
+	if (!ctx)
+		return;
+	if (ctx->state > DEMUX_PV_STATE_DOWN) {
+		ctx->state = DEMUX_PV_STATE_DOWNING;
+		if (flush)
+			flush_workqueue(ctx->wq);
+		if (ctx->has_smi) {
+			ib_destroy_qp(ctx->qp[0].qp);
+			ctx->qp[0].qp = NULL;
+			mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_SMI, 1);
+		}
+		ib_destroy_qp(ctx->qp[1].qp);
+		ctx->qp[1].qp = NULL;
+		mlx4_ib_free_pv_qp_bufs(ctx, IB_QPT_GSI, 1);
+		ib_dereg_mr(ctx->mr);
+		ctx->mr = NULL;
+		ib_dealloc_pd(ctx->pd);
+		ctx->pd = NULL;
+		ib_destroy_cq(ctx->cq);
+		ctx->cq = NULL;
+		ctx->state = DEMUX_PV_STATE_DOWN;
+	}
+}
+
+static int mlx4_ib_tunnels_update(struct mlx4_ib_dev *dev, int slave,
+				  int port, int do_init)
+{
+	int ret = 0;
+
+	if (!do_init) {
+		clean_vf_mcast(&dev->sriov.demux[port - 1], slave);
+		/* for master, destroy real sqp resources */
+		if (slave == mlx4_master_func_num(dev->dev))
+			destroy_pv_resources(dev, slave, port,
+					     dev->sriov.sqps[port - 1], 1);
+		/* destroy the tunnel qp resources */
+		destroy_pv_resources(dev, slave, port,
+				     dev->sriov.demux[port - 1].tun[slave], 1);
+		return 0;
+	}
+
+	/* create the tunnel qp resources */
+	ret = create_pv_resources(&dev->ib_dev, slave, port, 1,
+				  dev->sriov.demux[port - 1].tun[slave]);
+
+	/* for master, create the real sqp resources */
+	if (!ret && slave == mlx4_master_func_num(dev->dev))
+		ret = create_pv_resources(&dev->ib_dev, slave, port, 0,
+					  dev->sriov.sqps[port - 1]);
+	return ret;
+}
+
+void mlx4_ib_tunnels_update_work(struct work_struct *work)
+{
+	struct mlx4_ib_demux_work *dmxw;
+
+	dmxw = container_of(work, struct mlx4_ib_demux_work, work);
+	mlx4_ib_tunnels_update(dmxw->dev, dmxw->slave, (int) dmxw->port,
+			       dmxw->do_init);
+	kfree(dmxw);
+	return;
+}
+
+static int mlx4_ib_alloc_demux_ctx(struct mlx4_ib_dev *dev,
+				       struct mlx4_ib_demux_ctx *ctx,
+				       int port)
+{
+	char name[12];
+	int ret = 0;
+	int i;
+
+	ctx->tun = kcalloc(dev->dev->caps.sqp_demux,
+			   sizeof (struct mlx4_ib_demux_pv_ctx *), GFP_KERNEL);
+	if (!ctx->tun)
+		return -ENOMEM;
+
+	ctx->dev = dev;
+	ctx->port = port;
+	ctx->ib_dev = &dev->ib_dev;
+
+	for (i = 0;
+	     i < min(dev->dev->caps.sqp_demux, (u16)(dev->dev->num_vfs + 1));
+	     i++) {
+		struct mlx4_active_ports actv_ports =
+			mlx4_get_active_ports(dev->dev, i);
+
+		if (!test_bit(port - 1, actv_ports.ports))
+			continue;
+
+		ret = alloc_pv_object(dev, i, port, &ctx->tun[i]);
+		if (ret) {
+			ret = -ENOMEM;
+			goto err_mcg;
+		}
+	}
+
+	ret = mlx4_ib_mcg_port_init(ctx);
+	if (ret) {
+		pr_err("Failed initializing mcg para-virt (%d)\n", ret);
+		goto err_mcg;
+	}
+
+	snprintf(name, sizeof name, "mlx4_ibt%d", port);
+	ctx->wq = create_singlethread_workqueue(name);
+	if (!ctx->wq) {
+		pr_err("Failed to create tunnelling WQ for port %d\n", port);
+		ret = -ENOMEM;
+		goto err_wq;
+	}
+
+	snprintf(name, sizeof name, "mlx4_ibud%d", port);
+	ctx->ud_wq = create_singlethread_workqueue(name);
+	if (!ctx->ud_wq) {
+		pr_err("Failed to create up/down WQ for port %d\n", port);
+		ret = -ENOMEM;
+		goto err_udwq;
+	}
+
+	return 0;
+
+err_udwq:
+	destroy_workqueue(ctx->wq);
+	ctx->wq = NULL;
+
+err_wq:
+	mlx4_ib_mcg_port_cleanup(ctx, 1);
+err_mcg:
+	for (i = 0; i < dev->dev->caps.sqp_demux; i++)
+		free_pv_object(dev, i, port);
+	kfree(ctx->tun);
+	ctx->tun = NULL;
+	return ret;
+}
+
+static void mlx4_ib_free_sqp_ctx(struct mlx4_ib_demux_pv_ctx *sqp_ctx)
+{
+	if (sqp_ctx->state > DEMUX_PV_STATE_DOWN) {
+		sqp_ctx->state = DEMUX_PV_STATE_DOWNING;
+		flush_workqueue(sqp_ctx->wq);
+		if (sqp_ctx->has_smi) {
+			ib_destroy_qp(sqp_ctx->qp[0].qp);
+			sqp_ctx->qp[0].qp = NULL;
+			mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_SMI, 0);
+		}
+		ib_destroy_qp(sqp_ctx->qp[1].qp);
+		sqp_ctx->qp[1].qp = NULL;
+		mlx4_ib_free_pv_qp_bufs(sqp_ctx, IB_QPT_GSI, 0);
+		ib_dereg_mr(sqp_ctx->mr);
+		sqp_ctx->mr = NULL;
+		ib_dealloc_pd(sqp_ctx->pd);
+		sqp_ctx->pd = NULL;
+		ib_destroy_cq(sqp_ctx->cq);
+		sqp_ctx->cq = NULL;
+		sqp_ctx->state = DEMUX_PV_STATE_DOWN;
+	}
+}
+
+static void mlx4_ib_free_demux_ctx(struct mlx4_ib_demux_ctx *ctx)
+{
+	int i;
+	if (ctx) {
+		struct mlx4_ib_dev *dev = to_mdev(ctx->ib_dev);
+		mlx4_ib_mcg_port_cleanup(ctx, 1);
+		for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+			if (!ctx->tun[i])
+				continue;
+			if (ctx->tun[i]->state > DEMUX_PV_STATE_DOWN)
+				ctx->tun[i]->state = DEMUX_PV_STATE_DOWNING;
+		}
+		flush_workqueue(ctx->wq);
+		for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+			destroy_pv_resources(dev, i, ctx->port, ctx->tun[i], 0);
+			free_pv_object(dev, i, ctx->port);
+		}
+		kfree(ctx->tun);
+		destroy_workqueue(ctx->ud_wq);
+		destroy_workqueue(ctx->wq);
+	}
+}
+
+static void mlx4_ib_master_tunnels(struct mlx4_ib_dev *dev, int do_init)
+{
+	int i;
+
+	if (!mlx4_is_master(dev->dev))
+		return;
+	/* initialize or tear down tunnel QPs for the master */
+	for (i = 0; i < dev->dev->caps.num_ports; i++)
+		mlx4_ib_tunnels_update(dev, mlx4_master_func_num(dev->dev), i + 1, do_init);
+	return;
+}
+
+int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev)
+{
+	int i = 0;
+	int err;
+
+	if (!mlx4_is_mfunc(dev->dev))
+		return 0;
+
+	dev->sriov.is_going_down = 0;
+	spin_lock_init(&dev->sriov.going_down_lock);
+	mlx4_ib_cm_paravirt_init(dev);
+
+	mlx4_ib_warn(&dev->ib_dev, "multi-function enabled\n");
+
+	if (mlx4_is_slave(dev->dev)) {
+		mlx4_ib_warn(&dev->ib_dev, "operating in qp1 tunnel mode\n");
+		return 0;
+	}
+
+	for (i = 0; i < dev->dev->caps.sqp_demux; i++) {
+		if (i == mlx4_master_func_num(dev->dev))
+			mlx4_put_slave_node_guid(dev->dev, i, dev->ib_dev.node_guid);
+		else
+			mlx4_put_slave_node_guid(dev->dev, i, mlx4_ib_gen_node_guid());
+	}
+
+	err = mlx4_ib_init_alias_guid_service(dev);
+	if (err) {
+		mlx4_ib_warn(&dev->ib_dev, "Failed init alias guid process.\n");
+		goto paravirt_err;
+	}
+	err = mlx4_ib_device_register_sysfs(dev);
+	if (err) {
+		mlx4_ib_warn(&dev->ib_dev, "Failed to register sysfs\n");
+		goto sysfs_err;
+	}
+
+	mlx4_ib_warn(&dev->ib_dev, "initializing demux service for %d qp1 clients\n",
+		     dev->dev->caps.sqp_demux);
+	for (i = 0; i < dev->num_ports; i++) {
+		union ib_gid gid;
+		err = __mlx4_ib_query_gid(&dev->ib_dev, i + 1, 0, &gid, 1);
+		if (err)
+			goto demux_err;
+		dev->sriov.demux[i].guid_cache[0] = gid.global.interface_id;
+		err = alloc_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1,
+				      &dev->sriov.sqps[i]);
+		if (err)
+			goto demux_err;
+		err = mlx4_ib_alloc_demux_ctx(dev, &dev->sriov.demux[i], i + 1);
+		if (err)
+			goto free_pv;
+	}
+	mlx4_ib_master_tunnels(dev, 1);
+	return 0;
+
+free_pv:
+	free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
+demux_err:
+	while (--i >= 0) {
+		free_pv_object(dev, mlx4_master_func_num(dev->dev), i + 1);
+		mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+	}
+	mlx4_ib_device_unregister_sysfs(dev);
+
+sysfs_err:
+	mlx4_ib_destroy_alias_guid_service(dev);
+
+paravirt_err:
+	mlx4_ib_cm_paravirt_clean(dev, -1);
+
+	return err;
+}
+
+void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev)
+{
+	int i;
+	unsigned long flags;
+
+	if (!mlx4_is_mfunc(dev->dev))
+		return;
+
+	spin_lock_irqsave(&dev->sriov.going_down_lock, flags);
+	dev->sriov.is_going_down = 1;
+	spin_unlock_irqrestore(&dev->sriov.going_down_lock, flags);
+	if (mlx4_is_master(dev->dev)) {
+		for (i = 0; i < dev->num_ports; i++) {
+			flush_workqueue(dev->sriov.demux[i].ud_wq);
+			mlx4_ib_free_sqp_ctx(dev->sriov.sqps[i]);
+			kfree(dev->sriov.sqps[i]);
+			dev->sriov.sqps[i] = NULL;
+			mlx4_ib_free_demux_ctx(&dev->sriov.demux[i]);
+		}
+
+		mlx4_ib_cm_paravirt_clean(dev, -1);
+		mlx4_ib_destroy_alias_guid_service(dev);
+		mlx4_ib_device_unregister_sysfs(dev);
+	}
+}
diff --git a/drivers/infiniband/hw/mlx4/main.c b/drivers/infiniband/hw/mlx4/main.c
index bcf50648fa1..0f7027e7db1 100644
--- a/drivers/infiniband/hw/mlx4/main.c
+++ b/drivers/infiniband/hw/mlx4/main.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2006, 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -32,30 +33,57 @@
 
 #include <linux/module.h>
 #include <linux/init.h>
+#include <linux/slab.h>
 #include <linux/errno.h>
+#include <linux/netdevice.h>
+#include <linux/inetdevice.h>
+#include <linux/rtnetlink.h>
+#include <linux/if_vlan.h>
+#include <net/ipv6.h>
+#include <net/addrconf.h>
 
 #include <rdma/ib_smi.h>
 #include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
 
 #include <linux/mlx4/driver.h>
 #include <linux/mlx4/cmd.h>
+#include <linux/mlx4/qp.h>
 
 #include "mlx4_ib.h"
 #include "user.h"
 
-#define DRV_NAME	"mlx4_ib"
-#define DRV_VERSION	"1.0"
-#define DRV_RELDATE	"April 4, 2008"
+#define DRV_NAME	MLX4_IB_DRV_NAME
+#define DRV_VERSION	"2.2-1"
+#define DRV_RELDATE	"Feb 2014"
+
+#define MLX4_IB_FLOW_MAX_PRIO 0xFFF
+#define MLX4_IB_FLOW_QPN_MASK 0xFFFFFF
 
 MODULE_AUTHOR("Roland Dreier");
 MODULE_DESCRIPTION("Mellanox ConnectX HCA InfiniBand driver");
 MODULE_LICENSE("Dual BSD/GPL");
 MODULE_VERSION(DRV_VERSION);
 
+int mlx4_ib_sm_guid_assign = 1;
+module_param_named(sm_guid_assign, mlx4_ib_sm_guid_assign, int, 0444);
+MODULE_PARM_DESC(sm_guid_assign, "Enable SM alias_GUID assignment if sm_guid_assign > 0 (Default: 1)");
+
 static const char mlx4_ib_version[] =
 	DRV_NAME ": Mellanox ConnectX InfiniBand driver v"
 	DRV_VERSION " (" DRV_RELDATE ")\n";
 
+struct update_gid_work {
+	struct work_struct	work;
+	union ib_gid		gids[128];
+	struct mlx4_ib_dev     *dev;
+	int			port;
+};
+
+static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init);
+
+static struct workqueue_struct *wq;
+
 static void init_query_mad(struct ib_smp *mad)
 {
 	mad->base_version  = 1;
@@ -64,6 +92,33 @@ static void init_query_mad(struct ib_smp *mad)
 	mad->method	   = IB_MGMT_METHOD_GET;
 }
 
+static union ib_gid zgid;
+
+static int check_flow_steering_support(struct mlx4_dev *dev)
+{
+	int eth_num_ports = 0;
+	int ib_num_ports = 0;
+
+	int dmfs = dev->caps.steering_mode == MLX4_STEERING_MODE_DEVICE_MANAGED;
+
+	if (dmfs) {
+		int i;
+		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_ETH)
+			eth_num_ports++;
+		mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
+			ib_num_ports++;
+		dmfs &= (!ib_num_ports ||
+			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_DMFS_IPOIB)) &&
+			(!eth_num_ports ||
+			 (dev->caps.flags2 & MLX4_DEV_CAP_FLAG2_FS_EN));
+		if (ib_num_ports && mlx4_is_mfunc(dev)) {
+			pr_warn("Device managed flow steering is unavailable for IB port in multifunction env.\n");
+			dmfs = 0;
+		}
+	}
+	return dmfs;
+}
+
 static int mlx4_ib_query_device(struct ib_device *ibdev,
 				struct ib_device_attr *props)
 {
@@ -80,7 +135,8 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 	init_query_mad(in_mad);
 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 
-	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	err = mlx4_MAD_IFC(to_mdev(ibdev), MLX4_MAD_IFC_IGNORE_KEYS,
+			   1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
 
@@ -102,40 +158,60 @@ static int mlx4_ib_query_device(struct ib_device *ibdev,
 		props->device_cap_flags |= IB_DEVICE_UD_AV_PORT_ENFORCE;
 	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_IPOIB_CSUM)
 		props->device_cap_flags |= IB_DEVICE_UD_IP_CSUM;
-	if (dev->dev->caps.max_gso_sz)
+	if (dev->dev->caps.max_gso_sz && dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_BLH)
 		props->device_cap_flags |= IB_DEVICE_UD_TSO;
+	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_RESERVED_LKEY)
+		props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
+	if ((dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_LOCAL_INV) &&
+	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_REMOTE_INV) &&
+	    (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_FAST_REG_WR))
+		props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC)
+		props->device_cap_flags |= IB_DEVICE_XRC;
+	if (dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW)
+		props->device_cap_flags |= IB_DEVICE_MEM_WINDOW;
+	if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
+		if (dev->dev->caps.bmme_flags & MLX4_BMME_FLAG_WIN_TYPE_2B)
+			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2B;
+		else
+			props->device_cap_flags |= IB_DEVICE_MEM_WINDOW_TYPE_2A;
+	if (dev->steering_support ==  MLX4_STEERING_MODE_DEVICE_MANAGED)
+		props->device_cap_flags |= IB_DEVICE_MANAGED_FLOW_STEERING;
+	}
 
 	props->vendor_id	   = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
 		0xffffff;
-	props->vendor_part_id	   = be16_to_cpup((__be16 *) (out_mad->data + 30));
+	props->vendor_part_id	   = dev->dev->pdev->device;
 	props->hw_ver		   = be32_to_cpup((__be32 *) (out_mad->data + 32));
 	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
 
 	props->max_mr_size	   = ~0ull;
 	props->page_size_cap	   = dev->dev->caps.page_size_cap;
-	props->max_qp		   = dev->dev->caps.num_qps - dev->dev->caps.reserved_qps;
-	props->max_qp_wr	   = dev->dev->caps.max_wqes;
+	props->max_qp		   = dev->dev->quotas.qp;
+	props->max_qp_wr	   = dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE;
 	props->max_sge		   = min(dev->dev->caps.max_sq_sg,
 					 dev->dev->caps.max_rq_sg);
-	props->max_cq		   = dev->dev->caps.num_cqs - dev->dev->caps.reserved_cqs;
+	props->max_cq		   = dev->dev->quotas.cq;
 	props->max_cqe		   = dev->dev->caps.max_cqes;
-	props->max_mr		   = dev->dev->caps.num_mpts - dev->dev->caps.reserved_mrws;
+	props->max_mr		   = dev->dev->quotas.mpt;
 	props->max_pd		   = dev->dev->caps.num_pds - dev->dev->caps.reserved_pds;
 	props->max_qp_rd_atom	   = dev->dev->caps.max_qp_dest_rdma;
 	props->max_qp_init_rd_atom = dev->dev->caps.max_qp_init_rdma;
 	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
-	props->max_srq		   = dev->dev->caps.num_srqs - dev->dev->caps.reserved_srqs;
+	props->max_srq		   = dev->dev->quotas.srq;
 	props->max_srq_wr	   = dev->dev->caps.max_srq_wqes - 1;
 	props->max_srq_sge	   = dev->dev->caps.max_srq_sge;
+	props->max_fast_reg_page_list_len = MLX4_MAX_FAST_REG_PAGES;
 	props->local_ca_ack_delay  = dev->dev->caps.local_ca_ack_delay;
 	props->atomic_cap	   = dev->dev->caps.flags & MLX4_DEV_CAP_FLAG_ATOMIC ?
 		IB_ATOMIC_HCA : IB_ATOMIC_NONE;
+	props->masked_atomic_cap   = props->atomic_cap;
 	props->max_pkeys	   = dev->dev->caps.pkey_table_len[1];
 	props->max_mcast_grp	   = dev->dev->caps.num_mgms + dev->dev->caps.num_amgms;
 	props->max_mcast_qp_attach = dev->dev->caps.num_qp_per_mgm;
 	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
 					   props->max_mcast_grp;
-	props->max_map_per_fmr = (1 << (32 - ilog2(dev->dev->caps.num_mpts))) - 1;
+	props->max_map_per_fmr = dev->dev->caps.max_fmr_maps;
 
 out:
 	kfree(in_mad);
@@ -144,11 +220,22 @@ out:
 	return err;
 }
 
-static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
-			      struct ib_port_attr *props)
+static enum rdma_link_layer
+mlx4_ib_port_link_layer(struct ib_device *device, u8 port_num)
+{
+	struct mlx4_dev *dev = to_mdev(device)->dev;
+
+	return dev->caps.port_mask[port_num] == MLX4_PORT_TYPE_IB ?
+		IB_LINK_LAYER_INFINIBAND : IB_LINK_LAYER_ETHERNET;
+}
+
+static int ib_link_query_port(struct ib_device *ibdev, u8 port,
+			      struct ib_port_attr *props, int netw_view)
 {
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
+	int ext_active_speed;
+	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 	int err = -ENOMEM;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -156,16 +243,19 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 	if (!in_mad || !out_mad)
 		goto out;
 
-	memset(props, 0, sizeof *props);
-
 	init_query_mad(in_mad);
 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 	in_mad->attr_mod = cpu_to_be32(port);
 
-	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
+		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
+				in_mad, out_mad);
 	if (err)
 		goto out;
 
+
 	props->lid		= be16_to_cpup((__be16 *) (out_mad->data + 16));
 	props->lmc		= out_mad->data[34] & 0x7;
 	props->sm_lid		= be16_to_cpup((__be16 *) (out_mad->data + 18));
@@ -173,7 +263,10 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 	props->state		= out_mad->data[32] & 0xf;
 	props->phys_state	= out_mad->data[33] >> 4;
 	props->port_cap_flags	= be32_to_cpup((__be32 *) (out_mad->data + 20));
-	props->gid_tbl_len	= to_mdev(ibdev)->dev->caps.gid_table_len[port];
+	if (netw_view)
+		props->gid_tbl_len = out_mad->data[50];
+	else
+		props->gid_tbl_len = to_mdev(ibdev)->dev->caps.gid_table_len[port];
 	props->max_msg_sz	= to_mdev(ibdev)->dev->caps.max_msg_sz;
 	props->pkey_tbl_len	= to_mdev(ibdev)->dev->caps.pkey_table_len[port];
 	props->bad_pkey_cntr	= be16_to_cpup((__be16 *) (out_mad->data + 46));
@@ -186,19 +279,132 @@ static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
 	props->max_vl_num	= out_mad->data[37] >> 4;
 	props->init_type_reply	= out_mad->data[41] >> 4;
 
+	/* Check if extended speeds (EDR/FDR/...) are supported */
+	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
+		ext_active_speed = out_mad->data[62] >> 4;
+
+		switch (ext_active_speed) {
+		case 1:
+			props->active_speed = IB_SPEED_FDR;
+			break;
+		case 2:
+			props->active_speed = IB_SPEED_EDR;
+			break;
+		}
+	}
+
+	/* If reported active speed is QDR, check if is FDR-10 */
+	if (props->active_speed == IB_SPEED_QDR) {
+		init_query_mad(in_mad);
+		in_mad->attr_id = MLX4_ATTR_EXTENDED_PORT_INFO;
+		in_mad->attr_mod = cpu_to_be32(port);
+
+		err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port,
+				   NULL, NULL, in_mad, out_mad);
+		if (err)
+			goto out;
+
+		/* Checking LinkSpeedActive for FDR-10 */
+		if (out_mad->data[15] & 0x1)
+			props->active_speed = IB_SPEED_FDR10;
+	}
+
+	/* Avoid wrong speed value returned by FW if the IB link is down. */
+	if (props->state == IB_PORT_DOWN)
+		 props->active_speed = IB_SPEED_SDR;
+
 out:
 	kfree(in_mad);
 	kfree(out_mad);
+	return err;
+}
 
+static u8 state_to_phys_state(enum ib_port_state state)
+{
+	return state == IB_PORT_ACTIVE ? 5 : 3;
+}
+
+static int eth_link_query_port(struct ib_device *ibdev, u8 port,
+			       struct ib_port_attr *props, int netw_view)
+{
+
+	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
+	struct mlx4_ib_iboe *iboe = &mdev->iboe;
+	struct net_device *ndev;
+	enum ib_mtu tmp;
+	struct mlx4_cmd_mailbox *mailbox;
+	int err = 0;
+
+	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+
+	err = mlx4_cmd_box(mdev->dev, 0, mailbox->dma, port, 0,
+			   MLX4_CMD_QUERY_PORT, MLX4_CMD_TIME_CLASS_B,
+			   MLX4_CMD_WRAPPED);
+	if (err)
+		goto out;
+
+	props->active_width	=  (((u8 *)mailbox->buf)[5] == 0x40) ?
+						IB_WIDTH_4X : IB_WIDTH_1X;
+	props->active_speed	= IB_SPEED_QDR;
+	props->port_cap_flags	= IB_PORT_CM_SUP | IB_PORT_IP_BASED_GIDS;
+	props->gid_tbl_len	= mdev->dev->caps.gid_table_len[port];
+	props->max_msg_sz	= mdev->dev->caps.max_msg_sz;
+	props->pkey_tbl_len	= 1;
+	props->max_mtu		= IB_MTU_4096;
+	props->max_vl_num	= 2;
+	props->state		= IB_PORT_DOWN;
+	props->phys_state	= state_to_phys_state(props->state);
+	props->active_mtu	= IB_MTU_256;
+	spin_lock(&iboe->lock);
+	ndev = iboe->netdevs[port - 1];
+	if (!ndev)
+		goto out_unlock;
+
+	tmp = iboe_get_mtu(ndev->mtu);
+	props->active_mtu = tmp ? min(props->max_mtu, tmp) : IB_MTU_256;
+
+	props->state		= (netif_running(ndev) && netif_carrier_ok(ndev)) ?
+					IB_PORT_ACTIVE : IB_PORT_DOWN;
+	props->phys_state	= state_to_phys_state(props->state);
+out_unlock:
+	spin_unlock(&iboe->lock);
+out:
+	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
 	return err;
 }
 
-static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
-			     union ib_gid *gid)
+int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+			 struct ib_port_attr *props, int netw_view)
+{
+	int err;
+
+	memset(props, 0, sizeof *props);
+
+	err = mlx4_ib_port_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND ?
+		ib_link_query_port(ibdev, port, props, netw_view) :
+				eth_link_query_port(ibdev, port, props, netw_view);
+
+	return err;
+}
+
+static int mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+			      struct ib_port_attr *props)
+{
+	/* returns host view */
+	return __mlx4_ib_query_port(ibdev, port, props, 0);
+}
+
+int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+			union ib_gid *gid, int netw_view)
 {
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	int clear = 0;
+	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -209,33 +415,68 @@ static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
 	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
 	in_mad->attr_mod = cpu_to_be32(port);
 
-	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (mlx4_is_mfunc(dev->dev) && netw_view)
+		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
 
 	memcpy(gid->raw, out_mad->data + 8, 8);
 
+	if (mlx4_is_mfunc(dev->dev) && !netw_view) {
+		if (index) {
+			/* For any index > 0, return the null guid */
+			err = 0;
+			clear = 1;
+			goto out;
+		}
+	}
+
 	init_query_mad(in_mad);
 	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
 	in_mad->attr_mod = cpu_to_be32(index / 8);
 
-	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	err = mlx4_MAD_IFC(dev, mad_ifc_flags, port,
+			   NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
 
 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 
 out:
+	if (clear)
+		memset(gid->raw + 8, 0, 8);
 	kfree(in_mad);
 	kfree(out_mad);
 	return err;
 }
 
-static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
-			      u16 *pkey)
+static int iboe_query_gid(struct ib_device *ibdev, u8 port, int index,
+			  union ib_gid *gid)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+
+	*gid = dev->iboe.gid_table[port - 1][index];
+
+	return 0;
+}
+
+static int mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+			     union ib_gid *gid)
+{
+	if (rdma_port_get_link_layer(ibdev, port) == IB_LINK_LAYER_INFINIBAND)
+		return __mlx4_ib_query_gid(ibdev, port, index, gid, 0);
+	else
+		return iboe_query_gid(ibdev, port, index, gid);
+}
+
+int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			 u16 *pkey, int netw_view)
 {
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
+	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 	int err = -ENOMEM;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -247,7 +488,11 @@ static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
 	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
 	in_mad->attr_mod = cpu_to_be32(index / 32);
 
-	err = mlx4_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (mlx4_is_mfunc(to_mdev(ibdev)->dev) && netw_view)
+		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
+
+	err = mlx4_MAD_IFC(to_mdev(ibdev), mad_ifc_flags, port, NULL, NULL,
+			   in_mad, out_mad);
 	if (err)
 		goto out;
 
@@ -259,23 +504,49 @@ out:
 	return err;
 }
 
+static int mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
+{
+	return __mlx4_ib_query_pkey(ibdev, port, index, pkey, 0);
+}
+
 static int mlx4_ib_modify_device(struct ib_device *ibdev, int mask,
 				 struct ib_device_modify *props)
 {
+	struct mlx4_cmd_mailbox *mailbox;
+	unsigned long flags;
+
 	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
 		return -EOPNOTSUPP;
 
-	if (mask & IB_DEVICE_MODIFY_NODE_DESC) {
-		spin_lock(&to_mdev(ibdev)->sm_lock);
-		memcpy(ibdev->node_desc, props->node_desc, 64);
-		spin_unlock(&to_mdev(ibdev)->sm_lock);
-	}
+	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
+		return 0;
+
+	if (mlx4_is_slave(to_mdev(ibdev)->dev))
+		return -EOPNOTSUPP;
+
+	spin_lock_irqsave(&to_mdev(ibdev)->sm_lock, flags);
+	memcpy(ibdev->node_desc, props->node_desc, 64);
+	spin_unlock_irqrestore(&to_mdev(ibdev)->sm_lock, flags);
+
+	/*
+	 * If possible, pass node desc to FW, so it can generate
+	 * a 144 trap.  If cmd fails, just ignore.
+	 */
+	mailbox = mlx4_alloc_cmd_mailbox(to_mdev(ibdev)->dev);
+	if (IS_ERR(mailbox))
+		return 0;
+
+	memcpy(mailbox->buf, props->node_desc, 64);
+	mlx4_cmd(to_mdev(ibdev)->dev, mailbox->dma, 1, 0,
+		 MLX4_CMD_SET_NODE, MLX4_CMD_TIME_CLASS_A, MLX4_CMD_NATIVE);
+
+	mlx4_free_cmd_mailbox(to_mdev(ibdev)->dev, mailbox);
 
 	return 0;
 }
 
-static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
-			 u32 cap_mask)
+static int mlx4_ib_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
+			    u32 cap_mask)
 {
 	struct mlx4_cmd_mailbox *mailbox;
 	int err;
@@ -284,8 +555,6 @@ static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 	if (IS_ERR(mailbox))
 		return PTR_ERR(mailbox);
 
-	memset(mailbox->buf, 0, 256);
-
 	if (dev->dev->flags & MLX4_FLAG_OLD_PORT_CMDS) {
 		*(u8 *) mailbox->buf	     = !!reset_qkey_viols << 6;
 		((__be32 *) mailbox->buf)[2] = cpu_to_be32(cap_mask);
@@ -295,7 +564,7 @@ static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 	}
 
 	err = mlx4_cmd(dev->dev, mailbox->dma, port, 0, MLX4_CMD_SET_PORT,
-		       MLX4_CMD_TIME_CLASS_B);
+		       MLX4_CMD_TIME_CLASS_B, MLX4_CMD_WRAPPED);
 
 	mlx4_free_cmd_mailbox(dev->dev, mailbox);
 	return err;
@@ -304,11 +573,20 @@ static int mlx4_SET_PORT(struct mlx4_ib_dev *dev, u8 port, int reset_qkey_viols,
 static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 			       struct ib_port_modify *props)
 {
+	struct mlx4_ib_dev *mdev = to_mdev(ibdev);
+	u8 is_eth = mdev->dev->caps.port_type[port] == MLX4_PORT_TYPE_ETH;
 	struct ib_port_attr attr;
 	u32 cap_mask;
 	int err;
 
-	mutex_lock(&to_mdev(ibdev)->cap_mask_mutex);
+	/* return OK if this is RoCE. CM calls ib_modify_port() regardless
+	 * of whether port link layer is ETH or IB. For ETH ports, qkey
+	 * violations and port capabilities are not meaningful.
+	 */
+	if (is_eth)
+		return 0;
+
+	mutex_lock(&mdev->cap_mask_mutex);
 
 	err = mlx4_ib_query_port(ibdev, port, &attr);
 	if (err)
@@ -317,9 +595,9 @@ static int mlx4_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
 	cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 		~props->clr_port_cap_mask;
 
-	err = mlx4_SET_PORT(to_mdev(ibdev), port,
-			    !!(mask & IB_PORT_RESET_QKEY_CNTR),
-			    cap_mask);
+	err = mlx4_ib_SET_PORT(mdev, port,
+			       !!(mask & IB_PORT_RESET_QKEY_CNTR),
+			       cap_mask);
 
 out:
 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
@@ -331,12 +609,24 @@ static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 {
 	struct mlx4_ib_dev *dev = to_mdev(ibdev);
 	struct mlx4_ib_ucontext *context;
+	struct mlx4_ib_alloc_ucontext_resp_v3 resp_v3;
 	struct mlx4_ib_alloc_ucontext_resp resp;
 	int err;
 
-	resp.qp_tab_size      = dev->dev->caps.num_qps;
-	resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
-	resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
+	if (!dev->ib_active)
+		return ERR_PTR(-EAGAIN);
+
+	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION) {
+		resp_v3.qp_tab_size      = dev->dev->caps.num_qps;
+		resp_v3.bf_reg_size      = dev->dev->caps.bf_reg_size;
+		resp_v3.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
+	} else {
+		resp.dev_caps	      = dev->dev->caps.userspace_caps;
+		resp.qp_tab_size      = dev->dev->caps.num_qps;
+		resp.bf_reg_size      = dev->dev->caps.bf_reg_size;
+		resp.bf_regs_per_page = dev->dev->caps.bf_regs_per_page;
+		resp.cqe_size	      = dev->dev->caps.cqe_size;
+	}
 
 	context = kmalloc(sizeof *context, GFP_KERNEL);
 	if (!context)
@@ -351,7 +641,11 @@ static struct ib_ucontext *mlx4_ib_alloc_ucontext(struct ib_device *ibdev,
 	INIT_LIST_HEAD(&context->db_page_list);
 	mutex_init(&context->db_page_mutex);
 
-	err = ib_copy_to_udata(udata, &resp, sizeof resp);
+	if (ibdev->uverbs_abi_ver == MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION)
+		err = ib_copy_to_udata(udata, &resp_v3, sizeof(resp_v3));
+	else
+		err = ib_copy_to_udata(udata, &resp, sizeof(resp));
+
 	if (err) {
 		mlx4_uar_free(to_mdev(ibdev)->dev, &context->uar);
 		kfree(context);
@@ -386,8 +680,7 @@ static int mlx4_ib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
 				       PAGE_SIZE, vma->vm_page_prot))
 			return -EAGAIN;
 	} else if (vma->vm_pgoff == 1 && dev->dev->caps.bf_reg_size != 0) {
-		/* FIXME want pgprot_writecombine() for BlueFlame pages */
-		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
 
 		if (io_remap_pfn_range(vma, vma->vm_start,
 				       to_mucontext(context)->uar.pfn +
@@ -435,24 +728,564 @@ static int mlx4_ib_dealloc_pd(struct ib_pd *pd)
 	return 0;
 }
 
+static struct ib_xrcd *mlx4_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata)
+{
+	struct mlx4_ib_xrcd *xrcd;
+	int err;
+
+	if (!(to_mdev(ibdev)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = kmalloc(sizeof *xrcd, GFP_KERNEL);
+	if (!xrcd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx4_xrcd_alloc(to_mdev(ibdev)->dev, &xrcd->xrcdn);
+	if (err)
+		goto err1;
+
+	xrcd->pd = ib_alloc_pd(ibdev);
+	if (IS_ERR(xrcd->pd)) {
+		err = PTR_ERR(xrcd->pd);
+		goto err2;
+	}
+
+	xrcd->cq = ib_create_cq(ibdev, NULL, NULL, xrcd, 1, 0);
+	if (IS_ERR(xrcd->cq)) {
+		err = PTR_ERR(xrcd->cq);
+		goto err3;
+	}
+
+	return &xrcd->ibxrcd;
+
+err3:
+	ib_dealloc_pd(xrcd->pd);
+err2:
+	mlx4_xrcd_free(to_mdev(ibdev)->dev, xrcd->xrcdn);
+err1:
+	kfree(xrcd);
+	return ERR_PTR(err);
+}
+
+static int mlx4_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	ib_destroy_cq(to_mxrcd(xrcd)->cq);
+	ib_dealloc_pd(to_mxrcd(xrcd)->pd);
+	mlx4_xrcd_free(to_mdev(xrcd->device)->dev, to_mxrcd(xrcd)->xrcdn);
+	kfree(xrcd);
+
+	return 0;
+}
+
+static int add_gid_entry(struct ib_qp *ibqp, union ib_gid *gid)
+{
+	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
+	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
+	struct mlx4_ib_gid_entry *ge;
+
+	ge = kzalloc(sizeof *ge, GFP_KERNEL);
+	if (!ge)
+		return -ENOMEM;
+
+	ge->gid = *gid;
+	if (mlx4_ib_add_mc(mdev, mqp, gid)) {
+		ge->port = mqp->port;
+		ge->added = 1;
+	}
+
+	mutex_lock(&mqp->mutex);
+	list_add_tail(&ge->list, &mqp->gid_list);
+	mutex_unlock(&mqp->mutex);
+
+	return 0;
+}
+
+int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+		   union ib_gid *gid)
+{
+	struct net_device *ndev;
+	int ret = 0;
+
+	if (!mqp->port)
+		return 0;
+
+	spin_lock(&mdev->iboe.lock);
+	ndev = mdev->iboe.netdevs[mqp->port - 1];
+	if (ndev)
+		dev_hold(ndev);
+	spin_unlock(&mdev->iboe.lock);
+
+	if (ndev) {
+		ret = 1;
+		dev_put(ndev);
+	}
+
+	return ret;
+}
+
+struct mlx4_ib_steering {
+	struct list_head list;
+	u64 reg_id;
+	union ib_gid gid;
+};
+
+static int parse_flow_attr(struct mlx4_dev *dev,
+			   u32 qp_num,
+			   union ib_flow_spec *ib_spec,
+			   struct _rule_hw *mlx4_spec)
+{
+	enum mlx4_net_trans_rule_id type;
+
+	switch (ib_spec->type) {
+	case IB_FLOW_SPEC_ETH:
+		type = MLX4_NET_TRANS_RULE_ID_ETH;
+		memcpy(mlx4_spec->eth.dst_mac, ib_spec->eth.val.dst_mac,
+		       ETH_ALEN);
+		memcpy(mlx4_spec->eth.dst_mac_msk, ib_spec->eth.mask.dst_mac,
+		       ETH_ALEN);
+		mlx4_spec->eth.vlan_tag = ib_spec->eth.val.vlan_tag;
+		mlx4_spec->eth.vlan_tag_msk = ib_spec->eth.mask.vlan_tag;
+		break;
+	case IB_FLOW_SPEC_IB:
+		type = MLX4_NET_TRANS_RULE_ID_IB;
+		mlx4_spec->ib.l3_qpn =
+			cpu_to_be32(qp_num);
+		mlx4_spec->ib.qpn_mask =
+			cpu_to_be32(MLX4_IB_FLOW_QPN_MASK);
+		break;
+
+
+	case IB_FLOW_SPEC_IPV4:
+		type = MLX4_NET_TRANS_RULE_ID_IPV4;
+		mlx4_spec->ipv4.src_ip = ib_spec->ipv4.val.src_ip;
+		mlx4_spec->ipv4.src_ip_msk = ib_spec->ipv4.mask.src_ip;
+		mlx4_spec->ipv4.dst_ip = ib_spec->ipv4.val.dst_ip;
+		mlx4_spec->ipv4.dst_ip_msk = ib_spec->ipv4.mask.dst_ip;
+		break;
+
+	case IB_FLOW_SPEC_TCP:
+	case IB_FLOW_SPEC_UDP:
+		type = ib_spec->type == IB_FLOW_SPEC_TCP ?
+					MLX4_NET_TRANS_RULE_ID_TCP :
+					MLX4_NET_TRANS_RULE_ID_UDP;
+		mlx4_spec->tcp_udp.dst_port = ib_spec->tcp_udp.val.dst_port;
+		mlx4_spec->tcp_udp.dst_port_msk = ib_spec->tcp_udp.mask.dst_port;
+		mlx4_spec->tcp_udp.src_port = ib_spec->tcp_udp.val.src_port;
+		mlx4_spec->tcp_udp.src_port_msk = ib_spec->tcp_udp.mask.src_port;
+		break;
+
+	default:
+		return -EINVAL;
+	}
+	if (mlx4_map_sw_to_hw_steering_id(dev, type) < 0 ||
+	    mlx4_hw_rule_sz(dev, type) < 0)
+		return -EINVAL;
+	mlx4_spec->id = cpu_to_be16(mlx4_map_sw_to_hw_steering_id(dev, type));
+	mlx4_spec->size = mlx4_hw_rule_sz(dev, type) >> 2;
+	return mlx4_hw_rule_sz(dev, type);
+}
+
+struct default_rules {
+	__u32 mandatory_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u32 mandatory_not_fields[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u32 rules_create_list[IB_FLOW_SPEC_SUPPORT_LAYERS];
+	__u8  link_layer;
+};
+static const struct default_rules default_table[] = {
+	{
+		.mandatory_fields = {IB_FLOW_SPEC_IPV4},
+		.mandatory_not_fields = {IB_FLOW_SPEC_ETH},
+		.rules_create_list = {IB_FLOW_SPEC_IB},
+		.link_layer = IB_LINK_LAYER_INFINIBAND
+	}
+};
+
+static int __mlx4_ib_default_rules_match(struct ib_qp *qp,
+					 struct ib_flow_attr *flow_attr)
+{
+	int i, j, k;
+	void *ib_flow;
+	const struct default_rules *pdefault_rules = default_table;
+	u8 link_layer = rdma_port_get_link_layer(qp->device, flow_attr->port);
+
+	for (i = 0; i < sizeof(default_table)/sizeof(default_table[0]); i++,
+	     pdefault_rules++) {
+		__u32 field_types[IB_FLOW_SPEC_SUPPORT_LAYERS];
+		memset(&field_types, 0, sizeof(field_types));
+
+		if (link_layer != pdefault_rules->link_layer)
+			continue;
+
+		ib_flow = flow_attr + 1;
+		/* we assume the specs are sorted */
+		for (j = 0, k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS &&
+		     j < flow_attr->num_of_specs; k++) {
+			union ib_flow_spec *current_flow =
+				(union ib_flow_spec *)ib_flow;
+
+			/* same layer but different type */
+			if (((current_flow->type & IB_FLOW_SPEC_LAYER_MASK) ==
+			     (pdefault_rules->mandatory_fields[k] &
+			      IB_FLOW_SPEC_LAYER_MASK)) &&
+			    (current_flow->type !=
+			     pdefault_rules->mandatory_fields[k]))
+				goto out;
+
+			/* same layer, try match next one */
+			if (current_flow->type ==
+			    pdefault_rules->mandatory_fields[k]) {
+				j++;
+				ib_flow +=
+					((union ib_flow_spec *)ib_flow)->size;
+			}
+		}
+
+		ib_flow = flow_attr + 1;
+		for (j = 0; j < flow_attr->num_of_specs;
+		     j++, ib_flow += ((union ib_flow_spec *)ib_flow)->size)
+			for (k = 0; k < IB_FLOW_SPEC_SUPPORT_LAYERS; k++)
+				/* same layer and same type */
+				if (((union ib_flow_spec *)ib_flow)->type ==
+				    pdefault_rules->mandatory_not_fields[k])
+					goto out;
+
+		return i;
+	}
+out:
+	return -1;
+}
+
+static int __mlx4_ib_create_default_rules(
+		struct mlx4_ib_dev *mdev,
+		struct ib_qp *qp,
+		const struct default_rules *pdefault_rules,
+		struct _rule_hw *mlx4_spec) {
+	int size = 0;
+	int i;
+
+	for (i = 0; i < sizeof(pdefault_rules->rules_create_list)/
+			sizeof(pdefault_rules->rules_create_list[0]); i++) {
+		int ret;
+		union ib_flow_spec ib_spec;
+		switch (pdefault_rules->rules_create_list[i]) {
+		case 0:
+			/* no rule */
+			continue;
+		case IB_FLOW_SPEC_IB:
+			ib_spec.type = IB_FLOW_SPEC_IB;
+			ib_spec.size = sizeof(struct ib_flow_spec_ib);
+
+			break;
+		default:
+			/* invalid rule */
+			return -EINVAL;
+		}
+		/* We must put empty rule, qpn is being ignored */
+		ret = parse_flow_attr(mdev->dev, 0, &ib_spec,
+				      mlx4_spec);
+		if (ret < 0) {
+			pr_info("invalid parsing\n");
+			return -EINVAL;
+		}
+
+		mlx4_spec = (void *)mlx4_spec + ret;
+		size += ret;
+	}
+	return size;
+}
+
+static int __mlx4_ib_create_flow(struct ib_qp *qp, struct ib_flow_attr *flow_attr,
+			  int domain,
+			  enum mlx4_net_trans_promisc_mode flow_type,
+			  u64 *reg_id)
+{
+	int ret, i;
+	int size = 0;
+	void *ib_flow;
+	struct mlx4_ib_dev *mdev = to_mdev(qp->device);
+	struct mlx4_cmd_mailbox *mailbox;
+	struct mlx4_net_trans_rule_hw_ctrl *ctrl;
+	int default_flow;
+
+	static const u16 __mlx4_domain[] = {
+		[IB_FLOW_DOMAIN_USER] = MLX4_DOMAIN_UVERBS,
+		[IB_FLOW_DOMAIN_ETHTOOL] = MLX4_DOMAIN_ETHTOOL,
+		[IB_FLOW_DOMAIN_RFS] = MLX4_DOMAIN_RFS,
+		[IB_FLOW_DOMAIN_NIC] = MLX4_DOMAIN_NIC,
+	};
+
+	if (flow_attr->priority > MLX4_IB_FLOW_MAX_PRIO) {
+		pr_err("Invalid priority value %d\n", flow_attr->priority);
+		return -EINVAL;
+	}
+
+	if (domain >= IB_FLOW_DOMAIN_NUM) {
+		pr_err("Invalid domain value %d\n", domain);
+		return -EINVAL;
+	}
+
+	if (mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type) < 0)
+		return -EINVAL;
+
+	mailbox = mlx4_alloc_cmd_mailbox(mdev->dev);
+	if (IS_ERR(mailbox))
+		return PTR_ERR(mailbox);
+	ctrl = mailbox->buf;
+
+	ctrl->prio = cpu_to_be16(__mlx4_domain[domain] |
+				 flow_attr->priority);
+	ctrl->type = mlx4_map_sw_to_hw_steering_mode(mdev->dev, flow_type);
+	ctrl->port = flow_attr->port;
+	ctrl->qpn = cpu_to_be32(qp->qp_num);
+
+	ib_flow = flow_attr + 1;
+	size += sizeof(struct mlx4_net_trans_rule_hw_ctrl);
+	/* Add default flows */
+	default_flow = __mlx4_ib_default_rules_match(qp, flow_attr);
+	if (default_flow >= 0) {
+		ret = __mlx4_ib_create_default_rules(
+				mdev, qp, default_table + default_flow,
+				mailbox->buf + size);
+		if (ret < 0) {
+			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+			return -EINVAL;
+		}
+		size += ret;
+	}
+	for (i = 0; i < flow_attr->num_of_specs; i++) {
+		ret = parse_flow_attr(mdev->dev, qp->qp_num, ib_flow,
+				      mailbox->buf + size);
+		if (ret < 0) {
+			mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+			return -EINVAL;
+		}
+		ib_flow += ((union ib_flow_spec *) ib_flow)->size;
+		size += ret;
+	}
+
+	ret = mlx4_cmd_imm(mdev->dev, mailbox->dma, reg_id, size >> 2, 0,
+			   MLX4_QP_FLOW_STEERING_ATTACH, MLX4_CMD_TIME_CLASS_A,
+			   MLX4_CMD_NATIVE);
+	if (ret == -ENOMEM)
+		pr_err("mcg table is full. Fail to register network rule.\n");
+	else if (ret == -ENXIO)
+		pr_err("Device managed flow steering is disabled. Fail to register network rule.\n");
+	else if (ret)
+		pr_err("Invalid argumant. Fail to register network rule.\n");
+
+	mlx4_free_cmd_mailbox(mdev->dev, mailbox);
+	return ret;
+}
+
+static int __mlx4_ib_destroy_flow(struct mlx4_dev *dev, u64 reg_id)
+{
+	int err;
+	err = mlx4_cmd(dev, reg_id, 0, 0,
+		       MLX4_QP_FLOW_STEERING_DETACH, MLX4_CMD_TIME_CLASS_A,
+		       MLX4_CMD_NATIVE);
+	if (err)
+		pr_err("Fail to detach network rule. registration id = 0x%llx\n",
+		       reg_id);
+	return err;
+}
+
+static struct ib_flow *mlx4_ib_create_flow(struct ib_qp *qp,
+				    struct ib_flow_attr *flow_attr,
+				    int domain)
+{
+	int err = 0, i = 0;
+	struct mlx4_ib_flow *mflow;
+	enum mlx4_net_trans_promisc_mode type[2];
+
+	memset(type, 0, sizeof(type));
+
+	mflow = kzalloc(sizeof(*mflow), GFP_KERNEL);
+	if (!mflow) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	switch (flow_attr->type) {
+	case IB_FLOW_ATTR_NORMAL:
+		type[0] = MLX4_FS_REGULAR;
+		break;
+
+	case IB_FLOW_ATTR_ALL_DEFAULT:
+		type[0] = MLX4_FS_ALL_DEFAULT;
+		break;
+
+	case IB_FLOW_ATTR_MC_DEFAULT:
+		type[0] = MLX4_FS_MC_DEFAULT;
+		break;
+
+	case IB_FLOW_ATTR_SNIFFER:
+		type[0] = MLX4_FS_UC_SNIFFER;
+		type[1] = MLX4_FS_MC_SNIFFER;
+		break;
+
+	default:
+		err = -EINVAL;
+		goto err_free;
+	}
+
+	while (i < ARRAY_SIZE(type) && type[i]) {
+		err = __mlx4_ib_create_flow(qp, flow_attr, domain, type[i],
+					    &mflow->reg_id[i]);
+		if (err)
+			goto err_free;
+		i++;
+	}
+
+	return &mflow->ibflow;
+
+err_free:
+	kfree(mflow);
+	return ERR_PTR(err);
+}
+
+static int mlx4_ib_destroy_flow(struct ib_flow *flow_id)
+{
+	int err, ret = 0;
+	int i = 0;
+	struct mlx4_ib_dev *mdev = to_mdev(flow_id->qp->device);
+	struct mlx4_ib_flow *mflow = to_mflow(flow_id);
+
+	while (i < ARRAY_SIZE(mflow->reg_id) && mflow->reg_id[i]) {
+		err = __mlx4_ib_destroy_flow(mdev->dev, mflow->reg_id[i]);
+		if (err)
+			ret = err;
+		i++;
+	}
+
+	kfree(mflow);
+	return ret;
+}
+
 static int mlx4_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 {
-	return mlx4_multicast_attach(to_mdev(ibqp->device)->dev,
-				     &to_mqp(ibqp)->mqp, gid->raw,
-				     !!(to_mqp(ibqp)->flags &
-					MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK));
+	int err;
+	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
+	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
+	u64 reg_id;
+	struct mlx4_ib_steering *ib_steering = NULL;
+	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
+		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
+
+	if (mdev->dev->caps.steering_mode ==
+	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
+		ib_steering = kmalloc(sizeof(*ib_steering), GFP_KERNEL);
+		if (!ib_steering)
+			return -ENOMEM;
+	}
+
+	err = mlx4_multicast_attach(mdev->dev, &mqp->mqp, gid->raw, mqp->port,
+				    !!(mqp->flags &
+				       MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK),
+				    prot, &reg_id);
+	if (err)
+		goto err_malloc;
+
+	err = add_gid_entry(ibqp, gid);
+	if (err)
+		goto err_add;
+
+	if (ib_steering) {
+		memcpy(ib_steering->gid.raw, gid->raw, 16);
+		ib_steering->reg_id = reg_id;
+		mutex_lock(&mqp->mutex);
+		list_add(&ib_steering->list, &mqp->steering_rules);
+		mutex_unlock(&mqp->mutex);
+	}
+	return 0;
+
+err_add:
+	mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
+			      prot, reg_id);
+err_malloc:
+	kfree(ib_steering);
+
+	return err;
+}
+
+static struct mlx4_ib_gid_entry *find_gid_entry(struct mlx4_ib_qp *qp, u8 *raw)
+{
+	struct mlx4_ib_gid_entry *ge;
+	struct mlx4_ib_gid_entry *tmp;
+	struct mlx4_ib_gid_entry *ret = NULL;
+
+	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
+		if (!memcmp(raw, ge->gid.raw, 16)) {
+			ret = ge;
+			break;
+		}
+	}
+
+	return ret;
 }
 
 static int mlx4_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 {
-	return mlx4_multicast_detach(to_mdev(ibqp->device)->dev,
-				     &to_mqp(ibqp)->mqp, gid->raw);
+	int err;
+	struct mlx4_ib_dev *mdev = to_mdev(ibqp->device);
+	struct mlx4_ib_qp *mqp = to_mqp(ibqp);
+	struct net_device *ndev;
+	struct mlx4_ib_gid_entry *ge;
+	u64 reg_id = 0;
+	enum mlx4_protocol prot = (gid->raw[1] == 0x0e) ?
+		MLX4_PROT_IB_IPV4 : MLX4_PROT_IB_IPV6;
+
+	if (mdev->dev->caps.steering_mode ==
+	    MLX4_STEERING_MODE_DEVICE_MANAGED) {
+		struct mlx4_ib_steering *ib_steering;
+
+		mutex_lock(&mqp->mutex);
+		list_for_each_entry(ib_steering, &mqp->steering_rules, list) {
+			if (!memcmp(ib_steering->gid.raw, gid->raw, 16)) {
+				list_del(&ib_steering->list);
+				break;
+			}
+		}
+		mutex_unlock(&mqp->mutex);
+		if (&ib_steering->list == &mqp->steering_rules) {
+			pr_err("Couldn't find reg_id for mgid. Steering rule is left attached\n");
+			return -EINVAL;
+		}
+		reg_id = ib_steering->reg_id;
+		kfree(ib_steering);
+	}
+
+	err = mlx4_multicast_detach(mdev->dev, &mqp->mqp, gid->raw,
+				    prot, reg_id);
+	if (err)
+		return err;
+
+	mutex_lock(&mqp->mutex);
+	ge = find_gid_entry(mqp, gid->raw);
+	if (ge) {
+		spin_lock(&mdev->iboe.lock);
+		ndev = ge->added ? mdev->iboe.netdevs[ge->port - 1] : NULL;
+		if (ndev)
+			dev_hold(ndev);
+		spin_unlock(&mdev->iboe.lock);
+		if (ndev)
+			dev_put(ndev);
+		list_del(&ge->list);
+		kfree(ge);
+	} else
+		pr_warn("could not find mgid entry\n");
+
+	mutex_unlock(&mqp->mutex);
+
+	return 0;
 }
 
 static int init_node_data(struct mlx4_ib_dev *dev)
 {
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
+	int mad_ifc_flags = MLX4_MAD_IFC_IGNORE_KEYS;
 	int err = -ENOMEM;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
@@ -462,8 +1295,10 @@ static int init_node_data(struct mlx4_ib_dev *dev)
 
 	init_query_mad(in_mad);
 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
+	if (mlx4_is_master(dev->dev))
+		mad_ifc_flags |= MLX4_MAD_IFC_NET_VIEW;
 
-	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
 
@@ -471,7 +1306,7 @@ static int init_node_data(struct mlx4_ib_dev *dev)
 
 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 
-	err = mlx4_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	err = mlx4_MAD_IFC(dev, mad_ifc_flags, 1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
 
@@ -531,31 +1366,620 @@ static struct device_attribute *mlx4_class_attributes[] = {
 	&dev_attr_board_id
 };
 
-static void *mlx4_ib_add(struct mlx4_dev *dev)
+static void mlx4_addrconf_ifid_eui48(u8 *eui, u16 vlan_id,
+				     struct net_device *dev)
 {
-	static int mlx4_ib_version_printed;
+	memcpy(eui, dev->dev_addr, 3);
+	memcpy(eui + 5, dev->dev_addr + 3, 3);
+	if (vlan_id < 0x1000) {
+		eui[3] = vlan_id >> 8;
+		eui[4] = vlan_id & 0xff;
+	} else {
+		eui[3] = 0xff;
+		eui[4] = 0xfe;
+	}
+	eui[0] ^= 2;
+}
+
+static void update_gids_task(struct work_struct *work)
+{
+	struct update_gid_work *gw = container_of(work, struct update_gid_work, work);
+	struct mlx4_cmd_mailbox *mailbox;
+	union ib_gid *gids;
+	int err;
+	struct mlx4_dev	*dev = gw->dev->dev;
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(mailbox)) {
+		pr_warn("update gid table failed %ld\n", PTR_ERR(mailbox));
+		return;
+	}
+
+	gids = mailbox->buf;
+	memcpy(gids, gw->gids, sizeof gw->gids);
+
+	err = mlx4_cmd(dev, mailbox->dma, MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
+		       1, MLX4_CMD_SET_PORT, MLX4_CMD_TIME_CLASS_B,
+		       MLX4_CMD_WRAPPED);
+	if (err)
+		pr_warn("set port command failed\n");
+	else
+		mlx4_ib_dispatch_event(gw->dev, gw->port, IB_EVENT_GID_CHANGE);
+
+	mlx4_free_cmd_mailbox(dev, mailbox);
+	kfree(gw);
+}
+
+static void reset_gids_task(struct work_struct *work)
+{
+	struct update_gid_work *gw =
+			container_of(work, struct update_gid_work, work);
+	struct mlx4_cmd_mailbox *mailbox;
+	union ib_gid *gids;
+	int err;
+	struct mlx4_dev	*dev = gw->dev->dev;
+
+	mailbox = mlx4_alloc_cmd_mailbox(dev);
+	if (IS_ERR(mailbox)) {
+		pr_warn("reset gid table failed\n");
+		goto free;
+	}
+
+	gids = mailbox->buf;
+	memcpy(gids, gw->gids, sizeof(gw->gids));
+
+	if (mlx4_ib_port_link_layer(&gw->dev->ib_dev, gw->port) ==
+				    IB_LINK_LAYER_ETHERNET) {
+		err = mlx4_cmd(dev, mailbox->dma,
+			       MLX4_SET_PORT_GID_TABLE << 8 | gw->port,
+			       1, MLX4_CMD_SET_PORT,
+			       MLX4_CMD_TIME_CLASS_B,
+			       MLX4_CMD_WRAPPED);
+		if (err)
+			pr_warn(KERN_WARNING
+				"set port %d command failed\n", gw->port);
+	}
+
+	mlx4_free_cmd_mailbox(dev, mailbox);
+free:
+	kfree(gw);
+}
+
+static int update_gid_table(struct mlx4_ib_dev *dev, int port,
+			    union ib_gid *gid, int clear,
+			    int default_gid)
+{
+	struct update_gid_work *work;
+	int i;
+	int need_update = 0;
+	int free = -1;
+	int found = -1;
+	int max_gids;
+
+	if (default_gid) {
+		free = 0;
+	} else {
+		max_gids = dev->dev->caps.gid_table_len[port];
+		for (i = 1; i < max_gids; ++i) {
+			if (!memcmp(&dev->iboe.gid_table[port - 1][i], gid,
+				    sizeof(*gid)))
+				found = i;
+
+			if (clear) {
+				if (found >= 0) {
+					need_update = 1;
+					dev->iboe.gid_table[port - 1][found] =
+						zgid;
+					break;
+				}
+			} else {
+				if (found >= 0)
+					break;
+
+				if (free < 0 &&
+				    !memcmp(&dev->iboe.gid_table[port - 1][i],
+					    &zgid, sizeof(*gid)))
+					free = i;
+			}
+		}
+	}
+
+	if (found == -1 && !clear && free >= 0) {
+		dev->iboe.gid_table[port - 1][free] = *gid;
+		need_update = 1;
+	}
+
+	if (!need_update)
+		return 0;
+
+	work = kzalloc(sizeof(*work), GFP_ATOMIC);
+	if (!work)
+		return -ENOMEM;
+
+	memcpy(work->gids, dev->iboe.gid_table[port - 1], sizeof(work->gids));
+	INIT_WORK(&work->work, update_gids_task);
+	work->port = port;
+	work->dev = dev;
+	queue_work(wq, &work->work);
+
+	return 0;
+}
+
+static void mlx4_make_default_gid(struct  net_device *dev, union ib_gid *gid)
+{
+	gid->global.subnet_prefix = cpu_to_be64(0xfe80000000000000LL);
+	mlx4_addrconf_ifid_eui48(&gid->raw[8], 0xffff, dev);
+}
+
+
+static int reset_gid_table(struct mlx4_ib_dev *dev, u8 port)
+{
+	struct update_gid_work *work;
+
+	work = kzalloc(sizeof(*work), GFP_ATOMIC);
+	if (!work)
+		return -ENOMEM;
+
+	memset(dev->iboe.gid_table[port - 1], 0, sizeof(work->gids));
+	memset(work->gids, 0, sizeof(work->gids));
+	INIT_WORK(&work->work, reset_gids_task);
+	work->dev = dev;
+	work->port = port;
+	queue_work(wq, &work->work);
+	return 0;
+}
+
+static int mlx4_ib_addr_event(int event, struct net_device *event_netdev,
+			      struct mlx4_ib_dev *ibdev, union ib_gid *gid)
+{
+	struct mlx4_ib_iboe *iboe;
+	int port = 0;
+	struct net_device *real_dev = rdma_vlan_dev_real_dev(event_netdev) ?
+				rdma_vlan_dev_real_dev(event_netdev) :
+				event_netdev;
+	union ib_gid default_gid;
+
+	mlx4_make_default_gid(real_dev, &default_gid);
+
+	if (!memcmp(gid, &default_gid, sizeof(*gid)))
+		return 0;
+
+	if (event != NETDEV_DOWN && event != NETDEV_UP)
+		return 0;
+
+	if ((real_dev != event_netdev) &&
+	    (event == NETDEV_DOWN) &&
+	    rdma_link_local_addr((struct in6_addr *)gid))
+		return 0;
+
+	iboe = &ibdev->iboe;
+	spin_lock(&iboe->lock);
+
+	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
+		if ((netif_is_bond_master(real_dev) &&
+		     (real_dev == iboe->masters[port - 1])) ||
+		     (!netif_is_bond_master(real_dev) &&
+		     (real_dev == iboe->netdevs[port - 1])))
+			update_gid_table(ibdev, port, gid,
+					 event == NETDEV_DOWN, 0);
+
+	spin_unlock(&iboe->lock);
+	return 0;
+
+}
+
+static u8 mlx4_ib_get_dev_port(struct net_device *dev,
+			       struct mlx4_ib_dev *ibdev)
+{
+	u8 port = 0;
+	struct mlx4_ib_iboe *iboe;
+	struct net_device *real_dev = rdma_vlan_dev_real_dev(dev) ?
+				rdma_vlan_dev_real_dev(dev) : dev;
+
+	iboe = &ibdev->iboe;
+
+	for (port = 1; port <= ibdev->dev->caps.num_ports; ++port)
+		if ((netif_is_bond_master(real_dev) &&
+		     (real_dev == iboe->masters[port - 1])) ||
+		     (!netif_is_bond_master(real_dev) &&
+		     (real_dev == iboe->netdevs[port - 1])))
+			break;
+
+	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
+		return 0;
+	else
+		return port;
+}
+
+static int mlx4_ib_inet_event(struct notifier_block *this, unsigned long event,
+				void *ptr)
+{
+	struct mlx4_ib_dev *ibdev;
+	struct in_ifaddr *ifa = ptr;
+	union ib_gid gid;
+	struct net_device *event_netdev = ifa->ifa_dev->dev;
+
+	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
+
+	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet);
+
+	mlx4_ib_addr_event(event, event_netdev, ibdev, &gid);
+	return NOTIFY_DONE;
+}
+
+#if IS_ENABLED(CONFIG_IPV6)
+static int mlx4_ib_inet6_event(struct notifier_block *this, unsigned long event,
+				void *ptr)
+{
+	struct mlx4_ib_dev *ibdev;
+	struct inet6_ifaddr *ifa = ptr;
+	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
+	struct net_device *event_netdev = ifa->idev->dev;
+
+	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb_inet6);
+
+	mlx4_ib_addr_event(event, event_netdev, ibdev, gid);
+	return NOTIFY_DONE;
+}
+#endif
+
+#define MLX4_IB_INVALID_MAC	((u64)-1)
+static void mlx4_ib_update_qps(struct mlx4_ib_dev *ibdev,
+			       struct net_device *dev,
+			       int port)
+{
+	u64 new_smac = 0;
+	u64 release_mac = MLX4_IB_INVALID_MAC;
+	struct mlx4_ib_qp *qp;
+
+	read_lock(&dev_base_lock);
+	new_smac = mlx4_mac_to_u64(dev->dev_addr);
+	read_unlock(&dev_base_lock);
+
+	mutex_lock(&ibdev->qp1_proxy_lock[port - 1]);
+	qp = ibdev->qp1_proxy[port - 1];
+	if (qp) {
+		int new_smac_index;
+		u64 old_smac = qp->pri.smac;
+		struct mlx4_update_qp_params update_params;
+
+		if (new_smac == old_smac)
+			goto unlock;
+
+		new_smac_index = mlx4_register_mac(ibdev->dev, port, new_smac);
+
+		if (new_smac_index < 0)
+			goto unlock;
+
+		update_params.smac_index = new_smac_index;
+		if (mlx4_update_qp(ibdev->dev, &qp->mqp, MLX4_UPDATE_QP_SMAC,
+				   &update_params)) {
+			release_mac = new_smac;
+			goto unlock;
+		}
+
+		qp->pri.smac = new_smac;
+		qp->pri.smac_index = new_smac_index;
+
+		release_mac = old_smac;
+	}
+
+unlock:
+	mutex_unlock(&ibdev->qp1_proxy_lock[port - 1]);
+	if (release_mac != MLX4_IB_INVALID_MAC)
+		mlx4_unregister_mac(ibdev->dev, port, release_mac);
+}
+
+static void mlx4_ib_get_dev_addr(struct net_device *dev,
+				 struct mlx4_ib_dev *ibdev, u8 port)
+{
+	struct in_device *in_dev;
+#if IS_ENABLED(CONFIG_IPV6)
+	struct inet6_dev *in6_dev;
+	union ib_gid  *pgid;
+	struct inet6_ifaddr *ifp;
+#endif
+	union ib_gid gid;
+
+
+	if ((port == 0) || (port > ibdev->dev->caps.num_ports))
+		return;
+
+	/* IPv4 gids */
+	in_dev = in_dev_get(dev);
+	if (in_dev) {
+		for_ifa(in_dev) {
+			/*ifa->ifa_address;*/
+			ipv6_addr_set_v4mapped(ifa->ifa_address,
+					       (struct in6_addr *)&gid);
+			update_gid_table(ibdev, port, &gid, 0, 0);
+		}
+		endfor_ifa(in_dev);
+		in_dev_put(in_dev);
+	}
+#if IS_ENABLED(CONFIG_IPV6)
+	/* IPv6 gids */
+	in6_dev = in6_dev_get(dev);
+	if (in6_dev) {
+		read_lock_bh(&in6_dev->lock);
+		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
+			pgid = (union ib_gid *)&ifp->addr;
+			update_gid_table(ibdev, port, pgid, 0, 0);
+		}
+		read_unlock_bh(&in6_dev->lock);
+		in6_dev_put(in6_dev);
+	}
+#endif
+}
+
+static void mlx4_ib_set_default_gid(struct mlx4_ib_dev *ibdev,
+				 struct  net_device *dev, u8 port)
+{
+	union ib_gid gid;
+	mlx4_make_default_gid(dev, &gid);
+	update_gid_table(ibdev, port, &gid, 0, 1);
+}
+
+static int mlx4_ib_init_gid_table(struct mlx4_ib_dev *ibdev)
+{
+	struct	net_device *dev;
+	struct mlx4_ib_iboe *iboe = &ibdev->iboe;
+	int i;
+
+	for (i = 1; i <= ibdev->num_ports; ++i)
+		if (reset_gid_table(ibdev, i))
+			return -1;
+
+	read_lock(&dev_base_lock);
+	spin_lock(&iboe->lock);
+
+	for_each_netdev(&init_net, dev) {
+		u8 port = mlx4_ib_get_dev_port(dev, ibdev);
+		if (port)
+			mlx4_ib_get_dev_addr(dev, ibdev, port);
+	}
+
+	spin_unlock(&iboe->lock);
+	read_unlock(&dev_base_lock);
+
+	return 0;
+}
+
+static void mlx4_ib_scan_netdevs(struct mlx4_ib_dev *ibdev,
+				 struct net_device *dev,
+				 unsigned long event)
+
+{
+	struct mlx4_ib_iboe *iboe;
+	int update_qps_port = -1;
+	int port;
+
+	iboe = &ibdev->iboe;
+
+	spin_lock(&iboe->lock);
+	mlx4_foreach_ib_transport_port(port, ibdev->dev) {
+		enum ib_port_state	port_state = IB_PORT_NOP;
+		struct net_device *old_master = iboe->masters[port - 1];
+		struct net_device *curr_netdev;
+		struct net_device *curr_master;
+
+		iboe->netdevs[port - 1] =
+			mlx4_get_protocol_dev(ibdev->dev, MLX4_PROT_ETH, port);
+		if (iboe->netdevs[port - 1])
+			mlx4_ib_set_default_gid(ibdev,
+						iboe->netdevs[port - 1], port);
+		curr_netdev = iboe->netdevs[port - 1];
+
+		if (iboe->netdevs[port - 1] &&
+		    netif_is_bond_slave(iboe->netdevs[port - 1])) {
+			iboe->masters[port - 1] = netdev_master_upper_dev_get(
+				iboe->netdevs[port - 1]);
+		} else {
+			iboe->masters[port - 1] = NULL;
+		}
+		curr_master = iboe->masters[port - 1];
+
+		if (dev == iboe->netdevs[port - 1] &&
+		    (event == NETDEV_CHANGEADDR || event == NETDEV_REGISTER ||
+		     event == NETDEV_UP || event == NETDEV_CHANGE))
+			update_qps_port = port;
+
+		if (curr_netdev) {
+			port_state = (netif_running(curr_netdev) && netif_carrier_ok(curr_netdev)) ?
+						IB_PORT_ACTIVE : IB_PORT_DOWN;
+			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
+		} else {
+			reset_gid_table(ibdev, port);
+		}
+		/* if using bonding/team and a slave port is down, we don't the bond IP
+		 * based gids in the table since flows that select port by gid may get
+		 * the down port.
+		 */
+		if (curr_master && (port_state == IB_PORT_DOWN)) {
+			reset_gid_table(ibdev, port);
+			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
+		}
+		/* if bonding is used it is possible that we add it to masters
+		 * only after IP address is assigned to the net bonding
+		 * interface.
+		*/
+		if (curr_master && (old_master != curr_master)) {
+			reset_gid_table(ibdev, port);
+			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
+			mlx4_ib_get_dev_addr(curr_master, ibdev, port);
+		}
+
+		if (!curr_master && (old_master != curr_master)) {
+			reset_gid_table(ibdev, port);
+			mlx4_ib_set_default_gid(ibdev, curr_netdev, port);
+			mlx4_ib_get_dev_addr(curr_netdev, ibdev, port);
+		}
+	}
+
+	spin_unlock(&iboe->lock);
+
+	if (update_qps_port > 0)
+		mlx4_ib_update_qps(ibdev, dev, update_qps_port);
+}
+
+static int mlx4_ib_netdev_event(struct notifier_block *this,
+				unsigned long event, void *ptr)
+{
+	struct net_device *dev = netdev_notifier_info_to_dev(ptr);
 	struct mlx4_ib_dev *ibdev;
+
+	if (!net_eq(dev_net(dev), &init_net))
+		return NOTIFY_DONE;
+
+	ibdev = container_of(this, struct mlx4_ib_dev, iboe.nb);
+	mlx4_ib_scan_netdevs(ibdev, dev, event);
+
+	return NOTIFY_DONE;
+}
+
+static void init_pkeys(struct mlx4_ib_dev *ibdev)
+{
+	int port;
+	int slave;
 	int i;
 
+	if (mlx4_is_master(ibdev->dev)) {
+		for (slave = 0; slave <= ibdev->dev->num_vfs; ++slave) {
+			for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
+				for (i = 0;
+				     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
+				     ++i) {
+					ibdev->pkeys.virt2phys_pkey[slave][port - 1][i] =
+					/* master has the identity virt2phys pkey mapping */
+						(slave == mlx4_master_func_num(ibdev->dev) || !i) ? i :
+							ibdev->dev->phys_caps.pkey_phys_table_len[port] - 1;
+					mlx4_sync_pkey_table(ibdev->dev, slave, port, i,
+							     ibdev->pkeys.virt2phys_pkey[slave][port - 1][i]);
+				}
+			}
+		}
+		/* initialize pkey cache */
+		for (port = 1; port <= ibdev->dev->caps.num_ports; ++port) {
+			for (i = 0;
+			     i < ibdev->dev->phys_caps.pkey_phys_table_len[port];
+			     ++i)
+				ibdev->pkeys.phys_pkey_cache[port-1][i] =
+					(i) ? 0 : 0xFFFF;
+		}
+	}
+}
+
+static void mlx4_ib_alloc_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
+{
+	char name[80];
+	int eq_per_port = 0;
+	int added_eqs = 0;
+	int total_eqs = 0;
+	int i, j, eq;
+
+	/* Legacy mode or comp_pool is not large enough */
+	if (dev->caps.comp_pool == 0 ||
+	    dev->caps.num_ports > dev->caps.comp_pool)
+		return;
+
+	eq_per_port = rounddown_pow_of_two(dev->caps.comp_pool/
+					dev->caps.num_ports);
+
+	/* Init eq table */
+	added_eqs = 0;
+	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
+		added_eqs += eq_per_port;
+
+	total_eqs = dev->caps.num_comp_vectors + added_eqs;
+
+	ibdev->eq_table = kzalloc(total_eqs * sizeof(int), GFP_KERNEL);
+	if (!ibdev->eq_table)
+		return;
 
-	if (!mlx4_ib_version_printed) {
-		printk(KERN_INFO "%s", mlx4_ib_version);
-		++mlx4_ib_version_printed;
+	ibdev->eq_added = added_eqs;
+
+	eq = 0;
+	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB) {
+		for (j = 0; j < eq_per_port; j++) {
+			snprintf(name, sizeof(name), "mlx4-ib-%d-%d@%s",
+				 i, j, dev->pdev->bus->name);
+			/* Set IRQ for specific name (per ring) */
+			if (mlx4_assign_eq(dev, name, NULL,
+					   &ibdev->eq_table[eq])) {
+				/* Use legacy (same as mlx4_en driver) */
+				pr_warn("Can't allocate EQ %d; reverting to legacy\n", eq);
+				ibdev->eq_table[eq] =
+					(eq % dev->caps.num_comp_vectors);
+			}
+			eq++;
+		}
 	}
 
+	/* Fill the reset of the vector with legacy EQ */
+	for (i = 0, eq = added_eqs; i < dev->caps.num_comp_vectors; i++)
+		ibdev->eq_table[eq++] = i;
+
+	/* Advertise the new number of EQs to clients */
+	ibdev->ib_dev.num_comp_vectors = total_eqs;
+}
+
+static void mlx4_ib_free_eqs(struct mlx4_dev *dev, struct mlx4_ib_dev *ibdev)
+{
+	int i;
+
+	/* no additional eqs were added */
+	if (!ibdev->eq_table)
+		return;
+
+	/* Reset the advertised EQ number */
+	ibdev->ib_dev.num_comp_vectors = dev->caps.num_comp_vectors;
+
+	/* Free only the added eqs */
+	for (i = 0; i < ibdev->eq_added; i++) {
+		/* Don't free legacy eqs if used */
+		if (ibdev->eq_table[i] <= dev->caps.num_comp_vectors)
+			continue;
+		mlx4_release_eq(dev, ibdev->eq_table[i]);
+	}
+
+	kfree(ibdev->eq_table);
+}
+
+static void *mlx4_ib_add(struct mlx4_dev *dev)
+{
+	struct mlx4_ib_dev *ibdev;
+	int num_ports = 0;
+	int i, j;
+	int err;
+	struct mlx4_ib_iboe *iboe;
+	int ib_num_ports = 0;
+
+	pr_info_once("%s", mlx4_ib_version);
+
+	num_ports = 0;
+	mlx4_foreach_ib_transport_port(i, dev)
+		num_ports++;
+
+	/* No point in registering a device with no ports... */
+	if (num_ports == 0)
+		return NULL;
+
 	ibdev = (struct mlx4_ib_dev *) ib_alloc_device(sizeof *ibdev);
 	if (!ibdev) {
 		dev_err(&dev->pdev->dev, "Device struct alloc failed\n");
 		return NULL;
 	}
 
+	iboe = &ibdev->iboe;
+
 	if (mlx4_pd_alloc(dev, &ibdev->priv_pdn))
 		goto err_dealloc;
 
 	if (mlx4_uar_alloc(dev, &ibdev->priv_uar))
 		goto err_pd;
 
-	ibdev->uar_map = ioremap(ibdev->priv_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
+	ibdev->uar_map = ioremap((phys_addr_t) ibdev->priv_uar.pfn << PAGE_SHIFT,
+				 PAGE_SIZE);
 	if (!ibdev->uar_map)
 		goto err_uar;
 	MLX4_INIT_DOORBELL_LOCK(&ibdev->uar_lock);
@@ -565,11 +1989,17 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	strlcpy(ibdev->ib_dev.name, "mlx4_%d", IB_DEVICE_NAME_MAX);
 	ibdev->ib_dev.owner		= THIS_MODULE;
 	ibdev->ib_dev.node_type		= RDMA_NODE_IB_CA;
-	ibdev->ib_dev.phys_port_cnt	= dev->caps.num_ports;
-	ibdev->ib_dev.num_comp_vectors	= 1;
+	ibdev->ib_dev.local_dma_lkey	= dev->caps.reserved_lkey;
+	ibdev->num_ports		= num_ports;
+	ibdev->ib_dev.phys_port_cnt     = ibdev->num_ports;
+	ibdev->ib_dev.num_comp_vectors	= dev->caps.num_comp_vectors;
 	ibdev->ib_dev.dma_device	= &dev->pdev->dev;
 
-	ibdev->ib_dev.uverbs_abi_ver	= MLX4_IB_UVERBS_ABI_VERSION;
+	if (dev->caps.userspace_caps)
+		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_ABI_VERSION;
+	else
+		ibdev->ib_dev.uverbs_abi_ver = MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION;
+
 	ibdev->ib_dev.uverbs_cmd_mask	=
 		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
@@ -591,10 +2021,13 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
 		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
-		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ);
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
 
 	ibdev->ib_dev.query_device	= mlx4_ib_query_device;
 	ibdev->ib_dev.query_port	= mlx4_ib_query_port;
+	ibdev->ib_dev.get_link_layer	= mlx4_ib_port_link_layer;
 	ibdev->ib_dev.query_gid		= mlx4_ib_query_gid;
 	ibdev->ib_dev.query_pkey	= mlx4_ib_query_pkey;
 	ibdev->ib_dev.modify_device	= mlx4_ib_modify_device;
@@ -627,38 +2060,208 @@ static void *mlx4_ib_add(struct mlx4_dev *dev)
 	ibdev->ib_dev.get_dma_mr	= mlx4_ib_get_dma_mr;
 	ibdev->ib_dev.reg_user_mr	= mlx4_ib_reg_user_mr;
 	ibdev->ib_dev.dereg_mr		= mlx4_ib_dereg_mr;
+	ibdev->ib_dev.alloc_fast_reg_mr = mlx4_ib_alloc_fast_reg_mr;
+	ibdev->ib_dev.alloc_fast_reg_page_list = mlx4_ib_alloc_fast_reg_page_list;
+	ibdev->ib_dev.free_fast_reg_page_list  = mlx4_ib_free_fast_reg_page_list;
 	ibdev->ib_dev.attach_mcast	= mlx4_ib_mcg_attach;
 	ibdev->ib_dev.detach_mcast	= mlx4_ib_mcg_detach;
 	ibdev->ib_dev.process_mad	= mlx4_ib_process_mad;
 
-	ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
-	ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
-	ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
-	ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
+	if (!mlx4_is_slave(ibdev->dev)) {
+		ibdev->ib_dev.alloc_fmr		= mlx4_ib_fmr_alloc;
+		ibdev->ib_dev.map_phys_fmr	= mlx4_ib_map_phys_fmr;
+		ibdev->ib_dev.unmap_fmr		= mlx4_ib_unmap_fmr;
+		ibdev->ib_dev.dealloc_fmr	= mlx4_ib_fmr_dealloc;
+	}
+
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_MEM_WINDOW ||
+	    dev->caps.bmme_flags & MLX4_BMME_FLAG_TYPE_2_WIN) {
+		ibdev->ib_dev.alloc_mw = mlx4_ib_alloc_mw;
+		ibdev->ib_dev.bind_mw = mlx4_ib_bind_mw;
+		ibdev->ib_dev.dealloc_mw = mlx4_ib_dealloc_mw;
+
+		ibdev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_ALLOC_MW) |
+			(1ull << IB_USER_VERBS_CMD_DEALLOC_MW);
+	}
+
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC) {
+		ibdev->ib_dev.alloc_xrcd = mlx4_ib_alloc_xrcd;
+		ibdev->ib_dev.dealloc_xrcd = mlx4_ib_dealloc_xrcd;
+		ibdev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
+			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+	}
+
+	if (check_flow_steering_support(dev)) {
+		ibdev->steering_support = MLX4_STEERING_MODE_DEVICE_MANAGED;
+		ibdev->ib_dev.create_flow	= mlx4_ib_create_flow;
+		ibdev->ib_dev.destroy_flow	= mlx4_ib_destroy_flow;
+
+		ibdev->ib_dev.uverbs_ex_cmd_mask	|=
+			(1ull << IB_USER_VERBS_EX_CMD_CREATE_FLOW) |
+			(1ull << IB_USER_VERBS_EX_CMD_DESTROY_FLOW);
+	}
+
+	mlx4_ib_alloc_eqs(dev, ibdev);
+
+	spin_lock_init(&iboe->lock);
 
 	if (init_node_data(ibdev))
 		goto err_map;
 
+	for (i = 0; i < ibdev->num_ports; ++i) {
+		mutex_init(&ibdev->qp1_proxy_lock[i]);
+		if (mlx4_ib_port_link_layer(&ibdev->ib_dev, i + 1) ==
+						IB_LINK_LAYER_ETHERNET) {
+			err = mlx4_counter_alloc(ibdev->dev, &ibdev->counters[i]);
+			if (err)
+				ibdev->counters[i] = -1;
+		} else {
+			ibdev->counters[i] = -1;
+		}
+	}
+
+	mlx4_foreach_port(i, dev, MLX4_PORT_TYPE_IB)
+		ib_num_ports++;
+
 	spin_lock_init(&ibdev->sm_lock);
 	mutex_init(&ibdev->cap_mask_mutex);
 
-	if (ib_register_device(&ibdev->ib_dev))
-		goto err_map;
+	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED &&
+	    ib_num_ports) {
+		ibdev->steer_qpn_count = MLX4_IB_UC_MAX_NUM_QPS;
+		err = mlx4_qp_reserve_range(dev, ibdev->steer_qpn_count,
+					    MLX4_IB_UC_STEER_QPN_ALIGN,
+					    &ibdev->steer_qpn_base);
+		if (err)
+			goto err_counter;
+
+		ibdev->ib_uc_qpns_bitmap =
+			kmalloc(BITS_TO_LONGS(ibdev->steer_qpn_count) *
+				sizeof(long),
+				GFP_KERNEL);
+		if (!ibdev->ib_uc_qpns_bitmap) {
+			dev_err(&dev->pdev->dev, "bit map alloc failed\n");
+			goto err_steer_qp_release;
+		}
+
+		bitmap_zero(ibdev->ib_uc_qpns_bitmap, ibdev->steer_qpn_count);
+
+		err = mlx4_FLOW_STEERING_IB_UC_QP_RANGE(
+				dev, ibdev->steer_qpn_base,
+				ibdev->steer_qpn_base +
+				ibdev->steer_qpn_count - 1);
+		if (err)
+			goto err_steer_free_bitmap;
+	}
+
+	if (ib_register_device(&ibdev->ib_dev, NULL))
+		goto err_steer_free_bitmap;
 
 	if (mlx4_ib_mad_init(ibdev))
 		goto err_reg;
 
-	for (i = 0; i < ARRAY_SIZE(mlx4_class_attributes); ++i) {
+	if (mlx4_ib_init_sriov(ibdev))
+		goto err_mad;
+
+	if (dev->caps.flags & MLX4_DEV_CAP_FLAG_IBOE) {
+		if (!iboe->nb.notifier_call) {
+			iboe->nb.notifier_call = mlx4_ib_netdev_event;
+			err = register_netdevice_notifier(&iboe->nb);
+			if (err) {
+				iboe->nb.notifier_call = NULL;
+				goto err_notif;
+			}
+		}
+		if (!iboe->nb_inet.notifier_call) {
+			iboe->nb_inet.notifier_call = mlx4_ib_inet_event;
+			err = register_inetaddr_notifier(&iboe->nb_inet);
+			if (err) {
+				iboe->nb_inet.notifier_call = NULL;
+				goto err_notif;
+			}
+		}
+#if IS_ENABLED(CONFIG_IPV6)
+		if (!iboe->nb_inet6.notifier_call) {
+			iboe->nb_inet6.notifier_call = mlx4_ib_inet6_event;
+			err = register_inet6addr_notifier(&iboe->nb_inet6);
+			if (err) {
+				iboe->nb_inet6.notifier_call = NULL;
+				goto err_notif;
+			}
+		}
+#endif
+		for (i = 1 ; i <= ibdev->num_ports ; ++i)
+			reset_gid_table(ibdev, i);
+		rtnl_lock();
+		mlx4_ib_scan_netdevs(ibdev, NULL, 0);
+		rtnl_unlock();
+		mlx4_ib_init_gid_table(ibdev);
+	}
+
+	for (j = 0; j < ARRAY_SIZE(mlx4_class_attributes); ++j) {
 		if (device_create_file(&ibdev->ib_dev.dev,
-				       mlx4_class_attributes[i]))
-			goto err_reg;
+				       mlx4_class_attributes[j]))
+			goto err_notif;
 	}
 
+	ibdev->ib_active = true;
+
+	if (mlx4_is_mfunc(ibdev->dev))
+		init_pkeys(ibdev);
+
+	/* create paravirt contexts for any VFs which are active */
+	if (mlx4_is_master(ibdev->dev)) {
+		for (j = 0; j < MLX4_MFUNC_MAX; j++) {
+			if (j == mlx4_master_func_num(ibdev->dev))
+				continue;
+			if (mlx4_is_slave_active(ibdev->dev, j))
+				do_slave_init(ibdev, j, 1);
+		}
+	}
 	return ibdev;
 
+err_notif:
+	if (ibdev->iboe.nb.notifier_call) {
+		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb.notifier_call = NULL;
+	}
+	if (ibdev->iboe.nb_inet.notifier_call) {
+		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb_inet.notifier_call = NULL;
+	}
+#if IS_ENABLED(CONFIG_IPV6)
+	if (ibdev->iboe.nb_inet6.notifier_call) {
+		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb_inet6.notifier_call = NULL;
+	}
+#endif
+	flush_workqueue(wq);
+
+	mlx4_ib_close_sriov(ibdev);
+
+err_mad:
+	mlx4_ib_mad_cleanup(ibdev);
+
 err_reg:
 	ib_unregister_device(&ibdev->ib_dev);
 
+err_steer_free_bitmap:
+	kfree(ibdev->ib_uc_qpns_bitmap);
+
+err_steer_qp_release:
+	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED)
+		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
+				      ibdev->steer_qpn_count);
+err_counter:
+	for (; i; --i)
+		if (ibdev->counters[i - 1] != -1)
+			mlx4_counter_free(ibdev->dev, ibdev->counters[i - 1]);
+
 err_map:
 	iounmap(ibdev->uar_map);
 
@@ -674,64 +2277,279 @@ err_dealloc:
 	return NULL;
 }
 
+int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn)
+{
+	int offset;
+
+	WARN_ON(!dev->ib_uc_qpns_bitmap);
+
+	offset = bitmap_find_free_region(dev->ib_uc_qpns_bitmap,
+					 dev->steer_qpn_count,
+					 get_count_order(count));
+	if (offset < 0)
+		return offset;
+
+	*qpn = dev->steer_qpn_base + offset;
+	return 0;
+}
+
+void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count)
+{
+	if (!qpn ||
+	    dev->steering_support != MLX4_STEERING_MODE_DEVICE_MANAGED)
+		return;
+
+	BUG_ON(qpn < dev->steer_qpn_base);
+
+	bitmap_release_region(dev->ib_uc_qpns_bitmap,
+			      qpn - dev->steer_qpn_base,
+			      get_count_order(count));
+}
+
+int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+			 int is_attach)
+{
+	int err;
+	size_t flow_size;
+	struct ib_flow_attr *flow = NULL;
+	struct ib_flow_spec_ib *ib_spec;
+
+	if (is_attach) {
+		flow_size = sizeof(struct ib_flow_attr) +
+			    sizeof(struct ib_flow_spec_ib);
+		flow = kzalloc(flow_size, GFP_KERNEL);
+		if (!flow)
+			return -ENOMEM;
+		flow->port = mqp->port;
+		flow->num_of_specs = 1;
+		flow->size = flow_size;
+		ib_spec = (struct ib_flow_spec_ib *)(flow + 1);
+		ib_spec->type = IB_FLOW_SPEC_IB;
+		ib_spec->size = sizeof(struct ib_flow_spec_ib);
+		/* Add an empty rule for IB L2 */
+		memset(&ib_spec->mask, 0, sizeof(ib_spec->mask));
+
+		err = __mlx4_ib_create_flow(&mqp->ibqp, flow,
+					    IB_FLOW_DOMAIN_NIC,
+					    MLX4_FS_REGULAR,
+					    &mqp->reg_id);
+	} else {
+		err = __mlx4_ib_destroy_flow(mdev->dev, mqp->reg_id);
+	}
+	kfree(flow);
+	return err;
+}
+
 static void mlx4_ib_remove(struct mlx4_dev *dev, void *ibdev_ptr)
 {
 	struct mlx4_ib_dev *ibdev = ibdev_ptr;
 	int p;
 
-	for (p = 1; p <= dev->caps.num_ports; ++p)
-		mlx4_CLOSE_PORT(dev, p);
-
+	mlx4_ib_close_sriov(ibdev);
 	mlx4_ib_mad_cleanup(ibdev);
 	ib_unregister_device(&ibdev->ib_dev);
+	if (ibdev->iboe.nb.notifier_call) {
+		if (unregister_netdevice_notifier(&ibdev->iboe.nb))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb.notifier_call = NULL;
+	}
+
+	if (ibdev->steering_support == MLX4_STEERING_MODE_DEVICE_MANAGED) {
+		mlx4_qp_release_range(dev, ibdev->steer_qpn_base,
+				      ibdev->steer_qpn_count);
+		kfree(ibdev->ib_uc_qpns_bitmap);
+	}
+
+	if (ibdev->iboe.nb_inet.notifier_call) {
+		if (unregister_inetaddr_notifier(&ibdev->iboe.nb_inet))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb_inet.notifier_call = NULL;
+	}
+#if IS_ENABLED(CONFIG_IPV6)
+	if (ibdev->iboe.nb_inet6.notifier_call) {
+		if (unregister_inet6addr_notifier(&ibdev->iboe.nb_inet6))
+			pr_warn("failure unregistering notifier\n");
+		ibdev->iboe.nb_inet6.notifier_call = NULL;
+	}
+#endif
+
 	iounmap(ibdev->uar_map);
+	for (p = 0; p < ibdev->num_ports; ++p)
+		if (ibdev->counters[p] != -1)
+			mlx4_counter_free(ibdev->dev, ibdev->counters[p]);
+	mlx4_foreach_port(p, dev, MLX4_PORT_TYPE_IB)
+		mlx4_CLOSE_PORT(dev, p);
+
+	mlx4_ib_free_eqs(dev, ibdev);
+
 	mlx4_uar_free(dev, &ibdev->priv_uar);
 	mlx4_pd_free(dev, ibdev->priv_pdn);
 	ib_dealloc_device(&ibdev->ib_dev);
 }
 
+static void do_slave_init(struct mlx4_ib_dev *ibdev, int slave, int do_init)
+{
+	struct mlx4_ib_demux_work **dm = NULL;
+	struct mlx4_dev *dev = ibdev->dev;
+	int i;
+	unsigned long flags;
+	struct mlx4_active_ports actv_ports;
+	unsigned int ports;
+	unsigned int first_port;
+
+	if (!mlx4_is_master(dev))
+		return;
+
+	actv_ports = mlx4_get_active_ports(dev, slave);
+	ports = bitmap_weight(actv_ports.ports, dev->caps.num_ports);
+	first_port = find_first_bit(actv_ports.ports, dev->caps.num_ports);
+
+	dm = kcalloc(ports, sizeof(*dm), GFP_ATOMIC);
+	if (!dm) {
+		pr_err("failed to allocate memory for tunneling qp update\n");
+		goto out;
+	}
+
+	for (i = 0; i < ports; i++) {
+		dm[i] = kmalloc(sizeof (struct mlx4_ib_demux_work), GFP_ATOMIC);
+		if (!dm[i]) {
+			pr_err("failed to allocate memory for tunneling qp update work struct\n");
+			for (i = 0; i < dev->caps.num_ports; i++) {
+				if (dm[i])
+					kfree(dm[i]);
+			}
+			goto out;
+		}
+	}
+	/* initialize or tear down tunnel QPs for the slave */
+	for (i = 0; i < ports; i++) {
+		INIT_WORK(&dm[i]->work, mlx4_ib_tunnels_update_work);
+		dm[i]->port = first_port + i + 1;
+		dm[i]->slave = slave;
+		dm[i]->do_init = do_init;
+		dm[i]->dev = ibdev;
+		spin_lock_irqsave(&ibdev->sriov.going_down_lock, flags);
+		if (!ibdev->sriov.is_going_down)
+			queue_work(ibdev->sriov.demux[i].ud_wq, &dm[i]->work);
+		spin_unlock_irqrestore(&ibdev->sriov.going_down_lock, flags);
+	}
+out:
+	kfree(dm);
+	return;
+}
+
 static void mlx4_ib_event(struct mlx4_dev *dev, void *ibdev_ptr,
-			  enum mlx4_dev_event event, int port)
+			  enum mlx4_dev_event event, unsigned long param)
 {
 	struct ib_event ibev;
+	struct mlx4_ib_dev *ibdev = to_mdev((struct ib_device *) ibdev_ptr);
+	struct mlx4_eqe *eqe = NULL;
+	struct ib_event_work *ew;
+	int p = 0;
+
+	if (event == MLX4_DEV_EVENT_PORT_MGMT_CHANGE)
+		eqe = (struct mlx4_eqe *)param;
+	else
+		p = (int) param;
 
 	switch (event) {
 	case MLX4_DEV_EVENT_PORT_UP:
+		if (p > ibdev->num_ports)
+			return;
+		if (mlx4_is_master(dev) &&
+		    rdma_port_get_link_layer(&ibdev->ib_dev, p) ==
+			IB_LINK_LAYER_INFINIBAND) {
+			mlx4_ib_invalidate_all_guid_record(ibdev, p);
+		}
 		ibev.event = IB_EVENT_PORT_ACTIVE;
 		break;
 
 	case MLX4_DEV_EVENT_PORT_DOWN:
+		if (p > ibdev->num_ports)
+			return;
 		ibev.event = IB_EVENT_PORT_ERR;
 		break;
 
 	case MLX4_DEV_EVENT_CATASTROPHIC_ERROR:
+		ibdev->ib_active = false;
 		ibev.event = IB_EVENT_DEVICE_FATAL;
 		break;
 
+	case MLX4_DEV_EVENT_PORT_MGMT_CHANGE:
+		ew = kmalloc(sizeof *ew, GFP_ATOMIC);
+		if (!ew) {
+			pr_err("failed to allocate memory for events work\n");
+			break;
+		}
+
+		INIT_WORK(&ew->work, handle_port_mgmt_change_event);
+		memcpy(&ew->ib_eqe, eqe, sizeof *eqe);
+		ew->ib_dev = ibdev;
+		/* need to queue only for port owner, which uses GEN_EQE */
+		if (mlx4_is_master(dev))
+			queue_work(wq, &ew->work);
+		else
+			handle_port_mgmt_change_event(&ew->work);
+		return;
+
+	case MLX4_DEV_EVENT_SLAVE_INIT:
+		/* here, p is the slave id */
+		do_slave_init(ibdev, p, 1);
+		return;
+
+	case MLX4_DEV_EVENT_SLAVE_SHUTDOWN:
+		/* here, p is the slave id */
+		do_slave_init(ibdev, p, 0);
+		return;
+
 	default:
 		return;
 	}
 
 	ibev.device	      = ibdev_ptr;
-	ibev.element.port_num = port;
+	ibev.element.port_num = (u8) p;
 
 	ib_dispatch_event(&ibev);
 }
 
 static struct mlx4_interface mlx4_ib_interface = {
-	.add	= mlx4_ib_add,
-	.remove	= mlx4_ib_remove,
-	.event	= mlx4_ib_event
+	.add		= mlx4_ib_add,
+	.remove		= mlx4_ib_remove,
+	.event		= mlx4_ib_event,
+	.protocol	= MLX4_PROT_IB_IPV6
 };
 
 static int __init mlx4_ib_init(void)
 {
-	return mlx4_register_interface(&mlx4_ib_interface);
+	int err;
+
+	wq = create_singlethread_workqueue("mlx4_ib");
+	if (!wq)
+		return -ENOMEM;
+
+	err = mlx4_ib_mcg_init();
+	if (err)
+		goto clean_wq;
+
+	err = mlx4_register_interface(&mlx4_ib_interface);
+	if (err)
+		goto clean_mcg;
+
+	return 0;
+
+clean_mcg:
+	mlx4_ib_mcg_destroy();
+
+clean_wq:
+	destroy_workqueue(wq);
+	return err;
 }
 
 static void __exit mlx4_ib_cleanup(void)
 {
 	mlx4_unregister_interface(&mlx4_ib_interface);
+	mlx4_ib_mcg_destroy();
+	destroy_workqueue(wq);
 }
 
 module_init(mlx4_ib_init);
diff --git a/drivers/infiniband/hw/mlx4/mcg.c b/drivers/infiniband/hw/mlx4/mcg.c
new file mode 100644
index 00000000000..ed327e6c8fd
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/mcg.c
@@ -0,0 +1,1257 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_cache.h>
+#include <rdma/ib_sa.h>
+
+#include <linux/mlx4/cmd.h>
+#include <linux/rbtree.h>
+#include <linux/delay.h>
+
+#include "mlx4_ib.h"
+
+#define MAX_VFS		80
+#define MAX_PEND_REQS_PER_FUNC 4
+#define MAD_TIMEOUT_MS	2000
+
+#define mcg_warn(fmt, arg...)	pr_warn("MCG WARNING: " fmt, ##arg)
+#define mcg_error(fmt, arg...)	pr_err(fmt, ##arg)
+#define mcg_warn_group(group, format, arg...) \
+	pr_warn("%s-%d: %16s (port %d): WARNING: " format, __func__, __LINE__,\
+	(group)->name, group->demux->port, ## arg)
+
+#define mcg_error_group(group, format, arg...) \
+	pr_err("  %16s: " format, (group)->name, ## arg)
+
+
+static union ib_gid mgid0;
+
+static struct workqueue_struct *clean_wq;
+
+enum mcast_state {
+	MCAST_NOT_MEMBER = 0,
+	MCAST_MEMBER,
+};
+
+enum mcast_group_state {
+	MCAST_IDLE,
+	MCAST_JOIN_SENT,
+	MCAST_LEAVE_SENT,
+	MCAST_RESP_READY
+};
+
+struct mcast_member {
+	enum mcast_state state;
+	uint8_t			join_state;
+	int			num_pend_reqs;
+	struct list_head	pending;
+};
+
+struct ib_sa_mcmember_data {
+	union ib_gid	mgid;
+	union ib_gid	port_gid;
+	__be32		qkey;
+	__be16		mlid;
+	u8		mtusel_mtu;
+	u8		tclass;
+	__be16		pkey;
+	u8		ratesel_rate;
+	u8		lifetmsel_lifetm;
+	__be32		sl_flowlabel_hoplimit;
+	u8		scope_join_state;
+	u8		proxy_join;
+	u8		reserved[2];
+};
+
+struct mcast_group {
+	struct ib_sa_mcmember_data rec;
+	struct rb_node		node;
+	struct list_head	mgid0_list;
+	struct mlx4_ib_demux_ctx *demux;
+	struct mcast_member	func[MAX_VFS];
+	struct mutex		lock;
+	struct work_struct	work;
+	struct list_head	pending_list;
+	int			members[3];
+	enum mcast_group_state	state;
+	enum mcast_group_state	prev_state;
+	struct ib_sa_mad	response_sa_mad;
+	__be64			last_req_tid;
+
+	char			name[33]; /* MGID string */
+	struct device_attribute	dentry;
+
+	/* refcount is the reference count for the following:
+	   1. Each queued request
+	   2. Each invocation of the worker thread
+	   3. Membership of the port at the SA
+	*/
+	atomic_t		refcount;
+
+	/* delayed work to clean pending SM request */
+	struct delayed_work	timeout_work;
+	struct list_head	cleanup_list;
+};
+
+struct mcast_req {
+	int			func;
+	struct ib_sa_mad	sa_mad;
+	struct list_head	group_list;
+	struct list_head	func_list;
+	struct mcast_group	*group;
+	int			clean;
+};
+
+
+#define safe_atomic_dec(ref) \
+	do {\
+		if (atomic_dec_and_test(ref)) \
+			mcg_warn_group(group, "did not expect to reach zero\n"); \
+	} while (0)
+
+static const char *get_state_string(enum mcast_group_state state)
+{
+	switch (state) {
+	case MCAST_IDLE:
+		return "MCAST_IDLE";
+	case MCAST_JOIN_SENT:
+		return "MCAST_JOIN_SENT";
+	case MCAST_LEAVE_SENT:
+		return "MCAST_LEAVE_SENT";
+	case MCAST_RESP_READY:
+		return "MCAST_RESP_READY";
+	}
+	return "Invalid State";
+}
+
+static struct mcast_group *mcast_find(struct mlx4_ib_demux_ctx *ctx,
+				      union ib_gid *mgid)
+{
+	struct rb_node *node = ctx->mcg_table.rb_node;
+	struct mcast_group *group;
+	int ret;
+
+	while (node) {
+		group = rb_entry(node, struct mcast_group, node);
+		ret = memcmp(mgid->raw, group->rec.mgid.raw, sizeof *mgid);
+		if (!ret)
+			return group;
+
+		if (ret < 0)
+			node = node->rb_left;
+		else
+			node = node->rb_right;
+	}
+	return NULL;
+}
+
+static struct mcast_group *mcast_insert(struct mlx4_ib_demux_ctx *ctx,
+					struct mcast_group *group)
+{
+	struct rb_node **link = &ctx->mcg_table.rb_node;
+	struct rb_node *parent = NULL;
+	struct mcast_group *cur_group;
+	int ret;
+
+	while (*link) {
+		parent = *link;
+		cur_group = rb_entry(parent, struct mcast_group, node);
+
+		ret = memcmp(group->rec.mgid.raw, cur_group->rec.mgid.raw,
+			     sizeof group->rec.mgid);
+		if (ret < 0)
+			link = &(*link)->rb_left;
+		else if (ret > 0)
+			link = &(*link)->rb_right;
+		else
+			return cur_group;
+	}
+	rb_link_node(&group->node, parent, link);
+	rb_insert_color(&group->node, &ctx->mcg_table);
+	return NULL;
+}
+
+static int send_mad_to_wire(struct mlx4_ib_demux_ctx *ctx, struct ib_mad *mad)
+{
+	struct mlx4_ib_dev *dev = ctx->dev;
+	struct ib_ah_attr	ah_attr;
+
+	spin_lock(&dev->sm_lock);
+	if (!dev->sm_ah[ctx->port - 1]) {
+		/* port is not yet Active, sm_ah not ready */
+		spin_unlock(&dev->sm_lock);
+		return -EAGAIN;
+	}
+	mlx4_ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);
+	spin_unlock(&dev->sm_lock);
+	return mlx4_ib_send_to_wire(dev, mlx4_master_func_num(dev->dev),
+				    ctx->port, IB_QPT_GSI, 0, 1, IB_QP1_QKEY,
+				    &ah_attr, NULL, mad);
+}
+
+static int send_mad_to_slave(int slave, struct mlx4_ib_demux_ctx *ctx,
+			     struct ib_mad *mad)
+{
+	struct mlx4_ib_dev *dev = ctx->dev;
+	struct ib_mad_agent *agent = dev->send_agent[ctx->port - 1][1];
+	struct ib_wc wc;
+	struct ib_ah_attr ah_attr;
+
+	/* Our agent might not yet be registered when mads start to arrive */
+	if (!agent)
+		return -EAGAIN;
+
+	ib_query_ah(dev->sm_ah[ctx->port - 1], &ah_attr);
+
+	if (ib_find_cached_pkey(&dev->ib_dev, ctx->port, IB_DEFAULT_PKEY_FULL, &wc.pkey_index))
+		return -EINVAL;
+	wc.sl = 0;
+	wc.dlid_path_bits = 0;
+	wc.port_num = ctx->port;
+	wc.slid = ah_attr.dlid;  /* opensm lid */
+	wc.src_qp = 1;
+	return mlx4_ib_send_to_slave(dev, slave, ctx->port, IB_QPT_GSI, &wc, NULL, mad);
+}
+
+static int send_join_to_wire(struct mcast_group *group, struct ib_sa_mad *sa_mad)
+{
+	struct ib_sa_mad mad;
+	struct ib_sa_mcmember_data *sa_mad_data = (struct ib_sa_mcmember_data *)&mad.data;
+	int ret;
+
+	/* we rely on a mad request as arrived from a VF */
+	memcpy(&mad, sa_mad, sizeof mad);
+
+	/* fix port GID to be the real one (slave 0) */
+	sa_mad_data->port_gid.global.interface_id = group->demux->guid_cache[0];
+
+	/* assign our own TID */
+	mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
+	group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */
+
+	ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
+	/* set timeout handler */
+	if (!ret) {
+		/* calls mlx4_ib_mcg_timeout_handler */
+		queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
+				msecs_to_jiffies(MAD_TIMEOUT_MS));
+	}
+
+	return ret;
+}
+
+static int send_leave_to_wire(struct mcast_group *group, u8 join_state)
+{
+	struct ib_sa_mad mad;
+	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
+	int ret;
+
+	memset(&mad, 0, sizeof mad);
+	mad.mad_hdr.base_version = 1;
+	mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
+	mad.mad_hdr.class_version = 2;
+	mad.mad_hdr.method = IB_SA_METHOD_DELETE;
+	mad.mad_hdr.status = cpu_to_be16(0);
+	mad.mad_hdr.class_specific = cpu_to_be16(0);
+	mad.mad_hdr.tid = mlx4_ib_get_new_demux_tid(group->demux);
+	group->last_req_tid = mad.mad_hdr.tid; /* keep it for later validation */
+	mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
+	mad.mad_hdr.attr_mod = cpu_to_be32(0);
+	mad.sa_hdr.sm_key = 0x0;
+	mad.sa_hdr.attr_offset = cpu_to_be16(7);
+	mad.sa_hdr.comp_mask = IB_SA_MCMEMBER_REC_MGID |
+		IB_SA_MCMEMBER_REC_PORT_GID | IB_SA_MCMEMBER_REC_JOIN_STATE;
+
+	*sa_data = group->rec;
+	sa_data->scope_join_state = join_state;
+
+	ret = send_mad_to_wire(group->demux, (struct ib_mad *)&mad);
+	if (ret)
+		group->state = MCAST_IDLE;
+
+	/* set timeout handler */
+	if (!ret) {
+		/* calls mlx4_ib_mcg_timeout_handler */
+		queue_delayed_work(group->demux->mcg_wq, &group->timeout_work,
+				msecs_to_jiffies(MAD_TIMEOUT_MS));
+	}
+
+	return ret;
+}
+
+static int send_reply_to_slave(int slave, struct mcast_group *group,
+		struct ib_sa_mad *req_sa_mad, u16 status)
+{
+	struct ib_sa_mad mad;
+	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)&mad.data;
+	struct ib_sa_mcmember_data *req_sa_data = (struct ib_sa_mcmember_data *)&req_sa_mad->data;
+	int ret;
+
+	memset(&mad, 0, sizeof mad);
+	mad.mad_hdr.base_version = 1;
+	mad.mad_hdr.mgmt_class = IB_MGMT_CLASS_SUBN_ADM;
+	mad.mad_hdr.class_version = 2;
+	mad.mad_hdr.method = IB_MGMT_METHOD_GET_RESP;
+	mad.mad_hdr.status = cpu_to_be16(status);
+	mad.mad_hdr.class_specific = cpu_to_be16(0);
+	mad.mad_hdr.tid = req_sa_mad->mad_hdr.tid;
+	*(u8 *)&mad.mad_hdr.tid = 0; /* resetting tid to 0 */
+	mad.mad_hdr.attr_id = cpu_to_be16(IB_SA_ATTR_MC_MEMBER_REC);
+	mad.mad_hdr.attr_mod = cpu_to_be32(0);
+	mad.sa_hdr.sm_key = req_sa_mad->sa_hdr.sm_key;
+	mad.sa_hdr.attr_offset = cpu_to_be16(7);
+	mad.sa_hdr.comp_mask = 0; /* ignored on responses, see IBTA spec */
+
+	*sa_data = group->rec;
+
+	/* reconstruct VF's requested join_state and port_gid */
+	sa_data->scope_join_state &= 0xf0;
+	sa_data->scope_join_state |= (group->func[slave].join_state & 0x0f);
+	memcpy(&sa_data->port_gid, &req_sa_data->port_gid, sizeof req_sa_data->port_gid);
+
+	ret = send_mad_to_slave(slave, group->demux, (struct ib_mad *)&mad);
+	return ret;
+}
+
+static int check_selector(ib_sa_comp_mask comp_mask,
+			  ib_sa_comp_mask selector_mask,
+			  ib_sa_comp_mask value_mask,
+			  u8 src_value, u8 dst_value)
+{
+	int err;
+	u8 selector = dst_value >> 6;
+	dst_value &= 0x3f;
+	src_value &= 0x3f;
+
+	if (!(comp_mask & selector_mask) || !(comp_mask & value_mask))
+		return 0;
+
+	switch (selector) {
+	case IB_SA_GT:
+		err = (src_value <= dst_value);
+		break;
+	case IB_SA_LT:
+		err = (src_value >= dst_value);
+		break;
+	case IB_SA_EQ:
+		err = (src_value != dst_value);
+		break;
+	default:
+		err = 0;
+		break;
+	}
+
+	return err;
+}
+
+static u16 cmp_rec(struct ib_sa_mcmember_data *src,
+		   struct ib_sa_mcmember_data *dst, ib_sa_comp_mask comp_mask)
+{
+	/* src is group record, dst is request record */
+	/* MGID must already match */
+	/* Port_GID we always replace to our Port_GID, so it is a match */
+
+#define MAD_STATUS_REQ_INVALID 0x0200
+	if (comp_mask & IB_SA_MCMEMBER_REC_QKEY && src->qkey != dst->qkey)
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_MLID && src->mlid != dst->mlid)
+		return MAD_STATUS_REQ_INVALID;
+	if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_MTU_SELECTOR,
+				 IB_SA_MCMEMBER_REC_MTU,
+				 src->mtusel_mtu, dst->mtusel_mtu))
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_TRAFFIC_CLASS &&
+	    src->tclass != dst->tclass)
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_PKEY && src->pkey != dst->pkey)
+		return MAD_STATUS_REQ_INVALID;
+	if (check_selector(comp_mask, IB_SA_MCMEMBER_REC_RATE_SELECTOR,
+				 IB_SA_MCMEMBER_REC_RATE,
+				 src->ratesel_rate, dst->ratesel_rate))
+		return MAD_STATUS_REQ_INVALID;
+	if (check_selector(comp_mask,
+				 IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME_SELECTOR,
+				 IB_SA_MCMEMBER_REC_PACKET_LIFE_TIME,
+				 src->lifetmsel_lifetm, dst->lifetmsel_lifetm))
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_SL &&
+			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0xf0000000) !=
+			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0xf0000000))
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_FLOW_LABEL &&
+			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x0fffff00) !=
+			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x0fffff00))
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_HOP_LIMIT &&
+			(be32_to_cpu(src->sl_flowlabel_hoplimit) & 0x000000ff) !=
+			(be32_to_cpu(dst->sl_flowlabel_hoplimit) & 0x000000ff))
+		return MAD_STATUS_REQ_INVALID;
+	if (comp_mask & IB_SA_MCMEMBER_REC_SCOPE &&
+			(src->scope_join_state & 0xf0) !=
+			(dst->scope_join_state & 0xf0))
+		return MAD_STATUS_REQ_INVALID;
+
+	/* join_state checked separately, proxy_join ignored */
+
+	return 0;
+}
+
+/* release group, return 1 if this was last release and group is destroyed
+ * timout work is canceled sync */
+static int release_group(struct mcast_group *group, int from_timeout_handler)
+{
+	struct mlx4_ib_demux_ctx *ctx = group->demux;
+	int nzgroup;
+
+	mutex_lock(&ctx->mcg_table_lock);
+	mutex_lock(&group->lock);
+	if (atomic_dec_and_test(&group->refcount)) {
+		if (!from_timeout_handler) {
+			if (group->state != MCAST_IDLE &&
+			    !cancel_delayed_work(&group->timeout_work)) {
+				atomic_inc(&group->refcount);
+				mutex_unlock(&group->lock);
+				mutex_unlock(&ctx->mcg_table_lock);
+				return 0;
+			}
+		}
+
+		nzgroup = memcmp(&group->rec.mgid, &mgid0, sizeof mgid0);
+		if (nzgroup)
+			del_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
+		if (!list_empty(&group->pending_list))
+			mcg_warn_group(group, "releasing a group with non empty pending list\n");
+		if (nzgroup)
+			rb_erase(&group->node, &ctx->mcg_table);
+		list_del_init(&group->mgid0_list);
+		mutex_unlock(&group->lock);
+		mutex_unlock(&ctx->mcg_table_lock);
+		kfree(group);
+		return 1;
+	} else {
+		mutex_unlock(&group->lock);
+		mutex_unlock(&ctx->mcg_table_lock);
+	}
+	return 0;
+}
+
+static void adjust_membership(struct mcast_group *group, u8 join_state, int inc)
+{
+	int i;
+
+	for (i = 0; i < 3; i++, join_state >>= 1)
+		if (join_state & 0x1)
+			group->members[i] += inc;
+}
+
+static u8 get_leave_state(struct mcast_group *group)
+{
+	u8 leave_state = 0;
+	int i;
+
+	for (i = 0; i < 3; i++)
+		if (!group->members[i])
+			leave_state |= (1 << i);
+
+	return leave_state & (group->rec.scope_join_state & 7);
+}
+
+static int join_group(struct mcast_group *group, int slave, u8 join_mask)
+{
+	int ret = 0;
+	u8 join_state;
+
+	/* remove bits that slave is already member of, and adjust */
+	join_state = join_mask & (~group->func[slave].join_state);
+	adjust_membership(group, join_state, 1);
+	group->func[slave].join_state |= join_state;
+	if (group->func[slave].state != MCAST_MEMBER && join_state) {
+		group->func[slave].state = MCAST_MEMBER;
+		ret = 1;
+	}
+	return ret;
+}
+
+static int leave_group(struct mcast_group *group, int slave, u8 leave_state)
+{
+	int ret = 0;
+
+	adjust_membership(group, leave_state, -1);
+	group->func[slave].join_state &= ~leave_state;
+	if (!group->func[slave].join_state) {
+		group->func[slave].state = MCAST_NOT_MEMBER;
+		ret = 1;
+	}
+	return ret;
+}
+
+static int check_leave(struct mcast_group *group, int slave, u8 leave_mask)
+{
+	if (group->func[slave].state != MCAST_MEMBER)
+		return MAD_STATUS_REQ_INVALID;
+
+	/* make sure we're not deleting unset bits */
+	if (~group->func[slave].join_state & leave_mask)
+		return MAD_STATUS_REQ_INVALID;
+
+	if (!leave_mask)
+		return MAD_STATUS_REQ_INVALID;
+
+	return 0;
+}
+
+static void mlx4_ib_mcg_timeout_handler(struct work_struct *work)
+{
+	struct delayed_work *delay = to_delayed_work(work);
+	struct mcast_group *group;
+	struct mcast_req *req = NULL;
+
+	group = container_of(delay, typeof(*group), timeout_work);
+
+	mutex_lock(&group->lock);
+	if (group->state == MCAST_JOIN_SENT) {
+		if (!list_empty(&group->pending_list)) {
+			req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+			list_del(&req->group_list);
+			list_del(&req->func_list);
+			--group->func[req->func].num_pend_reqs;
+			mutex_unlock(&group->lock);
+			kfree(req);
+			if (memcmp(&group->rec.mgid, &mgid0, sizeof mgid0)) {
+				if (release_group(group, 1))
+					return;
+			} else {
+				kfree(group);
+				return;
+			}
+			mutex_lock(&group->lock);
+		} else
+			mcg_warn_group(group, "DRIVER BUG\n");
+	} else if (group->state == MCAST_LEAVE_SENT) {
+		if (group->rec.scope_join_state & 7)
+			group->rec.scope_join_state &= 0xf8;
+		group->state = MCAST_IDLE;
+		mutex_unlock(&group->lock);
+		if (release_group(group, 1))
+			return;
+		mutex_lock(&group->lock);
+	} else
+		mcg_warn_group(group, "invalid state %s\n", get_state_string(group->state));
+	group->state = MCAST_IDLE;
+	atomic_inc(&group->refcount);
+	if (!queue_work(group->demux->mcg_wq, &group->work))
+		safe_atomic_dec(&group->refcount);
+
+	mutex_unlock(&group->lock);
+}
+
+static int handle_leave_req(struct mcast_group *group, u8 leave_mask,
+			    struct mcast_req *req)
+{
+	u16 status;
+
+	if (req->clean)
+		leave_mask = group->func[req->func].join_state;
+
+	status = check_leave(group, req->func, leave_mask);
+	if (!status)
+		leave_group(group, req->func, leave_mask);
+
+	if (!req->clean)
+		send_reply_to_slave(req->func, group, &req->sa_mad, status);
+	--group->func[req->func].num_pend_reqs;
+	list_del(&req->group_list);
+	list_del(&req->func_list);
+	kfree(req);
+	return 1;
+}
+
+static int handle_join_req(struct mcast_group *group, u8 join_mask,
+			   struct mcast_req *req)
+{
+	u8 group_join_state = group->rec.scope_join_state & 7;
+	int ref = 0;
+	u16 status;
+	struct ib_sa_mcmember_data *sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;
+
+	if (join_mask == (group_join_state & join_mask)) {
+		/* port's membership need not change */
+		status = cmp_rec(&group->rec, sa_data, req->sa_mad.sa_hdr.comp_mask);
+		if (!status)
+			join_group(group, req->func, join_mask);
+
+		--group->func[req->func].num_pend_reqs;
+		send_reply_to_slave(req->func, group, &req->sa_mad, status);
+		list_del(&req->group_list);
+		list_del(&req->func_list);
+		kfree(req);
+		++ref;
+	} else {
+		/* port's membership needs to be updated */
+		group->prev_state = group->state;
+		if (send_join_to_wire(group, &req->sa_mad)) {
+			--group->func[req->func].num_pend_reqs;
+			list_del(&req->group_list);
+			list_del(&req->func_list);
+			kfree(req);
+			ref = 1;
+			group->state = group->prev_state;
+		} else
+			group->state = MCAST_JOIN_SENT;
+	}
+
+	return ref;
+}
+
+static void mlx4_ib_mcg_work_handler(struct work_struct *work)
+{
+	struct mcast_group *group;
+	struct mcast_req *req = NULL;
+	struct ib_sa_mcmember_data *sa_data;
+	u8 req_join_state;
+	int rc = 1; /* release_count - this is for the scheduled work */
+	u16 status;
+	u8 method;
+
+	group = container_of(work, typeof(*group), work);
+
+	mutex_lock(&group->lock);
+
+	/* First, let's see if a response from SM is waiting regarding this group.
+	 * If so, we need to update the group's REC. If this is a bad response, we
+	 * may need to send a bad response to a VF waiting for it. If VF is waiting
+	 * and this is a good response, the VF will be answered later in this func. */
+	if (group->state == MCAST_RESP_READY) {
+		/* cancels mlx4_ib_mcg_timeout_handler */
+		cancel_delayed_work(&group->timeout_work);
+		status = be16_to_cpu(group->response_sa_mad.mad_hdr.status);
+		method = group->response_sa_mad.mad_hdr.method;
+		if (group->last_req_tid != group->response_sa_mad.mad_hdr.tid) {
+			mcg_warn_group(group, "Got MAD response to existing MGID but wrong TID, dropping. Resp TID=%llx, group TID=%llx\n",
+				be64_to_cpu(group->response_sa_mad.mad_hdr.tid),
+				be64_to_cpu(group->last_req_tid));
+			group->state = group->prev_state;
+			goto process_requests;
+		}
+		if (status) {
+			if (!list_empty(&group->pending_list))
+				req = list_first_entry(&group->pending_list,
+						struct mcast_req, group_list);
+			if ((method == IB_MGMT_METHOD_GET_RESP)) {
+					if (req) {
+						send_reply_to_slave(req->func, group, &req->sa_mad, status);
+						--group->func[req->func].num_pend_reqs;
+						list_del(&req->group_list);
+						list_del(&req->func_list);
+						kfree(req);
+						++rc;
+					} else
+						mcg_warn_group(group, "no request for failed join\n");
+			} else if (method == IB_SA_METHOD_DELETE_RESP && group->demux->flushing)
+				++rc;
+		} else {
+			u8 resp_join_state;
+			u8 cur_join_state;
+
+			resp_join_state = ((struct ib_sa_mcmember_data *)
+						group->response_sa_mad.data)->scope_join_state & 7;
+			cur_join_state = group->rec.scope_join_state & 7;
+
+			if (method == IB_MGMT_METHOD_GET_RESP) {
+				/* successfull join */
+				if (!cur_join_state && resp_join_state)
+					--rc;
+			} else if (!resp_join_state)
+					++rc;
+			memcpy(&group->rec, group->response_sa_mad.data, sizeof group->rec);
+		}
+		group->state = MCAST_IDLE;
+	}
+
+process_requests:
+	/* We should now go over pending join/leave requests, as long as we are idle. */
+	while (!list_empty(&group->pending_list) && group->state == MCAST_IDLE) {
+		req = list_first_entry(&group->pending_list, struct mcast_req,
+				       group_list);
+		sa_data = (struct ib_sa_mcmember_data *)req->sa_mad.data;
+		req_join_state = sa_data->scope_join_state & 0x7;
+
+		/* For a leave request, we will immediately answer the VF, and
+		 * update our internal counters. The actual leave will be sent
+		 * to SM later, if at all needed. We dequeue the request now. */
+		if (req->sa_mad.mad_hdr.method == IB_SA_METHOD_DELETE)
+			rc += handle_leave_req(group, req_join_state, req);
+		else
+			rc += handle_join_req(group, req_join_state, req);
+	}
+
+	/* Handle leaves */
+	if (group->state == MCAST_IDLE) {
+		req_join_state = get_leave_state(group);
+		if (req_join_state) {
+			group->rec.scope_join_state &= ~req_join_state;
+			group->prev_state = group->state;
+			if (send_leave_to_wire(group, req_join_state)) {
+				group->state = group->prev_state;
+				++rc;
+			} else
+				group->state = MCAST_LEAVE_SENT;
+		}
+	}
+
+	if (!list_empty(&group->pending_list) && group->state == MCAST_IDLE)
+		goto process_requests;
+	mutex_unlock(&group->lock);
+
+	while (rc--)
+		release_group(group, 0);
+}
+
+static struct mcast_group *search_relocate_mgid0_group(struct mlx4_ib_demux_ctx *ctx,
+						       __be64 tid,
+						       union ib_gid *new_mgid)
+{
+	struct mcast_group *group = NULL, *cur_group;
+	struct mcast_req *req;
+	struct list_head *pos;
+	struct list_head *n;
+
+	mutex_lock(&ctx->mcg_table_lock);
+	list_for_each_safe(pos, n, &ctx->mcg_mgid0_list) {
+		group = list_entry(pos, struct mcast_group, mgid0_list);
+		mutex_lock(&group->lock);
+		if (group->last_req_tid == tid) {
+			if (memcmp(new_mgid, &mgid0, sizeof mgid0)) {
+				group->rec.mgid = *new_mgid;
+				sprintf(group->name, "%016llx%016llx",
+						be64_to_cpu(group->rec.mgid.global.subnet_prefix),
+						be64_to_cpu(group->rec.mgid.global.interface_id));
+				list_del_init(&group->mgid0_list);
+				cur_group = mcast_insert(ctx, group);
+				if (cur_group) {
+					/* A race between our code and SM. Silently cleaning the new one */
+					req = list_first_entry(&group->pending_list,
+							       struct mcast_req, group_list);
+					--group->func[req->func].num_pend_reqs;
+					list_del(&req->group_list);
+					list_del(&req->func_list);
+					kfree(req);
+					mutex_unlock(&group->lock);
+					mutex_unlock(&ctx->mcg_table_lock);
+					release_group(group, 0);
+					return NULL;
+				}
+
+				atomic_inc(&group->refcount);
+				add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
+				mutex_unlock(&group->lock);
+				mutex_unlock(&ctx->mcg_table_lock);
+				return group;
+			} else {
+				struct mcast_req *tmp1, *tmp2;
+
+				list_del(&group->mgid0_list);
+				if (!list_empty(&group->pending_list) && group->state != MCAST_IDLE)
+					cancel_delayed_work_sync(&group->timeout_work);
+
+				list_for_each_entry_safe(tmp1, tmp2, &group->pending_list, group_list) {
+					list_del(&tmp1->group_list);
+					kfree(tmp1);
+				}
+				mutex_unlock(&group->lock);
+				mutex_unlock(&ctx->mcg_table_lock);
+				kfree(group);
+				return NULL;
+			}
+		}
+		mutex_unlock(&group->lock);
+	}
+	mutex_unlock(&ctx->mcg_table_lock);
+
+	return NULL;
+}
+
+static ssize_t sysfs_show_group(struct device *dev,
+		struct device_attribute *attr, char *buf);
+
+static struct mcast_group *acquire_group(struct mlx4_ib_demux_ctx *ctx,
+					 union ib_gid *mgid, int create,
+					 gfp_t gfp_mask)
+{
+	struct mcast_group *group, *cur_group;
+	int is_mgid0;
+	int i;
+
+	is_mgid0 = !memcmp(&mgid0, mgid, sizeof mgid0);
+	if (!is_mgid0) {
+		group = mcast_find(ctx, mgid);
+		if (group)
+			goto found;
+	}
+
+	if (!create)
+		return ERR_PTR(-ENOENT);
+
+	group = kzalloc(sizeof *group, gfp_mask);
+	if (!group)
+		return ERR_PTR(-ENOMEM);
+
+	group->demux = ctx;
+	group->rec.mgid = *mgid;
+	INIT_LIST_HEAD(&group->pending_list);
+	INIT_LIST_HEAD(&group->mgid0_list);
+	for (i = 0; i < MAX_VFS; ++i)
+		INIT_LIST_HEAD(&group->func[i].pending);
+	INIT_WORK(&group->work, mlx4_ib_mcg_work_handler);
+	INIT_DELAYED_WORK(&group->timeout_work, mlx4_ib_mcg_timeout_handler);
+	mutex_init(&group->lock);
+	sprintf(group->name, "%016llx%016llx",
+			be64_to_cpu(group->rec.mgid.global.subnet_prefix),
+			be64_to_cpu(group->rec.mgid.global.interface_id));
+	sysfs_attr_init(&group->dentry.attr);
+	group->dentry.show = sysfs_show_group;
+	group->dentry.store = NULL;
+	group->dentry.attr.name = group->name;
+	group->dentry.attr.mode = 0400;
+	group->state = MCAST_IDLE;
+
+	if (is_mgid0) {
+		list_add(&group->mgid0_list, &ctx->mcg_mgid0_list);
+		goto found;
+	}
+
+	cur_group = mcast_insert(ctx, group);
+	if (cur_group) {
+		mcg_warn("group just showed up %s - confused\n", cur_group->name);
+		kfree(group);
+		return ERR_PTR(-EINVAL);
+	}
+
+	add_sysfs_port_mcg_attr(ctx->dev, ctx->port, &group->dentry.attr);
+
+found:
+	atomic_inc(&group->refcount);
+	return group;
+}
+
+static void queue_req(struct mcast_req *req)
+{
+	struct mcast_group *group = req->group;
+
+	atomic_inc(&group->refcount); /* for the request */
+	atomic_inc(&group->refcount); /* for scheduling the work */
+	list_add_tail(&req->group_list, &group->pending_list);
+	list_add_tail(&req->func_list, &group->func[req->func].pending);
+	/* calls mlx4_ib_mcg_work_handler */
+	if (!queue_work(group->demux->mcg_wq, &group->work))
+		safe_atomic_dec(&group->refcount);
+}
+
+int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
+			      struct ib_sa_mad *mad)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)mad->data;
+	struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+	struct mcast_group *group;
+
+	switch (mad->mad_hdr.method) {
+	case IB_MGMT_METHOD_GET_RESP:
+	case IB_SA_METHOD_DELETE_RESP:
+		mutex_lock(&ctx->mcg_table_lock);
+		group = acquire_group(ctx, &rec->mgid, 0, GFP_KERNEL);
+		mutex_unlock(&ctx->mcg_table_lock);
+		if (IS_ERR(group)) {
+			if (mad->mad_hdr.method == IB_MGMT_METHOD_GET_RESP) {
+				__be64 tid = mad->mad_hdr.tid;
+				*(u8 *)(&tid) = (u8)slave; /* in group we kept the modified TID */
+				group = search_relocate_mgid0_group(ctx, tid, &rec->mgid);
+			} else
+				group = NULL;
+		}
+
+		if (!group)
+			return 1;
+
+		mutex_lock(&group->lock);
+		group->response_sa_mad = *mad;
+		group->prev_state = group->state;
+		group->state = MCAST_RESP_READY;
+		/* calls mlx4_ib_mcg_work_handler */
+		atomic_inc(&group->refcount);
+		if (!queue_work(ctx->mcg_wq, &group->work))
+			safe_atomic_dec(&group->refcount);
+		mutex_unlock(&group->lock);
+		release_group(group, 0);
+		return 1; /* consumed */
+	case IB_MGMT_METHOD_SET:
+	case IB_SA_METHOD_GET_TABLE:
+	case IB_SA_METHOD_GET_TABLE_RESP:
+	case IB_SA_METHOD_DELETE:
+		return 0; /* not consumed, pass-through to guest over tunnel */
+	default:
+		mcg_warn("In demux, port %d: unexpected MCMember method: 0x%x, dropping\n",
+			port, mad->mad_hdr.method);
+		return 1; /* consumed */
+	}
+}
+
+int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port,
+				  int slave, struct ib_sa_mad *sa_mad)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	struct ib_sa_mcmember_data *rec = (struct ib_sa_mcmember_data *)sa_mad->data;
+	struct mlx4_ib_demux_ctx *ctx = &dev->sriov.demux[port - 1];
+	struct mcast_group *group;
+	struct mcast_req *req;
+	int may_create = 0;
+
+	if (ctx->flushing)
+		return -EAGAIN;
+
+	switch (sa_mad->mad_hdr.method) {
+	case IB_MGMT_METHOD_SET:
+		may_create = 1;
+	case IB_SA_METHOD_DELETE:
+		req = kzalloc(sizeof *req, GFP_KERNEL);
+		if (!req)
+			return -ENOMEM;
+
+		req->func = slave;
+		req->sa_mad = *sa_mad;
+
+		mutex_lock(&ctx->mcg_table_lock);
+		group = acquire_group(ctx, &rec->mgid, may_create, GFP_KERNEL);
+		mutex_unlock(&ctx->mcg_table_lock);
+		if (IS_ERR(group)) {
+			kfree(req);
+			return PTR_ERR(group);
+		}
+		mutex_lock(&group->lock);
+		if (group->func[slave].num_pend_reqs > MAX_PEND_REQS_PER_FUNC) {
+			mutex_unlock(&group->lock);
+			mcg_warn_group(group, "Port %d, Func %d has too many pending requests (%d), dropping\n",
+				       port, slave, MAX_PEND_REQS_PER_FUNC);
+			release_group(group, 0);
+			kfree(req);
+			return -ENOMEM;
+		}
+		++group->func[slave].num_pend_reqs;
+		req->group = group;
+		queue_req(req);
+		mutex_unlock(&group->lock);
+		release_group(group, 0);
+		return 1; /* consumed */
+	case IB_SA_METHOD_GET_TABLE:
+	case IB_MGMT_METHOD_GET_RESP:
+	case IB_SA_METHOD_GET_TABLE_RESP:
+	case IB_SA_METHOD_DELETE_RESP:
+		return 0; /* not consumed, pass-through */
+	default:
+		mcg_warn("In multiplex, port %d, func %d: unexpected MCMember method: 0x%x, dropping\n",
+			port, slave, sa_mad->mad_hdr.method);
+		return 1; /* consumed */
+	}
+}
+
+static ssize_t sysfs_show_group(struct device *dev,
+		struct device_attribute *attr, char *buf)
+{
+	struct mcast_group *group =
+		container_of(attr, struct mcast_group, dentry);
+	struct mcast_req *req = NULL;
+	char pending_str[40];
+	char state_str[40];
+	ssize_t len = 0;
+	int f;
+
+	if (group->state == MCAST_IDLE)
+		sprintf(state_str, "%s", get_state_string(group->state));
+	else
+		sprintf(state_str, "%s(TID=0x%llx)",
+				get_state_string(group->state),
+				be64_to_cpu(group->last_req_tid));
+	if (list_empty(&group->pending_list)) {
+		sprintf(pending_str, "No");
+	} else {
+		req = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+		sprintf(pending_str, "Yes(TID=0x%llx)",
+				be64_to_cpu(req->sa_mad.mad_hdr.tid));
+	}
+	len += sprintf(buf + len, "%1d [%02d,%02d,%02d] %4d %4s %5s     ",
+			group->rec.scope_join_state & 0xf,
+			group->members[2], group->members[1], group->members[0],
+			atomic_read(&group->refcount),
+			pending_str,
+			state_str);
+	for (f = 0; f < MAX_VFS; ++f)
+		if (group->func[f].state == MCAST_MEMBER)
+			len += sprintf(buf + len, "%d[%1x] ",
+					f, group->func[f].join_state);
+
+	len += sprintf(buf + len, "\t\t(%4hx %4x %2x %2x %2x %2x %2x "
+		"%4x %4x %2x %2x)\n",
+		be16_to_cpu(group->rec.pkey),
+		be32_to_cpu(group->rec.qkey),
+		(group->rec.mtusel_mtu & 0xc0) >> 6,
+		group->rec.mtusel_mtu & 0x3f,
+		group->rec.tclass,
+		(group->rec.ratesel_rate & 0xc0) >> 6,
+		group->rec.ratesel_rate & 0x3f,
+		(be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0xf0000000) >> 28,
+		(be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x0fffff00) >> 8,
+		be32_to_cpu(group->rec.sl_flowlabel_hoplimit) & 0x000000ff,
+		group->rec.proxy_join);
+
+	return len;
+}
+
+int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx)
+{
+	char name[20];
+
+	atomic_set(&ctx->tid, 0);
+	sprintf(name, "mlx4_ib_mcg%d", ctx->port);
+	ctx->mcg_wq = create_singlethread_workqueue(name);
+	if (!ctx->mcg_wq)
+		return -ENOMEM;
+
+	mutex_init(&ctx->mcg_table_lock);
+	ctx->mcg_table = RB_ROOT;
+	INIT_LIST_HEAD(&ctx->mcg_mgid0_list);
+	ctx->flushing = 0;
+
+	return 0;
+}
+
+static void force_clean_group(struct mcast_group *group)
+{
+	struct mcast_req *req, *tmp
+		;
+	list_for_each_entry_safe(req, tmp, &group->pending_list, group_list) {
+		list_del(&req->group_list);
+		kfree(req);
+	}
+	del_sysfs_port_mcg_attr(group->demux->dev, group->demux->port, &group->dentry.attr);
+	rb_erase(&group->node, &group->demux->mcg_table);
+	kfree(group);
+}
+
+static void _mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
+{
+	int i;
+	struct rb_node *p;
+	struct mcast_group *group;
+	unsigned long end;
+	int count;
+
+	for (i = 0; i < MAX_VFS; ++i)
+		clean_vf_mcast(ctx, i);
+
+	end = jiffies + msecs_to_jiffies(MAD_TIMEOUT_MS + 3000);
+	do {
+		count = 0;
+		mutex_lock(&ctx->mcg_table_lock);
+		for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p))
+			++count;
+		mutex_unlock(&ctx->mcg_table_lock);
+		if (!count)
+			break;
+
+		msleep(1);
+	} while (time_after(end, jiffies));
+
+	flush_workqueue(ctx->mcg_wq);
+	if (destroy_wq)
+		destroy_workqueue(ctx->mcg_wq);
+
+	mutex_lock(&ctx->mcg_table_lock);
+	while ((p = rb_first(&ctx->mcg_table)) != NULL) {
+		group = rb_entry(p, struct mcast_group, node);
+		if (atomic_read(&group->refcount))
+			mcg_warn_group(group, "group refcount %d!!! (pointer %p)\n", atomic_read(&group->refcount), group);
+
+		force_clean_group(group);
+	}
+	mutex_unlock(&ctx->mcg_table_lock);
+}
+
+struct clean_work {
+	struct work_struct work;
+	struct mlx4_ib_demux_ctx *ctx;
+	int destroy_wq;
+};
+
+static void mcg_clean_task(struct work_struct *work)
+{
+	struct clean_work *cw = container_of(work, struct clean_work, work);
+
+	_mlx4_ib_mcg_port_cleanup(cw->ctx, cw->destroy_wq);
+	cw->ctx->flushing = 0;
+	kfree(cw);
+}
+
+void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq)
+{
+	struct clean_work *work;
+
+	if (ctx->flushing)
+		return;
+
+	ctx->flushing = 1;
+
+	if (destroy_wq) {
+		_mlx4_ib_mcg_port_cleanup(ctx, destroy_wq);
+		ctx->flushing = 0;
+		return;
+	}
+
+	work = kmalloc(sizeof *work, GFP_KERNEL);
+	if (!work) {
+		ctx->flushing = 0;
+		mcg_warn("failed allocating work for cleanup\n");
+		return;
+	}
+
+	work->ctx = ctx;
+	work->destroy_wq = destroy_wq;
+	INIT_WORK(&work->work, mcg_clean_task);
+	queue_work(clean_wq, &work->work);
+}
+
+static void build_leave_mad(struct mcast_req *req)
+{
+	struct ib_sa_mad *mad = &req->sa_mad;
+
+	mad->mad_hdr.method = IB_SA_METHOD_DELETE;
+}
+
+
+static void clear_pending_reqs(struct mcast_group *group, int vf)
+{
+	struct mcast_req *req, *tmp, *group_first = NULL;
+	int clear;
+	int pend = 0;
+
+	if (!list_empty(&group->pending_list))
+		group_first = list_first_entry(&group->pending_list, struct mcast_req, group_list);
+
+	list_for_each_entry_safe(req, tmp, &group->func[vf].pending, func_list) {
+		clear = 1;
+		if (group_first == req &&
+		    (group->state == MCAST_JOIN_SENT ||
+		     group->state == MCAST_LEAVE_SENT)) {
+			clear = cancel_delayed_work(&group->timeout_work);
+			pend = !clear;
+			group->state = MCAST_IDLE;
+		}
+		if (clear) {
+			--group->func[vf].num_pend_reqs;
+			list_del(&req->group_list);
+			list_del(&req->func_list);
+			kfree(req);
+			atomic_dec(&group->refcount);
+		}
+	}
+
+	if (!pend && (!list_empty(&group->func[vf].pending) || group->func[vf].num_pend_reqs)) {
+		mcg_warn_group(group, "DRIVER BUG: list_empty %d, num_pend_reqs %d\n",
+			       list_empty(&group->func[vf].pending), group->func[vf].num_pend_reqs);
+	}
+}
+
+static int push_deleteing_req(struct mcast_group *group, int slave)
+{
+	struct mcast_req *req;
+	struct mcast_req *pend_req;
+
+	if (!group->func[slave].join_state)
+		return 0;
+
+	req = kzalloc(sizeof *req, GFP_KERNEL);
+	if (!req) {
+		mcg_warn_group(group, "failed allocation - may leave stall groups\n");
+		return -ENOMEM;
+	}
+
+	if (!list_empty(&group->func[slave].pending)) {
+		pend_req = list_entry(group->func[slave].pending.prev, struct mcast_req, group_list);
+		if (pend_req->clean) {
+			kfree(req);
+			return 0;
+		}
+	}
+
+	req->clean = 1;
+	req->func = slave;
+	req->group = group;
+	++group->func[slave].num_pend_reqs;
+	build_leave_mad(req);
+	queue_req(req);
+	return 0;
+}
+
+void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave)
+{
+	struct mcast_group *group;
+	struct rb_node *p;
+
+	mutex_lock(&ctx->mcg_table_lock);
+	for (p = rb_first(&ctx->mcg_table); p; p = rb_next(p)) {
+		group = rb_entry(p, struct mcast_group, node);
+		mutex_lock(&group->lock);
+		if (atomic_read(&group->refcount)) {
+			/* clear pending requests of this VF */
+			clear_pending_reqs(group, slave);
+			push_deleteing_req(group, slave);
+		}
+		mutex_unlock(&group->lock);
+	}
+	mutex_unlock(&ctx->mcg_table_lock);
+}
+
+
+int mlx4_ib_mcg_init(void)
+{
+	clean_wq = create_singlethread_workqueue("mlx4_ib_mcg");
+	if (!clean_wq)
+		return -ENOMEM;
+
+	return 0;
+}
+
+void mlx4_ib_mcg_destroy(void)
+{
+	destroy_workqueue(clean_wq);
+}
diff --git a/drivers/infiniband/hw/mlx4/mlx4_ib.h b/drivers/infiniband/hw/mlx4/mlx4_ib.h
index c4cf5b69eef..369da3ca5d6 100644
--- a/drivers/infiniband/hw/mlx4/mlx4_ib.h
+++ b/drivers/infiniband/hw/mlx4/mlx4_ib.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2006, 2007 Cisco Systems.  All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -36,13 +37,39 @@
 #include <linux/compiler.h>
 #include <linux/list.h>
 #include <linux/mutex.h>
+#include <linux/idr.h>
 
 #include <rdma/ib_verbs.h>
 #include <rdma/ib_umem.h>
+#include <rdma/ib_mad.h>
+#include <rdma/ib_sa.h>
 
 #include <linux/mlx4/device.h>
 #include <linux/mlx4/doorbell.h>
 
+#define MLX4_IB_DRV_NAME	"mlx4_ib"
+
+#ifdef pr_fmt
+#undef pr_fmt
+#endif
+#define pr_fmt(fmt)	"<" MLX4_IB_DRV_NAME "> %s: " fmt, __func__
+
+#define mlx4_ib_warn(ibdev, format, arg...) \
+	dev_warn((ibdev)->dma_device, MLX4_IB_DRV_NAME ": " format, ## arg)
+
+enum {
+	MLX4_IB_SQ_MIN_WQE_SHIFT = 6,
+	MLX4_IB_MAX_HEADROOM	 = 2048
+};
+
+#define MLX4_IB_SQ_HEADROOM(shift)	((MLX4_IB_MAX_HEADROOM >> (shift)) + 1)
+#define MLX4_IB_SQ_MAX_SPARE		(MLX4_IB_SQ_HEADROOM(MLX4_IB_SQ_MIN_WQE_SHIFT))
+
+/*module param to indicate if SM assigns the alias_GUID*/
+extern int mlx4_ib_sm_guid_assign;
+
+#define MLX4_IB_UC_STEER_QPN_ALIGN 1
+#define MLX4_IB_UC_MAX_NUM_QPS     256
 struct mlx4_ib_ucontext {
 	struct ib_ucontext	ibucontext;
 	struct mlx4_uar		uar;
@@ -55,9 +82,17 @@ struct mlx4_ib_pd {
 	u32			pdn;
 };
 
+struct mlx4_ib_xrcd {
+	struct ib_xrcd		ibxrcd;
+	u32			xrcdn;
+	struct ib_pd	       *pd;
+	struct ib_cq	       *cq;
+};
+
 struct mlx4_ib_cq_buf {
 	struct mlx4_buf		buf;
 	struct mlx4_mtt		mtt;
+	int			entry_size;
 };
 
 struct mlx4_ib_cq_resize {
@@ -83,11 +118,28 @@ struct mlx4_ib_mr {
 	struct ib_umem	       *umem;
 };
 
+struct mlx4_ib_mw {
+	struct ib_mw		ibmw;
+	struct mlx4_mw		mmw;
+};
+
+struct mlx4_ib_fast_reg_page_list {
+	struct ib_fast_reg_page_list	ibfrpl;
+	__be64			       *mapped_page_list;
+	dma_addr_t			map;
+};
+
 struct mlx4_ib_fmr {
 	struct ib_fmr           ibfmr;
 	struct mlx4_fmr         mfmr;
 };
 
+struct mlx4_ib_flow {
+	struct ib_flow ibflow;
+	/* translating DMFS verbs sniffer rule to FW API requires two reg IDs */
+	u64 reg_id[2];
+};
+
 struct mlx4_ib_wq {
 	u64		       *wrid;
 	spinlock_t		lock;
@@ -101,8 +153,109 @@ struct mlx4_ib_wq {
 };
 
 enum mlx4_ib_qp_flags {
-	MLX4_IB_QP_LSO				= 1 << 0,
-	MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK	= 1 << 1,
+	MLX4_IB_QP_LSO = IB_QP_CREATE_IPOIB_UD_LSO,
+	MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK = IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK,
+	MLX4_IB_QP_NETIF = IB_QP_CREATE_NETIF_QP,
+	MLX4_IB_QP_CREATE_USE_GFP_NOIO = IB_QP_CREATE_USE_GFP_NOIO,
+	MLX4_IB_SRIOV_TUNNEL_QP = 1 << 30,
+	MLX4_IB_SRIOV_SQP = 1 << 31,
+};
+
+struct mlx4_ib_gid_entry {
+	struct list_head	list;
+	union ib_gid		gid;
+	int			added;
+	u8			port;
+};
+
+enum mlx4_ib_qp_type {
+	/*
+	 * IB_QPT_SMI and IB_QPT_GSI have to be the first two entries
+	 * here (and in that order) since the MAD layer uses them as
+	 * indices into a 2-entry table.
+	 */
+	MLX4_IB_QPT_SMI = IB_QPT_SMI,
+	MLX4_IB_QPT_GSI = IB_QPT_GSI,
+
+	MLX4_IB_QPT_RC = IB_QPT_RC,
+	MLX4_IB_QPT_UC = IB_QPT_UC,
+	MLX4_IB_QPT_UD = IB_QPT_UD,
+	MLX4_IB_QPT_RAW_IPV6 = IB_QPT_RAW_IPV6,
+	MLX4_IB_QPT_RAW_ETHERTYPE = IB_QPT_RAW_ETHERTYPE,
+	MLX4_IB_QPT_RAW_PACKET = IB_QPT_RAW_PACKET,
+	MLX4_IB_QPT_XRC_INI = IB_QPT_XRC_INI,
+	MLX4_IB_QPT_XRC_TGT = IB_QPT_XRC_TGT,
+
+	MLX4_IB_QPT_PROXY_SMI_OWNER	= 1 << 16,
+	MLX4_IB_QPT_PROXY_SMI		= 1 << 17,
+	MLX4_IB_QPT_PROXY_GSI		= 1 << 18,
+	MLX4_IB_QPT_TUN_SMI_OWNER	= 1 << 19,
+	MLX4_IB_QPT_TUN_SMI		= 1 << 20,
+	MLX4_IB_QPT_TUN_GSI		= 1 << 21,
+};
+
+#define MLX4_IB_QPT_ANY_SRIOV	(MLX4_IB_QPT_PROXY_SMI_OWNER | \
+	MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER | \
+	MLX4_IB_QPT_TUN_SMI | MLX4_IB_QPT_TUN_GSI)
+
+enum mlx4_ib_mad_ifc_flags {
+	MLX4_MAD_IFC_IGNORE_MKEY	= 1,
+	MLX4_MAD_IFC_IGNORE_BKEY	= 2,
+	MLX4_MAD_IFC_IGNORE_KEYS	= (MLX4_MAD_IFC_IGNORE_MKEY |
+					   MLX4_MAD_IFC_IGNORE_BKEY),
+	MLX4_MAD_IFC_NET_VIEW		= 4,
+};
+
+enum {
+	MLX4_NUM_TUNNEL_BUFS		= 256,
+};
+
+struct mlx4_ib_tunnel_header {
+	struct mlx4_av av;
+	__be32 remote_qpn;
+	__be32 qkey;
+	__be16 vlan;
+	u8 mac[6];
+	__be16 pkey_index;
+	u8 reserved[6];
+};
+
+struct mlx4_ib_buf {
+	void *addr;
+	dma_addr_t map;
+};
+
+struct mlx4_rcv_tunnel_hdr {
+	__be32 flags_src_qp; /* flags[6:5] is defined for VLANs:
+			      * 0x0 - no vlan was in the packet
+			      * 0x01 - C-VLAN was in the packet */
+	u8 g_ml_path; /* gid bit stands for ipv6/4 header in RoCE */
+	u8 reserved;
+	__be16 pkey_index;
+	__be16 sl_vid;
+	__be16 slid_mac_47_32;
+	__be32 mac_31_0;
+};
+
+struct mlx4_ib_proxy_sqp_hdr {
+	struct ib_grh grh;
+	struct mlx4_rcv_tunnel_hdr tun;
+}  __packed;
+
+struct mlx4_roce_smac_vlan_info {
+	u64 smac;
+	int smac_index;
+	int smac_port;
+	u64 candidate_smac;
+	int candidate_smac_index;
+	int candidate_smac_port;
+	u16 vid;
+	int vlan_index;
+	int vlan_port;
+	u16 candidate_vid;
+	int candidate_vlan_index;
+	int candidate_vlan_port;
+	int update_vid;
 };
 
 struct mlx4_ib_qp {
@@ -120,10 +273,12 @@ struct mlx4_ib_qp {
 	int			sq_spare_wqes;
 	struct mlx4_ib_wq	sq;
 
+	enum mlx4_ib_qp_type	mlx4_ib_qp_type;
 	struct ib_umem	       *umem;
 	struct mlx4_mtt		mtt;
 	int			buf_size;
 	struct mutex		mutex;
+	u16			xrcdn;
 	u32			flags;
 	u8			port;
 	u8			alt_port;
@@ -131,6 +286,13 @@ struct mlx4_ib_qp {
 	u8			resp_depth;
 	u8			sq_no_prefetch;
 	u8			state;
+	int			mlx_type;
+	struct list_head	gid_list;
+	struct list_head	steering_rules;
+	struct mlx4_ib_buf	*sqp_proxy_rcv;
+	struct mlx4_roce_smac_vlan_info pri;
+	struct mlx4_roce_smac_vlan_info alt;
+	u64			reg_id;
 };
 
 struct mlx4_ib_srq {
@@ -150,12 +312,191 @@ struct mlx4_ib_srq {
 
 struct mlx4_ib_ah {
 	struct ib_ah		ibah;
-	struct mlx4_av		av;
+	union mlx4_ext_av       av;
+};
+
+/****************************************/
+/* alias guid support */
+/****************************************/
+#define NUM_PORT_ALIAS_GUID		2
+#define NUM_ALIAS_GUID_IN_REC		8
+#define NUM_ALIAS_GUID_REC_IN_PORT	16
+#define GUID_REC_SIZE			8
+#define NUM_ALIAS_GUID_PER_PORT		128
+#define MLX4_NOT_SET_GUID		(0x00LL)
+#define MLX4_GUID_FOR_DELETE_VAL	(~(0x00LL))
+
+enum mlx4_guid_alias_rec_status {
+	MLX4_GUID_INFO_STATUS_IDLE,
+	MLX4_GUID_INFO_STATUS_SET,
+	MLX4_GUID_INFO_STATUS_PENDING,
+};
+
+enum mlx4_guid_alias_rec_ownership {
+	MLX4_GUID_DRIVER_ASSIGN,
+	MLX4_GUID_SYSADMIN_ASSIGN,
+	MLX4_GUID_NONE_ASSIGN, /*init state of each record*/
+};
+
+enum mlx4_guid_alias_rec_method {
+	MLX4_GUID_INFO_RECORD_SET	= IB_MGMT_METHOD_SET,
+	MLX4_GUID_INFO_RECORD_DELETE	= IB_SA_METHOD_DELETE,
+};
+
+struct mlx4_sriov_alias_guid_info_rec_det {
+	u8 all_recs[GUID_REC_SIZE * NUM_ALIAS_GUID_IN_REC];
+	ib_sa_comp_mask guid_indexes; /*indicates what from the 8 records are valid*/
+	enum mlx4_guid_alias_rec_status status; /*indicates the administraively status of the record.*/
+	u8 method; /*set or delete*/
+	enum mlx4_guid_alias_rec_ownership ownership; /*indicates who assign that alias_guid record*/
+};
+
+struct mlx4_sriov_alias_guid_port_rec_det {
+	struct mlx4_sriov_alias_guid_info_rec_det all_rec_per_port[NUM_ALIAS_GUID_REC_IN_PORT];
+	struct workqueue_struct *wq;
+	struct delayed_work alias_guid_work;
+	u8 port;
+	struct mlx4_sriov_alias_guid *parent;
+	struct list_head cb_list;
+};
+
+struct mlx4_sriov_alias_guid {
+	struct mlx4_sriov_alias_guid_port_rec_det ports_guid[MLX4_MAX_PORTS];
+	spinlock_t ag_work_lock;
+	struct ib_sa_client *sa_client;
+};
+
+struct mlx4_ib_demux_work {
+	struct work_struct	work;
+	struct mlx4_ib_dev     *dev;
+	int			slave;
+	int			do_init;
+	u8			port;
+
+};
+
+struct mlx4_ib_tun_tx_buf {
+	struct mlx4_ib_buf buf;
+	struct ib_ah *ah;
+};
+
+struct mlx4_ib_demux_pv_qp {
+	struct ib_qp *qp;
+	enum ib_qp_type proxy_qpt;
+	struct mlx4_ib_buf *ring;
+	struct mlx4_ib_tun_tx_buf *tx_ring;
+	spinlock_t tx_lock;
+	unsigned tx_ix_head;
+	unsigned tx_ix_tail;
+};
+
+enum mlx4_ib_demux_pv_state {
+	DEMUX_PV_STATE_DOWN,
+	DEMUX_PV_STATE_STARTING,
+	DEMUX_PV_STATE_ACTIVE,
+	DEMUX_PV_STATE_DOWNING,
+};
+
+struct mlx4_ib_demux_pv_ctx {
+	int port;
+	int slave;
+	enum mlx4_ib_demux_pv_state state;
+	int has_smi;
+	struct ib_device *ib_dev;
+	struct ib_cq *cq;
+	struct ib_pd *pd;
+	struct ib_mr *mr;
+	struct work_struct work;
+	struct workqueue_struct *wq;
+	struct mlx4_ib_demux_pv_qp qp[2];
+};
+
+struct mlx4_ib_demux_ctx {
+	struct ib_device *ib_dev;
+	int port;
+	struct workqueue_struct *wq;
+	struct workqueue_struct *ud_wq;
+	spinlock_t ud_lock;
+	__be64 subnet_prefix;
+	__be64 guid_cache[128];
+	struct mlx4_ib_dev *dev;
+	/* the following lock protects both mcg_table and mcg_mgid0_list */
+	struct mutex		mcg_table_lock;
+	struct rb_root		mcg_table;
+	struct list_head	mcg_mgid0_list;
+	struct workqueue_struct	*mcg_wq;
+	struct mlx4_ib_demux_pv_ctx **tun;
+	atomic_t tid;
+	int    flushing; /* flushing the work queue */
+};
+
+struct mlx4_ib_sriov {
+	struct mlx4_ib_demux_ctx demux[MLX4_MAX_PORTS];
+	struct mlx4_ib_demux_pv_ctx *sqps[MLX4_MAX_PORTS];
+	/* when using this spinlock you should use "irq" because
+	 * it may be called from interrupt context.*/
+	spinlock_t going_down_lock;
+	int is_going_down;
+
+	struct mlx4_sriov_alias_guid alias_guid;
+
+	/* CM paravirtualization fields */
+	struct list_head cm_list;
+	spinlock_t id_map_lock;
+	struct rb_root sl_id_map;
+	struct idr pv_id_table;
+};
+
+struct mlx4_ib_iboe {
+	spinlock_t		lock;
+	struct net_device      *netdevs[MLX4_MAX_PORTS];
+	struct net_device      *masters[MLX4_MAX_PORTS];
+	struct notifier_block 	nb;
+	struct notifier_block	nb_inet;
+	struct notifier_block	nb_inet6;
+	union ib_gid		gid_table[MLX4_MAX_PORTS][128];
+};
+
+struct pkey_mgt {
+	u8			virt2phys_pkey[MLX4_MFUNC_MAX][MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+	u16			phys_pkey_cache[MLX4_MAX_PORTS][MLX4_MAX_PORT_PKEYS];
+	struct list_head	pkey_port_list[MLX4_MFUNC_MAX];
+	struct kobject	       *device_parent[MLX4_MFUNC_MAX];
+};
+
+struct mlx4_ib_iov_sysfs_attr {
+	void *ctx;
+	struct kobject *kobj;
+	unsigned long data;
+	u32 entry_num;
+	char name[15];
+	struct device_attribute dentry;
+	struct device *dev;
+};
+
+struct mlx4_ib_iov_sysfs_attr_ar {
+	struct mlx4_ib_iov_sysfs_attr dentries[3 * NUM_ALIAS_GUID_PER_PORT + 1];
+};
+
+struct mlx4_ib_iov_port {
+	char name[100];
+	u8 num;
+	struct mlx4_ib_dev *dev;
+	struct list_head list;
+	struct mlx4_ib_iov_sysfs_attr_ar *dentr_ar;
+	struct ib_port_attr attr;
+	struct kobject	*cur_port;
+	struct kobject	*admin_alias_parent;
+	struct kobject	*gids_parent;
+	struct kobject	*pkeys_parent;
+	struct kobject	*mcgs_parent;
+	struct mlx4_ib_iov_sysfs_attr mcg_dentry;
 };
 
 struct mlx4_ib_dev {
 	struct ib_device	ib_dev;
 	struct mlx4_dev	       *dev;
+	int			num_ports;
 	void __iomem	       *uar_map;
 
 	struct mlx4_uar		priv_uar;
@@ -165,8 +506,39 @@ struct mlx4_ib_dev {
 	struct ib_mad_agent    *send_agent[MLX4_MAX_PORTS][2];
 	struct ib_ah	       *sm_ah[MLX4_MAX_PORTS];
 	spinlock_t		sm_lock;
+	struct mlx4_ib_sriov	sriov;
 
 	struct mutex		cap_mask_mutex;
+	bool			ib_active;
+	struct mlx4_ib_iboe	iboe;
+	int			counters[MLX4_MAX_PORTS];
+	int		       *eq_table;
+	int			eq_added;
+	struct kobject	       *iov_parent;
+	struct kobject	       *ports_parent;
+	struct kobject	       *dev_ports_parent[MLX4_MFUNC_MAX];
+	struct mlx4_ib_iov_port	iov_ports[MLX4_MAX_PORTS];
+	struct pkey_mgt		pkeys;
+	unsigned long *ib_uc_qpns_bitmap;
+	int steer_qpn_count;
+	int steer_qpn_base;
+	int steering_support;
+	struct mlx4_ib_qp      *qp1_proxy[MLX4_MAX_PORTS];
+	/* lock when destroying qp1_proxy and getting netdev events */
+	struct mutex		qp1_proxy_lock[MLX4_MAX_PORTS];
+};
+
+struct ib_event_work {
+	struct work_struct	work;
+	struct mlx4_ib_dev	*ib_dev;
+	struct mlx4_eqe		ib_eqe;
+};
+
+struct mlx4_ib_qp_tunnel_init_attr {
+	struct ib_qp_init_attr init_attr;
+	int slave;
+	enum ib_qp_type proxy_qp_type;
+	u8 port;
 };
 
 static inline struct mlx4_ib_dev *to_mdev(struct ib_device *ibdev)
@@ -184,6 +556,11 @@ static inline struct mlx4_ib_pd *to_mpd(struct ib_pd *ibpd)
 	return container_of(ibpd, struct mlx4_ib_pd, ibpd);
 }
 
+static inline struct mlx4_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
+{
+	return container_of(ibxrcd, struct mlx4_ib_xrcd, ibxrcd);
+}
+
 static inline struct mlx4_ib_cq *to_mcq(struct ib_cq *ibcq)
 {
 	return container_of(ibcq, struct mlx4_ib_cq, ibcq);
@@ -199,10 +576,26 @@ static inline struct mlx4_ib_mr *to_mmr(struct ib_mr *ibmr)
 	return container_of(ibmr, struct mlx4_ib_mr, ibmr);
 }
 
+static inline struct mlx4_ib_mw *to_mmw(struct ib_mw *ibmw)
+{
+	return container_of(ibmw, struct mlx4_ib_mw, ibmw);
+}
+
+static inline struct mlx4_ib_fast_reg_page_list *to_mfrpl(struct ib_fast_reg_page_list *ibfrpl)
+{
+	return container_of(ibfrpl, struct mlx4_ib_fast_reg_page_list, ibfrpl);
+}
+
 static inline struct mlx4_ib_fmr *to_mfmr(struct ib_fmr *ibfmr)
 {
 	return container_of(ibfmr, struct mlx4_ib_fmr, ibfmr);
 }
+
+static inline struct mlx4_ib_flow *to_mflow(struct ib_flow *ibflow)
+{
+	return container_of(ibflow, struct mlx4_ib_flow, ibflow);
+}
+
 static inline struct mlx4_ib_qp *to_mqp(struct ib_qp *ibqp)
 {
 	return container_of(ibqp, struct mlx4_ib_qp, ibqp);
@@ -228,6 +621,9 @@ static inline struct mlx4_ib_ah *to_mah(struct ib_ah *ibah)
 	return container_of(ibah, struct mlx4_ib_ah, ibah);
 }
 
+int mlx4_ib_init_sriov(struct mlx4_ib_dev *dev);
+void mlx4_ib_close_sriov(struct mlx4_ib_dev *dev);
+
 int mlx4_ib_db_map_user(struct mlx4_ib_ucontext *context, unsigned long virt,
 			struct mlx4_db *db);
 void mlx4_ib_db_unmap_user(struct mlx4_ib_ucontext *context, struct mlx4_db *db);
@@ -239,6 +635,15 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				  u64 virt_addr, int access_flags,
 				  struct ib_udata *udata);
 int mlx4_ib_dereg_mr(struct ib_mr *mr);
+struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type);
+int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
+		    struct ib_mw_bind *mw_bind);
+int mlx4_ib_dealloc_mw(struct ib_mw *mw);
+struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len);
+struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len);
+void mlx4_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
 
 int mlx4_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
 int mlx4_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
@@ -279,7 +684,7 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 		      struct ib_recv_wr **bad_wr);
 
-int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+int mlx4_MAD_IFC(struct mlx4_ib_dev *dev, int mad_ifc_flags,
 		 int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
 		 void *in_mad, void *response_mad);
 int mlx4_ib_process_mad(struct ib_device *ibdev, int mad_flags,	u8 port_num,
@@ -294,10 +699,94 @@ int mlx4_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list, int npages,
 			 u64 iova);
 int mlx4_ib_unmap_fmr(struct list_head *fmr_list);
 int mlx4_ib_fmr_dealloc(struct ib_fmr *fmr);
+int __mlx4_ib_query_port(struct ib_device *ibdev, u8 port,
+			 struct ib_port_attr *props, int netw_view);
+int __mlx4_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			 u16 *pkey, int netw_view);
+
+int __mlx4_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+			union ib_gid *gid, int netw_view);
 
-static inline int mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah)
+static inline bool mlx4_ib_ah_grh_present(struct mlx4_ib_ah *ah)
 {
-	return !!(ah->av.g_slid & 0x80);
+	u8 port = be32_to_cpu(ah->av.ib.port_pd) >> 24 & 3;
+
+	if (rdma_port_get_link_layer(ah->ibah.device, port) == IB_LINK_LAYER_ETHERNET)
+		return true;
+
+	return !!(ah->av.ib.g_slid & 0x80);
 }
 
+int mlx4_ib_mcg_port_init(struct mlx4_ib_demux_ctx *ctx);
+void mlx4_ib_mcg_port_cleanup(struct mlx4_ib_demux_ctx *ctx, int destroy_wq);
+void clean_vf_mcast(struct mlx4_ib_demux_ctx *ctx, int slave);
+int mlx4_ib_mcg_init(void);
+void mlx4_ib_mcg_destroy(void);
+
+int mlx4_ib_find_real_gid(struct ib_device *ibdev, u8 port, __be64 guid);
+
+int mlx4_ib_mcg_multiplex_handler(struct ib_device *ibdev, int port, int slave,
+				  struct ib_sa_mad *sa_mad);
+int mlx4_ib_mcg_demux_handler(struct ib_device *ibdev, int port, int slave,
+			      struct ib_sa_mad *mad);
+
+int mlx4_ib_add_mc(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+		   union ib_gid *gid);
+
+void mlx4_ib_dispatch_event(struct mlx4_ib_dev *dev, u8 port_num,
+			    enum ib_event_type type);
+
+void mlx4_ib_tunnels_update_work(struct work_struct *work);
+
+int mlx4_ib_send_to_slave(struct mlx4_ib_dev *dev, int slave, u8 port,
+			  enum ib_qp_type qpt, struct ib_wc *wc,
+			  struct ib_grh *grh, struct ib_mad *mad);
+
+int mlx4_ib_send_to_wire(struct mlx4_ib_dev *dev, int slave, u8 port,
+			 enum ib_qp_type dest_qpt, u16 pkey_index, u32 remote_qpn,
+			 u32 qkey, struct ib_ah_attr *attr, u8 *s_mac,
+			 struct ib_mad *mad);
+
+__be64 mlx4_ib_get_new_demux_tid(struct mlx4_ib_demux_ctx *ctx);
+
+int mlx4_ib_demux_cm_handler(struct ib_device *ibdev, int port, int *slave,
+		struct ib_mad *mad);
+
+int mlx4_ib_multiplex_cm_handler(struct ib_device *ibdev, int port, int slave_id,
+		struct ib_mad *mad);
+
+void mlx4_ib_cm_paravirt_init(struct mlx4_ib_dev *dev);
+void mlx4_ib_cm_paravirt_clean(struct mlx4_ib_dev *dev, int slave_id);
+
+/* alias guid support */
+void mlx4_ib_init_alias_guid_work(struct mlx4_ib_dev *dev, int port);
+int mlx4_ib_init_alias_guid_service(struct mlx4_ib_dev *dev);
+void mlx4_ib_destroy_alias_guid_service(struct mlx4_ib_dev *dev);
+void mlx4_ib_invalidate_all_guid_record(struct mlx4_ib_dev *dev, int port);
+
+void mlx4_ib_notify_slaves_on_guid_change(struct mlx4_ib_dev *dev,
+					  int block_num,
+					  u8 port_num, u8 *p_data);
+
+void mlx4_ib_update_cache_on_guid_change(struct mlx4_ib_dev *dev,
+					 int block_num, u8 port_num,
+					 u8 *p_data);
+
+int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+			    struct attribute *attr);
+void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+			     struct attribute *attr);
+ib_sa_comp_mask mlx4_ib_get_aguid_comp_mask_from_ix(int index);
+
+int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *device) ;
+
+void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device);
+
+__be64 mlx4_ib_gen_node_guid(void);
+
+int mlx4_ib_steer_qp_alloc(struct mlx4_ib_dev *dev, int count, int *qpn);
+void mlx4_ib_steer_qp_free(struct mlx4_ib_dev *dev, u32 qpn, int count);
+int mlx4_ib_steer_qp_reg(struct mlx4_ib_dev *mdev, struct mlx4_ib_qp *mqp,
+			 int is_attach);
+
 #endif /* MLX4_IB_H */
diff --git a/drivers/infiniband/hw/mlx4/mr.c b/drivers/infiniband/hw/mlx4/mr.c
index 68e92485fc7..cb2a8727f3f 100644
--- a/drivers/infiniband/hw/mlx4/mr.c
+++ b/drivers/infiniband/hw/mlx4/mr.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -30,6 +31,8 @@
  * SOFTWARE.
  */
 
+#include <linux/slab.h>
+
 #include "mlx4_ib.h"
 
 static u32 convert_access(int acc)
@@ -38,9 +41,19 @@ static u32 convert_access(int acc)
 	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX4_PERM_REMOTE_WRITE : 0) |
 	       (acc & IB_ACCESS_REMOTE_READ   ? MLX4_PERM_REMOTE_READ  : 0) |
 	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX4_PERM_LOCAL_WRITE  : 0) |
+	       (acc & IB_ACCESS_MW_BIND	      ? MLX4_PERM_BIND_MW      : 0) |
 	       MLX4_PERM_LOCAL_READ;
 }
 
+static enum mlx4_mw_type to_mlx4_type(enum ib_mw_type type)
+{
+	switch (type) {
+	case IB_MW_TYPE_1:	return MLX4_MW_TYPE_1;
+	case IB_MW_TYPE_2:	return MLX4_MW_TYPE_2;
+	default:		return -1;
+	}
+}
+
 struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
 {
 	struct mlx4_ib_mr *mr;
@@ -65,7 +78,7 @@ struct ib_mr *mlx4_ib_get_dma_mr(struct ib_pd *pd, int acc)
 	return &mr->ibmr;
 
 err_mr:
-	mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
+	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
 
 err_free:
 	kfree(mr);
@@ -77,11 +90,11 @@ int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
 			   struct ib_umem *umem)
 {
 	u64 *pages;
-	struct ib_umem_chunk *chunk;
-	int i, j, k;
+	int i, k, entry;
 	int n;
 	int len;
 	int err = 0;
+	struct scatterlist *sg;
 
 	pages = (u64 *) __get_free_page(GFP_KERNEL);
 	if (!pages)
@@ -89,26 +102,25 @@ int mlx4_ib_umem_write_mtt(struct mlx4_ib_dev *dev, struct mlx4_mtt *mtt,
 
 	i = n = 0;
 
-	list_for_each_entry(chunk, &umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> mtt->page_shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
-					umem->page_size * k;
-				/*
-				 * Be friendly to mlx4_write_mtt() and
-				 * pass it chunks of appropriate size.
-				 */
-				if (i == PAGE_SIZE / sizeof (u64)) {
-					err = mlx4_write_mtt(dev->dev, mtt, n,
-							     i, pages);
-					if (err)
-						goto out;
-					n += i;
-					i = 0;
-				}
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> mtt->page_shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = sg_dma_address(sg) +
+				umem->page_size * k;
+			/*
+			 * Be friendly to mlx4_write_mtt() and
+			 * pass it chunks of appropriate size.
+			 */
+			if (i == PAGE_SIZE / sizeof (u64)) {
+				err = mlx4_write_mtt(dev->dev, mtt, n,
+						     i, pages);
+				if (err)
+					goto out;
+				n += i;
+				i = 0;
 			}
 		}
+	}
 
 	if (i)
 		err = mlx4_write_mtt(dev->dev, mtt, n, i, pages);
@@ -160,7 +172,7 @@ struct ib_mr *mlx4_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	return &mr->ibmr;
 
 err_mr:
-	mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
+	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &mr->mmr);
 
 err_umem:
 	ib_umem_release(mr->umem);
@@ -174,8 +186,11 @@ err_free:
 int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
 {
 	struct mlx4_ib_mr *mr = to_mmr(ibmr);
+	int ret;
 
-	mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr);
+	ret = mlx4_mr_free(to_mdev(ibmr->device)->dev, &mr->mmr);
+	if (ret)
+		return ret;
 	if (mr->umem)
 		ib_umem_release(mr->umem);
 	kfree(mr);
@@ -183,6 +198,149 @@ int mlx4_ib_dereg_mr(struct ib_mr *ibmr)
 	return 0;
 }
 
+struct ib_mw *mlx4_ib_alloc_mw(struct ib_pd *pd, enum ib_mw_type type)
+{
+	struct mlx4_ib_dev *dev = to_mdev(pd->device);
+	struct mlx4_ib_mw *mw;
+	int err;
+
+	mw = kmalloc(sizeof(*mw), GFP_KERNEL);
+	if (!mw)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx4_mw_alloc(dev->dev, to_mpd(pd)->pdn,
+			    to_mlx4_type(type), &mw->mmw);
+	if (err)
+		goto err_free;
+
+	err = mlx4_mw_enable(dev->dev, &mw->mmw);
+	if (err)
+		goto err_mw;
+
+	mw->ibmw.rkey = mw->mmw.key;
+
+	return &mw->ibmw;
+
+err_mw:
+	mlx4_mw_free(dev->dev, &mw->mmw);
+
+err_free:
+	kfree(mw);
+
+	return ERR_PTR(err);
+}
+
+int mlx4_ib_bind_mw(struct ib_qp *qp, struct ib_mw *mw,
+		    struct ib_mw_bind *mw_bind)
+{
+	struct ib_send_wr  wr;
+	struct ib_send_wr *bad_wr;
+	int ret;
+
+	memset(&wr, 0, sizeof(wr));
+	wr.opcode               = IB_WR_BIND_MW;
+	wr.wr_id                = mw_bind->wr_id;
+	wr.send_flags           = mw_bind->send_flags;
+	wr.wr.bind_mw.mw        = mw;
+	wr.wr.bind_mw.bind_info = mw_bind->bind_info;
+	wr.wr.bind_mw.rkey      = ib_inc_rkey(mw->rkey);
+
+	ret = mlx4_ib_post_send(qp, &wr, &bad_wr);
+	if (!ret)
+		mw->rkey = wr.wr.bind_mw.rkey;
+
+	return ret;
+}
+
+int mlx4_ib_dealloc_mw(struct ib_mw *ibmw)
+{
+	struct mlx4_ib_mw *mw = to_mmw(ibmw);
+
+	mlx4_mw_free(to_mdev(ibmw->device)->dev, &mw->mmw);
+	kfree(mw);
+
+	return 0;
+}
+
+struct ib_mr *mlx4_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len)
+{
+	struct mlx4_ib_dev *dev = to_mdev(pd->device);
+	struct mlx4_ib_mr *mr;
+	int err;
+
+	mr = kmalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx4_mr_alloc(dev->dev, to_mpd(pd)->pdn, 0, 0, 0,
+			    max_page_list_len, 0, &mr->mmr);
+	if (err)
+		goto err_free;
+
+	err = mlx4_mr_enable(dev->dev, &mr->mmr);
+	if (err)
+		goto err_mr;
+
+	mr->ibmr.rkey = mr->ibmr.lkey = mr->mmr.key;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+err_mr:
+	(void) mlx4_mr_free(dev->dev, &mr->mmr);
+
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+struct ib_fast_reg_page_list *mlx4_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len)
+{
+	struct mlx4_ib_dev *dev = to_mdev(ibdev);
+	struct mlx4_ib_fast_reg_page_list *mfrpl;
+	int size = page_list_len * sizeof (u64);
+
+	if (page_list_len > MLX4_MAX_FAST_REG_PAGES)
+		return ERR_PTR(-EINVAL);
+
+	mfrpl = kmalloc(sizeof *mfrpl, GFP_KERNEL);
+	if (!mfrpl)
+		return ERR_PTR(-ENOMEM);
+
+	mfrpl->ibfrpl.page_list = kmalloc(size, GFP_KERNEL);
+	if (!mfrpl->ibfrpl.page_list)
+		goto err_free;
+
+	mfrpl->mapped_page_list = dma_alloc_coherent(&dev->dev->pdev->dev,
+						     size, &mfrpl->map,
+						     GFP_KERNEL);
+	if (!mfrpl->mapped_page_list)
+		goto err_free;
+
+	WARN_ON(mfrpl->map & 0x3f);
+
+	return &mfrpl->ibfrpl;
+
+err_free:
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+	return ERR_PTR(-ENOMEM);
+}
+
+void mlx4_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list)
+{
+	struct mlx4_ib_dev *dev = to_mdev(page_list->device);
+	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list);
+	int size = page_list->max_page_list_len * sizeof (u64);
+
+	dma_free_coherent(&dev->dev->pdev->dev, size, mfrpl->mapped_page_list,
+			  mfrpl->map);
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+}
+
 struct ib_fmr *mlx4_ib_fmr_alloc(struct ib_pd *pd, int acc,
 				 struct ib_fmr_attr *fmr_attr)
 {
@@ -209,7 +367,7 @@ struct ib_fmr *mlx4_ib_fmr_alloc(struct ib_pd *pd, int acc,
 	return &fmr->ibfmr;
 
 err_mr:
-	mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr);
+	(void) mlx4_mr_free(to_mdev(pd->device)->dev, &fmr->mfmr.mr);
 
 err_free:
 	kfree(fmr);
@@ -256,7 +414,7 @@ int mlx4_ib_unmap_fmr(struct list_head *fmr_list)
 
 	err = mlx4_SYNC_TPT(mdev);
 	if (err)
-		printk(KERN_WARNING "mlx4_ib: SYNC_TPT error %d when "
+		pr_warn("SYNC_TPT error %d when "
 		       "unmapping FMRs\n", err);
 
 	return 0;
diff --git a/drivers/infiniband/hw/mlx4/qp.c b/drivers/infiniband/hw/mlx4/qp.c
index 89eb6cbe592..67780452f0c 100644
--- a/drivers/infiniband/hw/mlx4/qp.c
+++ b/drivers/infiniband/hw/mlx4/qp.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -31,9 +32,13 @@
  */
 
 #include <linux/log2.h>
+#include <linux/slab.h>
+#include <linux/netdevice.h>
 
 #include <rdma/ib_cache.h>
 #include <rdma/ib_pack.h>
+#include <rdma/ib_addr.h>
+#include <rdma/ib_mad.h>
 
 #include <linux/mlx4/qp.h>
 
@@ -46,14 +51,24 @@ enum {
 
 enum {
 	MLX4_IB_DEFAULT_SCHED_QUEUE	= 0x83,
-	MLX4_IB_DEFAULT_QP0_SCHED_QUEUE	= 0x3f
+	MLX4_IB_DEFAULT_QP0_SCHED_QUEUE	= 0x3f,
+	MLX4_IB_LINK_TYPE_IB		= 0,
+	MLX4_IB_LINK_TYPE_ETH		= 1
 };
 
 enum {
 	/*
-	 * Largest possible UD header: send with GRH and immediate data.
+	 * Largest possible UD header: send with GRH and immediate
+	 * data plus 18 bytes for an Ethernet header with VLAN/802.1Q
+	 * tag.  (LRH would only use 8 bytes, so Ethernet is the
+	 * biggest case)
 	 */
-	MLX4_IB_UD_HEADER_SIZE		= 72
+	MLX4_IB_UD_HEADER_SIZE		= 82,
+	MLX4_IB_LSO_HEADER_SPARE	= 128,
+};
+
+enum {
+	MLX4_IB_IBOE_ETHERTYPE		= 0x8915
 };
 
 struct mlx4_ib_sqp {
@@ -66,18 +81,45 @@ struct mlx4_ib_sqp {
 };
 
 enum {
-	MLX4_IB_MIN_SQ_STRIDE = 6
+	MLX4_IB_MIN_SQ_STRIDE	= 6,
+	MLX4_IB_CACHE_LINE_SIZE	= 64,
+};
+
+enum {
+	MLX4_RAW_QP_MTU		= 7,
+	MLX4_RAW_QP_MSGMAX	= 31,
 };
 
+#ifndef ETH_ALEN
+#define ETH_ALEN        6
+#endif
+static inline u64 mlx4_mac_to_u64(u8 *addr)
+{
+	u64 mac = 0;
+	int i;
+
+	for (i = 0; i < ETH_ALEN; i++) {
+		mac <<= 8;
+		mac |= addr[i];
+	}
+	return mac;
+}
+
 static const __be32 mlx4_ib_opcode[] = {
-	[IB_WR_SEND]			= __constant_cpu_to_be32(MLX4_OPCODE_SEND),
-	[IB_WR_LSO]			= __constant_cpu_to_be32(MLX4_OPCODE_LSO),
-	[IB_WR_SEND_WITH_IMM]		= __constant_cpu_to_be32(MLX4_OPCODE_SEND_IMM),
-	[IB_WR_RDMA_WRITE]		= __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
-	[IB_WR_RDMA_WRITE_WITH_IMM]	= __constant_cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
-	[IB_WR_RDMA_READ]		= __constant_cpu_to_be32(MLX4_OPCODE_RDMA_READ),
-	[IB_WR_ATOMIC_CMP_AND_SWP]	= __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_CS),
-	[IB_WR_ATOMIC_FETCH_AND_ADD]	= __constant_cpu_to_be32(MLX4_OPCODE_ATOMIC_FA),
+	[IB_WR_SEND]				= cpu_to_be32(MLX4_OPCODE_SEND),
+	[IB_WR_LSO]				= cpu_to_be32(MLX4_OPCODE_LSO),
+	[IB_WR_SEND_WITH_IMM]			= cpu_to_be32(MLX4_OPCODE_SEND_IMM),
+	[IB_WR_RDMA_WRITE]			= cpu_to_be32(MLX4_OPCODE_RDMA_WRITE),
+	[IB_WR_RDMA_WRITE_WITH_IMM]		= cpu_to_be32(MLX4_OPCODE_RDMA_WRITE_IMM),
+	[IB_WR_RDMA_READ]			= cpu_to_be32(MLX4_OPCODE_RDMA_READ),
+	[IB_WR_ATOMIC_CMP_AND_SWP]		= cpu_to_be32(MLX4_OPCODE_ATOMIC_CS),
+	[IB_WR_ATOMIC_FETCH_AND_ADD]		= cpu_to_be32(MLX4_OPCODE_ATOMIC_FA),
+	[IB_WR_SEND_WITH_INV]			= cpu_to_be32(MLX4_OPCODE_SEND_INVAL),
+	[IB_WR_LOCAL_INV]			= cpu_to_be32(MLX4_OPCODE_LOCAL_INVAL),
+	[IB_WR_FAST_REG_MR]			= cpu_to_be32(MLX4_OPCODE_FMR),
+	[IB_WR_MASKED_ATOMIC_CMP_AND_SWP]	= cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_CS),
+	[IB_WR_MASKED_ATOMIC_FETCH_AND_ADD]	= cpu_to_be32(MLX4_OPCODE_MASKED_ATOMIC_FA),
+	[IB_WR_BIND_MW]				= cpu_to_be32(MLX4_OPCODE_BIND_MW),
 };
 
 static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
@@ -85,16 +127,62 @@ static struct mlx4_ib_sqp *to_msqp(struct mlx4_ib_qp *mqp)
 	return container_of(mqp, struct mlx4_ib_sqp, qp);
 }
 
+static int is_tunnel_qp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
+{
+	if (!mlx4_is_master(dev->dev))
+		return 0;
+
+	return qp->mqp.qpn >= dev->dev->phys_caps.base_tunnel_sqpn &&
+	       qp->mqp.qpn < dev->dev->phys_caps.base_tunnel_sqpn +
+		8 * MLX4_MFUNC_MAX;
+}
+
 static int is_sqp(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 {
-	return qp->mqp.qpn >= dev->dev->caps.sqp_start &&
-		qp->mqp.qpn <= dev->dev->caps.sqp_start + 3;
+	int proxy_sqp = 0;
+	int real_sqp = 0;
+	int i;
+	/* PPF or Native -- real SQP */
+	real_sqp = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+		    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+		    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 3);
+	if (real_sqp)
+		return 1;
+	/* VF or PF -- proxy SQP */
+	if (mlx4_is_mfunc(dev->dev)) {
+		for (i = 0; i < dev->dev->caps.num_ports; i++) {
+			if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i] ||
+			    qp->mqp.qpn == dev->dev->caps.qp1_proxy[i]) {
+				proxy_sqp = 1;
+				break;
+			}
+		}
+	}
+	return proxy_sqp;
 }
 
+/* used for INIT/CLOSE port logic */
 static int is_qp0(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
 {
-	return qp->mqp.qpn >= dev->dev->caps.sqp_start &&
-		qp->mqp.qpn <= dev->dev->caps.sqp_start + 1;
+	int proxy_qp0 = 0;
+	int real_qp0 = 0;
+	int i;
+	/* PPF or Native -- real QP0 */
+	real_qp0 = ((mlx4_is_master(dev->dev) || !mlx4_is_mfunc(dev->dev)) &&
+		    qp->mqp.qpn >= dev->dev->phys_caps.base_sqpn &&
+		    qp->mqp.qpn <= dev->dev->phys_caps.base_sqpn + 1);
+	if (real_qp0)
+		return 1;
+	/* VF or PF -- proxy QP0 */
+	if (mlx4_is_mfunc(dev->dev)) {
+		for (i = 0; i < dev->dev->caps.num_ports; i++) {
+			if (qp->mqp.qpn == dev->dev->caps.qp0_proxy[i]) {
+				proxy_qp0 = 1;
+				break;
+			}
+		}
+	}
+	return proxy_qp0;
 }
 
 static void *get_wqe(struct mlx4_ib_qp *qp, int offset)
@@ -236,7 +324,7 @@ static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type)
 			event.event = IB_EVENT_QP_ACCESS_ERR;
 			break;
 		default:
-			printk(KERN_WARNING "mlx4_ib: Unexpected event type %d "
+			pr_warn("Unexpected event type %d "
 			       "on QP %06x\n", type, qp->qpn);
 			return;
 		}
@@ -245,7 +333,7 @@ static void mlx4_ib_qp_event(struct mlx4_qp *qp, enum mlx4_event type)
 	}
 }
 
-static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
+static int send_wqe_overhead(enum mlx4_ib_qp_type type, u32 flags)
 {
 	/*
 	 * UD WQEs must have a datagram segment.
@@ -254,19 +342,29 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
 	 * header and space for the ICRC).
 	 */
 	switch (type) {
-	case IB_QPT_UD:
+	case MLX4_IB_QPT_UD:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
 			sizeof (struct mlx4_wqe_datagram_seg) +
-			((flags & MLX4_IB_QP_LSO) ? 64 : 0);
-	case IB_QPT_UC:
+			((flags & MLX4_IB_QP_LSO) ? MLX4_IB_LSO_HEADER_SPARE : 0);
+	case MLX4_IB_QPT_PROXY_SMI_OWNER:
+	case MLX4_IB_QPT_PROXY_SMI:
+	case MLX4_IB_QPT_PROXY_GSI:
+		return sizeof (struct mlx4_wqe_ctrl_seg) +
+			sizeof (struct mlx4_wqe_datagram_seg) + 64;
+	case MLX4_IB_QPT_TUN_SMI_OWNER:
+	case MLX4_IB_QPT_TUN_GSI:
+		return sizeof (struct mlx4_wqe_ctrl_seg) +
+			sizeof (struct mlx4_wqe_datagram_seg);
+
+	case MLX4_IB_QPT_UC:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
 			sizeof (struct mlx4_wqe_raddr_seg);
-	case IB_QPT_RC:
+	case MLX4_IB_QPT_RC:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
 			sizeof (struct mlx4_wqe_atomic_seg) +
 			sizeof (struct mlx4_wqe_raddr_seg);
-	case IB_QPT_SMI:
-	case IB_QPT_GSI:
+	case MLX4_IB_QPT_SMI:
+	case MLX4_IB_QPT_GSI:
 		return sizeof (struct mlx4_wqe_ctrl_seg) +
 			ALIGN(MLX4_IB_UD_HEADER_SIZE +
 			      DIV_ROUND_UP(MLX4_IB_UD_HEADER_SIZE,
@@ -282,15 +380,14 @@ static int send_wqe_overhead(enum ib_qp_type type, u32 flags)
 }
 
 static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
-		       int is_user, int has_srq, struct mlx4_ib_qp *qp)
+		       int is_user, int has_rq, struct mlx4_ib_qp *qp)
 {
 	/* Sanity check RQ size before proceeding */
-	if (cap->max_recv_wr  > dev->dev->caps.max_wqes  ||
-	    cap->max_recv_sge > dev->dev->caps.max_rq_sg)
+	if (cap->max_recv_wr > dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE ||
+	    cap->max_recv_sge > min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg))
 		return -EINVAL;
 
-	if (has_srq) {
-		/* QPs attached to an SRQ should have no RQ */
+	if (!has_rq) {
 		if (cap->max_recv_wr)
 			return -EINVAL;
 
@@ -305,20 +402,29 @@ static int set_rq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 		qp->rq.wqe_shift = ilog2(qp->rq.max_gs * sizeof (struct mlx4_wqe_data_seg));
 	}
 
-	cap->max_recv_wr  = qp->rq.max_post = qp->rq.wqe_cnt;
-	cap->max_recv_sge = qp->rq.max_gs;
+	/* leave userspace return values as they were, so as not to break ABI */
+	if (is_user) {
+		cap->max_recv_wr  = qp->rq.max_post = qp->rq.wqe_cnt;
+		cap->max_recv_sge = qp->rq.max_gs;
+	} else {
+		cap->max_recv_wr  = qp->rq.max_post =
+			min(dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE, qp->rq.wqe_cnt);
+		cap->max_recv_sge = min(qp->rq.max_gs,
+					min(dev->dev->caps.max_sq_sg,
+					    dev->dev->caps.max_rq_sg));
+	}
 
 	return 0;
 }
 
 static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
-			      enum ib_qp_type type, struct mlx4_ib_qp *qp)
+			      enum mlx4_ib_qp_type type, struct mlx4_ib_qp *qp)
 {
 	int s;
 
 	/* Sanity check SQ size before proceeding */
-	if (cap->max_send_wr	 > dev->dev->caps.max_wqes  ||
-	    cap->max_send_sge	 > dev->dev->caps.max_sq_sg ||
+	if (cap->max_send_wr  > (dev->dev->caps.max_wqes - MLX4_IB_SQ_MAX_SPARE) ||
+	    cap->max_send_sge > min(dev->dev->caps.max_sq_sg, dev->dev->caps.max_rq_sg) ||
 	    cap->max_inline_data + send_wqe_overhead(type, qp->flags) +
 	    sizeof (struct mlx4_wqe_inline_seg) > dev->dev->caps.max_sq_desc_sz)
 		return -EINVAL;
@@ -327,7 +433,8 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 	 * For MLX transport we need 2 extra S/G entries:
 	 * one for the header and one for the checksum at the end
 	 */
-	if ((type == IB_QPT_SMI || type == IB_QPT_GSI) &&
+	if ((type == MLX4_IB_QPT_SMI || type == MLX4_IB_QPT_GSI ||
+	     type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) &&
 	    cap->max_send_sge + 2 > dev->dev->caps.max_sq_sg)
 		return -EINVAL;
 
@@ -348,7 +455,7 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 	 * anymore, so we do this only if selective signaling is off.
 	 *
 	 * Further, on 32-bit platforms, we can't use vmap() to make
-	 * the QP buffer virtually contigious.  Thus we have to use
+	 * the QP buffer virtually contiguous.  Thus we have to use
 	 * constant-sized WRs to make sure a WR is always fully within
 	 * a single page-sized chunk.
 	 *
@@ -371,7 +478,9 @@ static int set_kernel_sq_size(struct mlx4_ib_dev *dev, struct ib_qp_cap *cap,
 	 */
 	if (dev->dev->caps.fw_ver >= MLX4_FW_VER_WQE_CTRL_NEC &&
 	    qp->sq_signal_bits && BITS_PER_LONG == 64 &&
-	    type != IB_QPT_SMI && type != IB_QPT_GSI)
+	    type != MLX4_IB_QPT_SMI && type != MLX4_IB_QPT_GSI &&
+	    !(type & (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_PROXY_SMI |
+		      MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER)))
 		qp->sq.wqe_shift = ilog2(64);
 	else
 		qp->sq.wqe_shift = ilog2(roundup_pow_of_two(s));
@@ -443,21 +552,157 @@ static int set_user_sq_size(struct mlx4_ib_dev *dev,
 	return 0;
 }
 
+static int alloc_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
+{
+	int i;
+
+	qp->sqp_proxy_rcv =
+		kmalloc(sizeof (struct mlx4_ib_buf) * qp->rq.wqe_cnt,
+			GFP_KERNEL);
+	if (!qp->sqp_proxy_rcv)
+		return -ENOMEM;
+	for (i = 0; i < qp->rq.wqe_cnt; i++) {
+		qp->sqp_proxy_rcv[i].addr =
+			kmalloc(sizeof (struct mlx4_ib_proxy_sqp_hdr),
+				GFP_KERNEL);
+		if (!qp->sqp_proxy_rcv[i].addr)
+			goto err;
+		qp->sqp_proxy_rcv[i].map =
+			ib_dma_map_single(dev, qp->sqp_proxy_rcv[i].addr,
+					  sizeof (struct mlx4_ib_proxy_sqp_hdr),
+					  DMA_FROM_DEVICE);
+	}
+	return 0;
+
+err:
+	while (i > 0) {
+		--i;
+		ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
+				    sizeof (struct mlx4_ib_proxy_sqp_hdr),
+				    DMA_FROM_DEVICE);
+		kfree(qp->sqp_proxy_rcv[i].addr);
+	}
+	kfree(qp->sqp_proxy_rcv);
+	qp->sqp_proxy_rcv = NULL;
+	return -ENOMEM;
+}
+
+static void free_proxy_bufs(struct ib_device *dev, struct mlx4_ib_qp *qp)
+{
+	int i;
+
+	for (i = 0; i < qp->rq.wqe_cnt; i++) {
+		ib_dma_unmap_single(dev, qp->sqp_proxy_rcv[i].map,
+				    sizeof (struct mlx4_ib_proxy_sqp_hdr),
+				    DMA_FROM_DEVICE);
+		kfree(qp->sqp_proxy_rcv[i].addr);
+	}
+	kfree(qp->sqp_proxy_rcv);
+}
+
+static int qp_has_rq(struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_XRC_INI || attr->qp_type == IB_QPT_XRC_TGT)
+		return 0;
+
+	return !attr->srq;
+}
+
+static int qp0_enabled_vf(struct mlx4_dev *dev, int qpn)
+{
+	int i;
+	for (i = 0; i < dev->caps.num_ports; i++) {
+		if (qpn == dev->caps.qp0_proxy[i])
+			return !!dev->caps.qp0_qkey[i];
+	}
+	return 0;
+}
+
 static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 			    struct ib_qp_init_attr *init_attr,
-			    struct ib_udata *udata, int sqpn, struct mlx4_ib_qp *qp)
+			    struct ib_udata *udata, int sqpn, struct mlx4_ib_qp **caller_qp,
+			    gfp_t gfp)
 {
+	int qpn;
 	int err;
+	struct mlx4_ib_sqp *sqp;
+	struct mlx4_ib_qp *qp;
+	enum mlx4_ib_qp_type qp_type = (enum mlx4_ib_qp_type) init_attr->qp_type;
+
+	/* When tunneling special qps, we use a plain UD qp */
+	if (sqpn) {
+		if (mlx4_is_mfunc(dev->dev) &&
+		    (!mlx4_is_master(dev->dev) ||
+		     !(init_attr->create_flags & MLX4_IB_SRIOV_SQP))) {
+			if (init_attr->qp_type == IB_QPT_GSI)
+				qp_type = MLX4_IB_QPT_PROXY_GSI;
+			else {
+				if (mlx4_is_master(dev->dev) ||
+				    qp0_enabled_vf(dev->dev, sqpn))
+					qp_type = MLX4_IB_QPT_PROXY_SMI_OWNER;
+				else
+					qp_type = MLX4_IB_QPT_PROXY_SMI;
+			}
+		}
+		qpn = sqpn;
+		/* add extra sg entry for tunneling */
+		init_attr->cap.max_recv_sge++;
+	} else if (init_attr->create_flags & MLX4_IB_SRIOV_TUNNEL_QP) {
+		struct mlx4_ib_qp_tunnel_init_attr *tnl_init =
+			container_of(init_attr,
+				     struct mlx4_ib_qp_tunnel_init_attr, init_attr);
+		if ((tnl_init->proxy_qp_type != IB_QPT_SMI &&
+		     tnl_init->proxy_qp_type != IB_QPT_GSI)   ||
+		    !mlx4_is_master(dev->dev))
+			return -EINVAL;
+		if (tnl_init->proxy_qp_type == IB_QPT_GSI)
+			qp_type = MLX4_IB_QPT_TUN_GSI;
+		else if (tnl_init->slave == mlx4_master_func_num(dev->dev) ||
+			 mlx4_vf_smi_enabled(dev->dev, tnl_init->slave,
+					     tnl_init->port))
+			qp_type = MLX4_IB_QPT_TUN_SMI_OWNER;
+		else
+			qp_type = MLX4_IB_QPT_TUN_SMI;
+		/* we are definitely in the PPF here, since we are creating
+		 * tunnel QPs. base_tunnel_sqpn is therefore valid. */
+		qpn = dev->dev->phys_caps.base_tunnel_sqpn + 8 * tnl_init->slave
+			+ tnl_init->proxy_qp_type * 2 + tnl_init->port - 1;
+		sqpn = qpn;
+	}
+
+	if (!*caller_qp) {
+		if (qp_type == MLX4_IB_QPT_SMI || qp_type == MLX4_IB_QPT_GSI ||
+		    (qp_type & (MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_SMI_OWNER |
+				MLX4_IB_QPT_PROXY_GSI | MLX4_IB_QPT_TUN_SMI_OWNER))) {
+			sqp = kzalloc(sizeof (struct mlx4_ib_sqp), gfp);
+			if (!sqp)
+				return -ENOMEM;
+			qp = &sqp->qp;
+			qp->pri.vid = 0xFFFF;
+			qp->alt.vid = 0xFFFF;
+		} else {
+			qp = kzalloc(sizeof (struct mlx4_ib_qp), gfp);
+			if (!qp)
+				return -ENOMEM;
+			qp->pri.vid = 0xFFFF;
+			qp->alt.vid = 0xFFFF;
+		}
+	} else
+		qp = *caller_qp;
+
+	qp->mlx4_ib_qp_type = qp_type;
 
 	mutex_init(&qp->mutex);
 	spin_lock_init(&qp->sq.lock);
 	spin_lock_init(&qp->rq.lock);
+	INIT_LIST_HEAD(&qp->gid_list);
+	INIT_LIST_HEAD(&qp->steering_rules);
 
 	qp->state	 = IB_QPS_RESET;
 	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
 		qp->sq_signal_bits = cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
 
-	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, !!init_attr->srq, qp);
+	err = set_rq_size(dev, &init_attr->cap, !!pd->uobject, qp_has_rq(init_attr), qp);
 	if (err)
 		goto err;
 
@@ -491,7 +736,7 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err_mtt;
 
-		if (!init_attr->srq) {
+		if (qp_has_rq(init_attr)) {
 			err = mlx4_ib_db_map_user(to_mucontext(pd->uobject->context),
 						  ucmd.db_addr, &qp->db);
 			if (err)
@@ -506,19 +751,27 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (init_attr->create_flags & IB_QP_CREATE_IPOIB_UD_LSO)
 			qp->flags |= MLX4_IB_QP_LSO;
 
-		err = set_kernel_sq_size(dev, &init_attr->cap, init_attr->qp_type, qp);
+		if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
+			if (dev->steering_support ==
+			    MLX4_STEERING_MODE_DEVICE_MANAGED)
+				qp->flags |= MLX4_IB_QP_NETIF;
+			else
+				goto err;
+		}
+
+		err = set_kernel_sq_size(dev, &init_attr->cap, qp_type, qp);
 		if (err)
 			goto err;
 
-		if (!init_attr->srq) {
-			err = mlx4_db_alloc(dev->dev, &qp->db, 0);
+		if (qp_has_rq(init_attr)) {
+			err = mlx4_db_alloc(dev->dev, &qp->db, 0, gfp);
 			if (err)
 				goto err;
 
 			*qp->db.db = 0;
 		}
 
-		if (mlx4_buf_alloc(dev->dev, qp->buf_size, PAGE_SIZE * 2, &qp->buf)) {
+		if (mlx4_buf_alloc(dev->dev, qp->buf_size, PAGE_SIZE * 2, &qp->buf, gfp)) {
 			err = -ENOMEM;
 			goto err_db;
 		}
@@ -528,22 +781,47 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 		if (err)
 			goto err_buf;
 
-		err = mlx4_buf_write_mtt(dev->dev, &qp->mtt, &qp->buf);
+		err = mlx4_buf_write_mtt(dev->dev, &qp->mtt, &qp->buf, gfp);
 		if (err)
 			goto err_mtt;
 
-		qp->sq.wrid  = kmalloc(qp->sq.wqe_cnt * sizeof (u64), GFP_KERNEL);
-		qp->rq.wrid  = kmalloc(qp->rq.wqe_cnt * sizeof (u64), GFP_KERNEL);
-
+		qp->sq.wrid  = kmalloc(qp->sq.wqe_cnt * sizeof (u64), gfp);
+		qp->rq.wrid  = kmalloc(qp->rq.wqe_cnt * sizeof (u64), gfp);
 		if (!qp->sq.wrid || !qp->rq.wrid) {
 			err = -ENOMEM;
 			goto err_wrid;
 		}
 	}
 
-	err = mlx4_qp_alloc(dev->dev, sqpn, &qp->mqp);
+	if (sqpn) {
+		if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+		    MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) {
+			if (alloc_proxy_bufs(pd->device, qp)) {
+				err = -ENOMEM;
+				goto err_wrid;
+			}
+		}
+	} else {
+		/* Raw packet QPNs must be aligned to 8 bits. If not, the WQE
+		 * BlueFlame setup flow wrongly causes VLAN insertion. */
+		if (init_attr->qp_type == IB_QPT_RAW_PACKET)
+			err = mlx4_qp_reserve_range(dev->dev, 1, 1 << 8, &qpn);
+		else
+			if (qp->flags & MLX4_IB_QP_NETIF)
+				err = mlx4_ib_steer_qp_alloc(dev, 1, &qpn);
+			else
+				err = mlx4_qp_reserve_range(dev->dev, 1, 1,
+							    &qpn);
+		if (err)
+			goto err_proxy;
+	}
+
+	err = mlx4_qp_alloc(dev->dev, qpn, &qp->mqp, gfp);
 	if (err)
-		goto err_wrid;
+		goto err_qpn;
+
+	if (init_attr->qp_type == IB_QPT_XRC_TGT)
+		qp->mqp.qpn |= (1 << 23);
 
 	/*
 	 * Hardware wants QPN written in big-endian order (after
@@ -553,14 +831,24 @@ static int create_qp_common(struct mlx4_ib_dev *dev, struct ib_pd *pd,
 	qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);
 
 	qp->mqp.event = mlx4_ib_qp_event;
-
+	if (!*caller_qp)
+		*caller_qp = qp;
 	return 0;
 
+err_qpn:
+	if (!sqpn) {
+		if (qp->flags & MLX4_IB_QP_NETIF)
+			mlx4_ib_steer_qp_free(dev, qpn, 1);
+		else
+			mlx4_qp_release_range(dev->dev, qpn, 1);
+	}
+err_proxy:
+	if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
+		free_proxy_bufs(pd->device, qp);
 err_wrid:
 	if (pd->uobject) {
-		if (!init_attr->srq)
-			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context),
-					      &qp->db);
+		if (qp_has_rq(init_attr))
+			mlx4_ib_db_unmap_user(to_mucontext(pd->uobject->context), &qp->db);
 	} else {
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
@@ -576,10 +864,12 @@ err_buf:
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
 
 err_db:
-	if (!pd->uobject && !init_attr->srq)
+	if (!pd->uobject && qp_has_rq(init_attr))
 		mlx4_db_free(dev->dev, &qp->db);
 
 err:
+	if (!*caller_qp)
+		kfree(qp);
 	return err;
 }
 
@@ -598,10 +888,12 @@ static enum mlx4_qp_state to_mlx4_state(enum ib_qp_state state)
 }
 
 static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
+	__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
 {
-	if (send_cq == recv_cq)
+	if (send_cq == recv_cq) {
 		spin_lock_irq(&send_cq->lock);
-	else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
+		__acquire(&recv_cq->lock);
+	} else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
 		spin_lock_irq(&send_cq->lock);
 		spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
 	} else {
@@ -611,10 +903,12 @@ static void mlx4_ib_lock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv
 }
 
 static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *recv_cq)
+	__releases(&send_cq->lock) __releases(&recv_cq->lock)
 {
-	if (send_cq == recv_cq)
+	if (send_cq == recv_cq) {
+		__release(&recv_cq->lock);
 		spin_unlock_irq(&send_cq->lock);
-	else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
+	} else if (send_cq->mcq.cqn < recv_cq->mcq.cqn) {
 		spin_unlock(&recv_cq->lock);
 		spin_unlock_irq(&send_cq->lock);
 	} else {
@@ -623,19 +917,76 @@ static void mlx4_ib_unlock_cqs(struct mlx4_ib_cq *send_cq, struct mlx4_ib_cq *re
 	}
 }
 
+static void del_gid_entries(struct mlx4_ib_qp *qp)
+{
+	struct mlx4_ib_gid_entry *ge, *tmp;
+
+	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
+		list_del(&ge->list);
+		kfree(ge);
+	}
+}
+
+static struct mlx4_ib_pd *get_pd(struct mlx4_ib_qp *qp)
+{
+	if (qp->ibqp.qp_type == IB_QPT_XRC_TGT)
+		return to_mpd(to_mxrcd(qp->ibqp.xrcd)->pd);
+	else
+		return to_mpd(qp->ibqp.pd);
+}
+
+static void get_cqs(struct mlx4_ib_qp *qp,
+		    struct mlx4_ib_cq **send_cq, struct mlx4_ib_cq **recv_cq)
+{
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_XRC_TGT:
+		*send_cq = to_mcq(to_mxrcd(qp->ibqp.xrcd)->cq);
+		*recv_cq = *send_cq;
+		break;
+	case IB_QPT_XRC_INI:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = *send_cq;
+		break;
+	default:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = to_mcq(qp->ibqp.recv_cq);
+		break;
+	}
+}
+
 static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 			      int is_user)
 {
 	struct mlx4_ib_cq *send_cq, *recv_cq;
 
-	if (qp->state != IB_QPS_RESET)
+	if (qp->state != IB_QPS_RESET) {
 		if (mlx4_qp_modify(dev->dev, NULL, to_mlx4_state(qp->state),
 				   MLX4_QP_STATE_RST, NULL, 0, 0, &qp->mqp))
-			printk(KERN_WARNING "mlx4_ib: modify QP %06x to RESET failed.\n",
+			pr_warn("modify QP %06x to RESET failed.\n",
 			       qp->mqp.qpn);
+		if (qp->pri.smac) {
+			mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
+			qp->pri.smac = 0;
+		}
+		if (qp->alt.smac) {
+			mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
+			qp->alt.smac = 0;
+		}
+		if (qp->pri.vid < 0x1000) {
+			mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port, qp->pri.vid);
+			qp->pri.vid = 0xFFFF;
+			qp->pri.candidate_vid = 0xFFFF;
+			qp->pri.update_vid = 0;
+		}
+		if (qp->alt.vid < 0x1000) {
+			mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port, qp->alt.vid);
+			qp->alt.vid = 0xFFFF;
+			qp->alt.candidate_vid = 0xFFFF;
+			qp->alt.update_vid = 0;
+		}
+	}
 
-	send_cq = to_mcq(qp->ibqp.send_cq);
-	recv_cq = to_mcq(qp->ibqp.recv_cq);
+	get_cqs(qp, &send_cq, &recv_cq);
 
 	mlx4_ib_lock_cqs(send_cq, recv_cq);
 
@@ -651,59 +1002,117 @@ static void destroy_qp_common(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp,
 	mlx4_ib_unlock_cqs(send_cq, recv_cq);
 
 	mlx4_qp_free(dev->dev, &qp->mqp);
+
+	if (!is_sqp(dev, qp) && !is_tunnel_qp(dev, qp)) {
+		if (qp->flags & MLX4_IB_QP_NETIF)
+			mlx4_ib_steer_qp_free(dev, qp->mqp.qpn, 1);
+		else
+			mlx4_qp_release_range(dev->dev, qp->mqp.qpn, 1);
+	}
+
 	mlx4_mtt_cleanup(dev->dev, &qp->mtt);
 
 	if (is_user) {
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_ib_db_unmap_user(to_mucontext(qp->ibqp.uobject->context),
 					      &qp->db);
 		ib_umem_release(qp->umem);
 	} else {
 		kfree(qp->sq.wrid);
 		kfree(qp->rq.wrid);
+		if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+		    MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI))
+			free_proxy_bufs(&dev->ib_dev, qp);
 		mlx4_buf_free(dev->dev, qp->buf_size, &qp->buf);
-		if (!qp->ibqp.srq)
+		if (qp->rq.wqe_cnt)
 			mlx4_db_free(dev->dev, &qp->db);
 	}
+
+	del_gid_entries(qp);
+}
+
+static u32 get_sqp_num(struct mlx4_ib_dev *dev, struct ib_qp_init_attr *attr)
+{
+	/* Native or PPF */
+	if (!mlx4_is_mfunc(dev->dev) ||
+	    (mlx4_is_master(dev->dev) &&
+	     attr->create_flags & MLX4_IB_SRIOV_SQP)) {
+		return  dev->dev->phys_caps.base_sqpn +
+			(attr->qp_type == IB_QPT_SMI ? 0 : 2) +
+			attr->port_num - 1;
+	}
+	/* PF or VF -- creating proxies */
+	if (attr->qp_type == IB_QPT_SMI)
+		return dev->dev->caps.qp0_proxy[attr->port_num - 1];
+	else
+		return dev->dev->caps.qp1_proxy[attr->port_num - 1];
 }
 
 struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 				struct ib_qp_init_attr *init_attr,
 				struct ib_udata *udata)
 {
-	struct mlx4_ib_dev *dev = to_mdev(pd->device);
-	struct mlx4_ib_sqp *sqp;
-	struct mlx4_ib_qp *qp;
+	struct mlx4_ib_qp *qp = NULL;
 	int err;
+	u16 xrcdn = 0;
+	gfp_t gfp;
 
+	gfp = (init_attr->create_flags & MLX4_IB_QP_CREATE_USE_GFP_NOIO) ?
+		GFP_NOIO : GFP_KERNEL;
 	/*
-	 * We only support LSO and multicast loopback blocking, and
-	 * only for kernel UD QPs.
+	 * We only support LSO, vendor flag1, and multicast loopback blocking,
+	 * and only for kernel UD QPs.
 	 */
-	if (init_attr->create_flags & ~(IB_QP_CREATE_IPOIB_UD_LSO |
-					IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK))
+	if (init_attr->create_flags & ~(MLX4_IB_QP_LSO |
+					MLX4_IB_QP_BLOCK_MULTICAST_LOOPBACK |
+					MLX4_IB_SRIOV_TUNNEL_QP |
+					MLX4_IB_SRIOV_SQP |
+					MLX4_IB_QP_NETIF |
+					MLX4_IB_QP_CREATE_USE_GFP_NOIO))
 		return ERR_PTR(-EINVAL);
 
+	if (init_attr->create_flags & IB_QP_CREATE_NETIF_QP) {
+		if (init_attr->qp_type != IB_QPT_UD)
+			return ERR_PTR(-EINVAL);
+	}
+
 	if (init_attr->create_flags &&
-	    (pd->uobject || init_attr->qp_type != IB_QPT_UD))
+	    (udata ||
+	     ((init_attr->create_flags & ~(MLX4_IB_SRIOV_SQP | MLX4_IB_QP_CREATE_USE_GFP_NOIO)) &&
+	      init_attr->qp_type != IB_QPT_UD) ||
+	     ((init_attr->create_flags & MLX4_IB_SRIOV_SQP) &&
+	      init_attr->qp_type > IB_QPT_GSI)))
 		return ERR_PTR(-EINVAL);
 
 	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+		pd = to_mxrcd(init_attr->xrcd)->pd;
+		xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+		init_attr->send_cq = to_mxrcd(init_attr->xrcd)->cq;
+		/* fall through */
+	case IB_QPT_XRC_INI:
+		if (!(to_mdev(pd->device)->dev->caps.flags & MLX4_DEV_CAP_FLAG_XRC))
+			return ERR_PTR(-ENOSYS);
+		init_attr->recv_cq = init_attr->send_cq;
+		/* fall through */
 	case IB_QPT_RC:
 	case IB_QPT_UC:
-	case IB_QPT_UD:
-	{
-		qp = kzalloc(sizeof *qp, GFP_KERNEL);
+	case IB_QPT_RAW_PACKET:
+		qp = kzalloc(sizeof *qp, gfp);
 		if (!qp)
 			return ERR_PTR(-ENOMEM);
-
-		err = create_qp_common(dev, pd, init_attr, udata, 0, qp);
-		if (err) {
-			kfree(qp);
+		qp->pri.vid = 0xFFFF;
+		qp->alt.vid = 0xFFFF;
+		/* fall through */
+	case IB_QPT_UD:
+	{
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr,
+				       udata, 0, &qp, gfp);
+		if (err)
 			return ERR_PTR(err);
-		}
 
 		qp->ibqp.qp_num = qp->mqp.qpn;
+		qp->xrcdn = xrcdn;
 
 		break;
 	}
@@ -711,24 +1120,14 @@ struct ib_qp *mlx4_ib_create_qp(struct ib_pd *pd,
 	case IB_QPT_GSI:
 	{
 		/* Userspace is not allowed to create special QPs: */
-		if (pd->uobject)
+		if (udata)
 			return ERR_PTR(-EINVAL);
 
-		sqp = kzalloc(sizeof *sqp, GFP_KERNEL);
-		if (!sqp)
-			return ERR_PTR(-ENOMEM);
-
-		qp = &sqp->qp;
-
-		err = create_qp_common(dev, pd, init_attr, udata,
-				       dev->dev->caps.sqp_start +
-				       (init_attr->qp_type == IB_QPT_SMI ? 0 : 2) +
-				       init_attr->port_num - 1,
-				       qp);
-		if (err) {
-			kfree(sqp);
+		err = create_qp_common(to_mdev(pd->device), pd, init_attr, udata,
+				       get_sqp_num(to_mdev(pd->device), init_attr),
+				       &qp, gfp);
+		if (err)
 			return ERR_PTR(err);
-		}
 
 		qp->port	= init_attr->port_num;
 		qp->ibqp.qp_num = init_attr->qp_type == IB_QPT_SMI ? 0 : 1;
@@ -747,11 +1146,19 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
 {
 	struct mlx4_ib_dev *dev = to_mdev(qp->device);
 	struct mlx4_ib_qp *mqp = to_mqp(qp);
+	struct mlx4_ib_pd *pd;
 
 	if (is_qp0(dev, mqp))
 		mlx4_CLOSE_PORT(dev->dev, mqp->port);
 
-	destroy_qp_common(dev, mqp, !!qp->pd->uobject);
+	if (dev->qp1_proxy[mqp->port - 1] == mqp) {
+		mutex_lock(&dev->qp1_proxy_lock[mqp->port - 1]);
+		dev->qp1_proxy[mqp->port - 1] = NULL;
+		mutex_unlock(&dev->qp1_proxy_lock[mqp->port - 1]);
+	}
+
+	pd = get_pd(mqp);
+	destroy_qp_common(dev, mqp, !!pd->ibpd.uobject);
 
 	if (is_sqp(dev, mqp))
 		kfree(to_msqp(mqp));
@@ -761,15 +1168,27 @@ int mlx4_ib_destroy_qp(struct ib_qp *qp)
 	return 0;
 }
 
-static int to_mlx4_st(enum ib_qp_type type)
+static int to_mlx4_st(struct mlx4_ib_dev *dev, enum mlx4_ib_qp_type type)
 {
 	switch (type) {
-	case IB_QPT_RC:		return MLX4_QP_ST_RC;
-	case IB_QPT_UC:		return MLX4_QP_ST_UC;
-	case IB_QPT_UD:		return MLX4_QP_ST_UD;
-	case IB_QPT_SMI:
-	case IB_QPT_GSI:	return MLX4_QP_ST_MLX;
-	default:		return -1;
+	case MLX4_IB_QPT_RC:		return MLX4_QP_ST_RC;
+	case MLX4_IB_QPT_UC:		return MLX4_QP_ST_UC;
+	case MLX4_IB_QPT_UD:		return MLX4_QP_ST_UD;
+	case MLX4_IB_QPT_XRC_INI:
+	case MLX4_IB_QPT_XRC_TGT:	return MLX4_QP_ST_XRC;
+	case MLX4_IB_QPT_SMI:
+	case MLX4_IB_QPT_GSI:
+	case MLX4_IB_QPT_RAW_PACKET:	return MLX4_QP_ST_MLX;
+
+	case MLX4_IB_QPT_PROXY_SMI_OWNER:
+	case MLX4_IB_QPT_TUN_SMI_OWNER:	return (mlx4_is_mfunc(dev->dev) ?
+						MLX4_QP_ST_MLX : -1);
+	case MLX4_IB_QPT_PROXY_SMI:
+	case MLX4_IB_QPT_TUN_SMI:
+	case MLX4_IB_QPT_PROXY_GSI:
+	case MLX4_IB_QPT_TUN_GSI:	return (mlx4_is_mfunc(dev->dev) ?
+						MLX4_QP_ST_UD : -1);
+	default:			return -1;
 	}
 }
 
@@ -819,9 +1238,17 @@ static void mlx4_set_sched(struct mlx4_qp_path *path, u8 port)
 	path->sched_queue = (path->sched_queue & 0xbf) | ((port - 1) << 6);
 }
 
-static int mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah,
-			 struct mlx4_qp_path *path, u8 port)
+static int _mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah,
+			  u64 smac, u16 vlan_tag, struct mlx4_qp_path *path,
+			  struct mlx4_roce_smac_vlan_info *smac_info, u8 port)
 {
+	int is_eth = rdma_port_get_link_layer(&dev->ib_dev, port) ==
+		IB_LINK_LAYER_ETHERNET;
+	int vidx;
+	int smac_index;
+	int err;
+
+
 	path->grh_mylmc     = ah->src_path_bits & 0x7f;
 	path->rlid	    = cpu_to_be16(ah->dlid);
 	if (ah->static_rate) {
@@ -831,11 +1258,10 @@ static int mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah,
 			--path->static_rate;
 	} else
 		path->static_rate = 0;
-	path->counter_index = 0xff;
 
 	if (ah->ah_flags & IB_AH_GRH) {
 		if (ah->grh.sgid_index >= dev->dev->caps.gid_table_len[port]) {
-			printk(KERN_ERR "sgid_index (%u) too large. max is %d\n",
+			pr_err("sgid_index (%u) too large. max is %d\n",
 			       ah->grh.sgid_index, dev->dev->caps.gid_table_len[port] - 1);
 			return -1;
 		}
@@ -849,9 +1275,146 @@ static int mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_ah_attr *ah,
 		memcpy(path->rgid, ah->grh.dgid.raw, 16);
 	}
 
-	path->sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE |
-		((port - 1) << 6) | ((ah->sl & 0xf) << 2);
+	if (is_eth) {
+		if (!(ah->ah_flags & IB_AH_GRH))
+			return -1;
+
+		path->sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE |
+			((port - 1) << 6) | ((ah->sl & 7) << 3);
+
+		path->feup |= MLX4_FEUP_FORCE_ETH_UP;
+		if (vlan_tag < 0x1000) {
+			if (smac_info->vid < 0x1000) {
+				/* both valid vlan ids */
+				if (smac_info->vid != vlan_tag) {
+					/* different VIDs.  unreg old and reg new */
+					err = mlx4_register_vlan(dev->dev, port, vlan_tag, &vidx);
+					if (err)
+						return err;
+					smac_info->candidate_vid = vlan_tag;
+					smac_info->candidate_vlan_index = vidx;
+					smac_info->candidate_vlan_port = port;
+					smac_info->update_vid = 1;
+					path->vlan_index = vidx;
+				} else {
+					path->vlan_index = smac_info->vlan_index;
+				}
+			} else {
+				/* no current vlan tag in qp */
+				err = mlx4_register_vlan(dev->dev, port, vlan_tag, &vidx);
+				if (err)
+					return err;
+				smac_info->candidate_vid = vlan_tag;
+				smac_info->candidate_vlan_index = vidx;
+				smac_info->candidate_vlan_port = port;
+				smac_info->update_vid = 1;
+				path->vlan_index = vidx;
+			}
+			path->feup |= MLX4_FVL_FORCE_ETH_VLAN;
+			path->fl = 1 << 6;
+		} else {
+			/* have current vlan tag. unregister it at modify-qp success */
+			if (smac_info->vid < 0x1000) {
+				smac_info->candidate_vid = 0xFFFF;
+				smac_info->update_vid = 1;
+			}
+		}
+
+		/* get smac_index for RoCE use.
+		 * If no smac was yet assigned, register one.
+		 * If one was already assigned, but the new mac differs,
+		 * unregister the old one and register the new one.
+		*/
+		if (!smac_info->smac || smac_info->smac != smac) {
+			/* register candidate now, unreg if needed, after success */
+			smac_index = mlx4_register_mac(dev->dev, port, smac);
+			if (smac_index >= 0) {
+				smac_info->candidate_smac_index = smac_index;
+				smac_info->candidate_smac = smac;
+				smac_info->candidate_smac_port = port;
+			} else {
+				return -EINVAL;
+			}
+		} else {
+			smac_index = smac_info->smac_index;
+		}
+
+		memcpy(path->dmac, ah->dmac, 6);
+		path->ackto = MLX4_IB_LINK_TYPE_ETH;
+		/* put MAC table smac index for IBoE */
+		path->grh_mylmc = (u8) (smac_index) | 0x80;
+	} else {
+		path->sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE |
+			((port - 1) << 6) | ((ah->sl & 0xf) << 2);
+	}
+
+	return 0;
+}
+
+static int mlx4_set_path(struct mlx4_ib_dev *dev, const struct ib_qp_attr *qp,
+			 enum ib_qp_attr_mask qp_attr_mask,
+			 struct mlx4_ib_qp *mqp,
+			 struct mlx4_qp_path *path, u8 port)
+{
+	return _mlx4_set_path(dev, &qp->ah_attr,
+			      mlx4_mac_to_u64((u8 *)qp->smac),
+			      (qp_attr_mask & IB_QP_VID) ? qp->vlan_id : 0xffff,
+			      path, &mqp->pri, port);
+}
+
+static int mlx4_set_alt_path(struct mlx4_ib_dev *dev,
+			     const struct ib_qp_attr *qp,
+			     enum ib_qp_attr_mask qp_attr_mask,
+			     struct mlx4_ib_qp *mqp,
+			     struct mlx4_qp_path *path, u8 port)
+{
+	return _mlx4_set_path(dev, &qp->alt_ah_attr,
+			      mlx4_mac_to_u64((u8 *)qp->alt_smac),
+			      (qp_attr_mask & IB_QP_ALT_VID) ?
+			      qp->alt_vlan_id : 0xffff,
+			      path, &mqp->alt, port);
+}
+
+static void update_mcg_macs(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp)
+{
+	struct mlx4_ib_gid_entry *ge, *tmp;
+
+	list_for_each_entry_safe(ge, tmp, &qp->gid_list, list) {
+		if (!ge->added && mlx4_ib_add_mc(dev, qp, &ge->gid)) {
+			ge->added = 1;
+			ge->port = qp->port;
+		}
+	}
+}
+
+static int handle_eth_ud_smac_index(struct mlx4_ib_dev *dev, struct mlx4_ib_qp *qp, u8 *smac,
+				    struct mlx4_qp_context *context)
+{
+	struct net_device *ndev;
+	u64 u64_mac;
+	int smac_index;
 
+
+	ndev = dev->iboe.netdevs[qp->port - 1];
+	if (ndev) {
+		smac = ndev->dev_addr;
+		u64_mac = mlx4_mac_to_u64(smac);
+	} else {
+		u64_mac = dev->dev->caps.def_mac[qp->port];
+	}
+
+	context->pri_path.sched_queue = MLX4_IB_DEFAULT_SCHED_QUEUE | ((qp->port - 1) << 6);
+	if (!qp->pri.smac) {
+		smac_index = mlx4_register_mac(dev->dev, qp->port, u64_mac);
+		if (smac_index >= 0) {
+			qp->pri.candidate_smac_index = smac_index;
+			qp->pri.candidate_smac = u64_mac;
+			qp->pri.candidate_smac_port = qp->port;
+			context->pri_path.grh_mylmc = 0x80 | (u8) smac_index;
+		} else {
+			return -ENOENT;
+		}
+	}
 	return 0;
 }
 
@@ -861,9 +1424,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 {
 	struct mlx4_ib_dev *dev = to_mdev(ibqp->device);
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
+	struct mlx4_ib_pd *pd;
+	struct mlx4_ib_cq *send_cq, *recv_cq;
 	struct mlx4_qp_context *context;
 	enum mlx4_qp_optpar optpar = 0;
 	int sqd_event;
+	int steer_qp = 0;
 	int err = -EINVAL;
 
 	context = kzalloc(sizeof *context, GFP_KERNEL);
@@ -871,8 +1437,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		return -ENOMEM;
 
 	context->flags = cpu_to_be32((to_mlx4_state(new_state) << 28) |
-				     (to_mlx4_st(ibqp->qp_type) << 16));
-	context->flags     |= cpu_to_be32(1 << 8); /* DE? */
+				     (to_mlx4_st(dev, qp->mlx4_ib_qp_type) << 16));
 
 	if (!(attr_mask & IB_QP_PATH_MIG_STATE))
 		context->flags |= cpu_to_be32(MLX4_QP_PM_MIGRATED << 11);
@@ -893,15 +1458,17 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 
 	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI)
 		context->mtu_msgmax = (IB_MTU_4096 << 5) | 11;
+	else if (ibqp->qp_type == IB_QPT_RAW_PACKET)
+		context->mtu_msgmax = (MLX4_RAW_QP_MTU << 5) | MLX4_RAW_QP_MSGMAX;
 	else if (ibqp->qp_type == IB_QPT_UD) {
 		if (qp->flags & MLX4_IB_QP_LSO)
 			context->mtu_msgmax = (IB_MTU_4096 << 5) |
 					      ilog2(dev->dev->caps.max_gso_sz);
 		else
-			context->mtu_msgmax = (IB_MTU_4096 << 5) | 11;
+			context->mtu_msgmax = (IB_MTU_4096 << 5) | 12;
 	} else if (attr_mask & IB_QP_PATH_MTU) {
 		if (attr->path_mtu < IB_MTU_256 || attr->path_mtu > IB_MTU_4096) {
-			printk(KERN_ERR "path MTU (%u) is invalid\n",
+			pr_err("path MTU (%u) is invalid\n",
 			       attr->path_mtu);
 			goto out;
 		}
@@ -917,8 +1484,12 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		context->sq_size_stride = ilog2(qp->sq.wqe_cnt) << 3;
 	context->sq_size_stride |= qp->sq.wqe_shift - 4;
 
-	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT) {
 		context->sq_size_stride |= !!qp->sq_no_prefetch << 7;
+		context->xrcd = cpu_to_be32((u32) qp->xrcdn);
+		if (ibqp->qp_type == IB_QPT_RAW_PACKET)
+			context->param3 |= cpu_to_be32(1 << 30);
+	}
 
 	if (qp->ibqp.uobject)
 		context->usr_page = cpu_to_be32(to_mucontext(ibqp->uobject->context)->uar.index);
@@ -936,14 +1507,31 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		}
 	}
 
+	if (cur_state == IB_QPS_INIT && new_state == IB_QPS_RTR) {
+		if (dev->counters[qp->port - 1] != -1) {
+			context->pri_path.counter_index =
+						dev->counters[qp->port - 1];
+			optpar |= MLX4_QP_OPTPAR_COUNTER_INDEX;
+		} else
+			context->pri_path.counter_index = 0xff;
+
+		if (qp->flags & MLX4_IB_QP_NETIF) {
+			mlx4_ib_steer_qp_reg(dev, qp, 1);
+			steer_qp = 1;
+		}
+	}
+
 	if (attr_mask & IB_QP_PKEY_INDEX) {
+		if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
+			context->pri_path.disable_pkey_check = 0x40;
 		context->pri_path.pkey_index = attr->pkey_index;
 		optpar |= MLX4_QP_OPTPAR_PKEY_INDEX;
 	}
 
 	if (attr_mask & IB_QP_AV) {
-		if (mlx4_set_path(dev, &attr->ah_attr, &context->pri_path,
-				  attr_mask & IB_QP_PORT ? attr->port_num : qp->port))
+		if (mlx4_set_path(dev, attr, attr_mask, qp, &context->pri_path,
+				  attr_mask & IB_QP_PORT ?
+				  attr->port_num : qp->port))
 			goto out;
 
 		optpar |= (MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH |
@@ -951,7 +1539,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	}
 
 	if (attr_mask & IB_QP_TIMEOUT) {
-		context->pri_path.ackto = attr->timeout << 3;
+		context->pri_path.ackto |= attr->timeout << 3;
 		optpar |= MLX4_QP_OPTPAR_ACK_TIMEOUT;
 	}
 
@@ -964,8 +1552,9 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		    dev->dev->caps.pkey_table_len[attr->alt_port_num])
 			goto out;
 
-		if (mlx4_set_path(dev, &attr->alt_ah_attr, &context->alt_path,
-				  attr->alt_port_num))
+		if (mlx4_set_alt_path(dev, attr, attr_mask, qp,
+				      &context->alt_path,
+				      attr->alt_port_num))
 			goto out;
 
 		context->alt_path.pkey_index = attr->alt_pkey_index;
@@ -973,8 +1562,16 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		optpar |= MLX4_QP_OPTPAR_ALT_ADDR_PATH;
 	}
 
-	context->pd	    = cpu_to_be32(to_mpd(ibqp->pd)->pdn);
-	context->params1    = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
+	pd = get_pd(qp);
+	get_cqs(qp, &send_cq, &recv_cq);
+	context->pd       = cpu_to_be32(pd->pdn);
+	context->cqn_send = cpu_to_be32(send_cq->mcq.cqn);
+	context->cqn_recv = cpu_to_be32(recv_cq->mcq.cqn);
+	context->params1  = cpu_to_be32(MLX4_IB_ACK_REQ_FREQ << 28);
+
+	/* Set "fast registration enabled" for all kernel QPs */
+	if (!qp->ibqp.uobject)
+		context->params1 |= cpu_to_be32(1 << 11);
 
 	if (attr_mask & IB_QP_RNR_RETRY) {
 		context->params1 |= cpu_to_be32(attr->rnr_retry << 13);
@@ -996,8 +1593,6 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_SQ_PSN)
 		context->next_send_psn = cpu_to_be32(attr->sq_psn);
 
-	context->cqn_send = cpu_to_be32(to_mcq(ibqp->send_cq)->mcq.cqn);
-
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
 		if (attr->max_dest_rd_atomic)
 			context->params2 |=
@@ -1020,36 +1615,92 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (attr_mask & IB_QP_RQ_PSN)
 		context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
 
-	context->cqn_recv = cpu_to_be32(to_mcq(ibqp->recv_cq)->mcq.cqn);
-
+	/* proxy and tunnel qp qkeys will be changed in modify-qp wrappers */
 	if (attr_mask & IB_QP_QKEY) {
-		context->qkey = cpu_to_be32(attr->qkey);
+		if (qp->mlx4_ib_qp_type &
+		    (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))
+			context->qkey = cpu_to_be32(IB_QP_SET_QKEY);
+		else {
+			if (mlx4_is_mfunc(dev->dev) &&
+			    !(qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV) &&
+			    (attr->qkey & MLX4_RESERVED_QKEY_MASK) ==
+			    MLX4_RESERVED_QKEY_BASE) {
+				pr_err("Cannot use reserved QKEY"
+				       " 0x%x (range 0xffff0000..0xffffffff"
+				       " is reserved)\n", attr->qkey);
+				err = -EINVAL;
+				goto out;
+			}
+			context->qkey = cpu_to_be32(attr->qkey);
+		}
 		optpar |= MLX4_QP_OPTPAR_Q_KEY;
 	}
 
 	if (ibqp->srq)
 		context->srqn = cpu_to_be32(1 << 24 | to_msrq(ibqp->srq)->msrq.srqn);
 
-	if (!ibqp->srq && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+	if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
 		context->db_rec_addr = cpu_to_be64(qp->db.dma);
 
 	if (cur_state == IB_QPS_INIT &&
 	    new_state == IB_QPS_RTR  &&
 	    (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI ||
-	     ibqp->qp_type == IB_QPT_UD)) {
+	     ibqp->qp_type == IB_QPT_UD ||
+	     ibqp->qp_type == IB_QPT_RAW_PACKET)) {
 		context->pri_path.sched_queue = (qp->port - 1) << 6;
-		if (is_qp0(dev, qp))
+		if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI ||
+		    qp->mlx4_ib_qp_type &
+		    (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER)) {
 			context->pri_path.sched_queue |= MLX4_IB_DEFAULT_QP0_SCHED_QUEUE;
-		else
+			if (qp->mlx4_ib_qp_type != MLX4_IB_QPT_SMI)
+				context->pri_path.fl = 0x80;
+		} else {
+			if (qp->mlx4_ib_qp_type & MLX4_IB_QPT_ANY_SRIOV)
+				context->pri_path.fl = 0x80;
 			context->pri_path.sched_queue |= MLX4_IB_DEFAULT_SCHED_QUEUE;
+		}
+		if (rdma_port_get_link_layer(&dev->ib_dev, qp->port) ==
+		    IB_LINK_LAYER_ETHERNET) {
+			if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_GSI ||
+			    qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI)
+				context->pri_path.feup = 1 << 7; /* don't fsm */
+			/* handle smac_index */
+			if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_UD ||
+			    qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI ||
+			    qp->mlx4_ib_qp_type == MLX4_IB_QPT_TUN_GSI) {
+				err = handle_eth_ud_smac_index(dev, qp, (u8 *)attr->smac, context);
+				if (err)
+					return -EINVAL;
+				if (qp->mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_GSI)
+					dev->qp1_proxy[qp->port - 1] = qp;
+			}
+		}
+	}
+
+	if (qp->ibqp.qp_type == IB_QPT_RAW_PACKET)
+		context->pri_path.ackto = (context->pri_path.ackto & 0xf8) |
+					MLX4_IB_LINK_TYPE_ETH;
+
+	if (ibqp->qp_type == IB_QPT_UD && (new_state == IB_QPS_RTR)) {
+		int is_eth = rdma_port_get_link_layer(
+				&dev->ib_dev, qp->port) ==
+				IB_LINK_LAYER_ETHERNET;
+		if (is_eth) {
+			context->pri_path.ackto = MLX4_IB_LINK_TYPE_ETH;
+			optpar |= MLX4_QP_OPTPAR_PRIMARY_ADDR_PATH;
+		}
 	}
 
+
 	if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD	&&
 	    attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY && attr->en_sqd_async_notify)
 		sqd_event = 1;
 	else
 		sqd_event = 0;
 
+	if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->rlkey |= (1 << 4);
+
 	/*
 	 * Before passing a kernel QP to the HW, make sure that the
 	 * ownership bits of the send queue are set and the SQ
@@ -1082,8 +1733,10 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 		qp->atomic_rd_en = attr->qp_access_flags;
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
 		qp->resp_depth = attr->max_dest_rd_atomic;
-	if (attr_mask & IB_QP_PORT)
+	if (attr_mask & IB_QP_PORT) {
 		qp->port = attr->port_num;
+		update_mcg_macs(dev, qp);
+	}
 	if (attr_mask & IB_QP_ALT_PATH)
 		qp->alt_port = attr->alt_port_num;
 
@@ -1097,7 +1750,7 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	if (is_qp0(dev, qp)) {
 		if (cur_state != IB_QPS_RTR && new_state == IB_QPS_RTR)
 			if (mlx4_INIT_PORT(dev->dev, qp->port))
-				printk(KERN_WARNING "INIT_PORT failed for port %d\n",
+				pr_warn("INIT_PORT failed for port %d\n",
 				       qp->port);
 
 		if (cur_state != IB_QPS_RESET && cur_state != IB_QPS_ERR &&
@@ -1109,23 +1762,113 @@ static int __mlx4_ib_modify_qp(struct ib_qp *ibqp,
 	 * If we moved a kernel QP to RESET, clean up all old CQ
 	 * entries and reinitialize the QP.
 	 */
-	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
-		mlx4_ib_cq_clean(to_mcq(ibqp->recv_cq), qp->mqp.qpn,
-				 ibqp->srq ? to_msrq(ibqp->srq): NULL);
-		if (ibqp->send_cq != ibqp->recv_cq)
-			mlx4_ib_cq_clean(to_mcq(ibqp->send_cq), qp->mqp.qpn, NULL);
+	if (new_state == IB_QPS_RESET) {
+		if (!ibqp->uobject) {
+			mlx4_ib_cq_clean(recv_cq, qp->mqp.qpn,
+					 ibqp->srq ? to_msrq(ibqp->srq) : NULL);
+			if (send_cq != recv_cq)
+				mlx4_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
+
+			qp->rq.head = 0;
+			qp->rq.tail = 0;
+			qp->sq.head = 0;
+			qp->sq.tail = 0;
+			qp->sq_next_wqe = 0;
+			if (qp->rq.wqe_cnt)
+				*qp->db.db  = 0;
+
+			if (qp->flags & MLX4_IB_QP_NETIF)
+				mlx4_ib_steer_qp_reg(dev, qp, 0);
+		}
+		if (qp->pri.smac) {
+			mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
+			qp->pri.smac = 0;
+		}
+		if (qp->alt.smac) {
+			mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
+			qp->alt.smac = 0;
+		}
+		if (qp->pri.vid < 0x1000) {
+			mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port, qp->pri.vid);
+			qp->pri.vid = 0xFFFF;
+			qp->pri.candidate_vid = 0xFFFF;
+			qp->pri.update_vid = 0;
+		}
 
-		qp->rq.head = 0;
-		qp->rq.tail = 0;
-		qp->sq.head = 0;
-		qp->sq.tail = 0;
-		qp->sq_next_wqe = 0;
-		if (!ibqp->srq)
-			*qp->db.db  = 0;
+		if (qp->alt.vid < 0x1000) {
+			mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port, qp->alt.vid);
+			qp->alt.vid = 0xFFFF;
+			qp->alt.candidate_vid = 0xFFFF;
+			qp->alt.update_vid = 0;
+		}
 	}
-
 out:
+	if (err && steer_qp)
+		mlx4_ib_steer_qp_reg(dev, qp, 0);
 	kfree(context);
+	if (qp->pri.candidate_smac) {
+		if (err) {
+			mlx4_unregister_mac(dev->dev, qp->pri.candidate_smac_port, qp->pri.candidate_smac);
+		} else {
+			if (qp->pri.smac)
+				mlx4_unregister_mac(dev->dev, qp->pri.smac_port, qp->pri.smac);
+			qp->pri.smac = qp->pri.candidate_smac;
+			qp->pri.smac_index = qp->pri.candidate_smac_index;
+			qp->pri.smac_port = qp->pri.candidate_smac_port;
+		}
+		qp->pri.candidate_smac = 0;
+		qp->pri.candidate_smac_index = 0;
+		qp->pri.candidate_smac_port = 0;
+	}
+	if (qp->alt.candidate_smac) {
+		if (err) {
+			mlx4_unregister_mac(dev->dev, qp->alt.candidate_smac_port, qp->alt.candidate_smac);
+		} else {
+			if (qp->alt.smac)
+				mlx4_unregister_mac(dev->dev, qp->alt.smac_port, qp->alt.smac);
+			qp->alt.smac = qp->alt.candidate_smac;
+			qp->alt.smac_index = qp->alt.candidate_smac_index;
+			qp->alt.smac_port = qp->alt.candidate_smac_port;
+		}
+		qp->alt.candidate_smac = 0;
+		qp->alt.candidate_smac_index = 0;
+		qp->alt.candidate_smac_port = 0;
+	}
+
+	if (qp->pri.update_vid) {
+		if (err) {
+			if (qp->pri.candidate_vid < 0x1000)
+				mlx4_unregister_vlan(dev->dev, qp->pri.candidate_vlan_port,
+						     qp->pri.candidate_vid);
+		} else {
+			if (qp->pri.vid < 0x1000)
+				mlx4_unregister_vlan(dev->dev, qp->pri.vlan_port,
+						     qp->pri.vid);
+			qp->pri.vid = qp->pri.candidate_vid;
+			qp->pri.vlan_port = qp->pri.candidate_vlan_port;
+			qp->pri.vlan_index =  qp->pri.candidate_vlan_index;
+		}
+		qp->pri.candidate_vid = 0xFFFF;
+		qp->pri.update_vid = 0;
+	}
+
+	if (qp->alt.update_vid) {
+		if (err) {
+			if (qp->alt.candidate_vid < 0x1000)
+				mlx4_unregister_vlan(dev->dev, qp->alt.candidate_vlan_port,
+						     qp->alt.candidate_vid);
+		} else {
+			if (qp->alt.vid < 0x1000)
+				mlx4_unregister_vlan(dev->dev, qp->alt.vlan_port,
+						     qp->alt.vid);
+			qp->alt.vid = qp->alt.candidate_vid;
+			qp->alt.vlan_port = qp->alt.candidate_vlan_port;
+			qp->alt.vlan_index =  qp->alt.candidate_vlan_index;
+		}
+		qp->alt.candidate_vid = 0xFFFF;
+		qp->alt.update_vid = 0;
+	}
+
 	return err;
 }
 
@@ -1136,33 +1879,69 @@ int mlx4_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	struct mlx4_ib_qp *qp = to_mqp(ibqp);
 	enum ib_qp_state cur_state, new_state;
 	int err = -EINVAL;
-
+	int ll;
 	mutex_lock(&qp->mutex);
 
 	cur_state = attr_mask & IB_QP_CUR_STATE ? attr->cur_qp_state : qp->state;
 	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
 
-	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask))
+	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
+		ll = IB_LINK_LAYER_UNSPECIFIED;
+	} else {
+		int port = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
+		ll = rdma_port_get_link_layer(&dev->ib_dev, port);
+	}
+
+	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
+				attr_mask, ll)) {
+		pr_debug("qpn 0x%x: invalid attribute mask specified "
+			 "for transition %d to %d. qp_type %d,"
+			 " attr_mask 0x%x\n",
+			 ibqp->qp_num, cur_state, new_state,
+			 ibqp->qp_type, attr_mask);
 		goto out;
+	}
 
 	if ((attr_mask & IB_QP_PORT) &&
-	    (attr->port_num == 0 || attr->port_num > dev->dev->caps.num_ports)) {
+	    (attr->port_num == 0 || attr->port_num > dev->num_ports)) {
+		pr_debug("qpn 0x%x: invalid port number (%d) specified "
+			 "for transition %d to %d. qp_type %d\n",
+			 ibqp->qp_num, attr->port_num, cur_state,
+			 new_state, ibqp->qp_type);
 		goto out;
 	}
 
+	if ((attr_mask & IB_QP_PORT) && (ibqp->qp_type == IB_QPT_RAW_PACKET) &&
+	    (rdma_port_get_link_layer(&dev->ib_dev, attr->port_num) !=
+	     IB_LINK_LAYER_ETHERNET))
+		goto out;
+
 	if (attr_mask & IB_QP_PKEY_INDEX) {
 		int p = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
-		if (attr->pkey_index >= dev->dev->caps.pkey_table_len[p])
+		if (attr->pkey_index >= dev->dev->caps.pkey_table_len[p]) {
+			pr_debug("qpn 0x%x: invalid pkey index (%d) specified "
+				 "for transition %d to %d. qp_type %d\n",
+				 ibqp->qp_num, attr->pkey_index, cur_state,
+				 new_state, ibqp->qp_type);
 			goto out;
+		}
 	}
 
 	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
 	    attr->max_rd_atomic > dev->dev->caps.max_qp_init_rdma) {
+		pr_debug("qpn 0x%x: max_rd_atomic (%d) too large. "
+			 "Transition %d to %d. qp_type %d\n",
+			 ibqp->qp_num, attr->max_rd_atomic, cur_state,
+			 new_state, ibqp->qp_type);
 		goto out;
 	}
 
 	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
 	    attr->max_dest_rd_atomic > dev->dev->caps.max_qp_dest_rdma) {
+		pr_debug("qpn 0x%x: max_dest_rd_atomic (%d) too large. "
+			 "Transition %d to %d. qp_type %d\n",
+			 ibqp->qp_num, attr->max_dest_rd_atomic, cur_state,
+			 new_state, ibqp->qp_type);
 		goto out;
 	}
 
@@ -1178,47 +1957,230 @@ out:
 	return err;
 }
 
+static int vf_get_qp0_qkey(struct mlx4_dev *dev, int qpn, u32 *qkey)
+{
+	int i;
+	for (i = 0; i < dev->caps.num_ports; i++) {
+		if (qpn == dev->caps.qp0_proxy[i] ||
+		    qpn == dev->caps.qp0_tunnel[i]) {
+			*qkey = dev->caps.qp0_qkey[i];
+			return 0;
+		}
+	}
+	return -EINVAL;
+}
+
+static int build_sriov_qp0_header(struct mlx4_ib_sqp *sqp,
+				  struct ib_send_wr *wr,
+				  void *wqe, unsigned *mlx_seg_len)
+{
+	struct mlx4_ib_dev *mdev = to_mdev(sqp->qp.ibqp.device);
+	struct ib_device *ib_dev = &mdev->ib_dev;
+	struct mlx4_wqe_mlx_seg *mlx = wqe;
+	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
+	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	u16 pkey;
+	u32 qkey;
+	int send_size;
+	int header_size;
+	int spc;
+	int i;
+
+	if (wr->opcode != IB_WR_SEND)
+		return -EINVAL;
+
+	send_size = 0;
+
+	for (i = 0; i < wr->num_sge; ++i)
+		send_size += wr->sg_list[i].length;
+
+	/* for proxy-qp0 sends, need to add in size of tunnel header */
+	/* for tunnel-qp0 sends, tunnel header is already in s/g list */
+	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER)
+		send_size += sizeof (struct mlx4_ib_tunnel_header);
+
+	ib_ud_header_init(send_size, 1, 0, 0, 0, 0, &sqp->ud_header);
+
+	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_PROXY_SMI_OWNER) {
+		sqp->ud_header.lrh.service_level =
+			be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
+		sqp->ud_header.lrh.destination_lid =
+			cpu_to_be16(ah->av.ib.g_slid & 0x7f);
+		sqp->ud_header.lrh.source_lid =
+			cpu_to_be16(ah->av.ib.g_slid & 0x7f);
+	}
+
+	mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
+
+	/* force loopback */
+	mlx->flags |= cpu_to_be32(MLX4_WQE_MLX_VL15 | 0x1 | MLX4_WQE_MLX_SLR);
+	mlx->rlid = sqp->ud_header.lrh.destination_lid;
+
+	sqp->ud_header.lrh.virtual_lane    = 0;
+	sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
+	ib_get_cached_pkey(ib_dev, sqp->qp.port, 0, &pkey);
+	sqp->ud_header.bth.pkey = cpu_to_be16(pkey);
+	if (sqp->qp.mlx4_ib_qp_type == MLX4_IB_QPT_TUN_SMI_OWNER)
+		sqp->ud_header.bth.destination_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+	else
+		sqp->ud_header.bth.destination_qpn =
+			cpu_to_be32(mdev->dev->caps.qp0_tunnel[sqp->qp.port - 1]);
+
+	sqp->ud_header.bth.psn = cpu_to_be32((sqp->send_psn++) & ((1 << 24) - 1));
+	if (mlx4_is_master(mdev->dev)) {
+		if (mlx4_get_parav_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
+			return -EINVAL;
+	} else {
+		if (vf_get_qp0_qkey(mdev->dev, sqp->qp.mqp.qpn, &qkey))
+			return -EINVAL;
+	}
+	sqp->ud_header.deth.qkey = cpu_to_be32(qkey);
+	sqp->ud_header.deth.source_qpn = cpu_to_be32(sqp->qp.mqp.qpn);
+
+	sqp->ud_header.bth.opcode        = IB_OPCODE_UD_SEND_ONLY;
+	sqp->ud_header.immediate_present = 0;
+
+	header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);
+
+	/*
+	 * Inline data segments may not cross a 64 byte boundary.  If
+	 * our UD header is bigger than the space available up to the
+	 * next 64 byte boundary in the WQE, use two inline data
+	 * segments to hold the UD header.
+	 */
+	spc = MLX4_INLINE_ALIGN -
+	      ((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1));
+	if (header_size <= spc) {
+		inl->byte_count = cpu_to_be32(1 << 31 | header_size);
+		memcpy(inl + 1, sqp->header_buf, header_size);
+		i = 1;
+	} else {
+		inl->byte_count = cpu_to_be32(1 << 31 | spc);
+		memcpy(inl + 1, sqp->header_buf, spc);
+
+		inl = (void *) (inl + 1) + spc;
+		memcpy(inl + 1, sqp->header_buf + spc, header_size - spc);
+		/*
+		 * Need a barrier here to make sure all the data is
+		 * visible before the byte_count field is set.
+		 * Otherwise the HCA prefetcher could grab the 64-byte
+		 * chunk with this inline segment and get a valid (!=
+		 * 0xffffffff) byte count but stale data, and end up
+		 * generating a packet with bad headers.
+		 *
+		 * The first inline segment's byte_count field doesn't
+		 * need a barrier, because it comes after a
+		 * control/MLX segment and therefore is at an offset
+		 * of 16 mod 64.
+		 */
+		wmb();
+		inl->byte_count = cpu_to_be32(1 << 31 | (header_size - spc));
+		i = 2;
+	}
+
+	*mlx_seg_len =
+	ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + header_size, 16);
+	return 0;
+}
+
 static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 			    void *wqe, unsigned *mlx_seg_len)
 {
-	struct ib_device *ib_dev = &to_mdev(sqp->qp.ibqp.device)->ib_dev;
+	struct ib_device *ib_dev = sqp->qp.ibqp.device;
 	struct mlx4_wqe_mlx_seg *mlx = wqe;
+	struct mlx4_wqe_ctrl_seg *ctrl = wqe;
 	struct mlx4_wqe_inline_seg *inl = wqe + sizeof *mlx;
 	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	union ib_gid sgid;
 	u16 pkey;
 	int send_size;
 	int header_size;
 	int spc;
 	int i;
+	int err = 0;
+	u16 vlan = 0xffff;
+	bool is_eth;
+	bool is_vlan = false;
+	bool is_grh;
 
 	send_size = 0;
 	for (i = 0; i < wr->num_sge; ++i)
 		send_size += wr->sg_list[i].length;
 
-	ib_ud_header_init(send_size, mlx4_ib_ah_grh_present(ah), &sqp->ud_header);
+	is_eth = rdma_port_get_link_layer(sqp->qp.ibqp.device, sqp->qp.port) == IB_LINK_LAYER_ETHERNET;
+	is_grh = mlx4_ib_ah_grh_present(ah);
+	if (is_eth) {
+		if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
+			/* When multi-function is enabled, the ib_core gid
+			 * indexes don't necessarily match the hw ones, so
+			 * we must use our own cache */
+			err = mlx4_get_roce_gid_from_slave(to_mdev(ib_dev)->dev,
+							   be32_to_cpu(ah->av.ib.port_pd) >> 24,
+							   ah->av.ib.gid_index, &sgid.raw[0]);
+			if (err)
+				return err;
+		} else  {
+			err = ib_get_cached_gid(ib_dev,
+						be32_to_cpu(ah->av.ib.port_pd) >> 24,
+						ah->av.ib.gid_index, &sgid);
+			if (err)
+				return err;
+		}
+
+		if (ah->av.eth.vlan != cpu_to_be16(0xffff)) {
+			vlan = be16_to_cpu(ah->av.eth.vlan) & 0x0fff;
+			is_vlan = 1;
+		}
+	}
+	ib_ud_header_init(send_size, !is_eth, is_eth, is_vlan, is_grh, 0, &sqp->ud_header);
+
+	if (!is_eth) {
+		sqp->ud_header.lrh.service_level =
+			be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 28;
+		sqp->ud_header.lrh.destination_lid = ah->av.ib.dlid;
+		sqp->ud_header.lrh.source_lid = cpu_to_be16(ah->av.ib.g_slid & 0x7f);
+	}
 
-	sqp->ud_header.lrh.service_level   =
-		be32_to_cpu(ah->av.sl_tclass_flowlabel) >> 28;
-	sqp->ud_header.lrh.destination_lid = ah->av.dlid;
-	sqp->ud_header.lrh.source_lid      = cpu_to_be16(ah->av.g_slid & 0x7f);
-	if (mlx4_ib_ah_grh_present(ah)) {
+	if (is_grh) {
 		sqp->ud_header.grh.traffic_class =
-			(be32_to_cpu(ah->av.sl_tclass_flowlabel) >> 20) & 0xff;
+			(be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 20) & 0xff;
 		sqp->ud_header.grh.flow_label    =
-			ah->av.sl_tclass_flowlabel & cpu_to_be32(0xfffff);
-		sqp->ud_header.grh.hop_limit     = ah->av.hop_limit;
-		ib_get_cached_gid(ib_dev, be32_to_cpu(ah->av.port_pd) >> 24,
-				  ah->av.gid_index, &sqp->ud_header.grh.source_gid);
+			ah->av.ib.sl_tclass_flowlabel & cpu_to_be32(0xfffff);
+		sqp->ud_header.grh.hop_limit     = ah->av.ib.hop_limit;
+		if (is_eth)
+			memcpy(sqp->ud_header.grh.source_gid.raw, sgid.raw, 16);
+		else {
+		if (mlx4_is_mfunc(to_mdev(ib_dev)->dev)) {
+			/* When multi-function is enabled, the ib_core gid
+			 * indexes don't necessarily match the hw ones, so
+			 * we must use our own cache */
+			sqp->ud_header.grh.source_gid.global.subnet_prefix =
+				to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+						       subnet_prefix;
+			sqp->ud_header.grh.source_gid.global.interface_id =
+				to_mdev(ib_dev)->sriov.demux[sqp->qp.port - 1].
+					       guid_cache[ah->av.ib.gid_index];
+		} else
+			ib_get_cached_gid(ib_dev,
+					  be32_to_cpu(ah->av.ib.port_pd) >> 24,
+					  ah->av.ib.gid_index,
+					  &sqp->ud_header.grh.source_gid);
+		}
 		memcpy(sqp->ud_header.grh.destination_gid.raw,
-		       ah->av.dgid, 16);
+		       ah->av.ib.dgid, 16);
 	}
 
 	mlx->flags &= cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE);
-	mlx->flags |= cpu_to_be32((!sqp->qp.ibqp.qp_num ? MLX4_WQE_MLX_VL15 : 0) |
-				  (sqp->ud_header.lrh.destination_lid ==
-				   IB_LID_PERMISSIVE ? MLX4_WQE_MLX_SLR : 0) |
-				  (sqp->ud_header.lrh.service_level << 8));
-	mlx->rlid   = sqp->ud_header.lrh.destination_lid;
+
+	if (!is_eth) {
+		mlx->flags |= cpu_to_be32((!sqp->qp.ibqp.qp_num ? MLX4_WQE_MLX_VL15 : 0) |
+					  (sqp->ud_header.lrh.destination_lid ==
+					   IB_LID_PERMISSIVE ? MLX4_WQE_MLX_SLR : 0) |
+					  (sqp->ud_header.lrh.service_level << 8));
+		if (ah->av.ib.port_pd & cpu_to_be32(0x80000000))
+			mlx->flags |= cpu_to_be32(0x1); /* force loopback */
+		mlx->rlid = sqp->ud_header.lrh.destination_lid;
+	}
 
 	switch (wr->opcode) {
 	case IB_WR_SEND:
@@ -1234,9 +2196,39 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 		return -EINVAL;
 	}
 
-	sqp->ud_header.lrh.virtual_lane    = !sqp->qp.ibqp.qp_num ? 15 : 0;
-	if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
-		sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
+	if (is_eth) {
+		u8 *smac;
+		struct in6_addr in6;
+
+		u16 pcp = (be32_to_cpu(ah->av.ib.sl_tclass_flowlabel) >> 29) << 13;
+
+		mlx->sched_prio = cpu_to_be16(pcp);
+
+		memcpy(sqp->ud_header.eth.dmac_h, ah->av.eth.mac, 6);
+		/* FIXME: cache smac value? */
+		memcpy(&ctrl->srcrb_flags16[0], ah->av.eth.mac, 2);
+		memcpy(&ctrl->imm, ah->av.eth.mac + 2, 4);
+		memcpy(&in6, sgid.raw, sizeof(in6));
+
+		if (!mlx4_is_mfunc(to_mdev(ib_dev)->dev))
+			smac = to_mdev(sqp->qp.ibqp.device)->
+				iboe.netdevs[sqp->qp.port - 1]->dev_addr;
+		else	/* use the src mac of the tunnel */
+			smac = ah->av.eth.s_mac;
+		memcpy(sqp->ud_header.eth.smac_h, smac, 6);
+		if (!memcmp(sqp->ud_header.eth.smac_h, sqp->ud_header.eth.dmac_h, 6))
+			mlx->flags |= cpu_to_be32(MLX4_WQE_CTRL_FORCE_LOOPBACK);
+		if (!is_vlan) {
+			sqp->ud_header.eth.type = cpu_to_be16(MLX4_IB_IBOE_ETHERTYPE);
+		} else {
+			sqp->ud_header.vlan.type = cpu_to_be16(MLX4_IB_IBOE_ETHERTYPE);
+			sqp->ud_header.vlan.tag = cpu_to_be16(vlan | pcp);
+		}
+	} else {
+		sqp->ud_header.lrh.virtual_lane    = !sqp->qp.ibqp.qp_num ? 15 : 0;
+		if (sqp->ud_header.lrh.destination_lid == IB_LID_PERMISSIVE)
+			sqp->ud_header.lrh.source_lid = IB_LID_PERMISSIVE;
+	}
 	sqp->ud_header.bth.solicited_event = !!(wr->send_flags & IB_SEND_SOLICITED);
 	if (!sqp->qp.ibqp.qp_num)
 		ib_get_cached_pkey(ib_dev, sqp->qp.port, sqp->pkey_index, &pkey);
@@ -1252,16 +2244,16 @@ static int build_mlx_header(struct mlx4_ib_sqp *sqp, struct ib_send_wr *wr,
 	header_size = ib_ud_header_pack(&sqp->ud_header, sqp->header_buf);
 
 	if (0) {
-		printk(KERN_ERR "built UD header of size %d:\n", header_size);
+		pr_err("built UD header of size %d:\n", header_size);
 		for (i = 0; i < header_size / 4; ++i) {
 			if (i % 8 == 0)
-				printk("  [%02x] ", i * 4);
-			printk(" %08x",
-			       be32_to_cpu(((__be32 *) sqp->header_buf)[i]));
+				pr_err("  [%02x] ", i * 4);
+			pr_cont(" %08x",
+				be32_to_cpu(((__be32 *) sqp->header_buf)[i]));
 			if ((i + 1) % 8 == 0)
-				printk("\n");
+				pr_cont("\n");
 		}
-		printk("\n");
+		pr_err("\n");
 	}
 
 	/*
@@ -1322,6 +2314,63 @@ static int mlx4_wq_overflow(struct mlx4_ib_wq *wq, int nreq, struct ib_cq *ib_cq
 	return cur + nreq >= wq->max_post;
 }
 
+static __be32 convert_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ?
+		cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC)       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ?
+		cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE) : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ?
+		cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ)  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_WRITE)  : 0) |
+		cpu_to_be32(MLX4_WQE_FMR_PERM_LOCAL_READ);
+}
+
+static void set_fmr_seg(struct mlx4_wqe_fmr_seg *fseg, struct ib_send_wr *wr)
+{
+	struct mlx4_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->wr.fast_reg.page_list);
+	int i;
+
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; ++i)
+		mfrpl->mapped_page_list[i] =
+			cpu_to_be64(wr->wr.fast_reg.page_list->page_list[i] |
+				    MLX4_MTT_FLAG_PRESENT);
+
+	fseg->flags		= convert_access(wr->wr.fast_reg.access_flags);
+	fseg->mem_key		= cpu_to_be32(wr->wr.fast_reg.rkey);
+	fseg->buf_list		= cpu_to_be64(mfrpl->map);
+	fseg->start_addr	= cpu_to_be64(wr->wr.fast_reg.iova_start);
+	fseg->reg_len		= cpu_to_be64(wr->wr.fast_reg.length);
+	fseg->offset		= 0; /* XXX -- is this just for ZBVA? */
+	fseg->page_size		= cpu_to_be32(wr->wr.fast_reg.page_shift);
+	fseg->reserved[0]	= 0;
+	fseg->reserved[1]	= 0;
+}
+
+static void set_bind_seg(struct mlx4_wqe_bind_seg *bseg, struct ib_send_wr *wr)
+{
+	bseg->flags1 =
+		convert_access(wr->wr.bind_mw.bind_info.mw_access_flags) &
+		cpu_to_be32(MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_READ  |
+			    MLX4_WQE_FMR_AND_BIND_PERM_REMOTE_WRITE |
+			    MLX4_WQE_FMR_AND_BIND_PERM_ATOMIC);
+	bseg->flags2 = 0;
+	if (wr->wr.bind_mw.mw->type == IB_MW_TYPE_2)
+		bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_TYPE_2);
+	if (wr->wr.bind_mw.bind_info.mw_access_flags & IB_ZERO_BASED)
+		bseg->flags2 |= cpu_to_be32(MLX4_WQE_BIND_ZERO_BASED);
+	bseg->new_rkey = cpu_to_be32(wr->wr.bind_mw.rkey);
+	bseg->lkey = cpu_to_be32(wr->wr.bind_mw.bind_info.mr->lkey);
+	bseg->addr = cpu_to_be64(wr->wr.bind_mw.bind_info.addr);
+	bseg->length = cpu_to_be64(wr->wr.bind_mw.bind_info.length);
+}
+
+static void set_local_inv_seg(struct mlx4_wqe_local_inval_seg *iseg, u32 rkey)
+{
+	memset(iseg, 0, sizeof(*iseg));
+	iseg->mem_key = cpu_to_be32(rkey);
+}
+
 static __always_inline void set_raddr_seg(struct mlx4_wqe_raddr_seg *rseg,
 					  u64 remote_addr, u32 rkey)
 {
@@ -1335,6 +2384,9 @@ static void set_atomic_seg(struct mlx4_wqe_atomic_seg *aseg, struct ib_send_wr *
 	if (wr->opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
 		aseg->swap_add = cpu_to_be64(wr->wr.atomic.swap);
 		aseg->compare  = cpu_to_be64(wr->wr.atomic.compare_add);
+	} else if (wr->opcode == IB_WR_MASKED_ATOMIC_FETCH_AND_ADD) {
+		aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
+		aseg->compare  = cpu_to_be64(wr->wr.atomic.compare_add_mask);
 	} else {
 		aseg->swap_add = cpu_to_be64(wr->wr.atomic.compare_add);
 		aseg->compare  = 0;
@@ -1342,12 +2394,85 @@ static void set_atomic_seg(struct mlx4_wqe_atomic_seg *aseg, struct ib_send_wr *
 
 }
 
+static void set_masked_atomic_seg(struct mlx4_wqe_masked_atomic_seg *aseg,
+				  struct ib_send_wr *wr)
+{
+	aseg->swap_add		= cpu_to_be64(wr->wr.atomic.swap);
+	aseg->swap_add_mask	= cpu_to_be64(wr->wr.atomic.swap_mask);
+	aseg->compare		= cpu_to_be64(wr->wr.atomic.compare_add);
+	aseg->compare_mask	= cpu_to_be64(wr->wr.atomic.compare_add_mask);
+}
+
 static void set_datagram_seg(struct mlx4_wqe_datagram_seg *dseg,
 			     struct ib_send_wr *wr)
 {
 	memcpy(dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof (struct mlx4_av));
 	dseg->dqpn = cpu_to_be32(wr->wr.ud.remote_qpn);
 	dseg->qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+	dseg->vlan = to_mah(wr->wr.ud.ah)->av.eth.vlan;
+	memcpy(dseg->mac, to_mah(wr->wr.ud.ah)->av.eth.mac, 6);
+}
+
+static void set_tunnel_datagram_seg(struct mlx4_ib_dev *dev,
+				    struct mlx4_wqe_datagram_seg *dseg,
+				    struct ib_send_wr *wr,
+				    enum mlx4_ib_qp_type qpt)
+{
+	union mlx4_ext_av *av = &to_mah(wr->wr.ud.ah)->av;
+	struct mlx4_av sqp_av = {0};
+	int port = *((u8 *) &av->ib.port_pd) & 0x3;
+
+	/* force loopback */
+	sqp_av.port_pd = av->ib.port_pd | cpu_to_be32(0x80000000);
+	sqp_av.g_slid = av->ib.g_slid & 0x7f; /* no GRH */
+	sqp_av.sl_tclass_flowlabel = av->ib.sl_tclass_flowlabel &
+			cpu_to_be32(0xf0000000);
+
+	memcpy(dseg->av, &sqp_av, sizeof (struct mlx4_av));
+	if (qpt == MLX4_IB_QPT_PROXY_GSI)
+		dseg->dqpn = cpu_to_be32(dev->dev->caps.qp1_tunnel[port - 1]);
+	else
+		dseg->dqpn = cpu_to_be32(dev->dev->caps.qp0_tunnel[port - 1]);
+	/* Use QKEY from the QP context, which is set by master */
+	dseg->qkey = cpu_to_be32(IB_QP_SET_QKEY);
+}
+
+static void build_tunnel_header(struct ib_send_wr *wr, void *wqe, unsigned *mlx_seg_len)
+{
+	struct mlx4_wqe_inline_seg *inl = wqe;
+	struct mlx4_ib_tunnel_header hdr;
+	struct mlx4_ib_ah *ah = to_mah(wr->wr.ud.ah);
+	int spc;
+	int i;
+
+	memcpy(&hdr.av, &ah->av, sizeof hdr.av);
+	hdr.remote_qpn = cpu_to_be32(wr->wr.ud.remote_qpn);
+	hdr.pkey_index = cpu_to_be16(wr->wr.ud.pkey_index);
+	hdr.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+	memcpy(hdr.mac, ah->av.eth.mac, 6);
+	hdr.vlan = ah->av.eth.vlan;
+
+	spc = MLX4_INLINE_ALIGN -
+		((unsigned long) (inl + 1) & (MLX4_INLINE_ALIGN - 1));
+	if (sizeof (hdr) <= spc) {
+		memcpy(inl + 1, &hdr, sizeof (hdr));
+		wmb();
+		inl->byte_count = cpu_to_be32(1 << 31 | sizeof (hdr));
+		i = 1;
+	} else {
+		memcpy(inl + 1, &hdr, spc);
+		wmb();
+		inl->byte_count = cpu_to_be32(1 << 31 | spc);
+
+		inl = (void *) (inl + 1) + spc;
+		memcpy(inl + 1, (void *) &hdr + spc, sizeof (hdr) - spc);
+		wmb();
+		inl->byte_count = cpu_to_be32(1 << 31 | (sizeof (hdr) - spc));
+		i = 2;
+	}
+
+	*mlx_seg_len =
+		ALIGN(i * sizeof (struct mlx4_wqe_inline_seg) + sizeof (hdr), 16);
 }
 
 static void set_mlx_icrc_seg(void *dseg)
@@ -1395,17 +2520,14 @@ static void __set_data_seg(struct mlx4_wqe_data_seg *dseg, struct ib_sge *sg)
 	dseg->addr       = cpu_to_be64(sg->addr);
 }
 
-static int build_lso_seg(struct mlx4_lso_seg *wqe, struct ib_send_wr *wr,
-			 struct mlx4_ib_qp *qp, unsigned *lso_seg_len)
+static int build_lso_seg(struct mlx4_wqe_lso_seg *wqe, struct ib_send_wr *wr,
+			 struct mlx4_ib_qp *qp, unsigned *lso_seg_len,
+			 __be32 *lso_hdr_sz, __be32 *blh)
 {
 	unsigned halign = ALIGN(sizeof *wqe + wr->wr.ud.hlen, 16);
 
-	/*
-	 * This is a temporary limitation and will be removed in
-	 * a forthcoming FW release:
-	 */
-	if (unlikely(halign > 64))
-		return -EINVAL;
+	if (unlikely(halign > MLX4_IB_CACHE_LINE_SIZE))
+		*blh = cpu_to_be32(1 << 6);
 
 	if (unlikely(!(qp->flags & MLX4_IB_QP_LSO) &&
 		     wr->num_sge > qp->sq.max_gs - (halign >> 4)))
@@ -1413,16 +2535,34 @@ static int build_lso_seg(struct mlx4_lso_seg *wqe, struct ib_send_wr *wr,
 
 	memcpy(wqe->header, wr->wr.ud.header, wr->wr.ud.hlen);
 
-	/* make sure LSO header is written before overwriting stamping */
-	wmb();
-
-	wqe->mss_hdr_size = cpu_to_be32((wr->wr.ud.mss - wr->wr.ud.hlen) << 16 |
-					wr->wr.ud.hlen);
-
+	*lso_hdr_sz  = cpu_to_be32((wr->wr.ud.mss - wr->wr.ud.hlen) << 16 |
+				   wr->wr.ud.hlen);
 	*lso_seg_len = halign;
 	return 0;
 }
 
+static __be32 send_ieth(struct ib_send_wr *wr)
+{
+	switch (wr->opcode) {
+	case IB_WR_SEND_WITH_IMM:
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		return wr->ex.imm_data;
+
+	case IB_WR_SEND_WITH_INV:
+		return cpu_to_be32(wr->ex.invalidate_rkey);
+
+	default:
+		return 0;
+	}
+}
+
+static void add_zero_len_inline(void *wqe)
+{
+	struct mlx4_wqe_inline_seg *inl = wqe;
+	memset(wqe, 0, 16);
+	inl->byte_count = cpu_to_be32(1 << 31);
+}
+
 int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		      struct ib_send_wr **bad_wr)
 {
@@ -1437,6 +2577,10 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	int uninitialized_var(stamp);
 	int uninitialized_var(size);
 	unsigned uninitialized_var(seglen);
+	__be32 dummy;
+	__be32 *lso_wqe;
+	__be32 uninitialized_var(lso_hdr_sz);
+	__be32 blh;
 	int i;
 
 	spin_lock_irqsave(&qp->sq.lock, flags);
@@ -1444,6 +2588,9 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 	ind = qp->sq_next_wqe;
 
 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
+		lso_wqe = &dummy;
+		blh = 0;
+
 		if (mlx4_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq)) {
 			err = -ENOMEM;
 			*bad_wr = wr;
@@ -1469,21 +2616,18 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 				     MLX4_WQE_CTRL_TCP_UDP_CSUM) : 0) |
 			qp->sq_signal_bits;
 
-		if (wr->opcode == IB_WR_SEND_WITH_IMM ||
-		    wr->opcode == IB_WR_RDMA_WRITE_WITH_IMM)
-			ctrl->imm = wr->ex.imm_data;
-		else
-			ctrl->imm = 0;
+		ctrl->imm = send_ieth(wr);
 
 		wqe += sizeof *ctrl;
 		size = sizeof *ctrl / 16;
 
-		switch (ibqp->qp_type) {
-		case IB_QPT_RC:
-		case IB_QPT_UC:
+		switch (qp->mlx4_ib_qp_type) {
+		case MLX4_IB_QPT_RC:
+		case MLX4_IB_QPT_UC:
 			switch (wr->opcode) {
 			case IB_WR_ATOMIC_CMP_AND_SWP:
 			case IB_WR_ATOMIC_FETCH_AND_ADD:
+			case IB_WR_MASKED_ATOMIC_FETCH_AND_ADD:
 				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
 					      wr->wr.atomic.rkey);
 				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
@@ -1496,6 +2640,19 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 
 				break;
 
+			case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
+				set_raddr_seg(wqe, wr->wr.atomic.remote_addr,
+					      wr->wr.atomic.rkey);
+				wqe  += sizeof (struct mlx4_wqe_raddr_seg);
+
+				set_masked_atomic_seg(wqe, wr);
+				wqe  += sizeof (struct mlx4_wqe_masked_atomic_seg);
+
+				size += (sizeof (struct mlx4_wqe_raddr_seg) +
+					 sizeof (struct mlx4_wqe_masked_atomic_seg)) / 16;
+
+				break;
+
 			case IB_WR_RDMA_READ:
 			case IB_WR_RDMA_WRITE:
 			case IB_WR_RDMA_WRITE_WITH_IMM:
@@ -1505,30 +2662,103 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 				size += sizeof (struct mlx4_wqe_raddr_seg) / 16;
 				break;
 
+			case IB_WR_LOCAL_INV:
+				ctrl->srcrb_flags |=
+					cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
+				set_local_inv_seg(wqe, wr->ex.invalidate_rkey);
+				wqe  += sizeof (struct mlx4_wqe_local_inval_seg);
+				size += sizeof (struct mlx4_wqe_local_inval_seg) / 16;
+				break;
+
+			case IB_WR_FAST_REG_MR:
+				ctrl->srcrb_flags |=
+					cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
+				set_fmr_seg(wqe, wr);
+				wqe  += sizeof (struct mlx4_wqe_fmr_seg);
+				size += sizeof (struct mlx4_wqe_fmr_seg) / 16;
+				break;
+
+			case IB_WR_BIND_MW:
+				ctrl->srcrb_flags |=
+					cpu_to_be32(MLX4_WQE_CTRL_STRONG_ORDER);
+				set_bind_seg(wqe, wr);
+				wqe  += sizeof(struct mlx4_wqe_bind_seg);
+				size += sizeof(struct mlx4_wqe_bind_seg) / 16;
+				break;
 			default:
 				/* No extra segments required for sends */
 				break;
 			}
 			break;
 
-		case IB_QPT_UD:
+		case MLX4_IB_QPT_TUN_SMI_OWNER:
+			err =  build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+			if (unlikely(err)) {
+				*bad_wr = wr;
+				goto out;
+			}
+			wqe  += seglen;
+			size += seglen / 16;
+			break;
+		case MLX4_IB_QPT_TUN_SMI:
+		case MLX4_IB_QPT_TUN_GSI:
+			/* this is a UD qp used in MAD responses to slaves. */
+			set_datagram_seg(wqe, wr);
+			/* set the forced-loopback bit in the data seg av */
+			*(__be32 *) wqe |= cpu_to_be32(0x80000000);
+			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
+			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
+			break;
+		case MLX4_IB_QPT_UD:
 			set_datagram_seg(wqe, wr);
 			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
 			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
 
 			if (wr->opcode == IB_WR_LSO) {
-				err = build_lso_seg(wqe, wr, qp, &seglen);
+				err = build_lso_seg(wqe, wr, qp, &seglen, &lso_hdr_sz, &blh);
 				if (unlikely(err)) {
 					*bad_wr = wr;
 					goto out;
 				}
+				lso_wqe = (__be32 *) wqe;
 				wqe  += seglen;
 				size += seglen / 16;
 			}
 			break;
 
-		case IB_QPT_SMI:
-		case IB_QPT_GSI:
+		case MLX4_IB_QPT_PROXY_SMI_OWNER:
+			err = build_sriov_qp0_header(to_msqp(qp), wr, ctrl, &seglen);
+			if (unlikely(err)) {
+				*bad_wr = wr;
+				goto out;
+			}
+			wqe  += seglen;
+			size += seglen / 16;
+			/* to start tunnel header on a cache-line boundary */
+			add_zero_len_inline(wqe);
+			wqe += 16;
+			size++;
+			build_tunnel_header(wr, wqe, &seglen);
+			wqe  += seglen;
+			size += seglen / 16;
+			break;
+		case MLX4_IB_QPT_PROXY_SMI:
+		case MLX4_IB_QPT_PROXY_GSI:
+			/* If we are tunneling special qps, this is a UD qp.
+			 * In this case we first add a UD segment targeting
+			 * the tunnel qp, and then add a header with address
+			 * information */
+			set_tunnel_datagram_seg(to_mdev(ibqp->device), wqe, wr,
+						qp->mlx4_ib_qp_type);
+			wqe  += sizeof (struct mlx4_wqe_datagram_seg);
+			size += sizeof (struct mlx4_wqe_datagram_seg) / 16;
+			build_tunnel_header(wr, wqe, &seglen);
+			wqe  += seglen;
+			size += seglen / 16;
+			break;
+
+		case MLX4_IB_QPT_SMI:
+		case MLX4_IB_QPT_GSI:
 			err = build_mlx_header(to_msqp(qp), wr, ctrl, &seglen);
 			if (unlikely(err)) {
 				*bad_wr = wr;
@@ -1554,8 +2784,10 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		size += wr->num_sge * (sizeof (struct mlx4_wqe_data_seg) / 16);
 
 		/* Add one more inline data segment for ICRC for MLX sends */
-		if (unlikely(qp->ibqp.qp_type == IB_QPT_SMI ||
-			     qp->ibqp.qp_type == IB_QPT_GSI)) {
+		if (unlikely(qp->mlx4_ib_qp_type == MLX4_IB_QPT_SMI ||
+			     qp->mlx4_ib_qp_type == MLX4_IB_QPT_GSI ||
+			     qp->mlx4_ib_qp_type &
+			     (MLX4_IB_QPT_PROXY_SMI_OWNER | MLX4_IB_QPT_TUN_SMI_OWNER))) {
 			set_mlx_icrc_seg(dseg + 1);
 			size += sizeof (struct mlx4_wqe_data_seg) / 16;
 		}
@@ -1563,6 +2795,14 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		for (i = wr->num_sge - 1; i >= 0; --i, --dseg)
 			set_data_seg(dseg, wr->sg_list + i);
 
+		/*
+		 * Possibly overwrite stamping in cacheline with LSO
+		 * segment only after making sure all data segments
+		 * are written.
+		 */
+		wmb();
+		*lso_wqe = lso_hdr_sz;
+
 		ctrl->fence_size = (wr->send_flags & IB_SEND_FENCE ?
 				    MLX4_WQE_CTRL_FENCE : 0) | size;
 
@@ -1574,12 +2814,13 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 		wmb();
 
 		if (wr->opcode < 0 || wr->opcode >= ARRAY_SIZE(mlx4_ib_opcode)) {
+			*bad_wr = wr;
 			err = -EINVAL;
 			goto out;
 		}
 
 		ctrl->owner_opcode = mlx4_ib_opcode[wr->opcode] |
-			(ind & qp->sq.wqe_cnt ? cpu_to_be32(1 << 31) : 0);
+			(ind & qp->sq.wqe_cnt ? cpu_to_be32(1 << 31) : 0) | blh;
 
 		stamp = ind + qp->sq_spare_wqes;
 		ind += DIV_ROUND_UP(size * 16, 1U << qp->sq.wqe_shift);
@@ -1597,7 +2838,6 @@ int mlx4_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
 			stamp_send_wqe(qp, stamp, size * 16);
 			ind = pad_wraparound(qp, ind);
 		}
-
 	}
 
 out:
@@ -1639,14 +2879,16 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 	int err = 0;
 	int nreq;
 	int ind;
+	int max_gs;
 	int i;
 
+	max_gs = qp->rq.max_gs;
 	spin_lock_irqsave(&qp->rq.lock, flags);
 
 	ind = qp->rq.head & (qp->rq.wqe_cnt - 1);
 
 	for (nreq = 0; wr; ++nreq, wr = wr->next) {
-		if (mlx4_wq_overflow(&qp->rq, nreq, qp->ibqp.send_cq)) {
+		if (mlx4_wq_overflow(&qp->rq, nreq, qp->ibqp.recv_cq)) {
 			err = -ENOMEM;
 			*bad_wr = wr;
 			goto out;
@@ -1660,10 +2902,25 @@ int mlx4_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
 
 		scat = get_recv_wqe(qp, ind);
 
+		if (qp->mlx4_ib_qp_type & (MLX4_IB_QPT_PROXY_SMI_OWNER |
+		    MLX4_IB_QPT_PROXY_SMI | MLX4_IB_QPT_PROXY_GSI)) {
+			ib_dma_sync_single_for_device(ibqp->device,
+						      qp->sqp_proxy_rcv[ind].map,
+						      sizeof (struct mlx4_ib_proxy_sqp_hdr),
+						      DMA_FROM_DEVICE);
+			scat->byte_count =
+				cpu_to_be32(sizeof (struct mlx4_ib_proxy_sqp_hdr));
+			/* use dma lkey from upper layer entry */
+			scat->lkey = cpu_to_be32(wr->sg_list->lkey);
+			scat->addr = cpu_to_be64(qp->sqp_proxy_rcv[ind].map);
+			scat++;
+			max_gs--;
+		}
+
 		for (i = 0; i < wr->num_sge; ++i)
 			__set_data_seg(scat + i, wr->sg_list + i);
 
-		if (i < qp->rq.max_gs) {
+		if (i < max_gs) {
 			scat[i].byte_count = 0;
 			scat[i].lkey       = cpu_to_be32(MLX4_INVALID_LKEY);
 			scat[i].addr       = 0;
@@ -1731,17 +2988,27 @@ static int to_ib_qp_access_flags(int mlx4_flags)
 	return ib_flags;
 }
 
-static void to_ib_ah_attr(struct mlx4_dev *dev, struct ib_ah_attr *ib_ah_attr,
+static void to_ib_ah_attr(struct mlx4_ib_dev *ibdev, struct ib_ah_attr *ib_ah_attr,
 				struct mlx4_qp_path *path)
 {
+	struct mlx4_dev *dev = ibdev->dev;
+	int is_eth;
+
 	memset(ib_ah_attr, 0, sizeof *ib_ah_attr);
 	ib_ah_attr->port_num	  = path->sched_queue & 0x40 ? 2 : 1;
 
 	if (ib_ah_attr->port_num == 0 || ib_ah_attr->port_num > dev->caps.num_ports)
 		return;
 
+	is_eth = rdma_port_get_link_layer(&ibdev->ib_dev, ib_ah_attr->port_num) ==
+		IB_LINK_LAYER_ETHERNET;
+	if (is_eth)
+		ib_ah_attr->sl = ((path->sched_queue >> 3) & 0x7) |
+		((path->sched_queue & 4) << 1);
+	else
+		ib_ah_attr->sl = (path->sched_queue >> 2) & 0xf;
+
 	ib_ah_attr->dlid	  = be16_to_cpu(path->rlid);
-	ib_ah_attr->sl		  = (path->sched_queue >> 2) & 0xf;
 	ib_ah_attr->src_path_bits = path->grh_mylmc & 0x7f;
 	ib_ah_attr->static_rate   = path->static_rate ? path->static_rate - 5 : 0;
 	ib_ah_attr->ah_flags      = (path->grh_mylmc & (1 << 7)) ? IB_AH_GRH : 0;
@@ -1794,8 +3061,8 @@ int mlx4_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr
 		to_ib_qp_access_flags(be32_to_cpu(context.params2));
 
 	if (qp->ibqp.qp_type == IB_QPT_RC || qp->ibqp.qp_type == IB_QPT_UC) {
-		to_ib_ah_attr(dev->dev, &qp_attr->ah_attr, &context.pri_path);
-		to_ib_ah_attr(dev->dev, &qp_attr->alt_ah_attr, &context.alt_path);
+		to_ib_ah_attr(dev, &qp_attr->ah_attr, &context.pri_path);
+		to_ib_ah_attr(dev, &qp_attr->alt_ah_attr, &context.alt_path);
 		qp_attr->alt_pkey_index = context.alt_path.pkey_index & 0x7f;
 		qp_attr->alt_port_num	= qp_attr->alt_ah_attr.port_num;
 	}
@@ -1848,6 +3115,13 @@ done:
 	if (qp->flags & MLX4_IB_QP_LSO)
 		qp_init_attr->create_flags |= IB_QP_CREATE_IPOIB_UD_LSO;
 
+	if (qp->flags & MLX4_IB_QP_NETIF)
+		qp_init_attr->create_flags |= IB_QP_CREATE_NETIF_QP;
+
+	qp_init_attr->sq_sig_type =
+		qp->sq_signal_bits == cpu_to_be32(MLX4_WQE_CTRL_CQ_UPDATE) ?
+		IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
+
 out:
 	mutex_unlock(&qp->mutex);
 	return err;
diff --git a/drivers/infiniband/hw/mlx4/srq.c b/drivers/infiniband/hw/mlx4/srq.c
index 12d6bc6f800..62d9285300a 100644
--- a/drivers/infiniband/hw/mlx4/srq.c
+++ b/drivers/infiniband/hw/mlx4/srq.c
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -32,6 +33,7 @@
 
 #include <linux/mlx4/qp.h>
 #include <linux/mlx4/srq.h>
+#include <linux/slab.h>
 
 #include "mlx4_ib.h"
 #include "user.h"
@@ -57,7 +59,7 @@ static void mlx4_ib_srq_event(struct mlx4_srq *srq, enum mlx4_event type)
 			event.event = IB_EVENT_SRQ_ERR;
 			break;
 		default:
-			printk(KERN_WARNING "mlx4_ib: Unexpected event type %d "
+			pr_warn("Unexpected event type %d "
 			       "on SRQ %06x\n", type, srq->srqn);
 			return;
 		}
@@ -73,6 +75,9 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 	struct mlx4_ib_dev *dev = to_mdev(pd->device);
 	struct mlx4_ib_srq *srq;
 	struct mlx4_wqe_srq_next_seg *next;
+	struct mlx4_wqe_data_seg *scatter;
+	u32 cqn;
+	u16 xrcdn;
 	int desc_size;
 	int buf_size;
 	int err;
@@ -129,13 +134,14 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 		if (err)
 			goto err_mtt;
 	} else {
-		err = mlx4_db_alloc(dev->dev, &srq->db, 0);
+		err = mlx4_db_alloc(dev->dev, &srq->db, 0, GFP_KERNEL);
 		if (err)
 			goto err_srq;
 
 		*srq->db.db = 0;
 
-		if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
+		if (mlx4_buf_alloc(dev->dev, buf_size, PAGE_SIZE * 2, &srq->buf,
+				   GFP_KERNEL)) {
 			err = -ENOMEM;
 			goto err_db;
 		}
@@ -148,6 +154,11 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 			next = get_wqe(srq, i);
 			next->next_wqe_index =
 				cpu_to_be16((i + 1) & (srq->msrq.max - 1));
+
+			for (scatter = (void *) (next + 1);
+			     (void *) scatter < (void *) next + desc_size;
+			     ++scatter)
+				scatter->lkey = cpu_to_be32(MLX4_INVALID_LKEY);
 		}
 
 		err = mlx4_mtt_init(dev->dev, srq->buf.npages, srq->buf.page_shift,
@@ -155,7 +166,7 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 		if (err)
 			goto err_buf;
 
-		err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf);
+		err = mlx4_buf_write_mtt(dev->dev, &srq->mtt, &srq->buf, GFP_KERNEL);
 		if (err)
 			goto err_mtt;
 
@@ -166,12 +177,18 @@ struct ib_srq *mlx4_ib_create_srq(struct ib_pd *pd,
 		}
 	}
 
-	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, &srq->mtt,
+	cqn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mcq(init_attr->ext.xrc.cq)->mcq.cqn : 0;
+	xrcdn = (init_attr->srq_type == IB_SRQT_XRC) ?
+		to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn :
+		(u16) dev->dev->caps.reserved_xrcds;
+	err = mlx4_srq_alloc(dev->dev, to_mpd(pd)->pdn, cqn, xrcdn, &srq->mtt,
 			     srq->db.dma, &srq->msrq);
 	if (err)
 		goto err_wrid;
 
 	srq->msrq.event = mlx4_ib_srq_event;
+	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
 
 	if (pd->uobject)
 		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof (__u32))) {
diff --git a/drivers/infiniband/hw/mlx4/sysfs.c b/drivers/infiniband/hw/mlx4/sysfs.c
new file mode 100644
index 00000000000..cb4c66e723b
--- /dev/null
+++ b/drivers/infiniband/hw/mlx4/sysfs.c
@@ -0,0 +1,906 @@
+/*
+ * Copyright (c) 2012 Mellanox Technologies.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*#include "core_priv.h"*/
+#include "mlx4_ib.h"
+#include <linux/slab.h>
+#include <linux/string.h>
+#include <linux/stat.h>
+
+#include <rdma/ib_mad.h>
+/*show_admin_alias_guid returns the administratively assigned value of that GUID.
+ * Values returned in buf parameter string:
+ *	0			- requests opensm to assign a value.
+ *	ffffffffffffffff	- delete this entry.
+ *	other			- value assigned by administrator.
+ */
+static ssize_t show_admin_alias_guid(struct device *dev,
+			      struct device_attribute *attr, char *buf)
+{
+	int record_num;/*0-15*/
+	int guid_index_in_rec; /*0 - 7*/
+	struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+		container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+	struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+	struct mlx4_ib_dev *mdev = port->dev;
+
+	record_num = mlx4_ib_iov_dentry->entry_num / 8 ;
+	guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8 ;
+
+	return sprintf(buf, "%llx\n",
+		       be64_to_cpu(*(__be64 *)&mdev->sriov.alias_guid.
+				   ports_guid[port->num - 1].
+				   all_rec_per_port[record_num].
+				   all_recs[8 * guid_index_in_rec]));
+}
+
+/* store_admin_alias_guid stores the (new) administratively assigned value of that GUID.
+ * Values in buf parameter string:
+ *	0			- requests opensm to assign a value.
+ *	0xffffffffffffffff	- delete this entry.
+ *	other			- guid value assigned by the administrator.
+ */
+static ssize_t store_admin_alias_guid(struct device *dev,
+				      struct device_attribute *attr,
+				      const char *buf, size_t count)
+{
+	int record_num;/*0-15*/
+	int guid_index_in_rec; /*0 - 7*/
+	struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+		container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+	struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+	struct mlx4_ib_dev *mdev = port->dev;
+	u64 sysadmin_ag_val;
+
+	record_num = mlx4_ib_iov_dentry->entry_num / 8;
+	guid_index_in_rec = mlx4_ib_iov_dentry->entry_num % 8;
+	if (0 == record_num && 0 == guid_index_in_rec) {
+		pr_err("GUID 0 block 0 is RO\n");
+		return count;
+	}
+	sscanf(buf, "%llx", &sysadmin_ag_val);
+	*(__be64 *)&mdev->sriov.alias_guid.ports_guid[port->num - 1].
+		all_rec_per_port[record_num].
+		all_recs[GUID_REC_SIZE * guid_index_in_rec] =
+			cpu_to_be64(sysadmin_ag_val);
+
+	/* Change the state to be pending for update */
+	mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].status
+		= MLX4_GUID_INFO_STATUS_IDLE ;
+
+	mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method
+		= MLX4_GUID_INFO_RECORD_SET;
+
+	switch (sysadmin_ag_val) {
+	case MLX4_GUID_FOR_DELETE_VAL:
+		mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].method
+			= MLX4_GUID_INFO_RECORD_DELETE;
+		mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+			= MLX4_GUID_SYSADMIN_ASSIGN;
+		break;
+	/* The sysadmin requests the SM to re-assign */
+	case MLX4_NOT_SET_GUID:
+		mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+			= MLX4_GUID_DRIVER_ASSIGN;
+		break;
+	/* The sysadmin requests a specific value.*/
+	default:
+		mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].ownership
+			= MLX4_GUID_SYSADMIN_ASSIGN;
+		break;
+	}
+
+	/* set the record index */
+	mdev->sriov.alias_guid.ports_guid[port->num - 1].all_rec_per_port[record_num].guid_indexes
+		= mlx4_ib_get_aguid_comp_mask_from_ix(guid_index_in_rec);
+
+	mlx4_ib_init_alias_guid_work(mdev, port->num - 1);
+
+	return count;
+}
+
+static ssize_t show_port_gid(struct device *dev,
+			     struct device_attribute *attr,
+			     char *buf)
+{
+	struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+		container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+	struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+	struct mlx4_ib_dev *mdev = port->dev;
+	union ib_gid gid;
+	ssize_t ret;
+
+	ret = __mlx4_ib_query_gid(&mdev->ib_dev, port->num,
+				  mlx4_ib_iov_dentry->entry_num, &gid, 1);
+	if (ret)
+		return ret;
+	ret = sprintf(buf, "%04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x\n",
+		      be16_to_cpu(((__be16 *) gid.raw)[0]),
+		      be16_to_cpu(((__be16 *) gid.raw)[1]),
+		      be16_to_cpu(((__be16 *) gid.raw)[2]),
+		      be16_to_cpu(((__be16 *) gid.raw)[3]),
+		      be16_to_cpu(((__be16 *) gid.raw)[4]),
+		      be16_to_cpu(((__be16 *) gid.raw)[5]),
+		      be16_to_cpu(((__be16 *) gid.raw)[6]),
+		      be16_to_cpu(((__be16 *) gid.raw)[7]));
+	return ret;
+}
+
+static ssize_t show_phys_port_pkey(struct device *dev,
+				   struct device_attribute *attr,
+				   char *buf)
+{
+	struct mlx4_ib_iov_sysfs_attr *mlx4_ib_iov_dentry =
+		container_of(attr, struct mlx4_ib_iov_sysfs_attr, dentry);
+	struct mlx4_ib_iov_port *port = mlx4_ib_iov_dentry->ctx;
+	struct mlx4_ib_dev *mdev = port->dev;
+	u16 pkey;
+	ssize_t ret;
+
+	ret = __mlx4_ib_query_pkey(&mdev->ib_dev, port->num,
+				   mlx4_ib_iov_dentry->entry_num, &pkey, 1);
+	if (ret)
+		return ret;
+
+	return sprintf(buf, "0x%04x\n", pkey);
+}
+
+#define DENTRY_REMOVE(_dentry)						\
+do {									\
+	sysfs_remove_file((_dentry)->kobj, &(_dentry)->dentry.attr);	\
+} while (0);
+
+static int create_sysfs_entry(void *_ctx, struct mlx4_ib_iov_sysfs_attr *_dentry,
+			      char *_name, struct kobject *_kobj,
+			      ssize_t (*show)(struct device *dev,
+					      struct device_attribute *attr,
+					      char *buf),
+			      ssize_t (*store)(struct device *dev,
+					       struct device_attribute *attr,
+					       const char *buf, size_t count)
+			      )
+{
+	int ret = 0;
+	struct mlx4_ib_iov_sysfs_attr *vdentry = _dentry;
+
+	vdentry->ctx = _ctx;
+	vdentry->dentry.show = show;
+	vdentry->dentry.store = store;
+	sysfs_attr_init(&vdentry->dentry.attr);
+	vdentry->dentry.attr.name = vdentry->name;
+	vdentry->dentry.attr.mode = 0;
+	vdentry->kobj = _kobj;
+	snprintf(vdentry->name, 15, "%s", _name);
+
+	if (vdentry->dentry.store)
+		vdentry->dentry.attr.mode |= S_IWUSR;
+
+	if (vdentry->dentry.show)
+		vdentry->dentry.attr.mode |= S_IRUGO;
+
+	ret = sysfs_create_file(vdentry->kobj, &vdentry->dentry.attr);
+	if (ret) {
+		pr_err("failed to create %s\n", vdentry->dentry.attr.name);
+		vdentry->ctx = NULL;
+		return ret;
+	}
+
+	return ret;
+}
+
+int add_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+		struct attribute *attr)
+{
+	struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1];
+	int ret;
+
+	ret = sysfs_create_file(port->mcgs_parent, attr);
+	if (ret)
+		pr_err("failed to create %s\n", attr->name);
+
+	return ret;
+}
+
+void del_sysfs_port_mcg_attr(struct mlx4_ib_dev *device, int port_num,
+		struct attribute *attr)
+{
+	struct mlx4_ib_iov_port *port = &device->iov_ports[port_num - 1];
+
+	sysfs_remove_file(port->mcgs_parent, attr);
+}
+
+static int add_port_entries(struct mlx4_ib_dev *device, int port_num)
+{
+	int i;
+	char buff[10];
+	struct mlx4_ib_iov_port *port = NULL;
+	int ret = 0 ;
+	struct ib_port_attr attr;
+
+	/* get the physical gid and pkey table sizes.*/
+	ret = __mlx4_ib_query_port(&device->ib_dev, port_num, &attr, 1);
+	if (ret)
+		goto err;
+
+	port = &device->iov_ports[port_num - 1];
+	port->dev = device;
+	port->num = port_num;
+	/* Directory structure:
+	 * iov -
+	 *   port num -
+	 *	admin_guids
+	 *	gids (operational)
+	 *	mcg_table
+	 */
+	port->dentr_ar = kzalloc(sizeof (struct mlx4_ib_iov_sysfs_attr_ar),
+				 GFP_KERNEL);
+	if (!port->dentr_ar) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	sprintf(buff, "%d", port_num);
+	port->cur_port = kobject_create_and_add(buff,
+				 kobject_get(device->ports_parent));
+	if (!port->cur_port) {
+		ret = -ENOMEM;
+		goto kobj_create_err;
+	}
+	/* admin GUIDs */
+	port->admin_alias_parent = kobject_create_and_add("admin_guids",
+						  kobject_get(port->cur_port));
+	if (!port->admin_alias_parent) {
+		ret = -ENOMEM;
+		goto err_admin_guids;
+	}
+	for (i = 0 ; i < attr.gid_tbl_len; i++) {
+		sprintf(buff, "%d", i);
+		port->dentr_ar->dentries[i].entry_num = i;
+		ret = create_sysfs_entry(port, &port->dentr_ar->dentries[i],
+					  buff, port->admin_alias_parent,
+					  show_admin_alias_guid, store_admin_alias_guid);
+		if (ret)
+			goto err_admin_alias_parent;
+	}
+
+	/* gids subdirectory (operational gids) */
+	port->gids_parent = kobject_create_and_add("gids",
+						  kobject_get(port->cur_port));
+	if (!port->gids_parent) {
+		ret = -ENOMEM;
+		goto err_gids;
+	}
+
+	for (i = 0 ; i < attr.gid_tbl_len; i++) {
+		sprintf(buff, "%d", i);
+		port->dentr_ar->dentries[attr.gid_tbl_len + i].entry_num = i;
+		ret = create_sysfs_entry(port,
+					 &port->dentr_ar->dentries[attr.gid_tbl_len + i],
+					 buff,
+					 port->gids_parent, show_port_gid, NULL);
+		if (ret)
+			goto err_gids_parent;
+	}
+
+	/* physical port pkey table */
+	port->pkeys_parent =
+		kobject_create_and_add("pkeys", kobject_get(port->cur_port));
+	if (!port->pkeys_parent) {
+		ret = -ENOMEM;
+		goto err_pkeys;
+	}
+
+	for (i = 0 ; i < attr.pkey_tbl_len; i++) {
+		sprintf(buff, "%d", i);
+		port->dentr_ar->dentries[2 * attr.gid_tbl_len + i].entry_num = i;
+		ret = create_sysfs_entry(port,
+					 &port->dentr_ar->dentries[2 * attr.gid_tbl_len + i],
+					 buff, port->pkeys_parent,
+					 show_phys_port_pkey, NULL);
+		if (ret)
+			goto err_pkeys_parent;
+	}
+
+	/* MCGs table */
+	port->mcgs_parent =
+		kobject_create_and_add("mcgs", kobject_get(port->cur_port));
+	if (!port->mcgs_parent) {
+		ret = -ENOMEM;
+		goto err_mcgs;
+	}
+	return 0;
+
+err_mcgs:
+	kobject_put(port->cur_port);
+
+err_pkeys_parent:
+	kobject_put(port->pkeys_parent);
+
+err_pkeys:
+	kobject_put(port->cur_port);
+
+err_gids_parent:
+	kobject_put(port->gids_parent);
+
+err_gids:
+	kobject_put(port->cur_port);
+
+err_admin_alias_parent:
+	kobject_put(port->admin_alias_parent);
+
+err_admin_guids:
+	kobject_put(port->cur_port);
+	kobject_put(port->cur_port); /* once more for create_and_add buff */
+
+kobj_create_err:
+	kobject_put(device->ports_parent);
+	kfree(port->dentr_ar);
+
+err:
+	pr_err("add_port_entries FAILED: for port:%d, error: %d\n",
+	       port_num, ret);
+	return ret;
+}
+
+static void get_name(struct mlx4_ib_dev *dev, char *name, int i, int max)
+{
+	char base_name[9];
+
+	/* pci_name format is: bus:dev:func -> xxxx:yy:zz.n */
+	strlcpy(name, pci_name(dev->dev->pdev), max);
+	strncpy(base_name, name, 8); /*till xxxx:yy:*/
+	base_name[8] = '\0';
+	/* with no ARI only 3 last bits are used so when the fn is higher than 8
+	 * need to add it to the dev num, so count in the last number will be
+	 * modulo 8 */
+	sprintf(name, "%s%.2d.%d", base_name, (i/8), (i%8));
+}
+
+struct mlx4_port {
+	struct kobject         kobj;
+	struct mlx4_ib_dev    *dev;
+	struct attribute_group pkey_group;
+	struct attribute_group gid_group;
+	struct device_attribute	enable_smi_admin;
+	struct device_attribute	smi_enabled;
+	int		       slave;
+	u8                     port_num;
+};
+
+
+static void mlx4_port_release(struct kobject *kobj)
+{
+	struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+	struct attribute *a;
+	int i;
+
+	for (i = 0; (a = p->pkey_group.attrs[i]); ++i)
+		kfree(a);
+	kfree(p->pkey_group.attrs);
+	for (i = 0; (a = p->gid_group.attrs[i]); ++i)
+		kfree(a);
+	kfree(p->gid_group.attrs);
+	kfree(p);
+}
+
+struct port_attribute {
+	struct attribute attr;
+	ssize_t (*show)(struct mlx4_port *, struct port_attribute *, char *buf);
+	ssize_t (*store)(struct mlx4_port *, struct port_attribute *,
+			 const char *buf, size_t count);
+};
+
+static ssize_t port_attr_show(struct kobject *kobj,
+			      struct attribute *attr, char *buf)
+{
+	struct port_attribute *port_attr =
+		container_of(attr, struct port_attribute, attr);
+	struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+
+	if (!port_attr->show)
+		return -EIO;
+	return port_attr->show(p, port_attr, buf);
+}
+
+static ssize_t port_attr_store(struct kobject *kobj,
+			       struct attribute *attr,
+			       const char *buf, size_t size)
+{
+	struct port_attribute *port_attr =
+		container_of(attr, struct port_attribute, attr);
+	struct mlx4_port *p = container_of(kobj, struct mlx4_port, kobj);
+
+	if (!port_attr->store)
+		return -EIO;
+	return port_attr->store(p, port_attr, buf, size);
+}
+
+static const struct sysfs_ops port_sysfs_ops = {
+	.show = port_attr_show,
+	.store = port_attr_store,
+};
+
+static struct kobj_type port_type = {
+	.release    = mlx4_port_release,
+	.sysfs_ops  = &port_sysfs_ops,
+};
+
+struct port_table_attribute {
+	struct port_attribute	attr;
+	char			name[8];
+	int			index;
+};
+
+static ssize_t show_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
+			      char *buf)
+{
+	struct port_table_attribute *tab_attr =
+		container_of(attr, struct port_table_attribute, attr);
+	ssize_t ret = -ENODEV;
+
+	if (p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1][tab_attr->index] >=
+	    (p->dev->dev->caps.pkey_table_len[p->port_num]))
+		ret = sprintf(buf, "none\n");
+	else
+		ret = sprintf(buf, "%d\n",
+			      p->dev->pkeys.virt2phys_pkey[p->slave]
+			      [p->port_num - 1][tab_attr->index]);
+	return ret;
+}
+
+static ssize_t store_port_pkey(struct mlx4_port *p, struct port_attribute *attr,
+			       const char *buf, size_t count)
+{
+	struct port_table_attribute *tab_attr =
+		container_of(attr, struct port_table_attribute, attr);
+	int idx;
+	int err;
+
+	/* do not allow remapping Dom0 virtual pkey table */
+	if (p->slave == mlx4_master_func_num(p->dev->dev))
+		return -EINVAL;
+
+	if (!strncasecmp(buf, "no", 2))
+		idx = p->dev->dev->phys_caps.pkey_phys_table_len[p->port_num] - 1;
+	else if (sscanf(buf, "%i", &idx) != 1 ||
+		 idx >= p->dev->dev->caps.pkey_table_len[p->port_num] ||
+		 idx < 0)
+		return -EINVAL;
+
+	p->dev->pkeys.virt2phys_pkey[p->slave][p->port_num - 1]
+				    [tab_attr->index] = idx;
+	mlx4_sync_pkey_table(p->dev->dev, p->slave, p->port_num,
+			     tab_attr->index, idx);
+	err = mlx4_gen_pkey_eqe(p->dev->dev, p->slave, p->port_num);
+	if (err) {
+		pr_err("mlx4_gen_pkey_eqe failed for slave %d,"
+		       " port %d, index %d\n", p->slave, p->port_num, idx);
+		return err;
+	}
+	return count;
+}
+
+static ssize_t show_port_gid_idx(struct mlx4_port *p,
+				 struct port_attribute *attr, char *buf)
+{
+	return sprintf(buf, "%d\n", p->slave);
+}
+
+static struct attribute **
+alloc_group_attrs(ssize_t (*show)(struct mlx4_port *,
+				  struct port_attribute *, char *buf),
+		  ssize_t (*store)(struct mlx4_port *, struct port_attribute *,
+				   const char *buf, size_t count),
+		  int len)
+{
+	struct attribute **tab_attr;
+	struct port_table_attribute *element;
+	int i;
+
+	tab_attr = kcalloc(1 + len, sizeof (struct attribute *), GFP_KERNEL);
+	if (!tab_attr)
+		return NULL;
+
+	for (i = 0; i < len; i++) {
+		element = kzalloc(sizeof (struct port_table_attribute),
+				  GFP_KERNEL);
+		if (!element)
+			goto err;
+		if (snprintf(element->name, sizeof (element->name),
+			     "%d", i) >= sizeof (element->name)) {
+			kfree(element);
+			goto err;
+		}
+		sysfs_attr_init(&element->attr.attr);
+		element->attr.attr.name  = element->name;
+		if (store) {
+			element->attr.attr.mode  = S_IWUSR | S_IRUGO;
+			element->attr.store	 = store;
+		} else
+			element->attr.attr.mode  = S_IRUGO;
+
+		element->attr.show       = show;
+		element->index		 = i;
+		tab_attr[i] = &element->attr.attr;
+	}
+	return tab_attr;
+
+err:
+	while (--i >= 0)
+		kfree(tab_attr[i]);
+	kfree(tab_attr);
+	return NULL;
+}
+
+static ssize_t sysfs_show_smi_enabled(struct device *dev,
+				      struct device_attribute *attr, char *buf)
+{
+	struct mlx4_port *p =
+		container_of(attr, struct mlx4_port, smi_enabled);
+	ssize_t len = 0;
+
+	if (mlx4_vf_smi_enabled(p->dev->dev, p->slave, p->port_num))
+		len = sprintf(buf, "%d\n", 1);
+	else
+		len = sprintf(buf, "%d\n", 0);
+
+	return len;
+}
+
+static ssize_t sysfs_show_enable_smi_admin(struct device *dev,
+					   struct device_attribute *attr,
+					   char *buf)
+{
+	struct mlx4_port *p =
+		container_of(attr, struct mlx4_port, enable_smi_admin);
+	ssize_t len = 0;
+
+	if (mlx4_vf_get_enable_smi_admin(p->dev->dev, p->slave, p->port_num))
+		len = sprintf(buf, "%d\n", 1);
+	else
+		len = sprintf(buf, "%d\n", 0);
+
+	return len;
+}
+
+static ssize_t sysfs_store_enable_smi_admin(struct device *dev,
+					    struct device_attribute *attr,
+					    const char *buf, size_t count)
+{
+	struct mlx4_port *p =
+		container_of(attr, struct mlx4_port, enable_smi_admin);
+	int enable;
+
+	if (sscanf(buf, "%i", &enable) != 1 ||
+	    enable < 0 || enable > 1)
+		return -EINVAL;
+
+	if (mlx4_vf_set_enable_smi_admin(p->dev->dev, p->slave, p->port_num, enable))
+		return -EINVAL;
+	return count;
+}
+
+static int add_vf_smi_entries(struct mlx4_port *p)
+{
+	int is_eth = rdma_port_get_link_layer(&p->dev->ib_dev, p->port_num) ==
+			IB_LINK_LAYER_ETHERNET;
+	int ret;
+
+	/* do not display entries if eth transport, or if master */
+	if (is_eth || p->slave == mlx4_master_func_num(p->dev->dev))
+		return 0;
+
+	sysfs_attr_init(&p->smi_enabled.attr);
+	p->smi_enabled.show = sysfs_show_smi_enabled;
+	p->smi_enabled.store = NULL;
+	p->smi_enabled.attr.name = "smi_enabled";
+	p->smi_enabled.attr.mode = 0444;
+	ret = sysfs_create_file(&p->kobj, &p->smi_enabled.attr);
+	if (ret) {
+		pr_err("failed to create smi_enabled\n");
+		return ret;
+	}
+
+	sysfs_attr_init(&p->enable_smi_admin.attr);
+	p->enable_smi_admin.show = sysfs_show_enable_smi_admin;
+	p->enable_smi_admin.store = sysfs_store_enable_smi_admin;
+	p->enable_smi_admin.attr.name = "enable_smi_admin";
+	p->enable_smi_admin.attr.mode = 0644;
+	ret = sysfs_create_file(&p->kobj, &p->enable_smi_admin.attr);
+	if (ret) {
+		pr_err("failed to create enable_smi_admin\n");
+		sysfs_remove_file(&p->kobj, &p->smi_enabled.attr);
+		return ret;
+	}
+	return 0;
+}
+
+static void remove_vf_smi_entries(struct mlx4_port *p)
+{
+	int is_eth = rdma_port_get_link_layer(&p->dev->ib_dev, p->port_num) ==
+			IB_LINK_LAYER_ETHERNET;
+
+	if (is_eth || p->slave == mlx4_master_func_num(p->dev->dev))
+		return;
+
+	sysfs_remove_file(&p->kobj, &p->smi_enabled.attr);
+	sysfs_remove_file(&p->kobj, &p->enable_smi_admin.attr);
+}
+
+static int add_port(struct mlx4_ib_dev *dev, int port_num, int slave)
+{
+	struct mlx4_port *p;
+	int i;
+	int ret;
+
+	p = kzalloc(sizeof *p, GFP_KERNEL);
+	if (!p)
+		return -ENOMEM;
+
+	p->dev = dev;
+	p->port_num = port_num;
+	p->slave = slave;
+
+	ret = kobject_init_and_add(&p->kobj, &port_type,
+				   kobject_get(dev->dev_ports_parent[slave]),
+				   "%d", port_num);
+	if (ret)
+		goto err_alloc;
+
+	p->pkey_group.name  = "pkey_idx";
+	p->pkey_group.attrs =
+		alloc_group_attrs(show_port_pkey, store_port_pkey,
+				  dev->dev->caps.pkey_table_len[port_num]);
+	if (!p->pkey_group.attrs) {
+		ret = -ENOMEM;
+		goto err_alloc;
+	}
+
+	ret = sysfs_create_group(&p->kobj, &p->pkey_group);
+	if (ret)
+		goto err_free_pkey;
+
+	p->gid_group.name  = "gid_idx";
+	p->gid_group.attrs = alloc_group_attrs(show_port_gid_idx, NULL, 1);
+	if (!p->gid_group.attrs) {
+		ret = -ENOMEM;
+		goto err_free_pkey;
+	}
+
+	ret = sysfs_create_group(&p->kobj, &p->gid_group);
+	if (ret)
+		goto err_free_gid;
+
+	ret = add_vf_smi_entries(p);
+	if (ret)
+		goto err_free_gid;
+
+	list_add_tail(&p->kobj.entry, &dev->pkeys.pkey_port_list[slave]);
+	return 0;
+
+err_free_gid:
+	kfree(p->gid_group.attrs[0]);
+	kfree(p->gid_group.attrs);
+
+err_free_pkey:
+	for (i = 0; i < dev->dev->caps.pkey_table_len[port_num]; ++i)
+		kfree(p->pkey_group.attrs[i]);
+	kfree(p->pkey_group.attrs);
+
+err_alloc:
+	kobject_put(dev->dev_ports_parent[slave]);
+	kfree(p);
+	return ret;
+}
+
+static int register_one_pkey_tree(struct mlx4_ib_dev *dev, int slave)
+{
+	char name[32];
+	int err;
+	int port;
+	struct kobject *p, *t;
+	struct mlx4_port *mport;
+	struct mlx4_active_ports actv_ports;
+
+	get_name(dev, name, slave, sizeof name);
+
+	dev->pkeys.device_parent[slave] =
+		kobject_create_and_add(name, kobject_get(dev->iov_parent));
+
+	if (!dev->pkeys.device_parent[slave]) {
+		err = -ENOMEM;
+		goto fail_dev;
+	}
+
+	INIT_LIST_HEAD(&dev->pkeys.pkey_port_list[slave]);
+
+	dev->dev_ports_parent[slave] =
+		kobject_create_and_add("ports",
+				       kobject_get(dev->pkeys.device_parent[slave]));
+
+	if (!dev->dev_ports_parent[slave]) {
+		err = -ENOMEM;
+		goto err_ports;
+	}
+
+	actv_ports = mlx4_get_active_ports(dev->dev, slave);
+
+	for (port = 1; port <= dev->dev->caps.num_ports; ++port) {
+		if (!test_bit(port - 1, actv_ports.ports))
+			continue;
+		err = add_port(dev, port, slave);
+		if (err)
+			goto err_add;
+	}
+	return 0;
+
+err_add:
+	list_for_each_entry_safe(p, t,
+				 &dev->pkeys.pkey_port_list[slave],
+				 entry) {
+		list_del(&p->entry);
+		mport = container_of(p, struct mlx4_port, kobj);
+		sysfs_remove_group(p, &mport->pkey_group);
+		sysfs_remove_group(p, &mport->gid_group);
+		remove_vf_smi_entries(mport);
+		kobject_put(p);
+	}
+	kobject_put(dev->dev_ports_parent[slave]);
+
+err_ports:
+	kobject_put(dev->pkeys.device_parent[slave]);
+	/* extra put for the device_parent create_and_add */
+	kobject_put(dev->pkeys.device_parent[slave]);
+
+fail_dev:
+	kobject_put(dev->iov_parent);
+	return err;
+}
+
+static int register_pkey_tree(struct mlx4_ib_dev *device)
+{
+	int i;
+
+	if (!mlx4_is_master(device->dev))
+		return 0;
+
+	for (i = 0; i <= device->dev->num_vfs; ++i)
+		register_one_pkey_tree(device, i);
+
+	return 0;
+}
+
+static void unregister_pkey_tree(struct mlx4_ib_dev *device)
+{
+	int slave;
+	struct kobject *p, *t;
+	struct mlx4_port *port;
+
+	if (!mlx4_is_master(device->dev))
+		return;
+
+	for (slave = device->dev->num_vfs; slave >= 0; --slave) {
+		list_for_each_entry_safe(p, t,
+					 &device->pkeys.pkey_port_list[slave],
+					 entry) {
+			list_del(&p->entry);
+			port = container_of(p, struct mlx4_port, kobj);
+			sysfs_remove_group(p, &port->pkey_group);
+			sysfs_remove_group(p, &port->gid_group);
+			remove_vf_smi_entries(port);
+			kobject_put(p);
+			kobject_put(device->dev_ports_parent[slave]);
+		}
+		kobject_put(device->dev_ports_parent[slave]);
+		kobject_put(device->pkeys.device_parent[slave]);
+		kobject_put(device->pkeys.device_parent[slave]);
+		kobject_put(device->iov_parent);
+	}
+}
+
+int mlx4_ib_device_register_sysfs(struct mlx4_ib_dev *dev)
+{
+	int i;
+	int ret = 0;
+
+	if (!mlx4_is_master(dev->dev))
+		return 0;
+
+	dev->iov_parent =
+		kobject_create_and_add("iov",
+				       kobject_get(dev->ib_dev.ports_parent->parent));
+	if (!dev->iov_parent) {
+		ret = -ENOMEM;
+		goto err;
+	}
+	dev->ports_parent =
+		kobject_create_and_add("ports",
+				       kobject_get(dev->iov_parent));
+	if (!dev->ports_parent) {
+		ret = -ENOMEM;
+		goto err_ports;
+	}
+
+	for (i = 1; i <= dev->ib_dev.phys_port_cnt; ++i) {
+		ret = add_port_entries(dev, i);
+		if (ret)
+			goto err_add_entries;
+	}
+
+	ret = register_pkey_tree(dev);
+	if (ret)
+		goto err_add_entries;
+	return 0;
+
+err_add_entries:
+	kobject_put(dev->ports_parent);
+
+err_ports:
+	kobject_put(dev->iov_parent);
+err:
+	kobject_put(dev->ib_dev.ports_parent->parent);
+	pr_err("mlx4_ib_device_register_sysfs error (%d)\n", ret);
+	return ret;
+}
+
+static void unregister_alias_guid_tree(struct mlx4_ib_dev *device)
+{
+	struct mlx4_ib_iov_port *p;
+	int i;
+
+	if (!mlx4_is_master(device->dev))
+		return;
+
+	for (i = 0; i < device->dev->caps.num_ports; i++) {
+		p = &device->iov_ports[i];
+		kobject_put(p->admin_alias_parent);
+		kobject_put(p->gids_parent);
+		kobject_put(p->pkeys_parent);
+		kobject_put(p->mcgs_parent);
+		kobject_put(p->cur_port);
+		kobject_put(p->cur_port);
+		kobject_put(p->cur_port);
+		kobject_put(p->cur_port);
+		kobject_put(p->cur_port);
+		kobject_put(p->dev->ports_parent);
+		kfree(p->dentr_ar);
+	}
+}
+
+void mlx4_ib_device_unregister_sysfs(struct mlx4_ib_dev *device)
+{
+	unregister_alias_guid_tree(device);
+	unregister_pkey_tree(device);
+	kobject_put(device->ports_parent);
+	kobject_put(device->iov_parent);
+	kobject_put(device->iov_parent);
+	kobject_put(device->ib_dev.ports_parent->parent);
+}
diff --git a/drivers/infiniband/hw/mlx4/user.h b/drivers/infiniband/hw/mlx4/user.h
index e2d11be4525..07e6769ef43 100644
--- a/drivers/infiniband/hw/mlx4/user.h
+++ b/drivers/infiniband/hw/mlx4/user.h
@@ -1,5 +1,6 @@
 /*
  * Copyright (c) 2007 Cisco Systems, Inc. All rights reserved.
+ * Copyright (c) 2007, 2008 Mellanox Technologies. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -39,7 +40,9 @@
  * Increment this value if any changes that break userspace ABI
  * compatibility are made.
  */
-#define MLX4_IB_UVERBS_ABI_VERSION	3
+
+#define MLX4_IB_UVERBS_NO_DEV_CAPS_ABI_VERSION	3
+#define MLX4_IB_UVERBS_ABI_VERSION		4
 
 /*
  * Make sure that all structs defined in this file remain laid out so
@@ -49,10 +52,18 @@
  * instead.
  */
 
+struct mlx4_ib_alloc_ucontext_resp_v3 {
+	__u32	qp_tab_size;
+	__u16	bf_reg_size;
+	__u16	bf_regs_per_page;
+};
+
 struct mlx4_ib_alloc_ucontext_resp {
+	__u32	dev_caps;
 	__u32	qp_tab_size;
 	__u16	bf_reg_size;
 	__u16	bf_regs_per_page;
+	__u32	cqe_size;
 };
 
 struct mlx4_ib_alloc_pd_resp {
diff --git a/drivers/infiniband/hw/mlx5/Kconfig b/drivers/infiniband/hw/mlx5/Kconfig
new file mode 100644
index 00000000000..10df386c634
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/Kconfig
@@ -0,0 +1,10 @@
+config MLX5_INFINIBAND
+	tristate "Mellanox Connect-IB HCA support"
+	depends on NETDEVICES && ETHERNET && PCI
+	select NET_VENDOR_MELLANOX
+	select MLX5_CORE
+	---help---
+	  This driver provides low-level InfiniBand support for
+	  Mellanox Connect-IB PCI Express host channel adapters (HCAs).
+	  This is required to use InfiniBand protocols such as
+	  IP-over-IB or SRP with these devices.
diff --git a/drivers/infiniband/hw/mlx5/Makefile b/drivers/infiniband/hw/mlx5/Makefile
new file mode 100644
index 00000000000..4ea0135af48
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/Makefile
@@ -0,0 +1,3 @@
+obj-$(CONFIG_MLX5_INFINIBAND)	+= mlx5_ib.o
+
+mlx5_ib-y :=	main.o cq.o doorbell.o qp.o mem.o srq.o mr.o ah.o mad.o
diff --git a/drivers/infiniband/hw/mlx5/ah.c b/drivers/infiniband/hw/mlx5/ah.c
new file mode 100644
index 00000000000..39ab0caefdf
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/ah.c
@@ -0,0 +1,92 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "mlx5_ib.h"
+
+struct ib_ah *create_ib_ah(struct ib_ah_attr *ah_attr,
+			   struct mlx5_ib_ah *ah)
+{
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		memcpy(ah->av.rgid, &ah_attr->grh.dgid, 16);
+		ah->av.grh_gid_fl = cpu_to_be32(ah_attr->grh.flow_label |
+						(1 << 30) |
+						ah_attr->grh.sgid_index << 20);
+		ah->av.hop_limit = ah_attr->grh.hop_limit;
+		ah->av.tclass = ah_attr->grh.traffic_class;
+	}
+
+	ah->av.rlid = cpu_to_be16(ah_attr->dlid);
+	ah->av.fl_mlid = ah_attr->src_path_bits & 0x7f;
+	ah->av.stat_rate_sl = (ah_attr->static_rate << 4) | (ah_attr->sl & 0xf);
+
+	return &ah->ibah;
+}
+
+struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr)
+{
+	struct mlx5_ib_ah *ah;
+
+	ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	return create_ib_ah(ah_attr, ah); /* never fails */
+}
+
+int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
+{
+	struct mlx5_ib_ah *ah = to_mah(ibah);
+	u32 tmp;
+
+	memset(ah_attr, 0, sizeof(*ah_attr));
+
+	tmp = be32_to_cpu(ah->av.grh_gid_fl);
+	if (tmp & (1 << 30)) {
+		ah_attr->ah_flags = IB_AH_GRH;
+		ah_attr->grh.sgid_index = (tmp >> 20) & 0xff;
+		ah_attr->grh.flow_label = tmp & 0xfffff;
+		memcpy(&ah_attr->grh.dgid, ah->av.rgid, 16);
+		ah_attr->grh.hop_limit = ah->av.hop_limit;
+		ah_attr->grh.traffic_class = ah->av.tclass;
+	}
+	ah_attr->dlid = be16_to_cpu(ah->av.rlid);
+	ah_attr->static_rate = ah->av.stat_rate_sl >> 4;
+	ah_attr->sl = ah->av.stat_rate_sl & 0xf;
+
+	return 0;
+}
+
+int mlx5_ib_destroy_ah(struct ib_ah *ah)
+{
+	kfree(to_mah(ah));
+	return 0;
+}
diff --git a/drivers/infiniband/hw/mlx5/cq.c b/drivers/infiniband/hw/mlx5/cq.c
new file mode 100644
index 00000000000..8ae4f896cb4
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/cq.c
@@ -0,0 +1,1187 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/kref.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_user_verbs.h>
+#include "mlx5_ib.h"
+#include "user.h"
+
+static void mlx5_ib_cq_comp(struct mlx5_core_cq *cq)
+{
+	struct ib_cq *ibcq = &to_mibcq(cq)->ibcq;
+
+	ibcq->comp_handler(ibcq, ibcq->cq_context);
+}
+
+static void mlx5_ib_cq_event(struct mlx5_core_cq *mcq, enum mlx5_event type)
+{
+	struct mlx5_ib_cq *cq = container_of(mcq, struct mlx5_ib_cq, mcq);
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct ib_cq *ibcq = &cq->ibcq;
+	struct ib_event event;
+
+	if (type != MLX5_EVENT_TYPE_CQ_ERROR) {
+		mlx5_ib_warn(dev, "Unexpected event type %d on CQ %06x\n",
+			     type, mcq->cqn);
+		return;
+	}
+
+	if (ibcq->event_handler) {
+		event.device     = &dev->ib_dev;
+		event.event      = IB_EVENT_CQ_ERR;
+		event.element.cq = ibcq;
+		ibcq->event_handler(&event, ibcq->cq_context);
+	}
+}
+
+static void *get_cqe_from_buf(struct mlx5_ib_cq_buf *buf, int n, int size)
+{
+	return mlx5_buf_offset(&buf->buf, n * size);
+}
+
+static void *get_cqe(struct mlx5_ib_cq *cq, int n)
+{
+	return get_cqe_from_buf(&cq->buf, n, cq->mcq.cqe_sz);
+}
+
+static u8 sw_ownership_bit(int n, int nent)
+{
+	return (n & nent) ? 1 : 0;
+}
+
+static void *get_sw_cqe(struct mlx5_ib_cq *cq, int n)
+{
+	void *cqe = get_cqe(cq, n & cq->ibcq.cqe);
+	struct mlx5_cqe64 *cqe64;
+
+	cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+
+	if (likely((cqe64->op_own) >> 4 != MLX5_CQE_INVALID) &&
+	    !((cqe64->op_own & MLX5_CQE_OWNER_MASK) ^ !!(n & (cq->ibcq.cqe + 1)))) {
+		return cqe;
+	} else {
+		return NULL;
+	}
+}
+
+static void *next_cqe_sw(struct mlx5_ib_cq *cq)
+{
+	return get_sw_cqe(cq, cq->mcq.cons_index);
+}
+
+static enum ib_wc_opcode get_umr_comp(struct mlx5_ib_wq *wq, int idx)
+{
+	switch (wq->wr_data[idx]) {
+	case MLX5_IB_WR_UMR:
+		return 0;
+
+	case IB_WR_LOCAL_INV:
+		return IB_WC_LOCAL_INV;
+
+	case IB_WR_FAST_REG_MR:
+		return IB_WC_FAST_REG_MR;
+
+	default:
+		pr_warn("unknown completion status\n");
+		return 0;
+	}
+}
+
+static void handle_good_req(struct ib_wc *wc, struct mlx5_cqe64 *cqe,
+			    struct mlx5_ib_wq *wq, int idx)
+{
+	wc->wc_flags = 0;
+	switch (be32_to_cpu(cqe->sop_drop_qpn) >> 24) {
+	case MLX5_OPCODE_RDMA_WRITE_IMM:
+		wc->wc_flags |= IB_WC_WITH_IMM;
+	case MLX5_OPCODE_RDMA_WRITE:
+		wc->opcode    = IB_WC_RDMA_WRITE;
+		break;
+	case MLX5_OPCODE_SEND_IMM:
+		wc->wc_flags |= IB_WC_WITH_IMM;
+	case MLX5_OPCODE_SEND:
+	case MLX5_OPCODE_SEND_INVAL:
+		wc->opcode    = IB_WC_SEND;
+		break;
+	case MLX5_OPCODE_RDMA_READ:
+		wc->opcode    = IB_WC_RDMA_READ;
+		wc->byte_len  = be32_to_cpu(cqe->byte_cnt);
+		break;
+	case MLX5_OPCODE_ATOMIC_CS:
+		wc->opcode    = IB_WC_COMP_SWAP;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_FA:
+		wc->opcode    = IB_WC_FETCH_ADD;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_MASKED_CS:
+		wc->opcode    = IB_WC_MASKED_COMP_SWAP;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_ATOMIC_MASKED_FA:
+		wc->opcode    = IB_WC_MASKED_FETCH_ADD;
+		wc->byte_len  = 8;
+		break;
+	case MLX5_OPCODE_BIND_MW:
+		wc->opcode    = IB_WC_BIND_MW;
+		break;
+	case MLX5_OPCODE_UMR:
+		wc->opcode = get_umr_comp(wq, idx);
+		break;
+	}
+}
+
+enum {
+	MLX5_GRH_IN_BUFFER = 1,
+	MLX5_GRH_IN_CQE	   = 2,
+};
+
+static void handle_responder(struct ib_wc *wc, struct mlx5_cqe64 *cqe,
+			     struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->ibqp.device);
+	struct mlx5_ib_srq *srq;
+	struct mlx5_ib_wq *wq;
+	u16 wqe_ctr;
+	u8 g;
+
+	if (qp->ibqp.srq || qp->ibqp.xrcd) {
+		struct mlx5_core_srq *msrq = NULL;
+
+		if (qp->ibqp.xrcd) {
+			msrq = mlx5_core_get_srq(&dev->mdev,
+						 be32_to_cpu(cqe->srqn));
+			srq = to_mibsrq(msrq);
+		} else {
+			srq = to_msrq(qp->ibqp.srq);
+		}
+		if (srq) {
+			wqe_ctr = be16_to_cpu(cqe->wqe_counter);
+			wc->wr_id = srq->wrid[wqe_ctr];
+			mlx5_ib_free_srq_wqe(srq, wqe_ctr);
+			if (msrq && atomic_dec_and_test(&msrq->refcount))
+				complete(&msrq->free);
+		}
+	} else {
+		wq	  = &qp->rq;
+		wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+		++wq->tail;
+	}
+	wc->byte_len = be32_to_cpu(cqe->byte_cnt);
+
+	switch (cqe->op_own >> 4) {
+	case MLX5_CQE_RESP_WR_IMM:
+		wc->opcode	= IB_WC_RECV_RDMA_WITH_IMM;
+		wc->wc_flags	= IB_WC_WITH_IMM;
+		wc->ex.imm_data = cqe->imm_inval_pkey;
+		break;
+	case MLX5_CQE_RESP_SEND:
+		wc->opcode   = IB_WC_RECV;
+		wc->wc_flags = 0;
+		break;
+	case MLX5_CQE_RESP_SEND_IMM:
+		wc->opcode	= IB_WC_RECV;
+		wc->wc_flags	= IB_WC_WITH_IMM;
+		wc->ex.imm_data = cqe->imm_inval_pkey;
+		break;
+	case MLX5_CQE_RESP_SEND_INV:
+		wc->opcode	= IB_WC_RECV;
+		wc->wc_flags	= IB_WC_WITH_INVALIDATE;
+		wc->ex.invalidate_rkey = be32_to_cpu(cqe->imm_inval_pkey);
+		break;
+	}
+	wc->slid	   = be16_to_cpu(cqe->slid);
+	wc->sl		   = (be32_to_cpu(cqe->flags_rqpn) >> 24) & 0xf;
+	wc->src_qp	   = be32_to_cpu(cqe->flags_rqpn) & 0xffffff;
+	wc->dlid_path_bits = cqe->ml_path;
+	g = (be32_to_cpu(cqe->flags_rqpn) >> 28) & 3;
+	wc->wc_flags |= g ? IB_WC_GRH : 0;
+	wc->pkey_index     = be32_to_cpu(cqe->imm_inval_pkey) & 0xffff;
+}
+
+static void dump_cqe(struct mlx5_ib_dev *dev, struct mlx5_err_cqe *cqe)
+{
+	__be32 *p = (__be32 *)cqe;
+	int i;
+
+	mlx5_ib_warn(dev, "dump error cqe\n");
+	for (i = 0; i < sizeof(*cqe) / 16; i++, p += 4)
+		pr_info("%08x %08x %08x %08x\n", be32_to_cpu(p[0]),
+			be32_to_cpu(p[1]), be32_to_cpu(p[2]),
+			be32_to_cpu(p[3]));
+}
+
+static void mlx5_handle_error_cqe(struct mlx5_ib_dev *dev,
+				  struct mlx5_err_cqe *cqe,
+				  struct ib_wc *wc)
+{
+	int dump = 1;
+
+	switch (cqe->syndrome) {
+	case MLX5_CQE_SYNDROME_LOCAL_LENGTH_ERR:
+		wc->status = IB_WC_LOC_LEN_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_QP_OP_ERR:
+		wc->status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_PROT_ERR:
+		wc->status = IB_WC_LOC_PROT_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_WR_FLUSH_ERR:
+		dump = 0;
+		wc->status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_MW_BIND_ERR:
+		wc->status = IB_WC_MW_BIND_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_BAD_RESP_ERR:
+		wc->status = IB_WC_BAD_RESP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_LOCAL_ACCESS_ERR:
+		wc->status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_INVAL_REQ_ERR:
+		wc->status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_ACCESS_ERR:
+		wc->status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_OP_ERR:
+		wc->status = IB_WC_REM_OP_ERR;
+		break;
+	case MLX5_CQE_SYNDROME_TRANSPORT_RETRY_EXC_ERR:
+		wc->status = IB_WC_RETRY_EXC_ERR;
+		dump = 0;
+		break;
+	case MLX5_CQE_SYNDROME_RNR_RETRY_EXC_ERR:
+		wc->status = IB_WC_RNR_RETRY_EXC_ERR;
+		dump = 0;
+		break;
+	case MLX5_CQE_SYNDROME_REMOTE_ABORTED_ERR:
+		wc->status = IB_WC_REM_ABORT_ERR;
+		break;
+	default:
+		wc->status = IB_WC_GENERAL_ERR;
+		break;
+	}
+
+	wc->vendor_err = cqe->vendor_err_synd;
+	if (dump)
+		dump_cqe(dev, cqe);
+}
+
+static int is_atomic_response(struct mlx5_ib_qp *qp, uint16_t idx)
+{
+	/* TBD: waiting decision
+	*/
+	return 0;
+}
+
+static void *mlx5_get_atomic_laddr(struct mlx5_ib_qp *qp, uint16_t idx)
+{
+	struct mlx5_wqe_data_seg *dpseg;
+	void *addr;
+
+	dpseg = mlx5_get_send_wqe(qp, idx) + sizeof(struct mlx5_wqe_ctrl_seg) +
+		sizeof(struct mlx5_wqe_raddr_seg) +
+		sizeof(struct mlx5_wqe_atomic_seg);
+	addr = (void *)(unsigned long)be64_to_cpu(dpseg->addr);
+	return addr;
+}
+
+static void handle_atomic(struct mlx5_ib_qp *qp, struct mlx5_cqe64 *cqe64,
+			  uint16_t idx)
+{
+	void *addr;
+	int byte_count;
+	int i;
+
+	if (!is_atomic_response(qp, idx))
+		return;
+
+	byte_count = be32_to_cpu(cqe64->byte_cnt);
+	addr = mlx5_get_atomic_laddr(qp, idx);
+
+	if (byte_count == 4) {
+		*(uint32_t *)addr = be32_to_cpu(*((__be32 *)addr));
+	} else {
+		for (i = 0; i < byte_count; i += 8) {
+			*(uint64_t *)addr = be64_to_cpu(*((__be64 *)addr));
+			addr += 8;
+		}
+	}
+
+	return;
+}
+
+static void handle_atomics(struct mlx5_ib_qp *qp, struct mlx5_cqe64 *cqe64,
+			   u16 tail, u16 head)
+{
+	int idx;
+
+	do {
+		idx = tail & (qp->sq.wqe_cnt - 1);
+		handle_atomic(qp, cqe64, idx);
+		if (idx == head)
+			break;
+
+		tail = qp->sq.w_list[idx].next;
+	} while (1);
+	tail = qp->sq.w_list[idx].next;
+	qp->sq.last_poll = tail;
+}
+
+static void free_cq_buf(struct mlx5_ib_dev *dev, struct mlx5_ib_cq_buf *buf)
+{
+	mlx5_buf_free(&dev->mdev, &buf->buf);
+}
+
+static void get_sig_err_item(struct mlx5_sig_err_cqe *cqe,
+			     struct ib_sig_err *item)
+{
+	u16 syndrome = be16_to_cpu(cqe->syndrome);
+
+#define GUARD_ERR   (1 << 13)
+#define APPTAG_ERR  (1 << 12)
+#define REFTAG_ERR  (1 << 11)
+
+	if (syndrome & GUARD_ERR) {
+		item->err_type = IB_SIG_BAD_GUARD;
+		item->expected = be32_to_cpu(cqe->expected_trans_sig) >> 16;
+		item->actual = be32_to_cpu(cqe->actual_trans_sig) >> 16;
+	} else
+	if (syndrome & REFTAG_ERR) {
+		item->err_type = IB_SIG_BAD_REFTAG;
+		item->expected = be32_to_cpu(cqe->expected_reftag);
+		item->actual = be32_to_cpu(cqe->actual_reftag);
+	} else
+	if (syndrome & APPTAG_ERR) {
+		item->err_type = IB_SIG_BAD_APPTAG;
+		item->expected = be32_to_cpu(cqe->expected_trans_sig) & 0xffff;
+		item->actual = be32_to_cpu(cqe->actual_trans_sig) & 0xffff;
+	} else {
+		pr_err("Got signature completion error with bad syndrome %04x\n",
+		       syndrome);
+	}
+
+	item->sig_err_offset = be64_to_cpu(cqe->err_offset);
+	item->key = be32_to_cpu(cqe->mkey);
+}
+
+static int mlx5_poll_one(struct mlx5_ib_cq *cq,
+			 struct mlx5_ib_qp **cur_qp,
+			 struct ib_wc *wc)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct mlx5_err_cqe *err_cqe;
+	struct mlx5_cqe64 *cqe64;
+	struct mlx5_core_qp *mqp;
+	struct mlx5_ib_wq *wq;
+	struct mlx5_sig_err_cqe *sig_err_cqe;
+	struct mlx5_core_mr *mmr;
+	struct mlx5_ib_mr *mr;
+	uint8_t opcode;
+	uint32_t qpn;
+	u16 wqe_ctr;
+	void *cqe;
+	int idx;
+
+repoll:
+	cqe = next_cqe_sw(cq);
+	if (!cqe)
+		return -EAGAIN;
+
+	cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+
+	++cq->mcq.cons_index;
+
+	/* Make sure we read CQ entry contents after we've checked the
+	 * ownership bit.
+	 */
+	rmb();
+
+	opcode = cqe64->op_own >> 4;
+	if (unlikely(opcode == MLX5_CQE_RESIZE_CQ)) {
+		if (likely(cq->resize_buf)) {
+			free_cq_buf(dev, &cq->buf);
+			cq->buf = *cq->resize_buf;
+			kfree(cq->resize_buf);
+			cq->resize_buf = NULL;
+			goto repoll;
+		} else {
+			mlx5_ib_warn(dev, "unexpected resize cqe\n");
+		}
+	}
+
+	qpn = ntohl(cqe64->sop_drop_qpn) & 0xffffff;
+	if (!*cur_qp || (qpn != (*cur_qp)->ibqp.qp_num)) {
+		/* We do not have to take the QP table lock here,
+		 * because CQs will be locked while QPs are removed
+		 * from the table.
+		 */
+		mqp = __mlx5_qp_lookup(&dev->mdev, qpn);
+		if (unlikely(!mqp)) {
+			mlx5_ib_warn(dev, "CQE@CQ %06x for unknown QPN %6x\n",
+				     cq->mcq.cqn, qpn);
+			return -EINVAL;
+		}
+
+		*cur_qp = to_mibqp(mqp);
+	}
+
+	wc->qp  = &(*cur_qp)->ibqp;
+	switch (opcode) {
+	case MLX5_CQE_REQ:
+		wq = &(*cur_qp)->sq;
+		wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+		idx = wqe_ctr & (wq->wqe_cnt - 1);
+		handle_good_req(wc, cqe64, wq, idx);
+		handle_atomics(*cur_qp, cqe64, wq->last_poll, idx);
+		wc->wr_id = wq->wrid[idx];
+		wq->tail = wq->wqe_head[idx] + 1;
+		wc->status = IB_WC_SUCCESS;
+		break;
+	case MLX5_CQE_RESP_WR_IMM:
+	case MLX5_CQE_RESP_SEND:
+	case MLX5_CQE_RESP_SEND_IMM:
+	case MLX5_CQE_RESP_SEND_INV:
+		handle_responder(wc, cqe64, *cur_qp);
+		wc->status = IB_WC_SUCCESS;
+		break;
+	case MLX5_CQE_RESIZE_CQ:
+		break;
+	case MLX5_CQE_REQ_ERR:
+	case MLX5_CQE_RESP_ERR:
+		err_cqe = (struct mlx5_err_cqe *)cqe64;
+		mlx5_handle_error_cqe(dev, err_cqe, wc);
+		mlx5_ib_dbg(dev, "%s error cqe on cqn 0x%x:\n",
+			    opcode == MLX5_CQE_REQ_ERR ?
+			    "Requestor" : "Responder", cq->mcq.cqn);
+		mlx5_ib_dbg(dev, "syndrome 0x%x, vendor syndrome 0x%x\n",
+			    err_cqe->syndrome, err_cqe->vendor_err_synd);
+		if (opcode == MLX5_CQE_REQ_ERR) {
+			wq = &(*cur_qp)->sq;
+			wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+			idx = wqe_ctr & (wq->wqe_cnt - 1);
+			wc->wr_id = wq->wrid[idx];
+			wq->tail = wq->wqe_head[idx] + 1;
+		} else {
+			struct mlx5_ib_srq *srq;
+
+			if ((*cur_qp)->ibqp.srq) {
+				srq = to_msrq((*cur_qp)->ibqp.srq);
+				wqe_ctr = be16_to_cpu(cqe64->wqe_counter);
+				wc->wr_id = srq->wrid[wqe_ctr];
+				mlx5_ib_free_srq_wqe(srq, wqe_ctr);
+			} else {
+				wq = &(*cur_qp)->rq;
+				wc->wr_id = wq->wrid[wq->tail & (wq->wqe_cnt - 1)];
+				++wq->tail;
+			}
+		}
+		break;
+	case MLX5_CQE_SIG_ERR:
+		sig_err_cqe = (struct mlx5_sig_err_cqe *)cqe64;
+
+		read_lock(&dev->mdev.priv.mr_table.lock);
+		mmr = __mlx5_mr_lookup(&dev->mdev,
+				       mlx5_base_mkey(be32_to_cpu(sig_err_cqe->mkey)));
+		if (unlikely(!mmr)) {
+			read_unlock(&dev->mdev.priv.mr_table.lock);
+			mlx5_ib_warn(dev, "CQE@CQ %06x for unknown MR %6x\n",
+				     cq->mcq.cqn, be32_to_cpu(sig_err_cqe->mkey));
+			return -EINVAL;
+		}
+
+		mr = to_mibmr(mmr);
+		get_sig_err_item(sig_err_cqe, &mr->sig->err_item);
+		mr->sig->sig_err_exists = true;
+		mr->sig->sigerr_count++;
+
+		mlx5_ib_warn(dev, "CQN: 0x%x Got SIGERR on key: 0x%x err_type %x err_offset %llx expected %x actual %x\n",
+			     cq->mcq.cqn, mr->sig->err_item.key,
+			     mr->sig->err_item.err_type,
+			     mr->sig->err_item.sig_err_offset,
+			     mr->sig->err_item.expected,
+			     mr->sig->err_item.actual);
+
+		read_unlock(&dev->mdev.priv.mr_table.lock);
+		goto repoll;
+	}
+
+	return 0;
+}
+
+int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	struct mlx5_ib_cq *cq = to_mcq(ibcq);
+	struct mlx5_ib_qp *cur_qp = NULL;
+	unsigned long flags;
+	int npolled;
+	int err = 0;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	for (npolled = 0; npolled < num_entries; npolled++) {
+		err = mlx5_poll_one(cq, &cur_qp, wc + npolled);
+		if (err)
+			break;
+	}
+
+	if (npolled)
+		mlx5_cq_set_ci(&cq->mcq);
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	if (err == 0 || err == -EAGAIN)
+		return npolled;
+	else
+		return err;
+}
+
+int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags)
+{
+	mlx5_cq_arm(&to_mcq(ibcq)->mcq,
+		    (flags & IB_CQ_SOLICITED_MASK) == IB_CQ_SOLICITED ?
+		    MLX5_CQ_DB_REQ_NOT_SOL : MLX5_CQ_DB_REQ_NOT,
+		    to_mdev(ibcq->device)->mdev.priv.uuari.uars[0].map,
+		    MLX5_GET_DOORBELL_LOCK(&to_mdev(ibcq->device)->mdev.priv.cq_uar_lock));
+
+	return 0;
+}
+
+static int alloc_cq_buf(struct mlx5_ib_dev *dev, struct mlx5_ib_cq_buf *buf,
+			int nent, int cqe_size)
+{
+	int err;
+
+	err = mlx5_buf_alloc(&dev->mdev, nent * cqe_size,
+			     PAGE_SIZE * 2, &buf->buf);
+	if (err)
+		return err;
+
+	buf->cqe_size = cqe_size;
+	buf->nent = nent;
+
+	return 0;
+}
+
+static int create_cq_user(struct mlx5_ib_dev *dev, struct ib_udata *udata,
+			  struct ib_ucontext *context, struct mlx5_ib_cq *cq,
+			  int entries, struct mlx5_create_cq_mbox_in **cqb,
+			  int *cqe_size, int *index, int *inlen)
+{
+	struct mlx5_ib_create_cq ucmd;
+	size_t ucmdlen;
+	int page_shift;
+	int npages;
+	int ncont;
+	int err;
+
+	ucmdlen =
+		(udata->inlen - sizeof(struct ib_uverbs_cmd_hdr) <
+		 sizeof(ucmd)) ? (sizeof(ucmd) -
+				  sizeof(ucmd.reserved)) : sizeof(ucmd);
+
+	if (ib_copy_from_udata(&ucmd, udata, ucmdlen))
+		return -EFAULT;
+
+	if (ucmdlen == sizeof(ucmd) &&
+	    ucmd.reserved != 0)
+		return -EINVAL;
+
+	if (ucmd.cqe_size != 64 && ucmd.cqe_size != 128)
+		return -EINVAL;
+
+	*cqe_size = ucmd.cqe_size;
+
+	cq->buf.umem = ib_umem_get(context, ucmd.buf_addr,
+				   entries * ucmd.cqe_size,
+				   IB_ACCESS_LOCAL_WRITE, 1);
+	if (IS_ERR(cq->buf.umem)) {
+		err = PTR_ERR(cq->buf.umem);
+		return err;
+	}
+
+	err = mlx5_ib_db_map_user(to_mucontext(context), ucmd.db_addr,
+				  &cq->db);
+	if (err)
+		goto err_umem;
+
+	mlx5_ib_cont_pages(cq->buf.umem, ucmd.buf_addr, &npages, &page_shift,
+			   &ncont, NULL);
+	mlx5_ib_dbg(dev, "addr 0x%llx, size %u, npages %d, page_shift %d, ncont %d\n",
+		    ucmd.buf_addr, entries * ucmd.cqe_size, npages, page_shift, ncont);
+
+	*inlen = sizeof(**cqb) + sizeof(*(*cqb)->pas) * ncont;
+	*cqb = mlx5_vzalloc(*inlen);
+	if (!*cqb) {
+		err = -ENOMEM;
+		goto err_db;
+	}
+	mlx5_ib_populate_pas(dev, cq->buf.umem, page_shift, (*cqb)->pas, 0);
+	(*cqb)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+
+	*index = to_mucontext(context)->uuari.uars[0].index;
+
+	return 0;
+
+err_db:
+	mlx5_ib_db_unmap_user(to_mucontext(context), &cq->db);
+
+err_umem:
+	ib_umem_release(cq->buf.umem);
+	return err;
+}
+
+static void destroy_cq_user(struct mlx5_ib_cq *cq, struct ib_ucontext *context)
+{
+	mlx5_ib_db_unmap_user(to_mucontext(context), &cq->db);
+	ib_umem_release(cq->buf.umem);
+}
+
+static void init_cq_buf(struct mlx5_ib_cq *cq, struct mlx5_ib_cq_buf *buf)
+{
+	int i;
+	void *cqe;
+	struct mlx5_cqe64 *cqe64;
+
+	for (i = 0; i < buf->nent; i++) {
+		cqe = get_cqe_from_buf(buf, i, buf->cqe_size);
+		cqe64 = buf->cqe_size == 64 ? cqe : cqe + 64;
+		cqe64->op_own = MLX5_CQE_INVALID << 4;
+	}
+}
+
+static int create_cq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+			    int entries, int cqe_size,
+			    struct mlx5_create_cq_mbox_in **cqb,
+			    int *index, int *inlen)
+{
+	int err;
+
+	err = mlx5_db_alloc(&dev->mdev, &cq->db);
+	if (err)
+		return err;
+
+	cq->mcq.set_ci_db  = cq->db.db;
+	cq->mcq.arm_db     = cq->db.db + 1;
+	*cq->mcq.set_ci_db = 0;
+	*cq->mcq.arm_db    = 0;
+	cq->mcq.cqe_sz = cqe_size;
+
+	err = alloc_cq_buf(dev, &cq->buf, entries, cqe_size);
+	if (err)
+		goto err_db;
+
+	init_cq_buf(cq, &cq->buf);
+
+	*inlen = sizeof(**cqb) + sizeof(*(*cqb)->pas) * cq->buf.buf.npages;
+	*cqb = mlx5_vzalloc(*inlen);
+	if (!*cqb) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	mlx5_fill_page_array(&cq->buf.buf, (*cqb)->pas);
+
+	(*cqb)->ctx.log_pg_sz = cq->buf.buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	*index = dev->mdev.priv.uuari.uars[0].index;
+
+	return 0;
+
+err_buf:
+	free_cq_buf(dev, &cq->buf);
+
+err_db:
+	mlx5_db_free(&dev->mdev, &cq->db);
+	return err;
+}
+
+static void destroy_cq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq)
+{
+	free_cq_buf(dev, &cq->buf);
+	mlx5_db_free(&dev->mdev, &cq->db);
+}
+
+struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, int entries,
+				int vector, struct ib_ucontext *context,
+				struct ib_udata *udata)
+{
+	struct mlx5_create_cq_mbox_in *cqb = NULL;
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_cq *cq;
+	int uninitialized_var(index);
+	int uninitialized_var(inlen);
+	int cqe_size;
+	int irqn;
+	int eqn;
+	int err;
+
+	if (entries < 0)
+		return ERR_PTR(-EINVAL);
+
+	entries = roundup_pow_of_two(entries + 1);
+	if (entries > dev->mdev.caps.max_cqes)
+		return ERR_PTR(-EINVAL);
+
+	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-ENOMEM);
+
+	cq->ibcq.cqe = entries - 1;
+	mutex_init(&cq->resize_mutex);
+	spin_lock_init(&cq->lock);
+	cq->resize_buf = NULL;
+	cq->resize_umem = NULL;
+
+	if (context) {
+		err = create_cq_user(dev, udata, context, cq, entries,
+				     &cqb, &cqe_size, &index, &inlen);
+		if (err)
+			goto err_create;
+	} else {
+		/* for now choose 64 bytes till we have a proper interface */
+		cqe_size = 64;
+		err = create_cq_kernel(dev, cq, entries, cqe_size, &cqb,
+				       &index, &inlen);
+		if (err)
+			goto err_create;
+	}
+
+	cq->cqe_size = cqe_size;
+	cqb->ctx.cqe_sz_flags = cqe_sz_to_mlx_sz(cqe_size) << 5;
+	cqb->ctx.log_sz_usr_page = cpu_to_be32((ilog2(entries) << 24) | index);
+	err = mlx5_vector2eqn(dev, vector, &eqn, &irqn);
+	if (err)
+		goto err_cqb;
+
+	cqb->ctx.c_eqn = cpu_to_be16(eqn);
+	cqb->ctx.db_record_addr = cpu_to_be64(cq->db.dma);
+
+	err = mlx5_core_create_cq(&dev->mdev, &cq->mcq, cqb, inlen);
+	if (err)
+		goto err_cqb;
+
+	mlx5_ib_dbg(dev, "cqn 0x%x\n", cq->mcq.cqn);
+	cq->mcq.irqn = irqn;
+	cq->mcq.comp  = mlx5_ib_cq_comp;
+	cq->mcq.event = mlx5_ib_cq_event;
+
+	if (context)
+		if (ib_copy_to_udata(udata, &cq->mcq.cqn, sizeof(__u32))) {
+			err = -EFAULT;
+			goto err_cmd;
+		}
+
+
+	mlx5_vfree(cqb);
+	return &cq->ibcq;
+
+err_cmd:
+	mlx5_core_destroy_cq(&dev->mdev, &cq->mcq);
+
+err_cqb:
+	mlx5_vfree(cqb);
+	if (context)
+		destroy_cq_user(cq, context);
+	else
+		destroy_cq_kernel(dev, cq);
+
+err_create:
+	kfree(cq);
+
+	return ERR_PTR(err);
+}
+
+
+int mlx5_ib_destroy_cq(struct ib_cq *cq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->device);
+	struct mlx5_ib_cq *mcq = to_mcq(cq);
+	struct ib_ucontext *context = NULL;
+
+	if (cq->uobject)
+		context = cq->uobject->context;
+
+	mlx5_core_destroy_cq(&dev->mdev, &mcq->mcq);
+	if (context)
+		destroy_cq_user(mcq, context);
+	else
+		destroy_cq_kernel(dev, mcq);
+
+	kfree(mcq);
+
+	return 0;
+}
+
+static int is_equal_rsn(struct mlx5_cqe64 *cqe64, u32 rsn)
+{
+	return rsn == (ntohl(cqe64->sop_drop_qpn) & 0xffffff);
+}
+
+void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 rsn, struct mlx5_ib_srq *srq)
+{
+	struct mlx5_cqe64 *cqe64, *dest64;
+	void *cqe, *dest;
+	u32 prod_index;
+	int nfreed = 0;
+	u8 owner_bit;
+
+	if (!cq)
+		return;
+
+	/* First we need to find the current producer index, so we
+	 * know where to start cleaning from.  It doesn't matter if HW
+	 * adds new entries after this loop -- the QP we're worried
+	 * about is already in RESET, so the new entries won't come
+	 * from our QP and therefore don't need to be checked.
+	 */
+	for (prod_index = cq->mcq.cons_index; get_sw_cqe(cq, prod_index); prod_index++)
+		if (prod_index == cq->mcq.cons_index + cq->ibcq.cqe)
+			break;
+
+	/* Now sweep backwards through the CQ, removing CQ entries
+	 * that match our QP by copying older entries on top of them.
+	 */
+	while ((int) --prod_index - (int) cq->mcq.cons_index >= 0) {
+		cqe = get_cqe(cq, prod_index & cq->ibcq.cqe);
+		cqe64 = (cq->mcq.cqe_sz == 64) ? cqe : cqe + 64;
+		if (is_equal_rsn(cqe64, rsn)) {
+			if (srq && (ntohl(cqe64->srqn) & 0xffffff))
+				mlx5_ib_free_srq_wqe(srq, be16_to_cpu(cqe64->wqe_counter));
+			++nfreed;
+		} else if (nfreed) {
+			dest = get_cqe(cq, (prod_index + nfreed) & cq->ibcq.cqe);
+			dest64 = (cq->mcq.cqe_sz == 64) ? dest : dest + 64;
+			owner_bit = dest64->op_own & MLX5_CQE_OWNER_MASK;
+			memcpy(dest, cqe, cq->mcq.cqe_sz);
+			dest64->op_own = owner_bit |
+				(dest64->op_own & ~MLX5_CQE_OWNER_MASK);
+		}
+	}
+
+	if (nfreed) {
+		cq->mcq.cons_index += nfreed;
+		/* Make sure update of buffer contents is done before
+		 * updating consumer index.
+		 */
+		wmb();
+		mlx5_cq_set_ci(&cq->mcq);
+	}
+}
+
+void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq)
+{
+	if (!cq)
+		return;
+
+	spin_lock_irq(&cq->lock);
+	__mlx5_ib_cq_clean(cq, qpn, srq);
+	spin_unlock_irq(&cq->lock);
+}
+
+int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period)
+{
+	struct mlx5_modify_cq_mbox_in *in;
+	struct mlx5_ib_dev *dev = to_mdev(cq->device);
+	struct mlx5_ib_cq *mcq = to_mcq(cq);
+	int err;
+	u32 fsel;
+
+	if (!(dev->mdev.caps.flags & MLX5_DEV_CAP_FLAG_CQ_MODER))
+		return -ENOSYS;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	in->cqn = cpu_to_be32(mcq->mcq.cqn);
+	fsel = (MLX5_CQ_MODIFY_PERIOD | MLX5_CQ_MODIFY_COUNT);
+	in->ctx.cq_period = cpu_to_be16(cq_period);
+	in->ctx.cq_max_count = cpu_to_be16(cq_count);
+	in->field_select = cpu_to_be32(fsel);
+	err = mlx5_core_modify_cq(&dev->mdev, &mcq->mcq, in, sizeof(*in));
+	kfree(in);
+
+	if (err)
+		mlx5_ib_warn(dev, "modify cq 0x%x failed\n", mcq->mcq.cqn);
+
+	return err;
+}
+
+static int resize_user(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+		       int entries, struct ib_udata *udata, int *npas,
+		       int *page_shift, int *cqe_size)
+{
+	struct mlx5_ib_resize_cq ucmd;
+	struct ib_umem *umem;
+	int err;
+	int npages;
+	struct ib_ucontext *context = cq->buf.umem->context;
+
+	err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
+	if (err)
+		return err;
+
+	if (ucmd.reserved0 || ucmd.reserved1)
+		return -EINVAL;
+
+	umem = ib_umem_get(context, ucmd.buf_addr, entries * ucmd.cqe_size,
+			   IB_ACCESS_LOCAL_WRITE, 1);
+	if (IS_ERR(umem)) {
+		err = PTR_ERR(umem);
+		return err;
+	}
+
+	mlx5_ib_cont_pages(umem, ucmd.buf_addr, &npages, page_shift,
+			   npas, NULL);
+
+	cq->resize_umem = umem;
+	*cqe_size = ucmd.cqe_size;
+
+	return 0;
+}
+
+static void un_resize_user(struct mlx5_ib_cq *cq)
+{
+	ib_umem_release(cq->resize_umem);
+}
+
+static int resize_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq,
+			 int entries, int cqe_size)
+{
+	int err;
+
+	cq->resize_buf = kzalloc(sizeof(*cq->resize_buf), GFP_KERNEL);
+	if (!cq->resize_buf)
+		return -ENOMEM;
+
+	err = alloc_cq_buf(dev, cq->resize_buf, entries, cqe_size);
+	if (err)
+		goto ex;
+
+	init_cq_buf(cq, cq->resize_buf);
+
+	return 0;
+
+ex:
+	kfree(cq->resize_buf);
+	return err;
+}
+
+static void un_resize_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_cq *cq)
+{
+	free_cq_buf(dev, cq->resize_buf);
+	cq->resize_buf = NULL;
+}
+
+static int copy_resize_cqes(struct mlx5_ib_cq *cq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(cq->ibcq.device);
+	struct mlx5_cqe64 *scqe64;
+	struct mlx5_cqe64 *dcqe64;
+	void *start_cqe;
+	void *scqe;
+	void *dcqe;
+	int ssize;
+	int dsize;
+	int i;
+	u8 sw_own;
+
+	ssize = cq->buf.cqe_size;
+	dsize = cq->resize_buf->cqe_size;
+	if (ssize != dsize) {
+		mlx5_ib_warn(dev, "resize from different cqe size is not supported\n");
+		return -EINVAL;
+	}
+
+	i = cq->mcq.cons_index;
+	scqe = get_sw_cqe(cq, i);
+	scqe64 = ssize == 64 ? scqe : scqe + 64;
+	start_cqe = scqe;
+	if (!scqe) {
+		mlx5_ib_warn(dev, "expected cqe in sw ownership\n");
+		return -EINVAL;
+	}
+
+	while ((scqe64->op_own >> 4) != MLX5_CQE_RESIZE_CQ) {
+		dcqe = get_cqe_from_buf(cq->resize_buf,
+					(i + 1) & (cq->resize_buf->nent),
+					dsize);
+		dcqe64 = dsize == 64 ? dcqe : dcqe + 64;
+		sw_own = sw_ownership_bit(i + 1, cq->resize_buf->nent);
+		memcpy(dcqe, scqe, dsize);
+		dcqe64->op_own = (dcqe64->op_own & ~MLX5_CQE_OWNER_MASK) | sw_own;
+
+		++i;
+		scqe = get_sw_cqe(cq, i);
+		scqe64 = ssize == 64 ? scqe : scqe + 64;
+		if (!scqe) {
+			mlx5_ib_warn(dev, "expected cqe in sw ownership\n");
+			return -EINVAL;
+		}
+
+		if (scqe == start_cqe) {
+			pr_warn("resize CQ failed to get resize CQE, CQN 0x%x\n",
+				cq->mcq.cqn);
+			return -ENOMEM;
+		}
+	}
+	++cq->mcq.cons_index;
+	return 0;
+}
+
+int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibcq->device);
+	struct mlx5_ib_cq *cq = to_mcq(ibcq);
+	struct mlx5_modify_cq_mbox_in *in;
+	int err;
+	int npas;
+	int page_shift;
+	int inlen;
+	int uninitialized_var(cqe_size);
+	unsigned long flags;
+
+	if (!(dev->mdev.caps.flags & MLX5_DEV_CAP_FLAG_RESIZE_CQ)) {
+		pr_info("Firmware does not support resize CQ\n");
+		return -ENOSYS;
+	}
+
+	if (entries < 1)
+		return -EINVAL;
+
+	entries = roundup_pow_of_two(entries + 1);
+	if (entries > dev->mdev.caps.max_cqes + 1)
+		return -EINVAL;
+
+	if (entries == ibcq->cqe + 1)
+		return 0;
+
+	mutex_lock(&cq->resize_mutex);
+	if (udata) {
+		err = resize_user(dev, cq, entries, udata, &npas, &page_shift,
+				  &cqe_size);
+	} else {
+		cqe_size = 64;
+		err = resize_kernel(dev, cq, entries, cqe_size);
+		if (!err) {
+			npas = cq->resize_buf->buf.npages;
+			page_shift = cq->resize_buf->buf.page_shift;
+		}
+	}
+
+	if (err)
+		goto ex;
+
+	inlen = sizeof(*in) + npas * sizeof(in->pas[0]);
+	in = mlx5_vzalloc(inlen);
+	if (!in) {
+		err = -ENOMEM;
+		goto ex_resize;
+	}
+
+	if (udata)
+		mlx5_ib_populate_pas(dev, cq->resize_umem, page_shift,
+				     in->pas, 0);
+	else
+		mlx5_fill_page_array(&cq->resize_buf->buf, in->pas);
+
+	in->field_select = cpu_to_be32(MLX5_MODIFY_CQ_MASK_LOG_SIZE  |
+				       MLX5_MODIFY_CQ_MASK_PG_OFFSET |
+				       MLX5_MODIFY_CQ_MASK_PG_SIZE);
+	in->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	in->ctx.cqe_sz_flags = cqe_sz_to_mlx_sz(cqe_size) << 5;
+	in->ctx.page_offset = 0;
+	in->ctx.log_sz_usr_page = cpu_to_be32(ilog2(entries) << 24);
+	in->hdr.opmod = cpu_to_be16(MLX5_CQ_OPMOD_RESIZE);
+	in->cqn = cpu_to_be32(cq->mcq.cqn);
+
+	err = mlx5_core_modify_cq(&dev->mdev, &cq->mcq, in, inlen);
+	if (err)
+		goto ex_alloc;
+
+	if (udata) {
+		cq->ibcq.cqe = entries - 1;
+		ib_umem_release(cq->buf.umem);
+		cq->buf.umem = cq->resize_umem;
+		cq->resize_umem = NULL;
+	} else {
+		struct mlx5_ib_cq_buf tbuf;
+		int resized = 0;
+
+		spin_lock_irqsave(&cq->lock, flags);
+		if (cq->resize_buf) {
+			err = copy_resize_cqes(cq);
+			if (!err) {
+				tbuf = cq->buf;
+				cq->buf = *cq->resize_buf;
+				kfree(cq->resize_buf);
+				cq->resize_buf = NULL;
+				resized = 1;
+			}
+		}
+		cq->ibcq.cqe = entries - 1;
+		spin_unlock_irqrestore(&cq->lock, flags);
+		if (resized)
+			free_cq_buf(dev, &tbuf);
+	}
+	mutex_unlock(&cq->resize_mutex);
+
+	mlx5_vfree(in);
+	return 0;
+
+ex_alloc:
+	mlx5_vfree(in);
+
+ex_resize:
+	if (udata)
+		un_resize_user(cq);
+	else
+		un_resize_kernel(dev, cq);
+ex:
+	mutex_unlock(&cq->resize_mutex);
+	return err;
+}
+
+int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq)
+{
+	struct mlx5_ib_cq *cq;
+
+	if (!ibcq)
+		return 128;
+
+	cq = to_mcq(ibcq);
+	return cq->cqe_size;
+}
diff --git a/drivers/infiniband/hw/mlx5/doorbell.c b/drivers/infiniband/hw/mlx5/doorbell.c
new file mode 100644
index 00000000000..ece028fc47d
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/doorbell.c
@@ -0,0 +1,98 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/kref.h>
+#include <linux/slab.h>
+#include <rdma/ib_umem.h>
+
+#include "mlx5_ib.h"
+
+struct mlx5_ib_user_db_page {
+	struct list_head	list;
+	struct ib_umem	       *umem;
+	unsigned long		user_virt;
+	int			refcnt;
+};
+
+int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
+			struct mlx5_db *db)
+{
+	struct mlx5_ib_user_db_page *page;
+	int err = 0;
+
+	mutex_lock(&context->db_page_mutex);
+
+	list_for_each_entry(page, &context->db_page_list, list)
+		if (page->user_virt == (virt & PAGE_MASK))
+			goto found;
+
+	page = kmalloc(sizeof(*page), GFP_KERNEL);
+	if (!page) {
+		err = -ENOMEM;
+		goto out;
+	}
+
+	page->user_virt = (virt & PAGE_MASK);
+	page->refcnt    = 0;
+	page->umem      = ib_umem_get(&context->ibucontext, virt & PAGE_MASK,
+				      PAGE_SIZE, 0, 0);
+	if (IS_ERR(page->umem)) {
+		err = PTR_ERR(page->umem);
+		kfree(page);
+		goto out;
+	}
+
+	list_add(&page->list, &context->db_page_list);
+
+found:
+	db->dma = sg_dma_address(page->umem->sg_head.sgl) + (virt & ~PAGE_MASK);
+	db->u.user_page = page;
+	++page->refcnt;
+
+out:
+	mutex_unlock(&context->db_page_mutex);
+
+	return err;
+}
+
+void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db)
+{
+	mutex_lock(&context->db_page_mutex);
+
+	if (!--db->u.user_page->refcnt) {
+		list_del(&db->u.user_page->list);
+		ib_umem_release(db->u.user_page->umem);
+		kfree(db->u.user_page);
+	}
+
+	mutex_unlock(&context->db_page_mutex);
+}
diff --git a/drivers/infiniband/hw/mlx5/mad.c b/drivers/infiniband/hw/mlx5/mad.c
new file mode 100644
index 00000000000..5c8938be0e0
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/mad.c
@@ -0,0 +1,139 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mlx5/cmd.h>
+#include <rdma/ib_mad.h>
+#include <rdma/ib_smi.h>
+#include "mlx5_ib.h"
+
+enum {
+	MLX5_IB_VENDOR_CLASS1 = 0x9,
+	MLX5_IB_VENDOR_CLASS2 = 0xa
+};
+
+int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+		 int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
+		 void *in_mad, void *response_mad)
+{
+	u8 op_modifier = 0;
+
+	/* Key check traps can't be generated unless we have in_wc to
+	 * tell us where to send the trap.
+	 */
+	if (ignore_mkey || !in_wc)
+		op_modifier |= 0x1;
+	if (ignore_bkey || !in_wc)
+		op_modifier |= 0x2;
+
+	return mlx5_core_mad_ifc(&dev->mdev, in_mad, response_mad, op_modifier, port);
+}
+
+int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	u16 slid;
+	int err;
+
+	slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP && slid == 0)
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	    in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		if (in_mad->mad_hdr.method   != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_SET &&
+		    in_mad->mad_hdr.method   != IB_MGMT_METHOD_TRAP_REPRESS)
+			return IB_MAD_RESULT_SUCCESS;
+
+		/* Don't process SMInfo queries -- the SMA can't handle them.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_SM_INFO)
+			return IB_MAD_RESULT_SUCCESS;
+	} else if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_PERF_MGMT ||
+		   in_mad->mad_hdr.mgmt_class == MLX5_IB_VENDOR_CLASS1   ||
+		   in_mad->mad_hdr.mgmt_class == MLX5_IB_VENDOR_CLASS2   ||
+		   in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_CONG_MGMT) {
+		if (in_mad->mad_hdr.method  != IB_MGMT_METHOD_GET &&
+		    in_mad->mad_hdr.method  != IB_MGMT_METHOD_SET)
+			return IB_MAD_RESULT_SUCCESS;
+	} else {
+		return IB_MAD_RESULT_SUCCESS;
+	}
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev),
+			   mad_flags & IB_MAD_IGNORE_MKEY,
+			   mad_flags & IB_MAD_IGNORE_BKEY,
+			   port_num, in_wc, in_grh, in_mad, out_mad);
+	if (err)
+		return IB_MAD_RESULT_FAILURE;
+
+	/* set return bit in status of directed route responses */
+	if (in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		out_mad->mad_hdr.status |= cpu_to_be16(1 << 15);
+
+	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP_REPRESS)
+		/* no response for trap repress */
+		return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	u16 packet_error;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+
+	packet_error = be16_to_cpu(out_mad->status);
+
+	dev->mdev.caps.ext_port_cap[port - 1] = (!err && !packet_error) ?
+		MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO : 0;
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
diff --git a/drivers/infiniband/hw/mlx5/main.c b/drivers/infiniband/hw/mlx5/main.c
new file mode 100644
index 00000000000..364d4b6937f
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/main.c
@@ -0,0 +1,1543 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <asm-generic/kmap_types.h>
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/slab.h>
+#include <linux/io-mapping.h>
+#include <linux/sched.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_smi.h>
+#include <rdma/ib_umem.h>
+#include "user.h"
+#include "mlx5_ib.h"
+
+#define DRIVER_NAME "mlx5_ib"
+#define DRIVER_VERSION "2.2-1"
+#define DRIVER_RELDATE	"Feb 2014"
+
+MODULE_AUTHOR("Eli Cohen <eli@mellanox.com>");
+MODULE_DESCRIPTION("Mellanox Connect-IB HCA IB driver");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRIVER_VERSION);
+
+static int prof_sel = 2;
+module_param_named(prof_sel, prof_sel, int, 0444);
+MODULE_PARM_DESC(prof_sel, "profile selector. Valid range 0 - 2");
+
+static char mlx5_version[] =
+	DRIVER_NAME ": Mellanox Connect-IB Infiniband driver v"
+	DRIVER_VERSION " (" DRIVER_RELDATE ")\n";
+
+static struct mlx5_profile profile[] = {
+	[0] = {
+		.mask		= 0,
+	},
+	[1] = {
+		.mask		= MLX5_PROF_MASK_QP_SIZE,
+		.log_max_qp	= 12,
+	},
+	[2] = {
+		.mask		= MLX5_PROF_MASK_QP_SIZE |
+				  MLX5_PROF_MASK_MR_CACHE,
+		.log_max_qp	= 17,
+		.mr_cache[0]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[1]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[2]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[3]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[4]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[5]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[6]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[7]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[8]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[9]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[10]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[11]	= {
+			.size	= 500,
+			.limit	= 250
+		},
+		.mr_cache[12]	= {
+			.size	= 64,
+			.limit	= 32
+		},
+		.mr_cache[13]	= {
+			.size	= 32,
+			.limit	= 16
+		},
+		.mr_cache[14]	= {
+			.size	= 16,
+			.limit	= 8
+		},
+		.mr_cache[15]	= {
+			.size	= 8,
+			.limit	= 4
+		},
+	},
+};
+
+int mlx5_vector2eqn(struct mlx5_ib_dev *dev, int vector, int *eqn, int *irqn)
+{
+	struct mlx5_eq_table *table = &dev->mdev.priv.eq_table;
+	struct mlx5_eq *eq, *n;
+	int err = -ENOENT;
+
+	spin_lock(&table->lock);
+	list_for_each_entry_safe(eq, n, &dev->eqs_list, list) {
+		if (eq->index == vector) {
+			*eqn = eq->eqn;
+			*irqn = eq->irqn;
+			err = 0;
+			break;
+		}
+	}
+	spin_unlock(&table->lock);
+
+	return err;
+}
+
+static int alloc_comp_eqs(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_eq_table *table = &dev->mdev.priv.eq_table;
+	char name[MLX5_MAX_EQ_NAME];
+	struct mlx5_eq *eq, *n;
+	int ncomp_vec;
+	int nent;
+	int err;
+	int i;
+
+	INIT_LIST_HEAD(&dev->eqs_list);
+	ncomp_vec = table->num_comp_vectors;
+	nent = MLX5_COMP_EQ_SIZE;
+	for (i = 0; i < ncomp_vec; i++) {
+		eq = kzalloc(sizeof(*eq), GFP_KERNEL);
+		if (!eq) {
+			err = -ENOMEM;
+			goto clean;
+		}
+
+		snprintf(name, MLX5_MAX_EQ_NAME, "mlx5_comp%d", i);
+		err = mlx5_create_map_eq(&dev->mdev, eq,
+					 i + MLX5_EQ_VEC_COMP_BASE, nent, 0,
+					 name, &dev->mdev.priv.uuari.uars[0]);
+		if (err) {
+			kfree(eq);
+			goto clean;
+		}
+		mlx5_ib_dbg(dev, "allocated completion EQN %d\n", eq->eqn);
+		eq->index = i;
+		spin_lock(&table->lock);
+		list_add_tail(&eq->list, &dev->eqs_list);
+		spin_unlock(&table->lock);
+	}
+
+	dev->num_comp_vectors = ncomp_vec;
+	return 0;
+
+clean:
+	spin_lock(&table->lock);
+	list_for_each_entry_safe(eq, n, &dev->eqs_list, list) {
+		list_del(&eq->list);
+		spin_unlock(&table->lock);
+		if (mlx5_destroy_unmap_eq(&dev->mdev, eq))
+			mlx5_ib_warn(dev, "failed to destroy EQ 0x%x\n", eq->eqn);
+		kfree(eq);
+		spin_lock(&table->lock);
+	}
+	spin_unlock(&table->lock);
+	return err;
+}
+
+static void free_comp_eqs(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_eq_table *table = &dev->mdev.priv.eq_table;
+	struct mlx5_eq *eq, *n;
+
+	spin_lock(&table->lock);
+	list_for_each_entry_safe(eq, n, &dev->eqs_list, list) {
+		list_del(&eq->list);
+		spin_unlock(&table->lock);
+		if (mlx5_destroy_unmap_eq(&dev->mdev, eq))
+			mlx5_ib_warn(dev, "failed to destroy EQ 0x%x\n", eq->eqn);
+		kfree(eq);
+		spin_lock(&table->lock);
+	}
+	spin_unlock(&table->lock);
+}
+
+static int mlx5_ib_query_device(struct ib_device *ibdev,
+				struct ib_device_attr *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+	int max_rq_sg;
+	int max_sq_sg;
+	u64 flags;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memset(props, 0, sizeof(*props));
+
+	props->fw_ver = ((u64)fw_rev_maj(&dev->mdev) << 32) |
+		(fw_rev_min(&dev->mdev) << 16) |
+		fw_rev_sub(&dev->mdev);
+	props->device_cap_flags    = IB_DEVICE_CHANGE_PHY_PORT |
+		IB_DEVICE_PORT_ACTIVE_EVENT		|
+		IB_DEVICE_SYS_IMAGE_GUID		|
+		IB_DEVICE_RC_RNR_NAK_GEN;
+	flags = dev->mdev.caps.flags;
+	if (flags & MLX5_DEV_CAP_FLAG_BAD_PKEY_CNTR)
+		props->device_cap_flags |= IB_DEVICE_BAD_PKEY_CNTR;
+	if (flags & MLX5_DEV_CAP_FLAG_BAD_QKEY_CNTR)
+		props->device_cap_flags |= IB_DEVICE_BAD_QKEY_CNTR;
+	if (flags & MLX5_DEV_CAP_FLAG_APM)
+		props->device_cap_flags |= IB_DEVICE_AUTO_PATH_MIG;
+	props->device_cap_flags |= IB_DEVICE_LOCAL_DMA_LKEY;
+	if (flags & MLX5_DEV_CAP_FLAG_XRC)
+		props->device_cap_flags |= IB_DEVICE_XRC;
+	props->device_cap_flags |= IB_DEVICE_MEM_MGT_EXTENSIONS;
+	if (flags & MLX5_DEV_CAP_FLAG_SIG_HAND_OVER) {
+		props->device_cap_flags |= IB_DEVICE_SIGNATURE_HANDOVER;
+		/* At this stage no support for signature handover */
+		props->sig_prot_cap = IB_PROT_T10DIF_TYPE_1 |
+				      IB_PROT_T10DIF_TYPE_2 |
+				      IB_PROT_T10DIF_TYPE_3;
+		props->sig_guard_cap = IB_GUARD_T10DIF_CRC |
+				       IB_GUARD_T10DIF_CSUM;
+	}
+	if (flags & MLX5_DEV_CAP_FLAG_BLOCK_MCAST)
+		props->device_cap_flags |= IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
+
+	props->vendor_id	   = be32_to_cpup((__be32 *)(out_mad->data + 36)) &
+		0xffffff;
+	props->vendor_part_id	   = be16_to_cpup((__be16 *)(out_mad->data + 30));
+	props->hw_ver		   = be32_to_cpup((__be32 *)(out_mad->data + 32));
+	memcpy(&props->sys_image_guid, out_mad->data +	4, 8);
+
+	props->max_mr_size	   = ~0ull;
+	props->page_size_cap	   = dev->mdev.caps.min_page_sz;
+	props->max_qp		   = 1 << dev->mdev.caps.log_max_qp;
+	props->max_qp_wr	   = dev->mdev.caps.max_wqes;
+	max_rq_sg = dev->mdev.caps.max_rq_desc_sz / sizeof(struct mlx5_wqe_data_seg);
+	max_sq_sg = (dev->mdev.caps.max_sq_desc_sz - sizeof(struct mlx5_wqe_ctrl_seg)) /
+		sizeof(struct mlx5_wqe_data_seg);
+	props->max_sge = min(max_rq_sg, max_sq_sg);
+	props->max_cq		   = 1 << dev->mdev.caps.log_max_cq;
+	props->max_cqe		   = dev->mdev.caps.max_cqes - 1;
+	props->max_mr		   = 1 << dev->mdev.caps.log_max_mkey;
+	props->max_pd		   = 1 << dev->mdev.caps.log_max_pd;
+	props->max_qp_rd_atom	   = dev->mdev.caps.max_ra_req_qp;
+	props->max_qp_init_rd_atom = dev->mdev.caps.max_ra_res_qp;
+	props->max_res_rd_atom	   = props->max_qp_rd_atom * props->max_qp;
+	props->max_srq		   = 1 << dev->mdev.caps.log_max_srq;
+	props->max_srq_wr	   = dev->mdev.caps.max_srq_wqes - 1;
+	props->max_srq_sge	   = max_rq_sg - 1;
+	props->max_fast_reg_page_list_len = (unsigned int)-1;
+	props->local_ca_ack_delay  = dev->mdev.caps.local_ca_ack_delay;
+	props->atomic_cap	   = IB_ATOMIC_NONE;
+	props->masked_atomic_cap   = IB_ATOMIC_NONE;
+	props->max_pkeys	   = be16_to_cpup((__be16 *)(out_mad->data + 28));
+	props->max_mcast_grp	   = 1 << dev->mdev.caps.log_max_mcg;
+	props->max_mcast_qp_attach = dev->mdev.caps.max_qp_mcg;
+	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
+					   props->max_mcast_grp;
+	props->max_map_per_fmr = INT_MAX; /* no limit in ConnectIB */
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+
+	return err;
+}
+
+int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
+		       struct ib_port_attr *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int ext_active_speed;
+	int err = -ENOMEM;
+
+	if (port < 1 || port > dev->mdev.caps.num_ports) {
+		mlx5_ib_warn(dev, "invalid port number %d\n", port);
+		return -EINVAL;
+	}
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	memset(props, 0, sizeof(*props));
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(dev, 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err) {
+		mlx5_ib_warn(dev, "err %d\n", err);
+		goto out;
+	}
+
+
+	props->lid		= be16_to_cpup((__be16 *)(out_mad->data + 16));
+	props->lmc		= out_mad->data[34] & 0x7;
+	props->sm_lid		= be16_to_cpup((__be16 *)(out_mad->data + 18));
+	props->sm_sl		= out_mad->data[36] & 0xf;
+	props->state		= out_mad->data[32] & 0xf;
+	props->phys_state	= out_mad->data[33] >> 4;
+	props->port_cap_flags	= be32_to_cpup((__be32 *)(out_mad->data + 20));
+	props->gid_tbl_len	= out_mad->data[50];
+	props->max_msg_sz	= 1 << to_mdev(ibdev)->mdev.caps.log_max_msg;
+	props->pkey_tbl_len	= to_mdev(ibdev)->mdev.caps.port[port - 1].pkey_table_len;
+	props->bad_pkey_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 46));
+	props->qkey_viol_cntr	= be16_to_cpup((__be16 *)(out_mad->data + 48));
+	props->active_width	= out_mad->data[31] & 0xf;
+	props->active_speed	= out_mad->data[35] >> 4;
+	props->max_mtu		= out_mad->data[41] & 0xf;
+	props->active_mtu	= out_mad->data[36] >> 4;
+	props->subnet_timeout	= out_mad->data[51] & 0x1f;
+	props->max_vl_num	= out_mad->data[37] >> 4;
+	props->init_type_reply	= out_mad->data[41] >> 4;
+
+	/* Check if extended speeds (EDR/FDR/...) are supported */
+	if (props->port_cap_flags & IB_PORT_EXTENDED_SPEEDS_SUP) {
+		ext_active_speed = out_mad->data[62] >> 4;
+
+		switch (ext_active_speed) {
+		case 1:
+			props->active_speed = 16; /* FDR */
+			break;
+		case 2:
+			props->active_speed = 32; /* EDR */
+			break;
+		}
+	}
+
+	/* If reported active speed is QDR, check if is FDR-10 */
+	if (props->active_speed == 4) {
+		if (dev->mdev.caps.ext_port_cap[port - 1] &
+		    MLX_EXT_PORT_CAP_FLAG_EXTENDED_PORT_INFO) {
+			init_query_mad(in_mad);
+			in_mad->attr_id = MLX5_ATTR_EXTENDED_PORT_INFO;
+			in_mad->attr_mod = cpu_to_be32(port);
+
+			err = mlx5_MAD_IFC(dev, 1, 1, port,
+					   NULL, NULL, in_mad, out_mad);
+			if (err)
+				goto out;
+
+			/* Checking LinkSpeedActive for FDR-10 */
+			if (out_mad->data[15] & 0x1)
+				props->active_speed = 8;
+		}
+	}
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+
+	return err;
+}
+
+static int mlx5_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+			     union ib_gid *gid)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PORT_INFO;
+	in_mad->attr_mod = cpu_to_be32(port);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(gid->raw, out_mad->data + 8, 8);
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_GUID_INFO;
+	in_mad->attr_mod = cpu_to_be32(index / 8);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static int mlx5_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			      u16 *pkey)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id  = IB_SMP_ATTR_PKEY_TABLE;
+	in_mad->attr_mod = cpu_to_be32(index / 32);
+
+	err = mlx5_MAD_IFC(to_mdev(ibdev), 1, 1, port, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	*pkey = be16_to_cpu(((__be16 *)out_mad->data)[index % 32]);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+struct mlx5_reg_node_desc {
+	u8	desc[64];
+};
+
+static int mlx5_ib_modify_device(struct ib_device *ibdev, int mask,
+				 struct ib_device_modify *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_reg_node_desc in;
+	struct mlx5_reg_node_desc out;
+	int err;
+
+	if (mask & ~IB_DEVICE_MODIFY_NODE_DESC)
+		return -EOPNOTSUPP;
+
+	if (!(mask & IB_DEVICE_MODIFY_NODE_DESC))
+		return 0;
+
+	/*
+	 * If possible, pass node desc to FW, so it can generate
+	 * a 144 trap.  If cmd fails, just ignore.
+	 */
+	memcpy(&in, props->node_desc, 64);
+	err = mlx5_core_access_reg(&dev->mdev, &in, sizeof(in), &out,
+				   sizeof(out), MLX5_REG_NODE_DESC, 0, 1);
+	if (err)
+		return err;
+
+	memcpy(ibdev->node_desc, props->node_desc, 64);
+
+	return err;
+}
+
+static int mlx5_ib_modify_port(struct ib_device *ibdev, u8 port, int mask,
+			       struct ib_port_modify *props)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct ib_port_attr attr;
+	u32 tmp;
+	int err;
+
+	mutex_lock(&dev->cap_mask_mutex);
+
+	err = mlx5_ib_query_port(ibdev, port, &attr);
+	if (err)
+		goto out;
+
+	tmp = (attr.port_cap_flags | props->set_port_cap_mask) &
+		~props->clr_port_cap_mask;
+
+	err = mlx5_set_port_caps(&dev->mdev, port, tmp);
+
+out:
+	mutex_unlock(&dev->cap_mask_mutex);
+	return err;
+}
+
+static struct ib_ucontext *mlx5_ib_alloc_ucontext(struct ib_device *ibdev,
+						  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_alloc_ucontext_req_v2 req;
+	struct mlx5_ib_alloc_ucontext_resp resp;
+	struct mlx5_ib_ucontext *context;
+	struct mlx5_uuar_info *uuari;
+	struct mlx5_uar *uars;
+	int gross_uuars;
+	int num_uars;
+	int ver;
+	int uuarn;
+	int err;
+	int i;
+	int reqlen;
+
+	if (!dev->ib_active)
+		return ERR_PTR(-EAGAIN);
+
+	memset(&req, 0, sizeof(req));
+	reqlen = udata->inlen - sizeof(struct ib_uverbs_cmd_hdr);
+	if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req))
+		ver = 0;
+	else if (reqlen == sizeof(struct mlx5_ib_alloc_ucontext_req_v2))
+		ver = 2;
+	else
+		return ERR_PTR(-EINVAL);
+
+	err = ib_copy_from_udata(&req, udata, reqlen);
+	if (err)
+		return ERR_PTR(err);
+
+	if (req.flags || req.reserved)
+		return ERR_PTR(-EINVAL);
+
+	if (req.total_num_uuars > MLX5_MAX_UUARS)
+		return ERR_PTR(-ENOMEM);
+
+	if (req.total_num_uuars == 0)
+		return ERR_PTR(-EINVAL);
+
+	req.total_num_uuars = ALIGN(req.total_num_uuars,
+				    MLX5_NON_FP_BF_REGS_PER_PAGE);
+	if (req.num_low_latency_uuars > req.total_num_uuars - 1)
+		return ERR_PTR(-EINVAL);
+
+	num_uars = req.total_num_uuars / MLX5_NON_FP_BF_REGS_PER_PAGE;
+	gross_uuars = num_uars * MLX5_BF_REGS_PER_PAGE;
+	resp.qp_tab_size      = 1 << dev->mdev.caps.log_max_qp;
+	resp.bf_reg_size      = dev->mdev.caps.bf_reg_size;
+	resp.cache_line_size  = L1_CACHE_BYTES;
+	resp.max_sq_desc_sz = dev->mdev.caps.max_sq_desc_sz;
+	resp.max_rq_desc_sz = dev->mdev.caps.max_rq_desc_sz;
+	resp.max_send_wqebb = dev->mdev.caps.max_wqes;
+	resp.max_recv_wr = dev->mdev.caps.max_wqes;
+	resp.max_srq_recv_wr = dev->mdev.caps.max_srq_wqes;
+
+	context = kzalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return ERR_PTR(-ENOMEM);
+
+	uuari = &context->uuari;
+	mutex_init(&uuari->lock);
+	uars = kcalloc(num_uars, sizeof(*uars), GFP_KERNEL);
+	if (!uars) {
+		err = -ENOMEM;
+		goto out_ctx;
+	}
+
+	uuari->bitmap = kcalloc(BITS_TO_LONGS(gross_uuars),
+				sizeof(*uuari->bitmap),
+				GFP_KERNEL);
+	if (!uuari->bitmap) {
+		err = -ENOMEM;
+		goto out_uar_ctx;
+	}
+	/*
+	 * clear all fast path uuars
+	 */
+	for (i = 0; i < gross_uuars; i++) {
+		uuarn = i & 3;
+		if (uuarn == 2 || uuarn == 3)
+			set_bit(i, uuari->bitmap);
+	}
+
+	uuari->count = kcalloc(gross_uuars, sizeof(*uuari->count), GFP_KERNEL);
+	if (!uuari->count) {
+		err = -ENOMEM;
+		goto out_bitmap;
+	}
+
+	for (i = 0; i < num_uars; i++) {
+		err = mlx5_cmd_alloc_uar(&dev->mdev, &uars[i].index);
+		if (err)
+			goto out_count;
+	}
+
+	INIT_LIST_HEAD(&context->db_page_list);
+	mutex_init(&context->db_page_mutex);
+
+	resp.tot_uuars = req.total_num_uuars;
+	resp.num_ports = dev->mdev.caps.num_ports;
+	err = ib_copy_to_udata(udata, &resp,
+			       sizeof(resp) - sizeof(resp.reserved));
+	if (err)
+		goto out_uars;
+
+	uuari->ver = ver;
+	uuari->num_low_latency_uuars = req.num_low_latency_uuars;
+	uuari->uars = uars;
+	uuari->num_uars = num_uars;
+	return &context->ibucontext;
+
+out_uars:
+	for (i--; i >= 0; i--)
+		mlx5_cmd_free_uar(&dev->mdev, uars[i].index);
+out_count:
+	kfree(uuari->count);
+
+out_bitmap:
+	kfree(uuari->bitmap);
+
+out_uar_ctx:
+	kfree(uars);
+
+out_ctx:
+	kfree(context);
+	return ERR_PTR(err);
+}
+
+static int mlx5_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
+{
+	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
+	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
+	struct mlx5_uuar_info *uuari = &context->uuari;
+	int i;
+
+	for (i = 0; i < uuari->num_uars; i++) {
+		if (mlx5_cmd_free_uar(&dev->mdev, uuari->uars[i].index))
+			mlx5_ib_warn(dev, "failed to free UAR 0x%x\n", uuari->uars[i].index);
+	}
+
+	kfree(uuari->count);
+	kfree(uuari->bitmap);
+	kfree(uuari->uars);
+	kfree(context);
+
+	return 0;
+}
+
+static phys_addr_t uar_index2pfn(struct mlx5_ib_dev *dev, int index)
+{
+	return (pci_resource_start(dev->mdev.pdev, 0) >> PAGE_SHIFT) + index;
+}
+
+static int get_command(unsigned long offset)
+{
+	return (offset >> MLX5_IB_MMAP_CMD_SHIFT) & MLX5_IB_MMAP_CMD_MASK;
+}
+
+static int get_arg(unsigned long offset)
+{
+	return offset & ((1 << MLX5_IB_MMAP_CMD_SHIFT) - 1);
+}
+
+static int get_index(unsigned long offset)
+{
+	return get_arg(offset);
+}
+
+static int mlx5_ib_mmap(struct ib_ucontext *ibcontext, struct vm_area_struct *vma)
+{
+	struct mlx5_ib_ucontext *context = to_mucontext(ibcontext);
+	struct mlx5_ib_dev *dev = to_mdev(ibcontext->device);
+	struct mlx5_uuar_info *uuari = &context->uuari;
+	unsigned long command;
+	unsigned long idx;
+	phys_addr_t pfn;
+
+	command = get_command(vma->vm_pgoff);
+	switch (command) {
+	case MLX5_IB_MMAP_REGULAR_PAGE:
+		if (vma->vm_end - vma->vm_start != PAGE_SIZE)
+			return -EINVAL;
+
+		idx = get_index(vma->vm_pgoff);
+		pfn = uar_index2pfn(dev, uuari->uars[idx].index);
+		mlx5_ib_dbg(dev, "uar idx 0x%lx, pfn 0x%llx\n", idx,
+			    (unsigned long long)pfn);
+
+		if (idx >= uuari->num_uars)
+			return -EINVAL;
+
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+		if (io_remap_pfn_range(vma, vma->vm_start, pfn,
+				       PAGE_SIZE, vma->vm_page_prot))
+			return -EAGAIN;
+
+		mlx5_ib_dbg(dev, "mapped WC at 0x%lx, PA 0x%llx\n",
+			    vma->vm_start,
+			    (unsigned long long)pfn << PAGE_SHIFT);
+		break;
+
+	case MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES:
+		return -ENOSYS;
+
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int alloc_pa_mkey(struct mlx5_ib_dev *dev, u32 *key, u32 pdn)
+{
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_mkey_seg *seg;
+	struct mlx5_core_mr mr;
+	int err;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	seg = &in->seg;
+	seg->flags = MLX5_PERM_LOCAL_READ | MLX5_ACCESS_MODE_PA;
+	seg->flags_pd = cpu_to_be32(pdn | MLX5_MKEY_LEN64);
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	seg->start_addr = 0;
+
+	err = mlx5_core_create_mkey(&dev->mdev, &mr, in, sizeof(*in),
+				    NULL, NULL, NULL);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to create mkey, %d\n", err);
+		goto err_in;
+	}
+
+	kfree(in);
+	*key = mr.key;
+
+	return 0;
+
+err_in:
+	kfree(in);
+
+	return err;
+}
+
+static void free_pa_mkey(struct mlx5_ib_dev *dev, u32 key)
+{
+	struct mlx5_core_mr mr;
+	int err;
+
+	memset(&mr, 0, sizeof(mr));
+	mr.key = key;
+	err = mlx5_core_destroy_mkey(&dev->mdev, &mr);
+	if (err)
+		mlx5_ib_warn(dev, "failed to destroy mkey 0x%x\n", key);
+}
+
+static struct ib_pd *mlx5_ib_alloc_pd(struct ib_device *ibdev,
+				      struct ib_ucontext *context,
+				      struct ib_udata *udata)
+{
+	struct mlx5_ib_alloc_pd_resp resp;
+	struct mlx5_ib_pd *pd;
+	int err;
+
+	pd = kmalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx5_core_alloc_pd(&to_mdev(ibdev)->mdev, &pd->pdn);
+	if (err) {
+		kfree(pd);
+		return ERR_PTR(err);
+	}
+
+	if (context) {
+		resp.pdn = pd->pdn;
+		if (ib_copy_to_udata(udata, &resp, sizeof(resp))) {
+			mlx5_core_dealloc_pd(&to_mdev(ibdev)->mdev, pd->pdn);
+			kfree(pd);
+			return ERR_PTR(-EFAULT);
+		}
+	} else {
+		err = alloc_pa_mkey(to_mdev(ibdev), &pd->pa_lkey, pd->pdn);
+		if (err) {
+			mlx5_core_dealloc_pd(&to_mdev(ibdev)->mdev, pd->pdn);
+			kfree(pd);
+			return ERR_PTR(err);
+		}
+	}
+
+	return &pd->ibpd;
+}
+
+static int mlx5_ib_dealloc_pd(struct ib_pd *pd)
+{
+	struct mlx5_ib_dev *mdev = to_mdev(pd->device);
+	struct mlx5_ib_pd *mpd = to_mpd(pd);
+
+	if (!pd->uobject)
+		free_pa_mkey(mdev, mpd->pa_lkey);
+
+	mlx5_core_dealloc_pd(&mdev->mdev, mpd->pdn);
+	kfree(mpd);
+
+	return 0;
+}
+
+static int mlx5_ib_mcg_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	int err;
+
+	err = mlx5_core_attach_mcg(&dev->mdev, gid, ibqp->qp_num);
+	if (err)
+		mlx5_ib_warn(dev, "failed attaching QPN 0x%x, MGID %pI6\n",
+			     ibqp->qp_num, gid->raw);
+
+	return err;
+}
+
+static int mlx5_ib_mcg_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	int err;
+
+	err = mlx5_core_detach_mcg(&dev->mdev, gid, ibqp->qp_num);
+	if (err)
+		mlx5_ib_warn(dev, "failed detaching QPN 0x%x, MGID %pI6\n",
+			     ibqp->qp_num, gid->raw);
+
+	return err;
+}
+
+static int init_node_data(struct mlx5_ib_dev *dev)
+{
+	struct ib_smp *in_mad  = NULL;
+	struct ib_smp *out_mad = NULL;
+	int err = -ENOMEM;
+
+	in_mad  = kzalloc(sizeof(*in_mad), GFP_KERNEL);
+	out_mad = kmalloc(sizeof(*out_mad), GFP_KERNEL);
+	if (!in_mad || !out_mad)
+		goto out;
+
+	init_query_mad(in_mad);
+	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
+
+	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
+
+	err = mlx5_MAD_IFC(dev, 1, 1, 1, NULL, NULL, in_mad, out_mad);
+	if (err)
+		goto out;
+
+	dev->mdev.rev_id = be32_to_cpup((__be32 *)(out_mad->data + 32));
+	memcpy(&dev->ib_dev.node_guid, out_mad->data + 12, 8);
+
+out:
+	kfree(in_mad);
+	kfree(out_mad);
+	return err;
+}
+
+static ssize_t show_fw_pages(struct device *device, struct device_attribute *attr,
+			     char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+
+	return sprintf(buf, "%d\n", dev->mdev.priv.fw_pages);
+}
+
+static ssize_t show_reg_pages(struct device *device,
+			      struct device_attribute *attr, char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+
+	return sprintf(buf, "%d\n", dev->mdev.priv.reg_pages);
+}
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "MT%d\n", dev->mdev.pdev->device);
+}
+
+static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
+			   char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%d.%d.%d\n", fw_rev_maj(&dev->mdev),
+		       fw_rev_min(&dev->mdev), fw_rev_sub(&dev->mdev));
+}
+
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%x\n", dev->mdev.rev_id);
+}
+
+static ssize_t show_board(struct device *device, struct device_attribute *attr,
+			  char *buf)
+{
+	struct mlx5_ib_dev *dev =
+		container_of(device, struct mlx5_ib_dev, ib_dev.dev);
+	return sprintf(buf, "%.*s\n", MLX5_BOARD_ID_LEN,
+		       dev->mdev.board_id);
+}
+
+static DEVICE_ATTR(hw_rev,   S_IRUGO, show_rev,    NULL);
+static DEVICE_ATTR(fw_ver,   S_IRUGO, show_fw_ver, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca,    NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_board,  NULL);
+static DEVICE_ATTR(fw_pages, S_IRUGO, show_fw_pages, NULL);
+static DEVICE_ATTR(reg_pages, S_IRUGO, show_reg_pages, NULL);
+
+static struct device_attribute *mlx5_class_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_fw_ver,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+	&dev_attr_fw_pages,
+	&dev_attr_reg_pages,
+};
+
+static void mlx5_ib_event(struct mlx5_core_dev *dev, enum mlx5_dev_event event,
+			  void *data)
+{
+	struct mlx5_ib_dev *ibdev = container_of(dev, struct mlx5_ib_dev, mdev);
+	struct ib_event ibev;
+	u8 port = 0;
+
+	switch (event) {
+	case MLX5_DEV_EVENT_SYS_ERROR:
+		ibdev->ib_active = false;
+		ibev.event = IB_EVENT_DEVICE_FATAL;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_UP:
+		ibev.event = IB_EVENT_PORT_ACTIVE;
+		port = *(u8 *)data;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_DOWN:
+		ibev.event = IB_EVENT_PORT_ERR;
+		port = *(u8 *)data;
+		break;
+
+	case MLX5_DEV_EVENT_PORT_INITIALIZED:
+		/* not used by ULPs */
+		return;
+
+	case MLX5_DEV_EVENT_LID_CHANGE:
+		ibev.event = IB_EVENT_LID_CHANGE;
+		port = *(u8 *)data;
+		break;
+
+	case MLX5_DEV_EVENT_PKEY_CHANGE:
+		ibev.event = IB_EVENT_PKEY_CHANGE;
+		port = *(u8 *)data;
+		break;
+
+	case MLX5_DEV_EVENT_GUID_CHANGE:
+		ibev.event = IB_EVENT_GID_CHANGE;
+		port = *(u8 *)data;
+		break;
+
+	case MLX5_DEV_EVENT_CLIENT_REREG:
+		ibev.event = IB_EVENT_CLIENT_REREGISTER;
+		port = *(u8 *)data;
+		break;
+	}
+
+	ibev.device	      = &ibdev->ib_dev;
+	ibev.element.port_num = port;
+
+	if (port < 1 || port > ibdev->num_ports) {
+		mlx5_ib_warn(ibdev, "warning: event on port %d\n", port);
+		return;
+	}
+
+	if (ibdev->ib_active)
+		ib_dispatch_event(&ibev);
+}
+
+static void get_ext_port_caps(struct mlx5_ib_dev *dev)
+{
+	int port;
+
+	for (port = 1; port <= dev->mdev.caps.num_ports; port++)
+		mlx5_query_ext_port_caps(dev, port);
+}
+
+static int get_port_caps(struct mlx5_ib_dev *dev)
+{
+	struct ib_device_attr *dprops = NULL;
+	struct ib_port_attr *pprops = NULL;
+	int err = 0;
+	int port;
+
+	pprops = kmalloc(sizeof(*pprops), GFP_KERNEL);
+	if (!pprops)
+		goto out;
+
+	dprops = kmalloc(sizeof(*dprops), GFP_KERNEL);
+	if (!dprops)
+		goto out;
+
+	err = mlx5_ib_query_device(&dev->ib_dev, dprops);
+	if (err) {
+		mlx5_ib_warn(dev, "query_device failed %d\n", err);
+		goto out;
+	}
+
+	for (port = 1; port <= dev->mdev.caps.num_ports; port++) {
+		err = mlx5_ib_query_port(&dev->ib_dev, port, pprops);
+		if (err) {
+			mlx5_ib_warn(dev, "query_port %d failed %d\n", port, err);
+			break;
+		}
+		dev->mdev.caps.port[port - 1].pkey_table_len = dprops->max_pkeys;
+		dev->mdev.caps.port[port - 1].gid_table_len = pprops->gid_tbl_len;
+		mlx5_ib_dbg(dev, "pkey_table_len %d, gid_table_len %d\n",
+			    dprops->max_pkeys, pprops->gid_tbl_len);
+	}
+
+out:
+	kfree(pprops);
+	kfree(dprops);
+
+	return err;
+}
+
+static void destroy_umrc_res(struct mlx5_ib_dev *dev)
+{
+	int err;
+
+	err = mlx5_mr_cache_cleanup(dev);
+	if (err)
+		mlx5_ib_warn(dev, "mr cache cleanup failed\n");
+
+	mlx5_ib_destroy_qp(dev->umrc.qp);
+	ib_destroy_cq(dev->umrc.cq);
+	ib_dereg_mr(dev->umrc.mr);
+	ib_dealloc_pd(dev->umrc.pd);
+}
+
+enum {
+	MAX_UMR_WR = 128,
+};
+
+static int create_umr_res(struct mlx5_ib_dev *dev)
+{
+	struct ib_qp_init_attr *init_attr = NULL;
+	struct ib_qp_attr *attr = NULL;
+	struct ib_pd *pd;
+	struct ib_cq *cq;
+	struct ib_qp *qp;
+	struct ib_mr *mr;
+	int ret;
+
+	attr = kzalloc(sizeof(*attr), GFP_KERNEL);
+	init_attr = kzalloc(sizeof(*init_attr), GFP_KERNEL);
+	if (!attr || !init_attr) {
+		ret = -ENOMEM;
+		goto error_0;
+	}
+
+	pd = ib_alloc_pd(&dev->ib_dev);
+	if (IS_ERR(pd)) {
+		mlx5_ib_dbg(dev, "Couldn't create PD for sync UMR QP\n");
+		ret = PTR_ERR(pd);
+		goto error_0;
+	}
+
+	mr = ib_get_dma_mr(pd,  IB_ACCESS_LOCAL_WRITE);
+	if (IS_ERR(mr)) {
+		mlx5_ib_dbg(dev, "Couldn't create DMA MR for sync UMR QP\n");
+		ret = PTR_ERR(mr);
+		goto error_1;
+	}
+
+	cq = ib_create_cq(&dev->ib_dev, mlx5_umr_cq_handler, NULL, NULL, 128,
+			  0);
+	if (IS_ERR(cq)) {
+		mlx5_ib_dbg(dev, "Couldn't create CQ for sync UMR QP\n");
+		ret = PTR_ERR(cq);
+		goto error_2;
+	}
+	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+
+	init_attr->send_cq = cq;
+	init_attr->recv_cq = cq;
+	init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
+	init_attr->cap.max_send_wr = MAX_UMR_WR;
+	init_attr->cap.max_send_sge = 1;
+	init_attr->qp_type = MLX5_IB_QPT_REG_UMR;
+	init_attr->port_num = 1;
+	qp = mlx5_ib_create_qp(pd, init_attr, NULL);
+	if (IS_ERR(qp)) {
+		mlx5_ib_dbg(dev, "Couldn't create sync UMR QP\n");
+		ret = PTR_ERR(qp);
+		goto error_3;
+	}
+	qp->device     = &dev->ib_dev;
+	qp->real_qp    = qp;
+	qp->uobject    = NULL;
+	qp->qp_type    = MLX5_IB_QPT_REG_UMR;
+
+	attr->qp_state = IB_QPS_INIT;
+	attr->port_num = 1;
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE | IB_QP_PKEY_INDEX |
+				IB_QP_PORT, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify UMR QP\n");
+		goto error_4;
+	}
+
+	memset(attr, 0, sizeof(*attr));
+	attr->qp_state = IB_QPS_RTR;
+	attr->path_mtu = IB_MTU_256;
+
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rtr\n");
+		goto error_4;
+	}
+
+	memset(attr, 0, sizeof(*attr));
+	attr->qp_state = IB_QPS_RTS;
+	ret = mlx5_ib_modify_qp(qp, attr, IB_QP_STATE, NULL);
+	if (ret) {
+		mlx5_ib_dbg(dev, "Couldn't modify umr QP to rts\n");
+		goto error_4;
+	}
+
+	dev->umrc.qp = qp;
+	dev->umrc.cq = cq;
+	dev->umrc.mr = mr;
+	dev->umrc.pd = pd;
+
+	sema_init(&dev->umrc.sem, MAX_UMR_WR);
+	ret = mlx5_mr_cache_init(dev);
+	if (ret) {
+		mlx5_ib_warn(dev, "mr cache init failed %d\n", ret);
+		goto error_4;
+	}
+
+	kfree(attr);
+	kfree(init_attr);
+
+	return 0;
+
+error_4:
+	mlx5_ib_destroy_qp(qp);
+
+error_3:
+	ib_destroy_cq(cq);
+
+error_2:
+	ib_dereg_mr(mr);
+
+error_1:
+	ib_dealloc_pd(pd);
+
+error_0:
+	kfree(attr);
+	kfree(init_attr);
+	return ret;
+}
+
+static int create_dev_resources(struct mlx5_ib_resources *devr)
+{
+	struct ib_srq_init_attr attr;
+	struct mlx5_ib_dev *dev;
+	int ret = 0;
+
+	dev = container_of(devr, struct mlx5_ib_dev, devr);
+
+	devr->p0 = mlx5_ib_alloc_pd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->p0)) {
+		ret = PTR_ERR(devr->p0);
+		goto error0;
+	}
+	devr->p0->device  = &dev->ib_dev;
+	devr->p0->uobject = NULL;
+	atomic_set(&devr->p0->usecnt, 0);
+
+	devr->c0 = mlx5_ib_create_cq(&dev->ib_dev, 1, 0, NULL, NULL);
+	if (IS_ERR(devr->c0)) {
+		ret = PTR_ERR(devr->c0);
+		goto error1;
+	}
+	devr->c0->device        = &dev->ib_dev;
+	devr->c0->uobject       = NULL;
+	devr->c0->comp_handler  = NULL;
+	devr->c0->event_handler = NULL;
+	devr->c0->cq_context    = NULL;
+	atomic_set(&devr->c0->usecnt, 0);
+
+	devr->x0 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->x0)) {
+		ret = PTR_ERR(devr->x0);
+		goto error2;
+	}
+	devr->x0->device = &dev->ib_dev;
+	devr->x0->inode = NULL;
+	atomic_set(&devr->x0->usecnt, 0);
+	mutex_init(&devr->x0->tgt_qp_mutex);
+	INIT_LIST_HEAD(&devr->x0->tgt_qp_list);
+
+	devr->x1 = mlx5_ib_alloc_xrcd(&dev->ib_dev, NULL, NULL);
+	if (IS_ERR(devr->x1)) {
+		ret = PTR_ERR(devr->x1);
+		goto error3;
+	}
+	devr->x1->device = &dev->ib_dev;
+	devr->x1->inode = NULL;
+	atomic_set(&devr->x1->usecnt, 0);
+	mutex_init(&devr->x1->tgt_qp_mutex);
+	INIT_LIST_HEAD(&devr->x1->tgt_qp_list);
+
+	memset(&attr, 0, sizeof(attr));
+	attr.attr.max_sge = 1;
+	attr.attr.max_wr = 1;
+	attr.srq_type = IB_SRQT_XRC;
+	attr.ext.xrc.cq = devr->c0;
+	attr.ext.xrc.xrcd = devr->x0;
+
+	devr->s0 = mlx5_ib_create_srq(devr->p0, &attr, NULL);
+	if (IS_ERR(devr->s0)) {
+		ret = PTR_ERR(devr->s0);
+		goto error4;
+	}
+	devr->s0->device	= &dev->ib_dev;
+	devr->s0->pd		= devr->p0;
+	devr->s0->uobject       = NULL;
+	devr->s0->event_handler = NULL;
+	devr->s0->srq_context   = NULL;
+	devr->s0->srq_type      = IB_SRQT_XRC;
+	devr->s0->ext.xrc.xrcd	= devr->x0;
+	devr->s0->ext.xrc.cq	= devr->c0;
+	atomic_inc(&devr->s0->ext.xrc.xrcd->usecnt);
+	atomic_inc(&devr->s0->ext.xrc.cq->usecnt);
+	atomic_inc(&devr->p0->usecnt);
+	atomic_set(&devr->s0->usecnt, 0);
+
+	return 0;
+
+error4:
+	mlx5_ib_dealloc_xrcd(devr->x1);
+error3:
+	mlx5_ib_dealloc_xrcd(devr->x0);
+error2:
+	mlx5_ib_destroy_cq(devr->c0);
+error1:
+	mlx5_ib_dealloc_pd(devr->p0);
+error0:
+	return ret;
+}
+
+static void destroy_dev_resources(struct mlx5_ib_resources *devr)
+{
+	mlx5_ib_destroy_srq(devr->s0);
+	mlx5_ib_dealloc_xrcd(devr->x0);
+	mlx5_ib_dealloc_xrcd(devr->x1);
+	mlx5_ib_destroy_cq(devr->c0);
+	mlx5_ib_dealloc_pd(devr->p0);
+}
+
+static int init_one(struct pci_dev *pdev,
+		    const struct pci_device_id *id)
+{
+	struct mlx5_core_dev *mdev;
+	struct mlx5_ib_dev *dev;
+	int err;
+	int i;
+
+	printk_once(KERN_INFO "%s", mlx5_version);
+
+	dev = (struct mlx5_ib_dev *)ib_alloc_device(sizeof(*dev));
+	if (!dev)
+		return -ENOMEM;
+
+	mdev = &dev->mdev;
+	mdev->event = mlx5_ib_event;
+	if (prof_sel >= ARRAY_SIZE(profile)) {
+		pr_warn("selected pofile out of range, selceting default\n");
+		prof_sel = 0;
+	}
+	mdev->profile = &profile[prof_sel];
+	err = mlx5_dev_init(mdev, pdev);
+	if (err)
+		goto err_free;
+
+	err = get_port_caps(dev);
+	if (err)
+		goto err_cleanup;
+
+	get_ext_port_caps(dev);
+
+	err = alloc_comp_eqs(dev);
+	if (err)
+		goto err_cleanup;
+
+	MLX5_INIT_DOORBELL_LOCK(&dev->uar_lock);
+
+	strlcpy(dev->ib_dev.name, "mlx5_%d", IB_DEVICE_NAME_MAX);
+	dev->ib_dev.owner		= THIS_MODULE;
+	dev->ib_dev.node_type		= RDMA_NODE_IB_CA;
+	dev->ib_dev.local_dma_lkey	= mdev->caps.reserved_lkey;
+	dev->num_ports		= mdev->caps.num_ports;
+	dev->ib_dev.phys_port_cnt     = dev->num_ports;
+	dev->ib_dev.num_comp_vectors	= dev->num_comp_vectors;
+	dev->ib_dev.dma_device	= &mdev->pdev->dev;
+
+	dev->ib_dev.uverbs_abi_ver	= MLX5_IB_UVERBS_ABI_VERSION;
+	dev->ib_dev.uverbs_cmd_mask	=
+		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)	|
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)		|
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)		|
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)		|
+		(1ull << IB_USER_VERBS_CMD_REG_MR)		|
+		(1ull << IB_USER_VERBS_CMD_DEREG_MR)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL)	|
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP)		|
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)		|
+		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)	|
+		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)	|
+		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)		|
+		(1ull << IB_USER_VERBS_CMD_CREATE_XSRQ)		|
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
+
+	dev->ib_dev.query_device	= mlx5_ib_query_device;
+	dev->ib_dev.query_port		= mlx5_ib_query_port;
+	dev->ib_dev.query_gid		= mlx5_ib_query_gid;
+	dev->ib_dev.query_pkey		= mlx5_ib_query_pkey;
+	dev->ib_dev.modify_device	= mlx5_ib_modify_device;
+	dev->ib_dev.modify_port		= mlx5_ib_modify_port;
+	dev->ib_dev.alloc_ucontext	= mlx5_ib_alloc_ucontext;
+	dev->ib_dev.dealloc_ucontext	= mlx5_ib_dealloc_ucontext;
+	dev->ib_dev.mmap		= mlx5_ib_mmap;
+	dev->ib_dev.alloc_pd		= mlx5_ib_alloc_pd;
+	dev->ib_dev.dealloc_pd		= mlx5_ib_dealloc_pd;
+	dev->ib_dev.create_ah		= mlx5_ib_create_ah;
+	dev->ib_dev.query_ah		= mlx5_ib_query_ah;
+	dev->ib_dev.destroy_ah		= mlx5_ib_destroy_ah;
+	dev->ib_dev.create_srq		= mlx5_ib_create_srq;
+	dev->ib_dev.modify_srq		= mlx5_ib_modify_srq;
+	dev->ib_dev.query_srq		= mlx5_ib_query_srq;
+	dev->ib_dev.destroy_srq		= mlx5_ib_destroy_srq;
+	dev->ib_dev.post_srq_recv	= mlx5_ib_post_srq_recv;
+	dev->ib_dev.create_qp		= mlx5_ib_create_qp;
+	dev->ib_dev.modify_qp		= mlx5_ib_modify_qp;
+	dev->ib_dev.query_qp		= mlx5_ib_query_qp;
+	dev->ib_dev.destroy_qp		= mlx5_ib_destroy_qp;
+	dev->ib_dev.post_send		= mlx5_ib_post_send;
+	dev->ib_dev.post_recv		= mlx5_ib_post_recv;
+	dev->ib_dev.create_cq		= mlx5_ib_create_cq;
+	dev->ib_dev.modify_cq		= mlx5_ib_modify_cq;
+	dev->ib_dev.resize_cq		= mlx5_ib_resize_cq;
+	dev->ib_dev.destroy_cq		= mlx5_ib_destroy_cq;
+	dev->ib_dev.poll_cq		= mlx5_ib_poll_cq;
+	dev->ib_dev.req_notify_cq	= mlx5_ib_arm_cq;
+	dev->ib_dev.get_dma_mr		= mlx5_ib_get_dma_mr;
+	dev->ib_dev.reg_user_mr		= mlx5_ib_reg_user_mr;
+	dev->ib_dev.dereg_mr		= mlx5_ib_dereg_mr;
+	dev->ib_dev.destroy_mr		= mlx5_ib_destroy_mr;
+	dev->ib_dev.attach_mcast	= mlx5_ib_mcg_attach;
+	dev->ib_dev.detach_mcast	= mlx5_ib_mcg_detach;
+	dev->ib_dev.process_mad		= mlx5_ib_process_mad;
+	dev->ib_dev.create_mr		= mlx5_ib_create_mr;
+	dev->ib_dev.alloc_fast_reg_mr	= mlx5_ib_alloc_fast_reg_mr;
+	dev->ib_dev.alloc_fast_reg_page_list = mlx5_ib_alloc_fast_reg_page_list;
+	dev->ib_dev.free_fast_reg_page_list  = mlx5_ib_free_fast_reg_page_list;
+	dev->ib_dev.check_mr_status	= mlx5_ib_check_mr_status;
+
+	if (mdev->caps.flags & MLX5_DEV_CAP_FLAG_XRC) {
+		dev->ib_dev.alloc_xrcd = mlx5_ib_alloc_xrcd;
+		dev->ib_dev.dealloc_xrcd = mlx5_ib_dealloc_xrcd;
+		dev->ib_dev.uverbs_cmd_mask |=
+			(1ull << IB_USER_VERBS_CMD_OPEN_XRCD) |
+			(1ull << IB_USER_VERBS_CMD_CLOSE_XRCD);
+	}
+
+	err = init_node_data(dev);
+	if (err)
+		goto err_eqs;
+
+	mutex_init(&dev->cap_mask_mutex);
+	spin_lock_init(&dev->mr_lock);
+
+	err = create_dev_resources(&dev->devr);
+	if (err)
+		goto err_eqs;
+
+	err = ib_register_device(&dev->ib_dev, NULL);
+	if (err)
+		goto err_rsrc;
+
+	err = create_umr_res(dev);
+	if (err)
+		goto err_dev;
+
+	for (i = 0; i < ARRAY_SIZE(mlx5_class_attributes); i++) {
+		err = device_create_file(&dev->ib_dev.dev,
+					 mlx5_class_attributes[i]);
+		if (err)
+			goto err_umrc;
+	}
+
+	dev->ib_active = true;
+
+	return 0;
+
+err_umrc:
+	destroy_umrc_res(dev);
+
+err_dev:
+	ib_unregister_device(&dev->ib_dev);
+
+err_rsrc:
+	destroy_dev_resources(&dev->devr);
+
+err_eqs:
+	free_comp_eqs(dev);
+
+err_cleanup:
+	mlx5_dev_cleanup(mdev);
+
+err_free:
+	ib_dealloc_device((struct ib_device *)dev);
+
+	return err;
+}
+
+static void remove_one(struct pci_dev *pdev)
+{
+	struct mlx5_ib_dev *dev = mlx5_pci2ibdev(pdev);
+
+	destroy_umrc_res(dev);
+	ib_unregister_device(&dev->ib_dev);
+	destroy_dev_resources(&dev->devr);
+	free_comp_eqs(dev);
+	mlx5_dev_cleanup(&dev->mdev);
+	ib_dealloc_device(&dev->ib_dev);
+}
+
+static DEFINE_PCI_DEVICE_TABLE(mlx5_ib_pci_table) = {
+	{ PCI_VDEVICE(MELLANOX, 4113) }, /* MT4113 Connect-IB */
+	{ 0, }
+};
+
+MODULE_DEVICE_TABLE(pci, mlx5_ib_pci_table);
+
+static struct pci_driver mlx5_ib_driver = {
+	.name		= DRIVER_NAME,
+	.id_table	= mlx5_ib_pci_table,
+	.probe		= init_one,
+	.remove		= remove_one
+};
+
+static int __init mlx5_ib_init(void)
+{
+	return pci_register_driver(&mlx5_ib_driver);
+}
+
+static void __exit mlx5_ib_cleanup(void)
+{
+	pci_unregister_driver(&mlx5_ib_driver);
+}
+
+module_init(mlx5_ib_init);
+module_exit(mlx5_ib_cleanup);
diff --git a/drivers/infiniband/hw/mlx5/mem.c b/drivers/infiniband/hw/mlx5/mem.c
new file mode 100644
index 00000000000..8499aec94db
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/mem.c
@@ -0,0 +1,162 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <rdma/ib_umem.h>
+#include "mlx5_ib.h"
+
+/* @umem: umem object to scan
+ * @addr: ib virtual address requested by the user
+ * @count: number of PAGE_SIZE pages covered by umem
+ * @shift: page shift for the compound pages found in the region
+ * @ncont: number of compund pages
+ * @order: log2 of the number of compound pages
+ */
+void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
+			int *ncont, int *order)
+{
+	unsigned long tmp;
+	unsigned long m;
+	int i, k;
+	u64 base = 0;
+	int p = 0;
+	int skip;
+	int mask;
+	u64 len;
+	u64 pfn;
+	struct scatterlist *sg;
+	int entry;
+
+	addr = addr >> PAGE_SHIFT;
+	tmp = (unsigned long)addr;
+	m = find_first_bit(&tmp, sizeof(tmp));
+	skip = 1 << m;
+	mask = skip - 1;
+	i = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		pfn = sg_dma_address(sg) >> PAGE_SHIFT;
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				tmp = (unsigned long)pfn;
+				m = min(m, find_first_bit(&tmp, sizeof(tmp)));
+				skip = 1 << m;
+				mask = skip - 1;
+				base = pfn;
+				p = 0;
+			} else {
+				if (base + p != pfn) {
+					tmp = (unsigned long)p;
+					m = find_first_bit(&tmp, sizeof(tmp));
+					skip = 1 << m;
+					mask = skip - 1;
+					base = pfn;
+					p = 0;
+				}
+			}
+			p++;
+			i++;
+		}
+	}
+
+	if (i) {
+		m = min_t(unsigned long, ilog2(roundup_pow_of_two(i)), m);
+
+		if (order)
+			*order = ilog2(roundup_pow_of_two(i) >> m);
+
+		*ncont = DIV_ROUND_UP(i, (1 << m));
+	} else {
+		m  = 0;
+
+		if (order)
+			*order = 0;
+
+		*ncont = 0;
+	}
+	*shift = PAGE_SHIFT + m;
+	*count = i;
+}
+
+void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			  int page_shift, __be64 *pas, int umr)
+{
+	int shift = page_shift - PAGE_SHIFT;
+	int mask = (1 << shift) - 1;
+	int i, k;
+	u64 cur = 0;
+	u64 base;
+	int len;
+	struct scatterlist *sg;
+	int entry;
+
+	i = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		len = sg_dma_len(sg) >> PAGE_SHIFT;
+		base = sg_dma_address(sg);
+		for (k = 0; k < len; k++) {
+			if (!(i & mask)) {
+				cur = base + (k << PAGE_SHIFT);
+				if (umr)
+					cur |= 3;
+
+				pas[i >> shift] = cpu_to_be64(cur);
+				mlx5_ib_dbg(dev, "pas[%d] 0x%llx\n",
+					    i >> shift, be64_to_cpu(pas[i >> shift]));
+			}  else
+				mlx5_ib_dbg(dev, "=====> 0x%llx\n",
+					    base + (k << PAGE_SHIFT));
+			i++;
+		}
+	}
+}
+
+int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset)
+{
+	u64 page_size;
+	u64 page_mask;
+	u64 off_size;
+	u64 off_mask;
+	u64 buf_off;
+
+	page_size = 1 << page_shift;
+	page_mask = page_size - 1;
+	buf_off = addr & page_mask;
+	off_size = page_size >> 6;
+	off_mask = off_size - 1;
+
+	if (buf_off & off_mask)
+		return -EINVAL;
+
+	*offset = buf_off >> ilog2(off_size);
+	return 0;
+}
diff --git a/drivers/infiniband/hw/mlx5/mlx5_ib.h b/drivers/infiniband/hw/mlx5/mlx5_ib.h
new file mode 100644
index 00000000000..f2ccf1a5a29
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/mlx5_ib.h
@@ -0,0 +1,574 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef MLX5_IB_H
+#define MLX5_IB_H
+
+#include <linux/kernel.h>
+#include <linux/sched.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_smi.h>
+#include <linux/mlx5/driver.h>
+#include <linux/mlx5/cq.h>
+#include <linux/mlx5/qp.h>
+#include <linux/mlx5/srq.h>
+#include <linux/types.h>
+
+#define mlx5_ib_dbg(dev, format, arg...)				\
+pr_debug("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	 __LINE__, current->pid, ##arg)
+
+#define mlx5_ib_err(dev, format, arg...)				\
+pr_err("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	__LINE__, current->pid, ##arg)
+
+#define mlx5_ib_warn(dev, format, arg...)				\
+pr_warn("%s:%s:%d:(pid %d): " format, (dev)->ib_dev.name, __func__,	\
+	__LINE__, current->pid, ##arg)
+
+enum {
+	MLX5_IB_MMAP_CMD_SHIFT	= 8,
+	MLX5_IB_MMAP_CMD_MASK	= 0xff,
+};
+
+enum mlx5_ib_mmap_cmd {
+	MLX5_IB_MMAP_REGULAR_PAGE		= 0,
+	MLX5_IB_MMAP_GET_CONTIGUOUS_PAGES	= 1, /* always last */
+};
+
+enum {
+	MLX5_RES_SCAT_DATA32_CQE	= 0x1,
+	MLX5_RES_SCAT_DATA64_CQE	= 0x2,
+	MLX5_REQ_SCAT_DATA32_CQE	= 0x11,
+	MLX5_REQ_SCAT_DATA64_CQE	= 0x22,
+};
+
+enum mlx5_ib_latency_class {
+	MLX5_IB_LATENCY_CLASS_LOW,
+	MLX5_IB_LATENCY_CLASS_MEDIUM,
+	MLX5_IB_LATENCY_CLASS_HIGH,
+	MLX5_IB_LATENCY_CLASS_FAST_PATH
+};
+
+enum mlx5_ib_mad_ifc_flags {
+	MLX5_MAD_IFC_IGNORE_MKEY	= 1,
+	MLX5_MAD_IFC_IGNORE_BKEY	= 2,
+	MLX5_MAD_IFC_NET_VIEW		= 4,
+};
+
+struct mlx5_ib_ucontext {
+	struct ib_ucontext	ibucontext;
+	struct list_head	db_page_list;
+
+	/* protect doorbell record alloc/free
+	 */
+	struct mutex		db_page_mutex;
+	struct mlx5_uuar_info	uuari;
+};
+
+static inline struct mlx5_ib_ucontext *to_mucontext(struct ib_ucontext *ibucontext)
+{
+	return container_of(ibucontext, struct mlx5_ib_ucontext, ibucontext);
+}
+
+struct mlx5_ib_pd {
+	struct ib_pd		ibpd;
+	u32			pdn;
+	u32			pa_lkey;
+};
+
+/* Use macros here so that don't have to duplicate
+ * enum ib_send_flags and enum ib_qp_type for low-level driver
+ */
+
+#define MLX5_IB_SEND_UMR_UNREG	IB_SEND_RESERVED_START
+#define MLX5_IB_QPT_REG_UMR	IB_QPT_RESERVED1
+#define MLX5_IB_WR_UMR		IB_WR_RESERVED1
+
+struct wr_list {
+	u16	opcode;
+	u16	next;
+};
+
+struct mlx5_ib_wq {
+	u64		       *wrid;
+	u32		       *wr_data;
+	struct wr_list	       *w_list;
+	unsigned	       *wqe_head;
+	u16		        unsig_count;
+
+	/* serialize post to the work queue
+	 */
+	spinlock_t		lock;
+	int			wqe_cnt;
+	int			max_post;
+	int			max_gs;
+	int			offset;
+	int			wqe_shift;
+	unsigned		head;
+	unsigned		tail;
+	u16			cur_post;
+	u16			last_poll;
+	void		       *qend;
+};
+
+enum {
+	MLX5_QP_USER,
+	MLX5_QP_KERNEL,
+	MLX5_QP_EMPTY
+};
+
+struct mlx5_ib_qp {
+	struct ib_qp		ibqp;
+	struct mlx5_core_qp	mqp;
+	struct mlx5_buf		buf;
+
+	struct mlx5_db		db;
+	struct mlx5_ib_wq	rq;
+
+	u32			doorbell_qpn;
+	u8			sq_signal_bits;
+	u8			fm_cache;
+	int			sq_max_wqes_per_wr;
+	int			sq_spare_wqes;
+	struct mlx5_ib_wq	sq;
+
+	struct ib_umem	       *umem;
+	int			buf_size;
+
+	/* serialize qp state modifications
+	 */
+	struct mutex		mutex;
+	u16			xrcdn;
+	u32			flags;
+	u8			port;
+	u8			alt_port;
+	u8			atomic_rd_en;
+	u8			resp_depth;
+	u8			state;
+	int			mlx_type;
+	int			wq_sig;
+	int			scat_cqe;
+	int			max_inline_data;
+	struct mlx5_bf	       *bf;
+	int			has_rq;
+
+	/* only for user space QPs. For kernel
+	 * we have it from the bf object
+	 */
+	int			uuarn;
+
+	int			create_type;
+	u32			pa_lkey;
+
+	/* Store signature errors */
+	bool			signature_en;
+};
+
+struct mlx5_ib_cq_buf {
+	struct mlx5_buf		buf;
+	struct ib_umem		*umem;
+	int			cqe_size;
+	int			nent;
+};
+
+enum mlx5_ib_qp_flags {
+	MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK     = 1 << 0,
+	MLX5_IB_QP_SIGNATURE_HANDLING           = 1 << 1,
+};
+
+struct mlx5_shared_mr_info {
+	int mr_id;
+	struct ib_umem		*umem;
+};
+
+struct mlx5_ib_cq {
+	struct ib_cq		ibcq;
+	struct mlx5_core_cq	mcq;
+	struct mlx5_ib_cq_buf	buf;
+	struct mlx5_db		db;
+
+	/* serialize access to the CQ
+	 */
+	spinlock_t		lock;
+
+	/* protect resize cq
+	 */
+	struct mutex		resize_mutex;
+	struct mlx5_ib_cq_buf  *resize_buf;
+	struct ib_umem	       *resize_umem;
+	int			cqe_size;
+};
+
+struct mlx5_ib_srq {
+	struct ib_srq		ibsrq;
+	struct mlx5_core_srq	msrq;
+	struct mlx5_buf		buf;
+	struct mlx5_db		db;
+	u64		       *wrid;
+	/* protect SRQ hanlding
+	 */
+	spinlock_t		lock;
+	int			head;
+	int			tail;
+	u16			wqe_ctr;
+	struct ib_umem	       *umem;
+	/* serialize arming a SRQ
+	 */
+	struct mutex		mutex;
+	int			wq_sig;
+};
+
+struct mlx5_ib_xrcd {
+	struct ib_xrcd		ibxrcd;
+	u32			xrcdn;
+};
+
+struct mlx5_ib_mr {
+	struct ib_mr		ibmr;
+	struct mlx5_core_mr	mmr;
+	struct ib_umem	       *umem;
+	struct mlx5_shared_mr_info	*smr_info;
+	struct list_head	list;
+	int			order;
+	int			umred;
+	__be64			*pas;
+	dma_addr_t		dma;
+	int			npages;
+	struct mlx5_ib_dev     *dev;
+	struct mlx5_create_mkey_mbox_out out;
+	struct mlx5_core_sig_ctx    *sig;
+};
+
+struct mlx5_ib_fast_reg_page_list {
+	struct ib_fast_reg_page_list	ibfrpl;
+	__be64			       *mapped_page_list;
+	dma_addr_t			map;
+};
+
+struct mlx5_ib_umr_context {
+	enum ib_wc_status	status;
+	struct completion	done;
+};
+
+static inline void mlx5_ib_init_umr_context(struct mlx5_ib_umr_context *context)
+{
+	context->status = -1;
+	init_completion(&context->done);
+}
+
+struct umr_common {
+	struct ib_pd	*pd;
+	struct ib_cq	*cq;
+	struct ib_qp	*qp;
+	struct ib_mr	*mr;
+	/* control access to UMR QP
+	 */
+	struct semaphore	sem;
+};
+
+enum {
+	MLX5_FMR_INVALID,
+	MLX5_FMR_VALID,
+	MLX5_FMR_BUSY,
+};
+
+struct mlx5_ib_fmr {
+	struct ib_fmr			ibfmr;
+	struct mlx5_core_mr		mr;
+	int				access_flags;
+	int				state;
+	/* protect fmr state
+	 */
+	spinlock_t			lock;
+	u64				wrid;
+	struct ib_send_wr		wr[2];
+	u8				page_shift;
+	struct ib_fast_reg_page_list	page_list;
+};
+
+struct mlx5_cache_ent {
+	struct list_head	head;
+	/* sync access to the cahce entry
+	 */
+	spinlock_t		lock;
+
+
+	struct dentry	       *dir;
+	char                    name[4];
+	u32                     order;
+	u32			size;
+	u32                     cur;
+	u32                     miss;
+	u32			limit;
+
+	struct dentry          *fsize;
+	struct dentry          *fcur;
+	struct dentry          *fmiss;
+	struct dentry          *flimit;
+
+	struct mlx5_ib_dev     *dev;
+	struct work_struct	work;
+	struct delayed_work	dwork;
+	int			pending;
+};
+
+struct mlx5_mr_cache {
+	struct workqueue_struct *wq;
+	struct mlx5_cache_ent	ent[MAX_MR_CACHE_ENTRIES];
+	int			stopped;
+	struct dentry		*root;
+	unsigned long		last_add;
+};
+
+struct mlx5_ib_resources {
+	struct ib_cq	*c0;
+	struct ib_xrcd	*x0;
+	struct ib_xrcd	*x1;
+	struct ib_pd	*p0;
+	struct ib_srq	*s0;
+};
+
+struct mlx5_ib_dev {
+	struct ib_device		ib_dev;
+	struct mlx5_core_dev		mdev;
+	MLX5_DECLARE_DOORBELL_LOCK(uar_lock);
+	struct list_head		eqs_list;
+	int				num_ports;
+	int				num_comp_vectors;
+	/* serialize update of capability mask
+	 */
+	struct mutex			cap_mask_mutex;
+	bool				ib_active;
+	struct umr_common		umrc;
+	/* sync used page count stats
+	 */
+	spinlock_t			mr_lock;
+	struct mlx5_ib_resources	devr;
+	struct mlx5_mr_cache		cache;
+	struct timer_list		delay_timer;
+	int				fill_delay;
+};
+
+static inline struct mlx5_ib_cq *to_mibcq(struct mlx5_core_cq *mcq)
+{
+	return container_of(mcq, struct mlx5_ib_cq, mcq);
+}
+
+static inline struct mlx5_ib_xrcd *to_mxrcd(struct ib_xrcd *ibxrcd)
+{
+	return container_of(ibxrcd, struct mlx5_ib_xrcd, ibxrcd);
+}
+
+static inline struct mlx5_ib_dev *to_mdev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct mlx5_ib_dev, ib_dev);
+}
+
+static inline struct mlx5_ib_fmr *to_mfmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct mlx5_ib_fmr, ibfmr);
+}
+
+static inline struct mlx5_ib_cq *to_mcq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct mlx5_ib_cq, ibcq);
+}
+
+static inline struct mlx5_ib_qp *to_mibqp(struct mlx5_core_qp *mqp)
+{
+	return container_of(mqp, struct mlx5_ib_qp, mqp);
+}
+
+static inline struct mlx5_ib_mr *to_mibmr(struct mlx5_core_mr *mmr)
+{
+	return container_of(mmr, struct mlx5_ib_mr, mmr);
+}
+
+static inline struct mlx5_ib_pd *to_mpd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct mlx5_ib_pd, ibpd);
+}
+
+static inline struct mlx5_ib_srq *to_msrq(struct ib_srq *ibsrq)
+{
+	return container_of(ibsrq, struct mlx5_ib_srq, ibsrq);
+}
+
+static inline struct mlx5_ib_qp *to_mqp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct mlx5_ib_qp, ibqp);
+}
+
+static inline struct mlx5_ib_srq *to_mibsrq(struct mlx5_core_srq *msrq)
+{
+	return container_of(msrq, struct mlx5_ib_srq, msrq);
+}
+
+static inline struct mlx5_ib_mr *to_mmr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct mlx5_ib_mr, ibmr);
+}
+
+static inline struct mlx5_ib_fast_reg_page_list *to_mfrpl(struct ib_fast_reg_page_list *ibfrpl)
+{
+	return container_of(ibfrpl, struct mlx5_ib_fast_reg_page_list, ibfrpl);
+}
+
+struct mlx5_ib_ah {
+	struct ib_ah		ibah;
+	struct mlx5_av		av;
+};
+
+static inline struct mlx5_ib_ah *to_mah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct mlx5_ib_ah, ibah);
+}
+
+static inline struct mlx5_ib_dev *mlx5_core2ibdev(struct mlx5_core_dev *dev)
+{
+	return container_of(dev, struct mlx5_ib_dev, mdev);
+}
+
+static inline struct mlx5_ib_dev *mlx5_pci2ibdev(struct pci_dev *pdev)
+{
+	return mlx5_core2ibdev(pci2mlx5_core_dev(pdev));
+}
+
+int mlx5_ib_db_map_user(struct mlx5_ib_ucontext *context, unsigned long virt,
+			struct mlx5_db *db);
+void mlx5_ib_db_unmap_user(struct mlx5_ib_ucontext *context, struct mlx5_db *db);
+void __mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
+void mlx5_ib_cq_clean(struct mlx5_ib_cq *cq, u32 qpn, struct mlx5_ib_srq *srq);
+void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index);
+int mlx5_MAD_IFC(struct mlx5_ib_dev *dev, int ignore_mkey, int ignore_bkey,
+		 int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
+		 void *in_mad, void *response_mad);
+struct ib_ah *create_ib_ah(struct ib_ah_attr *ah_attr,
+			   struct mlx5_ib_ah *ah);
+struct ib_ah *mlx5_ib_create_ah(struct ib_pd *pd, struct ib_ah_attr *ah_attr);
+int mlx5_ib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr);
+int mlx5_ib_destroy_ah(struct ib_ah *ah);
+struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
+				  struct ib_srq_init_attr *init_attr,
+				  struct ib_udata *udata);
+int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata);
+int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr);
+int mlx5_ib_destroy_srq(struct ib_srq *srq);
+int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			  struct ib_recv_wr **bad_wr);
+struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
+				struct ib_qp_init_attr *init_attr,
+				struct ib_udata *udata);
+int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask, struct ib_udata *udata);
+int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr);
+int mlx5_ib_destroy_qp(struct ib_qp *qp);
+int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr);
+int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr);
+void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n);
+struct ib_cq *mlx5_ib_create_cq(struct ib_device *ibdev, int entries,
+				int vector, struct ib_ucontext *context,
+				struct ib_udata *udata);
+int mlx5_ib_destroy_cq(struct ib_cq *cq);
+int mlx5_ib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc);
+int mlx5_ib_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags flags);
+int mlx5_ib_modify_cq(struct ib_cq *cq, u16 cq_count, u16 cq_period);
+int mlx5_ib_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *udata);
+struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc);
+struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				  u64 virt_addr, int access_flags,
+				  struct ib_udata *udata);
+int mlx5_ib_dereg_mr(struct ib_mr *ibmr);
+int mlx5_ib_destroy_mr(struct ib_mr *ibmr);
+struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd,
+				struct ib_mr_init_attr *mr_init_attr);
+struct ib_mr *mlx5_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len);
+struct ib_fast_reg_page_list *mlx5_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len);
+void mlx5_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list);
+struct ib_fmr *mlx5_ib_fmr_alloc(struct ib_pd *pd, int acc,
+				 struct ib_fmr_attr *fmr_attr);
+int mlx5_ib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		      int npages, u64 iova);
+int mlx5_ib_unmap_fmr(struct list_head *fmr_list);
+int mlx5_ib_fmr_dealloc(struct ib_fmr *ibfmr);
+int mlx5_ib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+			struct ib_wc *in_wc, struct ib_grh *in_grh,
+			struct ib_mad *in_mad, struct ib_mad *out_mad);
+struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata);
+int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd);
+int mlx5_vector2eqn(struct mlx5_ib_dev *dev, int vector, int *eqn, int *irqn);
+int mlx5_ib_get_buf_offset(u64 addr, int page_shift, u32 *offset);
+int mlx5_query_ext_port_caps(struct mlx5_ib_dev *dev, u8 port);
+int mlx5_ib_query_port(struct ib_device *ibdev, u8 port,
+		       struct ib_port_attr *props);
+int mlx5_ib_init_fmr(struct mlx5_ib_dev *dev);
+void mlx5_ib_cleanup_fmr(struct mlx5_ib_dev *dev);
+void mlx5_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift,
+			int *ncont, int *order);
+void mlx5_ib_populate_pas(struct mlx5_ib_dev *dev, struct ib_umem *umem,
+			  int page_shift, __be64 *pas, int umr);
+void mlx5_ib_copy_pas(u64 *old, u64 *new, int step, int num);
+int mlx5_ib_get_cqe_size(struct mlx5_ib_dev *dev, struct ib_cq *ibcq);
+int mlx5_mr_cache_init(struct mlx5_ib_dev *dev);
+int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev);
+int mlx5_mr_ib_cont_pages(struct ib_umem *umem, u64 addr, int *count, int *shift);
+void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context);
+int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
+			    struct ib_mr_status *mr_status);
+
+static inline void init_query_mad(struct ib_smp *mad)
+{
+	mad->base_version  = 1;
+	mad->mgmt_class    = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	mad->class_version = 1;
+	mad->method	   = IB_MGMT_METHOD_GET;
+}
+
+static inline u8 convert_access(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
+	       MLX5_PERM_LOCAL_READ;
+}
+
+#endif /* MLX5_IB_H */
diff --git a/drivers/infiniband/hw/mlx5/mr.c b/drivers/infiniband/hw/mlx5/mr.c
new file mode 100644
index 00000000000..afa873bd028
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/mr.c
@@ -0,0 +1,1250 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+
+#include <linux/kref.h>
+#include <linux/random.h>
+#include <linux/debugfs.h>
+#include <linux/export.h>
+#include <linux/delay.h>
+#include <rdma/ib_umem.h>
+#include "mlx5_ib.h"
+
+enum {
+	MAX_PENDING_REG_MR = 8,
+};
+
+enum {
+	MLX5_UMR_ALIGN	= 2048
+};
+
+static __be64 *mr_align(__be64 *ptr, int align)
+{
+	unsigned long mask = align - 1;
+
+	return (__be64 *)(((unsigned long)ptr + mask) & ~mask);
+}
+
+static int order2idx(struct mlx5_ib_dev *dev, int order)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+
+	if (order < cache->ent[0].order)
+		return 0;
+	else
+		return order - cache->ent[0].order;
+}
+
+static void reg_mr_callback(int status, void *context)
+{
+	struct mlx5_ib_mr *mr = context;
+	struct mlx5_ib_dev *dev = mr->dev;
+	struct mlx5_mr_cache *cache = &dev->cache;
+	int c = order2idx(dev, mr->order);
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	u8 key;
+	unsigned long flags;
+	struct mlx5_mr_table *table = &dev->mdev.priv.mr_table;
+	int err;
+
+	spin_lock_irqsave(&ent->lock, flags);
+	ent->pending--;
+	spin_unlock_irqrestore(&ent->lock, flags);
+	if (status) {
+		mlx5_ib_warn(dev, "async reg mr failed. status %d\n", status);
+		kfree(mr);
+		dev->fill_delay = 1;
+		mod_timer(&dev->delay_timer, jiffies + HZ);
+		return;
+	}
+
+	if (mr->out.hdr.status) {
+		mlx5_ib_warn(dev, "failed - status %d, syndorme 0x%x\n",
+			     mr->out.hdr.status,
+			     be32_to_cpu(mr->out.hdr.syndrome));
+		kfree(mr);
+		dev->fill_delay = 1;
+		mod_timer(&dev->delay_timer, jiffies + HZ);
+		return;
+	}
+
+	spin_lock_irqsave(&dev->mdev.priv.mkey_lock, flags);
+	key = dev->mdev.priv.mkey_key++;
+	spin_unlock_irqrestore(&dev->mdev.priv.mkey_lock, flags);
+	mr->mmr.key = mlx5_idx_to_mkey(be32_to_cpu(mr->out.mkey) & 0xffffff) | key;
+
+	cache->last_add = jiffies;
+
+	spin_lock_irqsave(&ent->lock, flags);
+	list_add_tail(&mr->list, &ent->head);
+	ent->cur++;
+	ent->size++;
+	spin_unlock_irqrestore(&ent->lock, flags);
+
+	write_lock_irqsave(&table->lock, flags);
+	err = radix_tree_insert(&table->tree, mlx5_base_mkey(mr->mmr.key),
+				&mr->mmr);
+	if (err)
+		pr_err("Error inserting to mr tree. 0x%x\n", -err);
+	write_unlock_irqrestore(&table->lock, flags);
+}
+
+static int add_keys(struct mlx5_ib_dev *dev, int c, int num)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int npages = 1 << ent->order;
+	int err = 0;
+	int i;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	for (i = 0; i < num; i++) {
+		if (ent->pending >= MAX_PENDING_REG_MR) {
+			err = -EAGAIN;
+			break;
+		}
+
+		mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+		if (!mr) {
+			err = -ENOMEM;
+			break;
+		}
+		mr->order = ent->order;
+		mr->umred = 1;
+		mr->dev = dev;
+		in->seg.status = 1 << 6;
+		in->seg.xlt_oct_size = cpu_to_be32((npages + 1) / 2);
+		in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+		in->seg.flags = MLX5_ACCESS_MODE_MTT | MLX5_PERM_UMR_EN;
+		in->seg.log2_page_size = 12;
+
+		spin_lock_irq(&ent->lock);
+		ent->pending++;
+		spin_unlock_irq(&ent->lock);
+		err = mlx5_core_create_mkey(&dev->mdev, &mr->mmr, in,
+					    sizeof(*in), reg_mr_callback,
+					    mr, &mr->out);
+		if (err) {
+			mlx5_ib_warn(dev, "create mkey failed %d\n", err);
+			kfree(mr);
+			break;
+		}
+	}
+
+	kfree(in);
+	return err;
+}
+
+static void remove_keys(struct mlx5_ib_dev *dev, int c, int num)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_ib_mr *mr;
+	int err;
+	int i;
+
+	for (i = 0; i < num; i++) {
+		spin_lock_irq(&ent->lock);
+		if (list_empty(&ent->head)) {
+			spin_unlock_irq(&ent->lock);
+			return;
+		}
+		mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list);
+		list_del(&mr->list);
+		ent->cur--;
+		ent->size--;
+		spin_unlock_irq(&ent->lock);
+		err = mlx5_core_destroy_mkey(&dev->mdev, &mr->mmr);
+		if (err)
+			mlx5_ib_warn(dev, "failed destroy mkey\n");
+		else
+			kfree(mr);
+	}
+}
+
+static ssize_t size_write(struct file *filp, const char __user *buf,
+			  size_t count, loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	struct mlx5_ib_dev *dev = ent->dev;
+	char lbuf[20];
+	u32 var;
+	int err;
+	int c;
+
+	if (copy_from_user(lbuf, buf, sizeof(lbuf)))
+		return -EFAULT;
+
+	c = order2idx(dev, ent->order);
+	lbuf[sizeof(lbuf) - 1] = 0;
+
+	if (sscanf(lbuf, "%u", &var) != 1)
+		return -EINVAL;
+
+	if (var < ent->limit)
+		return -EINVAL;
+
+	if (var > ent->size) {
+		do {
+			err = add_keys(dev, c, var - ent->size);
+			if (err && err != -EAGAIN)
+				return err;
+
+			usleep_range(3000, 5000);
+		} while (err);
+	} else if (var < ent->size) {
+		remove_keys(dev, c, ent->size - var);
+	}
+
+	return count;
+}
+
+static ssize_t size_read(struct file *filp, char __user *buf, size_t count,
+			 loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	char lbuf[20];
+	int err;
+
+	if (*pos)
+		return 0;
+
+	err = snprintf(lbuf, sizeof(lbuf), "%d\n", ent->size);
+	if (err < 0)
+		return err;
+
+	if (copy_to_user(buf, lbuf, err))
+		return -EFAULT;
+
+	*pos += err;
+
+	return err;
+}
+
+static const struct file_operations size_fops = {
+	.owner	= THIS_MODULE,
+	.open	= simple_open,
+	.write	= size_write,
+	.read	= size_read,
+};
+
+static ssize_t limit_write(struct file *filp, const char __user *buf,
+			   size_t count, loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	struct mlx5_ib_dev *dev = ent->dev;
+	char lbuf[20];
+	u32 var;
+	int err;
+	int c;
+
+	if (copy_from_user(lbuf, buf, sizeof(lbuf)))
+		return -EFAULT;
+
+	c = order2idx(dev, ent->order);
+	lbuf[sizeof(lbuf) - 1] = 0;
+
+	if (sscanf(lbuf, "%u", &var) != 1)
+		return -EINVAL;
+
+	if (var > ent->size)
+		return -EINVAL;
+
+	ent->limit = var;
+
+	if (ent->cur < ent->limit) {
+		err = add_keys(dev, c, 2 * ent->limit - ent->cur);
+		if (err)
+			return err;
+	}
+
+	return count;
+}
+
+static ssize_t limit_read(struct file *filp, char __user *buf, size_t count,
+			  loff_t *pos)
+{
+	struct mlx5_cache_ent *ent = filp->private_data;
+	char lbuf[20];
+	int err;
+
+	if (*pos)
+		return 0;
+
+	err = snprintf(lbuf, sizeof(lbuf), "%d\n", ent->limit);
+	if (err < 0)
+		return err;
+
+	if (copy_to_user(buf, lbuf, err))
+		return -EFAULT;
+
+	*pos += err;
+
+	return err;
+}
+
+static const struct file_operations limit_fops = {
+	.owner	= THIS_MODULE,
+	.open	= simple_open,
+	.write	= limit_write,
+	.read	= limit_read,
+};
+
+static int someone_adding(struct mlx5_mr_cache *cache)
+{
+	int i;
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		if (cache->ent[i].cur < cache->ent[i].limit)
+			return 1;
+	}
+
+	return 0;
+}
+
+static void __cache_work_func(struct mlx5_cache_ent *ent)
+{
+	struct mlx5_ib_dev *dev = ent->dev;
+	struct mlx5_mr_cache *cache = &dev->cache;
+	int i = order2idx(dev, ent->order);
+	int err;
+
+	if (cache->stopped)
+		return;
+
+	ent = &dev->cache.ent[i];
+	if (ent->cur < 2 * ent->limit && !dev->fill_delay) {
+		err = add_keys(dev, i, 1);
+		if (ent->cur < 2 * ent->limit) {
+			if (err == -EAGAIN) {
+				mlx5_ib_dbg(dev, "returned eagain, order %d\n",
+					    i + 2);
+				queue_delayed_work(cache->wq, &ent->dwork,
+						   msecs_to_jiffies(3));
+			} else if (err) {
+				mlx5_ib_warn(dev, "command failed order %d, err %d\n",
+					     i + 2, err);
+				queue_delayed_work(cache->wq, &ent->dwork,
+						   msecs_to_jiffies(1000));
+			} else {
+				queue_work(cache->wq, &ent->work);
+			}
+		}
+	} else if (ent->cur > 2 * ent->limit) {
+		if (!someone_adding(cache) &&
+		    time_after(jiffies, cache->last_add + 300 * HZ)) {
+			remove_keys(dev, i, 1);
+			if (ent->cur > ent->limit)
+				queue_work(cache->wq, &ent->work);
+		} else {
+			queue_delayed_work(cache->wq, &ent->dwork, 300 * HZ);
+		}
+	}
+}
+
+static void delayed_cache_work_func(struct work_struct *work)
+{
+	struct mlx5_cache_ent *ent;
+
+	ent = container_of(work, struct mlx5_cache_ent, dwork.work);
+	__cache_work_func(ent);
+}
+
+static void cache_work_func(struct work_struct *work)
+{
+	struct mlx5_cache_ent *ent;
+
+	ent = container_of(work, struct mlx5_cache_ent, work);
+	__cache_work_func(ent);
+}
+
+static struct mlx5_ib_mr *alloc_cached_mr(struct mlx5_ib_dev *dev, int order)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_ib_mr *mr = NULL;
+	struct mlx5_cache_ent *ent;
+	int c;
+	int i;
+
+	c = order2idx(dev, order);
+	if (c < 0 || c >= MAX_MR_CACHE_ENTRIES) {
+		mlx5_ib_warn(dev, "order %d, cache index %d\n", order, c);
+		return NULL;
+	}
+
+	for (i = c; i < MAX_MR_CACHE_ENTRIES; i++) {
+		ent = &cache->ent[i];
+
+		mlx5_ib_dbg(dev, "order %d, cache index %d\n", ent->order, i);
+
+		spin_lock_irq(&ent->lock);
+		if (!list_empty(&ent->head)) {
+			mr = list_first_entry(&ent->head, struct mlx5_ib_mr,
+					      list);
+			list_del(&mr->list);
+			ent->cur--;
+			spin_unlock_irq(&ent->lock);
+			if (ent->cur < ent->limit)
+				queue_work(cache->wq, &ent->work);
+			break;
+		}
+		spin_unlock_irq(&ent->lock);
+
+		queue_work(cache->wq, &ent->work);
+
+		if (mr)
+			break;
+	}
+
+	if (!mr)
+		cache->ent[c].miss++;
+
+	return mr;
+}
+
+static void free_cached_mr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int shrink = 0;
+	int c;
+
+	c = order2idx(dev, mr->order);
+	if (c < 0 || c >= MAX_MR_CACHE_ENTRIES) {
+		mlx5_ib_warn(dev, "order %d, cache index %d\n", mr->order, c);
+		return;
+	}
+	ent = &cache->ent[c];
+	spin_lock_irq(&ent->lock);
+	list_add_tail(&mr->list, &ent->head);
+	ent->cur++;
+	if (ent->cur > 2 * ent->limit)
+		shrink = 1;
+	spin_unlock_irq(&ent->lock);
+
+	if (shrink)
+		queue_work(cache->wq, &ent->work);
+}
+
+static void clean_keys(struct mlx5_ib_dev *dev, int c)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent = &cache->ent[c];
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	cancel_delayed_work(&ent->dwork);
+	while (1) {
+		spin_lock_irq(&ent->lock);
+		if (list_empty(&ent->head)) {
+			spin_unlock_irq(&ent->lock);
+			return;
+		}
+		mr = list_first_entry(&ent->head, struct mlx5_ib_mr, list);
+		list_del(&mr->list);
+		ent->cur--;
+		ent->size--;
+		spin_unlock_irq(&ent->lock);
+		err = mlx5_core_destroy_mkey(&dev->mdev, &mr->mmr);
+		if (err)
+			mlx5_ib_warn(dev, "failed destroy mkey\n");
+		else
+			kfree(mr);
+	}
+}
+
+static int mlx5_mr_cache_debugfs_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int i;
+
+	if (!mlx5_debugfs_root)
+		return 0;
+
+	cache->root = debugfs_create_dir("mr_cache", dev->mdev.priv.dbg_root);
+	if (!cache->root)
+		return -ENOMEM;
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		ent = &cache->ent[i];
+		sprintf(ent->name, "%d", ent->order);
+		ent->dir = debugfs_create_dir(ent->name,  cache->root);
+		if (!ent->dir)
+			return -ENOMEM;
+
+		ent->fsize = debugfs_create_file("size", 0600, ent->dir, ent,
+						 &size_fops);
+		if (!ent->fsize)
+			return -ENOMEM;
+
+		ent->flimit = debugfs_create_file("limit", 0600, ent->dir, ent,
+						  &limit_fops);
+		if (!ent->flimit)
+			return -ENOMEM;
+
+		ent->fcur = debugfs_create_u32("cur", 0400, ent->dir,
+					       &ent->cur);
+		if (!ent->fcur)
+			return -ENOMEM;
+
+		ent->fmiss = debugfs_create_u32("miss", 0600, ent->dir,
+						&ent->miss);
+		if (!ent->fmiss)
+			return -ENOMEM;
+	}
+
+	return 0;
+}
+
+static void mlx5_mr_cache_debugfs_cleanup(struct mlx5_ib_dev *dev)
+{
+	if (!mlx5_debugfs_root)
+		return;
+
+	debugfs_remove_recursive(dev->cache.root);
+}
+
+static void delay_time_func(unsigned long ctx)
+{
+	struct mlx5_ib_dev *dev = (struct mlx5_ib_dev *)ctx;
+
+	dev->fill_delay = 0;
+}
+
+int mlx5_mr_cache_init(struct mlx5_ib_dev *dev)
+{
+	struct mlx5_mr_cache *cache = &dev->cache;
+	struct mlx5_cache_ent *ent;
+	int limit;
+	int err;
+	int i;
+
+	cache->wq = create_singlethread_workqueue("mkey_cache");
+	if (!cache->wq) {
+		mlx5_ib_warn(dev, "failed to create work queue\n");
+		return -ENOMEM;
+	}
+
+	setup_timer(&dev->delay_timer, delay_time_func, (unsigned long)dev);
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++) {
+		INIT_LIST_HEAD(&cache->ent[i].head);
+		spin_lock_init(&cache->ent[i].lock);
+
+		ent = &cache->ent[i];
+		INIT_LIST_HEAD(&ent->head);
+		spin_lock_init(&ent->lock);
+		ent->order = i + 2;
+		ent->dev = dev;
+
+		if (dev->mdev.profile->mask & MLX5_PROF_MASK_MR_CACHE)
+			limit = dev->mdev.profile->mr_cache[i].limit;
+		else
+			limit = 0;
+
+		INIT_WORK(&ent->work, cache_work_func);
+		INIT_DELAYED_WORK(&ent->dwork, delayed_cache_work_func);
+		ent->limit = limit;
+		queue_work(cache->wq, &ent->work);
+	}
+
+	err = mlx5_mr_cache_debugfs_init(dev);
+	if (err)
+		mlx5_ib_warn(dev, "cache debugfs failure\n");
+
+	return 0;
+}
+
+int mlx5_mr_cache_cleanup(struct mlx5_ib_dev *dev)
+{
+	int i;
+
+	dev->cache.stopped = 1;
+	flush_workqueue(dev->cache.wq);
+
+	mlx5_mr_cache_debugfs_cleanup(dev);
+
+	for (i = 0; i < MAX_MR_CACHE_ENTRIES; i++)
+		clean_keys(dev, i);
+
+	destroy_workqueue(dev->cache.wq);
+	del_timer_sync(&dev->delay_timer);
+
+	return 0;
+}
+
+struct ib_mr *mlx5_ib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_core_dev *mdev = &dev->mdev;
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_mkey_seg *seg;
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	seg = &in->seg;
+	seg->flags = convert_access(acc) | MLX5_ACCESS_MODE_PA;
+	seg->flags_pd = cpu_to_be32(to_mpd(pd)->pdn | MLX5_MKEY_LEN64);
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	seg->start_addr = 0;
+
+	err = mlx5_core_create_mkey(mdev, &mr->mmr, in, sizeof(*in), NULL, NULL,
+				    NULL);
+	if (err)
+		goto err_in;
+
+	kfree(in);
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+err_in:
+	kfree(in);
+
+err_free:
+	kfree(mr);
+
+	return ERR_PTR(err);
+}
+
+static int get_octo_len(u64 addr, u64 len, int page_size)
+{
+	u64 offset;
+	int npages;
+
+	offset = addr & (page_size - 1);
+	npages = ALIGN(len + offset, page_size) >> ilog2(page_size);
+	return (npages + 1) / 2;
+}
+
+static int use_umr(int order)
+{
+	return order <= 17;
+}
+
+static void prep_umr_reg_wqe(struct ib_pd *pd, struct ib_send_wr *wr,
+			     struct ib_sge *sg, u64 dma, int n, u32 key,
+			     int page_shift, u64 virt_addr, u64 len,
+			     int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct ib_mr *mr = dev->umrc.mr;
+
+	sg->addr = dma;
+	sg->length = ALIGN(sizeof(u64) * n, 64);
+	sg->lkey = mr->lkey;
+
+	wr->next = NULL;
+	wr->send_flags = 0;
+	wr->sg_list = sg;
+	if (n)
+		wr->num_sge = 1;
+	else
+		wr->num_sge = 0;
+
+	wr->opcode = MLX5_IB_WR_UMR;
+	wr->wr.fast_reg.page_list_len = n;
+	wr->wr.fast_reg.page_shift = page_shift;
+	wr->wr.fast_reg.rkey = key;
+	wr->wr.fast_reg.iova_start = virt_addr;
+	wr->wr.fast_reg.length = len;
+	wr->wr.fast_reg.access_flags = access_flags;
+	wr->wr.fast_reg.page_list = (struct ib_fast_reg_page_list *)pd;
+}
+
+static void prep_umr_unreg_wqe(struct mlx5_ib_dev *dev,
+			       struct ib_send_wr *wr, u32 key)
+{
+	wr->send_flags = MLX5_IB_SEND_UMR_UNREG;
+	wr->opcode = MLX5_IB_WR_UMR;
+	wr->wr.fast_reg.rkey = key;
+}
+
+void mlx5_umr_cq_handler(struct ib_cq *cq, void *cq_context)
+{
+	struct mlx5_ib_umr_context *context;
+	struct ib_wc wc;
+	int err;
+
+	while (1) {
+		err = ib_poll_cq(cq, 1, &wc);
+		if (err < 0) {
+			pr_warn("poll cq error %d\n", err);
+			return;
+		}
+		if (err == 0)
+			break;
+
+		context = (struct mlx5_ib_umr_context *) (unsigned long) wc.wr_id;
+		context->status = wc.status;
+		complete(&context->done);
+	}
+	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
+}
+
+static struct mlx5_ib_mr *reg_umr(struct ib_pd *pd, struct ib_umem *umem,
+				  u64 virt_addr, u64 len, int npages,
+				  int page_shift, int order, int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct device *ddev = dev->ib_dev.dma_device;
+	struct umr_common *umrc = &dev->umrc;
+	struct mlx5_ib_umr_context umr_context;
+	struct ib_send_wr wr, *bad;
+	struct mlx5_ib_mr *mr;
+	struct ib_sge sg;
+	int size = sizeof(u64) * npages;
+	int err = 0;
+	int i;
+
+	for (i = 0; i < 1; i++) {
+		mr = alloc_cached_mr(dev, order);
+		if (mr)
+			break;
+
+		err = add_keys(dev, order2idx(dev, order), 1);
+		if (err && err != -EAGAIN) {
+			mlx5_ib_warn(dev, "add_keys failed, err %d\n", err);
+			break;
+		}
+	}
+
+	if (!mr)
+		return ERR_PTR(-EAGAIN);
+
+	mr->pas = kmalloc(size + MLX5_UMR_ALIGN - 1, GFP_KERNEL);
+	if (!mr->pas) {
+		err = -ENOMEM;
+		goto free_mr;
+	}
+
+	mlx5_ib_populate_pas(dev, umem, page_shift,
+			     mr_align(mr->pas, MLX5_UMR_ALIGN), 1);
+
+	mr->dma = dma_map_single(ddev, mr_align(mr->pas, MLX5_UMR_ALIGN), size,
+				 DMA_TO_DEVICE);
+	if (dma_mapping_error(ddev, mr->dma)) {
+		err = -ENOMEM;
+		goto free_pas;
+	}
+
+	memset(&wr, 0, sizeof(wr));
+	wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_reg_wqe(pd, &wr, &sg, mr->dma, npages, mr->mmr.key, page_shift, virt_addr, len, access_flags);
+
+	mlx5_ib_init_umr_context(&umr_context);
+	down(&umrc->sem);
+	err = ib_post_send(umrc->qp, &wr, &bad);
+	if (err) {
+		mlx5_ib_warn(dev, "post send failed, err %d\n", err);
+		goto unmap_dma;
+	} else {
+		wait_for_completion(&umr_context.done);
+		if (umr_context.status != IB_WC_SUCCESS) {
+			mlx5_ib_warn(dev, "reg umr failed\n");
+			err = -EFAULT;
+		}
+	}
+
+	mr->mmr.iova = virt_addr;
+	mr->mmr.size = len;
+	mr->mmr.pd = to_mpd(pd)->pdn;
+
+unmap_dma:
+	up(&umrc->sem);
+	dma_unmap_single(ddev, mr->dma, size, DMA_TO_DEVICE);
+
+free_pas:
+	kfree(mr->pas);
+
+free_mr:
+	if (err) {
+		free_cached_mr(dev, mr);
+		return ERR_PTR(err);
+	}
+
+	return mr;
+}
+
+static struct mlx5_ib_mr *reg_create(struct ib_pd *pd, u64 virt_addr,
+				     u64 length, struct ib_umem *umem,
+				     int npages, int page_shift,
+				     int access_flags)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int inlen;
+	int err;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	inlen = sizeof(*in) + sizeof(*in->pas) * ((npages + 1) / 2) * 2;
+	in = mlx5_vzalloc(inlen);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_1;
+	}
+	mlx5_ib_populate_pas(dev, umem, page_shift, in->pas, 0);
+
+	in->seg.flags = convert_access(access_flags) |
+		MLX5_ACCESS_MODE_MTT;
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	in->seg.start_addr = cpu_to_be64(virt_addr);
+	in->seg.len = cpu_to_be64(length);
+	in->seg.bsfs_octo_size = 0;
+	in->seg.xlt_oct_size = cpu_to_be32(get_octo_len(virt_addr, length, 1 << page_shift));
+	in->seg.log2_page_size = page_shift;
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->xlat_oct_act_size = cpu_to_be32(get_octo_len(virt_addr, length,
+							 1 << page_shift));
+	err = mlx5_core_create_mkey(&dev->mdev, &mr->mmr, in, inlen, NULL,
+				    NULL, NULL);
+	if (err) {
+		mlx5_ib_warn(dev, "create mkey failed\n");
+		goto err_2;
+	}
+	mr->umem = umem;
+	mlx5_vfree(in);
+
+	mlx5_ib_dbg(dev, "mkey = 0x%x\n", mr->mmr.key);
+
+	return mr;
+
+err_2:
+	mlx5_vfree(in);
+
+err_1:
+	kfree(mr);
+
+	return ERR_PTR(err);
+}
+
+struct ib_mr *mlx5_ib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+				  u64 virt_addr, int access_flags,
+				  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_mr *mr = NULL;
+	struct ib_umem *umem;
+	int page_shift;
+	int npages;
+	int ncont;
+	int order;
+	int err;
+
+	mlx5_ib_dbg(dev, "start 0x%llx, virt_addr 0x%llx, length 0x%llx\n",
+		    start, virt_addr, length);
+	umem = ib_umem_get(pd->uobject->context, start, length, access_flags,
+			   0);
+	if (IS_ERR(umem)) {
+		mlx5_ib_dbg(dev, "umem get failed\n");
+		return (void *)umem;
+	}
+
+	mlx5_ib_cont_pages(umem, start, &npages, &page_shift, &ncont, &order);
+	if (!npages) {
+		mlx5_ib_warn(dev, "avoid zero region\n");
+		err = -EINVAL;
+		goto error;
+	}
+
+	mlx5_ib_dbg(dev, "npages %d, ncont %d, order %d, page_shift %d\n",
+		    npages, ncont, order, page_shift);
+
+	if (use_umr(order)) {
+		mr = reg_umr(pd, umem, virt_addr, length, ncont, page_shift,
+			     order, access_flags);
+		if (PTR_ERR(mr) == -EAGAIN) {
+			mlx5_ib_dbg(dev, "cache empty for order %d", order);
+			mr = NULL;
+		}
+	}
+
+	if (!mr)
+		mr = reg_create(pd, virt_addr, length, umem, ncont, page_shift,
+				access_flags);
+
+	if (IS_ERR(mr)) {
+		err = PTR_ERR(mr);
+		goto error;
+	}
+
+	mlx5_ib_dbg(dev, "mkey 0x%x\n", mr->mmr.key);
+
+	mr->umem = umem;
+	mr->npages = npages;
+	spin_lock(&dev->mr_lock);
+	dev->mdev.priv.reg_pages += npages;
+	spin_unlock(&dev->mr_lock);
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+
+	return &mr->ibmr;
+
+error:
+	ib_umem_release(umem);
+	return ERR_PTR(err);
+}
+
+static int unreg_umr(struct mlx5_ib_dev *dev, struct mlx5_ib_mr *mr)
+{
+	struct umr_common *umrc = &dev->umrc;
+	struct mlx5_ib_umr_context umr_context;
+	struct ib_send_wr wr, *bad;
+	int err;
+
+	memset(&wr, 0, sizeof(wr));
+	wr.wr_id = (u64)(unsigned long)&umr_context;
+	prep_umr_unreg_wqe(dev, &wr, mr->mmr.key);
+
+	mlx5_ib_init_umr_context(&umr_context);
+	down(&umrc->sem);
+	err = ib_post_send(umrc->qp, &wr, &bad);
+	if (err) {
+		up(&umrc->sem);
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto error;
+	} else {
+		wait_for_completion(&umr_context.done);
+		up(&umrc->sem);
+	}
+	if (umr_context.status != IB_WC_SUCCESS) {
+		mlx5_ib_warn(dev, "unreg umr failed\n");
+		err = -EFAULT;
+		goto error;
+	}
+	return 0;
+
+error:
+	return err;
+}
+
+int mlx5_ib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibmr->device);
+	struct mlx5_ib_mr *mr = to_mmr(ibmr);
+	struct ib_umem *umem = mr->umem;
+	int npages = mr->npages;
+	int umred = mr->umred;
+	int err;
+
+	if (!umred) {
+		err = mlx5_core_destroy_mkey(&dev->mdev, &mr->mmr);
+		if (err) {
+			mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n",
+				     mr->mmr.key, err);
+			return err;
+		}
+	} else {
+		err = unreg_umr(dev, mr);
+		if (err) {
+			mlx5_ib_warn(dev, "failed unregister\n");
+			return err;
+		}
+		free_cached_mr(dev, mr);
+	}
+
+	if (umem) {
+		ib_umem_release(umem);
+		spin_lock(&dev->mr_lock);
+		dev->mdev.priv.reg_pages -= npages;
+		spin_unlock(&dev->mr_lock);
+	}
+
+	if (!umred)
+		kfree(mr);
+
+	return 0;
+}
+
+struct ib_mr *mlx5_ib_create_mr(struct ib_pd *pd,
+				struct ib_mr_init_attr *mr_init_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int access_mode, err;
+	int ndescs = roundup(mr_init_attr->max_reg_descriptors, 4);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	in->seg.status = 1 << 6; /* free */
+	in->seg.xlt_oct_size = cpu_to_be32(ndescs);
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	access_mode = MLX5_ACCESS_MODE_MTT;
+
+	if (mr_init_attr->flags & IB_MR_SIGNATURE_EN) {
+		u32 psv_index[2];
+
+		in->seg.flags_pd = cpu_to_be32(be32_to_cpu(in->seg.flags_pd) |
+							   MLX5_MKEY_BSF_EN);
+		in->seg.bsfs_octo_size = cpu_to_be32(MLX5_MKEY_BSF_OCTO_SIZE);
+		mr->sig = kzalloc(sizeof(*mr->sig), GFP_KERNEL);
+		if (!mr->sig) {
+			err = -ENOMEM;
+			goto err_free_in;
+		}
+
+		/* create mem & wire PSVs */
+		err = mlx5_core_create_psv(&dev->mdev, to_mpd(pd)->pdn,
+					   2, psv_index);
+		if (err)
+			goto err_free_sig;
+
+		access_mode = MLX5_ACCESS_MODE_KLM;
+		mr->sig->psv_memory.psv_idx = psv_index[0];
+		mr->sig->psv_wire.psv_idx = psv_index[1];
+
+		mr->sig->sig_status_checked = true;
+		mr->sig->sig_err_exists = false;
+		/* Next UMR, Arm SIGERR */
+		++mr->sig->sigerr_count;
+	}
+
+	in->seg.flags = MLX5_PERM_UMR_EN | access_mode;
+	err = mlx5_core_create_mkey(&dev->mdev, &mr->mmr, in, sizeof(*in),
+				    NULL, NULL, NULL);
+	if (err)
+		goto err_destroy_psv;
+
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+	kfree(in);
+
+	return &mr->ibmr;
+
+err_destroy_psv:
+	if (mr->sig) {
+		if (mlx5_core_destroy_psv(&dev->mdev,
+					  mr->sig->psv_memory.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy mem psv %d\n",
+				     mr->sig->psv_memory.psv_idx);
+		if (mlx5_core_destroy_psv(&dev->mdev,
+					  mr->sig->psv_wire.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy wire psv %d\n",
+				     mr->sig->psv_wire.psv_idx);
+	}
+err_free_sig:
+	kfree(mr->sig);
+err_free_in:
+	kfree(in);
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+int mlx5_ib_destroy_mr(struct ib_mr *ibmr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibmr->device);
+	struct mlx5_ib_mr *mr = to_mmr(ibmr);
+	int err;
+
+	if (mr->sig) {
+		if (mlx5_core_destroy_psv(&dev->mdev,
+					  mr->sig->psv_memory.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy mem psv %d\n",
+				     mr->sig->psv_memory.psv_idx);
+		if (mlx5_core_destroy_psv(&dev->mdev,
+					  mr->sig->psv_wire.psv_idx))
+			mlx5_ib_warn(dev, "failed to destroy wire psv %d\n",
+				     mr->sig->psv_wire.psv_idx);
+		kfree(mr->sig);
+	}
+
+	err = mlx5_core_destroy_mkey(&dev->mdev, &mr->mmr);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to destroy mkey 0x%x (%d)\n",
+			     mr->mmr.key, err);
+		return err;
+	}
+
+	kfree(mr);
+
+	return err;
+}
+
+struct ib_mr *mlx5_ib_alloc_fast_reg_mr(struct ib_pd *pd,
+					int max_page_list_len)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_create_mkey_mbox_in *in;
+	struct mlx5_ib_mr *mr;
+	int err;
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in) {
+		err = -ENOMEM;
+		goto err_free;
+	}
+
+	in->seg.status = 1 << 6; /* free */
+	in->seg.xlt_oct_size = cpu_to_be32((max_page_list_len + 1) / 2);
+	in->seg.qpn_mkey7_0 = cpu_to_be32(0xffffff << 8);
+	in->seg.flags = MLX5_PERM_UMR_EN | MLX5_ACCESS_MODE_MTT;
+	in->seg.flags_pd = cpu_to_be32(to_mpd(pd)->pdn);
+	/*
+	 * TBD not needed - issue 197292 */
+	in->seg.log2_page_size = PAGE_SHIFT;
+
+	err = mlx5_core_create_mkey(&dev->mdev, &mr->mmr, in, sizeof(*in), NULL,
+				    NULL, NULL);
+	kfree(in);
+	if (err)
+		goto err_free;
+
+	mr->ibmr.lkey = mr->mmr.key;
+	mr->ibmr.rkey = mr->mmr.key;
+	mr->umem = NULL;
+
+	return &mr->ibmr;
+
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+struct ib_fast_reg_page_list *mlx5_ib_alloc_fast_reg_page_list(struct ib_device *ibdev,
+							       int page_list_len)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl;
+	int size = page_list_len * sizeof(u64);
+
+	mfrpl = kmalloc(sizeof(*mfrpl), GFP_KERNEL);
+	if (!mfrpl)
+		return ERR_PTR(-ENOMEM);
+
+	mfrpl->ibfrpl.page_list = kmalloc(size, GFP_KERNEL);
+	if (!mfrpl->ibfrpl.page_list)
+		goto err_free;
+
+	mfrpl->mapped_page_list = dma_alloc_coherent(ibdev->dma_device,
+						     size, &mfrpl->map,
+						     GFP_KERNEL);
+	if (!mfrpl->mapped_page_list)
+		goto err_free;
+
+	WARN_ON(mfrpl->map & 0x3f);
+
+	return &mfrpl->ibfrpl;
+
+err_free:
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+	return ERR_PTR(-ENOMEM);
+}
+
+void mlx5_ib_free_fast_reg_page_list(struct ib_fast_reg_page_list *page_list)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(page_list);
+	struct mlx5_ib_dev *dev = to_mdev(page_list->device);
+	int size = page_list->max_page_list_len * sizeof(u64);
+
+	dma_free_coherent(&dev->mdev.pdev->dev, size, mfrpl->mapped_page_list,
+			  mfrpl->map);
+	kfree(mfrpl->ibfrpl.page_list);
+	kfree(mfrpl);
+}
+
+int mlx5_ib_check_mr_status(struct ib_mr *ibmr, u32 check_mask,
+			    struct ib_mr_status *mr_status)
+{
+	struct mlx5_ib_mr *mmr = to_mmr(ibmr);
+	int ret = 0;
+
+	if (check_mask & ~IB_MR_CHECK_SIG_STATUS) {
+		pr_err("Invalid status check mask\n");
+		ret = -EINVAL;
+		goto done;
+	}
+
+	mr_status->fail_status = 0;
+	if (check_mask & IB_MR_CHECK_SIG_STATUS) {
+		if (!mmr->sig) {
+			ret = -EINVAL;
+			pr_err("signature status check requested on a non-signature enabled MR\n");
+			goto done;
+		}
+
+		mmr->sig->sig_status_checked = true;
+		if (!mmr->sig->sig_err_exists)
+			goto done;
+
+		if (ibmr->lkey == mmr->sig->err_item.key)
+			memcpy(&mr_status->sig_err, &mmr->sig->err_item,
+			       sizeof(mr_status->sig_err));
+		else {
+			mr_status->sig_err.err_type = IB_SIG_BAD_GUARD;
+			mr_status->sig_err.sig_err_offset = 0;
+			mr_status->sig_err.key = mmr->sig->err_item.key;
+		}
+
+		mmr->sig->sig_err_exists = false;
+		mr_status->fail_status |= IB_MR_CHECK_SIG_STATUS;
+	}
+
+done:
+	return ret;
+}
diff --git a/drivers/infiniband/hw/mlx5/qp.c b/drivers/infiniband/hw/mlx5/qp.c
new file mode 100644
index 00000000000..bbbcf389272
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/qp.c
@@ -0,0 +1,3048 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <rdma/ib_umem.h>
+#include "mlx5_ib.h"
+#include "user.h"
+
+/* not supported currently */
+static int wq_signature;
+
+enum {
+	MLX5_IB_ACK_REQ_FREQ	= 8,
+};
+
+enum {
+	MLX5_IB_DEFAULT_SCHED_QUEUE	= 0x83,
+	MLX5_IB_DEFAULT_QP0_SCHED_QUEUE	= 0x3f,
+	MLX5_IB_LINK_TYPE_IB		= 0,
+	MLX5_IB_LINK_TYPE_ETH		= 1
+};
+
+enum {
+	MLX5_IB_SQ_STRIDE	= 6,
+	MLX5_IB_CACHE_LINE_SIZE	= 64,
+};
+
+static const u32 mlx5_ib_opcode[] = {
+	[IB_WR_SEND]				= MLX5_OPCODE_SEND,
+	[IB_WR_SEND_WITH_IMM]			= MLX5_OPCODE_SEND_IMM,
+	[IB_WR_RDMA_WRITE]			= MLX5_OPCODE_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM]		= MLX5_OPCODE_RDMA_WRITE_IMM,
+	[IB_WR_RDMA_READ]			= MLX5_OPCODE_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP]		= MLX5_OPCODE_ATOMIC_CS,
+	[IB_WR_ATOMIC_FETCH_AND_ADD]		= MLX5_OPCODE_ATOMIC_FA,
+	[IB_WR_SEND_WITH_INV]			= MLX5_OPCODE_SEND_INVAL,
+	[IB_WR_LOCAL_INV]			= MLX5_OPCODE_UMR,
+	[IB_WR_FAST_REG_MR]			= MLX5_OPCODE_UMR,
+	[IB_WR_MASKED_ATOMIC_CMP_AND_SWP]	= MLX5_OPCODE_ATOMIC_MASKED_CS,
+	[IB_WR_MASKED_ATOMIC_FETCH_AND_ADD]	= MLX5_OPCODE_ATOMIC_MASKED_FA,
+	[MLX5_IB_WR_UMR]			= MLX5_OPCODE_UMR,
+};
+
+struct umr_wr {
+	u64				virt_addr;
+	struct ib_pd		       *pd;
+	unsigned int			page_shift;
+	unsigned int			npages;
+	u32				length;
+	int				access_flags;
+	u32				mkey;
+};
+
+static int is_qp0(enum ib_qp_type qp_type)
+{
+	return qp_type == IB_QPT_SMI;
+}
+
+static int is_qp1(enum ib_qp_type qp_type)
+{
+	return qp_type == IB_QPT_GSI;
+}
+
+static int is_sqp(enum ib_qp_type qp_type)
+{
+	return is_qp0(qp_type) || is_qp1(qp_type);
+}
+
+static void *get_wqe(struct mlx5_ib_qp *qp, int offset)
+{
+	return mlx5_buf_offset(&qp->buf, offset);
+}
+
+static void *get_recv_wqe(struct mlx5_ib_qp *qp, int n)
+{
+	return get_wqe(qp, qp->rq.offset + (n << qp->rq.wqe_shift));
+}
+
+void *mlx5_get_send_wqe(struct mlx5_ib_qp *qp, int n)
+{
+	return get_wqe(qp, qp->sq.offset + (n << MLX5_IB_SQ_STRIDE));
+}
+
+static void mlx5_ib_qp_event(struct mlx5_core_qp *qp, int type)
+{
+	struct ib_qp *ibqp = &to_mibqp(qp)->ibqp;
+	struct ib_event event;
+
+	if (type == MLX5_EVENT_TYPE_PATH_MIG)
+		to_mibqp(qp)->port = to_mibqp(qp)->alt_port;
+
+	if (ibqp->event_handler) {
+		event.device     = ibqp->device;
+		event.element.qp = ibqp;
+		switch (type) {
+		case MLX5_EVENT_TYPE_PATH_MIG:
+			event.event = IB_EVENT_PATH_MIG;
+			break;
+		case MLX5_EVENT_TYPE_COMM_EST:
+			event.event = IB_EVENT_COMM_EST;
+			break;
+		case MLX5_EVENT_TYPE_SQ_DRAINED:
+			event.event = IB_EVENT_SQ_DRAINED;
+			break;
+		case MLX5_EVENT_TYPE_SRQ_LAST_WQE:
+			event.event = IB_EVENT_QP_LAST_WQE_REACHED;
+			break;
+		case MLX5_EVENT_TYPE_WQ_CATAS_ERROR:
+			event.event = IB_EVENT_QP_FATAL;
+			break;
+		case MLX5_EVENT_TYPE_PATH_MIG_FAILED:
+			event.event = IB_EVENT_PATH_MIG_ERR;
+			break;
+		case MLX5_EVENT_TYPE_WQ_INVAL_REQ_ERROR:
+			event.event = IB_EVENT_QP_REQ_ERR;
+			break;
+		case MLX5_EVENT_TYPE_WQ_ACCESS_ERROR:
+			event.event = IB_EVENT_QP_ACCESS_ERR;
+			break;
+		default:
+			pr_warn("mlx5_ib: Unexpected event type %d on QP %06x\n", type, qp->qpn);
+			return;
+		}
+
+		ibqp->event_handler(&event, ibqp->qp_context);
+	}
+}
+
+static int set_rq_size(struct mlx5_ib_dev *dev, struct ib_qp_cap *cap,
+		       int has_rq, struct mlx5_ib_qp *qp, struct mlx5_ib_create_qp *ucmd)
+{
+	int wqe_size;
+	int wq_size;
+
+	/* Sanity check RQ size before proceeding */
+	if (cap->max_recv_wr  > dev->mdev.caps.max_wqes)
+		return -EINVAL;
+
+	if (!has_rq) {
+		qp->rq.max_gs = 0;
+		qp->rq.wqe_cnt = 0;
+		qp->rq.wqe_shift = 0;
+	} else {
+		if (ucmd) {
+			qp->rq.wqe_cnt = ucmd->rq_wqe_count;
+			qp->rq.wqe_shift = ucmd->rq_wqe_shift;
+			qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig;
+			qp->rq.max_post = qp->rq.wqe_cnt;
+		} else {
+			wqe_size = qp->wq_sig ? sizeof(struct mlx5_wqe_signature_seg) : 0;
+			wqe_size += cap->max_recv_sge * sizeof(struct mlx5_wqe_data_seg);
+			wqe_size = roundup_pow_of_two(wqe_size);
+			wq_size = roundup_pow_of_two(cap->max_recv_wr) * wqe_size;
+			wq_size = max_t(int, wq_size, MLX5_SEND_WQE_BB);
+			qp->rq.wqe_cnt = wq_size / wqe_size;
+			if (wqe_size > dev->mdev.caps.max_rq_desc_sz) {
+				mlx5_ib_dbg(dev, "wqe_size %d, max %d\n",
+					    wqe_size,
+					    dev->mdev.caps.max_rq_desc_sz);
+				return -EINVAL;
+			}
+			qp->rq.wqe_shift = ilog2(wqe_size);
+			qp->rq.max_gs = (1 << qp->rq.wqe_shift) / sizeof(struct mlx5_wqe_data_seg) - qp->wq_sig;
+			qp->rq.max_post = qp->rq.wqe_cnt;
+		}
+	}
+
+	return 0;
+}
+
+static int sq_overhead(enum ib_qp_type qp_type)
+{
+	int size = 0;
+
+	switch (qp_type) {
+	case IB_QPT_XRC_INI:
+		size += sizeof(struct mlx5_wqe_xrc_seg);
+		/* fall through */
+	case IB_QPT_RC:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_atomic_seg) +
+			sizeof(struct mlx5_wqe_raddr_seg);
+		break;
+
+	case IB_QPT_XRC_TGT:
+		return 0;
+
+	case IB_QPT_UC:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_raddr_seg) +
+			sizeof(struct mlx5_wqe_umr_ctrl_seg) +
+			sizeof(struct mlx5_mkey_seg);
+		break;
+
+	case IB_QPT_UD:
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_datagram_seg);
+		break;
+
+	case MLX5_IB_QPT_REG_UMR:
+		size += sizeof(struct mlx5_wqe_ctrl_seg) +
+			sizeof(struct mlx5_wqe_umr_ctrl_seg) +
+			sizeof(struct mlx5_mkey_seg);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	return size;
+}
+
+static int calc_send_wqe(struct ib_qp_init_attr *attr)
+{
+	int inl_size = 0;
+	int size;
+
+	size = sq_overhead(attr->qp_type);
+	if (size < 0)
+		return size;
+
+	if (attr->cap.max_inline_data) {
+		inl_size = size + sizeof(struct mlx5_wqe_inline_seg) +
+			attr->cap.max_inline_data;
+	}
+
+	size += attr->cap.max_send_sge * sizeof(struct mlx5_wqe_data_seg);
+	if (attr->create_flags & IB_QP_CREATE_SIGNATURE_EN &&
+	    ALIGN(max_t(int, inl_size, size), MLX5_SEND_WQE_BB) < MLX5_SIG_WQE_SIZE)
+			return MLX5_SIG_WQE_SIZE;
+	else
+		return ALIGN(max_t(int, inl_size, size), MLX5_SEND_WQE_BB);
+}
+
+static int calc_sq_size(struct mlx5_ib_dev *dev, struct ib_qp_init_attr *attr,
+			struct mlx5_ib_qp *qp)
+{
+	int wqe_size;
+	int wq_size;
+
+	if (!attr->cap.max_send_wr)
+		return 0;
+
+	wqe_size = calc_send_wqe(attr);
+	mlx5_ib_dbg(dev, "wqe_size %d\n", wqe_size);
+	if (wqe_size < 0)
+		return wqe_size;
+
+	if (wqe_size > dev->mdev.caps.max_sq_desc_sz) {
+		mlx5_ib_dbg(dev, "wqe_size(%d) > max_sq_desc_sz(%d)\n",
+			    wqe_size, dev->mdev.caps.max_sq_desc_sz);
+		return -EINVAL;
+	}
+
+	qp->max_inline_data = wqe_size - sq_overhead(attr->qp_type) -
+		sizeof(struct mlx5_wqe_inline_seg);
+	attr->cap.max_inline_data = qp->max_inline_data;
+
+	if (attr->create_flags & IB_QP_CREATE_SIGNATURE_EN)
+		qp->signature_en = true;
+
+	wq_size = roundup_pow_of_two(attr->cap.max_send_wr * wqe_size);
+	qp->sq.wqe_cnt = wq_size / MLX5_SEND_WQE_BB;
+	if (qp->sq.wqe_cnt > dev->mdev.caps.max_wqes) {
+		mlx5_ib_dbg(dev, "wqe count(%d) exceeds limits(%d)\n",
+			    qp->sq.wqe_cnt, dev->mdev.caps.max_wqes);
+		return -ENOMEM;
+	}
+	qp->sq.wqe_shift = ilog2(MLX5_SEND_WQE_BB);
+	qp->sq.max_gs = attr->cap.max_send_sge;
+	qp->sq.max_post = wq_size / wqe_size;
+	attr->cap.max_send_wr = qp->sq.max_post;
+
+	return wq_size;
+}
+
+static int set_user_buf_size(struct mlx5_ib_dev *dev,
+			    struct mlx5_ib_qp *qp,
+			    struct mlx5_ib_create_qp *ucmd)
+{
+	int desc_sz = 1 << qp->sq.wqe_shift;
+
+	if (desc_sz > dev->mdev.caps.max_sq_desc_sz) {
+		mlx5_ib_warn(dev, "desc_sz %d, max_sq_desc_sz %d\n",
+			     desc_sz, dev->mdev.caps.max_sq_desc_sz);
+		return -EINVAL;
+	}
+
+	if (ucmd->sq_wqe_count && ((1 << ilog2(ucmd->sq_wqe_count)) != ucmd->sq_wqe_count)) {
+		mlx5_ib_warn(dev, "sq_wqe_count %d, sq_wqe_count %d\n",
+			     ucmd->sq_wqe_count, ucmd->sq_wqe_count);
+		return -EINVAL;
+	}
+
+	qp->sq.wqe_cnt = ucmd->sq_wqe_count;
+
+	if (qp->sq.wqe_cnt > dev->mdev.caps.max_wqes) {
+		mlx5_ib_warn(dev, "wqe_cnt %d, max_wqes %d\n",
+			     qp->sq.wqe_cnt, dev->mdev.caps.max_wqes);
+		return -EINVAL;
+	}
+
+	qp->buf_size = (qp->rq.wqe_cnt << qp->rq.wqe_shift) +
+		(qp->sq.wqe_cnt << 6);
+
+	return 0;
+}
+
+static int qp_has_rq(struct ib_qp_init_attr *attr)
+{
+	if (attr->qp_type == IB_QPT_XRC_INI ||
+	    attr->qp_type == IB_QPT_XRC_TGT || attr->srq ||
+	    attr->qp_type == MLX5_IB_QPT_REG_UMR ||
+	    !attr->cap.max_recv_wr)
+		return 0;
+
+	return 1;
+}
+
+static int first_med_uuar(void)
+{
+	return 1;
+}
+
+static int next_uuar(int n)
+{
+	n++;
+
+	while (((n % 4) & 2))
+		n++;
+
+	return n;
+}
+
+static int num_med_uuar(struct mlx5_uuar_info *uuari)
+{
+	int n;
+
+	n = uuari->num_uars * MLX5_NON_FP_BF_REGS_PER_PAGE -
+		uuari->num_low_latency_uuars - 1;
+
+	return n >= 0 ? n : 0;
+}
+
+static int max_uuari(struct mlx5_uuar_info *uuari)
+{
+	return uuari->num_uars * 4;
+}
+
+static int first_hi_uuar(struct mlx5_uuar_info *uuari)
+{
+	int med;
+	int i;
+	int t;
+
+	med = num_med_uuar(uuari);
+	for (t = 0, i = first_med_uuar();; i = next_uuar(i)) {
+		t++;
+		if (t == med)
+			return next_uuar(i);
+	}
+
+	return 0;
+}
+
+static int alloc_high_class_uuar(struct mlx5_uuar_info *uuari)
+{
+	int i;
+
+	for (i = first_hi_uuar(uuari); i < max_uuari(uuari); i = next_uuar(i)) {
+		if (!test_bit(i, uuari->bitmap)) {
+			set_bit(i, uuari->bitmap);
+			uuari->count[i]++;
+			return i;
+		}
+	}
+
+	return -ENOMEM;
+}
+
+static int alloc_med_class_uuar(struct mlx5_uuar_info *uuari)
+{
+	int minidx = first_med_uuar();
+	int i;
+
+	for (i = first_med_uuar(); i < first_hi_uuar(uuari); i = next_uuar(i)) {
+		if (uuari->count[i] < uuari->count[minidx])
+			minidx = i;
+	}
+
+	uuari->count[minidx]++;
+	return minidx;
+}
+
+static int alloc_uuar(struct mlx5_uuar_info *uuari,
+		      enum mlx5_ib_latency_class lat)
+{
+	int uuarn = -EINVAL;
+
+	mutex_lock(&uuari->lock);
+	switch (lat) {
+	case MLX5_IB_LATENCY_CLASS_LOW:
+		uuarn = 0;
+		uuari->count[uuarn]++;
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_MEDIUM:
+		if (uuari->ver < 2)
+			uuarn = -ENOMEM;
+		else
+			uuarn = alloc_med_class_uuar(uuari);
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_HIGH:
+		if (uuari->ver < 2)
+			uuarn = -ENOMEM;
+		else
+			uuarn = alloc_high_class_uuar(uuari);
+		break;
+
+	case MLX5_IB_LATENCY_CLASS_FAST_PATH:
+		uuarn = 2;
+		break;
+	}
+	mutex_unlock(&uuari->lock);
+
+	return uuarn;
+}
+
+static void free_med_class_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	clear_bit(uuarn, uuari->bitmap);
+	--uuari->count[uuarn];
+}
+
+static void free_high_class_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	clear_bit(uuarn, uuari->bitmap);
+	--uuari->count[uuarn];
+}
+
+static void free_uuar(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	int nuuars = uuari->num_uars * MLX5_BF_REGS_PER_PAGE;
+	int high_uuar = nuuars - uuari->num_low_latency_uuars;
+
+	mutex_lock(&uuari->lock);
+	if (uuarn == 0) {
+		--uuari->count[uuarn];
+		goto out;
+	}
+
+	if (uuarn < high_uuar) {
+		free_med_class_uuar(uuari, uuarn);
+		goto out;
+	}
+
+	free_high_class_uuar(uuari, uuarn);
+
+out:
+	mutex_unlock(&uuari->lock);
+}
+
+static enum mlx5_qp_state to_mlx5_state(enum ib_qp_state state)
+{
+	switch (state) {
+	case IB_QPS_RESET:	return MLX5_QP_STATE_RST;
+	case IB_QPS_INIT:	return MLX5_QP_STATE_INIT;
+	case IB_QPS_RTR:	return MLX5_QP_STATE_RTR;
+	case IB_QPS_RTS:	return MLX5_QP_STATE_RTS;
+	case IB_QPS_SQD:	return MLX5_QP_STATE_SQD;
+	case IB_QPS_SQE:	return MLX5_QP_STATE_SQER;
+	case IB_QPS_ERR:	return MLX5_QP_STATE_ERR;
+	default:		return -1;
+	}
+}
+
+static int to_mlx5_st(enum ib_qp_type type)
+{
+	switch (type) {
+	case IB_QPT_RC:			return MLX5_QP_ST_RC;
+	case IB_QPT_UC:			return MLX5_QP_ST_UC;
+	case IB_QPT_UD:			return MLX5_QP_ST_UD;
+	case MLX5_IB_QPT_REG_UMR:	return MLX5_QP_ST_REG_UMR;
+	case IB_QPT_XRC_INI:
+	case IB_QPT_XRC_TGT:		return MLX5_QP_ST_XRC;
+	case IB_QPT_SMI:		return MLX5_QP_ST_QP0;
+	case IB_QPT_GSI:		return MLX5_QP_ST_QP1;
+	case IB_QPT_RAW_IPV6:		return MLX5_QP_ST_RAW_IPV6;
+	case IB_QPT_RAW_ETHERTYPE:	return MLX5_QP_ST_RAW_ETHERTYPE;
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:		return -EINVAL;
+	}
+}
+
+static int uuarn_to_uar_index(struct mlx5_uuar_info *uuari, int uuarn)
+{
+	return uuari->uars[uuarn / MLX5_BF_REGS_PER_PAGE].index;
+}
+
+static int create_user_qp(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			  struct mlx5_ib_qp *qp, struct ib_udata *udata,
+			  struct mlx5_create_qp_mbox_in **in,
+			  struct mlx5_ib_create_qp_resp *resp, int *inlen)
+{
+	struct mlx5_ib_ucontext *context;
+	struct mlx5_ib_create_qp ucmd;
+	int page_shift = 0;
+	int uar_index;
+	int npages;
+	u32 offset = 0;
+	int uuarn;
+	int ncont = 0;
+	int err;
+
+	err = ib_copy_from_udata(&ucmd, udata, sizeof(ucmd));
+	if (err) {
+		mlx5_ib_dbg(dev, "copy failed\n");
+		return err;
+	}
+
+	context = to_mucontext(pd->uobject->context);
+	/*
+	 * TBD: should come from the verbs when we have the API
+	 */
+	uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_HIGH);
+	if (uuarn < 0) {
+		mlx5_ib_dbg(dev, "failed to allocate low latency UUAR\n");
+		mlx5_ib_dbg(dev, "reverting to medium latency\n");
+		uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_MEDIUM);
+		if (uuarn < 0) {
+			mlx5_ib_dbg(dev, "failed to allocate medium latency UUAR\n");
+			mlx5_ib_dbg(dev, "reverting to high latency\n");
+			uuarn = alloc_uuar(&context->uuari, MLX5_IB_LATENCY_CLASS_LOW);
+			if (uuarn < 0) {
+				mlx5_ib_warn(dev, "uuar allocation failed\n");
+				return uuarn;
+			}
+		}
+	}
+
+	uar_index = uuarn_to_uar_index(&context->uuari, uuarn);
+	mlx5_ib_dbg(dev, "uuarn 0x%x, uar_index 0x%x\n", uuarn, uar_index);
+
+	qp->rq.offset = 0;
+	qp->sq.wqe_shift = ilog2(MLX5_SEND_WQE_BB);
+	qp->sq.offset = qp->rq.wqe_cnt << qp->rq.wqe_shift;
+
+	err = set_user_buf_size(dev, qp, &ucmd);
+	if (err)
+		goto err_uuar;
+
+	if (ucmd.buf_addr && qp->buf_size) {
+		qp->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr,
+				       qp->buf_size, 0, 0);
+		if (IS_ERR(qp->umem)) {
+			mlx5_ib_dbg(dev, "umem_get failed\n");
+			err = PTR_ERR(qp->umem);
+			goto err_uuar;
+		}
+	} else {
+		qp->umem = NULL;
+	}
+
+	if (qp->umem) {
+		mlx5_ib_cont_pages(qp->umem, ucmd.buf_addr, &npages, &page_shift,
+				   &ncont, NULL);
+		err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift, &offset);
+		if (err) {
+			mlx5_ib_warn(dev, "bad offset\n");
+			goto err_umem;
+		}
+		mlx5_ib_dbg(dev, "addr 0x%llx, size %d, npages %d, page_shift %d, ncont %d, offset %d\n",
+			    ucmd.buf_addr, qp->buf_size, npages, page_shift, ncont, offset);
+	}
+
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * ncont;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_umem;
+	}
+	if (qp->umem)
+		mlx5_ib_populate_pas(dev, qp->umem, page_shift, (*in)->pas, 0);
+	(*in)->ctx.log_pg_sz_remote_qpn =
+		cpu_to_be32((page_shift - MLX5_ADAPTER_PAGE_SHIFT) << 24);
+	(*in)->ctx.params2 = cpu_to_be32(offset << 6);
+
+	(*in)->ctx.qp_counter_set_usr_page = cpu_to_be32(uar_index);
+	resp->uuar_index = uuarn;
+	qp->uuarn = uuarn;
+
+	err = mlx5_ib_db_map_user(context, ucmd.db_addr, &qp->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "map failed\n");
+		goto err_free;
+	}
+
+	err = ib_copy_to_udata(udata, resp, sizeof(*resp));
+	if (err) {
+		mlx5_ib_dbg(dev, "copy failed\n");
+		goto err_unmap;
+	}
+	qp->create_type = MLX5_QP_USER;
+
+	return 0;
+
+err_unmap:
+	mlx5_ib_db_unmap_user(context, &qp->db);
+
+err_free:
+	mlx5_vfree(*in);
+
+err_umem:
+	if (qp->umem)
+		ib_umem_release(qp->umem);
+
+err_uuar:
+	free_uuar(&context->uuari, uuarn);
+	return err;
+}
+
+static void destroy_qp_user(struct ib_pd *pd, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_ucontext *context;
+
+	context = to_mucontext(pd->uobject->context);
+	mlx5_ib_db_unmap_user(context, &qp->db);
+	if (qp->umem)
+		ib_umem_release(qp->umem);
+	free_uuar(&context->uuari, qp->uuarn);
+}
+
+static int create_kernel_qp(struct mlx5_ib_dev *dev,
+			    struct ib_qp_init_attr *init_attr,
+			    struct mlx5_ib_qp *qp,
+			    struct mlx5_create_qp_mbox_in **in, int *inlen)
+{
+	enum mlx5_ib_latency_class lc = MLX5_IB_LATENCY_CLASS_LOW;
+	struct mlx5_uuar_info *uuari;
+	int uar_index;
+	int uuarn;
+	int err;
+
+	uuari = &dev->mdev.priv.uuari;
+	if (init_attr->create_flags & ~(IB_QP_CREATE_SIGNATURE_EN | IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK))
+		return -EINVAL;
+
+	if (init_attr->qp_type == MLX5_IB_QPT_REG_UMR)
+		lc = MLX5_IB_LATENCY_CLASS_FAST_PATH;
+
+	uuarn = alloc_uuar(uuari, lc);
+	if (uuarn < 0) {
+		mlx5_ib_dbg(dev, "\n");
+		return -ENOMEM;
+	}
+
+	qp->bf = &uuari->bfs[uuarn];
+	uar_index = qp->bf->uar->index;
+
+	err = calc_sq_size(dev, init_attr, qp);
+	if (err < 0) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_uuar;
+	}
+
+	qp->rq.offset = 0;
+	qp->sq.offset = qp->rq.wqe_cnt << qp->rq.wqe_shift;
+	qp->buf_size = err + (qp->rq.wqe_cnt << qp->rq.wqe_shift);
+
+	err = mlx5_buf_alloc(&dev->mdev, qp->buf_size, PAGE_SIZE * 2, &qp->buf);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_uuar;
+	}
+
+	qp->sq.qend = mlx5_get_send_wqe(qp, qp->sq.wqe_cnt);
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * qp->buf.npages;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	(*in)->ctx.qp_counter_set_usr_page = cpu_to_be32(uar_index);
+	(*in)->ctx.log_pg_sz_remote_qpn =
+		cpu_to_be32((qp->buf.page_shift - MLX5_ADAPTER_PAGE_SHIFT) << 24);
+	/* Set "fast registration enabled" for all kernel QPs */
+	(*in)->ctx.params1 |= cpu_to_be32(1 << 11);
+	(*in)->ctx.sq_crq_size |= cpu_to_be16(1 << 4);
+
+	mlx5_fill_page_array(&qp->buf, (*in)->pas);
+
+	err = mlx5_db_alloc(&dev->mdev, &qp->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		goto err_free;
+	}
+
+	qp->db.db[0] = 0;
+	qp->db.db[1] = 0;
+
+	qp->sq.wrid = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wrid), GFP_KERNEL);
+	qp->sq.wr_data = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wr_data), GFP_KERNEL);
+	qp->rq.wrid = kmalloc(qp->rq.wqe_cnt * sizeof(*qp->rq.wrid), GFP_KERNEL);
+	qp->sq.w_list = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.w_list), GFP_KERNEL);
+	qp->sq.wqe_head = kmalloc(qp->sq.wqe_cnt * sizeof(*qp->sq.wqe_head), GFP_KERNEL);
+
+	if (!qp->sq.wrid || !qp->sq.wr_data || !qp->rq.wrid ||
+	    !qp->sq.w_list || !qp->sq.wqe_head) {
+		err = -ENOMEM;
+		goto err_wrid;
+	}
+	qp->create_type = MLX5_QP_KERNEL;
+
+	return 0;
+
+err_wrid:
+	mlx5_db_free(&dev->mdev, &qp->db);
+	kfree(qp->sq.wqe_head);
+	kfree(qp->sq.w_list);
+	kfree(qp->sq.wrid);
+	kfree(qp->sq.wr_data);
+	kfree(qp->rq.wrid);
+
+err_free:
+	mlx5_vfree(*in);
+
+err_buf:
+	mlx5_buf_free(&dev->mdev, &qp->buf);
+
+err_uuar:
+	free_uuar(&dev->mdev.priv.uuari, uuarn);
+	return err;
+}
+
+static void destroy_qp_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
+{
+	mlx5_db_free(&dev->mdev, &qp->db);
+	kfree(qp->sq.wqe_head);
+	kfree(qp->sq.w_list);
+	kfree(qp->sq.wrid);
+	kfree(qp->sq.wr_data);
+	kfree(qp->rq.wrid);
+	mlx5_buf_free(&dev->mdev, &qp->buf);
+	free_uuar(&dev->mdev.priv.uuari, qp->bf->uuarn);
+}
+
+static __be32 get_rx_type(struct mlx5_ib_qp *qp, struct ib_qp_init_attr *attr)
+{
+	if (attr->srq || (attr->qp_type == IB_QPT_XRC_TGT) ||
+	    (attr->qp_type == IB_QPT_XRC_INI))
+		return cpu_to_be32(MLX5_SRQ_RQ);
+	else if (!qp->has_rq)
+		return cpu_to_be32(MLX5_ZERO_LEN_RQ);
+	else
+		return cpu_to_be32(MLX5_NON_ZERO_RQ);
+}
+
+static int is_connected(enum ib_qp_type qp_type)
+{
+	if (qp_type == IB_QPT_RC || qp_type == IB_QPT_UC)
+		return 1;
+
+	return 0;
+}
+
+static int create_qp_common(struct mlx5_ib_dev *dev, struct ib_pd *pd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_resources *devr = &dev->devr;
+	struct mlx5_ib_create_qp_resp resp;
+	struct mlx5_create_qp_mbox_in *in;
+	struct mlx5_ib_create_qp ucmd;
+	int inlen = sizeof(*in);
+	int err;
+
+	mutex_init(&qp->mutex);
+	spin_lock_init(&qp->sq.lock);
+	spin_lock_init(&qp->rq.lock);
+
+	if (init_attr->create_flags & IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK) {
+		if (!(dev->mdev.caps.flags & MLX5_DEV_CAP_FLAG_BLOCK_MCAST)) {
+			mlx5_ib_dbg(dev, "block multicast loopback isn't supported\n");
+			return -EINVAL;
+		} else {
+			qp->flags |= MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK;
+		}
+	}
+
+	if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR)
+		qp->sq_signal_bits = MLX5_WQE_CTRL_CQ_UPDATE;
+
+	if (pd && pd->uobject) {
+		if (ib_copy_from_udata(&ucmd, udata, sizeof(ucmd))) {
+			mlx5_ib_dbg(dev, "copy failed\n");
+			return -EFAULT;
+		}
+
+		qp->wq_sig = !!(ucmd.flags & MLX5_QP_FLAG_SIGNATURE);
+		qp->scat_cqe = !!(ucmd.flags & MLX5_QP_FLAG_SCATTER_CQE);
+	} else {
+		qp->wq_sig = !!wq_signature;
+	}
+
+	qp->has_rq = qp_has_rq(init_attr);
+	err = set_rq_size(dev, &init_attr->cap, qp->has_rq,
+			  qp, (pd && pd->uobject) ? &ucmd : NULL);
+	if (err) {
+		mlx5_ib_dbg(dev, "err %d\n", err);
+		return err;
+	}
+
+	if (pd) {
+		if (pd->uobject) {
+			mlx5_ib_dbg(dev, "requested sq_wqe_count (%d)\n", ucmd.sq_wqe_count);
+			if (ucmd.rq_wqe_shift != qp->rq.wqe_shift ||
+			    ucmd.rq_wqe_count != qp->rq.wqe_cnt) {
+				mlx5_ib_dbg(dev, "invalid rq params\n");
+				return -EINVAL;
+			}
+			if (ucmd.sq_wqe_count > dev->mdev.caps.max_wqes) {
+				mlx5_ib_dbg(dev, "requested sq_wqe_count (%d) > max allowed (%d)\n",
+					    ucmd.sq_wqe_count, dev->mdev.caps.max_wqes);
+				return -EINVAL;
+			}
+			err = create_user_qp(dev, pd, qp, udata, &in, &resp, &inlen);
+			if (err)
+				mlx5_ib_dbg(dev, "err %d\n", err);
+		} else {
+			err = create_kernel_qp(dev, init_attr, qp, &in, &inlen);
+			if (err)
+				mlx5_ib_dbg(dev, "err %d\n", err);
+			else
+				qp->pa_lkey = to_mpd(pd)->pa_lkey;
+		}
+
+		if (err)
+			return err;
+	} else {
+		in = mlx5_vzalloc(sizeof(*in));
+		if (!in)
+			return -ENOMEM;
+
+		qp->create_type = MLX5_QP_EMPTY;
+	}
+
+	if (is_sqp(init_attr->qp_type))
+		qp->port = init_attr->port_num;
+
+	in->ctx.flags = cpu_to_be32(to_mlx5_st(init_attr->qp_type) << 16 |
+				    MLX5_QP_PM_MIGRATED << 11);
+
+	if (init_attr->qp_type != MLX5_IB_QPT_REG_UMR)
+		in->ctx.flags_pd = cpu_to_be32(to_mpd(pd ? pd : devr->p0)->pdn);
+	else
+		in->ctx.flags_pd = cpu_to_be32(MLX5_QP_LAT_SENSITIVE);
+
+	if (qp->wq_sig)
+		in->ctx.flags_pd |= cpu_to_be32(MLX5_QP_ENABLE_SIG);
+
+	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+		in->ctx.flags_pd |= cpu_to_be32(MLX5_QP_BLOCK_MCAST);
+
+	if (qp->scat_cqe && is_connected(init_attr->qp_type)) {
+		int rcqe_sz;
+		int scqe_sz;
+
+		rcqe_sz = mlx5_ib_get_cqe_size(dev, init_attr->recv_cq);
+		scqe_sz = mlx5_ib_get_cqe_size(dev, init_attr->send_cq);
+
+		if (rcqe_sz == 128)
+			in->ctx.cs_res = MLX5_RES_SCAT_DATA64_CQE;
+		else
+			in->ctx.cs_res = MLX5_RES_SCAT_DATA32_CQE;
+
+		if (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR) {
+			if (scqe_sz == 128)
+				in->ctx.cs_req = MLX5_REQ_SCAT_DATA64_CQE;
+			else
+				in->ctx.cs_req = MLX5_REQ_SCAT_DATA32_CQE;
+		}
+	}
+
+	if (qp->rq.wqe_cnt) {
+		in->ctx.rq_size_stride = (qp->rq.wqe_shift - 4);
+		in->ctx.rq_size_stride |= ilog2(qp->rq.wqe_cnt) << 3;
+	}
+
+	in->ctx.rq_type_srqn = get_rx_type(qp, init_attr);
+
+	if (qp->sq.wqe_cnt)
+		in->ctx.sq_crq_size |= cpu_to_be16(ilog2(qp->sq.wqe_cnt) << 11);
+	else
+		in->ctx.sq_crq_size |= cpu_to_be16(0x8000);
+
+	/* Set default resources */
+	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.cqn_send = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		in->ctx.xrcd = cpu_to_be32(to_mxrcd(init_attr->xrcd)->xrcdn);
+		break;
+	case IB_QPT_XRC_INI:
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(devr->c0)->mcq.cqn);
+		in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x1)->xrcdn);
+		in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		break;
+	default:
+		if (init_attr->srq) {
+			in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x0)->xrcdn);
+			in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(init_attr->srq)->msrq.srqn);
+		} else {
+			in->ctx.xrcd = cpu_to_be32(to_mxrcd(devr->x1)->xrcdn);
+			in->ctx.rq_type_srqn |= cpu_to_be32(to_msrq(devr->s0)->msrq.srqn);
+		}
+	}
+
+	if (init_attr->send_cq)
+		in->ctx.cqn_send = cpu_to_be32(to_mcq(init_attr->send_cq)->mcq.cqn);
+
+	if (init_attr->recv_cq)
+		in->ctx.cqn_recv = cpu_to_be32(to_mcq(init_attr->recv_cq)->mcq.cqn);
+
+	in->ctx.db_rec_addr = cpu_to_be64(qp->db.dma);
+
+	err = mlx5_core_create_qp(&dev->mdev, &qp->mqp, in, inlen);
+	if (err) {
+		mlx5_ib_dbg(dev, "create qp failed\n");
+		goto err_create;
+	}
+
+	mlx5_vfree(in);
+	/* Hardware wants QPN written in big-endian order (after
+	 * shifting) for send doorbell.  Precompute this value to save
+	 * a little bit when posting sends.
+	 */
+	qp->doorbell_qpn = swab32(qp->mqp.qpn << 8);
+
+	qp->mqp.event = mlx5_ib_qp_event;
+
+	return 0;
+
+err_create:
+	if (qp->create_type == MLX5_QP_USER)
+		destroy_qp_user(pd, qp);
+	else if (qp->create_type == MLX5_QP_KERNEL)
+		destroy_qp_kernel(dev, qp);
+
+	mlx5_vfree(in);
+	return err;
+}
+
+static void mlx5_ib_lock_cqs(struct mlx5_ib_cq *send_cq, struct mlx5_ib_cq *recv_cq)
+	__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
+{
+	if (send_cq) {
+		if (recv_cq) {
+			if (send_cq->mcq.cqn < recv_cq->mcq.cqn)  {
+				spin_lock_irq(&send_cq->lock);
+				spin_lock_nested(&recv_cq->lock,
+						 SINGLE_DEPTH_NESTING);
+			} else if (send_cq->mcq.cqn == recv_cq->mcq.cqn) {
+				spin_lock_irq(&send_cq->lock);
+				__acquire(&recv_cq->lock);
+			} else {
+				spin_lock_irq(&recv_cq->lock);
+				spin_lock_nested(&send_cq->lock,
+						 SINGLE_DEPTH_NESTING);
+			}
+		} else {
+			spin_lock_irq(&send_cq->lock);
+		}
+	} else if (recv_cq) {
+		spin_lock_irq(&recv_cq->lock);
+	}
+}
+
+static void mlx5_ib_unlock_cqs(struct mlx5_ib_cq *send_cq, struct mlx5_ib_cq *recv_cq)
+	__releases(&send_cq->lock) __releases(&recv_cq->lock)
+{
+	if (send_cq) {
+		if (recv_cq) {
+			if (send_cq->mcq.cqn < recv_cq->mcq.cqn)  {
+				spin_unlock(&recv_cq->lock);
+				spin_unlock_irq(&send_cq->lock);
+			} else if (send_cq->mcq.cqn == recv_cq->mcq.cqn) {
+				__release(&recv_cq->lock);
+				spin_unlock_irq(&send_cq->lock);
+			} else {
+				spin_unlock(&send_cq->lock);
+				spin_unlock_irq(&recv_cq->lock);
+			}
+		} else {
+			spin_unlock_irq(&send_cq->lock);
+		}
+	} else if (recv_cq) {
+		spin_unlock_irq(&recv_cq->lock);
+	}
+}
+
+static struct mlx5_ib_pd *get_pd(struct mlx5_ib_qp *qp)
+{
+	return to_mpd(qp->ibqp.pd);
+}
+
+static void get_cqs(struct mlx5_ib_qp *qp,
+		    struct mlx5_ib_cq **send_cq, struct mlx5_ib_cq **recv_cq)
+{
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_XRC_TGT:
+		*send_cq = NULL;
+		*recv_cq = NULL;
+		break;
+	case MLX5_IB_QPT_REG_UMR:
+	case IB_QPT_XRC_INI:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = NULL;
+		break;
+
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_RAW_IPV6:
+	case IB_QPT_RAW_ETHERTYPE:
+		*send_cq = to_mcq(qp->ibqp.send_cq);
+		*recv_cq = to_mcq(qp->ibqp.recv_cq);
+		break;
+
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:
+		*send_cq = NULL;
+		*recv_cq = NULL;
+		break;
+	}
+}
+
+static void destroy_qp_common(struct mlx5_ib_dev *dev, struct mlx5_ib_qp *qp)
+{
+	struct mlx5_ib_cq *send_cq, *recv_cq;
+	struct mlx5_modify_qp_mbox_in *in;
+	int err;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return;
+	if (qp->state != IB_QPS_RESET)
+		if (mlx5_core_qp_modify(&dev->mdev, to_mlx5_state(qp->state),
+					MLX5_QP_STATE_RST, in, sizeof(*in), &qp->mqp))
+			mlx5_ib_warn(dev, "mlx5_ib: modify QP %06x to RESET failed\n",
+				     qp->mqp.qpn);
+
+	get_cqs(qp, &send_cq, &recv_cq);
+
+	if (qp->create_type == MLX5_QP_KERNEL) {
+		mlx5_ib_lock_cqs(send_cq, recv_cq);
+		__mlx5_ib_cq_clean(recv_cq, qp->mqp.qpn,
+				   qp->ibqp.srq ? to_msrq(qp->ibqp.srq) : NULL);
+		if (send_cq != recv_cq)
+			__mlx5_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
+		mlx5_ib_unlock_cqs(send_cq, recv_cq);
+	}
+
+	err = mlx5_core_destroy_qp(&dev->mdev, &qp->mqp);
+	if (err)
+		mlx5_ib_warn(dev, "failed to destroy QP 0x%x\n", qp->mqp.qpn);
+	kfree(in);
+
+
+	if (qp->create_type == MLX5_QP_KERNEL)
+		destroy_qp_kernel(dev, qp);
+	else if (qp->create_type == MLX5_QP_USER)
+		destroy_qp_user(&get_pd(qp)->ibpd, qp);
+}
+
+static const char *ib_qp_type_str(enum ib_qp_type type)
+{
+	switch (type) {
+	case IB_QPT_SMI:
+		return "IB_QPT_SMI";
+	case IB_QPT_GSI:
+		return "IB_QPT_GSI";
+	case IB_QPT_RC:
+		return "IB_QPT_RC";
+	case IB_QPT_UC:
+		return "IB_QPT_UC";
+	case IB_QPT_UD:
+		return "IB_QPT_UD";
+	case IB_QPT_RAW_IPV6:
+		return "IB_QPT_RAW_IPV6";
+	case IB_QPT_RAW_ETHERTYPE:
+		return "IB_QPT_RAW_ETHERTYPE";
+	case IB_QPT_XRC_INI:
+		return "IB_QPT_XRC_INI";
+	case IB_QPT_XRC_TGT:
+		return "IB_QPT_XRC_TGT";
+	case IB_QPT_RAW_PACKET:
+		return "IB_QPT_RAW_PACKET";
+	case MLX5_IB_QPT_REG_UMR:
+		return "MLX5_IB_QPT_REG_UMR";
+	case IB_QPT_MAX:
+	default:
+		return "Invalid QP type";
+	}
+}
+
+struct ib_qp *mlx5_ib_create_qp(struct ib_pd *pd,
+				struct ib_qp_init_attr *init_attr,
+				struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev;
+	struct mlx5_ib_qp *qp;
+	u16 xrcdn = 0;
+	int err;
+
+	if (pd) {
+		dev = to_mdev(pd->device);
+	} else {
+		/* being cautious here */
+		if (init_attr->qp_type != IB_QPT_XRC_TGT &&
+		    init_attr->qp_type != MLX5_IB_QPT_REG_UMR) {
+			pr_warn("%s: no PD for transport %s\n", __func__,
+				ib_qp_type_str(init_attr->qp_type));
+			return ERR_PTR(-EINVAL);
+		}
+		dev = to_mdev(to_mxrcd(init_attr->xrcd)->ibxrcd.device);
+	}
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_XRC_TGT:
+	case IB_QPT_XRC_INI:
+		if (!(dev->mdev.caps.flags & MLX5_DEV_CAP_FLAG_XRC)) {
+			mlx5_ib_dbg(dev, "XRC not supported\n");
+			return ERR_PTR(-ENOSYS);
+		}
+		init_attr->recv_cq = NULL;
+		if (init_attr->qp_type == IB_QPT_XRC_TGT) {
+			xrcdn = to_mxrcd(init_attr->xrcd)->xrcdn;
+			init_attr->send_cq = NULL;
+		}
+
+		/* fall through */
+	case IB_QPT_RC:
+	case IB_QPT_UC:
+	case IB_QPT_UD:
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+	case MLX5_IB_QPT_REG_UMR:
+		qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+		if (!qp)
+			return ERR_PTR(-ENOMEM);
+
+		err = create_qp_common(dev, pd, init_attr, udata, qp);
+		if (err) {
+			mlx5_ib_dbg(dev, "create_qp_common failed\n");
+			kfree(qp);
+			return ERR_PTR(err);
+		}
+
+		if (is_qp0(init_attr->qp_type))
+			qp->ibqp.qp_num = 0;
+		else if (is_qp1(init_attr->qp_type))
+			qp->ibqp.qp_num = 1;
+		else
+			qp->ibqp.qp_num = qp->mqp.qpn;
+
+		mlx5_ib_dbg(dev, "ib qpnum 0x%x, mlx qpn 0x%x, rcqn 0x%x, scqn 0x%x\n",
+			    qp->ibqp.qp_num, qp->mqp.qpn, to_mcq(init_attr->recv_cq)->mcq.cqn,
+			    to_mcq(init_attr->send_cq)->mcq.cqn);
+
+		qp->xrcdn = xrcdn;
+
+		break;
+
+	case IB_QPT_RAW_IPV6:
+	case IB_QPT_RAW_ETHERTYPE:
+	case IB_QPT_RAW_PACKET:
+	case IB_QPT_MAX:
+	default:
+		mlx5_ib_dbg(dev, "unsupported qp type %d\n",
+			    init_attr->qp_type);
+		/* Don't support raw QPs */
+		return ERR_PTR(-EINVAL);
+	}
+
+	return &qp->ibqp;
+}
+
+int mlx5_ib_destroy_qp(struct ib_qp *qp)
+{
+	struct mlx5_ib_dev *dev = to_mdev(qp->device);
+	struct mlx5_ib_qp *mqp = to_mqp(qp);
+
+	destroy_qp_common(dev, mqp);
+
+	kfree(mqp);
+
+	return 0;
+}
+
+static __be32 to_mlx5_access_flags(struct mlx5_ib_qp *qp, const struct ib_qp_attr *attr,
+				   int attr_mask)
+{
+	u32 hw_access_flags = 0;
+	u8 dest_rd_atomic;
+	u32 access_flags;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		dest_rd_atomic = attr->max_dest_rd_atomic;
+	else
+		dest_rd_atomic = qp->resp_depth;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		access_flags = attr->qp_access_flags;
+	else
+		access_flags = qp->atomic_rd_en;
+
+	if (!dest_rd_atomic)
+		access_flags &= IB_ACCESS_REMOTE_WRITE;
+
+	if (access_flags & IB_ACCESS_REMOTE_READ)
+		hw_access_flags |= MLX5_QP_BIT_RRE;
+	if (access_flags & IB_ACCESS_REMOTE_ATOMIC)
+		hw_access_flags |= (MLX5_QP_BIT_RAE | MLX5_ATOMIC_MODE_CX);
+	if (access_flags & IB_ACCESS_REMOTE_WRITE)
+		hw_access_flags |= MLX5_QP_BIT_RWE;
+
+	return cpu_to_be32(hw_access_flags);
+}
+
+enum {
+	MLX5_PATH_FLAG_FL	= 1 << 0,
+	MLX5_PATH_FLAG_FREE_AR	= 1 << 1,
+	MLX5_PATH_FLAG_COUNTER	= 1 << 2,
+};
+
+static int ib_rate_to_mlx5(struct mlx5_ib_dev *dev, u8 rate)
+{
+	if (rate == IB_RATE_PORT_CURRENT) {
+		return 0;
+	} else if (rate < IB_RATE_2_5_GBPS || rate > IB_RATE_300_GBPS) {
+		return -EINVAL;
+	} else {
+		while (rate != IB_RATE_2_5_GBPS &&
+		       !(1 << (rate + MLX5_STAT_RATE_OFFSET) &
+			 dev->mdev.caps.stat_rate_support))
+			--rate;
+	}
+
+	return rate + MLX5_STAT_RATE_OFFSET;
+}
+
+static int mlx5_set_path(struct mlx5_ib_dev *dev, const struct ib_ah_attr *ah,
+			 struct mlx5_qp_path *path, u8 port, int attr_mask,
+			 u32 path_flags, const struct ib_qp_attr *attr)
+{
+	int err;
+
+	path->fl = (path_flags & MLX5_PATH_FLAG_FL) ? 0x80 : 0;
+	path->free_ar = (path_flags & MLX5_PATH_FLAG_FREE_AR) ? 0x80 : 0;
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		path->pkey_index = attr->pkey_index;
+
+	path->grh_mlid	= ah->src_path_bits & 0x7f;
+	path->rlid	= cpu_to_be16(ah->dlid);
+
+	if (ah->ah_flags & IB_AH_GRH) {
+		path->grh_mlid |= 1 << 7;
+		path->mgid_index = ah->grh.sgid_index;
+		path->hop_limit  = ah->grh.hop_limit;
+		path->tclass_flowlabel =
+			cpu_to_be32((ah->grh.traffic_class << 20) |
+				    (ah->grh.flow_label));
+		memcpy(path->rgid, ah->grh.dgid.raw, 16);
+	}
+
+	err = ib_rate_to_mlx5(dev, ah->static_rate);
+	if (err < 0)
+		return err;
+	path->static_rate = err;
+	path->port = port;
+
+	if (ah->ah_flags & IB_AH_GRH) {
+		if (ah->grh.sgid_index >= dev->mdev.caps.port[port - 1].gid_table_len) {
+			pr_err(KERN_ERR "sgid_index (%u) too large. max is %d\n",
+			       ah->grh.sgid_index, dev->mdev.caps.port[port - 1].gid_table_len);
+			return -EINVAL;
+		}
+
+		path->grh_mlid |= 1 << 7;
+		path->mgid_index = ah->grh.sgid_index;
+		path->hop_limit  = ah->grh.hop_limit;
+		path->tclass_flowlabel =
+			cpu_to_be32((ah->grh.traffic_class << 20) |
+				    (ah->grh.flow_label));
+		memcpy(path->rgid, ah->grh.dgid.raw, 16);
+	}
+
+	if (attr_mask & IB_QP_TIMEOUT)
+		path->ackto_lt = attr->timeout << 3;
+
+	path->sl = ah->sl & 0xf;
+
+	return 0;
+}
+
+static enum mlx5_qp_optpar opt_mask[MLX5_QP_NUM_STATE][MLX5_QP_NUM_STATE][MLX5_QP_ST_MAX] = {
+	[MLX5_QP_STATE_INIT] = {
+		[MLX5_QP_STATE_INIT] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_PKEY_INDEX	|
+					  MLX5_QP_OPTPAR_Q_KEY		|
+					  MLX5_QP_OPTPAR_PRI_PORT,
+		},
+		[MLX5_QP_STATE_RTR] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH  |
+					  MLX5_QP_OPTPAR_RRE            |
+					  MLX5_QP_OPTPAR_RAE            |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH  |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_PKEY_INDEX     |
+					  MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_MLX] = MLX5_QP_OPTPAR_PKEY_INDEX	|
+					   MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_XRC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH |
+					  MLX5_QP_OPTPAR_RRE            |
+					  MLX5_QP_OPTPAR_RAE            |
+					  MLX5_QP_OPTPAR_RWE            |
+					  MLX5_QP_OPTPAR_PKEY_INDEX,
+		},
+	},
+	[MLX5_QP_STATE_RTR] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH	|
+					  MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_RNR_TIMEOUT,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_ALT_ADDR_PATH	|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_Q_KEY,
+		},
+	},
+	[MLX5_QP_STATE_RTS] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_RC] = MLX5_QP_OPTPAR_RRE		|
+					  MLX5_QP_OPTPAR_RAE		|
+					  MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_RNR_TIMEOUT	|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_ALT_ADDR_PATH,
+			[MLX5_QP_ST_UC] = MLX5_QP_OPTPAR_RWE		|
+					  MLX5_QP_OPTPAR_PM_STATE	|
+					  MLX5_QP_OPTPAR_ALT_ADDR_PATH,
+			[MLX5_QP_ST_UD] = MLX5_QP_OPTPAR_Q_KEY		|
+					  MLX5_QP_OPTPAR_SRQN		|
+					  MLX5_QP_OPTPAR_CQN_RCV,
+		},
+	},
+	[MLX5_QP_STATE_SQER] = {
+		[MLX5_QP_STATE_RTS] = {
+			[MLX5_QP_ST_UD]	 = MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_MLX] = MLX5_QP_OPTPAR_Q_KEY,
+			[MLX5_QP_ST_UC]	 = MLX5_QP_OPTPAR_RWE,
+			[MLX5_QP_ST_RC]	 = MLX5_QP_OPTPAR_RNR_TIMEOUT	|
+					   MLX5_QP_OPTPAR_RWE		|
+					   MLX5_QP_OPTPAR_RAE		|
+					   MLX5_QP_OPTPAR_RRE,
+		},
+	},
+};
+
+static int ib_nr_to_mlx5_nr(int ib_mask)
+{
+	switch (ib_mask) {
+	case IB_QP_STATE:
+		return 0;
+	case IB_QP_CUR_STATE:
+		return 0;
+	case IB_QP_EN_SQD_ASYNC_NOTIFY:
+		return 0;
+	case IB_QP_ACCESS_FLAGS:
+		return MLX5_QP_OPTPAR_RWE | MLX5_QP_OPTPAR_RRE |
+			MLX5_QP_OPTPAR_RAE;
+	case IB_QP_PKEY_INDEX:
+		return MLX5_QP_OPTPAR_PKEY_INDEX;
+	case IB_QP_PORT:
+		return MLX5_QP_OPTPAR_PRI_PORT;
+	case IB_QP_QKEY:
+		return MLX5_QP_OPTPAR_Q_KEY;
+	case IB_QP_AV:
+		return MLX5_QP_OPTPAR_PRIMARY_ADDR_PATH |
+			MLX5_QP_OPTPAR_PRI_PORT;
+	case IB_QP_PATH_MTU:
+		return 0;
+	case IB_QP_TIMEOUT:
+		return MLX5_QP_OPTPAR_ACK_TIMEOUT;
+	case IB_QP_RETRY_CNT:
+		return MLX5_QP_OPTPAR_RETRY_COUNT;
+	case IB_QP_RNR_RETRY:
+		return MLX5_QP_OPTPAR_RNR_RETRY;
+	case IB_QP_RQ_PSN:
+		return 0;
+	case IB_QP_MAX_QP_RD_ATOMIC:
+		return MLX5_QP_OPTPAR_SRA_MAX;
+	case IB_QP_ALT_PATH:
+		return MLX5_QP_OPTPAR_ALT_ADDR_PATH;
+	case IB_QP_MIN_RNR_TIMER:
+		return MLX5_QP_OPTPAR_RNR_TIMEOUT;
+	case IB_QP_SQ_PSN:
+		return 0;
+	case IB_QP_MAX_DEST_RD_ATOMIC:
+		return MLX5_QP_OPTPAR_RRA_MAX | MLX5_QP_OPTPAR_RWE |
+			MLX5_QP_OPTPAR_RRE | MLX5_QP_OPTPAR_RAE;
+	case IB_QP_PATH_MIG_STATE:
+		return MLX5_QP_OPTPAR_PM_STATE;
+	case IB_QP_CAP:
+		return 0;
+	case IB_QP_DEST_QPN:
+		return 0;
+	}
+	return 0;
+}
+
+static int ib_mask_to_mlx5_opt(int ib_mask)
+{
+	int result = 0;
+	int i;
+
+	for (i = 0; i < 8 * sizeof(int); i++) {
+		if ((1 << i) & ib_mask)
+			result |= ib_nr_to_mlx5_nr(1 << i);
+	}
+
+	return result;
+}
+
+static int __mlx5_ib_modify_qp(struct ib_qp *ibqp,
+			       const struct ib_qp_attr *attr, int attr_mask,
+			       enum ib_qp_state cur_state, enum ib_qp_state new_state)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_ib_cq *send_cq, *recv_cq;
+	struct mlx5_qp_context *context;
+	struct mlx5_modify_qp_mbox_in *in;
+	struct mlx5_ib_pd *pd;
+	enum mlx5_qp_state mlx5_cur, mlx5_new;
+	enum mlx5_qp_optpar optpar;
+	int sqd_event;
+	int mlx5_st;
+	int err;
+
+	in = kzalloc(sizeof(*in), GFP_KERNEL);
+	if (!in)
+		return -ENOMEM;
+
+	context = &in->ctx;
+	err = to_mlx5_st(ibqp->qp_type);
+	if (err < 0)
+		goto out;
+
+	context->flags = cpu_to_be32(err << 16);
+
+	if (!(attr_mask & IB_QP_PATH_MIG_STATE)) {
+		context->flags |= cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
+	} else {
+		switch (attr->path_mig_state) {
+		case IB_MIG_MIGRATED:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_MIGRATED << 11);
+			break;
+		case IB_MIG_REARM:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_REARM << 11);
+			break;
+		case IB_MIG_ARMED:
+			context->flags |= cpu_to_be32(MLX5_QP_PM_ARMED << 11);
+			break;
+		}
+	}
+
+	if (ibqp->qp_type == IB_QPT_GSI || ibqp->qp_type == IB_QPT_SMI) {
+		context->mtu_msgmax = (IB_MTU_256 << 5) | 8;
+	} else if (ibqp->qp_type == IB_QPT_UD ||
+		   ibqp->qp_type == MLX5_IB_QPT_REG_UMR) {
+		context->mtu_msgmax = (IB_MTU_4096 << 5) | 12;
+	} else if (attr_mask & IB_QP_PATH_MTU) {
+		if (attr->path_mtu < IB_MTU_256 ||
+		    attr->path_mtu > IB_MTU_4096) {
+			mlx5_ib_warn(dev, "invalid mtu %d\n", attr->path_mtu);
+			err = -EINVAL;
+			goto out;
+		}
+		context->mtu_msgmax = (attr->path_mtu << 5) | dev->mdev.caps.log_max_msg;
+	}
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		context->log_pg_sz_remote_qpn = cpu_to_be32(attr->dest_qp_num);
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		context->pri_path.pkey_index = attr->pkey_index;
+
+	/* todo implement counter_index functionality */
+
+	if (is_sqp(ibqp->qp_type))
+		context->pri_path.port = qp->port;
+
+	if (attr_mask & IB_QP_PORT)
+		context->pri_path.port = attr->port_num;
+
+	if (attr_mask & IB_QP_AV) {
+		err = mlx5_set_path(dev, &attr->ah_attr, &context->pri_path,
+				    attr_mask & IB_QP_PORT ? attr->port_num : qp->port,
+				    attr_mask, 0, attr);
+		if (err)
+			goto out;
+	}
+
+	if (attr_mask & IB_QP_TIMEOUT)
+		context->pri_path.ackto_lt |= attr->timeout << 3;
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		err = mlx5_set_path(dev, &attr->alt_ah_attr, &context->alt_path,
+				    attr->alt_port_num, attr_mask, 0, attr);
+		if (err)
+			goto out;
+	}
+
+	pd = get_pd(qp);
+	get_cqs(qp, &send_cq, &recv_cq);
+
+	context->flags_pd = cpu_to_be32(pd ? pd->pdn : to_mpd(dev->devr.p0)->pdn);
+	context->cqn_send = send_cq ? cpu_to_be32(send_cq->mcq.cqn) : 0;
+	context->cqn_recv = recv_cq ? cpu_to_be32(recv_cq->mcq.cqn) : 0;
+	context->params1  = cpu_to_be32(MLX5_IB_ACK_REQ_FREQ << 28);
+
+	if (attr_mask & IB_QP_RNR_RETRY)
+		context->params1 |= cpu_to_be32(attr->rnr_retry << 13);
+
+	if (attr_mask & IB_QP_RETRY_CNT)
+		context->params1 |= cpu_to_be32(attr->retry_cnt << 16);
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+		if (attr->max_rd_atomic)
+			context->params1 |=
+				cpu_to_be32(fls(attr->max_rd_atomic - 1) << 21);
+	}
+
+	if (attr_mask & IB_QP_SQ_PSN)
+		context->next_send_psn = cpu_to_be32(attr->sq_psn);
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
+		if (attr->max_dest_rd_atomic)
+			context->params2 |=
+				cpu_to_be32(fls(attr->max_dest_rd_atomic - 1) << 21);
+	}
+
+	if (attr_mask & (IB_QP_ACCESS_FLAGS | IB_QP_MAX_DEST_RD_ATOMIC))
+		context->params2 |= to_mlx5_access_flags(qp, attr, attr_mask);
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		context->rnr_nextrecvpsn |= cpu_to_be32(attr->min_rnr_timer << 24);
+
+	if (attr_mask & IB_QP_RQ_PSN)
+		context->rnr_nextrecvpsn |= cpu_to_be32(attr->rq_psn);
+
+	if (attr_mask & IB_QP_QKEY)
+		context->qkey = cpu_to_be32(attr->qkey);
+
+	if (qp->rq.wqe_cnt && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->db_rec_addr = cpu_to_be64(qp->db.dma);
+
+	if (cur_state == IB_QPS_RTS && new_state == IB_QPS_SQD	&&
+	    attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY && attr->en_sqd_async_notify)
+		sqd_event = 1;
+	else
+		sqd_event = 0;
+
+	if (!ibqp->uobject && cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		context->sq_crq_size |= cpu_to_be16(1 << 4);
+
+
+	mlx5_cur = to_mlx5_state(cur_state);
+	mlx5_new = to_mlx5_state(new_state);
+	mlx5_st = to_mlx5_st(ibqp->qp_type);
+	if (mlx5_st < 0)
+		goto out;
+
+	optpar = ib_mask_to_mlx5_opt(attr_mask);
+	optpar &= opt_mask[mlx5_cur][mlx5_new][mlx5_st];
+	in->optparam = cpu_to_be32(optpar);
+	err = mlx5_core_qp_modify(&dev->mdev, to_mlx5_state(cur_state),
+				  to_mlx5_state(new_state), in, sqd_event,
+				  &qp->mqp);
+	if (err)
+		goto out;
+
+	qp->state = new_state;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		qp->atomic_rd_en = attr->qp_access_flags;
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		qp->resp_depth = attr->max_dest_rd_atomic;
+	if (attr_mask & IB_QP_PORT)
+		qp->port = attr->port_num;
+	if (attr_mask & IB_QP_ALT_PATH)
+		qp->alt_port = attr->alt_port_num;
+
+	/*
+	 * If we moved a kernel QP to RESET, clean up all old CQ
+	 * entries and reinitialize the QP.
+	 */
+	if (new_state == IB_QPS_RESET && !ibqp->uobject) {
+		mlx5_ib_cq_clean(recv_cq, qp->mqp.qpn,
+				 ibqp->srq ? to_msrq(ibqp->srq) : NULL);
+		if (send_cq != recv_cq)
+			mlx5_ib_cq_clean(send_cq, qp->mqp.qpn, NULL);
+
+		qp->rq.head = 0;
+		qp->rq.tail = 0;
+		qp->sq.head = 0;
+		qp->sq.tail = 0;
+		qp->sq.cur_post = 0;
+		qp->sq.last_poll = 0;
+		qp->db.db[MLX5_RCV_DBR] = 0;
+		qp->db.db[MLX5_SND_DBR] = 0;
+	}
+
+out:
+	kfree(in);
+	return err;
+}
+
+int mlx5_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	enum ib_qp_state cur_state, new_state;
+	int err = -EINVAL;
+	int port;
+
+	mutex_lock(&qp->mutex);
+
+	cur_state = attr_mask & IB_QP_CUR_STATE ? attr->cur_qp_state : qp->state;
+	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
+
+	if (ibqp->qp_type != MLX5_IB_QPT_REG_UMR &&
+	    !ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask,
+				IB_LINK_LAYER_UNSPECIFIED))
+		goto out;
+
+	if ((attr_mask & IB_QP_PORT) &&
+	    (attr->port_num == 0 || attr->port_num > dev->mdev.caps.num_ports))
+		goto out;
+
+	if (attr_mask & IB_QP_PKEY_INDEX) {
+		port = attr_mask & IB_QP_PORT ? attr->port_num : qp->port;
+		if (attr->pkey_index >= dev->mdev.caps.port[port - 1].pkey_table_len)
+			goto out;
+	}
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC &&
+	    attr->max_rd_atomic > dev->mdev.caps.max_ra_res_qp)
+		goto out;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC &&
+	    attr->max_dest_rd_atomic > dev->mdev.caps.max_ra_req_qp)
+		goto out;
+
+	if (cur_state == new_state && cur_state == IB_QPS_RESET) {
+		err = 0;
+		goto out;
+	}
+
+	err = __mlx5_ib_modify_qp(ibqp, attr, attr_mask, cur_state, new_state);
+
+out:
+	mutex_unlock(&qp->mutex);
+	return err;
+}
+
+static int mlx5_wq_overflow(struct mlx5_ib_wq *wq, int nreq, struct ib_cq *ib_cq)
+{
+	struct mlx5_ib_cq *cq;
+	unsigned cur;
+
+	cur = wq->head - wq->tail;
+	if (likely(cur + nreq < wq->max_post))
+		return 0;
+
+	cq = to_mcq(ib_cq);
+	spin_lock(&cq->lock);
+	cur = wq->head - wq->tail;
+	spin_unlock(&cq->lock);
+
+	return cur + nreq >= wq->max_post;
+}
+
+static __always_inline void set_raddr_seg(struct mlx5_wqe_raddr_seg *rseg,
+					  u64 remote_addr, u32 rkey)
+{
+	rseg->raddr    = cpu_to_be64(remote_addr);
+	rseg->rkey     = cpu_to_be32(rkey);
+	rseg->reserved = 0;
+}
+
+static void set_datagram_seg(struct mlx5_wqe_datagram_seg *dseg,
+			     struct ib_send_wr *wr)
+{
+	memcpy(&dseg->av, &to_mah(wr->wr.ud.ah)->av, sizeof(struct mlx5_av));
+	dseg->av.dqp_dct = cpu_to_be32(wr->wr.ud.remote_qpn | MLX5_EXTENDED_UD_AV);
+	dseg->av.key.qkey.qkey = cpu_to_be32(wr->wr.ud.remote_qkey);
+}
+
+static void set_data_ptr_seg(struct mlx5_wqe_data_seg *dseg, struct ib_sge *sg)
+{
+	dseg->byte_count = cpu_to_be32(sg->length);
+	dseg->lkey       = cpu_to_be32(sg->lkey);
+	dseg->addr       = cpu_to_be64(sg->addr);
+}
+
+static __be16 get_klm_octo(int npages)
+{
+	return cpu_to_be16(ALIGN(npages, 8) / 2);
+}
+
+static __be64 frwr_mkey_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_LEN		|
+		MLX5_MKEY_MASK_PAGE_SIZE	|
+		MLX5_MKEY_MASK_START_ADDR	|
+		MLX5_MKEY_MASK_EN_RINVAL	|
+		MLX5_MKEY_MASK_KEY		|
+		MLX5_MKEY_MASK_LR		|
+		MLX5_MKEY_MASK_LW		|
+		MLX5_MKEY_MASK_RR		|
+		MLX5_MKEY_MASK_RW		|
+		MLX5_MKEY_MASK_A		|
+		MLX5_MKEY_MASK_SMALL_FENCE	|
+		MLX5_MKEY_MASK_FREE;
+
+	return cpu_to_be64(result);
+}
+
+static __be64 sig_mkey_mask(void)
+{
+	u64 result;
+
+	result = MLX5_MKEY_MASK_LEN		|
+		MLX5_MKEY_MASK_PAGE_SIZE	|
+		MLX5_MKEY_MASK_START_ADDR	|
+		MLX5_MKEY_MASK_EN_SIGERR	|
+		MLX5_MKEY_MASK_EN_RINVAL	|
+		MLX5_MKEY_MASK_KEY		|
+		MLX5_MKEY_MASK_LR		|
+		MLX5_MKEY_MASK_LW		|
+		MLX5_MKEY_MASK_RR		|
+		MLX5_MKEY_MASK_RW		|
+		MLX5_MKEY_MASK_SMALL_FENCE	|
+		MLX5_MKEY_MASK_FREE		|
+		MLX5_MKEY_MASK_BSF_EN;
+
+	return cpu_to_be64(result);
+}
+
+static void set_frwr_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				 struct ib_send_wr *wr, int li)
+{
+	memset(umr, 0, sizeof(*umr));
+
+	if (li) {
+		umr->mkey_mask = cpu_to_be64(MLX5_MKEY_MASK_FREE);
+		umr->flags = 1 << 7;
+		return;
+	}
+
+	umr->flags = (1 << 5); /* fail if not free */
+	umr->klm_octowords = get_klm_octo(wr->wr.fast_reg.page_list_len);
+	umr->mkey_mask = frwr_mkey_mask();
+}
+
+static void set_reg_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				struct ib_send_wr *wr)
+{
+	struct umr_wr *umrwr = (struct umr_wr *)&wr->wr.fast_reg;
+	u64 mask;
+
+	memset(umr, 0, sizeof(*umr));
+
+	if (!(wr->send_flags & MLX5_IB_SEND_UMR_UNREG)) {
+		umr->flags = 1 << 5; /* fail if not free */
+		umr->klm_octowords = get_klm_octo(umrwr->npages);
+		mask =  MLX5_MKEY_MASK_LEN		|
+			MLX5_MKEY_MASK_PAGE_SIZE	|
+			MLX5_MKEY_MASK_START_ADDR	|
+			MLX5_MKEY_MASK_PD		|
+			MLX5_MKEY_MASK_LR		|
+			MLX5_MKEY_MASK_LW		|
+			MLX5_MKEY_MASK_KEY		|
+			MLX5_MKEY_MASK_RR		|
+			MLX5_MKEY_MASK_RW		|
+			MLX5_MKEY_MASK_A		|
+			MLX5_MKEY_MASK_FREE;
+		umr->mkey_mask = cpu_to_be64(mask);
+	} else {
+		umr->flags = 2 << 5; /* fail if free */
+		mask = MLX5_MKEY_MASK_FREE;
+		umr->mkey_mask = cpu_to_be64(mask);
+	}
+
+	if (!wr->num_sge)
+		umr->flags |= (1 << 7); /* inline */
+}
+
+static u8 get_umr_flags(int acc)
+{
+	return (acc & IB_ACCESS_REMOTE_ATOMIC ? MLX5_PERM_ATOMIC       : 0) |
+	       (acc & IB_ACCESS_REMOTE_WRITE  ? MLX5_PERM_REMOTE_WRITE : 0) |
+	       (acc & IB_ACCESS_REMOTE_READ   ? MLX5_PERM_REMOTE_READ  : 0) |
+	       (acc & IB_ACCESS_LOCAL_WRITE   ? MLX5_PERM_LOCAL_WRITE  : 0) |
+		MLX5_PERM_LOCAL_READ | MLX5_PERM_UMR_EN;
+}
+
+static void set_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr,
+			     int li, int *writ)
+{
+	memset(seg, 0, sizeof(*seg));
+	if (li) {
+		seg->status = 1 << 6;
+		return;
+	}
+
+	seg->flags = get_umr_flags(wr->wr.fast_reg.access_flags) |
+		     MLX5_ACCESS_MODE_MTT;
+	*writ = seg->flags & (MLX5_PERM_LOCAL_WRITE | IB_ACCESS_REMOTE_WRITE);
+	seg->qpn_mkey7_0 = cpu_to_be32((wr->wr.fast_reg.rkey & 0xff) | 0xffffff00);
+	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL);
+	seg->start_addr = cpu_to_be64(wr->wr.fast_reg.iova_start);
+	seg->len = cpu_to_be64(wr->wr.fast_reg.length);
+	seg->xlt_oct_size = cpu_to_be32((wr->wr.fast_reg.page_list_len + 1) / 2);
+	seg->log2_page_size = wr->wr.fast_reg.page_shift;
+}
+
+static void set_reg_mkey_segment(struct mlx5_mkey_seg *seg, struct ib_send_wr *wr)
+{
+	memset(seg, 0, sizeof(*seg));
+	if (wr->send_flags & MLX5_IB_SEND_UMR_UNREG) {
+		seg->status = 1 << 6;
+		return;
+	}
+
+	seg->flags = convert_access(wr->wr.fast_reg.access_flags);
+	seg->flags_pd = cpu_to_be32(to_mpd((struct ib_pd *)wr->wr.fast_reg.page_list)->pdn);
+	seg->start_addr = cpu_to_be64(wr->wr.fast_reg.iova_start);
+	seg->len = cpu_to_be64(wr->wr.fast_reg.length);
+	seg->log2_page_size = wr->wr.fast_reg.page_shift;
+	seg->qpn_mkey7_0 = cpu_to_be32(0xffffff00 |
+				       mlx5_mkey_variant(wr->wr.fast_reg.rkey));
+}
+
+static void set_frwr_pages(struct mlx5_wqe_data_seg *dseg,
+			   struct ib_send_wr *wr,
+			   struct mlx5_core_dev *mdev,
+			   struct mlx5_ib_pd *pd,
+			   int writ)
+{
+	struct mlx5_ib_fast_reg_page_list *mfrpl = to_mfrpl(wr->wr.fast_reg.page_list);
+	u64 *page_list = wr->wr.fast_reg.page_list->page_list;
+	u64 perm = MLX5_EN_RD | (writ ? MLX5_EN_WR : 0);
+	int i;
+
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++)
+		mfrpl->mapped_page_list[i] = cpu_to_be64(page_list[i] | perm);
+	dseg->addr = cpu_to_be64(mfrpl->map);
+	dseg->byte_count = cpu_to_be32(ALIGN(sizeof(u64) * wr->wr.fast_reg.page_list_len, 64));
+	dseg->lkey = cpu_to_be32(pd->pa_lkey);
+}
+
+static __be32 send_ieth(struct ib_send_wr *wr)
+{
+	switch (wr->opcode) {
+	case IB_WR_SEND_WITH_IMM:
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		return wr->ex.imm_data;
+
+	case IB_WR_SEND_WITH_INV:
+		return cpu_to_be32(wr->ex.invalidate_rkey);
+
+	default:
+		return 0;
+	}
+}
+
+static u8 calc_sig(void *wqe, int size)
+{
+	u8 *p = wqe;
+	u8 res = 0;
+	int i;
+
+	for (i = 0; i < size; i++)
+		res ^= p[i];
+
+	return ~res;
+}
+
+static u8 wq_sig(void *wqe)
+{
+	return calc_sig(wqe, (*((u8 *)wqe + 8) & 0x3f) << 4);
+}
+
+static int set_data_inl_seg(struct mlx5_ib_qp *qp, struct ib_send_wr *wr,
+			    void *wqe, int *sz)
+{
+	struct mlx5_wqe_inline_seg *seg;
+	void *qend = qp->sq.qend;
+	void *addr;
+	int inl = 0;
+	int copy;
+	int len;
+	int i;
+
+	seg = wqe;
+	wqe += sizeof(*seg);
+	for (i = 0; i < wr->num_sge; i++) {
+		addr = (void *)(unsigned long)(wr->sg_list[i].addr);
+		len  = wr->sg_list[i].length;
+		inl += len;
+
+		if (unlikely(inl > qp->max_inline_data))
+			return -ENOMEM;
+
+		if (unlikely(wqe + len > qend)) {
+			copy = qend - wqe;
+			memcpy(wqe, addr, copy);
+			addr += copy;
+			len -= copy;
+			wqe = mlx5_get_send_wqe(qp, 0);
+		}
+		memcpy(wqe, addr, len);
+		wqe += len;
+	}
+
+	seg->byte_count = cpu_to_be32(inl | MLX5_INLINE_SEG);
+
+	*sz = ALIGN(inl + sizeof(seg->byte_count), 16) / 16;
+
+	return 0;
+}
+
+static u16 prot_field_size(enum ib_signature_type type)
+{
+	switch (type) {
+	case IB_SIG_TYPE_T10_DIF:
+		return MLX5_DIF_SIZE;
+	default:
+		return 0;
+	}
+}
+
+static u8 bs_selector(int block_size)
+{
+	switch (block_size) {
+	case 512:	    return 0x1;
+	case 520:	    return 0x2;
+	case 4096:	    return 0x3;
+	case 4160:	    return 0x4;
+	case 1073741824:    return 0x5;
+	default:	    return 0;
+	}
+}
+
+static int format_selector(struct ib_sig_attrs *attr,
+			   struct ib_sig_domain *domain,
+			   int *selector)
+{
+
+#define FORMAT_DIF_NONE		0
+#define FORMAT_DIF_CRC_INC	8
+#define FORMAT_DIF_CRC_NO_INC	12
+#define FORMAT_DIF_CSUM_INC	13
+#define FORMAT_DIF_CSUM_NO_INC	14
+
+	switch (domain->sig.dif.type) {
+	case IB_T10DIF_NONE:
+		/* No DIF */
+		*selector = FORMAT_DIF_NONE;
+		break;
+	case IB_T10DIF_TYPE1: /* Fall through */
+	case IB_T10DIF_TYPE2:
+		switch (domain->sig.dif.bg_type) {
+		case IB_T10DIF_CRC:
+			*selector = FORMAT_DIF_CRC_INC;
+			break;
+		case IB_T10DIF_CSUM:
+			*selector = FORMAT_DIF_CSUM_INC;
+			break;
+		default:
+			return 1;
+		}
+		break;
+	case IB_T10DIF_TYPE3:
+		switch (domain->sig.dif.bg_type) {
+		case IB_T10DIF_CRC:
+			*selector = domain->sig.dif.type3_inc_reftag ?
+					   FORMAT_DIF_CRC_INC :
+					   FORMAT_DIF_CRC_NO_INC;
+			break;
+		case IB_T10DIF_CSUM:
+			*selector = domain->sig.dif.type3_inc_reftag ?
+					   FORMAT_DIF_CSUM_INC :
+					   FORMAT_DIF_CSUM_NO_INC;
+			break;
+		default:
+			return 1;
+		}
+		break;
+	default:
+		return 1;
+	}
+
+	return 0;
+}
+
+static int mlx5_set_bsf(struct ib_mr *sig_mr,
+			struct ib_sig_attrs *sig_attrs,
+			struct mlx5_bsf *bsf, u32 data_size)
+{
+	struct mlx5_core_sig_ctx *msig = to_mmr(sig_mr)->sig;
+	struct mlx5_bsf_basic *basic = &bsf->basic;
+	struct ib_sig_domain *mem = &sig_attrs->mem;
+	struct ib_sig_domain *wire = &sig_attrs->wire;
+	int ret, selector;
+
+	memset(bsf, 0, sizeof(*bsf));
+	switch (sig_attrs->mem.sig_type) {
+	case IB_SIG_TYPE_T10_DIF:
+		if (sig_attrs->wire.sig_type != IB_SIG_TYPE_T10_DIF)
+			return -EINVAL;
+
+		/* Input domain check byte mask */
+		basic->check_byte_mask = sig_attrs->check_mask;
+		if (mem->sig.dif.pi_interval == wire->sig.dif.pi_interval &&
+		    mem->sig.dif.type == wire->sig.dif.type) {
+			/* Same block structure */
+			basic->bsf_size_sbs = 1 << 4;
+			if (mem->sig.dif.bg_type == wire->sig.dif.bg_type)
+				basic->wire.copy_byte_mask |= 0xc0;
+			if (mem->sig.dif.app_tag == wire->sig.dif.app_tag)
+				basic->wire.copy_byte_mask |= 0x30;
+			if (mem->sig.dif.ref_tag == wire->sig.dif.ref_tag)
+				basic->wire.copy_byte_mask |= 0x0f;
+		} else
+			basic->wire.bs_selector = bs_selector(wire->sig.dif.pi_interval);
+
+		basic->mem.bs_selector = bs_selector(mem->sig.dif.pi_interval);
+		basic->raw_data_size = cpu_to_be32(data_size);
+
+		ret = format_selector(sig_attrs, mem, &selector);
+		if (ret)
+			return -EINVAL;
+		basic->m_bfs_psv = cpu_to_be32(selector << 24 |
+					       msig->psv_memory.psv_idx);
+
+		ret = format_selector(sig_attrs, wire, &selector);
+		if (ret)
+			return -EINVAL;
+		basic->w_bfs_psv = cpu_to_be32(selector << 24 |
+					       msig->psv_wire.psv_idx);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+static int set_sig_data_segment(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
+				void **seg, int *size)
+{
+	struct ib_sig_attrs *sig_attrs = wr->wr.sig_handover.sig_attrs;
+	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	struct mlx5_bsf *bsf;
+	u32 data_len = wr->sg_list->length;
+	u32 data_key = wr->sg_list->lkey;
+	u64 data_va = wr->sg_list->addr;
+	int ret;
+	int wqe_size;
+
+	if (!wr->wr.sig_handover.prot ||
+	    (data_key == wr->wr.sig_handover.prot->lkey &&
+	     data_va == wr->wr.sig_handover.prot->addr &&
+	     data_len == wr->wr.sig_handover.prot->length)) {
+		/**
+		 * Source domain doesn't contain signature information
+		 * or data and protection are interleaved in memory.
+		 * So need construct:
+		 *                  ------------------
+		 *                 |     data_klm     |
+		 *                  ------------------
+		 *                 |       BSF        |
+		 *                  ------------------
+		 **/
+		struct mlx5_klm *data_klm = *seg;
+
+		data_klm->bcount = cpu_to_be32(data_len);
+		data_klm->key = cpu_to_be32(data_key);
+		data_klm->va = cpu_to_be64(data_va);
+		wqe_size = ALIGN(sizeof(*data_klm), 64);
+	} else {
+		/**
+		 * Source domain contains signature information
+		 * So need construct a strided block format:
+		 *               ---------------------------
+		 *              |     stride_block_ctrl     |
+		 *               ---------------------------
+		 *              |          data_klm         |
+		 *               ---------------------------
+		 *              |          prot_klm         |
+		 *               ---------------------------
+		 *              |             BSF           |
+		 *               ---------------------------
+		 **/
+		struct mlx5_stride_block_ctrl_seg *sblock_ctrl;
+		struct mlx5_stride_block_entry *data_sentry;
+		struct mlx5_stride_block_entry *prot_sentry;
+		u32 prot_key = wr->wr.sig_handover.prot->lkey;
+		u64 prot_va = wr->wr.sig_handover.prot->addr;
+		u16 block_size = sig_attrs->mem.sig.dif.pi_interval;
+		int prot_size;
+
+		sblock_ctrl = *seg;
+		data_sentry = (void *)sblock_ctrl + sizeof(*sblock_ctrl);
+		prot_sentry = (void *)data_sentry + sizeof(*data_sentry);
+
+		prot_size = prot_field_size(sig_attrs->mem.sig_type);
+		if (!prot_size) {
+			pr_err("Bad block size given: %u\n", block_size);
+			return -EINVAL;
+		}
+		sblock_ctrl->bcount_per_cycle = cpu_to_be32(block_size +
+							    prot_size);
+		sblock_ctrl->op = cpu_to_be32(MLX5_STRIDE_BLOCK_OP);
+		sblock_ctrl->repeat_count = cpu_to_be32(data_len / block_size);
+		sblock_ctrl->num_entries = cpu_to_be16(2);
+
+		data_sentry->bcount = cpu_to_be16(block_size);
+		data_sentry->key = cpu_to_be32(data_key);
+		data_sentry->va = cpu_to_be64(data_va);
+		data_sentry->stride = cpu_to_be16(block_size);
+
+		prot_sentry->bcount = cpu_to_be16(prot_size);
+		prot_sentry->key = cpu_to_be32(prot_key);
+		prot_sentry->va = cpu_to_be64(prot_va);
+		prot_sentry->stride = cpu_to_be16(prot_size);
+
+		wqe_size = ALIGN(sizeof(*sblock_ctrl) + sizeof(*data_sentry) +
+				 sizeof(*prot_sentry), 64);
+	}
+
+	*seg += wqe_size;
+	*size += wqe_size / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	bsf = *seg;
+	ret = mlx5_set_bsf(sig_mr, sig_attrs, bsf, data_len);
+	if (ret)
+		return -EINVAL;
+
+	*seg += sizeof(*bsf);
+	*size += sizeof(*bsf) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	return 0;
+}
+
+static void set_sig_mkey_segment(struct mlx5_mkey_seg *seg,
+				 struct ib_send_wr *wr, u32 nelements,
+				 u32 length, u32 pdn)
+{
+	struct ib_mr *sig_mr = wr->wr.sig_handover.sig_mr;
+	u32 sig_key = sig_mr->rkey;
+	u8 sigerr = to_mmr(sig_mr)->sig->sigerr_count & 1;
+
+	memset(seg, 0, sizeof(*seg));
+
+	seg->flags = get_umr_flags(wr->wr.sig_handover.access_flags) |
+				   MLX5_ACCESS_MODE_KLM;
+	seg->qpn_mkey7_0 = cpu_to_be32((sig_key & 0xff) | 0xffffff00);
+	seg->flags_pd = cpu_to_be32(MLX5_MKEY_REMOTE_INVAL | sigerr << 26 |
+				    MLX5_MKEY_BSF_EN | pdn);
+	seg->len = cpu_to_be64(length);
+	seg->xlt_oct_size = cpu_to_be32(be16_to_cpu(get_klm_octo(nelements)));
+	seg->bsfs_octo_size = cpu_to_be32(MLX5_MKEY_BSF_OCTO_SIZE);
+}
+
+static void set_sig_umr_segment(struct mlx5_wqe_umr_ctrl_seg *umr,
+				struct ib_send_wr *wr, u32 nelements)
+{
+	memset(umr, 0, sizeof(*umr));
+
+	umr->flags = MLX5_FLAGS_INLINE | MLX5_FLAGS_CHECK_FREE;
+	umr->klm_octowords = get_klm_octo(nelements);
+	umr->bsf_octowords = cpu_to_be16(MLX5_MKEY_BSF_OCTO_SIZE);
+	umr->mkey_mask = sig_mkey_mask();
+}
+
+
+static int set_sig_umr_wr(struct ib_send_wr *wr, struct mlx5_ib_qp *qp,
+			  void **seg, int *size)
+{
+	struct mlx5_ib_mr *sig_mr = to_mmr(wr->wr.sig_handover.sig_mr);
+	u32 pdn = get_pd(qp)->pdn;
+	u32 klm_oct_size;
+	int region_len, ret;
+
+	if (unlikely(wr->num_sge != 1) ||
+	    unlikely(wr->wr.sig_handover.access_flags &
+		     IB_ACCESS_REMOTE_ATOMIC) ||
+	    unlikely(!sig_mr->sig) || unlikely(!qp->signature_en) ||
+	    unlikely(!sig_mr->sig->sig_status_checked))
+		return -EINVAL;
+
+	/* length of the protected region, data + protection */
+	region_len = wr->sg_list->length;
+	if (wr->wr.sig_handover.prot &&
+	    (wr->wr.sig_handover.prot->lkey != wr->sg_list->lkey  ||
+	     wr->wr.sig_handover.prot->addr != wr->sg_list->addr  ||
+	     wr->wr.sig_handover.prot->length != wr->sg_list->length))
+		region_len += wr->wr.sig_handover.prot->length;
+
+	/**
+	 * KLM octoword size - if protection was provided
+	 * then we use strided block format (3 octowords),
+	 * else we use single KLM (1 octoword)
+	 **/
+	klm_oct_size = wr->wr.sig_handover.prot ? 3 : 1;
+
+	set_sig_umr_segment(*seg, wr, klm_oct_size);
+	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	set_sig_mkey_segment(*seg, wr, klm_oct_size, region_len, pdn);
+	*seg += sizeof(struct mlx5_mkey_seg);
+	*size += sizeof(struct mlx5_mkey_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+
+	ret = set_sig_data_segment(wr, qp, seg, size);
+	if (ret)
+		return ret;
+
+	sig_mr->sig->sig_status_checked = false;
+	return 0;
+}
+
+static int set_psv_wr(struct ib_sig_domain *domain,
+		      u32 psv_idx, void **seg, int *size)
+{
+	struct mlx5_seg_set_psv *psv_seg = *seg;
+
+	memset(psv_seg, 0, sizeof(*psv_seg));
+	psv_seg->psv_num = cpu_to_be32(psv_idx);
+	switch (domain->sig_type) {
+	case IB_SIG_TYPE_T10_DIF:
+		psv_seg->transient_sig = cpu_to_be32(domain->sig.dif.bg << 16 |
+						     domain->sig.dif.app_tag);
+		psv_seg->ref_tag = cpu_to_be32(domain->sig.dif.ref_tag);
+
+		*seg += sizeof(*psv_seg);
+		*size += sizeof(*psv_seg) / 16;
+		break;
+
+	default:
+		pr_err("Bad signature type given.\n");
+		return 1;
+	}
+
+	return 0;
+}
+
+static int set_frwr_li_wr(void **seg, struct ib_send_wr *wr, int *size,
+			  struct mlx5_core_dev *mdev, struct mlx5_ib_pd *pd, struct mlx5_ib_qp *qp)
+{
+	int writ = 0;
+	int li;
+
+	li = wr->opcode == IB_WR_LOCAL_INV ? 1 : 0;
+	if (unlikely(wr->send_flags & IB_SEND_INLINE))
+		return -EINVAL;
+
+	set_frwr_umr_segment(*seg, wr, li);
+	*seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+	*size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+	set_mkey_segment(*seg, wr, li, &writ);
+	*seg += sizeof(struct mlx5_mkey_seg);
+	*size += sizeof(struct mlx5_mkey_seg) / 16;
+	if (unlikely((*seg == qp->sq.qend)))
+		*seg = mlx5_get_send_wqe(qp, 0);
+	if (!li) {
+		if (unlikely(wr->wr.fast_reg.page_list_len >
+			     wr->wr.fast_reg.page_list->max_page_list_len))
+			return	-ENOMEM;
+
+		set_frwr_pages(*seg, wr, mdev, pd, writ);
+		*seg += sizeof(struct mlx5_wqe_data_seg);
+		*size += (sizeof(struct mlx5_wqe_data_seg) / 16);
+	}
+	return 0;
+}
+
+static void dump_wqe(struct mlx5_ib_qp *qp, int idx, int size_16)
+{
+	__be32 *p = NULL;
+	int tidx = idx;
+	int i, j;
+
+	pr_debug("dump wqe at %p\n", mlx5_get_send_wqe(qp, tidx));
+	for (i = 0, j = 0; i < size_16 * 4; i += 4, j += 4) {
+		if ((i & 0xf) == 0) {
+			void *buf = mlx5_get_send_wqe(qp, tidx);
+			tidx = (tidx + 1) & (qp->sq.wqe_cnt - 1);
+			p = buf;
+			j = 0;
+		}
+		pr_debug("%08x %08x %08x %08x\n", be32_to_cpu(p[j]),
+			 be32_to_cpu(p[j + 1]), be32_to_cpu(p[j + 2]),
+			 be32_to_cpu(p[j + 3]));
+	}
+}
+
+static void mlx5_bf_copy(u64 __iomem *dst, u64 *src,
+			 unsigned bytecnt, struct mlx5_ib_qp *qp)
+{
+	while (bytecnt > 0) {
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		__iowrite64_copy(dst++, src++, 8);
+		bytecnt -= 64;
+		if (unlikely(src == qp->sq.qend))
+			src = mlx5_get_send_wqe(qp, 0);
+	}
+}
+
+static u8 get_fence(u8 fence, struct ib_send_wr *wr)
+{
+	if (unlikely(wr->opcode == IB_WR_LOCAL_INV &&
+		     wr->send_flags & IB_SEND_FENCE))
+		return MLX5_FENCE_MODE_STRONG_ORDERING;
+
+	if (unlikely(fence)) {
+		if (wr->send_flags & IB_SEND_FENCE)
+			return MLX5_FENCE_MODE_SMALL_AND_FENCE;
+		else
+			return fence;
+
+	} else {
+		return 0;
+	}
+}
+
+static int begin_wqe(struct mlx5_ib_qp *qp, void **seg,
+		     struct mlx5_wqe_ctrl_seg **ctrl,
+		     struct ib_send_wr *wr, int *idx,
+		     int *size, int nreq)
+{
+	int err = 0;
+
+	if (unlikely(mlx5_wq_overflow(&qp->sq, nreq, qp->ibqp.send_cq))) {
+		err = -ENOMEM;
+		return err;
+	}
+
+	*idx = qp->sq.cur_post & (qp->sq.wqe_cnt - 1);
+	*seg = mlx5_get_send_wqe(qp, *idx);
+	*ctrl = *seg;
+	*(uint32_t *)(*seg + 8) = 0;
+	(*ctrl)->imm = send_ieth(wr);
+	(*ctrl)->fm_ce_se = qp->sq_signal_bits |
+		(wr->send_flags & IB_SEND_SIGNALED ?
+		 MLX5_WQE_CTRL_CQ_UPDATE : 0) |
+		(wr->send_flags & IB_SEND_SOLICITED ?
+		 MLX5_WQE_CTRL_SOLICITED : 0);
+
+	*seg += sizeof(**ctrl);
+	*size = sizeof(**ctrl) / 16;
+
+	return err;
+}
+
+static void finish_wqe(struct mlx5_ib_qp *qp,
+		       struct mlx5_wqe_ctrl_seg *ctrl,
+		       u8 size, unsigned idx, u64 wr_id,
+		       int nreq, u8 fence, u8 next_fence,
+		       u32 mlx5_opcode)
+{
+	u8 opmod = 0;
+
+	ctrl->opmod_idx_opcode = cpu_to_be32(((u32)(qp->sq.cur_post) << 8) |
+					     mlx5_opcode | ((u32)opmod << 24));
+	ctrl->qpn_ds = cpu_to_be32(size | (qp->mqp.qpn << 8));
+	ctrl->fm_ce_se |= fence;
+	qp->fm_cache = next_fence;
+	if (unlikely(qp->wq_sig))
+		ctrl->signature = wq_sig(ctrl);
+
+	qp->sq.wrid[idx] = wr_id;
+	qp->sq.w_list[idx].opcode = mlx5_opcode;
+	qp->sq.wqe_head[idx] = qp->sq.head + nreq;
+	qp->sq.cur_post += DIV_ROUND_UP(size * 16, MLX5_SEND_WQE_BB);
+	qp->sq.w_list[idx].next = qp->sq.cur_post;
+}
+
+
+int mlx5_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		      struct ib_send_wr **bad_wr)
+{
+	struct mlx5_wqe_ctrl_seg *ctrl = NULL;  /* compiler warning */
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_core_dev *mdev = &dev->mdev;
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_ib_mr *mr;
+	struct mlx5_wqe_data_seg *dpseg;
+	struct mlx5_wqe_xrc_seg *xrc;
+	struct mlx5_bf *bf = qp->bf;
+	int uninitialized_var(size);
+	void *qend = qp->sq.qend;
+	unsigned long flags;
+	unsigned idx;
+	int err = 0;
+	int inl = 0;
+	int num_sge;
+	void *seg;
+	int nreq;
+	int i;
+	u8 next_fence = 0;
+	u8 fence;
+
+	spin_lock_irqsave(&qp->sq.lock, flags);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (unlikely(wr->opcode >= sizeof(mlx5_ib_opcode) / sizeof(mlx5_ib_opcode[0]))) {
+			mlx5_ib_warn(dev, "\n");
+			err = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		fence = qp->fm_cache;
+		num_sge = wr->num_sge;
+		if (unlikely(num_sge > qp->sq.max_gs)) {
+			mlx5_ib_warn(dev, "\n");
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		err = begin_wqe(qp, &seg, &ctrl, wr, &idx, &size, nreq);
+		if (err) {
+			mlx5_ib_warn(dev, "\n");
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		switch (ibqp->qp_type) {
+		case IB_QPT_XRC_INI:
+			xrc = seg;
+			xrc->xrc_srqn = htonl(wr->xrc_remote_srq_num);
+			seg += sizeof(*xrc);
+			size += sizeof(*xrc) / 16;
+			/* fall through */
+		case IB_QPT_RC:
+			switch (wr->opcode) {
+			case IB_WR_RDMA_READ:
+			case IB_WR_RDMA_WRITE:
+			case IB_WR_RDMA_WRITE_WITH_IMM:
+				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
+					      wr->wr.rdma.rkey);
+				seg  += sizeof(struct mlx5_wqe_raddr_seg);
+				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
+				break;
+
+			case IB_WR_ATOMIC_CMP_AND_SWP:
+			case IB_WR_ATOMIC_FETCH_AND_ADD:
+			case IB_WR_MASKED_ATOMIC_CMP_AND_SWP:
+				mlx5_ib_warn(dev, "Atomic operations are not supported yet\n");
+				err = -ENOSYS;
+				*bad_wr = wr;
+				goto out;
+
+			case IB_WR_LOCAL_INV:
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				qp->sq.wr_data[idx] = IB_WR_LOCAL_INV;
+				ctrl->imm = cpu_to_be32(wr->ex.invalidate_rkey);
+				err = set_frwr_li_wr(&seg, wr, &size, mdev, to_mpd(ibqp->pd), qp);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+				num_sge = 0;
+				break;
+
+			case IB_WR_FAST_REG_MR:
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				qp->sq.wr_data[idx] = IB_WR_FAST_REG_MR;
+				ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+				err = set_frwr_li_wr(&seg, wr, &size, mdev, to_mpd(ibqp->pd), qp);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+				num_sge = 0;
+				break;
+
+			case IB_WR_REG_SIG_MR:
+				qp->sq.wr_data[idx] = IB_WR_REG_SIG_MR;
+				mr = to_mmr(wr->wr.sig_handover.sig_mr);
+
+				ctrl->imm = cpu_to_be32(mr->ibmr.rkey);
+				err = set_sig_umr_wr(wr, qp, &seg, &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_UMR);
+				/*
+				 * SET_PSV WQEs are not signaled and solicited
+				 * on error
+				 */
+				wr->send_flags &= ~IB_SEND_SIGNALED;
+				wr->send_flags |= IB_SEND_SOLICITED;
+				err = begin_wqe(qp, &seg, &ctrl, wr,
+						&idx, &size, nreq);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					err = -ENOMEM;
+					*bad_wr = wr;
+					goto out;
+				}
+
+				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->mem,
+						 mr->sig->psv_memory.psv_idx, &seg,
+						 &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_SET_PSV);
+				err = begin_wqe(qp, &seg, &ctrl, wr,
+						&idx, &size, nreq);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					err = -ENOMEM;
+					*bad_wr = wr;
+					goto out;
+				}
+
+				next_fence = MLX5_FENCE_MODE_INITIATOR_SMALL;
+				err = set_psv_wr(&wr->wr.sig_handover.sig_attrs->wire,
+						 mr->sig->psv_wire.psv_idx, &seg,
+						 &size);
+				if (err) {
+					mlx5_ib_warn(dev, "\n");
+					*bad_wr = wr;
+					goto out;
+				}
+
+				finish_wqe(qp, ctrl, size, idx, wr->wr_id,
+					   nreq, get_fence(fence, wr),
+					   next_fence, MLX5_OPCODE_SET_PSV);
+				num_sge = 0;
+				goto skip_psv;
+
+			default:
+				break;
+			}
+			break;
+
+		case IB_QPT_UC:
+			switch (wr->opcode) {
+			case IB_WR_RDMA_WRITE:
+			case IB_WR_RDMA_WRITE_WITH_IMM:
+				set_raddr_seg(seg, wr->wr.rdma.remote_addr,
+					      wr->wr.rdma.rkey);
+				seg  += sizeof(struct mlx5_wqe_raddr_seg);
+				size += sizeof(struct mlx5_wqe_raddr_seg) / 16;
+				break;
+
+			default:
+				break;
+			}
+			break;
+
+		case IB_QPT_UD:
+		case IB_QPT_SMI:
+		case IB_QPT_GSI:
+			set_datagram_seg(seg, wr);
+			seg  += sizeof(struct mlx5_wqe_datagram_seg);
+			size += sizeof(struct mlx5_wqe_datagram_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			break;
+
+		case MLX5_IB_QPT_REG_UMR:
+			if (wr->opcode != MLX5_IB_WR_UMR) {
+				err = -EINVAL;
+				mlx5_ib_warn(dev, "bad opcode\n");
+				goto out;
+			}
+			qp->sq.wr_data[idx] = MLX5_IB_WR_UMR;
+			ctrl->imm = cpu_to_be32(wr->wr.fast_reg.rkey);
+			set_reg_umr_segment(seg, wr);
+			seg += sizeof(struct mlx5_wqe_umr_ctrl_seg);
+			size += sizeof(struct mlx5_wqe_umr_ctrl_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			set_reg_mkey_segment(seg, wr);
+			seg += sizeof(struct mlx5_mkey_seg);
+			size += sizeof(struct mlx5_mkey_seg) / 16;
+			if (unlikely((seg == qend)))
+				seg = mlx5_get_send_wqe(qp, 0);
+			break;
+
+		default:
+			break;
+		}
+
+		if (wr->send_flags & IB_SEND_INLINE && num_sge) {
+			int uninitialized_var(sz);
+
+			err = set_data_inl_seg(qp, wr, seg, &sz);
+			if (unlikely(err)) {
+				mlx5_ib_warn(dev, "\n");
+				*bad_wr = wr;
+				goto out;
+			}
+			inl = 1;
+			size += sz;
+		} else {
+			dpseg = seg;
+			for (i = 0; i < num_sge; i++) {
+				if (unlikely(dpseg == qend)) {
+					seg = mlx5_get_send_wqe(qp, 0);
+					dpseg = seg;
+				}
+				if (likely(wr->sg_list[i].length)) {
+					set_data_ptr_seg(dpseg, wr->sg_list + i);
+					size += sizeof(struct mlx5_wqe_data_seg) / 16;
+					dpseg++;
+				}
+			}
+		}
+
+		finish_wqe(qp, ctrl, size, idx, wr->wr_id, nreq,
+			   get_fence(fence, wr), next_fence,
+			   mlx5_ib_opcode[wr->opcode]);
+skip_psv:
+		if (0)
+			dump_wqe(qp, idx, size);
+	}
+
+out:
+	if (likely(nreq)) {
+		qp->sq.head += nreq;
+
+		/* Make sure that descriptors are written before
+		 * updating doorbell record and ringing the doorbell
+		 */
+		wmb();
+
+		qp->db.db[MLX5_SND_DBR] = cpu_to_be32(qp->sq.cur_post);
+
+		/* Make sure doorbell record is visible to the HCA before
+		 * we hit doorbell */
+		wmb();
+
+		if (bf->need_lock)
+			spin_lock(&bf->lock);
+
+		/* TBD enable WC */
+		if (0 && nreq == 1 && bf->uuarn && inl && size > 1 && size <= bf->buf_size / 16) {
+			mlx5_bf_copy(bf->reg + bf->offset, (u64 *)ctrl, ALIGN(size * 16, 64), qp);
+			/* wc_wmb(); */
+		} else {
+			mlx5_write64((__be32 *)ctrl, bf->regreg + bf->offset,
+				     MLX5_GET_DOORBELL_LOCK(&bf->lock32));
+			/* Make sure doorbells don't leak out of SQ spinlock
+			 * and reach the HCA out of order.
+			 */
+			mmiowb();
+		}
+		bf->offset ^= bf->buf_size;
+		if (bf->need_lock)
+			spin_unlock(&bf->lock);
+	}
+
+	spin_unlock_irqrestore(&qp->sq.lock, flags);
+
+	return err;
+}
+
+static void set_sig_seg(struct mlx5_rwqe_sig *sig, int size)
+{
+	sig->signature = calc_sig(sig, size);
+}
+
+int mlx5_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		      struct ib_recv_wr **bad_wr)
+{
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_wqe_data_seg *scat;
+	struct mlx5_rwqe_sig *sig;
+	unsigned long flags;
+	int err = 0;
+	int nreq;
+	int ind;
+	int i;
+
+	spin_lock_irqsave(&qp->rq.lock, flags);
+
+	ind = qp->rq.head & (qp->rq.wqe_cnt - 1);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (mlx5_wq_overflow(&qp->rq, nreq, qp->ibqp.recv_cq)) {
+			err = -ENOMEM;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		if (unlikely(wr->num_sge > qp->rq.max_gs)) {
+			err = -EINVAL;
+			*bad_wr = wr;
+			goto out;
+		}
+
+		scat = get_recv_wqe(qp, ind);
+		if (qp->wq_sig)
+			scat++;
+
+		for (i = 0; i < wr->num_sge; i++)
+			set_data_ptr_seg(scat + i, wr->sg_list + i);
+
+		if (i < qp->rq.max_gs) {
+			scat[i].byte_count = 0;
+			scat[i].lkey       = cpu_to_be32(MLX5_INVALID_LKEY);
+			scat[i].addr       = 0;
+		}
+
+		if (qp->wq_sig) {
+			sig = (struct mlx5_rwqe_sig *)scat;
+			set_sig_seg(sig, (qp->rq.max_gs + 1) << 2);
+		}
+
+		qp->rq.wrid[ind] = wr->wr_id;
+
+		ind = (ind + 1) & (qp->rq.wqe_cnt - 1);
+	}
+
+out:
+	if (likely(nreq)) {
+		qp->rq.head += nreq;
+
+		/* Make sure that descriptors are written before
+		 * doorbell record.
+		 */
+		wmb();
+
+		*qp->db.db = cpu_to_be32(qp->rq.head & 0xffff);
+	}
+
+	spin_unlock_irqrestore(&qp->rq.lock, flags);
+
+	return err;
+}
+
+static inline enum ib_qp_state to_ib_qp_state(enum mlx5_qp_state mlx5_state)
+{
+	switch (mlx5_state) {
+	case MLX5_QP_STATE_RST:      return IB_QPS_RESET;
+	case MLX5_QP_STATE_INIT:     return IB_QPS_INIT;
+	case MLX5_QP_STATE_RTR:      return IB_QPS_RTR;
+	case MLX5_QP_STATE_RTS:      return IB_QPS_RTS;
+	case MLX5_QP_STATE_SQ_DRAINING:
+	case MLX5_QP_STATE_SQD:      return IB_QPS_SQD;
+	case MLX5_QP_STATE_SQER:     return IB_QPS_SQE;
+	case MLX5_QP_STATE_ERR:      return IB_QPS_ERR;
+	default:		     return -1;
+	}
+}
+
+static inline enum ib_mig_state to_ib_mig_state(int mlx5_mig_state)
+{
+	switch (mlx5_mig_state) {
+	case MLX5_QP_PM_ARMED:		return IB_MIG_ARMED;
+	case MLX5_QP_PM_REARM:		return IB_MIG_REARM;
+	case MLX5_QP_PM_MIGRATED:	return IB_MIG_MIGRATED;
+	default: return -1;
+	}
+}
+
+static int to_ib_qp_access_flags(int mlx5_flags)
+{
+	int ib_flags = 0;
+
+	if (mlx5_flags & MLX5_QP_BIT_RRE)
+		ib_flags |= IB_ACCESS_REMOTE_READ;
+	if (mlx5_flags & MLX5_QP_BIT_RWE)
+		ib_flags |= IB_ACCESS_REMOTE_WRITE;
+	if (mlx5_flags & MLX5_QP_BIT_RAE)
+		ib_flags |= IB_ACCESS_REMOTE_ATOMIC;
+
+	return ib_flags;
+}
+
+static void to_ib_ah_attr(struct mlx5_ib_dev *ibdev, struct ib_ah_attr *ib_ah_attr,
+				struct mlx5_qp_path *path)
+{
+	struct mlx5_core_dev *dev = &ibdev->mdev;
+
+	memset(ib_ah_attr, 0, sizeof(*ib_ah_attr));
+	ib_ah_attr->port_num	  = path->port;
+
+	if (ib_ah_attr->port_num == 0 || ib_ah_attr->port_num > dev->caps.num_ports)
+		return;
+
+	ib_ah_attr->sl = path->sl & 0xf;
+
+	ib_ah_attr->dlid	  = be16_to_cpu(path->rlid);
+	ib_ah_attr->src_path_bits = path->grh_mlid & 0x7f;
+	ib_ah_attr->static_rate   = path->static_rate ? path->static_rate - 5 : 0;
+	ib_ah_attr->ah_flags      = (path->grh_mlid & (1 << 7)) ? IB_AH_GRH : 0;
+	if (ib_ah_attr->ah_flags) {
+		ib_ah_attr->grh.sgid_index = path->mgid_index;
+		ib_ah_attr->grh.hop_limit  = path->hop_limit;
+		ib_ah_attr->grh.traffic_class =
+			(be32_to_cpu(path->tclass_flowlabel) >> 20) & 0xff;
+		ib_ah_attr->grh.flow_label =
+			be32_to_cpu(path->tclass_flowlabel) & 0xfffff;
+		memcpy(ib_ah_attr->grh.dgid.raw,
+		       path->rgid, sizeof(ib_ah_attr->grh.dgid.raw));
+	}
+}
+
+int mlx5_ib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_mask,
+		     struct ib_qp_init_attr *qp_init_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibqp->device);
+	struct mlx5_ib_qp *qp = to_mqp(ibqp);
+	struct mlx5_query_qp_mbox_out *outb;
+	struct mlx5_qp_context *context;
+	int mlx5_state;
+	int err = 0;
+
+	mutex_lock(&qp->mutex);
+	outb = kzalloc(sizeof(*outb), GFP_KERNEL);
+	if (!outb) {
+		err = -ENOMEM;
+		goto out;
+	}
+	context = &outb->ctx;
+	err = mlx5_core_qp_query(&dev->mdev, &qp->mqp, outb, sizeof(*outb));
+	if (err)
+		goto out_free;
+
+	mlx5_state = be32_to_cpu(context->flags) >> 28;
+
+	qp->state		     = to_ib_qp_state(mlx5_state);
+	qp_attr->qp_state	     = qp->state;
+	qp_attr->path_mtu	     = context->mtu_msgmax >> 5;
+	qp_attr->path_mig_state	     =
+		to_ib_mig_state((be32_to_cpu(context->flags) >> 11) & 0x3);
+	qp_attr->qkey		     = be32_to_cpu(context->qkey);
+	qp_attr->rq_psn		     = be32_to_cpu(context->rnr_nextrecvpsn) & 0xffffff;
+	qp_attr->sq_psn		     = be32_to_cpu(context->next_send_psn) & 0xffffff;
+	qp_attr->dest_qp_num	     = be32_to_cpu(context->log_pg_sz_remote_qpn) & 0xffffff;
+	qp_attr->qp_access_flags     =
+		to_ib_qp_access_flags(be32_to_cpu(context->params2));
+
+	if (qp->ibqp.qp_type == IB_QPT_RC || qp->ibqp.qp_type == IB_QPT_UC) {
+		to_ib_ah_attr(dev, &qp_attr->ah_attr, &context->pri_path);
+		to_ib_ah_attr(dev, &qp_attr->alt_ah_attr, &context->alt_path);
+		qp_attr->alt_pkey_index = context->alt_path.pkey_index & 0x7f;
+		qp_attr->alt_port_num	= qp_attr->alt_ah_attr.port_num;
+	}
+
+	qp_attr->pkey_index = context->pri_path.pkey_index & 0x7f;
+	qp_attr->port_num = context->pri_path.port;
+
+	/* qp_attr->en_sqd_async_notify is only applicable in modify qp */
+	qp_attr->sq_draining = mlx5_state == MLX5_QP_STATE_SQ_DRAINING;
+
+	qp_attr->max_rd_atomic = 1 << ((be32_to_cpu(context->params1) >> 21) & 0x7);
+
+	qp_attr->max_dest_rd_atomic =
+		1 << ((be32_to_cpu(context->params2) >> 21) & 0x7);
+	qp_attr->min_rnr_timer	    =
+		(be32_to_cpu(context->rnr_nextrecvpsn) >> 24) & 0x1f;
+	qp_attr->timeout	    = context->pri_path.ackto_lt >> 3;
+	qp_attr->retry_cnt	    = (be32_to_cpu(context->params1) >> 16) & 0x7;
+	qp_attr->rnr_retry	    = (be32_to_cpu(context->params1) >> 13) & 0x7;
+	qp_attr->alt_timeout	    = context->alt_path.ackto_lt >> 3;
+	qp_attr->cur_qp_state	     = qp_attr->qp_state;
+	qp_attr->cap.max_recv_wr     = qp->rq.wqe_cnt;
+	qp_attr->cap.max_recv_sge    = qp->rq.max_gs;
+
+	if (!ibqp->uobject) {
+		qp_attr->cap.max_send_wr  = qp->sq.wqe_cnt;
+		qp_attr->cap.max_send_sge = qp->sq.max_gs;
+	} else {
+		qp_attr->cap.max_send_wr  = 0;
+		qp_attr->cap.max_send_sge = 0;
+	}
+
+	/* We don't support inline sends for kernel QPs (yet), and we
+	 * don't know what userspace's value should be.
+	 */
+	qp_attr->cap.max_inline_data = 0;
+
+	qp_init_attr->cap	     = qp_attr->cap;
+
+	qp_init_attr->create_flags = 0;
+	if (qp->flags & MLX5_IB_QP_BLOCK_MULTICAST_LOOPBACK)
+		qp_init_attr->create_flags |= IB_QP_CREATE_BLOCK_MULTICAST_LOOPBACK;
+
+	qp_init_attr->sq_sig_type = qp->sq_signal_bits & MLX5_WQE_CTRL_CQ_UPDATE ?
+		IB_SIGNAL_ALL_WR : IB_SIGNAL_REQ_WR;
+
+out_free:
+	kfree(outb);
+
+out:
+	mutex_unlock(&qp->mutex);
+	return err;
+}
+
+struct ib_xrcd *mlx5_ib_alloc_xrcd(struct ib_device *ibdev,
+					  struct ib_ucontext *context,
+					  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibdev);
+	struct mlx5_ib_xrcd *xrcd;
+	int err;
+
+	if (!(dev->mdev.caps.flags & MLX5_DEV_CAP_FLAG_XRC))
+		return ERR_PTR(-ENOSYS);
+
+	xrcd = kmalloc(sizeof(*xrcd), GFP_KERNEL);
+	if (!xrcd)
+		return ERR_PTR(-ENOMEM);
+
+	err = mlx5_core_xrcd_alloc(&dev->mdev, &xrcd->xrcdn);
+	if (err) {
+		kfree(xrcd);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	return &xrcd->ibxrcd;
+}
+
+int mlx5_ib_dealloc_xrcd(struct ib_xrcd *xrcd)
+{
+	struct mlx5_ib_dev *dev = to_mdev(xrcd->device);
+	u32 xrcdn = to_mxrcd(xrcd)->xrcdn;
+	int err;
+
+	err = mlx5_core_xrcd_dealloc(&dev->mdev, xrcdn);
+	if (err) {
+		mlx5_ib_warn(dev, "failed to dealloc xrcdn 0x%x\n", xrcdn);
+		return err;
+	}
+
+	kfree(xrcd);
+
+	return 0;
+}
diff --git a/drivers/infiniband/hw/mlx5/srq.c b/drivers/infiniband/hw/mlx5/srq.c
new file mode 100644
index 00000000000..384af6dec5e
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/srq.c
@@ -0,0 +1,485 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/mlx5/qp.h>
+#include <linux/mlx5/srq.h>
+#include <linux/slab.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "mlx5_ib.h"
+#include "user.h"
+
+/* not supported currently */
+static int srq_signature;
+
+static void *get_wqe(struct mlx5_ib_srq *srq, int n)
+{
+	return mlx5_buf_offset(&srq->buf, n << srq->msrq.wqe_shift);
+}
+
+static void mlx5_ib_srq_event(struct mlx5_core_srq *srq, enum mlx5_event type)
+{
+	struct ib_event event;
+	struct ib_srq *ibsrq = &to_mibsrq(srq)->ibsrq;
+
+	if (ibsrq->event_handler) {
+		event.device      = ibsrq->device;
+		event.element.srq = ibsrq;
+		switch (type) {
+		case MLX5_EVENT_TYPE_SRQ_RQ_LIMIT:
+			event.event = IB_EVENT_SRQ_LIMIT_REACHED;
+			break;
+		case MLX5_EVENT_TYPE_SRQ_CATAS_ERROR:
+			event.event = IB_EVENT_SRQ_ERR;
+			break;
+		default:
+			pr_warn("mlx5_ib: Unexpected event type %d on SRQ %06x\n",
+				type, srq->srqn);
+			return;
+		}
+
+		ibsrq->event_handler(&event, ibsrq->srq_context);
+	}
+}
+
+static int create_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq,
+			   struct mlx5_create_srq_mbox_in **in,
+			   struct ib_udata *udata, int buf_size, int *inlen)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_create_srq ucmd;
+	size_t ucmdlen;
+	int err;
+	int npages;
+	int page_shift;
+	int ncont;
+	u32 offset;
+
+	ucmdlen =
+		(udata->inlen - sizeof(struct ib_uverbs_cmd_hdr) <
+		 sizeof(ucmd)) ? (sizeof(ucmd) -
+				  sizeof(ucmd.reserved)) : sizeof(ucmd);
+
+	if (ib_copy_from_udata(&ucmd, udata, ucmdlen)) {
+		mlx5_ib_dbg(dev, "failed copy udata\n");
+		return -EFAULT;
+	}
+
+	if (ucmdlen == sizeof(ucmd) &&
+	    ucmd.reserved != 0)
+		return -EINVAL;
+
+	srq->wq_sig = !!(ucmd.flags & MLX5_SRQ_FLAG_SIGNATURE);
+
+	srq->umem = ib_umem_get(pd->uobject->context, ucmd.buf_addr, buf_size,
+				0, 0);
+	if (IS_ERR(srq->umem)) {
+		mlx5_ib_dbg(dev, "failed umem get, size %d\n", buf_size);
+		err = PTR_ERR(srq->umem);
+		return err;
+	}
+
+	mlx5_ib_cont_pages(srq->umem, ucmd.buf_addr, &npages,
+			   &page_shift, &ncont, NULL);
+	err = mlx5_ib_get_buf_offset(ucmd.buf_addr, page_shift,
+				     &offset);
+	if (err) {
+		mlx5_ib_warn(dev, "bad offset\n");
+		goto err_umem;
+	}
+
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * ncont;
+	*in = mlx5_vzalloc(*inlen);
+	if (!(*in)) {
+		err = -ENOMEM;
+		goto err_umem;
+	}
+
+	mlx5_ib_populate_pas(dev, srq->umem, page_shift, (*in)->pas, 0);
+
+	err = mlx5_ib_db_map_user(to_mucontext(pd->uobject->context),
+				  ucmd.db_addr, &srq->db);
+	if (err) {
+		mlx5_ib_dbg(dev, "map doorbell failed\n");
+		goto err_in;
+	}
+
+	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+	(*in)->ctx.pgoff_cqn = cpu_to_be32(offset << 26);
+
+	return 0;
+
+err_in:
+	mlx5_vfree(*in);
+
+err_umem:
+	ib_umem_release(srq->umem);
+
+	return err;
+}
+
+static int create_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq,
+			     struct mlx5_create_srq_mbox_in **in, int buf_size,
+			     int *inlen)
+{
+	int err;
+	int i;
+	struct mlx5_wqe_srq_next_seg *next;
+	int page_shift;
+	int npages;
+
+	err = mlx5_db_alloc(&dev->mdev, &srq->db);
+	if (err) {
+		mlx5_ib_warn(dev, "alloc dbell rec failed\n");
+		return err;
+	}
+
+	*srq->db.db = 0;
+
+	if (mlx5_buf_alloc(&dev->mdev, buf_size, PAGE_SIZE * 2, &srq->buf)) {
+		mlx5_ib_dbg(dev, "buf alloc failed\n");
+		err = -ENOMEM;
+		goto err_db;
+	}
+	page_shift = srq->buf.page_shift;
+
+	srq->head    = 0;
+	srq->tail    = srq->msrq.max - 1;
+	srq->wqe_ctr = 0;
+
+	for (i = 0; i < srq->msrq.max; i++) {
+		next = get_wqe(srq, i);
+		next->next_wqe_index =
+			cpu_to_be16((i + 1) & (srq->msrq.max - 1));
+	}
+
+	npages = DIV_ROUND_UP(srq->buf.npages, 1 << (page_shift - PAGE_SHIFT));
+	mlx5_ib_dbg(dev, "buf_size %d, page_shift %d, npages %d, calc npages %d\n",
+		    buf_size, page_shift, srq->buf.npages, npages);
+	*inlen = sizeof(**in) + sizeof(*(*in)->pas) * npages;
+	*in = mlx5_vzalloc(*inlen);
+	if (!*in) {
+		err = -ENOMEM;
+		goto err_buf;
+	}
+	mlx5_fill_page_array(&srq->buf, (*in)->pas);
+
+	srq->wrid = kmalloc(srq->msrq.max * sizeof(u64), GFP_KERNEL);
+	if (!srq->wrid) {
+		mlx5_ib_dbg(dev, "kmalloc failed %lu\n",
+			    (unsigned long)(srq->msrq.max * sizeof(u64)));
+		err = -ENOMEM;
+		goto err_in;
+	}
+	srq->wq_sig = !!srq_signature;
+
+	(*in)->ctx.log_pg_sz = page_shift - MLX5_ADAPTER_PAGE_SHIFT;
+
+	return 0;
+
+err_in:
+	mlx5_vfree(*in);
+
+err_buf:
+	mlx5_buf_free(&dev->mdev, &srq->buf);
+
+err_db:
+	mlx5_db_free(&dev->mdev, &srq->db);
+	return err;
+}
+
+static void destroy_srq_user(struct ib_pd *pd, struct mlx5_ib_srq *srq)
+{
+	mlx5_ib_db_unmap_user(to_mucontext(pd->uobject->context), &srq->db);
+	ib_umem_release(srq->umem);
+}
+
+
+static void destroy_srq_kernel(struct mlx5_ib_dev *dev, struct mlx5_ib_srq *srq)
+{
+	kfree(srq->wrid);
+	mlx5_buf_free(&dev->mdev, &srq->buf);
+	mlx5_db_free(&dev->mdev, &srq->db);
+}
+
+struct ib_srq *mlx5_ib_create_srq(struct ib_pd *pd,
+				  struct ib_srq_init_attr *init_attr,
+				  struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(pd->device);
+	struct mlx5_ib_srq *srq;
+	int desc_size;
+	int buf_size;
+	int err;
+	struct mlx5_create_srq_mbox_in *uninitialized_var(in);
+	int uninitialized_var(inlen);
+	int is_xrc;
+	u32 flgs, xrcdn;
+
+	/* Sanity check SRQ size before proceeding */
+	if (init_attr->attr.max_wr >= dev->mdev.caps.max_srq_wqes) {
+		mlx5_ib_dbg(dev, "max_wr %d, cap %d\n",
+			    init_attr->attr.max_wr,
+			    dev->mdev.caps.max_srq_wqes);
+		return ERR_PTR(-EINVAL);
+	}
+
+	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&srq->mutex);
+	spin_lock_init(&srq->lock);
+	srq->msrq.max    = roundup_pow_of_two(init_attr->attr.max_wr + 1);
+	srq->msrq.max_gs = init_attr->attr.max_sge;
+
+	desc_size = sizeof(struct mlx5_wqe_srq_next_seg) +
+		    srq->msrq.max_gs * sizeof(struct mlx5_wqe_data_seg);
+	desc_size = roundup_pow_of_two(desc_size);
+	desc_size = max_t(int, 32, desc_size);
+	srq->msrq.max_avail_gather = (desc_size - sizeof(struct mlx5_wqe_srq_next_seg)) /
+		sizeof(struct mlx5_wqe_data_seg);
+	srq->msrq.wqe_shift = ilog2(desc_size);
+	buf_size = srq->msrq.max * desc_size;
+	mlx5_ib_dbg(dev, "desc_size 0x%x, req wr 0x%x, srq size 0x%x, max_gs 0x%x, max_avail_gather 0x%x\n",
+		    desc_size, init_attr->attr.max_wr, srq->msrq.max, srq->msrq.max_gs,
+		    srq->msrq.max_avail_gather);
+
+	if (pd->uobject)
+		err = create_srq_user(pd, srq, &in, udata, buf_size, &inlen);
+	else
+		err = create_srq_kernel(dev, srq, &in, buf_size, &inlen);
+
+	if (err) {
+		mlx5_ib_warn(dev, "create srq %s failed, err %d\n",
+			     pd->uobject ? "user" : "kernel", err);
+		goto err_srq;
+	}
+
+	is_xrc = (init_attr->srq_type == IB_SRQT_XRC);
+	in->ctx.state_log_sz = ilog2(srq->msrq.max);
+	flgs = ((srq->msrq.wqe_shift - 4) | (is_xrc << 5) | (srq->wq_sig << 7)) << 24;
+	xrcdn = 0;
+	if (is_xrc) {
+		xrcdn = to_mxrcd(init_attr->ext.xrc.xrcd)->xrcdn;
+		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(init_attr->ext.xrc.cq)->mcq.cqn);
+	} else if (init_attr->srq_type == IB_SRQT_BASIC) {
+		xrcdn = to_mxrcd(dev->devr.x0)->xrcdn;
+		in->ctx.pgoff_cqn |= cpu_to_be32(to_mcq(dev->devr.c0)->mcq.cqn);
+	}
+
+	in->ctx.flags_xrcd = cpu_to_be32((flgs & 0xFF000000) | (xrcdn & 0xFFFFFF));
+
+	in->ctx.pd = cpu_to_be32(to_mpd(pd)->pdn);
+	in->ctx.db_record = cpu_to_be64(srq->db.dma);
+	err = mlx5_core_create_srq(&dev->mdev, &srq->msrq, in, inlen);
+	mlx5_vfree(in);
+	if (err) {
+		mlx5_ib_dbg(dev, "create SRQ failed, err %d\n", err);
+		goto err_usr_kern_srq;
+	}
+
+	mlx5_ib_dbg(dev, "create SRQ with srqn 0x%x\n", srq->msrq.srqn);
+
+	srq->msrq.event = mlx5_ib_srq_event;
+	srq->ibsrq.ext.xrc.srq_num = srq->msrq.srqn;
+
+	if (pd->uobject)
+		if (ib_copy_to_udata(udata, &srq->msrq.srqn, sizeof(__u32))) {
+			mlx5_ib_dbg(dev, "copy to user failed\n");
+			err = -EFAULT;
+			goto err_core;
+		}
+
+	init_attr->attr.max_wr = srq->msrq.max - 1;
+
+	return &srq->ibsrq;
+
+err_core:
+	mlx5_core_destroy_srq(&dev->mdev, &srq->msrq);
+
+err_usr_kern_srq:
+	if (pd->uobject)
+		destroy_srq_user(pd, srq);
+	else
+		destroy_srq_kernel(dev, srq);
+
+err_srq:
+	kfree(srq);
+
+	return ERR_PTR(err);
+}
+
+int mlx5_ib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		       enum ib_srq_attr_mask attr_mask, struct ib_udata *udata)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	int ret;
+
+	/* We don't support resizing SRQs yet */
+	if (attr_mask & IB_SRQ_MAX_WR)
+		return -EINVAL;
+
+	if (attr_mask & IB_SRQ_LIMIT) {
+		if (attr->srq_limit >= srq->msrq.max)
+			return -EINVAL;
+
+		mutex_lock(&srq->mutex);
+		ret = mlx5_core_arm_srq(&dev->mdev, &srq->msrq, attr->srq_limit, 1);
+		mutex_unlock(&srq->mutex);
+
+		if (ret)
+			return ret;
+	}
+
+	return 0;
+}
+
+int mlx5_ib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
+{
+	struct mlx5_ib_dev *dev = to_mdev(ibsrq->device);
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	int ret;
+	struct mlx5_query_srq_mbox_out *out;
+
+	out = kzalloc(sizeof(*out), GFP_KERNEL);
+	if (!out)
+		return -ENOMEM;
+
+	ret = mlx5_core_query_srq(&dev->mdev, &srq->msrq, out);
+	if (ret)
+		goto out_box;
+
+	srq_attr->srq_limit = be16_to_cpu(out->ctx.lwm);
+	srq_attr->max_wr    = srq->msrq.max - 1;
+	srq_attr->max_sge   = srq->msrq.max_gs;
+
+out_box:
+	kfree(out);
+	return ret;
+}
+
+int mlx5_ib_destroy_srq(struct ib_srq *srq)
+{
+	struct mlx5_ib_dev *dev = to_mdev(srq->device);
+	struct mlx5_ib_srq *msrq = to_msrq(srq);
+
+	mlx5_core_destroy_srq(&dev->mdev, &msrq->msrq);
+
+	if (srq->uobject) {
+		mlx5_ib_db_unmap_user(to_mucontext(srq->uobject->context), &msrq->db);
+		ib_umem_release(msrq->umem);
+	} else {
+		destroy_srq_kernel(dev, msrq);
+	}
+
+	kfree(srq);
+	return 0;
+}
+
+void mlx5_ib_free_srq_wqe(struct mlx5_ib_srq *srq, int wqe_index)
+{
+	struct mlx5_wqe_srq_next_seg *next;
+
+	/* always called with interrupts disabled. */
+	spin_lock(&srq->lock);
+
+	next = get_wqe(srq, srq->tail);
+	next->next_wqe_index = cpu_to_be16(wqe_index);
+	srq->tail = wqe_index;
+
+	spin_unlock(&srq->lock);
+}
+
+int mlx5_ib_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			  struct ib_recv_wr **bad_wr)
+{
+	struct mlx5_ib_srq *srq = to_msrq(ibsrq);
+	struct mlx5_wqe_srq_next_seg *next;
+	struct mlx5_wqe_data_seg *scat;
+	unsigned long flags;
+	int err = 0;
+	int nreq;
+	int i;
+
+	spin_lock_irqsave(&srq->lock, flags);
+
+	for (nreq = 0; wr; nreq++, wr = wr->next) {
+		if (unlikely(wr->num_sge > srq->msrq.max_gs)) {
+			err = -EINVAL;
+			*bad_wr = wr;
+			break;
+		}
+
+		if (unlikely(srq->head == srq->tail)) {
+			err = -ENOMEM;
+			*bad_wr = wr;
+			break;
+		}
+
+		srq->wrid[srq->head] = wr->wr_id;
+
+		next      = get_wqe(srq, srq->head);
+		srq->head = be16_to_cpu(next->next_wqe_index);
+		scat      = (struct mlx5_wqe_data_seg *)(next + 1);
+
+		for (i = 0; i < wr->num_sge; i++) {
+			scat[i].byte_count = cpu_to_be32(wr->sg_list[i].length);
+			scat[i].lkey       = cpu_to_be32(wr->sg_list[i].lkey);
+			scat[i].addr       = cpu_to_be64(wr->sg_list[i].addr);
+		}
+
+		if (i < srq->msrq.max_avail_gather) {
+			scat[i].byte_count = 0;
+			scat[i].lkey       = cpu_to_be32(MLX5_INVALID_LKEY);
+			scat[i].addr       = 0;
+		}
+	}
+
+	if (likely(nreq)) {
+		srq->wqe_ctr += nreq;
+
+		/* Make sure that descriptors are written before
+		 * doorbell record.
+		 */
+		wmb();
+
+		*srq->db.db = cpu_to_be32(srq->wqe_ctr);
+	}
+
+	spin_unlock_irqrestore(&srq->lock, flags);
+
+	return err;
+}
diff --git a/drivers/infiniband/hw/mlx5/user.h b/drivers/infiniband/hw/mlx5/user.h
new file mode 100644
index 00000000000..d0ba264ac1e
--- /dev/null
+++ b/drivers/infiniband/hw/mlx5/user.h
@@ -0,0 +1,133 @@
+/*
+ * Copyright (c) 2013, Mellanox Technologies inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef MLX5_IB_USER_H
+#define MLX5_IB_USER_H
+
+#include <linux/types.h>
+
+enum {
+	MLX5_QP_FLAG_SIGNATURE		= 1 << 0,
+	MLX5_QP_FLAG_SCATTER_CQE	= 1 << 1,
+};
+
+enum {
+	MLX5_SRQ_FLAG_SIGNATURE		= 1 << 0,
+};
+
+
+/* Increment this value if any changes that break userspace ABI
+ * compatibility are made.
+ */
+#define MLX5_IB_UVERBS_ABI_VERSION	1
+
+/* Make sure that all structs defined in this file remain laid out so
+ * that they pack the same way on 32-bit and 64-bit architectures (to
+ * avoid incompatibility between 32-bit userspace and 64-bit kernels).
+ * In particular do not use pointer types -- pass pointers in __u64
+ * instead.
+ */
+
+struct mlx5_ib_alloc_ucontext_req {
+	__u32	total_num_uuars;
+	__u32	num_low_latency_uuars;
+};
+
+struct mlx5_ib_alloc_ucontext_req_v2 {
+	__u32	total_num_uuars;
+	__u32	num_low_latency_uuars;
+	__u32	flags;
+	__u32	reserved;
+};
+
+struct mlx5_ib_alloc_ucontext_resp {
+	__u32	qp_tab_size;
+	__u32	bf_reg_size;
+	__u32	tot_uuars;
+	__u32	cache_line_size;
+	__u16	max_sq_desc_sz;
+	__u16	max_rq_desc_sz;
+	__u32	max_send_wqebb;
+	__u32	max_recv_wr;
+	__u32	max_srq_recv_wr;
+	__u16	num_ports;
+	__u16	reserved;
+};
+
+struct mlx5_ib_alloc_pd_resp {
+	__u32	pdn;
+};
+
+struct mlx5_ib_create_cq {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	cqe_size;
+	__u32	reserved; /* explicit padding (optional on i386) */
+};
+
+struct mlx5_ib_create_cq_resp {
+	__u32	cqn;
+	__u32	reserved;
+};
+
+struct mlx5_ib_resize_cq {
+	__u64	buf_addr;
+	__u16	cqe_size;
+	__u16	reserved0;
+	__u32	reserved1;
+};
+
+struct mlx5_ib_create_srq {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	flags;
+	__u32	reserved; /* explicit padding (optional on i386) */
+};
+
+struct mlx5_ib_create_srq_resp {
+	__u32	srqn;
+	__u32	reserved;
+};
+
+struct mlx5_ib_create_qp {
+	__u64	buf_addr;
+	__u64	db_addr;
+	__u32	sq_wqe_count;
+	__u32	rq_wqe_count;
+	__u32	rq_wqe_shift;
+	__u32	flags;
+};
+
+struct mlx5_ib_create_qp_resp {
+	__u32	uuar_index;
+};
+#endif /* MLX5_IB_USER_H */
diff --git a/drivers/infiniband/hw/mthca/Kconfig b/drivers/infiniband/hw/mthca/Kconfig
index 03efc074967..da314c3fec2 100644
--- a/drivers/infiniband/hw/mthca/Kconfig
+++ b/drivers/infiniband/hw/mthca/Kconfig
@@ -7,7 +7,7 @@ config INFINIBAND_MTHCA
 	  ("Tavor") and the MT25208 PCI Express HCA ("Arbel").
 
 config INFINIBAND_MTHCA_DEBUG
-	bool "Verbose debugging output" if EMBEDDED
+	bool "Verbose debugging output" if EXPERT
 	depends on INFINIBAND_MTHCA
 	default y
 	---help---
diff --git a/drivers/infiniband/hw/mthca/mthca_allocator.c b/drivers/infiniband/hw/mthca/mthca_allocator.c
index c5ccc2daab6..b4e0cf4e95c 100644
--- a/drivers/infiniband/hw/mthca/mthca_allocator.c
+++ b/drivers/infiniband/hw/mthca/mthca_allocator.c
@@ -211,7 +211,7 @@ int mthca_buf_alloc(struct mthca_dev *dev, int size, int max_direct,
 		if (!buf->direct.buf)
 			return -ENOMEM;
 
-		pci_unmap_addr_set(&buf->direct, mapping, t);
+		dma_unmap_addr_set(&buf->direct, mapping, t);
 
 		memset(buf->direct.buf, 0, size);
 
@@ -251,7 +251,7 @@ int mthca_buf_alloc(struct mthca_dev *dev, int size, int max_direct,
 				goto err_free;
 
 			dma_list[i] = t;
-			pci_unmap_addr_set(&buf->page_list[i], mapping, t);
+			dma_unmap_addr_set(&buf->page_list[i], mapping, t);
 
 			clear_page(buf->page_list[i].buf);
 		}
@@ -289,12 +289,12 @@ void mthca_buf_free(struct mthca_dev *dev, int size, union mthca_buf *buf,
 
 	if (is_direct)
 		dma_free_coherent(&dev->pdev->dev, size, buf->direct.buf,
-				  pci_unmap_addr(&buf->direct, mapping));
+				  dma_unmap_addr(&buf->direct, mapping));
 	else {
 		for (i = 0; i < (size + PAGE_SIZE - 1) / PAGE_SIZE; ++i)
 			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
 					  buf->page_list[i].buf,
-					  pci_unmap_addr(&buf->page_list[i],
+					  dma_unmap_addr(&buf->page_list[i],
 							 mapping));
 		kfree(buf->page_list);
 	}
diff --git a/drivers/infiniband/hw/mthca/mthca_catas.c b/drivers/infiniband/hw/mthca/mthca_catas.c
index cc440f90000..712d2a30fbe 100644
--- a/drivers/infiniband/hw/mthca/mthca_catas.c
+++ b/drivers/infiniband/hw/mthca/mthca_catas.c
@@ -31,6 +31,7 @@
  */
 
 #include <linux/jiffies.h>
+#include <linux/module.h>
 #include <linux/timer.h>
 #include <linux/workqueue.h>
 
@@ -68,11 +69,16 @@ static void catas_reset(struct work_struct *work)
 	spin_unlock_irq(&catas_lock);
 
 	list_for_each_entry_safe(dev, tmpdev, &tlist, catas_err.list) {
+		struct pci_dev *pdev = dev->pdev;
 		ret = __mthca_restart_one(dev->pdev);
+		/* 'dev' now is not valid */
 		if (ret)
-			mthca_err(dev, "Reset failed (%d)\n", ret);
-		else
-			mthca_dbg(dev, "Reset succeeded\n");
+			printk(KERN_ERR "mthca %s: Reset failed (%d)\n",
+			       pci_name(pdev), ret);
+		else {
+			struct mthca_dev *d = pci_get_drvdata(pdev);
+			mthca_dbg(d, "Reset succeeded\n");
+		}
 	}
 
 	mutex_unlock(&mthca_device_mutex);
@@ -88,6 +94,7 @@ static void handle_catas(struct mthca_dev *dev)
 	event.device = &dev->ib_dev;
 	event.event  = IB_EVENT_DEVICE_FATAL;
 	event.element.port_num = 0;
+	dev->active = false;
 
 	ib_dispatch_event(&event);
 
@@ -140,7 +147,7 @@ static void poll_catas(unsigned long dev_ptr)
 
 void mthca_start_catas_poll(struct mthca_dev *dev)
 {
-	unsigned long addr;
+	phys_addr_t addr;
 
 	init_timer(&dev->catas_err.timer);
 	dev->catas_err.map  = NULL;
@@ -149,18 +156,11 @@ void mthca_start_catas_poll(struct mthca_dev *dev)
 		((pci_resource_len(dev->pdev, 0) - 1) &
 		 dev->catas_err.addr);
 
-	if (!request_mem_region(addr, dev->catas_err.size * 4,
-				DRV_NAME)) {
-		mthca_warn(dev, "couldn't request catastrophic error region "
-			   "at 0x%lx/0x%x\n", addr, dev->catas_err.size * 4);
-		return;
-	}
-
 	dev->catas_err.map = ioremap(addr, dev->catas_err.size * 4);
 	if (!dev->catas_err.map) {
 		mthca_warn(dev, "couldn't map catastrophic error region "
-			   "at 0x%lx/0x%x\n", addr, dev->catas_err.size * 4);
-		release_mem_region(addr, dev->catas_err.size * 4);
+			   "at 0x%llx/0x%x\n", (unsigned long long) addr,
+			   dev->catas_err.size * 4);
 		return;
 	}
 
@@ -175,13 +175,8 @@ void mthca_stop_catas_poll(struct mthca_dev *dev)
 {
 	del_timer_sync(&dev->catas_err.timer);
 
-	if (dev->catas_err.map) {
+	if (dev->catas_err.map)
 		iounmap(dev->catas_err.map);
-		release_mem_region(pci_resource_start(dev->pdev, 0) +
-				   ((pci_resource_len(dev->pdev, 0) - 1) &
-				    dev->catas_err.addr),
-				   dev->catas_err.size * 4);
-	}
 
 	spin_lock_irq(&catas_lock);
 	list_del(&dev->catas_err.list);
diff --git a/drivers/infiniband/hw/mthca/mthca_cmd.c b/drivers/infiniband/hw/mthca/mthca_cmd.c
index c33e1c53c79..9d3e5c1ac60 100644
--- a/drivers/infiniband/hw/mthca/mthca_cmd.c
+++ b/drivers/infiniband/hw/mthca/mthca_cmd.c
@@ -36,6 +36,8 @@
 #include <linux/pci.h>
 #include <linux/errno.h>
 #include <linux/sched.h>
+#include <linux/module.h>
+#include <linux/slab.h>
 #include <asm/io.h>
 #include <rdma/ib_mad.h>
 
@@ -157,13 +159,15 @@ enum {
 enum {
 	CMD_TIME_CLASS_A = (HZ + 999) / 1000 + 1,
 	CMD_TIME_CLASS_B = (HZ +  99) /  100 + 1,
-	CMD_TIME_CLASS_C = (HZ +   9) /   10 + 1
+	CMD_TIME_CLASS_C = (HZ +   9) /   10 + 1,
+	CMD_TIME_CLASS_D = 60 * HZ
 };
 #else
 enum {
 	CMD_TIME_CLASS_A = 60 * HZ,
 	CMD_TIME_CLASS_B = 60 * HZ,
-	CMD_TIME_CLASS_C = 60 * HZ
+	CMD_TIME_CLASS_C = 60 * HZ,
+	CMD_TIME_CLASS_D = 60 * HZ
 };
 #endif
 
@@ -298,6 +302,38 @@ static int mthca_cmd_post(struct mthca_dev *dev,
 	return err;
 }
 
+
+static int mthca_status_to_errno(u8 status)
+{
+	static const int trans_table[] = {
+		[MTHCA_CMD_STAT_INTERNAL_ERR]   = -EIO,
+		[MTHCA_CMD_STAT_BAD_OP]         = -EPERM,
+		[MTHCA_CMD_STAT_BAD_PARAM]      = -EINVAL,
+		[MTHCA_CMD_STAT_BAD_SYS_STATE]  = -ENXIO,
+		[MTHCA_CMD_STAT_BAD_RESOURCE]   = -EBADF,
+		[MTHCA_CMD_STAT_RESOURCE_BUSY]  = -EBUSY,
+		[MTHCA_CMD_STAT_DDR_MEM_ERR]    = -ENOMEM,
+		[MTHCA_CMD_STAT_EXCEED_LIM]     = -ENOMEM,
+		[MTHCA_CMD_STAT_BAD_RES_STATE]  = -EBADF,
+		[MTHCA_CMD_STAT_BAD_INDEX]      = -EBADF,
+		[MTHCA_CMD_STAT_BAD_NVMEM]      = -EFAULT,
+		[MTHCA_CMD_STAT_BAD_QPEE_STATE] = -EINVAL,
+		[MTHCA_CMD_STAT_BAD_SEG_PARAM]  = -EFAULT,
+		[MTHCA_CMD_STAT_REG_BOUND]      = -EBUSY,
+		[MTHCA_CMD_STAT_LAM_NOT_PRE]    = -EAGAIN,
+		[MTHCA_CMD_STAT_BAD_PKT]        = -EBADMSG,
+		[MTHCA_CMD_STAT_BAD_SIZE]       = -ENOMEM,
+	};
+
+	if (status >= ARRAY_SIZE(trans_table) ||
+			(status != MTHCA_CMD_STAT_OK
+			 && trans_table[status] == 0))
+		return -EINVAL;
+
+	return trans_table[status];
+}
+
+
 static int mthca_cmd_poll(struct mthca_dev *dev,
 			  u64 in_param,
 			  u64 *out_param,
@@ -305,11 +341,11 @@ static int mthca_cmd_poll(struct mthca_dev *dev,
 			  u32 in_modifier,
 			  u8 op_modifier,
 			  u16 op,
-			  unsigned long timeout,
-			  u8 *status)
+			  unsigned long timeout)
 {
 	int err = 0;
 	unsigned long end;
+	u8 status;
 
 	down(&dev->cmd.poll_sem);
 
@@ -338,7 +374,12 @@ static int mthca_cmd_poll(struct mthca_dev *dev,
 			(u64) be32_to_cpu((__force __be32)
 					  __raw_readl(dev->hcr + HCR_OUT_PARAM_OFFSET + 4));
 
-	*status = be32_to_cpu((__force __be32) __raw_readl(dev->hcr + HCR_STATUS_OFFSET)) >> 24;
+	status = be32_to_cpu((__force __be32) __raw_readl(dev->hcr + HCR_STATUS_OFFSET)) >> 24;
+	if (status) {
+		mthca_dbg(dev, "Command %02x completed with status %02x\n",
+			  op, status);
+		err = mthca_status_to_errno(status);
+	}
 
 out:
 	up(&dev->cmd.poll_sem);
@@ -371,8 +412,7 @@ static int mthca_cmd_wait(struct mthca_dev *dev,
 			  u32 in_modifier,
 			  u8 op_modifier,
 			  u16 op,
-			  unsigned long timeout,
-			  u8 *status)
+			  unsigned long timeout)
 {
 	int err = 0;
 	struct mthca_cmd_context *context;
@@ -404,10 +444,11 @@ static int mthca_cmd_wait(struct mthca_dev *dev,
 	if (err)
 		goto out;
 
-	*status = context->status;
-	if (*status)
+	if (context->status) {
 		mthca_dbg(dev, "Command %02x completed with status %02x\n",
-			  op, *status);
+			  op, context->status);
+		err = mthca_status_to_errno(context->status);
+	}
 
 	if (out_is_imm)
 		*out_param = context->out_param;
@@ -429,17 +470,16 @@ static int mthca_cmd_box(struct mthca_dev *dev,
 			 u32 in_modifier,
 			 u8 op_modifier,
 			 u16 op,
-			 unsigned long timeout,
-			 u8 *status)
+			 unsigned long timeout)
 {
 	if (dev->cmd.flags & MTHCA_CMD_USE_EVENTS)
 		return mthca_cmd_wait(dev, in_param, &out_param, 0,
 				      in_modifier, op_modifier, op,
-				      timeout, status);
+				      timeout);
 	else
 		return mthca_cmd_poll(dev, in_param, &out_param, 0,
 				      in_modifier, op_modifier, op,
-				      timeout, status);
+				      timeout);
 }
 
 /* Invoke a command with no output parameter */
@@ -448,11 +488,10 @@ static int mthca_cmd(struct mthca_dev *dev,
 		     u32 in_modifier,
 		     u8 op_modifier,
 		     u16 op,
-		     unsigned long timeout,
-		     u8 *status)
+		     unsigned long timeout)
 {
 	return mthca_cmd_box(dev, in_param, 0, in_modifier,
-			     op_modifier, op, timeout, status);
+			     op_modifier, op, timeout);
 }
 
 /*
@@ -466,17 +505,16 @@ static int mthca_cmd_imm(struct mthca_dev *dev,
 			 u32 in_modifier,
 			 u8 op_modifier,
 			 u16 op,
-			 unsigned long timeout,
-			 u8 *status)
+			 unsigned long timeout)
 {
 	if (dev->cmd.flags & MTHCA_CMD_USE_EVENTS)
 		return mthca_cmd_wait(dev, in_param, out_param, 1,
 				      in_modifier, op_modifier, op,
-				      timeout, status);
+				      timeout);
 	else
 		return mthca_cmd_poll(dev, in_param, out_param, 1,
 				      in_modifier, op_modifier, op,
-				      timeout, status);
+				      timeout);
 }
 
 int mthca_cmd_init(struct mthca_dev *dev)
@@ -593,14 +631,14 @@ void mthca_free_mailbox(struct mthca_dev *dev, struct mthca_mailbox *mailbox)
 	kfree(mailbox);
 }
 
-int mthca_SYS_EN(struct mthca_dev *dev, u8 *status)
+int mthca_SYS_EN(struct mthca_dev *dev)
 {
 	u64 out;
 	int ret;
 
-	ret = mthca_cmd_imm(dev, 0, &out, 0, 0, CMD_SYS_EN, HZ, status);
+	ret = mthca_cmd_imm(dev, 0, &out, 0, 0, CMD_SYS_EN, CMD_TIME_CLASS_D);
 
-	if (*status == MTHCA_CMD_STAT_DDR_MEM_ERR)
+	if (ret == -ENOMEM)
 		mthca_warn(dev, "SYS_EN DDR error: syn=%x, sock=%d, "
 			   "sladdr=%d, SPD source=%s\n",
 			   (int) (out >> 6) & 0xf, (int) (out >> 4) & 3,
@@ -609,13 +647,13 @@ int mthca_SYS_EN(struct mthca_dev *dev, u8 *status)
 	return ret;
 }
 
-int mthca_SYS_DIS(struct mthca_dev *dev, u8 *status)
+int mthca_SYS_DIS(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, HZ, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, CMD_TIME_CLASS_C);
 }
 
 static int mthca_map_cmd(struct mthca_dev *dev, u16 op, struct mthca_icm *icm,
-			 u64 virt, u8 *status)
+			 u64 virt)
 {
 	struct mthca_mailbox *mailbox;
 	struct mthca_icm_iter iter;
@@ -663,8 +701,8 @@ static int mthca_map_cmd(struct mthca_dev *dev, u16 op, struct mthca_icm *icm,
 
 			if (++nent == MTHCA_MAILBOX_SIZE / 16) {
 				err = mthca_cmd(dev, mailbox->dma, nent, 0, op,
-						CMD_TIME_CLASS_B, status);
-				if (err || *status)
+						CMD_TIME_CLASS_B);
+				if (err)
 					goto out;
 				nent = 0;
 			}
@@ -673,7 +711,7 @@ static int mthca_map_cmd(struct mthca_dev *dev, u16 op, struct mthca_icm *icm,
 
 	if (nent)
 		err = mthca_cmd(dev, mailbox->dma, nent, 0, op,
-				CMD_TIME_CLASS_B, status);
+				CMD_TIME_CLASS_B);
 
 	switch (op) {
 	case CMD_MAP_FA:
@@ -693,24 +731,24 @@ out:
 	return err;
 }
 
-int mthca_MAP_FA(struct mthca_dev *dev, struct mthca_icm *icm, u8 *status)
+int mthca_MAP_FA(struct mthca_dev *dev, struct mthca_icm *icm)
 {
-	return mthca_map_cmd(dev, CMD_MAP_FA, icm, -1, status);
+	return mthca_map_cmd(dev, CMD_MAP_FA, icm, -1);
 }
 
-int mthca_UNMAP_FA(struct mthca_dev *dev, u8 *status)
+int mthca_UNMAP_FA(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_FA, CMD_TIME_CLASS_B, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_FA, CMD_TIME_CLASS_B);
 }
 
-int mthca_RUN_FW(struct mthca_dev *dev, u8 *status)
+int mthca_RUN_FW(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_RUN_FW, CMD_TIME_CLASS_A, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_RUN_FW, CMD_TIME_CLASS_A);
 }
 
 static void mthca_setup_cmd_doorbells(struct mthca_dev *dev, u64 base)
 {
-	unsigned long addr;
+	phys_addr_t addr;
 	u16 max_off = 0;
 	int i;
 
@@ -734,7 +772,7 @@ static void mthca_setup_cmd_doorbells(struct mthca_dev *dev, u64 base)
 	mthca_dbg(dev, "Mapped doorbell page for posting FW commands\n");
 }
 
-int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status)
+int mthca_QUERY_FW(struct mthca_dev *dev)
 {
 	struct mthca_mailbox *mailbox;
 	u32 *outbox;
@@ -768,7 +806,7 @@ int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status)
 	outbox = mailbox->buf;
 
 	err = mthca_cmd_box(dev, 0, mailbox->dma, 0, 0, CMD_QUERY_FW,
-			    CMD_TIME_CLASS_A, status);
+			    CMD_TIME_CLASS_A);
 
 	if (err)
 		goto out;
@@ -840,7 +878,7 @@ out:
 	return err;
 }
 
-int mthca_ENABLE_LAM(struct mthca_dev *dev, u8 *status)
+int mthca_ENABLE_LAM(struct mthca_dev *dev)
 {
 	struct mthca_mailbox *mailbox;
 	u8 info;
@@ -861,14 +899,11 @@ int mthca_ENABLE_LAM(struct mthca_dev *dev, u8 *status)
 	outbox = mailbox->buf;
 
 	err = mthca_cmd_box(dev, 0, mailbox->dma, 0, 0, CMD_ENABLE_LAM,
-			    CMD_TIME_CLASS_C, status);
+			    CMD_TIME_CLASS_C);
 
 	if (err)
 		goto out;
 
-	if (*status == MTHCA_CMD_STAT_LAM_NOT_PRE)
-		goto out;
-
 	MTHCA_GET(dev->ddr_start, outbox, ENABLE_LAM_START_OFFSET);
 	MTHCA_GET(dev->ddr_end,   outbox, ENABLE_LAM_END_OFFSET);
 	MTHCA_GET(info,           outbox, ENABLE_LAM_INFO_OFFSET);
@@ -893,12 +928,12 @@ out:
 	return err;
 }
 
-int mthca_DISABLE_LAM(struct mthca_dev *dev, u8 *status)
+int mthca_DISABLE_LAM(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, CMD_TIME_CLASS_C, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYS_DIS, CMD_TIME_CLASS_C);
 }
 
-int mthca_QUERY_DDR(struct mthca_dev *dev, u8 *status)
+int mthca_QUERY_DDR(struct mthca_dev *dev)
 {
 	struct mthca_mailbox *mailbox;
 	u8 info;
@@ -919,7 +954,7 @@ int mthca_QUERY_DDR(struct mthca_dev *dev, u8 *status)
 	outbox = mailbox->buf;
 
 	err = mthca_cmd_box(dev, 0, mailbox->dma, 0, 0, CMD_QUERY_DDR,
-			    CMD_TIME_CLASS_A, status);
+			    CMD_TIME_CLASS_A);
 
 	if (err)
 		goto out;
@@ -949,7 +984,7 @@ out:
 }
 
 int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
-			struct mthca_dev_lim *dev_lim, u8 *status)
+			struct mthca_dev_lim *dev_lim)
 {
 	struct mthca_mailbox *mailbox;
 	u32 *outbox;
@@ -1025,7 +1060,7 @@ int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
 	outbox = mailbox->buf;
 
 	err = mthca_cmd_box(dev, 0, mailbox->dma, 0, 0, CMD_QUERY_DEV_LIM,
-			    CMD_TIME_CLASS_A, status);
+			    CMD_TIME_CLASS_A);
 
 	if (err)
 		goto out;
@@ -1057,7 +1092,7 @@ int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
 	MTHCA_GET(field, outbox, QUERY_DEV_LIM_RSVD_MTT_OFFSET);
 	if (mthca_is_memfree(dev))
 		dev_lim->reserved_mtts = ALIGN((1 << (field >> 4)) * sizeof(u64),
-					       MTHCA_MTT_SEG_SIZE) / MTHCA_MTT_SEG_SIZE;
+					       dev->limits.mtt_seg_size) / dev->limits.mtt_seg_size;
 	else
 		dev_lim->reserved_mtts = 1 << (field >> 4);
 	MTHCA_GET(field, outbox, QUERY_DEV_LIM_MAX_MRW_SZ_OFFSET);
@@ -1229,7 +1264,7 @@ static void get_board_id(void *vsd, char *board_id)
 }
 
 int mthca_QUERY_ADAPTER(struct mthca_dev *dev,
-			struct mthca_adapter *adapter, u8 *status)
+			struct mthca_adapter *adapter)
 {
 	struct mthca_mailbox *mailbox;
 	u32 *outbox;
@@ -1248,7 +1283,7 @@ int mthca_QUERY_ADAPTER(struct mthca_dev *dev,
 	outbox = mailbox->buf;
 
 	err = mthca_cmd_box(dev, 0, mailbox->dma, 0, 0, CMD_QUERY_ADAPTER,
-			    CMD_TIME_CLASS_A, status);
+			    CMD_TIME_CLASS_A);
 
 	if (err)
 		goto out;
@@ -1272,8 +1307,7 @@ out:
 }
 
 int mthca_INIT_HCA(struct mthca_dev *dev,
-		   struct mthca_init_hca_param *param,
-		   u8 *status)
+		   struct mthca_init_hca_param *param)
 {
 	struct mthca_mailbox *mailbox;
 	__be32 *inbox;
@@ -1390,7 +1424,8 @@ int mthca_INIT_HCA(struct mthca_dev *dev,
 		MTHCA_PUT(inbox, param->uarc_base,   INIT_HCA_UAR_CTX_BASE_OFFSET);
 	}
 
-	err = mthca_cmd(dev, mailbox->dma, 0, 0, CMD_INIT_HCA, HZ, status);
+	err = mthca_cmd(dev, mailbox->dma, 0, 0,
+			CMD_INIT_HCA, CMD_TIME_CLASS_D);
 
 	mthca_free_mailbox(dev, mailbox);
 	return err;
@@ -1398,7 +1433,7 @@ int mthca_INIT_HCA(struct mthca_dev *dev,
 
 int mthca_INIT_IB(struct mthca_dev *dev,
 		  struct mthca_init_ib_param *param,
-		  int port, u8 *status)
+		  int port)
 {
 	struct mthca_mailbox *mailbox;
 	u32 *inbox;
@@ -1442,24 +1477,24 @@ int mthca_INIT_IB(struct mthca_dev *dev,
 	MTHCA_PUT(inbox, param->si_guid,   INIT_IB_SI_GUID_OFFSET);
 
 	err = mthca_cmd(dev, mailbox->dma, port, 0, CMD_INIT_IB,
-			CMD_TIME_CLASS_A, status);
+			CMD_TIME_CLASS_A);
 
 	mthca_free_mailbox(dev, mailbox);
 	return err;
 }
 
-int mthca_CLOSE_IB(struct mthca_dev *dev, int port, u8 *status)
+int mthca_CLOSE_IB(struct mthca_dev *dev, int port)
 {
-	return mthca_cmd(dev, 0, port, 0, CMD_CLOSE_IB, HZ, status);
+	return mthca_cmd(dev, 0, port, 0, CMD_CLOSE_IB, CMD_TIME_CLASS_A);
 }
 
-int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic, u8 *status)
+int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic)
 {
-	return mthca_cmd(dev, 0, 0, panic, CMD_CLOSE_HCA, HZ, status);
+	return mthca_cmd(dev, 0, 0, panic, CMD_CLOSE_HCA, CMD_TIME_CLASS_C);
 }
 
 int mthca_SET_IB(struct mthca_dev *dev, struct mthca_set_ib_param *param,
-		 int port, u8 *status)
+		 int port)
 {
 	struct mthca_mailbox *mailbox;
 	u32 *inbox;
@@ -1488,18 +1523,18 @@ int mthca_SET_IB(struct mthca_dev *dev, struct mthca_set_ib_param *param,
 	MTHCA_PUT(inbox, param->si_guid,  SET_IB_SI_GUID_OFFSET);
 
 	err = mthca_cmd(dev, mailbox->dma, port, 0, CMD_SET_IB,
-			CMD_TIME_CLASS_B, status);
+			CMD_TIME_CLASS_B);
 
 	mthca_free_mailbox(dev, mailbox);
 	return err;
 }
 
-int mthca_MAP_ICM(struct mthca_dev *dev, struct mthca_icm *icm, u64 virt, u8 *status)
+int mthca_MAP_ICM(struct mthca_dev *dev, struct mthca_icm *icm, u64 virt)
 {
-	return mthca_map_cmd(dev, CMD_MAP_ICM, icm, virt, status);
+	return mthca_map_cmd(dev, CMD_MAP_ICM, icm, virt);
 }
 
-int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt, u8 *status)
+int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt)
 {
 	struct mthca_mailbox *mailbox;
 	__be64 *inbox;
@@ -1514,7 +1549,7 @@ int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt, u8 *status
 	inbox[1] = cpu_to_be64(dma_addr);
 
 	err = mthca_cmd(dev, mailbox->dma, 1, 0, CMD_MAP_ICM,
-			CMD_TIME_CLASS_B, status);
+			CMD_TIME_CLASS_B);
 
 	mthca_free_mailbox(dev, mailbox);
 
@@ -1525,31 +1560,31 @@ int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt, u8 *status
 	return err;
 }
 
-int mthca_UNMAP_ICM(struct mthca_dev *dev, u64 virt, u32 page_count, u8 *status)
+int mthca_UNMAP_ICM(struct mthca_dev *dev, u64 virt, u32 page_count)
 {
 	mthca_dbg(dev, "Unmapping %d pages at %llx from ICM.\n",
 		  page_count, (unsigned long long) virt);
 
-	return mthca_cmd(dev, virt, page_count, 0, CMD_UNMAP_ICM, CMD_TIME_CLASS_B, status);
+	return mthca_cmd(dev, virt, page_count, 0,
+			CMD_UNMAP_ICM, CMD_TIME_CLASS_B);
 }
 
-int mthca_MAP_ICM_AUX(struct mthca_dev *dev, struct mthca_icm *icm, u8 *status)
+int mthca_MAP_ICM_AUX(struct mthca_dev *dev, struct mthca_icm *icm)
 {
-	return mthca_map_cmd(dev, CMD_MAP_ICM_AUX, icm, -1, status);
+	return mthca_map_cmd(dev, CMD_MAP_ICM_AUX, icm, -1);
 }
 
-int mthca_UNMAP_ICM_AUX(struct mthca_dev *dev, u8 *status)
+int mthca_UNMAP_ICM_AUX(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_ICM_AUX, CMD_TIME_CLASS_B, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_UNMAP_ICM_AUX, CMD_TIME_CLASS_B);
 }
 
-int mthca_SET_ICM_SIZE(struct mthca_dev *dev, u64 icm_size, u64 *aux_pages,
-		       u8 *status)
+int mthca_SET_ICM_SIZE(struct mthca_dev *dev, u64 icm_size, u64 *aux_pages)
 {
-	int ret = mthca_cmd_imm(dev, icm_size, aux_pages, 0, 0, CMD_SET_ICM_SIZE,
-				CMD_TIME_CLASS_A, status);
+	int ret = mthca_cmd_imm(dev, icm_size, aux_pages, 0,
+			0, CMD_SET_ICM_SIZE, CMD_TIME_CLASS_A);
 
-	if (ret || status)
+	if (ret)
 		return ret;
 
 	/*
@@ -1563,74 +1598,73 @@ int mthca_SET_ICM_SIZE(struct mthca_dev *dev, u64 icm_size, u64 *aux_pages,
 }
 
 int mthca_SW2HW_MPT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int mpt_index, u8 *status)
+		    int mpt_index)
 {
 	return mthca_cmd(dev, mailbox->dma, mpt_index, 0, CMD_SW2HW_MPT,
-			 CMD_TIME_CLASS_B, status);
+			 CMD_TIME_CLASS_B);
 }
 
 int mthca_HW2SW_MPT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int mpt_index, u8 *status)
+		    int mpt_index)
 {
 	return mthca_cmd_box(dev, 0, mailbox ? mailbox->dma : 0, mpt_index,
 			     !mailbox, CMD_HW2SW_MPT,
-			     CMD_TIME_CLASS_B, status);
+			     CMD_TIME_CLASS_B);
 }
 
 int mthca_WRITE_MTT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int num_mtt, u8 *status)
+		    int num_mtt)
 {
 	return mthca_cmd(dev, mailbox->dma, num_mtt, 0, CMD_WRITE_MTT,
-			 CMD_TIME_CLASS_B, status);
+			 CMD_TIME_CLASS_B);
 }
 
-int mthca_SYNC_TPT(struct mthca_dev *dev, u8 *status)
+int mthca_SYNC_TPT(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0, 0, CMD_SYNC_TPT, CMD_TIME_CLASS_B, status);
+	return mthca_cmd(dev, 0, 0, 0, CMD_SYNC_TPT, CMD_TIME_CLASS_B);
 }
 
 int mthca_MAP_EQ(struct mthca_dev *dev, u64 event_mask, int unmap,
-		 int eq_num, u8 *status)
+		 int eq_num)
 {
 	mthca_dbg(dev, "%s mask %016llx for eqn %d\n",
 		  unmap ? "Clearing" : "Setting",
 		  (unsigned long long) event_mask, eq_num);
 	return mthca_cmd(dev, event_mask, (unmap << 31) | eq_num,
-			 0, CMD_MAP_EQ, CMD_TIME_CLASS_B, status);
+			 0, CMD_MAP_EQ, CMD_TIME_CLASS_B);
 }
 
 int mthca_SW2HW_EQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int eq_num, u8 *status)
+		   int eq_num)
 {
 	return mthca_cmd(dev, mailbox->dma, eq_num, 0, CMD_SW2HW_EQ,
-			 CMD_TIME_CLASS_A, status);
+			 CMD_TIME_CLASS_A);
 }
 
 int mthca_HW2SW_EQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int eq_num, u8 *status)
+		   int eq_num)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, eq_num, 0,
 			     CMD_HW2SW_EQ,
-			     CMD_TIME_CLASS_A, status);
+			     CMD_TIME_CLASS_A);
 }
 
 int mthca_SW2HW_CQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int cq_num, u8 *status)
+		   int cq_num)
 {
 	return mthca_cmd(dev, mailbox->dma, cq_num, 0, CMD_SW2HW_CQ,
-			CMD_TIME_CLASS_A, status);
+			CMD_TIME_CLASS_A);
 }
 
 int mthca_HW2SW_CQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int cq_num, u8 *status)
+		   int cq_num)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, cq_num, 0,
 			     CMD_HW2SW_CQ,
-			     CMD_TIME_CLASS_A, status);
+			     CMD_TIME_CLASS_A);
 }
 
-int mthca_RESIZE_CQ(struct mthca_dev *dev, int cq_num, u32 lkey, u8 log_size,
-		    u8 *status)
+int mthca_RESIZE_CQ(struct mthca_dev *dev, int cq_num, u32 lkey, u8 log_size)
 {
 	struct mthca_mailbox *mailbox;
 	__be32 *inbox;
@@ -1654,44 +1688,43 @@ int mthca_RESIZE_CQ(struct mthca_dev *dev, int cq_num, u32 lkey, u8 log_size,
 	MTHCA_PUT(inbox, lkey,     RESIZE_CQ_LKEY_OFFSET);
 
 	err = mthca_cmd(dev, mailbox->dma, cq_num, 1, CMD_RESIZE_CQ,
-			CMD_TIME_CLASS_B, status);
+			CMD_TIME_CLASS_B);
 
 	mthca_free_mailbox(dev, mailbox);
 	return err;
 }
 
 int mthca_SW2HW_SRQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int srq_num, u8 *status)
+		    int srq_num)
 {
 	return mthca_cmd(dev, mailbox->dma, srq_num, 0, CMD_SW2HW_SRQ,
-			CMD_TIME_CLASS_A, status);
+			CMD_TIME_CLASS_A);
 }
 
 int mthca_HW2SW_SRQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int srq_num, u8 *status)
+		    int srq_num)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, srq_num, 0,
 			     CMD_HW2SW_SRQ,
-			     CMD_TIME_CLASS_A, status);
+			     CMD_TIME_CLASS_A);
 }
 
 int mthca_QUERY_SRQ(struct mthca_dev *dev, u32 num,
-		    struct mthca_mailbox *mailbox, u8 *status)
+		    struct mthca_mailbox *mailbox)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, num, 0,
-			     CMD_QUERY_SRQ, CMD_TIME_CLASS_A, status);
+			     CMD_QUERY_SRQ, CMD_TIME_CLASS_A);
 }
 
-int mthca_ARM_SRQ(struct mthca_dev *dev, int srq_num, int limit, u8 *status)
+int mthca_ARM_SRQ(struct mthca_dev *dev, int srq_num, int limit)
 {
 	return mthca_cmd(dev, limit, srq_num, 0, CMD_ARM_SRQ,
-			 CMD_TIME_CLASS_B, status);
+			 CMD_TIME_CLASS_B);
 }
 
 int mthca_MODIFY_QP(struct mthca_dev *dev, enum ib_qp_state cur,
 		    enum ib_qp_state next, u32 num, int is_ee,
-		    struct mthca_mailbox *mailbox, u32 optmask,
-		    u8 *status)
+		    struct mthca_mailbox *mailbox, u32 optmask)
 {
 	static const u16 op[IB_QPS_ERR + 1][IB_QPS_ERR + 1] = {
 		[IB_QPS_RESET] = {
@@ -1752,7 +1785,7 @@ int mthca_MODIFY_QP(struct mthca_dev *dev, enum ib_qp_state cur,
 
 		err = mthca_cmd_box(dev, 0, mailbox ? mailbox->dma : 0,
 				    (!!is_ee << 24) | num, op_mod,
-				    op[cur][next], CMD_TIME_CLASS_C, status);
+				    op[cur][next], CMD_TIME_CLASS_C);
 
 		if (0 && mailbox) {
 			int i;
@@ -1786,21 +1819,20 @@ int mthca_MODIFY_QP(struct mthca_dev *dev, enum ib_qp_state cur,
 		}
 
 		err = mthca_cmd(dev, mailbox->dma, optmask | (!!is_ee << 24) | num,
-				op_mod, op[cur][next], CMD_TIME_CLASS_C, status);
+				op_mod, op[cur][next], CMD_TIME_CLASS_C);
 	}
 
 	return err;
 }
 
 int mthca_QUERY_QP(struct mthca_dev *dev, u32 num, int is_ee,
-		   struct mthca_mailbox *mailbox, u8 *status)
+		   struct mthca_mailbox *mailbox)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, (!!is_ee << 24) | num, 0,
-			     CMD_QUERY_QPEE, CMD_TIME_CLASS_A, status);
+			     CMD_QUERY_QPEE, CMD_TIME_CLASS_A);
 }
 
-int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
-			  u8 *status)
+int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn)
 {
 	u8 op_mod;
 
@@ -1814,7 +1846,7 @@ int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
 	case IB_QPT_RAW_IPV6:
 		op_mod = 2;
 		break;
-	case IB_QPT_RAW_ETY:
+	case IB_QPT_RAW_ETHERTYPE:
 		op_mod = 3;
 		break;
 	default:
@@ -1822,12 +1854,12 @@ int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
 	}
 
 	return mthca_cmd(dev, 0, qpn, op_mod, CMD_CONF_SPECIAL_QP,
-			 CMD_TIME_CLASS_B, status);
+			 CMD_TIME_CLASS_B);
 }
 
 int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int ignore_bkey,
 		  int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
-		  void *in_mad, void *response_mad, u8 *status)
+		  void *in_mad, void *response_mad)
 {
 	struct mthca_mailbox *inmailbox, *outmailbox;
 	void *inbox;
@@ -1894,9 +1926,9 @@ int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int ignore_bkey,
 
 	err = mthca_cmd_box(dev, inmailbox->dma, outmailbox->dma,
 			    in_modifier, op_modifier,
-			    CMD_MAD_IFC, CMD_TIME_CLASS_C, status);
+			    CMD_MAD_IFC, CMD_TIME_CLASS_C);
 
-	if (!err && !*status)
+	if (!err)
 		memcpy(response_mad, outmailbox->buf, 256);
 
 	mthca_free_mailbox(dev, inmailbox);
@@ -1905,33 +1937,33 @@ int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int ignore_bkey,
 }
 
 int mthca_READ_MGM(struct mthca_dev *dev, int index,
-		   struct mthca_mailbox *mailbox, u8 *status)
+		   struct mthca_mailbox *mailbox)
 {
 	return mthca_cmd_box(dev, 0, mailbox->dma, index, 0,
-			     CMD_READ_MGM, CMD_TIME_CLASS_A, status);
+			     CMD_READ_MGM, CMD_TIME_CLASS_A);
 }
 
 int mthca_WRITE_MGM(struct mthca_dev *dev, int index,
-		    struct mthca_mailbox *mailbox, u8 *status)
+		    struct mthca_mailbox *mailbox)
 {
 	return mthca_cmd(dev, mailbox->dma, index, 0, CMD_WRITE_MGM,
-			 CMD_TIME_CLASS_A, status);
+			 CMD_TIME_CLASS_A);
 }
 
 int mthca_MGID_HASH(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    u16 *hash, u8 *status)
+		    u16 *hash)
 {
 	u64 imm;
 	int err;
 
 	err = mthca_cmd_imm(dev, mailbox->dma, &imm, 0, 0, CMD_MGID_HASH,
-			    CMD_TIME_CLASS_A, status);
+			    CMD_TIME_CLASS_A);
 
 	*hash = imm;
 	return err;
 }
 
-int mthca_NOP(struct mthca_dev *dev, u8 *status)
+int mthca_NOP(struct mthca_dev *dev)
 {
-	return mthca_cmd(dev, 0, 0x1f, 0, CMD_NOP, msecs_to_jiffies(100), status);
+	return mthca_cmd(dev, 0, 0x1f, 0, CMD_NOP, msecs_to_jiffies(100));
 }
diff --git a/drivers/infiniband/hw/mthca/mthca_cmd.h b/drivers/infiniband/hw/mthca/mthca_cmd.h
index 6efd3265f24..f952244c54d 100644
--- a/drivers/infiniband/hw/mthca/mthca_cmd.h
+++ b/drivers/infiniband/hw/mthca/mthca_cmd.h
@@ -252,79 +252,74 @@ struct mthca_mailbox *mthca_alloc_mailbox(struct mthca_dev *dev,
 					  gfp_t gfp_mask);
 void mthca_free_mailbox(struct mthca_dev *dev, struct mthca_mailbox *mailbox);
 
-int mthca_SYS_EN(struct mthca_dev *dev, u8 *status);
-int mthca_SYS_DIS(struct mthca_dev *dev, u8 *status);
-int mthca_MAP_FA(struct mthca_dev *dev, struct mthca_icm *icm, u8 *status);
-int mthca_UNMAP_FA(struct mthca_dev *dev, u8 *status);
-int mthca_RUN_FW(struct mthca_dev *dev, u8 *status);
-int mthca_QUERY_FW(struct mthca_dev *dev, u8 *status);
-int mthca_ENABLE_LAM(struct mthca_dev *dev, u8 *status);
-int mthca_DISABLE_LAM(struct mthca_dev *dev, u8 *status);
-int mthca_QUERY_DDR(struct mthca_dev *dev, u8 *status);
+int mthca_SYS_EN(struct mthca_dev *dev);
+int mthca_SYS_DIS(struct mthca_dev *dev);
+int mthca_MAP_FA(struct mthca_dev *dev, struct mthca_icm *icm);
+int mthca_UNMAP_FA(struct mthca_dev *dev);
+int mthca_RUN_FW(struct mthca_dev *dev);
+int mthca_QUERY_FW(struct mthca_dev *dev);
+int mthca_ENABLE_LAM(struct mthca_dev *dev);
+int mthca_DISABLE_LAM(struct mthca_dev *dev);
+int mthca_QUERY_DDR(struct mthca_dev *dev);
 int mthca_QUERY_DEV_LIM(struct mthca_dev *dev,
-			struct mthca_dev_lim *dev_lim, u8 *status);
+			struct mthca_dev_lim *dev_lim);
 int mthca_QUERY_ADAPTER(struct mthca_dev *dev,
-			struct mthca_adapter *adapter, u8 *status);
+			struct mthca_adapter *adapter);
 int mthca_INIT_HCA(struct mthca_dev *dev,
-		   struct mthca_init_hca_param *param,
-		   u8 *status);
+		   struct mthca_init_hca_param *param);
 int mthca_INIT_IB(struct mthca_dev *dev,
 		  struct mthca_init_ib_param *param,
-		  int port, u8 *status);
-int mthca_CLOSE_IB(struct mthca_dev *dev, int port, u8 *status);
-int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic, u8 *status);
+		  int port);
+int mthca_CLOSE_IB(struct mthca_dev *dev, int port);
+int mthca_CLOSE_HCA(struct mthca_dev *dev, int panic);
 int mthca_SET_IB(struct mthca_dev *dev, struct mthca_set_ib_param *param,
-		 int port, u8 *status);
-int mthca_MAP_ICM(struct mthca_dev *dev, struct mthca_icm *icm, u64 virt, u8 *status);
-int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt, u8 *status);
-int mthca_UNMAP_ICM(struct mthca_dev *dev, u64 virt, u32 page_count, u8 *status);
-int mthca_MAP_ICM_AUX(struct mthca_dev *dev, struct mthca_icm *icm, u8 *status);
-int mthca_UNMAP_ICM_AUX(struct mthca_dev *dev, u8 *status);
-int mthca_SET_ICM_SIZE(struct mthca_dev *dev, u64 icm_size, u64 *aux_pages,
-		       u8 *status);
+		 int port);
+int mthca_MAP_ICM(struct mthca_dev *dev, struct mthca_icm *icm, u64 virt);
+int mthca_MAP_ICM_page(struct mthca_dev *dev, u64 dma_addr, u64 virt);
+int mthca_UNMAP_ICM(struct mthca_dev *dev, u64 virt, u32 page_count);
+int mthca_MAP_ICM_AUX(struct mthca_dev *dev, struct mthca_icm *icm);
+int mthca_UNMAP_ICM_AUX(struct mthca_dev *dev);
+int mthca_SET_ICM_SIZE(struct mthca_dev *dev, u64 icm_size, u64 *aux_pages);
 int mthca_SW2HW_MPT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int mpt_index, u8 *status);
+		    int mpt_index);
 int mthca_HW2SW_MPT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int mpt_index, u8 *status);
+		    int mpt_index);
 int mthca_WRITE_MTT(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int num_mtt, u8 *status);
-int mthca_SYNC_TPT(struct mthca_dev *dev, u8 *status);
+		    int num_mtt);
+int mthca_SYNC_TPT(struct mthca_dev *dev);
 int mthca_MAP_EQ(struct mthca_dev *dev, u64 event_mask, int unmap,
-		 int eq_num, u8 *status);
+		 int eq_num);
 int mthca_SW2HW_EQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int eq_num, u8 *status);
+		   int eq_num);
 int mthca_HW2SW_EQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int eq_num, u8 *status);
+		   int eq_num);
 int mthca_SW2HW_CQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int cq_num, u8 *status);
+		   int cq_num);
 int mthca_HW2SW_CQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		   int cq_num, u8 *status);
-int mthca_RESIZE_CQ(struct mthca_dev *dev, int cq_num, u32 lkey, u8 log_size,
-		    u8 *status);
+		   int cq_num);
+int mthca_RESIZE_CQ(struct mthca_dev *dev, int cq_num, u32 lkey, u8 log_size);
 int mthca_SW2HW_SRQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int srq_num, u8 *status);
+		    int srq_num);
 int mthca_HW2SW_SRQ(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    int srq_num, u8 *status);
+		    int srq_num);
 int mthca_QUERY_SRQ(struct mthca_dev *dev, u32 num,
-		    struct mthca_mailbox *mailbox, u8 *status);
-int mthca_ARM_SRQ(struct mthca_dev *dev, int srq_num, int limit, u8 *status);
+		    struct mthca_mailbox *mailbox);
+int mthca_ARM_SRQ(struct mthca_dev *dev, int srq_num, int limit);
 int mthca_MODIFY_QP(struct mthca_dev *dev, enum ib_qp_state cur,
 		    enum ib_qp_state next, u32 num, int is_ee,
-		    struct mthca_mailbox *mailbox, u32 optmask,
-		    u8 *status);
+		    struct mthca_mailbox *mailbox, u32 optmask);
 int mthca_QUERY_QP(struct mthca_dev *dev, u32 num, int is_ee,
-		   struct mthca_mailbox *mailbox, u8 *status);
-int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn,
-			  u8 *status);
+		   struct mthca_mailbox *mailbox);
+int mthca_CONF_SPECIAL_QP(struct mthca_dev *dev, int type, u32 qpn);
 int mthca_MAD_IFC(struct mthca_dev *dev, int ignore_mkey, int ignore_bkey,
 		  int port, struct ib_wc *in_wc, struct ib_grh *in_grh,
-		  void *in_mad, void *response_mad, u8 *status);
+		  void *in_mad, void *response_mad);
 int mthca_READ_MGM(struct mthca_dev *dev, int index,
-		   struct mthca_mailbox *mailbox, u8 *status);
+		   struct mthca_mailbox *mailbox);
 int mthca_WRITE_MGM(struct mthca_dev *dev, int index,
-		    struct mthca_mailbox *mailbox, u8 *status);
+		    struct mthca_mailbox *mailbox);
 int mthca_MGID_HASH(struct mthca_dev *dev, struct mthca_mailbox *mailbox,
-		    u16 *hash, u8 *status);
-int mthca_NOP(struct mthca_dev *dev, u8 *status);
+		    u16 *hash);
+int mthca_NOP(struct mthca_dev *dev);
 
 #endif /* MTHCA_CMD_H */
diff --git a/drivers/infiniband/hw/mthca/mthca_config_reg.h b/drivers/infiniband/hw/mthca/mthca_config_reg.h
index 75671f75cac..155bc66395b 100644
--- a/drivers/infiniband/hw/mthca/mthca_config_reg.h
+++ b/drivers/infiniband/hw/mthca/mthca_config_reg.h
@@ -34,8 +34,6 @@
 #ifndef MTHCA_CONFIG_REG_H
 #define MTHCA_CONFIG_REG_H
 
-#include <asm/page.h>
-
 #define MTHCA_HCR_BASE         0x80680
 #define MTHCA_HCR_SIZE         0x0001c
 #define MTHCA_ECR_BASE         0x80700
diff --git a/drivers/infiniband/hw/mthca/mthca_cq.c b/drivers/infiniband/hw/mthca/mthca_cq.c
index d9f4735c2b3..40ba8333815 100644
--- a/drivers/infiniband/hw/mthca/mthca_cq.c
+++ b/drivers/infiniband/hw/mthca/mthca_cq.c
@@ -34,6 +34,7 @@
  * SOFTWARE.
  */
 
+#include <linux/gfp.h>
 #include <linux/hardirq.h>
 #include <linux/sched.h>
 
@@ -642,7 +643,8 @@ static inline int mthca_poll_one(struct mthca_dev *dev,
 		entry->wc_flags   |= cqe->g_mlpath & 0x80 ? IB_WC_GRH : 0;
 		checksum = (be32_to_cpu(cqe->rqpn) >> 24) |
 				((be32_to_cpu(cqe->my_ee) >> 16) & 0xff00);
-		entry->csum_ok = (cqe->sl_ipok & 1 && checksum == 0xffff);
+		entry->wc_flags	  |=  (cqe->sl_ipok & 1 && checksum == 0xffff) ?
+							IB_WC_IP_CSUM_OK : 0;
 	}
 
 	entry->status = IB_WC_SUCCESS;
@@ -778,7 +780,6 @@ int mthca_init_cq(struct mthca_dev *dev, int nent,
 	struct mthca_mailbox *mailbox;
 	struct mthca_cq_context *cq_context;
 	int err = -ENOMEM;
-	u8 status;
 
 	cq->ibcq.cqe  = nent - 1;
 	cq->is_kernel = !ctx;
@@ -846,19 +847,12 @@ int mthca_init_cq(struct mthca_dev *dev, int nent,
 		cq_context->state_db = cpu_to_be32(cq->arm_db_index);
 	}
 
-	err = mthca_SW2HW_CQ(dev, mailbox, cq->cqn, &status);
+	err = mthca_SW2HW_CQ(dev, mailbox, cq->cqn);
 	if (err) {
 		mthca_warn(dev, "SW2HW_CQ failed (%d)\n", err);
 		goto err_out_free_mr;
 	}
 
-	if (status) {
-		mthca_warn(dev, "SW2HW_CQ returned status 0x%02x\n",
-			   status);
-		err = -EINVAL;
-		goto err_out_free_mr;
-	}
-
 	spin_lock_irq(&dev->cq_table.lock);
 	if (mthca_array_set(&dev->cq_table.cq,
 			    cq->cqn & (dev->limits.num_cqs - 1),
@@ -914,7 +908,6 @@ void mthca_free_cq(struct mthca_dev *dev,
 {
 	struct mthca_mailbox *mailbox;
 	int err;
-	u8 status;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox)) {
@@ -922,11 +915,9 @@ void mthca_free_cq(struct mthca_dev *dev,
 		return;
 	}
 
-	err = mthca_HW2SW_CQ(dev, mailbox, cq->cqn, &status);
+	err = mthca_HW2SW_CQ(dev, mailbox, cq->cqn);
 	if (err)
 		mthca_warn(dev, "HW2SW_CQ failed (%d)\n", err);
-	else if (status)
-		mthca_warn(dev, "HW2SW_CQ returned status 0x%02x\n", status);
 
 	if (0) {
 		__be32 *ctx = mailbox->buf;
diff --git a/drivers/infiniband/hw/mthca/mthca_dev.h b/drivers/infiniband/hw/mthca/mthca_dev.h
index ee4d073c889..7e6a6d64ad4 100644
--- a/drivers/infiniband/hw/mthca/mthca_dev.h
+++ b/drivers/infiniband/hw/mthca/mthca_dev.h
@@ -159,6 +159,7 @@ struct mthca_limits {
 	int      reserved_eqs;
 	int      num_mpts;
 	int      num_mtt_segs;
+	int	 mtt_seg_size;
 	int      fmr_reserved_mtts;
 	int      reserved_mtts;
 	int      reserved_mrws;
@@ -202,6 +203,7 @@ struct mthca_pd_table {
 
 struct mthca_buddy {
 	unsigned long **bits;
+	int	       *num_free;
 	int             max_order;
 	spinlock_t      lock;
 };
@@ -355,6 +357,7 @@ struct mthca_dev {
 	struct ib_ah         *sm_ah[MTHCA_MAX_PORTS];
 	spinlock_t            sm_lock;
 	u8                    rate[MTHCA_MAX_PORTS];
+	bool		      active;
 };
 
 #ifdef CONFIG_INFINIBAND_MTHCA_DEBUG
diff --git a/drivers/infiniband/hw/mthca/mthca_eq.c b/drivers/infiniband/hw/mthca/mthca_eq.c
index 4e36aa7cb3d..69020173899 100644
--- a/drivers/infiniband/hw/mthca/mthca_eq.c
+++ b/drivers/infiniband/hw/mthca/mthca_eq.c
@@ -34,6 +34,7 @@
 #include <linux/errno.h>
 #include <linux/interrupt.h>
 #include <linux/pci.h>
+#include <linux/slab.h>
 
 #include "mthca_dev.h"
 #include "mthca_cmd.h"
@@ -356,7 +357,7 @@ static int mthca_eq_int(struct mthca_dev *dev, struct mthca_eq *eq)
 			mthca_warn(dev, "Unhandled event %02x(%02x) on EQ %d\n",
 				   eqe->type, eqe->subtype, eq->eqn);
 			break;
-		};
+		}
 
 		set_eqe_hw(eqe);
 		++eq->cons_index;
@@ -473,7 +474,6 @@ static int mthca_create_eq(struct mthca_dev *dev,
 	struct mthca_eq_context *eq_context;
 	int err = -ENOMEM;
 	int i;
-	u8 status;
 
 	eq->dev  = dev;
 	eq->nent = roundup_pow_of_two(max(nent, 2));
@@ -503,7 +503,7 @@ static int mthca_create_eq(struct mthca_dev *dev,
 			goto err_out_free_pages;
 
 		dma_list[i] = t;
-		pci_unmap_addr_set(&eq->page_list[i], mapping, t);
+		dma_unmap_addr_set(&eq->page_list[i], mapping, t);
 
 		clear_page(eq->page_list[i].buf);
 	}
@@ -542,15 +542,9 @@ static int mthca_create_eq(struct mthca_dev *dev,
 	eq_context->intr            = intr;
 	eq_context->lkey            = cpu_to_be32(eq->mr.ibmr.lkey);
 
-	err = mthca_SW2HW_EQ(dev, mailbox, eq->eqn, &status);
+	err = mthca_SW2HW_EQ(dev, mailbox, eq->eqn);
 	if (err) {
-		mthca_warn(dev, "SW2HW_EQ failed (%d)\n", err);
-		goto err_out_free_mr;
-	}
-	if (status) {
-		mthca_warn(dev, "SW2HW_EQ returned status 0x%02x\n",
-			   status);
-		err = -EINVAL;
+		mthca_warn(dev, "SW2HW_EQ returned %d\n", err);
 		goto err_out_free_mr;
 	}
 
@@ -578,7 +572,7 @@ static int mthca_create_eq(struct mthca_dev *dev,
 		if (eq->page_list[i].buf)
 			dma_free_coherent(&dev->pdev->dev, PAGE_SIZE,
 					  eq->page_list[i].buf,
-					  pci_unmap_addr(&eq->page_list[i],
+					  dma_unmap_addr(&eq->page_list[i],
 							 mapping));
 
 	mthca_free_mailbox(dev, mailbox);
@@ -596,7 +590,6 @@ static void mthca_free_eq(struct mthca_dev *dev,
 {
 	struct mthca_mailbox *mailbox;
 	int err;
-	u8 status;
 	int npages = (eq->nent * MTHCA_EQ_ENTRY_SIZE + PAGE_SIZE - 1) /
 		PAGE_SIZE;
 	int i;
@@ -605,11 +598,9 @@ static void mthca_free_eq(struct mthca_dev *dev,
 	if (IS_ERR(mailbox))
 		return;
 
-	err = mthca_HW2SW_EQ(dev, mailbox, eq->eqn, &status);
+	err = mthca_HW2SW_EQ(dev, mailbox, eq->eqn);
 	if (err)
-		mthca_warn(dev, "HW2SW_EQ failed (%d)\n", err);
-	if (status)
-		mthca_warn(dev, "HW2SW_EQ returned status 0x%02x\n", status);
+		mthca_warn(dev, "HW2SW_EQ returned %d\n", err);
 
 	dev->eq_table.arm_mask &= ~eq->eqn_mask;
 
@@ -628,7 +619,7 @@ static void mthca_free_eq(struct mthca_dev *dev,
 	for (i = 0; i < npages; ++i)
 		pci_free_consistent(dev->pdev, PAGE_SIZE,
 				    eq->page_list[i].buf,
-				    pci_unmap_addr(&eq->page_list[i], mapping));
+				    dma_unmap_addr(&eq->page_list[i], mapping));
 
 	kfree(eq->page_list);
 	mthca_free_mailbox(dev, mailbox);
@@ -641,38 +632,26 @@ static void mthca_free_irqs(struct mthca_dev *dev)
 	if (dev->eq_table.have_irq)
 		free_irq(dev->pdev->irq, dev);
 	for (i = 0; i < MTHCA_NUM_EQ; ++i)
-		if (dev->eq_table.eq[i].have_irq)
+		if (dev->eq_table.eq[i].have_irq) {
 			free_irq(dev->eq_table.eq[i].msi_x_vector,
 				 dev->eq_table.eq + i);
+			dev->eq_table.eq[i].have_irq = 0;
+		}
 }
 
 static int mthca_map_reg(struct mthca_dev *dev,
 			 unsigned long offset, unsigned long size,
 			 void __iomem **map)
 {
-	unsigned long base = pci_resource_start(dev->pdev, 0);
-
-	if (!request_mem_region(base + offset, size, DRV_NAME))
-		return -EBUSY;
+	phys_addr_t base = pci_resource_start(dev->pdev, 0);
 
 	*map = ioremap(base + offset, size);
-	if (!*map) {
-		release_mem_region(base + offset, size);
+	if (!*map)
 		return -ENOMEM;
-	}
 
 	return 0;
 }
 
-static void mthca_unmap_reg(struct mthca_dev *dev, unsigned long offset,
-			    unsigned long size, void __iomem *map)
-{
-	unsigned long base = pci_resource_start(dev->pdev, 0);
-
-	release_mem_region(base + offset, size);
-	iounmap(map);
-}
-
 static int mthca_map_eq_regs(struct mthca_dev *dev)
 {
 	if (mthca_is_memfree(dev)) {
@@ -699,9 +678,7 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 					dev->fw.arbel.eq_arm_base) + 4, 4,
 				  &dev->eq_regs.arbel.eq_arm)) {
 			mthca_err(dev, "Couldn't map EQ arm register, aborting.\n");
-			mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-					dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 
@@ -710,12 +687,8 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 				  MTHCA_EQ_SET_CI_SIZE,
 				  &dev->eq_regs.arbel.eq_set_ci_base)) {
 			mthca_err(dev, "Couldn't map EQ CI register, aborting.\n");
-			mthca_unmap_reg(dev, ((pci_resource_len(dev->pdev, 0) - 1) &
-					      dev->fw.arbel.eq_arm_base) + 4, 4,
-					dev->eq_regs.arbel.eq_arm);
-			mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-					dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->eq_regs.arbel.eq_arm);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 	} else {
@@ -731,8 +704,7 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 				  &dev->eq_regs.tavor.ecr_base)) {
 			mthca_err(dev, "Couldn't map ecr register, "
 				  "aborting.\n");
-			mthca_unmap_reg(dev, MTHCA_CLR_INT_BASE, MTHCA_CLR_INT_SIZE,
-					dev->clr_base);
+			iounmap(dev->clr_base);
 			return -ENOMEM;
 		}
 	}
@@ -744,29 +716,18 @@ static int mthca_map_eq_regs(struct mthca_dev *dev)
 static void mthca_unmap_eq_regs(struct mthca_dev *dev)
 {
 	if (mthca_is_memfree(dev)) {
-		mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-				dev->fw.arbel.eq_set_ci_base,
-				MTHCA_EQ_SET_CI_SIZE,
-				dev->eq_regs.arbel.eq_set_ci_base);
-		mthca_unmap_reg(dev, ((pci_resource_len(dev->pdev, 0) - 1) &
-				      dev->fw.arbel.eq_arm_base) + 4, 4,
-				dev->eq_regs.arbel.eq_arm);
-		mthca_unmap_reg(dev, (pci_resource_len(dev->pdev, 0) - 1) &
-				dev->fw.arbel.clr_int_base, MTHCA_CLR_INT_SIZE,
-				dev->clr_base);
+		iounmap(dev->eq_regs.arbel.eq_set_ci_base);
+		iounmap(dev->eq_regs.arbel.eq_arm);
+		iounmap(dev->clr_base);
 	} else {
-		mthca_unmap_reg(dev, MTHCA_ECR_BASE,
-				MTHCA_ECR_SIZE + MTHCA_ECR_CLR_SIZE,
-				dev->eq_regs.tavor.ecr_base);
-		mthca_unmap_reg(dev, MTHCA_CLR_INT_BASE, MTHCA_CLR_INT_SIZE,
-				dev->clr_base);
+		iounmap(dev->eq_regs.tavor.ecr_base);
+		iounmap(dev->clr_base);
 	}
 }
 
 int mthca_map_eq_icm(struct mthca_dev *dev, u64 icm_virt)
 {
 	int ret;
-	u8 status;
 
 	/*
 	 * We assume that mapping one page is enough for the whole EQ
@@ -780,14 +741,12 @@ int mthca_map_eq_icm(struct mthca_dev *dev, u64 icm_virt)
 		return -ENOMEM;
 	dev->eq_table.icm_dma  = pci_map_page(dev->pdev, dev->eq_table.icm_page, 0,
 					      PAGE_SIZE, PCI_DMA_BIDIRECTIONAL);
-	if (pci_dma_mapping_error(dev->eq_table.icm_dma)) {
+	if (pci_dma_mapping_error(dev->pdev, dev->eq_table.icm_dma)) {
 		__free_page(dev->eq_table.icm_page);
 		return -ENOMEM;
 	}
 
-	ret = mthca_MAP_ICM_page(dev, dev->eq_table.icm_dma, icm_virt, &status);
-	if (!ret && status)
-		ret = -EINVAL;
+	ret = mthca_MAP_ICM_page(dev, dev->eq_table.icm_dma, icm_virt);
 	if (ret) {
 		pci_unmap_page(dev->pdev, dev->eq_table.icm_dma, PAGE_SIZE,
 			       PCI_DMA_BIDIRECTIONAL);
@@ -799,9 +758,7 @@ int mthca_map_eq_icm(struct mthca_dev *dev, u64 icm_virt)
 
 void mthca_unmap_eq_icm(struct mthca_dev *dev)
 {
-	u8 status;
-
-	mthca_UNMAP_ICM(dev, dev->eq_table.icm_virt, 1, &status);
+	mthca_UNMAP_ICM(dev, dev->eq_table.icm_virt, 1);
 	pci_unmap_page(dev->pdev, dev->eq_table.icm_dma, PAGE_SIZE,
 		       PCI_DMA_BIDIRECTIONAL);
 	__free_page(dev->eq_table.icm_page);
@@ -810,7 +767,6 @@ void mthca_unmap_eq_icm(struct mthca_dev *dev)
 int mthca_init_eq_table(struct mthca_dev *dev)
 {
 	int err;
-	u8 status;
 	u8 intr;
 	int i;
 
@@ -858,49 +814,50 @@ int mthca_init_eq_table(struct mthca_dev *dev)
 
 	if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
 		static const char *eq_name[] = {
-			[MTHCA_EQ_COMP]  = DRV_NAME " (comp)",
-			[MTHCA_EQ_ASYNC] = DRV_NAME " (async)",
-			[MTHCA_EQ_CMD]   = DRV_NAME " (cmd)"
+			[MTHCA_EQ_COMP]  = DRV_NAME "-comp",
+			[MTHCA_EQ_ASYNC] = DRV_NAME "-async",
+			[MTHCA_EQ_CMD]   = DRV_NAME "-cmd"
 		};
 
 		for (i = 0; i < MTHCA_NUM_EQ; ++i) {
+			snprintf(dev->eq_table.eq[i].irq_name,
+				 IB_DEVICE_NAME_MAX,
+				 "%s@pci:%s", eq_name[i],
+				 pci_name(dev->pdev));
 			err = request_irq(dev->eq_table.eq[i].msi_x_vector,
 					  mthca_is_memfree(dev) ?
 					  mthca_arbel_msi_x_interrupt :
 					  mthca_tavor_msi_x_interrupt,
-					  0, eq_name[i], dev->eq_table.eq + i);
+					  0, dev->eq_table.eq[i].irq_name,
+					  dev->eq_table.eq + i);
 			if (err)
 				goto err_out_cmd;
 			dev->eq_table.eq[i].have_irq = 1;
 		}
 	} else {
+		snprintf(dev->eq_table.eq[0].irq_name, IB_DEVICE_NAME_MAX,
+			 DRV_NAME "@pci:%s", pci_name(dev->pdev));
 		err = request_irq(dev->pdev->irq,
 				  mthca_is_memfree(dev) ?
 				  mthca_arbel_interrupt :
 				  mthca_tavor_interrupt,
-				  IRQF_SHARED, DRV_NAME, dev);
+				  IRQF_SHARED, dev->eq_table.eq[0].irq_name, dev);
 		if (err)
 			goto err_out_cmd;
 		dev->eq_table.have_irq = 1;
 	}
 
 	err = mthca_MAP_EQ(dev, async_mask(dev),
-			   0, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+			   0, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn);
 	if (err)
 		mthca_warn(dev, "MAP_EQ for async EQ %d failed (%d)\n",
 			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, err);
-	if (status)
-		mthca_warn(dev, "MAP_EQ for async EQ %d returned status 0x%02x\n",
-			   dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, status);
 
 	err = mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
-			   0, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+			   0, dev->eq_table.eq[MTHCA_EQ_CMD].eqn);
 	if (err)
 		mthca_warn(dev, "MAP_EQ for cmd EQ %d failed (%d)\n",
 			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, err);
-	if (status)
-		mthca_warn(dev, "MAP_EQ for cmd EQ %d returned status 0x%02x\n",
-			   dev->eq_table.eq[MTHCA_EQ_CMD].eqn, status);
 
 	for (i = 0; i < MTHCA_NUM_EQ; ++i)
 		if (mthca_is_memfree(dev))
@@ -930,15 +887,14 @@ err_out_free:
 
 void mthca_cleanup_eq_table(struct mthca_dev *dev)
 {
-	u8 status;
 	int i;
 
 	mthca_free_irqs(dev);
 
 	mthca_MAP_EQ(dev, async_mask(dev),
-		     1, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn, &status);
+		     1, dev->eq_table.eq[MTHCA_EQ_ASYNC].eqn);
 	mthca_MAP_EQ(dev, MTHCA_CMD_EVENT_MASK,
-		     1, dev->eq_table.eq[MTHCA_EQ_CMD].eqn, &status);
+		     1, dev->eq_table.eq[MTHCA_EQ_CMD].eqn);
 
 	for (i = 0; i < MTHCA_NUM_EQ; ++i)
 		mthca_free_eq(dev, &dev->eq_table.eq[i]);
diff --git a/drivers/infiniband/hw/mthca/mthca_mad.c b/drivers/infiniband/hw/mthca/mthca_mad.c
index 640449582ab..b6f7f457fc5 100644
--- a/drivers/infiniband/hw/mthca/mthca_mad.c
+++ b/drivers/infiniband/hw/mthca/mthca_mad.c
@@ -104,7 +104,8 @@ static void update_sm_ah(struct mthca_dev *dev,
  */
 static void smp_snoop(struct ib_device *ibdev,
 		      u8 port_num,
-		      struct ib_mad *mad)
+		      struct ib_mad *mad,
+		      u16 prev_lid)
 {
 	struct ib_event event;
 
@@ -114,6 +115,7 @@ static void smp_snoop(struct ib_device *ibdev,
 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO) {
 			struct ib_port_info *pinfo =
 				(struct ib_port_info *) ((struct ib_smp *) mad)->data;
+			u16 lid = be16_to_cpu(pinfo->lid);
 
 			mthca_update_rate(to_mdev(ibdev), port_num);
 			update_sm_ah(to_mdev(ibdev), port_num,
@@ -123,12 +125,15 @@ static void smp_snoop(struct ib_device *ibdev,
 			event.device           = ibdev;
 			event.element.port_num = port_num;
 
-			if (pinfo->clientrereg_resv_subnetto & 0x80)
+			if (pinfo->clientrereg_resv_subnetto & 0x80) {
 				event.event    = IB_EVENT_CLIENT_REREGISTER;
-			else
-				event.event    = IB_EVENT_LID_CHANGE;
+				ib_dispatch_event(&event);
+			}
 
-			ib_dispatch_event(&event);
+			if (prev_lid != lid) {
+				event.event    = IB_EVENT_LID_CHANGE;
+				ib_dispatch_event(&event);
+			}
 		}
 
 		if (mad->mad_hdr.attr_id == IB_SMP_ATTR_PKEY_TABLE) {
@@ -166,6 +171,8 @@ static void forward_trap(struct mthca_dev *dev,
 	if (agent) {
 		send_buf = ib_create_send_mad(agent, qpn, 0, 0, IB_MGMT_MAD_HDR,
 					      IB_MGMT_MAD_DATA, GFP_ATOMIC);
+		if (IS_ERR(send_buf))
+			return;
 		/*
 		 * We rely here on the fact that MLX QPs don't use the
 		 * address handle after the send is posted (this is
@@ -194,8 +201,9 @@ int mthca_process_mad(struct ib_device *ibdev,
 		      struct ib_mad *out_mad)
 {
 	int err;
-	u8 status;
 	u16 slid = in_wc ? in_wc->slid : be16_to_cpu(IB_LID_PERMISSIVE);
+	u16 prev_lid = 0;
+	struct ib_port_attr pattr;
 
 	/* Forward locally generated traps to the SM */
 	if (in_mad->mad_hdr.method == IB_MGMT_METHOD_TRAP &&
@@ -233,26 +241,26 @@ int mthca_process_mad(struct ib_device *ibdev,
 			return IB_MAD_RESULT_SUCCESS;
 	} else
 		return IB_MAD_RESULT_SUCCESS;
+	if ((in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_LID_ROUTED ||
+	     in_mad->mad_hdr.mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) &&
+	    in_mad->mad_hdr.method == IB_MGMT_METHOD_SET &&
+	    in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
+	    !ib_query_port(ibdev, port_num, &pattr))
+		prev_lid = pattr.lid;
 
 	err = mthca_MAD_IFC(to_mdev(ibdev),
 			    mad_flags & IB_MAD_IGNORE_MKEY,
 			    mad_flags & IB_MAD_IGNORE_BKEY,
-			    port_num, in_wc, in_grh, in_mad, out_mad,
-			    &status);
-	if (err) {
-		mthca_err(to_mdev(ibdev), "MAD_IFC failed\n");
-		return IB_MAD_RESULT_FAILURE;
-	}
-	if (status == MTHCA_CMD_STAT_BAD_PKT)
+			    port_num, in_wc, in_grh, in_mad, out_mad);
+	if (err == -EBADMSG)
 		return IB_MAD_RESULT_SUCCESS;
-	if (status) {
-		mthca_err(to_mdev(ibdev), "MAD_IFC returned status %02x\n",
-			  status);
+	else if (err) {
+		mthca_err(to_mdev(ibdev), "MAD_IFC returned %d\n", err);
 		return IB_MAD_RESULT_FAILURE;
 	}
 
 	if (!out_mad->mad_hdr.status) {
-		smp_snoop(ibdev, port_num, in_mad);
+		smp_snoop(ibdev, port_num, in_mad, prev_lid);
 		node_desc_override(ibdev, out_mad);
 	}
 
diff --git a/drivers/infiniband/hw/mthca/mthca_main.c b/drivers/infiniband/hw/mthca/mthca_main.c
index fb9f91b60f3..ded76c101dd 100644
--- a/drivers/infiniband/hw/mthca/mthca_main.c
+++ b/drivers/infiniband/hw/mthca/mthca_main.c
@@ -37,6 +37,7 @@
 #include <linux/errno.h>
 #include <linux/pci.h>
 #include <linux/interrupt.h>
+#include <linux/gfp.h>
 
 #include "mthca_dev.h"
 #include "mthca_config_reg.h"
@@ -125,7 +126,11 @@ module_param_named(fmr_reserved_mtts, hca_profile.fmr_reserved_mtts, int, 0444);
 MODULE_PARM_DESC(fmr_reserved_mtts,
 		 "number of memory translation table segments reserved for FMR");
 
-static char mthca_version[] __devinitdata =
+static int log_mtts_per_seg = ilog2(MTHCA_MTT_SEG_SIZE / 8);
+module_param_named(log_mtts_per_seg, log_mtts_per_seg, int, 0444);
+MODULE_PARM_DESC(log_mtts_per_seg, "Log2 number of MTT entries per segment (1-5)");
+
+static char mthca_version[] =
 	DRV_NAME ": Mellanox InfiniBand HCA driver v"
 	DRV_VERSION " (" DRV_RELDATE ")\n";
 
@@ -144,7 +149,7 @@ static int mthca_tune_pci(struct mthca_dev *mdev)
 	} else if (!(mdev->mthca_flags & MTHCA_FLAG_PCIE))
 		mthca_info(mdev, "No PCI-X capability, not setting RBC.\n");
 
-	if (pci_find_capability(mdev->pdev, PCI_CAP_ID_EXP)) {
+	if (pci_is_pcie(mdev->pdev)) {
 		if (pcie_set_readrq(mdev->pdev, 4096)) {
 			mthca_err(mdev, "Couldn't write PCI Express read request, "
 				"aborting.\n");
@@ -160,18 +165,14 @@ static int mthca_tune_pci(struct mthca_dev *mdev)
 static int mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
 {
 	int err;
-	u8 status;
 
-	err = mthca_QUERY_DEV_LIM(mdev, dev_lim, &status);
+	mdev->limits.mtt_seg_size = (1 << log_mtts_per_seg) * 8;
+	err = mthca_QUERY_DEV_LIM(mdev, dev_lim);
 	if (err) {
-		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
+		mthca_err(mdev, "QUERY_DEV_LIM command returned %d"
+				", aborting.\n", err);
 		return err;
 	}
-	if (status) {
-		mthca_err(mdev, "QUERY_DEV_LIM returned status 0x%02x, "
-			  "aborting.\n", status);
-		return -EINVAL;
-	}
 	if (dev_lim->min_page_sz > PAGE_SIZE) {
 		mthca_err(mdev, "HCA minimum page size of %d bigger than "
 			  "kernel PAGE_SIZE of %ld, aborting.\n",
@@ -287,49 +288,32 @@ static int mthca_dev_lim(struct mthca_dev *mdev, struct mthca_dev_lim *dev_lim)
 static int mthca_init_tavor(struct mthca_dev *mdev)
 {
 	s64 size;
-	u8 status;
 	int err;
 	struct mthca_dev_lim        dev_lim;
 	struct mthca_profile        profile;
 	struct mthca_init_hca_param init_hca;
 
-	err = mthca_SYS_EN(mdev, &status);
+	err = mthca_SYS_EN(mdev);
 	if (err) {
-		mthca_err(mdev, "SYS_EN command failed, aborting.\n");
+		mthca_err(mdev, "SYS_EN command returned %d, aborting.\n", err);
 		return err;
 	}
-	if (status) {
-		mthca_err(mdev, "SYS_EN returned status 0x%02x, "
-			  "aborting.\n", status);
-		return -EINVAL;
-	}
 
-	err = mthca_QUERY_FW(mdev, &status);
+	err = mthca_QUERY_FW(mdev);
 	if (err) {
-		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
-		goto err_disable;
-	}
-	if (status) {
-		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
-			  "aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "QUERY_FW command returned %d,"
+				" aborting.\n", err);
 		goto err_disable;
 	}
-	err = mthca_QUERY_DDR(mdev, &status);
+	err = mthca_QUERY_DDR(mdev);
 	if (err) {
-		mthca_err(mdev, "QUERY_DDR command failed, aborting.\n");
-		goto err_disable;
-	}
-	if (status) {
-		mthca_err(mdev, "QUERY_DDR returned status 0x%02x, "
-			  "aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "QUERY_DDR command returned %d, aborting.\n", err);
 		goto err_disable;
 	}
 
 	err = mthca_dev_lim(mdev, &dev_lim);
 	if (err) {
-		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
+		mthca_err(mdev, "QUERY_DEV_LIM command returned %d, aborting.\n", err);
 		goto err_disable;
 	}
 
@@ -345,29 +329,22 @@ static int mthca_init_tavor(struct mthca_dev *mdev)
 		goto err_disable;
 	}
 
-	err = mthca_INIT_HCA(mdev, &init_hca, &status);
+	err = mthca_INIT_HCA(mdev, &init_hca);
 	if (err) {
-		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
-		goto err_disable;
-	}
-	if (status) {
-		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
-			  "aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "INIT_HCA command returned %d, aborting.\n", err);
 		goto err_disable;
 	}
 
 	return 0;
 
 err_disable:
-	mthca_SYS_DIS(mdev, &status);
+	mthca_SYS_DIS(mdev);
 
 	return err;
 }
 
 static int mthca_load_fw(struct mthca_dev *mdev)
 {
-	u8 status;
 	int err;
 
 	/* FIXME: use HCA-attached memory for FW if present */
@@ -380,31 +357,21 @@ static int mthca_load_fw(struct mthca_dev *mdev)
 		return -ENOMEM;
 	}
 
-	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm, &status);
+	err = mthca_MAP_FA(mdev, mdev->fw.arbel.fw_icm);
 	if (err) {
-		mthca_err(mdev, "MAP_FA command failed, aborting.\n");
-		goto err_free;
-	}
-	if (status) {
-		mthca_err(mdev, "MAP_FA returned status 0x%02x, aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "MAP_FA command returned %d, aborting.\n", err);
 		goto err_free;
 	}
-	err = mthca_RUN_FW(mdev, &status);
+	err = mthca_RUN_FW(mdev);
 	if (err) {
-		mthca_err(mdev, "RUN_FW command failed, aborting.\n");
-		goto err_unmap_fa;
-	}
-	if (status) {
-		mthca_err(mdev, "RUN_FW returned status 0x%02x, aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "RUN_FW command returned %d, aborting.\n", err);
 		goto err_unmap_fa;
 	}
 
 	return 0;
 
 err_unmap_fa:
-	mthca_UNMAP_FA(mdev, &status);
+	mthca_UNMAP_FA(mdev);
 
 err_free:
 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
@@ -417,19 +384,13 @@ static int mthca_init_icm(struct mthca_dev *mdev,
 			  u64 icm_size)
 {
 	u64 aux_pages;
-	u8 status;
 	int err;
 
-	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages, &status);
+	err = mthca_SET_ICM_SIZE(mdev, icm_size, &aux_pages);
 	if (err) {
-		mthca_err(mdev, "SET_ICM_SIZE command failed, aborting.\n");
+		mthca_err(mdev, "SET_ICM_SIZE command returned %d, aborting.\n", err);
 		return err;
 	}
-	if (status) {
-		mthca_err(mdev, "SET_ICM_SIZE returned status 0x%02x, "
-			  "aborting.\n", status);
-		return -EINVAL;
-	}
 
 	mthca_dbg(mdev, "%lld KB of HCA context requires %lld KB aux memory.\n",
 		  (unsigned long long) icm_size >> 10,
@@ -442,14 +403,9 @@ static int mthca_init_icm(struct mthca_dev *mdev,
 		return -ENOMEM;
 	}
 
-	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm, &status);
+	err = mthca_MAP_ICM_AUX(mdev, mdev->fw.arbel.aux_icm);
 	if (err) {
-		mthca_err(mdev, "MAP_ICM_AUX command failed, aborting.\n");
-		goto err_free_aux;
-	}
-	if (status) {
-		mthca_err(mdev, "MAP_ICM_AUX returned status 0x%02x, aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "MAP_ICM_AUX returned %d, aborting.\n", err);
 		goto err_free_aux;
 	}
 
@@ -460,11 +416,11 @@ static int mthca_init_icm(struct mthca_dev *mdev,
 	}
 
 	/* CPU writes to non-reserved MTTs, while HCA might DMA to reserved mtts */
-	mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * MTHCA_MTT_SEG_SIZE,
-					   dma_get_cache_alignment()) / MTHCA_MTT_SEG_SIZE;
+	mdev->limits.reserved_mtts = ALIGN(mdev->limits.reserved_mtts * mdev->limits.mtt_seg_size,
+					   dma_get_cache_alignment()) / mdev->limits.mtt_seg_size;
 
 	mdev->mr_table.mtt_table = mthca_alloc_icm_table(mdev, init_hca->mtt_base,
-							 MTHCA_MTT_SEG_SIZE,
+							 mdev->limits.mtt_seg_size,
 							 mdev->limits.num_mtt_segs,
 							 mdev->limits.reserved_mtts,
 							 1, 0);
@@ -590,7 +546,7 @@ err_unmap_eq:
 	mthca_unmap_eq_icm(mdev);
 
 err_unmap_aux:
-	mthca_UNMAP_ICM_AUX(mdev, &status);
+	mthca_UNMAP_ICM_AUX(mdev);
 
 err_free_aux:
 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
@@ -600,7 +556,6 @@ err_free_aux:
 
 static void mthca_free_icms(struct mthca_dev *mdev)
 {
-	u8 status;
 
 	mthca_free_icm_table(mdev, mdev->mcg_table.table);
 	if (mdev->mthca_flags & MTHCA_FLAG_SRQ)
@@ -613,7 +568,7 @@ static void mthca_free_icms(struct mthca_dev *mdev)
 	mthca_free_icm_table(mdev, mdev->mr_table.mtt_table);
 	mthca_unmap_eq_icm(mdev);
 
-	mthca_UNMAP_ICM_AUX(mdev, &status);
+	mthca_UNMAP_ICM_AUX(mdev);
 	mthca_free_icm(mdev, mdev->fw.arbel.aux_icm, 0);
 }
 
@@ -623,43 +578,32 @@ static int mthca_init_arbel(struct mthca_dev *mdev)
 	struct mthca_profile        profile;
 	struct mthca_init_hca_param init_hca;
 	s64 icm_size;
-	u8 status;
 	int err;
 
-	err = mthca_QUERY_FW(mdev, &status);
+	err = mthca_QUERY_FW(mdev);
 	if (err) {
-		mthca_err(mdev, "QUERY_FW command failed, aborting.\n");
+		mthca_err(mdev, "QUERY_FW command failed %d, aborting.\n", err);
 		return err;
 	}
-	if (status) {
-		mthca_err(mdev, "QUERY_FW returned status 0x%02x, "
-			  "aborting.\n", status);
-		return -EINVAL;
-	}
 
-	err = mthca_ENABLE_LAM(mdev, &status);
-	if (err) {
-		mthca_err(mdev, "ENABLE_LAM command failed, aborting.\n");
-		return err;
-	}
-	if (status == MTHCA_CMD_STAT_LAM_NOT_PRE) {
+	err = mthca_ENABLE_LAM(mdev);
+	if (err == -EAGAIN) {
 		mthca_dbg(mdev, "No HCA-attached memory (running in MemFree mode)\n");
 		mdev->mthca_flags |= MTHCA_FLAG_NO_LAM;
-	} else if (status) {
-		mthca_err(mdev, "ENABLE_LAM returned status 0x%02x, "
-			  "aborting.\n", status);
-		return -EINVAL;
+	} else if (err) {
+		mthca_err(mdev, "ENABLE_LAM returned %d, aborting.\n", err);
+		return err;
 	}
 
 	err = mthca_load_fw(mdev);
 	if (err) {
-		mthca_err(mdev, "Failed to start FW, aborting.\n");
+		mthca_err(mdev, "Loading FW returned %d, aborting.\n", err);
 		goto err_disable;
 	}
 
 	err = mthca_dev_lim(mdev, &dev_lim);
 	if (err) {
-		mthca_err(mdev, "QUERY_DEV_LIM command failed, aborting.\n");
+		mthca_err(mdev, "QUERY_DEV_LIM returned %d, aborting.\n", err);
 		goto err_stop_fw;
 	}
 
@@ -679,15 +623,9 @@ static int mthca_init_arbel(struct mthca_dev *mdev)
 	if (err)
 		goto err_stop_fw;
 
-	err = mthca_INIT_HCA(mdev, &init_hca, &status);
+	err = mthca_INIT_HCA(mdev, &init_hca);
 	if (err) {
-		mthca_err(mdev, "INIT_HCA command failed, aborting.\n");
-		goto err_free_icm;
-	}
-	if (status) {
-		mthca_err(mdev, "INIT_HCA returned status 0x%02x, "
-			  "aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "INIT_HCA command returned %d, aborting.\n", err);
 		goto err_free_icm;
 	}
 
@@ -697,37 +635,34 @@ err_free_icm:
 	mthca_free_icms(mdev);
 
 err_stop_fw:
-	mthca_UNMAP_FA(mdev, &status);
+	mthca_UNMAP_FA(mdev);
 	mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
 
 err_disable:
 	if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
-		mthca_DISABLE_LAM(mdev, &status);
+		mthca_DISABLE_LAM(mdev);
 
 	return err;
 }
 
 static void mthca_close_hca(struct mthca_dev *mdev)
 {
-	u8 status;
-
-	mthca_CLOSE_HCA(mdev, 0, &status);
+	mthca_CLOSE_HCA(mdev, 0);
 
 	if (mthca_is_memfree(mdev)) {
 		mthca_free_icms(mdev);
 
-		mthca_UNMAP_FA(mdev, &status);
+		mthca_UNMAP_FA(mdev);
 		mthca_free_icm(mdev, mdev->fw.arbel.fw_icm, 0);
 
 		if (!(mdev->mthca_flags & MTHCA_FLAG_NO_LAM))
-			mthca_DISABLE_LAM(mdev, &status);
+			mthca_DISABLE_LAM(mdev);
 	} else
-		mthca_SYS_DIS(mdev, &status);
+		mthca_SYS_DIS(mdev);
 }
 
 static int mthca_init_hca(struct mthca_dev *mdev)
 {
-	u8 status;
 	int err;
 	struct mthca_adapter adapter;
 
@@ -739,15 +674,9 @@ static int mthca_init_hca(struct mthca_dev *mdev)
 	if (err)
 		return err;
 
-	err = mthca_QUERY_ADAPTER(mdev, &adapter, &status);
+	err = mthca_QUERY_ADAPTER(mdev, &adapter);
 	if (err) {
-		mthca_err(mdev, "QUERY_ADAPTER command failed, aborting.\n");
-		goto err_close;
-	}
-	if (status) {
-		mthca_err(mdev, "QUERY_ADAPTER returned status 0x%02x, "
-			  "aborting.\n", status);
-		err = -EINVAL;
+		mthca_err(mdev, "QUERY_ADAPTER command returned %d, aborting.\n", err);
 		goto err_close;
 	}
 
@@ -766,7 +695,6 @@ err_close:
 static int mthca_setup_hca(struct mthca_dev *dev)
 {
 	int err;
-	u8 status;
 
 	MTHCA_INIT_DOORBELL_LOCK(&dev->doorbell_lock);
 
@@ -784,7 +712,7 @@ static int mthca_setup_hca(struct mthca_dev *dev)
 		goto err_uar_table_free;
 	}
 
-	dev->kar = ioremap(dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
+	dev->kar = ioremap((phys_addr_t) dev->driver_uar.pfn << PAGE_SHIFT, PAGE_SIZE);
 	if (!dev->kar) {
 		mthca_err(dev, "Couldn't map kernel access region, "
 			  "aborting.\n");
@@ -827,8 +755,8 @@ static int mthca_setup_hca(struct mthca_dev *dev)
 		goto err_eq_table_free;
 	}
 
-	err = mthca_NOP(dev, &status);
-	if (err || status) {
+	err = mthca_NOP(dev);
+	if (err) {
 		if (dev->mthca_flags & MTHCA_FLAG_MSI_X) {
 			mthca_warn(dev, "NOP command failed to generate interrupt "
 				   "(IRQ %d).\n",
@@ -921,58 +849,6 @@ err_uar_table_free:
 	return err;
 }
 
-static int mthca_request_regions(struct pci_dev *pdev, int ddr_hidden)
-{
-	int err;
-
-	/*
-	 * We can't just use pci_request_regions() because the MSI-X
-	 * table is right in the middle of the first BAR.  If we did
-	 * pci_request_region and grab all of the first BAR, then
-	 * setting up MSI-X would fail, since the PCI core wants to do
-	 * request_mem_region on the MSI-X vector table.
-	 *
-	 * So just request what we need right now, and request any
-	 * other regions we need when setting up EQs.
-	 */
-	if (!request_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-				MTHCA_HCR_SIZE, DRV_NAME))
-		return -EBUSY;
-
-	err = pci_request_region(pdev, 2, DRV_NAME);
-	if (err)
-		goto err_bar2_failed;
-
-	if (!ddr_hidden) {
-		err = pci_request_region(pdev, 4, DRV_NAME);
-		if (err)
-			goto err_bar4_failed;
-	}
-
-	return 0;
-
-err_bar4_failed:
-	pci_release_region(pdev, 2);
-
-err_bar2_failed:
-	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-			   MTHCA_HCR_SIZE);
-
-	return err;
-}
-
-static void mthca_release_regions(struct pci_dev *pdev,
-				  int ddr_hidden)
-{
-	if (!ddr_hidden)
-		pci_release_region(pdev, 4);
-
-	pci_release_region(pdev, 2);
-
-	release_mem_region(pci_resource_start(pdev, 0) + MTHCA_HCR_BASE,
-			   MTHCA_HCR_SIZE);
-}
-
 static int mthca_enable_msi_x(struct mthca_dev *mdev)
 {
 	struct msix_entry entries[3];
@@ -982,13 +858,9 @@ static int mthca_enable_msi_x(struct mthca_dev *mdev)
 	entries[1].entry = 1;
 	entries[2].entry = 2;
 
-	err = pci_enable_msix(mdev->pdev, entries, ARRAY_SIZE(entries));
-	if (err) {
-		if (err > 0)
-			mthca_info(mdev, "Only %d MSI-X vectors available, "
-				   "not using MSI-X\n", err);
+	err = pci_enable_msix_exact(mdev->pdev, entries, ARRAY_SIZE(entries));
+	if (err)
 		return err;
-	}
 
 	mdev->eq_table.eq[MTHCA_EQ_COMP ].msi_x_vector = entries[0].vector;
 	mdev->eq_table.eq[MTHCA_EQ_ASYNC].msi_x_vector = entries[1].vector;
@@ -1059,7 +931,7 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 	if (!(pci_resource_flags(pdev, 4) & IORESOURCE_MEM))
 		ddr_hidden = 1;
 
-	err = mthca_request_regions(pdev, ddr_hidden);
+	err = pci_request_regions(pdev, DRV_NAME);
 	if (err) {
 		dev_err(&pdev->dev, "Cannot obtain PCI resources, "
 			"aborting.\n");
@@ -1068,20 +940,20 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 
 	pci_set_master(pdev);
 
-	err = pci_set_dma_mask(pdev, DMA_64BIT_MASK);
+	err = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
 	if (err) {
 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit PCI DMA mask.\n");
-		err = pci_set_dma_mask(pdev, DMA_32BIT_MASK);
+		err = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
 		if (err) {
 			dev_err(&pdev->dev, "Can't set PCI DMA mask, aborting.\n");
 			goto err_free_res;
 		}
 	}
-	err = pci_set_consistent_dma_mask(pdev, DMA_64BIT_MASK);
+	err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
 	if (err) {
 		dev_warn(&pdev->dev, "Warning: couldn't set 64-bit "
 			 "consistent PCI DMA mask.\n");
-		err = pci_set_consistent_dma_mask(pdev, DMA_32BIT_MASK);
+		err = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
 		if (err) {
 			dev_err(&pdev->dev, "Can't set consistent PCI DMA mask, "
 				"aborting.\n");
@@ -1089,6 +961,9 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 		}
 	}
 
+	/* We can handle large RDMA requests, so allow larger segments. */
+	dma_set_max_seg_size(&pdev->dev, 1024 * 1024 * 1024);
+
 	mdev = (struct mthca_dev *) ib_alloc_device(sizeof *mdev);
 	if (!mdev) {
 		dev_err(&pdev->dev, "Device struct alloc failed, "
@@ -1163,6 +1038,8 @@ static int __mthca_init_one(struct pci_dev *pdev, int hca_type)
 	pci_set_drvdata(pdev, mdev);
 	mdev->hca_type = hca_type;
 
+	mdev->active = true;
+
 	return 0;
 
 err_unregister:
@@ -1196,7 +1073,7 @@ err_free_dev:
 	ib_dealloc_device(&mdev->ib_dev);
 
 err_free_res:
-	mthca_release_regions(pdev, ddr_hidden);
+	pci_release_regions(pdev);
 
 err_disable_pdev:
 	pci_disable_device(pdev);
@@ -1207,7 +1084,6 @@ err_disable_pdev:
 static void __mthca_remove_one(struct pci_dev *pdev)
 {
 	struct mthca_dev *mdev = pci_get_drvdata(pdev);
-	u8 status;
 	int p;
 
 	if (mdev) {
@@ -1215,7 +1091,7 @@ static void __mthca_remove_one(struct pci_dev *pdev)
 		mthca_unregister_device(mdev);
 
 		for (p = 1; p <= mdev->limits.num_ports; ++p)
-			mthca_CLOSE_IB(mdev, p, &status);
+			mthca_CLOSE_IB(mdev, p);
 
 		mthca_cleanup_mcg_table(mdev);
 		mthca_cleanup_av_table(mdev);
@@ -1240,8 +1116,7 @@ static void __mthca_remove_one(struct pci_dev *pdev)
 			pci_disable_msix(pdev);
 
 		ib_dealloc_device(&mdev->ib_dev);
-		mthca_release_regions(pdev, mdev->mthca_flags &
-				      MTHCA_FLAG_DDR_HIDDEN);
+		pci_release_regions(pdev);
 		pci_disable_device(pdev);
 		pci_set_drvdata(pdev, NULL);
 	}
@@ -1260,18 +1135,13 @@ int __mthca_restart_one(struct pci_dev *pdev)
 	return __mthca_init_one(pdev, hca_type);
 }
 
-static int __devinit mthca_init_one(struct pci_dev *pdev,
-				    const struct pci_device_id *id)
+static int mthca_init_one(struct pci_dev *pdev, const struct pci_device_id *id)
 {
-	static int mthca_version_printed = 0;
 	int ret;
 
 	mutex_lock(&mthca_device_mutex);
 
-	if (!mthca_version_printed) {
-		printk(KERN_INFO "%s", mthca_version);
-		++mthca_version_printed;
-	}
+	printk_once(KERN_INFO "%s", mthca_version);
 
 	if (id->driver_data >= ARRAY_SIZE(mthca_hca_table)) {
 		printk(KERN_ERR PFX "%s has invalid driver data %lx\n",
@@ -1287,7 +1157,7 @@ static int __devinit mthca_init_one(struct pci_dev *pdev,
 	return ret;
 }
 
-static void __devexit mthca_remove_one(struct pci_dev *pdev)
+static void mthca_remove_one(struct pci_dev *pdev)
 {
 	mutex_lock(&mthca_device_mutex);
 	__mthca_remove_one(pdev);
@@ -1324,7 +1194,7 @@ static struct pci_driver mthca_driver = {
 	.name		= DRV_NAME,
 	.id_table	= mthca_pci_table,
 	.probe		= mthca_init_one,
-	.remove		= __devexit_p(mthca_remove_one)
+	.remove		= mthca_remove_one,
 };
 
 static void __init __mthca_check_profile_val(const char *name, int *pval,
@@ -1368,6 +1238,12 @@ static void __init mthca_validate_profile(void)
 		printk(KERN_WARNING PFX "Corrected fmr_reserved_mtts to %d.\n",
 		       hca_profile.fmr_reserved_mtts);
 	}
+
+	if ((log_mtts_per_seg < 1) || (log_mtts_per_seg > 5)) {
+		printk(KERN_WARNING PFX "bad log_mtts_per_seg (%d). Using default - %d\n",
+		       log_mtts_per_seg, ilog2(MTHCA_MTT_SEG_SIZE / 8));
+		log_mtts_per_seg = ilog2(MTHCA_MTT_SEG_SIZE / 8);
+	}
 }
 
 static int __init mthca_init(void)
diff --git a/drivers/infiniband/hw/mthca/mthca_mcg.c b/drivers/infiniband/hw/mthca/mthca_mcg.c
index 3f5f9487920..6304ae8f4a6 100644
--- a/drivers/infiniband/hw/mthca/mthca_mcg.c
+++ b/drivers/infiniband/hw/mthca/mthca_mcg.c
@@ -31,7 +31,7 @@
  */
 
 #include <linux/string.h>
-#include <linux/slab.h>
+#include <linux/gfp.h>
 
 #include "mthca_dev.h"
 #include "mthca_cmd.h"
@@ -68,7 +68,6 @@ static int find_mgm(struct mthca_dev *dev,
 	struct mthca_mgm *mgm = mgm_mailbox->buf;
 	u8 *mgid;
 	int err;
-	u8 status;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox))
@@ -77,38 +76,22 @@ static int find_mgm(struct mthca_dev *dev,
 
 	memcpy(mgid, gid, 16);
 
-	err = mthca_MGID_HASH(dev, mailbox, hash, &status);
-	if (err)
-		goto out;
-	if (status) {
-		mthca_err(dev, "MGID_HASH returned status %02x\n", status);
-		err = -EINVAL;
+	err = mthca_MGID_HASH(dev, mailbox, hash);
+	if (err) {
+		mthca_err(dev, "MGID_HASH failed (%d)\n", err);
 		goto out;
 	}
 
 	if (0)
-		mthca_dbg(dev, "Hash for %04x:%04x:%04x:%04x:"
-			  "%04x:%04x:%04x:%04x is %04x\n",
-			  be16_to_cpu(((__be16 *) gid)[0]),
-			  be16_to_cpu(((__be16 *) gid)[1]),
-			  be16_to_cpu(((__be16 *) gid)[2]),
-			  be16_to_cpu(((__be16 *) gid)[3]),
-			  be16_to_cpu(((__be16 *) gid)[4]),
-			  be16_to_cpu(((__be16 *) gid)[5]),
-			  be16_to_cpu(((__be16 *) gid)[6]),
-			  be16_to_cpu(((__be16 *) gid)[7]),
-			  *hash);
+		mthca_dbg(dev, "Hash for %pI6 is %04x\n", gid, *hash);
 
 	*index = *hash;
 	*prev  = -1;
 
 	do {
-		err = mthca_READ_MGM(dev, *index, mgm_mailbox, &status);
-		if (err)
-			goto out;
-		if (status) {
-			mthca_err(dev, "READ_MGM returned status %02x\n", status);
-			err = -EINVAL;
+		err = mthca_READ_MGM(dev, *index, mgm_mailbox);
+		if (err) {
+			mthca_err(dev, "READ_MGM failed (%d)\n", err);
 			goto out;
 		}
 
@@ -144,7 +127,6 @@ int mthca_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	int link = 0;
 	int i;
 	int err;
-	u8 status;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox))
@@ -170,12 +152,9 @@ int mthca_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 			goto out;
 		}
 
-		err = mthca_READ_MGM(dev, index, mailbox, &status);
-		if (err)
-			goto out;
-		if (status) {
-			mthca_err(dev, "READ_MGM returned status %02x\n", status);
-			err = -EINVAL;
+		err = mthca_READ_MGM(dev, index, mailbox);
+		if (err) {
+			mthca_err(dev, "READ_MGM failed (%d)\n", err);
 			goto out;
 		}
 		memset(mgm, 0, sizeof *mgm);
@@ -199,11 +178,9 @@ int mthca_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 		goto out;
 	}
 
-	err = mthca_WRITE_MGM(dev, index, mailbox, &status);
-	if (err)
-		goto out;
-	if (status) {
-		mthca_err(dev, "WRITE_MGM returned status %02x\n", status);
+	err = mthca_WRITE_MGM(dev, index, mailbox);
+	if (err) {
+		mthca_err(dev, "WRITE_MGM failed %d\n", err);
 		err = -EINVAL;
 		goto out;
 	}
@@ -211,24 +188,17 @@ int mthca_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	if (!link)
 		goto out;
 
-	err = mthca_READ_MGM(dev, prev, mailbox, &status);
-	if (err)
-		goto out;
-	if (status) {
-		mthca_err(dev, "READ_MGM returned status %02x\n", status);
-		err = -EINVAL;
+	err = mthca_READ_MGM(dev, prev, mailbox);
+	if (err) {
+		mthca_err(dev, "READ_MGM failed %d\n", err);
 		goto out;
 	}
 
 	mgm->next_gid_index = cpu_to_be32(index << 6);
 
-	err = mthca_WRITE_MGM(dev, prev, mailbox, &status);
+	err = mthca_WRITE_MGM(dev, prev, mailbox);
 	if (err)
-		goto out;
-	if (status) {
-		mthca_err(dev, "WRITE_MGM returned status %02x\n", status);
-		err = -EINVAL;
-	}
+		mthca_err(dev, "WRITE_MGM returned %d\n", err);
 
  out:
 	if (err && link && index != -1) {
@@ -250,7 +220,6 @@ int mthca_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	int prev, index;
 	int i, loc;
 	int err;
-	u8 status;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox))
@@ -264,16 +233,7 @@ int mthca_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 		goto out;
 
 	if (index == -1) {
-		mthca_err(dev, "MGID %04x:%04x:%04x:%04x:%04x:%04x:%04x:%04x "
-			  "not found\n",
-			  be16_to_cpu(((__be16 *) gid->raw)[0]),
-			  be16_to_cpu(((__be16 *) gid->raw)[1]),
-			  be16_to_cpu(((__be16 *) gid->raw)[2]),
-			  be16_to_cpu(((__be16 *) gid->raw)[3]),
-			  be16_to_cpu(((__be16 *) gid->raw)[4]),
-			  be16_to_cpu(((__be16 *) gid->raw)[5]),
-			  be16_to_cpu(((__be16 *) gid->raw)[6]),
-			  be16_to_cpu(((__be16 *) gid->raw)[7]));
+		mthca_err(dev, "MGID %pI6 not found\n", gid->raw);
 		err = -EINVAL;
 		goto out;
 	}
@@ -294,12 +254,9 @@ int mthca_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	mgm->qp[loc]   = mgm->qp[i - 1];
 	mgm->qp[i - 1] = 0;
 
-	err = mthca_WRITE_MGM(dev, index, mailbox, &status);
-	if (err)
-		goto out;
-	if (status) {
-		mthca_err(dev, "WRITE_MGM returned status %02x\n", status);
-		err = -EINVAL;
+	err = mthca_WRITE_MGM(dev, index, mailbox);
+	if (err) {
+		mthca_err(dev, "WRITE_MGM returned %d\n", err);
 		goto out;
 	}
 
@@ -311,24 +268,17 @@ int mthca_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 		int amgm_index_to_free = be32_to_cpu(mgm->next_gid_index) >> 6;
 		if (amgm_index_to_free) {
 			err = mthca_READ_MGM(dev, amgm_index_to_free,
-					     mailbox, &status);
-			if (err)
-				goto out;
-			if (status) {
-				mthca_err(dev, "READ_MGM returned status %02x\n",
-					  status);
-				err = -EINVAL;
+					     mailbox);
+			if (err) {
+				mthca_err(dev, "READ_MGM returned %d\n", err);
 				goto out;
 			}
 		} else
 			memset(mgm->gid, 0, 16);
 
-		err = mthca_WRITE_MGM(dev, index, mailbox, &status);
-		if (err)
-			goto out;
-		if (status) {
-			mthca_err(dev, "WRITE_MGM returned status %02x\n", status);
-			err = -EINVAL;
+		err = mthca_WRITE_MGM(dev, index, mailbox);
+		if (err) {
+			mthca_err(dev, "WRITE_MGM returned %d\n", err);
 			goto out;
 		}
 		if (amgm_index_to_free) {
@@ -338,23 +288,17 @@ int mthca_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
 	} else {
 		/* Remove entry from AMGM */
 		int curr_next_index = be32_to_cpu(mgm->next_gid_index) >> 6;
-		err = mthca_READ_MGM(dev, prev, mailbox, &status);
-		if (err)
-			goto out;
-		if (status) {
-			mthca_err(dev, "READ_MGM returned status %02x\n", status);
-			err = -EINVAL;
+		err = mthca_READ_MGM(dev, prev, mailbox);
+		if (err) {
+			mthca_err(dev, "READ_MGM returned %d\n", err);
 			goto out;
 		}
 
 		mgm->next_gid_index = cpu_to_be32(curr_next_index << 6);
 
-		err = mthca_WRITE_MGM(dev, prev, mailbox, &status);
-		if (err)
-			goto out;
-		if (status) {
-			mthca_err(dev, "WRITE_MGM returned status %02x\n", status);
-			err = -EINVAL;
+		err = mthca_WRITE_MGM(dev, prev, mailbox);
+		if (err) {
+			mthca_err(dev, "WRITE_MGM returned %d\n", err);
 			goto out;
 		}
 		BUG_ON(index < dev->limits.num_mgms);
diff --git a/drivers/infiniband/hw/mthca/mthca_memfree.c b/drivers/infiniband/hw/mthca/mthca_memfree.c
index 1f7d1a29d2a..7d2e42dd692 100644
--- a/drivers/infiniband/hw/mthca/mthca_memfree.c
+++ b/drivers/infiniband/hw/mthca/mthca_memfree.c
@@ -35,6 +35,7 @@
 #include <linux/mm.h>
 #include <linux/scatterlist.h>
 #include <linux/sched.h>
+#include <linux/slab.h>
 
 #include <asm/page.h>
 
@@ -222,7 +223,6 @@ int mthca_table_get(struct mthca_dev *dev, struct mthca_icm_table *table, int ob
 {
 	int i = (obj & (table->num_obj - 1)) * table->obj_size / MTHCA_TABLE_CHUNK_SIZE;
 	int ret = 0;
-	u8 status;
 
 	mutex_lock(&table->mutex);
 
@@ -239,8 +239,8 @@ int mthca_table_get(struct mthca_dev *dev, struct mthca_icm_table *table, int ob
 		goto out;
 	}
 
-	if (mthca_MAP_ICM(dev, table->icm[i], table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
-			  &status) || status) {
+	if (mthca_MAP_ICM(dev, table->icm[i],
+			  table->virt + i * MTHCA_TABLE_CHUNK_SIZE)) {
 		mthca_free_icm(dev, table->icm[i], table->coherent);
 		table->icm[i] = NULL;
 		ret = -ENOMEM;
@@ -257,7 +257,6 @@ out:
 void mthca_table_put(struct mthca_dev *dev, struct mthca_icm_table *table, int obj)
 {
 	int i;
-	u8 status;
 
 	if (!mthca_is_memfree(dev))
 		return;
@@ -268,8 +267,7 @@ void mthca_table_put(struct mthca_dev *dev, struct mthca_icm_table *table, int o
 
 	if (--table->icm[i]->refcount == 0) {
 		mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
-				MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE,
-				&status);
+				MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE);
 		mthca_free_icm(dev, table->icm[i], table->coherent);
 		table->icm[i] = NULL;
 	}
@@ -365,7 +363,6 @@ struct mthca_icm_table *mthca_alloc_icm_table(struct mthca_dev *dev,
 	int num_icm;
 	unsigned chunk_size;
 	int i;
-	u8 status;
 
 	obj_per_chunk = MTHCA_TABLE_CHUNK_SIZE / obj_size;
 	num_icm = DIV_ROUND_UP(nobj, obj_per_chunk);
@@ -395,8 +392,8 @@ struct mthca_icm_table *mthca_alloc_icm_table(struct mthca_dev *dev,
 						__GFP_NOWARN, use_coherent);
 		if (!table->icm[i])
 			goto err;
-		if (mthca_MAP_ICM(dev, table->icm[i], virt + i * MTHCA_TABLE_CHUNK_SIZE,
-				  &status) || status) {
+		if (mthca_MAP_ICM(dev, table->icm[i],
+				  virt + i * MTHCA_TABLE_CHUNK_SIZE)) {
 			mthca_free_icm(dev, table->icm[i], table->coherent);
 			table->icm[i] = NULL;
 			goto err;
@@ -415,8 +412,7 @@ err:
 	for (i = 0; i < num_icm; ++i)
 		if (table->icm[i]) {
 			mthca_UNMAP_ICM(dev, virt + i * MTHCA_TABLE_CHUNK_SIZE,
-					MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE,
-					&status);
+					MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE);
 			mthca_free_icm(dev, table->icm[i], table->coherent);
 		}
 
@@ -428,13 +424,12 @@ err:
 void mthca_free_icm_table(struct mthca_dev *dev, struct mthca_icm_table *table)
 {
 	int i;
-	u8 status;
 
 	for (i = 0; i < table->num_icm; ++i)
 		if (table->icm[i]) {
-			mthca_UNMAP_ICM(dev, table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
-					MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE,
-					&status);
+			mthca_UNMAP_ICM(dev,
+					table->virt + i * MTHCA_TABLE_CHUNK_SIZE,
+					MTHCA_TABLE_CHUNK_SIZE / MTHCA_ICM_PAGE_SIZE);
 			mthca_free_icm(dev, table->icm[i], table->coherent);
 		}
 
@@ -453,7 +448,6 @@ int mthca_map_user_db(struct mthca_dev *dev, struct mthca_uar *uar,
 {
 	struct page *pages[1];
 	int ret = 0;
-	u8 status;
 	int i;
 
 	if (!mthca_is_memfree(dev))
@@ -493,9 +487,7 @@ int mthca_map_user_db(struct mthca_dev *dev, struct mthca_uar *uar,
 	}
 
 	ret = mthca_MAP_ICM_page(dev, sg_dma_address(&db_tab->page[i].mem),
-				 mthca_uarc_virt(dev, uar, i), &status);
-	if (!ret && status)
-		ret = -EINVAL;
+				 mthca_uarc_virt(dev, uar, i));
 	if (ret) {
 		pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE);
 		put_page(sg_page(&db_tab->page[i].mem));
@@ -556,14 +548,13 @@ void mthca_cleanup_user_db_tab(struct mthca_dev *dev, struct mthca_uar *uar,
 			       struct mthca_user_db_table *db_tab)
 {
 	int i;
-	u8 status;
 
 	if (!mthca_is_memfree(dev))
 		return;
 
 	for (i = 0; i < dev->uar_table.uarc_size / MTHCA_ICM_PAGE_SIZE; ++i) {
 		if (db_tab->page[i].uvirt) {
-			mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, uar, i), 1, &status);
+			mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, uar, i), 1);
 			pci_unmap_sg(dev->pdev, &db_tab->page[i].mem, 1, PCI_DMA_TODEVICE);
 			put_page(sg_page(&db_tab->page[i].mem));
 		}
@@ -580,7 +571,6 @@ int mthca_alloc_db(struct mthca_dev *dev, enum mthca_db_type type,
 	int i, j;
 	struct mthca_db_page *page;
 	int ret = 0;
-	u8 status;
 
 	mutex_lock(&dev->db_tab->mutex);
 
@@ -643,9 +633,7 @@ alloc:
 	memset(page->db_rec, 0, MTHCA_ICM_PAGE_SIZE);
 
 	ret = mthca_MAP_ICM_page(dev, page->mapping,
-				 mthca_uarc_virt(dev, &dev->driver_uar, i), &status);
-	if (!ret && status)
-		ret = -EINVAL;
+				 mthca_uarc_virt(dev, &dev->driver_uar, i));
 	if (ret) {
 		dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE,
 				  page->db_rec, page->mapping);
@@ -677,7 +665,6 @@ void mthca_free_db(struct mthca_dev *dev, int type, int db_index)
 {
 	int i, j;
 	struct mthca_db_page *page;
-	u8 status;
 
 	i = db_index / MTHCA_DB_REC_PER_PAGE;
 	j = db_index % MTHCA_DB_REC_PER_PAGE;
@@ -693,7 +680,7 @@ void mthca_free_db(struct mthca_dev *dev, int type, int db_index)
 
 	if (bitmap_empty(page->used, MTHCA_DB_REC_PER_PAGE) &&
 	    i >= dev->db_tab->max_group1 - 1) {
-		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status);
+		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1);
 
 		dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE,
 				  page->db_rec, page->mapping);
@@ -744,7 +731,6 @@ int mthca_init_db_tab(struct mthca_dev *dev)
 void mthca_cleanup_db_tab(struct mthca_dev *dev)
 {
 	int i;
-	u8 status;
 
 	if (!mthca_is_memfree(dev))
 		return;
@@ -762,7 +748,7 @@ void mthca_cleanup_db_tab(struct mthca_dev *dev)
 		if (!bitmap_empty(dev->db_tab->page[i].used, MTHCA_DB_REC_PER_PAGE))
 			mthca_warn(dev, "Kernel UARC page %d not empty\n", i);
 
-		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1, &status);
+		mthca_UNMAP_ICM(dev, mthca_uarc_virt(dev, &dev->driver_uar, i), 1);
 
 		dma_free_coherent(&dev->pdev->dev, MTHCA_ICM_PAGE_SIZE,
 				  dev->db_tab->page[i].db_rec,
diff --git a/drivers/infiniband/hw/mthca/mthca_mr.c b/drivers/infiniband/hw/mthca/mthca_mr.c
index 8489b1e81c0..ed9a989e501 100644
--- a/drivers/infiniband/hw/mthca/mthca_mr.c
+++ b/drivers/infiniband/hw/mthca/mthca_mr.c
@@ -89,23 +89,26 @@ static u32 mthca_buddy_alloc(struct mthca_buddy *buddy, int order)
 
 	spin_lock(&buddy->lock);
 
-	for (o = order; o <= buddy->max_order; ++o) {
-		m = 1 << (buddy->max_order - o);
-		seg = find_first_bit(buddy->bits[o], m);
-		if (seg < m)
-			goto found;
-	}
+	for (o = order; o <= buddy->max_order; ++o)
+		if (buddy->num_free[o]) {
+			m = 1 << (buddy->max_order - o);
+			seg = find_first_bit(buddy->bits[o], m);
+			if (seg < m)
+				goto found;
+		}
 
 	spin_unlock(&buddy->lock);
 	return -1;
 
  found:
 	clear_bit(seg, buddy->bits[o]);
+	--buddy->num_free[o];
 
 	while (o > order) {
 		--o;
 		seg <<= 1;
 		set_bit(seg ^ 1, buddy->bits[o]);
+		++buddy->num_free[o];
 	}
 
 	spin_unlock(&buddy->lock);
@@ -123,11 +126,13 @@ static void mthca_buddy_free(struct mthca_buddy *buddy, u32 seg, int order)
 
 	while (test_bit(seg ^ 1, buddy->bits[order])) {
 		clear_bit(seg ^ 1, buddy->bits[order]);
+		--buddy->num_free[order];
 		seg >>= 1;
 		++order;
 	}
 
 	set_bit(seg, buddy->bits[order]);
+	++buddy->num_free[order];
 
 	spin_unlock(&buddy->lock);
 }
@@ -141,7 +146,9 @@ static int mthca_buddy_init(struct mthca_buddy *buddy, int max_order)
 
 	buddy->bits = kzalloc((buddy->max_order + 1) * sizeof (long *),
 			      GFP_KERNEL);
-	if (!buddy->bits)
+	buddy->num_free = kcalloc((buddy->max_order + 1), sizeof *buddy->num_free,
+				  GFP_KERNEL);
+	if (!buddy->bits || !buddy->num_free)
 		goto err_out;
 
 	for (i = 0; i <= buddy->max_order; ++i) {
@@ -154,6 +161,7 @@ static int mthca_buddy_init(struct mthca_buddy *buddy, int max_order)
 	}
 
 	set_bit(0, buddy->bits[buddy->max_order]);
+	buddy->num_free[buddy->max_order] = 1;
 
 	return 0;
 
@@ -161,9 +169,10 @@ err_out_free:
 	for (i = 0; i <= buddy->max_order; ++i)
 		kfree(buddy->bits[i]);
 
+err_out:
 	kfree(buddy->bits);
+	kfree(buddy->num_free);
 
-err_out:
 	return -ENOMEM;
 }
 
@@ -175,6 +184,7 @@ static void mthca_buddy_cleanup(struct mthca_buddy *buddy)
 		kfree(buddy->bits[i]);
 
 	kfree(buddy->bits);
+	kfree(buddy->num_free);
 }
 
 static u32 mthca_alloc_mtt_range(struct mthca_dev *dev, int order,
@@ -210,7 +220,7 @@ static struct mthca_mtt *__mthca_alloc_mtt(struct mthca_dev *dev, int size,
 
 	mtt->buddy = buddy;
 	mtt->order = 0;
-	for (i = MTHCA_MTT_SEG_SIZE / 8; i < size; i <<= 1)
+	for (i = dev->limits.mtt_seg_size / 8; i < size; i <<= 1)
 		++mtt->order;
 
 	mtt->first_seg = mthca_alloc_mtt_range(dev, mtt->order, buddy);
@@ -247,7 +257,6 @@ static int __mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
 	struct mthca_mailbox *mailbox;
 	__be64 *mtt_entry;
 	int err = 0;
-	u8 status;
 	int i;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
@@ -257,7 +266,7 @@ static int __mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
 
 	while (list_len > 0) {
 		mtt_entry[0] = cpu_to_be64(dev->mr_table.mtt_base +
-					   mtt->first_seg * MTHCA_MTT_SEG_SIZE +
+					   mtt->first_seg * dev->limits.mtt_seg_size +
 					   start_index * 8);
 		mtt_entry[1] = 0;
 		for (i = 0; i < list_len && i < MTHCA_MAILBOX_SIZE / 8 - 2; ++i)
@@ -271,17 +280,11 @@ static int __mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
 		if (i & 1)
 			mtt_entry[i + 2] = 0;
 
-		err = mthca_WRITE_MTT(dev, mailbox, (i + 1) & ~1, &status);
+		err = mthca_WRITE_MTT(dev, mailbox, (i + 1) & ~1);
 		if (err) {
 			mthca_warn(dev, "WRITE_MTT failed (%d)\n", err);
 			goto out;
 		}
-		if (status) {
-			mthca_warn(dev, "WRITE_MTT returned status 0x%02x\n",
-				   status);
-			err = -EINVAL;
-			goto out;
-		}
 
 		list_len    -= i;
 		start_index += i;
@@ -316,7 +319,7 @@ static void mthca_tavor_write_mtt_seg(struct mthca_dev *dev,
 	u64 __iomem *mtts;
 	int i;
 
-	mtts = dev->mr_table.tavor_fmr.mtt_base + mtt->first_seg * MTHCA_MTT_SEG_SIZE +
+	mtts = dev->mr_table.tavor_fmr.mtt_base + mtt->first_seg * dev->limits.mtt_seg_size +
 		start_index * sizeof (u64);
 	for (i = 0; i < list_len; ++i)
 		mthca_write64_raw(cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT),
@@ -335,17 +338,21 @@ static void mthca_arbel_write_mtt_seg(struct mthca_dev *dev,
 	/* For Arbel, all MTTs must fit in the same page. */
 	BUG_ON(s / PAGE_SIZE != (s + list_len * sizeof(u64) - 1) / PAGE_SIZE);
 	/* Require full segments */
-	BUG_ON(s % MTHCA_MTT_SEG_SIZE);
+	BUG_ON(s % dev->limits.mtt_seg_size);
 
 	mtts = mthca_table_find(dev->mr_table.mtt_table, mtt->first_seg +
-				s / MTHCA_MTT_SEG_SIZE, &dma_handle);
+				s / dev->limits.mtt_seg_size, &dma_handle);
 
 	BUG_ON(!mtts);
 
+	dma_sync_single_for_cpu(&dev->pdev->dev, dma_handle,
+				list_len * sizeof (u64), DMA_TO_DEVICE);
+
 	for (i = 0; i < list_len; ++i)
 		mtts[i] = cpu_to_be64(buffer_list[i] | MTHCA_MTT_FLAG_PRESENT);
 
-	dma_sync_single(&dev->pdev->dev, dma_handle, list_len * sizeof (u64), DMA_TO_DEVICE);
+	dma_sync_single_for_device(&dev->pdev->dev, dma_handle,
+				   list_len * sizeof (u64), DMA_TO_DEVICE);
 }
 
 int mthca_write_mtt(struct mthca_dev *dev, struct mthca_mtt *mtt,
@@ -427,7 +434,6 @@ int mthca_mr_alloc(struct mthca_dev *dev, u32 pd, int buffer_size_shift,
 	u32 key;
 	int i;
 	int err;
-	u8 status;
 
 	WARN_ON(buffer_size_shift >= 32);
 
@@ -469,7 +475,7 @@ int mthca_mr_alloc(struct mthca_dev *dev, u32 pd, int buffer_size_shift,
 	if (mr->mtt)
 		mpt_entry->mtt_seg =
 			cpu_to_be64(dev->mr_table.mtt_base +
-				    mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE);
+				    mr->mtt->first_seg * dev->limits.mtt_seg_size);
 
 	if (0) {
 		mthca_dbg(dev, "Dumping MPT entry %08x:\n", mr->ibmr.lkey);
@@ -483,16 +489,10 @@ int mthca_mr_alloc(struct mthca_dev *dev, u32 pd, int buffer_size_shift,
 	}
 
 	err = mthca_SW2HW_MPT(dev, mailbox,
-			      key & (dev->limits.num_mpts - 1),
-			      &status);
+			      key & (dev->limits.num_mpts - 1));
 	if (err) {
 		mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
 		goto err_out_mailbox;
-	} else if (status) {
-		mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
-			   status);
-		err = -EINVAL;
-		goto err_out_mailbox;
 	}
 
 	mthca_free_mailbox(dev, mailbox);
@@ -553,17 +553,12 @@ static void mthca_free_region(struct mthca_dev *dev, u32 lkey)
 void mthca_free_mr(struct mthca_dev *dev, struct mthca_mr *mr)
 {
 	int err;
-	u8 status;
 
 	err = mthca_HW2SW_MPT(dev, NULL,
 			      key_to_hw_index(dev, mr->ibmr.lkey) &
-			      (dev->limits.num_mpts - 1),
-			      &status);
+			      (dev->limits.num_mpts - 1));
 	if (err)
 		mthca_warn(dev, "HW2SW_MPT failed (%d)\n", err);
-	else if (status)
-		mthca_warn(dev, "HW2SW_MPT returned status 0x%02x\n",
-			   status);
 
 	mthca_free_region(dev, mr->ibmr.lkey);
 	mthca_free_mtt(dev, mr->mtt);
@@ -576,7 +571,6 @@ int mthca_fmr_alloc(struct mthca_dev *dev, u32 pd,
 	struct mthca_mailbox *mailbox;
 	u64 mtt_seg;
 	u32 key, idx;
-	u8 status;
 	int list_len = mr->attr.max_pages;
 	int err = -ENOMEM;
 	int i;
@@ -616,7 +610,7 @@ int mthca_fmr_alloc(struct mthca_dev *dev, u32 pd,
 		goto err_out_table;
 	}
 
-	mtt_seg = mr->mtt->first_seg * MTHCA_MTT_SEG_SIZE;
+	mtt_seg = mr->mtt->first_seg * dev->limits.mtt_seg_size;
 
 	if (mthca_is_memfree(dev)) {
 		mr->mem.arbel.mtts = mthca_table_find(dev->mr_table.mtt_table,
@@ -658,18 +652,11 @@ int mthca_fmr_alloc(struct mthca_dev *dev, u32 pd,
 	}
 
 	err = mthca_SW2HW_MPT(dev, mailbox,
-			      key & (dev->limits.num_mpts - 1),
-			      &status);
+			      key & (dev->limits.num_mpts - 1));
 	if (err) {
 		mthca_warn(dev, "SW2HW_MPT failed (%d)\n", err);
 		goto err_out_mailbox_free;
 	}
-	if (status) {
-		mthca_warn(dev, "SW2HW_MPT returned status 0x%02x\n",
-			   status);
-		err = -EINVAL;
-		goto err_out_mailbox_free;
-	}
 
 	mthca_free_mailbox(dev, mailbox);
 	return 0;
@@ -793,12 +780,15 @@ int mthca_arbel_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
 
 	wmb();
 
+	dma_sync_single_for_cpu(&dev->pdev->dev, fmr->mem.arbel.dma_handle,
+				list_len * sizeof(u64), DMA_TO_DEVICE);
+
 	for (i = 0; i < list_len; ++i)
 		fmr->mem.arbel.mtts[i] = cpu_to_be64(page_list[i] |
 						     MTHCA_MTT_FLAG_PRESENT);
 
-	dma_sync_single(&dev->pdev->dev, fmr->mem.arbel.dma_handle,
-			list_len * sizeof(u64), DMA_TO_DEVICE);
+	dma_sync_single_for_device(&dev->pdev->dev, fmr->mem.arbel.dma_handle,
+				   list_len * sizeof(u64), DMA_TO_DEVICE);
 
 	fmr->mem.arbel.mpt->key    = cpu_to_be32(key);
 	fmr->mem.arbel.mpt->lkey   = cpu_to_be32(key);
@@ -836,7 +826,7 @@ void mthca_arbel_fmr_unmap(struct mthca_dev *dev, struct mthca_fmr *fmr)
 
 int mthca_init_mr_table(struct mthca_dev *dev)
 {
-	unsigned long addr;
+	phys_addr_t addr;
 	int mpts, mtts, err, i;
 
 	err = mthca_alloc_init(&dev->mr_table.mpt_alloc,
@@ -898,7 +888,7 @@ int mthca_init_mr_table(struct mthca_dev *dev)
 			 dev->mr_table.mtt_base);
 
 		dev->mr_table.tavor_fmr.mtt_base =
-			ioremap(addr, mtts * MTHCA_MTT_SEG_SIZE);
+			ioremap(addr, mtts * dev->limits.mtt_seg_size);
 		if (!dev->mr_table.tavor_fmr.mtt_base) {
 			mthca_warn(dev, "MTT ioremap for FMR failed.\n");
 			err = -ENOMEM;
diff --git a/drivers/infiniband/hw/mthca/mthca_profile.c b/drivers/infiniband/hw/mthca/mthca_profile.c
index d168c254061..8edb28a9a0e 100644
--- a/drivers/infiniband/hw/mthca/mthca_profile.c
+++ b/drivers/infiniband/hw/mthca/mthca_profile.c
@@ -94,7 +94,7 @@ s64 mthca_make_profile(struct mthca_dev *dev,
 	profile[MTHCA_RES_RDB].size  = MTHCA_RDB_ENTRY_SIZE;
 	profile[MTHCA_RES_MCG].size  = MTHCA_MGM_ENTRY_SIZE;
 	profile[MTHCA_RES_MPT].size  = dev_lim->mpt_entry_sz;
-	profile[MTHCA_RES_MTT].size  = MTHCA_MTT_SEG_SIZE;
+	profile[MTHCA_RES_MTT].size  = dev->limits.mtt_seg_size;
 	profile[MTHCA_RES_UAR].size  = dev_lim->uar_scratch_entry_sz;
 	profile[MTHCA_RES_UDAV].size = MTHCA_AV_SIZE;
 	profile[MTHCA_RES_UARC].size = request->uarc_size;
@@ -232,7 +232,7 @@ s64 mthca_make_profile(struct mthca_dev *dev,
 			dev->limits.num_mtt_segs = profile[i].num;
 			dev->mr_table.mtt_base   = profile[i].start;
 			init_hca->mtt_base       = profile[i].start;
-			init_hca->mtt_seg_sz     = ffs(MTHCA_MTT_SEG_SIZE) - 7;
+			init_hca->mtt_seg_sz     = ffs(dev->limits.mtt_seg_size) - 7;
 			break;
 		case MTHCA_RES_UAR:
 			dev->limits.num_uars       = profile[i].num;
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.c b/drivers/infiniband/hw/mthca/mthca_provider.c
index 87ad889e367..415f8e1a54d 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.c
+++ b/drivers/infiniband/hw/mthca/mthca_provider.c
@@ -39,7 +39,10 @@
 #include <rdma/ib_user_verbs.h>
 
 #include <linux/sched.h>
+#include <linux/slab.h>
+#include <linux/stat.h>
 #include <linux/mm.h>
+#include <linux/export.h>
 
 #include "mthca_dev.h"
 #include "mthca_cmd.h"
@@ -62,8 +65,6 @@ static int mthca_query_device(struct ib_device *ibdev,
 	int err = -ENOMEM;
 	struct mthca_dev *mdev = to_mdev(ibdev);
 
-	u8 status;
-
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
 	if (!in_mad || !out_mad)
@@ -77,14 +78,9 @@ static int mthca_query_device(struct ib_device *ibdev,
 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 
 	err = mthca_MAD_IFC(mdev, 1, 1,
-			    1, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	props->device_cap_flags    = mdev->device_cap_flags;
 	props->vendor_id           = be32_to_cpup((__be32 *) (out_mad->data + 36)) &
@@ -140,7 +136,6 @@ static int mthca_query_port(struct ib_device *ibdev,
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
-	u8 status;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -154,14 +149,9 @@ static int mthca_query_port(struct ib_device *ibdev,
 	in_mad->attr_mod = cpu_to_be32(port);
 
 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
-			    port, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    port, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	props->lid               = be16_to_cpup((__be16 *) (out_mad->data + 16));
 	props->lmc               = out_mad->data[34] & 0x7;
@@ -213,7 +203,6 @@ static int mthca_modify_port(struct ib_device *ibdev,
 	struct mthca_set_ib_param set_ib;
 	struct ib_port_attr attr;
 	int err;
-	u8 status;
 
 	if (mutex_lock_interruptible(&to_mdev(ibdev)->cap_mask_mutex))
 		return -ERESTARTSYS;
@@ -228,14 +217,9 @@ static int mthca_modify_port(struct ib_device *ibdev,
 	set_ib.cap_mask = (attr.port_cap_flags | props->set_port_cap_mask) &
 		~props->clr_port_cap_mask;
 
-	err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port, &status);
+	err = mthca_SET_IB(to_mdev(ibdev), &set_ib, port);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
-
 out:
 	mutex_unlock(&to_mdev(ibdev)->cap_mask_mutex);
 	return err;
@@ -247,7 +231,6 @@ static int mthca_query_pkey(struct ib_device *ibdev,
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
-	u8 status;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -259,14 +242,9 @@ static int mthca_query_pkey(struct ib_device *ibdev,
 	in_mad->attr_mod = cpu_to_be32(index / 32);
 
 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
-			    port, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    port, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	*pkey = be16_to_cpu(((__be16 *) out_mad->data)[index % 32]);
 
@@ -282,7 +260,6 @@ static int mthca_query_gid(struct ib_device *ibdev, u8 port,
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
-	u8 status;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -294,14 +271,9 @@ static int mthca_query_gid(struct ib_device *ibdev, u8 port,
 	in_mad->attr_mod = cpu_to_be32(port);
 
 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
-			    port, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    port, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	memcpy(gid->raw, out_mad->data + 8, 8);
 
@@ -310,14 +282,9 @@ static int mthca_query_gid(struct ib_device *ibdev, u8 port,
 	in_mad->attr_mod = cpu_to_be32(index / 8);
 
 	err = mthca_MAD_IFC(to_mdev(ibdev), 1, 1,
-			    port, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    port, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	memcpy(gid->raw + 8, out_mad->data + (index % 8) * 8, 8);
 
@@ -334,6 +301,9 @@ static struct ib_ucontext *mthca_alloc_ucontext(struct ib_device *ibdev,
 	struct mthca_ucontext           *context;
 	int                              err;
 
+	if (!(to_mdev(ibdev)->active))
+		return ERR_PTR(-EAGAIN);
+
 	memset(&uresp, 0, sizeof uresp);
 
 	uresp.qp_tab_size = to_mdev(ibdev)->limits.num_qps;
@@ -470,6 +440,9 @@ static struct ib_srq *mthca_create_srq(struct ib_pd *pd,
 	struct mthca_srq *srq;
 	int err;
 
+	if (init_attr->srq_type != IB_SRQT_BASIC)
+		return ERR_PTR(-ENOSYS);
+
 	srq = kmalloc(sizeof *srq, GFP_KERNEL);
 	if (!srq)
 		return ERR_PTR(-ENOMEM);
@@ -722,6 +695,7 @@ static struct ib_cq *mthca_create_cq(struct ib_device *ibdev, int entries,
 
 	if (context && ib_copy_to_udata(udata, &cq->cqn, sizeof (__u32))) {
 		mthca_free_cq(to_mdev(ibdev), cq);
+		err = -EFAULT;
 		goto err_free;
 	}
 
@@ -796,7 +770,6 @@ static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *uda
 	struct mthca_cq *cq = to_mcq(ibcq);
 	struct mthca_resize_cq ucmd;
 	u32 lkey;
-	u8 status;
 	int ret;
 
 	if (entries < 1 || entries > dev->limits.max_cqes)
@@ -823,9 +796,7 @@ static int mthca_resize_cq(struct ib_cq *ibcq, int entries, struct ib_udata *uda
 		lkey = ucmd.lkey;
 	}
 
-	ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries), &status);
-	if (status)
-		ret = -EINVAL;
+	ret = mthca_RESIZE_CQ(dev, cq->cqn, lkey, ilog2(entries));
 
 	if (ret) {
 		if (cq->resize_buf) {
@@ -1006,12 +977,12 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				       u64 virt, int acc, struct ib_udata *udata)
 {
 	struct mthca_dev *dev = to_mdev(pd->device);
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct mthca_mr *mr;
 	struct mthca_reg_mr ucmd;
 	u64 *pages;
 	int shift, n, len;
-	int i, j, k;
+	int i, k, entry;
 	int err = 0;
 	int write_mtt_size;
 
@@ -1039,10 +1010,7 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	}
 
 	shift = ffs(mr->umem->page_size) - 1;
-
-	n = 0;
-	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
-		n += chunk->nents;
+	n = mr->umem->nmap;
 
 	mr->mtt = mthca_alloc_mtt(dev, n);
 	if (IS_ERR(mr->mtt)) {
@@ -1060,25 +1028,24 @@ static struct ib_mr *mthca_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	write_mtt_size = min(mthca_write_mtt_size(dev), (int) (PAGE_SIZE / sizeof *pages));
 
-	list_for_each_entry(chunk, &mr->umem->chunk_list, list)
-		for (j = 0; j < chunk->nmap; ++j) {
-			len = sg_dma_len(&chunk->page_list[j]) >> shift;
-			for (k = 0; k < len; ++k) {
-				pages[i++] = sg_dma_address(&chunk->page_list[j]) +
-					mr->umem->page_size * k;
-				/*
-				 * Be friendly to write_mtt and pass it chunks
-				 * of appropriate size.
-				 */
-				if (i == write_mtt_size) {
-					err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
-					if (err)
-						goto mtt_done;
-					n += i;
-					i = 0;
-				}
+	for_each_sg(mr->umem->sg_head.sgl, sg, mr->umem->nmap, entry) {
+		len = sg_dma_len(sg) >> shift;
+		for (k = 0; k < len; ++k) {
+			pages[i++] = sg_dma_address(sg) +
+				mr->umem->page_size * k;
+			/*
+			 * Be friendly to write_mtt and pass it chunks
+			 * of appropriate size.
+			 */
+			if (i == write_mtt_size) {
+				err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
+				if (err)
+					goto mtt_done;
+				n += i;
+				i = 0;
 			}
 		}
+	}
 
 	if (i)
 		err = mthca_write_mtt(dev, mr->mtt, n, pages, i);
@@ -1157,7 +1124,6 @@ static int mthca_unmap_fmr(struct list_head *fmr_list)
 {
 	struct ib_fmr *fmr;
 	int err;
-	u8 status;
 	struct mthca_dev *mdev = NULL;
 
 	list_for_each_entry(fmr, fmr_list, list) {
@@ -1178,12 +1144,8 @@ static int mthca_unmap_fmr(struct list_head *fmr_list)
 		list_for_each_entry(fmr, fmr_list, list)
 			mthca_tavor_fmr_unmap(mdev, to_mfmr(fmr));
 
-	err = mthca_SYNC_TPT(mdev, &status);
-	if (err)
-		return err;
-	if (status)
-		return -EINVAL;
-	return 0;
+	err = mthca_SYNC_TPT(mdev);
+	return err;
 }
 
 static ssize_t show_rev(struct device *device, struct device_attribute *attr,
@@ -1249,7 +1211,6 @@ static int mthca_init_node_data(struct mthca_dev *dev)
 	struct ib_smp *in_mad  = NULL;
 	struct ib_smp *out_mad = NULL;
 	int err = -ENOMEM;
-	u8 status;
 
 	in_mad  = kzalloc(sizeof *in_mad, GFP_KERNEL);
 	out_mad = kmalloc(sizeof *out_mad, GFP_KERNEL);
@@ -1260,28 +1221,18 @@ static int mthca_init_node_data(struct mthca_dev *dev)
 	in_mad->attr_id = IB_SMP_ATTR_NODE_DESC;
 
 	err = mthca_MAD_IFC(dev, 1, 1,
-			    1, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	memcpy(dev->ib_dev.node_desc, out_mad->data, 64);
 
 	in_mad->attr_id = IB_SMP_ATTR_NODE_INFO;
 
 	err = mthca_MAD_IFC(dev, 1, 1,
-			    1, NULL, NULL, in_mad, out_mad,
-			    &status);
+			    1, NULL, NULL, in_mad, out_mad);
 	if (err)
 		goto out;
-	if (status) {
-		err = -EINVAL;
-		goto out;
-	}
 
 	if (mthca_is_memfree(dev))
 		dev->rev_id = be32_to_cpup((__be32 *) (out_mad->data + 32));
@@ -1399,7 +1350,7 @@ int mthca_register_device(struct mthca_dev *dev)
 
 	mutex_init(&dev->cap_mask_mutex);
 
-	ret = ib_register_device(&dev->ib_dev);
+	ret = ib_register_device(&dev->ib_dev, NULL);
 	if (ret)
 		return ret;
 
diff --git a/drivers/infiniband/hw/mthca/mthca_provider.h b/drivers/infiniband/hw/mthca/mthca_provider.h
index c621f8794b8..596acc45569 100644
--- a/drivers/infiniband/hw/mthca/mthca_provider.h
+++ b/drivers/infiniband/hw/mthca/mthca_provider.h
@@ -46,7 +46,7 @@
 
 struct mthca_buf_list {
 	void *buf;
-	DECLARE_PCI_UNMAP_ADDR(mapping)
+	DEFINE_DMA_UNMAP_ADDR(mapping);
 };
 
 union mthca_buf {
@@ -113,6 +113,7 @@ struct mthca_eq {
 	int                    nent;
 	struct mthca_buf_list *page_list;
 	struct mthca_mr        mr;
+	char		       irq_name[IB_DEVICE_NAME_MAX];
 };
 
 struct mthca_av;
diff --git a/drivers/infiniband/hw/mthca/mthca_qp.c b/drivers/infiniband/hw/mthca/mthca_qp.c
index f5081bfde6d..e354b2f04ad 100644
--- a/drivers/infiniband/hw/mthca/mthca_qp.c
+++ b/drivers/infiniband/hw/mthca/mthca_qp.c
@@ -247,7 +247,8 @@ void mthca_qp_event(struct mthca_dev *dev, u32 qpn,
 	spin_unlock(&dev->qp_table.lock);
 
 	if (!qp) {
-		mthca_warn(dev, "Async event for bogus QP %08x\n", qpn);
+		mthca_warn(dev, "Async event %d for bogus QP %08x\n",
+			   event_type, qpn);
 		return;
 	}
 
@@ -308,7 +309,6 @@ static void store_attrs(struct mthca_sqp *sqp, const struct ib_qp_attr *attr,
 static void init_port(struct mthca_dev *dev, int port)
 {
 	int err;
-	u8 status;
 	struct mthca_init_ib_param param;
 
 	memset(&param, 0, sizeof param);
@@ -319,11 +319,9 @@ static void init_port(struct mthca_dev *dev, int port)
 	param.gid_cap    = dev->limits.gid_table_len;
 	param.pkey_cap   = dev->limits.pkey_table_len;
 
-	err = mthca_INIT_IB(dev, &param, port, &status);
+	err = mthca_INIT_IB(dev, &param, port);
 	if (err)
 		mthca_warn(dev, "INIT_IB failed, return code %d.\n", err);
-	if (status)
-		mthca_warn(dev, "INIT_IB returned status %02x.\n", status);
 }
 
 static __be32 get_hw_access_flags(struct mthca_qp *qp, const struct ib_qp_attr *attr,
@@ -433,7 +431,6 @@ int mthca_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_m
 	struct mthca_qp_param *qp_param;
 	struct mthca_qp_context *context;
 	int mthca_state;
-	u8 status;
 
 	mutex_lock(&qp->mutex);
 
@@ -448,12 +445,9 @@ int mthca_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *qp_attr, int qp_attr_m
 		goto out;
 	}
 
-	err = mthca_QUERY_QP(dev, qp->qpn, 0, mailbox, &status);
-	if (err)
-		goto out_mailbox;
-	if (status) {
-		mthca_warn(dev, "QUERY_QP returned status %02x\n", status);
-		err = -EINVAL;
+	err = mthca_QUERY_QP(dev, qp->qpn, 0, mailbox);
+	if (err) {
+		mthca_warn(dev, "QUERY_QP failed (%d)\n", err);
 		goto out_mailbox;
 	}
 
@@ -508,6 +502,7 @@ done:
 	qp_attr->cap.max_inline_data = qp->max_inline_data;
 
 	qp_init_attr->cap	     = qp_attr->cap;
+	qp_init_attr->sq_sig_type    = qp->sq_policy;
 
 out_mailbox:
 	mthca_free_mailbox(dev, mailbox);
@@ -555,7 +550,6 @@ static int __mthca_modify_qp(struct ib_qp *ibqp,
 	struct mthca_qp_param *qp_param;
 	struct mthca_qp_context *qp_context;
 	u32 sqd_event = 0;
-	u8 status;
 	int err = -EINVAL;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
@@ -781,13 +775,10 @@ static int __mthca_modify_qp(struct ib_qp *ibqp,
 		sqd_event = 1 << 31;
 
 	err = mthca_MODIFY_QP(dev, cur_state, new_state, qp->qpn, 0,
-			      mailbox, sqd_event, &status);
-	if (err)
-		goto out_mailbox;
-	if (status) {
-		mthca_warn(dev, "modify QP %d->%d returned status %02x.\n",
-			   cur_state, new_state, status);
-		err = -EINVAL;
+			      mailbox, sqd_event);
+	if (err) {
+		mthca_warn(dev, "modify QP %d->%d returned %d.\n",
+			   cur_state, new_state, err);
 		goto out_mailbox;
 	}
 
@@ -817,7 +808,7 @@ static int __mthca_modify_qp(struct ib_qp *ibqp,
 		    cur_state != IB_QPS_ERR &&
 		    (new_state == IB_QPS_RESET ||
 		     new_state == IB_QPS_ERR))
-			mthca_CLOSE_IB(dev, qp->port, &status);
+			mthca_CLOSE_IB(dev, qp->port);
 	}
 
 	/*
@@ -869,7 +860,8 @@ int mthca_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr, int attr_mask,
 
 	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
 
-	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask)) {
+	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type, attr_mask,
+				IB_LINK_LAYER_UNSPECIFIED)) {
 		mthca_dbg(dev, "Bad QP transition (transport %d) "
 			  "%d->%d with attr 0x%08x\n",
 			  qp->transport, cur_state, new_state,
@@ -1319,10 +1311,12 @@ int mthca_alloc_qp(struct mthca_dev *dev,
 }
 
 static void mthca_lock_cqs(struct mthca_cq *send_cq, struct mthca_cq *recv_cq)
+	__acquires(&send_cq->lock) __acquires(&recv_cq->lock)
 {
-	if (send_cq == recv_cq)
+	if (send_cq == recv_cq) {
 		spin_lock_irq(&send_cq->lock);
-	else if (send_cq->cqn < recv_cq->cqn) {
+		__acquire(&recv_cq->lock);
+	} else if (send_cq->cqn < recv_cq->cqn) {
 		spin_lock_irq(&send_cq->lock);
 		spin_lock_nested(&recv_cq->lock, SINGLE_DEPTH_NESTING);
 	} else {
@@ -1332,10 +1326,12 @@ static void mthca_lock_cqs(struct mthca_cq *send_cq, struct mthca_cq *recv_cq)
 }
 
 static void mthca_unlock_cqs(struct mthca_cq *send_cq, struct mthca_cq *recv_cq)
+	__releases(&send_cq->lock) __releases(&recv_cq->lock)
 {
-	if (send_cq == recv_cq)
+	if (send_cq == recv_cq) {
+		__release(&recv_cq->lock);
 		spin_unlock_irq(&send_cq->lock);
-	else if (send_cq->cqn < recv_cq->cqn) {
+	} else if (send_cq->cqn < recv_cq->cqn) {
 		spin_unlock(&recv_cq->lock);
 		spin_unlock_irq(&send_cq->lock);
 	} else {
@@ -1425,7 +1421,6 @@ static inline int get_qp_refcount(struct mthca_dev *dev, struct mthca_qp *qp)
 void mthca_free_qp(struct mthca_dev *dev,
 		   struct mthca_qp *qp)
 {
-	u8 status;
 	struct mthca_cq *send_cq;
 	struct mthca_cq *recv_cq;
 
@@ -1450,7 +1445,7 @@ void mthca_free_qp(struct mthca_dev *dev,
 
 	if (qp->state != IB_QPS_RESET)
 		mthca_MODIFY_QP(dev, qp->state, IB_QPS_RESET, qp->qpn, 0,
-				NULL, 0, &status);
+				NULL, 0);
 
 	/*
 	 * If this is a userspace QP, the buffers, MR, CQs and so on
@@ -1489,8 +1484,8 @@ static int build_mlx_header(struct mthca_dev *dev, struct mthca_sqp *sqp,
 	int err;
 	u16 pkey;
 
-	ib_ud_header_init(256, /* assume a MAD */
-			  mthca_ah_grh_present(to_mah(wr->wr.ud.ah)),
+	ib_ud_header_init(256, /* assume a MAD */ 1, 0, 0,
+			  mthca_ah_grh_present(to_mah(wr->wr.ud.ah)), 0,
 			  &sqp->ud_header);
 
 	err = mthca_read_ah(dev, to_mah(wr->wr.ud.ah), &sqp->ud_header);
@@ -2259,7 +2254,6 @@ void mthca_free_err_wqe(struct mthca_dev *dev, struct mthca_qp *qp, int is_send,
 int mthca_init_qp_table(struct mthca_dev *dev)
 {
 	int err;
-	u8 status;
 	int i;
 
 	spin_lock_init(&dev->qp_table.lock);
@@ -2286,15 +2280,10 @@ int mthca_init_qp_table(struct mthca_dev *dev)
 
 	for (i = 0; i < 2; ++i) {
 		err = mthca_CONF_SPECIAL_QP(dev, i ? IB_QPT_GSI : IB_QPT_SMI,
-					    dev->qp_table.sqp_start + i * 2,
-					    &status);
-		if (err)
-			goto err_out;
-		if (status) {
+				    dev->qp_table.sqp_start + i * 2);
+		if (err) {
 			mthca_warn(dev, "CONF_SPECIAL_QP returned "
-				   "status %02x, aborting.\n",
-				   status);
-			err = -EINVAL;
+				   "%d, aborting.\n", err);
 			goto err_out;
 		}
 	}
@@ -2302,7 +2291,7 @@ int mthca_init_qp_table(struct mthca_dev *dev)
 
  err_out:
 	for (i = 0; i < 2; ++i)
-		mthca_CONF_SPECIAL_QP(dev, i, 0, &status);
+		mthca_CONF_SPECIAL_QP(dev, i, 0);
 
 	mthca_array_cleanup(&dev->qp_table.qp, dev->limits.num_qps);
 	mthca_alloc_cleanup(&dev->qp_table.alloc);
@@ -2313,10 +2302,9 @@ int mthca_init_qp_table(struct mthca_dev *dev)
 void mthca_cleanup_qp_table(struct mthca_dev *dev)
 {
 	int i;
-	u8 status;
 
 	for (i = 0; i < 2; ++i)
-		mthca_CONF_SPECIAL_QP(dev, i, 0, &status);
+		mthca_CONF_SPECIAL_QP(dev, i, 0);
 
 	mthca_array_cleanup(&dev->qp_table.qp, dev->limits.num_qps);
 	mthca_alloc_cleanup(&dev->qp_table.alloc);
diff --git a/drivers/infiniband/hw/mthca/mthca_reset.c b/drivers/infiniband/hw/mthca/mthca_reset.c
index acb6817f606..74c6a942604 100644
--- a/drivers/infiniband/hw/mthca/mthca_reset.c
+++ b/drivers/infiniband/hw/mthca/mthca_reset.c
@@ -30,7 +30,6 @@
  * SOFTWARE.
  */
 
-#include <linux/init.h>
 #include <linux/errno.h>
 #include <linux/pci.h>
 #include <linux/delay.h>
@@ -114,7 +113,7 @@ int mthca_reset(struct mthca_dev *mdev)
 	}
 
 	hca_pcix_cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_PCIX);
-	hca_pcie_cap = pci_find_capability(mdev->pdev, PCI_CAP_ID_EXP);
+	hca_pcie_cap = pci_pcie_cap(mdev->pdev);
 
 	if (bridge) {
 		bridge_header = kmalloc(256, GFP_KERNEL);
@@ -242,16 +241,16 @@ good:
 
 	if (hca_pcie_cap) {
 		devctl = hca_header[(hca_pcie_cap + PCI_EXP_DEVCTL) / 4];
-		if (pci_write_config_word(mdev->pdev, hca_pcie_cap + PCI_EXP_DEVCTL,
-					   devctl)) {
+		if (pcie_capability_write_word(mdev->pdev, PCI_EXP_DEVCTL,
+					       devctl)) {
 			err = -ENODEV;
 			mthca_err(mdev, "Couldn't restore HCA PCI Express "
 				  "Device Control register, aborting.\n");
 			goto out;
 		}
 		linkctl = hca_header[(hca_pcie_cap + PCI_EXP_LNKCTL) / 4];
-		if (pci_write_config_word(mdev->pdev, hca_pcie_cap + PCI_EXP_LNKCTL,
-					   linkctl)) {
+		if (pcie_capability_write_word(mdev->pdev, PCI_EXP_LNKCTL,
+					       linkctl)) {
 			err = -ENODEV;
 			mthca_err(mdev, "Couldn't restore HCA PCI Express "
 				  "Link control register, aborting.\n");
diff --git a/drivers/infiniband/hw/mthca/mthca_srq.c b/drivers/infiniband/hw/mthca/mthca_srq.c
index 4fabe62aab8..d22f970480c 100644
--- a/drivers/infiniband/hw/mthca/mthca_srq.c
+++ b/drivers/infiniband/hw/mthca/mthca_srq.c
@@ -200,7 +200,6 @@ int mthca_alloc_srq(struct mthca_dev *dev, struct mthca_pd *pd,
 		    struct ib_srq_attr *attr, struct mthca_srq *srq)
 {
 	struct mthca_mailbox *mailbox;
-	u8 status;
 	int ds;
 	int err;
 
@@ -266,18 +265,12 @@ int mthca_alloc_srq(struct mthca_dev *dev, struct mthca_pd *pd,
 	else
 		mthca_tavor_init_srq_context(dev, pd, srq, mailbox->buf);
 
-	err = mthca_SW2HW_SRQ(dev, mailbox, srq->srqn, &status);
+	err = mthca_SW2HW_SRQ(dev, mailbox, srq->srqn);
 
 	if (err) {
 		mthca_warn(dev, "SW2HW_SRQ failed (%d)\n", err);
 		goto err_out_free_buf;
 	}
-	if (status) {
-		mthca_warn(dev, "SW2HW_SRQ returned status 0x%02x\n",
-			   status);
-		err = -EINVAL;
-		goto err_out_free_buf;
-	}
 
 	spin_lock_irq(&dev->srq_table.lock);
 	if (mthca_array_set(&dev->srq_table.srq,
@@ -299,11 +292,9 @@ int mthca_alloc_srq(struct mthca_dev *dev, struct mthca_pd *pd,
 	return 0;
 
 err_out_free_srq:
-	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn, &status);
+	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn);
 	if (err)
 		mthca_warn(dev, "HW2SW_SRQ failed (%d)\n", err);
-	else if (status)
-		mthca_warn(dev, "HW2SW_SRQ returned status 0x%02x\n", status);
 
 err_out_free_buf:
 	if (!pd->ibpd.uobject)
@@ -340,7 +331,6 @@ void mthca_free_srq(struct mthca_dev *dev, struct mthca_srq *srq)
 {
 	struct mthca_mailbox *mailbox;
 	int err;
-	u8 status;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox)) {
@@ -348,11 +338,9 @@ void mthca_free_srq(struct mthca_dev *dev, struct mthca_srq *srq)
 		return;
 	}
 
-	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn, &status);
+	err = mthca_HW2SW_SRQ(dev, mailbox, srq->srqn);
 	if (err)
 		mthca_warn(dev, "HW2SW_SRQ failed (%d)\n", err);
-	else if (status)
-		mthca_warn(dev, "HW2SW_SRQ returned status 0x%02x\n", status);
 
 	spin_lock_irq(&dev->srq_table.lock);
 	mthca_array_clear(&dev->srq_table.srq,
@@ -378,8 +366,7 @@ int mthca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 {
 	struct mthca_dev *dev = to_mdev(ibsrq->device);
 	struct mthca_srq *srq = to_msrq(ibsrq);
-	int ret;
-	u8 status;
+	int ret = 0;
 
 	/* We don't support resizing SRQs (yet?) */
 	if (attr_mask & IB_SRQ_MAX_WR)
@@ -391,16 +378,11 @@ int mthca_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
 			return -EINVAL;
 
 		mutex_lock(&srq->mutex);
-		ret = mthca_ARM_SRQ(dev, srq->srqn, attr->srq_limit, &status);
+		ret = mthca_ARM_SRQ(dev, srq->srqn, attr->srq_limit);
 		mutex_unlock(&srq->mutex);
-
-		if (ret)
-			return ret;
-		if (status)
-			return -EINVAL;
 	}
 
-	return 0;
+	return ret;
 }
 
 int mthca_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
@@ -410,14 +392,13 @@ int mthca_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
 	struct mthca_mailbox *mailbox;
 	struct mthca_arbel_srq_context *arbel_ctx;
 	struct mthca_tavor_srq_context *tavor_ctx;
-	u8 status;
 	int err;
 
 	mailbox = mthca_alloc_mailbox(dev, GFP_KERNEL);
 	if (IS_ERR(mailbox))
 		return PTR_ERR(mailbox);
 
-	err = mthca_QUERY_SRQ(dev, srq->srqn, mailbox, &status);
+	err = mthca_QUERY_SRQ(dev, srq->srqn, mailbox);
 	if (err)
 		goto out;
 
diff --git a/drivers/infiniband/hw/nes/Kconfig b/drivers/infiniband/hw/nes/Kconfig
index d449eb6ec78..846dc97cf26 100644
--- a/drivers/infiniband/hw/nes/Kconfig
+++ b/drivers/infiniband/hw/nes/Kconfig
@@ -4,14 +4,13 @@ config INFINIBAND_NES
 	select LIBCRC32C
 	select INET_LRO
 	---help---
-	  This is a low-level driver for NetEffect RDMA enabled
-	  Network Interface Cards (RNIC).
+	  This is the RDMA Network Interface Card (RNIC) driver for
+	  NetEffect Ethernet Cluster Server Adapters.
 
 config INFINIBAND_NES_DEBUG
 	bool "Verbose debugging output"
 	depends on INFINIBAND_NES
 	default n
 	---help---
-	  This option causes the NetEffect RNIC driver to produce debug
-	  messages.  Select this if you are developing the driver
-	  or trying to diagnose a problem.
+	  This option enables debug messages from the NetEffect RNIC
+	  driver.  Select this if you are diagnosing a problem.
diff --git a/drivers/infiniband/hw/nes/Makefile b/drivers/infiniband/hw/nes/Makefile
index 35148513c47..97820c23ece 100644
--- a/drivers/infiniband/hw/nes/Makefile
+++ b/drivers/infiniband/hw/nes/Makefile
@@ -1,3 +1,3 @@
 obj-$(CONFIG_INFINIBAND_NES) += iw_nes.o
 
-iw_nes-objs := nes.o nes_hw.o nes_nic.o nes_utils.o nes_verbs.o nes_cm.o
+iw_nes-objs := nes.o nes_hw.o nes_nic.o nes_utils.o nes_verbs.o nes_cm.o nes_mgt.o
diff --git a/drivers/infiniband/hw/nes/nes.c b/drivers/infiniband/hw/nes/nes.c
index d2884e77809..3b2a6dc8ea9 100644
--- a/drivers/infiniband/hw/nes/nes.c
+++ b/drivers/infiniband/hw/nes/nes.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -44,6 +44,7 @@
 #include <linux/init.h>
 #include <linux/if_arp.h>
 #include <linux/highmem.h>
+#include <linux/slab.h>
 #include <asm/io.h>
 #include <asm/irq.h>
 #include <asm/byteorder.h>
@@ -67,30 +68,32 @@ MODULE_VERSION(DRV_VERSION);
 int max_mtu = 9000;
 int interrupt_mod_interval = 0;
 
-
 /* Interoperability */
 int mpa_version = 1;
-module_param(mpa_version, int, 0);
+module_param(mpa_version, int, 0644);
 MODULE_PARM_DESC(mpa_version, "MPA version to be used int MPA Req/Resp (0 or 1)");
 
 /* Interoperability */
 int disable_mpa_crc = 0;
-module_param(disable_mpa_crc, int, 0);
+module_param(disable_mpa_crc, int, 0644);
 MODULE_PARM_DESC(disable_mpa_crc, "Disable checking of MPA CRC");
 
-unsigned int send_first = 0;
-module_param(send_first, int, 0);
-MODULE_PARM_DESC(send_first, "Send RDMA Message First on Active Connection");
-
-
-unsigned int nes_drv_opt = 0;
-module_param(nes_drv_opt, int, 0);
+unsigned int nes_drv_opt = NES_DRV_OPT_DISABLE_INT_MOD | NES_DRV_OPT_ENABLE_PAU;
+module_param(nes_drv_opt, int, 0644);
 MODULE_PARM_DESC(nes_drv_opt, "Driver option parameters");
 
 unsigned int nes_debug_level = 0;
 module_param_named(debug_level, nes_debug_level, uint, 0644);
 MODULE_PARM_DESC(debug_level, "Enable debug output level");
 
+unsigned int wqm_quanta = 0x10000;
+module_param(wqm_quanta, int, 0644);
+MODULE_PARM_DESC(wqm_quanta, "WQM quanta");
+
+static bool limit_maxrdreqsz;
+module_param(limit_maxrdreqsz, bool, 0644);
+MODULE_PARM_DESC(limit_maxrdreqsz, "Limit max read request size to 256 Bytes");
+
 LIST_HEAD(nes_adapter_list);
 static LIST_HEAD(nes_dev_list);
 
@@ -101,12 +104,23 @@ static unsigned int sysfs_nonidx_addr;
 static unsigned int sysfs_idx_addr;
 
 static struct pci_device_id nes_pci_table[] = {
-	{PCI_VENDOR_ID_NETEFFECT, PCI_DEVICE_ID_NETEFFECT_NE020, PCI_ANY_ID, PCI_ANY_ID},
+	{ PCI_VDEVICE(NETEFFECT, PCI_DEVICE_ID_NETEFFECT_NE020), },
+	{ PCI_VDEVICE(NETEFFECT, PCI_DEVICE_ID_NETEFFECT_NE020_KR), },
 	{0}
 };
 
 MODULE_DEVICE_TABLE(pci, nes_pci_table);
 
+/* registered nes netlink callbacks */
+static struct ibnl_client_cbs nes_nl_cb_table[] = {
+	[RDMA_NL_IWPM_REG_PID] = {.dump = iwpm_register_pid_cb},
+	[RDMA_NL_IWPM_ADD_MAPPING] = {.dump = iwpm_add_mapping_cb},
+	[RDMA_NL_IWPM_QUERY_MAPPING] = {.dump = iwpm_add_and_query_mapping_cb},
+	[RDMA_NL_IWPM_HANDLE_ERR] = {.dump = iwpm_mapping_error_cb},
+	[RDMA_NL_IWPM_MAPINFO] = {.dump = iwpm_mapping_info_cb},
+	[RDMA_NL_IWPM_MAPINFO_NUM] = {.dump = iwpm_ack_mapping_info_cb}
+};
+
 static int nes_inetaddr_event(struct notifier_block *, unsigned long, void *);
 static int nes_net_event(struct notifier_block *, unsigned long, void *);
 static int nes_notifiers_registered;
@@ -120,9 +134,6 @@ static struct notifier_block nes_net_notifier = {
 	.notifier_call = nes_net_event
 };
 
-
-
-
 /**
  * nes_inetaddr_event
  */
@@ -133,21 +144,21 @@ static int nes_inetaddr_event(struct notifier_block *notifier,
 	struct net_device *event_netdev = ifa->ifa_dev->dev;
 	struct nes_device *nesdev;
 	struct net_device *netdev;
+	struct net_device *upper_dev;
 	struct nes_vnic *nesvnic;
-	unsigned int addr;
-	unsigned int mask;
-
-	addr = ntohl(ifa->ifa_address);
-	mask = ntohl(ifa->ifa_mask);
-	nes_debug(NES_DBG_NETDEV, "nes_inetaddr_event: ip address " NIPQUAD_FMT
-		  ", netmask " NIPQUAD_FMT ".\n",
-		  HIPQUAD(addr), HIPQUAD(mask));
+	unsigned int is_bonded;
+
+	nes_debug(NES_DBG_NETDEV, "nes_inetaddr_event: ip address %pI4, netmask %pI4.\n",
+		  &ifa->ifa_address, &ifa->ifa_mask);
 	list_for_each_entry(nesdev, &nes_dev_list, list) {
 		nes_debug(NES_DBG_NETDEV, "Nesdev list entry = 0x%p. (%s)\n",
 				nesdev, nesdev->netdev[0]->name);
 		netdev = nesdev->netdev[0];
 		nesvnic = netdev_priv(netdev);
-		if (netdev == event_netdev) {
+		upper_dev = netdev_master_upper_dev_get(netdev);
+		is_bonded = netif_is_bond_slave(netdev) &&
+			    (upper_dev == event_netdev);
+		if ((netdev == event_netdev) || is_bonded) {
 			if (nesvnic->rdma_enabled == 0) {
 				nes_debug(NES_DBG_NETDEV, "Returning without processing event for %s since"
 						" RDMA is not enabled.\n",
@@ -164,7 +175,10 @@ static int nes_inetaddr_event(struct notifier_block *notifier,
 					nes_manage_arp_cache(netdev, netdev->dev_addr,
 							ntohl(nesvnic->local_ipaddr), NES_ARP_DELETE);
 					nesvnic->local_ipaddr = 0;
-					return NOTIFY_OK;
+					if (is_bonded)
+						continue;
+					else
+						return NOTIFY_OK;
 					break;
 				case NETDEV_UP:
 					nes_debug(NES_DBG_NETDEV, "event:UP\n");
@@ -173,15 +187,24 @@ static int nes_inetaddr_event(struct notifier_block *notifier,
 						nes_debug(NES_DBG_NETDEV, "Interface already has local_ipaddr\n");
 						return NOTIFY_OK;
 					}
+					/* fall through */
+				case NETDEV_CHANGEADDR:
 					/* Add the address to the IP table */
-					nesvnic->local_ipaddr = ifa->ifa_address;
+					if (upper_dev)
+						nesvnic->local_ipaddr =
+							((struct in_device *)upper_dev->ip_ptr)->ifa_list->ifa_address;
+					else
+						nesvnic->local_ipaddr = ifa->ifa_address;
 
 					nes_write_indexed(nesdev,
 							NES_IDX_DST_IP_ADDR+(0x10*PCI_FUNC(nesdev->pcidev->devfn)),
-							ntohl(ifa->ifa_address));
+							ntohl(nesvnic->local_ipaddr));
 					nes_manage_arp_cache(netdev, netdev->dev_addr,
 							ntohl(nesvnic->local_ipaddr), NES_ARP_ADD);
-					return NOTIFY_OK;
+					if (is_bonded)
+						continue;
+					else
+						return NOTIFY_OK;
 					break;
 				default:
 					break;
@@ -254,13 +277,11 @@ static void nes_cqp_rem_ref_callback(struct nes_device *nesdev, struct nes_cqp_r
 	unsigned long flags;
 	struct nes_qp *nesqp = cqp_request->cqp_callback_pointer;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
-	u32 qp_id;
 
 	atomic_inc(&qps_destroyed);
 
 	/* Free the control structures */
 
-	qp_id = nesqp->hwqp.qp_id;
 	if (nesqp->pbl_vbase) {
 		pci_free_consistent(nesdev->pcidev, nesqp->qp_mem_size,
 				nesqp->hwqp.q2_vbase, nesqp->hwqp.q2_pbase);
@@ -276,6 +297,7 @@ static void nes_cqp_rem_ref_callback(struct nes_device *nesdev, struct nes_cqp_r
 	}
 	nes_free_resource(nesadapter, nesadapter->allocated_qps, nesqp->hwqp.qp_id);
 
+	nesadapter->qp_table[nesqp->hwqp.qp_id-NES_FIRST_QPN] = NULL;
 	kfree(nesqp->allocated_buffer);
 
 }
@@ -289,7 +311,6 @@ void nes_rem_ref(struct ib_qp *ibqp)
 	struct nes_qp *nesqp;
 	struct nes_vnic *nesvnic = to_nesvnic(ibqp->device);
 	struct nes_device *nesdev = nesvnic->nesdev;
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	struct nes_hw_cqp_wqe *cqp_wqe;
 	struct nes_cqp_request *cqp_request;
 	u32 opcode;
@@ -303,7 +324,8 @@ void nes_rem_ref(struct ib_qp *ibqp)
 	}
 
 	if (atomic_dec_and_test(&nesqp->refcount)) {
-		nesadapter->qp_table[nesqp->hwqp.qp_id-NES_FIRST_QPN] = NULL;
+		if (nesqp->pau_mode)
+			nes_destroy_pau_qp(nesdev, nesqp);
 
 		/* Destroy the QP */
 		cqp_request = nes_get_cqp_request(nesdev);
@@ -354,10 +376,8 @@ struct ib_qp *nes_get_qp(struct ib_device *device, int qpn)
  */
 static void nes_print_macaddr(struct net_device *netdev)
 {
-	DECLARE_MAC_BUF(mac);
-
-	nes_debug(NES_DBG_INIT, "%s: %s, IRQ %u\n",
-		  netdev->name, print_mac(mac, netdev->dev_addr), netdev->irq);
+	nes_debug(NES_DBG_INIT, "%s: %pM, IRQ %u\n",
+		  netdev->name, netdev->dev_addr, netdev->irq);
 }
 
 /**
@@ -435,12 +455,11 @@ static irqreturn_t nes_interrupt(int irq, void *dev_id)
 /**
  * nes_probe - Device initialization
  */
-static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_id *ent)
+static int nes_probe(struct pci_dev *pcidev, const struct pci_device_id *ent)
 {
 	struct net_device *netdev = NULL;
 	struct nes_device *nesdev = NULL;
 	int ret = 0;
-	struct nes_vnic *nesvnic = NULL;
 	void __iomem *mmio_regs = NULL;
 	u8 hw_rev;
 
@@ -479,23 +498,23 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 	}
 
 	if ((sizeof(dma_addr_t) > 4)) {
-		ret = pci_set_dma_mask(pcidev, DMA_64BIT_MASK);
+		ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(64));
 		if (ret < 0) {
 			printk(KERN_ERR PFX "64b DMA mask configuration failed\n");
 			goto bail2;
 		}
-		ret = pci_set_consistent_dma_mask(pcidev, DMA_64BIT_MASK);
+		ret = pci_set_consistent_dma_mask(pcidev, DMA_BIT_MASK(64));
 		if (ret) {
 			printk(KERN_ERR PFX "64b DMA consistent mask configuration failed\n");
 			goto bail2;
 		}
 	} else {
-		ret = pci_set_dma_mask(pcidev, DMA_32BIT_MASK);
+		ret = pci_set_dma_mask(pcidev, DMA_BIT_MASK(32));
 		if (ret < 0) {
 			printk(KERN_ERR PFX "32b DMA mask configuration failed\n");
 			goto bail2;
 		}
-		ret = pci_set_consistent_dma_mask(pcidev, DMA_32BIT_MASK);
+		ret = pci_set_consistent_dma_mask(pcidev, DMA_BIT_MASK(32));
 		if (ret) {
 			printk(KERN_ERR PFX "32b DMA consistent mask configuration failed\n");
 			goto bail2;
@@ -522,7 +541,8 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 	spin_lock_init(&nesdev->indexed_regs_lock);
 
 	/* Remap the PCI registers in adapter BAR0 to kernel VA space */
-	mmio_regs = ioremap_nocache(pci_resource_start(pcidev, BAR_0), sizeof(mmio_regs));
+	mmio_regs = ioremap_nocache(pci_resource_start(pcidev, BAR_0),
+				    pci_resource_len(pcidev, BAR_0));
 	if (mmio_regs == NULL) {
 		printk(KERN_ERR PFX "Unable to remap BAR0\n");
 		ret = -EIO;
@@ -559,12 +579,44 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 		goto bail5;
 	}
 	nesdev->nesadapter->et_rx_coalesce_usecs_irq = interrupt_mod_interval;
+	nesdev->nesadapter->wqm_quanta = wqm_quanta;
 
 	/* nesdev->base_doorbell_index =
 			nesdev->nesadapter->pd_config_base[PCI_FUNC(nesdev->pcidev->devfn)]; */
 	nesdev->base_doorbell_index = 1;
 	nesdev->doorbell_start = nesdev->nesadapter->doorbell_start;
-	nesdev->mac_index = PCI_FUNC(nesdev->pcidev->devfn) % nesdev->nesadapter->port_count;
+	if (nesdev->nesadapter->phy_type[0] == NES_PHY_TYPE_PUMA_1G) {
+		switch (PCI_FUNC(nesdev->pcidev->devfn) %
+			nesdev->nesadapter->port_count) {
+		case 1:
+			nesdev->mac_index = 2;
+			break;
+		case 2:
+			nesdev->mac_index = 1;
+			break;
+		case 3:
+			nesdev->mac_index = 3;
+			break;
+		case 0:
+		default:
+			nesdev->mac_index = 0;
+		}
+	} else {
+		nesdev->mac_index = PCI_FUNC(nesdev->pcidev->devfn) %
+						nesdev->nesadapter->port_count;
+	}
+
+	if ((limit_maxrdreqsz ||
+	     ((nesdev->nesadapter->phy_type[0] == NES_PHY_TYPE_GLADIUS) &&
+	      (hw_rev == NE020_REV1))) &&
+	    (pcie_get_readrq(pcidev) > 256)) {
+		if (pcie_set_readrq(pcidev, 256))
+			printk(KERN_ERR PFX "Unable to set max read request"
+				" to 256 bytes\n");
+		else
+			nes_debug(NES_DBG_INIT, "Max read request size set"
+				" to 256 bytes\n");
+	}
 
 	tasklet_init(&nesdev->dpc_tasklet, nes_dpc, (unsigned long)nesdev);
 
@@ -583,7 +635,7 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 	nesdev->int_req = (0x101 << PCI_FUNC(nesdev->pcidev->devfn)) |
 			(1 << (PCI_FUNC(nesdev->pcidev->devfn)+16));
 	if (PCI_FUNC(nesdev->pcidev->devfn) < 4) {
-		nesdev->int_req |= (1 << (PCI_FUNC(nesdev->pcidev->devfn)+24));
+		nesdev->int_req |= (1 << (PCI_FUNC(nesdev->mac_index)+24));
 	}
 
 	/* TODO: This really should be the first driver to load, not function 0 */
@@ -629,28 +681,40 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 	}
 	nes_notifiers_registered++;
 
-	/* Initialize network devices */
-		if ((netdev = nes_netdev_init(nesdev, mmio_regs)) == NULL) {
-			goto bail7;
-		}
+	if (ibnl_add_client(RDMA_NL_NES, RDMA_NL_IWPM_NUM_OPS, nes_nl_cb_table))
+		printk(KERN_ERR PFX "%s[%u]: Failed to add netlink callback\n",
+			__func__, __LINE__);
 
-		/* Register network device */
-		ret = register_netdev(netdev);
-		if (ret) {
-			printk(KERN_ERR PFX "Unable to register netdev, ret = %d\n", ret);
-			nes_netdev_destroy(netdev);
-			goto bail7;
-		}
+	ret = iwpm_init(RDMA_NL_NES);
+	if (ret) {
+		printk(KERN_ERR PFX "%s: port mapper initialization failed\n",
+				pci_name(pcidev));
+		goto bail7;
+	}
 
-		nes_print_macaddr(netdev);
-		/* create a CM core for this netdev */
-		nesvnic = netdev_priv(netdev);
+	INIT_DELAYED_WORK(&nesdev->work, nes_recheck_link_status);
+
+	/* Initialize network devices */
+	netdev = nes_netdev_init(nesdev, mmio_regs);
+	if (netdev == NULL) {
+		ret = -ENOMEM;
+		goto bail7;
+	}
+
+	/* Register network device */
+	ret = register_netdev(netdev);
+	if (ret) {
+		printk(KERN_ERR PFX "Unable to register netdev, ret = %d\n", ret);
+		nes_netdev_destroy(netdev);
+		goto bail7;
+	}
 
-		nesdev->netdev_count++;
-		nesdev->nesadapter->netdev_count++;
+	nes_print_macaddr(netdev);
 
+	nesdev->netdev_count++;
+	nesdev->nesadapter->netdev_count++;
 
-	printk(KERN_ERR PFX "%s: NetEffect RNIC driver successfully loaded.\n",
+	printk(KERN_INFO PFX "%s: NetEffect RNIC driver successfully loaded.\n",
 			pci_name(pcidev));
 	return 0;
 
@@ -666,6 +730,7 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 
 	nes_debug(NES_DBG_INIT, "netdev_count=%d, nesadapter->netdev_count=%d\n",
 			nesdev->netdev_count, nesdev->nesadapter->netdev_count);
+	ibnl_remove_client(RDMA_NL_NES);
 
 	nes_notifiers_registered--;
 	if (nes_notifiers_registered == 0) {
@@ -710,11 +775,12 @@ static int __devinit nes_probe(struct pci_dev *pcidev, const struct pci_device_i
 /**
  * nes_remove - unload from kernel
  */
-static void __devexit nes_remove(struct pci_dev *pcidev)
+static void nes_remove(struct pci_dev *pcidev)
 {
 	struct nes_device *nesdev = pci_get_drvdata(pcidev);
 	struct net_device *netdev;
 	int netdev_index = 0;
+	unsigned long flags;
 
 		if (nesdev->netdev_count) {
 			netdev = nesdev->netdev[netdev_index];
@@ -728,6 +794,8 @@ static void __devexit nes_remove(struct pci_dev *pcidev)
 				nesdev->nesadapter->netdev_count--;
 			}
 		}
+	ibnl_remove_client(RDMA_NL_NES);
+	iwpm_exit(RDMA_NL_NES);
 
 	nes_notifiers_registered--;
 	if (nes_notifiers_registered == 0) {
@@ -741,6 +809,14 @@ static void __devexit nes_remove(struct pci_dev *pcidev)
 	free_irq(pcidev->irq, nesdev);
 	tasklet_kill(&nesdev->dpc_tasklet);
 
+	spin_lock_irqsave(&nesdev->nesadapter->phy_lock, flags);
+	if (nesdev->link_recheck) {
+		spin_unlock_irqrestore(&nesdev->nesadapter->phy_lock, flags);
+		cancel_delayed_work_sync(&nesdev->work);
+	} else {
+		spin_unlock_irqrestore(&nesdev->nesadapter->phy_lock, flags);
+	}
+
 	/* Deallocate the Adapter Structure */
 	nes_destroy_adapter(nesdev->nesadapter);
 
@@ -762,7 +838,7 @@ static struct pci_driver nes_pci_driver = {
 	.name = DRV_NAME,
 	.id_table = nes_pci_table,
 	.probe = nes_probe,
-	.remove = __devexit_p(nes_remove),
+	.remove = nes_remove,
 };
 
 static ssize_t nes_show_adapter(struct device_driver *ddp, char *buf)
@@ -774,14 +850,14 @@ static ssize_t nes_show_adapter(struct device_driver *ddp, char *buf)
 
 	list_for_each_entry(nesdev, &nes_dev_list, list) {
 		if (i == ee_flsh_adapter) {
-			devfn      = nesdev->nesadapter->devfn;
-			bus_number = nesdev->nesadapter->bus_number;
+			devfn = nesdev->pcidev->devfn;
+			bus_number = nesdev->pcidev->bus->number;
 			break;
 		}
 		i++;
 	}
 
-	return snprintf(buf, PAGE_SIZE, "%x:%x", bus_number, devfn);
+	return snprintf(buf, PAGE_SIZE, "%x:%x\n", bus_number, devfn);
 }
 
 static ssize_t nes_store_adapter(struct device_driver *ddp,
@@ -1052,6 +1128,57 @@ static ssize_t nes_store_idx_data(struct device_driver *ddp,
 	return strnlen(buf, count);
 }
 
+
+/**
+ * nes_show_wqm_quanta
+ */
+static ssize_t nes_show_wqm_quanta(struct device_driver *ddp, char *buf)
+{
+	u32 wqm_quanta_value = 0xdead;
+	u32 i = 0;
+	struct nes_device *nesdev;
+
+	list_for_each_entry(nesdev, &nes_dev_list, list) {
+		if (i == ee_flsh_adapter) {
+			wqm_quanta_value = nesdev->nesadapter->wqm_quanta;
+			break;
+		}
+		i++;
+	}
+
+	return  snprintf(buf, PAGE_SIZE, "0x%X\n", wqm_quanta_value);
+}
+
+
+/**
+ * nes_store_wqm_quanta
+ */
+static ssize_t nes_store_wqm_quanta(struct device_driver *ddp,
+					const char *buf, size_t count)
+{
+	unsigned long wqm_quanta_value;
+	u32 wqm_config1;
+	u32 i = 0;
+	struct nes_device *nesdev;
+
+	if (kstrtoul(buf, 0, &wqm_quanta_value) < 0)
+		return -EINVAL;
+
+	list_for_each_entry(nesdev, &nes_dev_list, list) {
+		if (i == ee_flsh_adapter) {
+			nesdev->nesadapter->wqm_quanta = wqm_quanta_value;
+			wqm_config1 = nes_read_indexed(nesdev,
+						NES_IDX_WQM_CONFIG1);
+			nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG1,
+					((wqm_quanta_value << 1) |
+					(wqm_config1 & 0x00000001)));
+			break;
+		}
+		i++;
+	}
+	return strnlen(buf, count);
+}
+
 static DRIVER_ATTR(adapter, S_IRUSR | S_IWUSR,
 		   nes_show_adapter, nes_store_adapter);
 static DRIVER_ATTR(eeprom_cmd, S_IRUSR | S_IWUSR,
@@ -1070,6 +1197,8 @@ static DRIVER_ATTR(idx_addr, S_IRUSR | S_IWUSR,
 		   nes_show_idx_addr, nes_store_idx_addr);
 static DRIVER_ATTR(idx_data, S_IRUSR | S_IWUSR,
 		   nes_show_idx_data, nes_store_idx_data);
+static DRIVER_ATTR(wqm_quanta, S_IRUSR | S_IWUSR,
+		   nes_show_wqm_quanta, nes_store_wqm_quanta);
 
 static int nes_create_driver_sysfs(struct pci_driver *drv)
 {
@@ -1083,6 +1212,7 @@ static int nes_create_driver_sysfs(struct pci_driver *drv)
 	error |= driver_create_file(&drv->driver, &driver_attr_nonidx_data);
 	error |= driver_create_file(&drv->driver, &driver_attr_idx_addr);
 	error |= driver_create_file(&drv->driver, &driver_attr_idx_data);
+	error |= driver_create_file(&drv->driver, &driver_attr_wqm_quanta);
 	return error;
 }
 
@@ -1097,6 +1227,7 @@ static void nes_remove_driver_sysfs(struct pci_driver *drv)
 	driver_remove_file(&drv->driver, &driver_attr_nonidx_data);
 	driver_remove_file(&drv->driver, &driver_attr_idx_addr);
 	driver_remove_file(&drv->driver, &driver_attr_idx_data);
+	driver_remove_file(&drv->driver, &driver_attr_wqm_quanta);
 }
 
 /**
diff --git a/drivers/infiniband/hw/nes/nes.h b/drivers/infiniband/hw/nes/nes.h
index 39bd897b40c..bd9d132f11c 100644
--- a/drivers/infiniband/hw/nes/nes.h
+++ b/drivers/infiniband/hw/nes/nes.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -43,7 +43,6 @@
 #include <linux/dma-mapping.h>
 #include <linux/workqueue.h>
 #include <linux/slab.h>
-#include <linux/version.h>
 #include <asm/io.h>
 #include <linux/crc32c.h>
 
@@ -52,23 +51,24 @@
 #include <rdma/ib_pack.h>
 #include <rdma/rdma_cm.h>
 #include <rdma/iw_cm.h>
+#include <rdma/rdma_netlink.h>
+#include <rdma/iw_portmap.h>
 
 #define NES_SEND_FIRST_WRITE
 
 #define QUEUE_DISCONNECTS
 
-#define DRV_BUILD   "1"
-
 #define DRV_NAME    "iw_nes"
-#define DRV_VERSION "1.0 KO Build " DRV_BUILD
+#define DRV_VERSION "1.5.0.1"
 #define PFX         DRV_NAME ": "
 
 /*
  * NetEffect PCI vendor id and NE010 PCI device id.
  */
 #ifndef PCI_VENDOR_ID_NETEFFECT	/* not in pci.ids yet */
-#define PCI_VENDOR_ID_NETEFFECT       0x1678
-#define PCI_DEVICE_ID_NETEFFECT_NE020 0x0100
+#define PCI_VENDOR_ID_NETEFFECT          0x1678
+#define PCI_DEVICE_ID_NETEFFECT_NE020    0x0100
+#define PCI_DEVICE_ID_NETEFFECT_NE020_KR 0x0110
 #endif
 
 #define NE020_REV   4
@@ -104,6 +104,7 @@
 #define NES_DRV_OPT_NO_INLINE_DATA       0x00000080
 #define NES_DRV_OPT_DISABLE_INT_MOD      0x00000100
 #define NES_DRV_OPT_DISABLE_VIRT_WQ      0x00000200
+#define NES_DRV_OPT_ENABLE_PAU           0x00000400
 
 #define NES_AEQ_EVENT_TIMEOUT         2500
 #define NES_DISCONNECT_EVENT_TIMEOUT  2000
@@ -130,6 +131,8 @@
 #define NES_DBG_IW_RX       0x00020000
 #define NES_DBG_IW_TX       0x00040000
 #define NES_DBG_SHUTDOWN    0x00080000
+#define NES_DBG_PAU         0x00100000
+#define NES_DBG_NLMSG       0x00200000
 #define NES_DBG_RSVD1       0x10000000
 #define NES_DBG_RSVD2       0x20000000
 #define NES_DBG_RSVD3       0x40000000
@@ -138,14 +141,18 @@
 
 #ifdef CONFIG_INFINIBAND_NES_DEBUG
 #define nes_debug(level, fmt, args...) \
+do { \
 	if (level & nes_debug_level) \
-		printk(KERN_ERR PFX "%s[%u]: " fmt, __func__, __LINE__, ##args)
+		printk(KERN_ERR PFX "%s[%u]: " fmt, __func__, __LINE__, ##args); \
+} while (0)
 
-#define assert(expr)                                                \
-if (!(expr)) {                                                       \
-	printk(KERN_ERR PFX "Assertion failed! %s, %s, %s, line %d\n",  \
-		   #expr, __FILE__, __func__, __LINE__);                \
-}
+#define assert(expr) \
+do { \
+	if (!(expr)) { \
+		printk(KERN_ERR PFX "Assertion failed! %s, %s, %s, line %d\n", \
+			   #expr, __FILE__, __func__, __LINE__); \
+	} \
+} while (0)
 
 #define NES_EVENT_TIMEOUT   1200000
 #else
@@ -160,6 +167,7 @@ if (!(expr)) {                                                       \
 #include "nes_context.h"
 #include "nes_user.h"
 #include "nes_cm.h"
+#include "nes_mgt.h"
 
 extern int max_mtu;
 #define max_frame_len (max_mtu+ETH_HLEN)
@@ -167,10 +175,9 @@ extern int interrupt_mod_interval;
 extern int nes_if_count;
 extern int mpa_version;
 extern int disable_mpa_crc;
-extern unsigned int send_first;
 extern unsigned int nes_drv_opt;
 extern unsigned int nes_debug_level;
-
+extern unsigned int wqm_quanta;
 extern struct list_head nes_adapter_list;
 
 extern atomic_t cm_connects;
@@ -192,14 +199,16 @@ extern u32 cm_packets_created;
 extern u32 cm_packets_received;
 extern u32 cm_packets_dropped;
 extern u32 cm_packets_retrans;
-extern u32 cm_listens_created;
-extern u32 cm_listens_destroyed;
+extern atomic_t cm_listens_created;
+extern atomic_t cm_listens_destroyed;
 extern u32 cm_backlog_drops;
 extern atomic_t cm_loopbacks;
 extern atomic_t cm_nodes_created;
 extern atomic_t cm_nodes_destroyed;
 extern atomic_t cm_accel_dropped_pkts;
 extern atomic_t cm_resets_recvd;
+extern atomic_t pau_qps_created;
+extern atomic_t pau_qps_destroyed;
 
 extern u32 int_mod_timer_init;
 extern u32 int_mod_cq_depth_256;
@@ -260,13 +269,25 @@ struct nes_device {
 	u16                    base_doorbell_index;
 	u16                    currcq_count;
 	u16                    deepcq_count;
+	u8                     iw_status;
 	u8                     msi_enabled;
 	u8                     netdev_count;
 	u8                     napi_isr_ran;
 	u8                     disable_rx_flow_control;
 	u8                     disable_tx_flow_control;
+
+	struct delayed_work    work;
+	u8                     link_recheck;
 };
 
+/* Receive skb private area - must fit in skb->cb area */
+struct nes_rskb_cb {
+	u64                    busaddr;
+	u32                    maplen;
+	u32                    seqnum;
+	u8                     *data_start;
+	struct nes_qp          *nesqp;
+};
 
 static inline __le32 get_crc_value(struct nes_v4_quad *nes_quad)
 {
@@ -286,8 +307,8 @@ static inline __le32 get_crc_value(struct nes_v4_quad *nes_quad)
 static inline void
 set_wqe_64bit_value(__le32 *wqe_words, u32 index, u64 value)
 {
-	wqe_words[index]     = cpu_to_le32((u32) ((unsigned long)value));
-	wqe_words[index + 1] = cpu_to_le32((u32)(upper_32_bits((unsigned long)value)));
+	wqe_words[index]     = cpu_to_le32((u32) value);
+	wqe_words[index + 1] = cpu_to_le32(upper_32_bits(value));
 }
 
 static inline void
@@ -299,8 +320,8 @@ set_wqe_32bit_value(__le32 *wqe_words, u32 index, u32 value)
 static inline void
 nes_fill_init_cqp_wqe(struct nes_hw_cqp_wqe *cqp_wqe, struct nes_device *nesdev)
 {
-	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_COMP_CTX_LOW_IDX,
-			(u64)((unsigned long) &nesdev->cqp));
+	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_CTX_LOW_IDX]       = 0;
+	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_CTX_HIGH_IDX]      = 0;
 	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX]   = 0;
 	cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX]  = 0;
 	cqp_wqe->wqe_words[NES_CQP_STAG_WQE_PBL_BLK_COUNT_IDX] = 0;
@@ -380,11 +401,20 @@ static inline void nes_write8(void __iomem *addr, u8 val)
 	writeb(val, addr);
 }
 
-
+enum nes_resource {
+	NES_RESOURCE_MW = 1,
+	NES_RESOURCE_FAST_MR,
+	NES_RESOURCE_PHYS_MR,
+	NES_RESOURCE_USER_MR,
+	NES_RESOURCE_PD,
+	NES_RESOURCE_QP,
+	NES_RESOURCE_CQ,
+	NES_RESOURCE_ARP
+};
 
 static inline int nes_alloc_resource(struct nes_adapter *nesadapter,
 		unsigned long *resource_array, u32 max_resources,
-		u32 *req_resource_num, u32 *next)
+		u32 *req_resource_num, u32 *next, enum nes_resource resource_type)
 {
 	unsigned long flags;
 	u32 resource_num;
@@ -395,7 +425,7 @@ static inline int nes_alloc_resource(struct nes_adapter *nesadapter,
 	if (resource_num >= max_resources) {
 		resource_num = find_first_zero_bit(resource_array, max_resources);
 		if (resource_num >= max_resources) {
-			printk(KERN_ERR PFX "%s: No available resourcess.\n", __func__);
+			printk(KERN_ERR PFX "%s: No available resources [type=%u].\n", __func__, resource_type);
 			spin_unlock_irqrestore(&nesadapter->resource_lock, flags);
 			return -EMFILE;
 		}
@@ -504,6 +534,8 @@ void nes_nic_ce_handler(struct nes_device *, struct nes_hw_nic_cq *);
 void nes_iwarp_ce_handler(struct nes_device *, struct nes_hw_cq *);
 int nes_destroy_cqp(struct nes_device *);
 int nes_nic_cm_xmit(struct sk_buff *, struct net_device *);
+void nes_recheck_link_status(struct work_struct *work);
+void nes_terminate_timeout(unsigned long context);
 
 /* nes_nic.c */
 struct net_device *nes_netdev_init(struct nes_device *, void __iomem *);
@@ -522,9 +554,10 @@ int nes_cm_disconn(struct nes_qp *);
 void nes_cm_disconn_worker(void *);
 
 /* nes_verbs.c */
-int nes_hw_modify_qp(struct nes_device *, struct nes_qp *, u32, u32);
+int nes_hw_modify_qp(struct nes_device *, struct nes_qp *, u32, u32, u32);
 int nes_modify_qp(struct ib_qp *, struct ib_qp_attr *, int, struct ib_udata *);
 struct nes_ib_device *nes_init_ofa_device(struct net_device *);
+void  nes_port_ibevent(struct nes_vnic *nesvnic);
 void nes_destroy_ofa_device(struct nes_ib_device *);
 int nes_register_ofa_device(struct nes_ib_device *);
 
diff --git a/drivers/infiniband/hw/nes/nes_cm.c b/drivers/infiniband/hw/nes/nes_cm.c
index 6aa531d5276..6f09a72e78d 100644
--- a/drivers/infiniband/hw/nes/nes_cm.c
+++ b/drivers/infiniband/hw/nes/nes_cm.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2014 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -34,7 +34,7 @@
 
 #define TCPOPT_TIMESTAMP 8
 
-#include <asm/atomic.h>
+#include <linux/atomic.h>
 #include <linux/skbuff.h>
 #include <linux/ip.h>
 #include <linux/tcp.h>
@@ -52,10 +52,14 @@
 #include <linux/random.h>
 #include <linux/list.h>
 #include <linux/threads.h>
-
+#include <linux/highmem.h>
+#include <linux/slab.h>
+#include <net/arp.h>
 #include <net/neighbour.h>
 #include <net/route.h>
 #include <net/ip_fib.h>
+#include <net/tcp.h>
+#include <linux/fcntl.h>
 
 #include "nes.h"
 
@@ -65,8 +69,8 @@ u32 cm_packets_dropped;
 u32 cm_packets_retrans;
 u32 cm_packets_created;
 u32 cm_packets_received;
-u32 cm_listens_created;
-u32 cm_listens_destroyed;
+atomic_t cm_listens_created;
+atomic_t cm_listens_destroyed;
 u32 cm_backlog_drops;
 atomic_t cm_loopbacks;
 atomic_t cm_nodes_created;
@@ -75,35 +79,57 @@ atomic_t cm_accel_dropped_pkts;
 atomic_t cm_resets_recvd;
 
 static inline int mini_cm_accelerated(struct nes_cm_core *, struct nes_cm_node *);
-static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *,
-		struct nes_vnic *, struct nes_cm_info *);
-static int add_ref_cm_node(struct nes_cm_node *);
-static int rem_ref_cm_node(struct nes_cm_core *, struct nes_cm_node *);
+static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *, struct nes_vnic *, struct nes_cm_info *);
 static int mini_cm_del_listen(struct nes_cm_core *, struct nes_cm_listener *);
-static struct sk_buff *form_cm_frame(struct sk_buff *, struct nes_cm_node *,
-				     void *, u32, void *, u32, u8);
-static struct sk_buff *get_free_pkt(struct nes_cm_node *cm_node);
-
-static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *,
-					   struct nes_vnic *,
-					   struct ietf_mpa_frame *,
-					   struct nes_cm_info *);
-static int mini_cm_accept(struct nes_cm_core *, struct ietf_mpa_frame *,
-			  struct nes_cm_node *);
-static int mini_cm_reject(struct nes_cm_core *, struct ietf_mpa_frame *,
-			  struct nes_cm_node *);
+static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *, struct nes_vnic *, u16, void *, struct nes_cm_info *);
 static int mini_cm_close(struct nes_cm_core *, struct nes_cm_node *);
-static int mini_cm_recv_pkt(struct nes_cm_core *, struct nes_vnic *,
-			    struct sk_buff *);
+static int mini_cm_accept(struct nes_cm_core *, struct nes_cm_node *);
+static int mini_cm_reject(struct nes_cm_core *, struct nes_cm_node *);
+static int mini_cm_recv_pkt(struct nes_cm_core *, struct nes_vnic *, struct sk_buff *);
 static int mini_cm_dealloc_core(struct nes_cm_core *);
 static int mini_cm_get(struct nes_cm_core *);
 static int mini_cm_set(struct nes_cm_core *, u32, u32);
+
+static void form_cm_frame(struct sk_buff *, struct nes_cm_node *, void *, u32, void *, u32, u8);
+static int add_ref_cm_node(struct nes_cm_node *);
+static int rem_ref_cm_node(struct nes_cm_core *, struct nes_cm_node *);
+
 static int nes_cm_disconn_true(struct nes_qp *);
 static int nes_cm_post_event(struct nes_cm_event *event);
 static int nes_disconnect(struct nes_qp *nesqp, int abrupt);
 static void nes_disconnect_worker(struct work_struct *work);
-static int send_ack(struct nes_cm_node *cm_node);
+
+static int send_mpa_request(struct nes_cm_node *, struct sk_buff *);
+static int send_mpa_reject(struct nes_cm_node *);
+static int send_syn(struct nes_cm_node *, u32, struct sk_buff *);
+static int send_reset(struct nes_cm_node *, struct sk_buff *);
+static int send_ack(struct nes_cm_node *cm_node, struct sk_buff *skb);
 static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb);
+static void process_packet(struct nes_cm_node *, struct sk_buff *, struct nes_cm_core *);
+
+static void active_open_err(struct nes_cm_node *, struct sk_buff *, int);
+static void passive_open_err(struct nes_cm_node *, struct sk_buff *, int);
+static void cleanup_retrans_entry(struct nes_cm_node *);
+static void handle_rcv_mpa(struct nes_cm_node *, struct sk_buff *);
+static void free_retrans_entry(struct nes_cm_node *cm_node);
+static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph, struct sk_buff *skb, int optionsize, int passive);
+
+/* CM event handler functions */
+static void cm_event_connected(struct nes_cm_event *);
+static void cm_event_connect_error(struct nes_cm_event *);
+static void cm_event_reset(struct nes_cm_event *);
+static void cm_event_mpa_req(struct nes_cm_event *);
+static void cm_event_mpa_reject(struct nes_cm_event *);
+static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node);
+
+/* MPA build functions */
+static int cm_build_mpa_frame(struct nes_cm_node *, u8 **, u16 *, u8 *, u8);
+static void build_mpa_v2(struct nes_cm_node *, void *, u8);
+static void build_mpa_v1(struct nes_cm_node *, void *, u8);
+static void build_rdma0_msg(struct nes_cm_node *, struct nes_qp **);
+
+static void print_core(struct nes_cm_core *core);
+static void record_ird_ord(struct nes_cm_node *, u16, u16);
 
 /* External CM API Interface */
 /* instance of function pointers for client API */
@@ -132,12 +158,20 @@ atomic_t cm_connecteds;
 atomic_t cm_connect_reqs;
 atomic_t cm_rejects;
 
+int nes_add_ref_cm_node(struct nes_cm_node *cm_node)
+{
+	return add_ref_cm_node(cm_node);
+}
 
+int nes_rem_ref_cm_node(struct nes_cm_node *cm_node)
+{
+	return rem_ref_cm_node(cm_node->cm_core, cm_node);
+}
 /**
  * create_event
  */
-static struct nes_cm_event *create_event(struct nes_cm_node *cm_node,
-		enum nes_cm_event_type type)
+static struct nes_cm_event *create_event(struct nes_cm_node *	cm_node,
+					 enum nes_cm_event_type type)
 {
 	struct nes_cm_event *event;
 
@@ -158,11 +192,11 @@ static struct nes_cm_event *create_event(struct nes_cm_node *cm_node,
 	event->cm_info.loc_port = cm_node->loc_port;
 	event->cm_info.cm_id = cm_node->cm_id;
 
-	nes_debug(NES_DBG_CM, "Created event=%p, type=%u, dst_addr=%08x[%x],"
-			" src_addr=%08x[%x]\n",
-			event, type,
-			event->cm_info.loc_addr, event->cm_info.loc_port,
-			event->cm_info.rem_addr, event->cm_info.rem_port);
+	nes_debug(NES_DBG_CM, "cm_node=%p Created event=%p, type=%u, "
+		  "dst_addr=%08x[%x], src_addr=%08x[%x]\n",
+		  cm_node, event, type, event->cm_info.loc_addr,
+		  event->cm_info.loc_port, event->cm_info.rem_addr,
+		  event->cm_info.rem_port);
 
 	nes_cm_post_event(event);
 	return event;
@@ -172,27 +206,50 @@ static struct nes_cm_event *create_event(struct nes_cm_node *cm_node,
 /**
  * send_mpa_request
  */
-static int send_mpa_request(struct nes_cm_node *cm_node)
+static int send_mpa_request(struct nes_cm_node *cm_node, struct sk_buff *skb)
 {
-	struct sk_buff *skb;
-	int ret;
+	u8 start_addr = 0;
+	u8 *start_ptr = &start_addr;
+	u8 **start_buff = &start_ptr;
+	u16 buff_len = 0;
 
-	skb = get_free_pkt(cm_node);
 	if (!skb) {
-		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
+		nes_debug(NES_DBG_CM, "skb set to NULL\n");
 		return -1;
 	}
 
 	/* send an MPA Request frame */
-	form_cm_frame(skb, cm_node, NULL, 0, &cm_node->mpa_frame,
-			cm_node->mpa_frame_size, SET_ACK);
+	cm_build_mpa_frame(cm_node, start_buff, &buff_len, NULL, MPA_KEY_REQUEST);
+	form_cm_frame(skb, cm_node, NULL, 0, *start_buff, buff_len, SET_ACK);
 
-	ret = schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
-	if (ret < 0) {
-		return ret;
+	return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
+}
+
+
+
+static int send_mpa_reject(struct nes_cm_node *cm_node)
+{
+	struct sk_buff *skb = NULL;
+	u8 start_addr = 0;
+	u8 *start_ptr = &start_addr;
+	u8 **start_buff = &start_ptr;
+	u16 buff_len = 0;
+	struct ietf_mpa_v1 *mpa_frame;
+
+	skb = dev_alloc_skb(MAX_CM_BUFFER);
+	if (!skb) {
+		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
+		return -ENOMEM;
 	}
 
-	return 0;
+	/* send an MPA reject frame */
+	cm_build_mpa_frame(cm_node, start_buff, &buff_len, NULL, MPA_KEY_REPLY);
+	mpa_frame = (struct ietf_mpa_v1 *)*start_buff;
+	mpa_frame->flags |= IETF_MPA_FLAGS_REJECT;
+	form_cm_frame(skb, cm_node, NULL, 0, *start_buff, buff_len, SET_ACK | SET_FIN);
+
+	cm_node->state = NES_CM_STATE_FIN_WAIT1;
+	return schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
 }
 
 
@@ -200,65 +257,174 @@ static int send_mpa_request(struct nes_cm_node *cm_node)
  * recv_mpa - process a received TCP pkt, we are expecting an
  * IETF MPA frame
  */
-static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 len)
+static int parse_mpa(struct nes_cm_node *cm_node, u8 *buffer, u32 *type,
+		     u32 len)
 {
-	struct ietf_mpa_frame *mpa_frame;
+	struct ietf_mpa_v1 *mpa_frame;
+	struct ietf_mpa_v2 *mpa_v2_frame;
+	struct ietf_rtr_msg *rtr_msg;
+	int mpa_hdr_len;
+	int priv_data_len;
+
+	*type = NES_MPA_REQUEST_ACCEPT;
 
 	/* assume req frame is in tcp data payload */
-	if (len < sizeof(struct ietf_mpa_frame)) {
+	if (len < sizeof(struct ietf_mpa_v1)) {
 		nes_debug(NES_DBG_CM, "The received ietf buffer was too small (%x)\n", len);
-		return -1;
+		return -EINVAL;
 	}
 
-	mpa_frame = (struct ietf_mpa_frame *)buffer;
-	cm_node->mpa_frame_size = ntohs(mpa_frame->priv_data_len);
+	/* points to the beginning of the frame, which could be MPA V1 or V2 */
+	mpa_frame = (struct ietf_mpa_v1 *)buffer;
+	mpa_hdr_len = sizeof(struct ietf_mpa_v1);
+	priv_data_len = ntohs(mpa_frame->priv_data_len);
 
-	if (cm_node->mpa_frame_size + sizeof(struct ietf_mpa_frame) != len) {
-		nes_debug(NES_DBG_CM, "The received ietf buffer was not right"
-				" complete (%x + %x != %x)\n",
-				cm_node->mpa_frame_size, (u32)sizeof(struct ietf_mpa_frame), len);
-		return -1;
+	/* make sure mpa private data len is less than 512 bytes */
+	if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) {
+		nes_debug(NES_DBG_CM, "The received Length of Private"
+			  " Data field exceeds 512 octets\n");
+		return -EINVAL;
+	}
+	/*
+	 * make sure MPA receiver interoperate with the
+	 * received MPA version and MPA key information
+	 *
+	 */
+	if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) {
+		nes_debug(NES_DBG_CM, "The received mpa version"
+			  " is not supported\n");
+		return -EINVAL;
+	}
+	/*
+	* backwards compatibility only
+	*/
+	if (mpa_frame->rev > cm_node->mpa_frame_rev) {
+		nes_debug(NES_DBG_CM, "The received mpa version"
+			" can not be interoperated\n");
+		return -EINVAL;
+	} else {
+		cm_node->mpa_frame_rev = mpa_frame->rev;
 	}
 
-	/* copy entire MPA frame to our cm_node's frame */
-	memcpy(cm_node->mpa_frame_buf, buffer + sizeof(struct ietf_mpa_frame),
-			cm_node->mpa_frame_size);
+	if (cm_node->state != NES_CM_STATE_MPAREQ_SENT) {
+		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE)) {
+			nes_debug(NES_DBG_CM, "Unexpected MPA Key received \n");
+			return -EINVAL;
+		}
+	} else {
+		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE)) {
+			nes_debug(NES_DBG_CM, "Unexpected MPA Key received \n");
+			return -EINVAL;
+		}
+	}
 
-	return 0;
-}
+	if (priv_data_len + mpa_hdr_len != len) {
+		nes_debug(NES_DBG_CM, "The received ietf buffer was not right"
+			" complete (%x + %x != %x)\n",
+			priv_data_len, mpa_hdr_len, len);
+		return -EINVAL;
+	}
+	/* make sure it does not exceed the max size */
+	if (len > MAX_CM_BUFFER) {
+		nes_debug(NES_DBG_CM, "The received ietf buffer was too large"
+			" (%x + %x != %x)\n",
+			priv_data_len, mpa_hdr_len, len);
+		return -EINVAL;
+	}
+
+	cm_node->mpa_frame_size = priv_data_len;
+
+	switch (mpa_frame->rev) {
+	case IETF_MPA_V2: {
+		u16 ird_size;
+		u16 ord_size;
+		u16 rtr_ctrl_ird;
+		u16 rtr_ctrl_ord;
+
+		mpa_v2_frame = (struct ietf_mpa_v2 *)buffer;
+		mpa_hdr_len += IETF_RTR_MSG_SIZE;
+		cm_node->mpa_frame_size -= IETF_RTR_MSG_SIZE;
+		rtr_msg = &mpa_v2_frame->rtr_msg;
+
+		/* parse rtr message */
+		rtr_ctrl_ird = ntohs(rtr_msg->ctrl_ird);
+		rtr_ctrl_ord = ntohs(rtr_msg->ctrl_ord);
+		ird_size = rtr_ctrl_ird & IETF_NO_IRD_ORD;
+		ord_size = rtr_ctrl_ord & IETF_NO_IRD_ORD;
+
+		if (!(rtr_ctrl_ird & IETF_PEER_TO_PEER)) {
+			/* send reset */
+			return -EINVAL;
+		}
+		if (ird_size == IETF_NO_IRD_ORD || ord_size == IETF_NO_IRD_ORD)
+			cm_node->mpav2_ird_ord = IETF_NO_IRD_ORD;
+
+		if (cm_node->mpav2_ird_ord != IETF_NO_IRD_ORD) {
+			/* responder */
+			if (cm_node->state != NES_CM_STATE_MPAREQ_SENT) {
+				/* we are still negotiating */
+				if (ord_size > NES_MAX_IRD) {
+					cm_node->ird_size = NES_MAX_IRD;
+				} else {
+					cm_node->ird_size = ord_size;
+					if (ord_size == 0 &&
+					(rtr_ctrl_ord & IETF_RDMA0_READ)) {
+						cm_node->ird_size = 1;
+						nes_debug(NES_DBG_CM,
+						"%s: Remote peer doesn't support RDMA0_READ (ord=%u)\n",
+							__func__, ord_size);
+					}
+				}
+				if (ird_size > NES_MAX_ORD)
+					cm_node->ord_size = NES_MAX_ORD;
+				else
+					cm_node->ord_size = ird_size;
+			} else { /* initiator */
+				if (ord_size > NES_MAX_IRD) {
+					nes_debug(NES_DBG_CM,
+					"%s: Unable to support the requested (ord =%u)\n",
+							__func__, ord_size);
+					return -EINVAL;
+				}
+				cm_node->ird_size = ord_size;
 
+				if (ird_size > NES_MAX_ORD) {
+					cm_node->ord_size = NES_MAX_ORD;
+				} else {
+					if (ird_size == 0 &&
+					(rtr_ctrl_ord & IETF_RDMA0_READ)) {
+						nes_debug(NES_DBG_CM,
+						"%s: Remote peer doesn't support RDMA0_READ (ird=%u)\n",
+							__func__, ird_size);
+						return -EINVAL;
+					} else {
+						cm_node->ord_size = ird_size;
+					}
+				}
+			}
+		}
 
-/**
- * handle_exception_pkt - process an exception packet.
- * We have been in a TSA state, and we have now received SW
- * TCP/IP traffic should be a FIN request or IP pkt with options
- */
-static int handle_exception_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb)
-{
-	int ret = 0;
-	struct tcphdr *tcph = tcp_hdr(skb);
+		if (rtr_ctrl_ord & IETF_RDMA0_READ) {
+			cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
 
-	/* first check to see if this a FIN pkt */
-	if (tcph->fin) {
-		/* we need to ACK the FIN request */
-		send_ack(cm_node);
-
-		/* check which side we are (client/server) and set next state accordingly */
-		if (cm_node->tcp_cntxt.client)
-			cm_node->state = NES_CM_STATE_CLOSING;
-		else {
-			/* we are the server side */
-			cm_node->state = NES_CM_STATE_CLOSE_WAIT;
-			/* since this is a self contained CM we don't wait for */
-			/* an APP to close us, just send final FIN immediately */
-			ret = send_fin(cm_node, NULL);
-			cm_node->state = NES_CM_STATE_LAST_ACK;
+		} else if (rtr_ctrl_ord & IETF_RDMA0_WRITE) {
+			cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO;
+		} else {        /* Not supported RDMA0 operation */
+			return -EINVAL;
 		}
-	} else {
-		ret = -EINVAL;
+		break;
+	}
+	case IETF_MPA_V1:
+	default:
+		break;
 	}
 
-	return ret;
+	/* copy entire MPA frame to our cm_node's frame */
+	memcpy(cm_node->mpa_frame_buf, buffer + mpa_hdr_len, cm_node->mpa_frame_size);
+
+	if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT)
+		*type = NES_MPA_REQUEST_REJECT;
+	return 0;
 }
 
 
@@ -266,9 +432,9 @@ static int handle_exception_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb
  * form_cm_frame - get a free packet and build empty frame Use
  * node info to build.
  */
-static struct sk_buff *form_cm_frame(struct sk_buff *skb, struct nes_cm_node *cm_node,
-				     void *options, u32 optionsize, void *data,
-				     u32 datasize, u8 flags)
+static void form_cm_frame(struct sk_buff *skb,
+			  struct nes_cm_node *cm_node, void *options, u32 optionsize,
+			  void *data, u32 datasize, u8 flags)
 {
 	struct tcphdr *tcph;
 	struct iphdr *iph;
@@ -277,14 +443,14 @@ static struct sk_buff *form_cm_frame(struct sk_buff *skb, struct nes_cm_node *cm
 	u16 packetsize = sizeof(*iph);
 
 	packetsize += sizeof(*tcph);
-	packetsize +=  optionsize + datasize;
+	packetsize += optionsize + datasize;
 
+	skb_trim(skb, 0);
 	memset(skb->data, 0x00, ETH_HLEN + sizeof(*iph) + sizeof(*tcph));
 
-	skb->len = 0;
 	buf = skb_put(skb, packetsize + ETH_HLEN);
 
-	ethh = (struct ethhdr *) buf;
+	ethh = (struct ethhdr *)buf;
 	buf += ETH_HLEN;
 
 	iph = (struct iphdr *)buf;
@@ -292,10 +458,12 @@ static struct sk_buff *form_cm_frame(struct sk_buff *skb, struct nes_cm_node *cm
 	tcph = (struct tcphdr *)buf;
 	skb_reset_mac_header(skb);
 	skb_set_network_header(skb, ETH_HLEN);
-	skb_set_transport_header(skb, ETH_HLEN+sizeof(*iph));
+	skb_set_transport_header(skb, ETH_HLEN + sizeof(*iph));
 	buf += sizeof(*tcph);
 
 	skb->ip_summed = CHECKSUM_PARTIAL;
+	if (!(cm_node->netdev->features & NETIF_F_IP_CSUM))
+		skb->ip_summed = CHECKSUM_NONE;
 	skb->protocol = htons(0x800);
 	skb->data_len = 0;
 	skb->mac_len = ETH_HLEN;
@@ -305,37 +473,41 @@ static struct sk_buff *form_cm_frame(struct sk_buff *skb, struct nes_cm_node *cm
 	ethh->h_proto = htons(0x0800);
 
 	iph->version = IPVERSION;
-	iph->ihl = 5;		/* 5 * 4Byte words, IP headr len */
+	iph->ihl = 5;           /* 5 * 4Byte words, IP headr len */
 	iph->tos = 0;
 	iph->tot_len = htons(packetsize);
 	iph->id = htons(++cm_node->tcp_cntxt.loc_id);
 
 	iph->frag_off = htons(0x4000);
 	iph->ttl = 0x40;
-	iph->protocol = 0x06;	/* IPPROTO_TCP */
+	iph->protocol = 0x06;   /* IPPROTO_TCP */
 
-	iph->saddr = htonl(cm_node->loc_addr);
-	iph->daddr = htonl(cm_node->rem_addr);
+	iph->saddr = htonl(cm_node->mapped_loc_addr);
+	iph->daddr = htonl(cm_node->mapped_rem_addr);
 
-	tcph->source = htons(cm_node->loc_port);
-	tcph->dest = htons(cm_node->rem_port);
+	tcph->source = htons(cm_node->mapped_loc_port);
+	tcph->dest = htons(cm_node->mapped_rem_port);
 	tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
 
 	if (flags & SET_ACK) {
 		cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
 		tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
 		tcph->ack = 1;
-	} else
+	} else {
 		tcph->ack_seq = 0;
+	}
 
 	if (flags & SET_SYN) {
 		cm_node->tcp_cntxt.loc_seq_num++;
 		tcph->syn = 1;
-	} else
-		cm_node->tcp_cntxt.loc_seq_num += datasize;	/* data (no headers) */
+	} else {
+		cm_node->tcp_cntxt.loc_seq_num += datasize;
+	}
 
-	if (flags & SET_FIN)
+	if (flags & SET_FIN) {
+		cm_node->tcp_cntxt.loc_seq_num++;
 		tcph->fin = 1;
+	}
 
 	if (flags & SET_RST)
 		tcph->rst = 1;
@@ -351,10 +523,101 @@ static struct sk_buff *form_cm_frame(struct sk_buff *skb, struct nes_cm_node *cm
 
 	skb_shinfo(skb)->nr_frags = 0;
 	cm_packets_created++;
+}
+
+/*
+ * nes_create_sockaddr - Record ip addr and tcp port in a sockaddr struct
+ */
+static void nes_create_sockaddr(__be32 ip_addr, __be16 port,
+				struct sockaddr_storage *addr)
+{
+	struct sockaddr_in *nes_sockaddr = (struct sockaddr_in *)addr;
+	nes_sockaddr->sin_family = AF_INET;
+	memcpy(&nes_sockaddr->sin_addr.s_addr, &ip_addr, sizeof(__be32));
+	nes_sockaddr->sin_port = port;
+}
+
+/*
+ * nes_create_mapinfo - Create a mapinfo object in the port mapper data base
+ */
+static int nes_create_mapinfo(struct nes_cm_info *cm_info)
+{
+	struct sockaddr_storage local_sockaddr;
+	struct sockaddr_storage mapped_sockaddr;
+
+	nes_create_sockaddr(htonl(cm_info->loc_addr), htons(cm_info->loc_port),
+				&local_sockaddr);
+	nes_create_sockaddr(htonl(cm_info->mapped_loc_addr),
+			htons(cm_info->mapped_loc_port), &mapped_sockaddr);
+
+	return iwpm_create_mapinfo(&local_sockaddr,
+				&mapped_sockaddr, RDMA_NL_NES);
+}
+
+/*
+ * nes_remove_mapinfo - Remove a mapinfo object from the port mapper data base
+ *                      and send a remove mapping op message to
+ *                      the userspace port mapper
+ */
+static int nes_remove_mapinfo(u32 loc_addr, u16 loc_port,
+			u32 mapped_loc_addr, u16 mapped_loc_port)
+{
+	struct sockaddr_storage local_sockaddr;
+	struct sockaddr_storage mapped_sockaddr;
+
+	nes_create_sockaddr(htonl(loc_addr), htons(loc_port), &local_sockaddr);
+	nes_create_sockaddr(htonl(mapped_loc_addr), htons(mapped_loc_port),
+				&mapped_sockaddr);
+
+	iwpm_remove_mapinfo(&local_sockaddr, &mapped_sockaddr);
+	return iwpm_remove_mapping(&local_sockaddr, RDMA_NL_NES);
+}
+
+/*
+ * nes_form_pm_msg - Form a port mapper message with mapping info
+ */
+static void nes_form_pm_msg(struct nes_cm_info *cm_info,
+				struct iwpm_sa_data *pm_msg)
+{
+	nes_create_sockaddr(htonl(cm_info->loc_addr), htons(cm_info->loc_port),
+				&pm_msg->loc_addr);
+	nes_create_sockaddr(htonl(cm_info->rem_addr), htons(cm_info->rem_port),
+				&pm_msg->rem_addr);
+}
 
-	return skb;
+/*
+ * nes_form_reg_msg - Form a port mapper message with dev info
+ */
+static void nes_form_reg_msg(struct nes_vnic *nesvnic,
+			struct iwpm_dev_data *pm_msg)
+{
+	memcpy(pm_msg->dev_name, nesvnic->nesibdev->ibdev.name,
+				IWPM_DEVNAME_SIZE);
+	memcpy(pm_msg->if_name, nesvnic->netdev->name, IWPM_IFNAME_SIZE);
 }
 
+/*
+ * nes_record_pm_msg - Save the received mapping info
+ */
+static void nes_record_pm_msg(struct nes_cm_info *cm_info,
+			struct iwpm_sa_data *pm_msg)
+{
+	struct sockaddr_in *mapped_loc_addr =
+			(struct sockaddr_in *)&pm_msg->mapped_loc_addr;
+	struct sockaddr_in *mapped_rem_addr =
+			(struct sockaddr_in *)&pm_msg->mapped_rem_addr;
+
+	if (mapped_loc_addr->sin_family == AF_INET) {
+		cm_info->mapped_loc_addr =
+			ntohl(mapped_loc_addr->sin_addr.s_addr);
+		cm_info->mapped_loc_port = ntohs(mapped_loc_addr->sin_port);
+	}
+	if (mapped_rem_addr->sin_family == AF_INET) {
+		cm_info->mapped_rem_addr =
+			ntohl(mapped_rem_addr->sin_addr.s_addr);
+		cm_info->mapped_rem_port = ntohs(mapped_rem_addr->sin_port);
+	}
+}
 
 /**
  * print_core - dump a cm core
@@ -367,9 +630,8 @@ static void print_core(struct nes_cm_core *core)
 		return;
 	nes_debug(NES_DBG_CM, "---------------------------------------------\n");
 
-	nes_debug(NES_DBG_CM, "State         : %u \n",  core->state);
+	nes_debug(NES_DBG_CM, "State         : %u \n", core->state);
 
-	nes_debug(NES_DBG_CM, "Tx Free cnt   : %u \n", skb_queue_len(&core->tx_free_list));
 	nes_debug(NES_DBG_CM, "Listen Nodes  : %u \n", atomic_read(&core->listen_node_cnt));
 	nes_debug(NES_DBG_CM, "Active Nodes  : %u \n", atomic_read(&core->node_cnt));
 
@@ -378,6 +640,162 @@ static void print_core(struct nes_cm_core *core)
 	nes_debug(NES_DBG_CM, "-------------- end core ---------------\n");
 }
 
+static void record_ird_ord(struct nes_cm_node *cm_node,
+					u16 conn_ird, u16 conn_ord)
+{
+	if (conn_ird > NES_MAX_IRD)
+		conn_ird = NES_MAX_IRD;
+
+	if (conn_ord > NES_MAX_ORD)
+		conn_ord = NES_MAX_ORD;
+
+	cm_node->ird_size = conn_ird;
+	cm_node->ord_size = conn_ord;
+}
+
+/**
+ * cm_build_mpa_frame - build a MPA V1 frame or MPA V2 frame
+ */
+static int cm_build_mpa_frame(struct nes_cm_node *cm_node, u8 **start_buff,
+			      u16 *buff_len, u8 *pci_mem, u8 mpa_key)
+{
+	int ret = 0;
+
+	*start_buff = (pci_mem) ? pci_mem : &cm_node->mpa_frame_buf[0];
+
+	switch (cm_node->mpa_frame_rev) {
+	case IETF_MPA_V1:
+		*start_buff = (u8 *)*start_buff + sizeof(struct ietf_rtr_msg);
+		*buff_len = sizeof(struct ietf_mpa_v1) + cm_node->mpa_frame_size;
+		build_mpa_v1(cm_node, *start_buff, mpa_key);
+		break;
+	case IETF_MPA_V2:
+		*buff_len = sizeof(struct ietf_mpa_v2) + cm_node->mpa_frame_size;
+		build_mpa_v2(cm_node, *start_buff, mpa_key);
+		break;
+	default:
+		ret = -EINVAL;
+	}
+	return ret;
+}
+
+/**
+ * build_mpa_v2 - build a MPA V2 frame
+ */
+static void build_mpa_v2(struct nes_cm_node *cm_node,
+			 void *start_addr, u8 mpa_key)
+{
+	struct ietf_mpa_v2 *mpa_frame = (struct ietf_mpa_v2 *)start_addr;
+	struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg;
+	u16 ctrl_ird;
+	u16 ctrl_ord;
+
+	/* initialize the upper 5 bytes of the frame */
+	build_mpa_v1(cm_node, start_addr, mpa_key);
+	mpa_frame->flags |= IETF_MPA_V2_FLAG; /* set a bit to indicate MPA V2 */
+	mpa_frame->priv_data_len += htons(IETF_RTR_MSG_SIZE);
+
+	/* initialize RTR msg */
+	if (cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) {
+		ctrl_ird = IETF_NO_IRD_ORD;
+		ctrl_ord = IETF_NO_IRD_ORD;
+	} else {
+		ctrl_ird = cm_node->ird_size & IETF_NO_IRD_ORD;
+		ctrl_ord = cm_node->ord_size & IETF_NO_IRD_ORD;
+	}
+	ctrl_ird |= IETF_PEER_TO_PEER;
+	ctrl_ird |= IETF_FLPDU_ZERO_LEN;
+
+	switch (mpa_key) {
+	case MPA_KEY_REQUEST:
+		ctrl_ord |= IETF_RDMA0_WRITE;
+		ctrl_ord |= IETF_RDMA0_READ;
+		break;
+	case MPA_KEY_REPLY:
+		switch (cm_node->send_rdma0_op) {
+		case SEND_RDMA_WRITE_ZERO:
+			ctrl_ord |= IETF_RDMA0_WRITE;
+			break;
+		case SEND_RDMA_READ_ZERO:
+			ctrl_ord |= IETF_RDMA0_READ;
+			break;
+		}
+	}
+	rtr_msg->ctrl_ird = htons(ctrl_ird);
+	rtr_msg->ctrl_ord = htons(ctrl_ord);
+}
+
+/**
+ * build_mpa_v1 - build a MPA V1 frame
+ */
+static void build_mpa_v1(struct nes_cm_node *cm_node, void *start_addr, u8 mpa_key)
+{
+	struct ietf_mpa_v1 *mpa_frame = (struct ietf_mpa_v1 *)start_addr;
+
+	switch (mpa_key) {
+	case MPA_KEY_REQUEST:
+		memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
+		break;
+	case MPA_KEY_REPLY:
+		memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
+		break;
+	}
+	mpa_frame->flags = IETF_MPA_FLAGS_CRC;
+	mpa_frame->rev = cm_node->mpa_frame_rev;
+	mpa_frame->priv_data_len = htons(cm_node->mpa_frame_size);
+}
+
+static void build_rdma0_msg(struct nes_cm_node *cm_node, struct nes_qp **nesqp_addr)
+{
+	u64 u64temp;
+	struct nes_qp *nesqp = *nesqp_addr;
+	struct nes_hw_qp_wqe *wqe = &nesqp->hwqp.sq_vbase[0];
+
+	u64temp = (unsigned long)nesqp->nesuqp_addr;
+	u64temp |= NES_SW_CONTEXT_ALIGN >> 1;
+	set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX, u64temp);
+
+	wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_LOW_IDX] = 0;
+	wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_HIGH_IDX] = 0;
+
+	switch (cm_node->send_rdma0_op) {
+	case SEND_RDMA_WRITE_ZERO:
+		nes_debug(NES_DBG_CM, "Sending first write.\n");
+		wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
+			cpu_to_le32(NES_IWARP_SQ_OP_RDMAW);
+		wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] = 0;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] = 0;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 0;
+		break;
+
+	case SEND_RDMA_READ_ZERO:
+	default:
+		if (cm_node->send_rdma0_op != SEND_RDMA_READ_ZERO)
+			WARN(1, "Unsupported RDMA0 len operation=%u\n",
+			     cm_node->send_rdma0_op);
+		nes_debug(NES_DBG_CM, "Sending first rdma operation.\n");
+		wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
+			cpu_to_le32(NES_IWARP_SQ_OP_RDMAR);
+		wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX] = 1;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_TO_HIGH_IDX] = 0;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX] = 0;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_STAG_IDX] = 1;
+		wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 1;
+		break;
+	}
+
+	if (nesqp->sq_kmapped) {
+		nesqp->sq_kmapped = 0;
+		kunmap(nesqp->page);
+	}
+
+	/*use the reserved spot on the WQ for the extra first WQE*/
+	nesqp->nesqp_context->ird_ord_sizes &= cpu_to_le32(~(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
+							     NES_QPCONTEXT_ORDIRD_WRPDU |
+							     NES_QPCONTEXT_ORDIRD_ALSMM));
+	nesqp->skip_lsmm = 1;
+	nesqp->hwqp.sq_tail = 0;
+}
 
 /**
  * schedule_nes_timer
@@ -385,20 +803,17 @@ static void print_core(struct nes_cm_core *core)
  *			rem_ref_cm_node(cm_core, cm_node);add_ref_cm_node(cm_node);
  */
 int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb,
-		enum nes_timer_type type, int send_retrans,
-		int close_when_complete)
+		       enum nes_timer_type type, int send_retrans,
+		       int close_when_complete)
 {
-	unsigned long  flags;
-	struct nes_cm_core *cm_core;
+	unsigned long flags;
+	struct nes_cm_core *cm_core = cm_node->cm_core;
 	struct nes_timer_entry *new_send;
 	int ret = 0;
-	u32 was_timer_set;
 
-	if (!cm_node)
-		return -EINVAL;
 	new_send = kzalloc(sizeof(*new_send), GFP_ATOMIC);
 	if (!new_send)
-		return -1;
+		return -ENOMEM;
 
 	/* new_send->timetosend = currenttime */
 	new_send->retrycount = NES_DEFAULT_RETRYS;
@@ -411,255 +826,262 @@ int schedule_nes_timer(struct nes_cm_node *cm_node, struct sk_buff *skb,
 	new_send->close_when_complete = close_when_complete;
 
 	if (type == NES_TIMER_TYPE_CLOSE) {
-		new_send->timetosend += (HZ/2);	/* TODO: decide on the correct value here */
-		spin_lock_irqsave(&cm_node->recv_list_lock, flags);
-		list_add_tail(&new_send->list, &cm_node->recv_list);
-		spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
+		new_send->timetosend += (HZ / 10);
+		if (cm_node->recv_entry) {
+			kfree(new_send);
+			WARN_ON(1);
+			return -EINVAL;
+		}
+		cm_node->recv_entry = new_send;
 	}
 
 	if (type == NES_TIMER_TYPE_SEND) {
 		new_send->seq_num = ntohl(tcp_hdr(skb)->seq);
 		atomic_inc(&new_send->skb->users);
+		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
+		cm_node->send_entry = new_send;
+		add_ref_cm_node(cm_node);
+		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
+		new_send->timetosend = jiffies + NES_RETRY_TIMEOUT;
 
 		ret = nes_nic_cm_xmit(new_send->skb, cm_node->netdev);
 		if (ret != NETDEV_TX_OK) {
-			nes_debug(NES_DBG_CM, "Error sending packet %p (jiffies = %lu)\n",
-					new_send, jiffies);
-			atomic_dec(&new_send->skb->users);
+			nes_debug(NES_DBG_CM, "Error sending packet %p "
+				  "(jiffies = %lu)\n", new_send, jiffies);
 			new_send->timetosend = jiffies;
+			ret = NETDEV_TX_OK;
 		} else {
 			cm_packets_sent++;
 			if (!send_retrans) {
+				cleanup_retrans_entry(cm_node);
 				if (close_when_complete)
-					rem_ref_cm_node(cm_node->cm_core, cm_node);
-				dev_kfree_skb_any(new_send->skb);
-				kfree(new_send);
+					rem_ref_cm_node(cm_core, cm_node);
 				return ret;
 			}
-			new_send->timetosend = jiffies + NES_RETRY_TIMEOUT;
 		}
-		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-		list_add_tail(&new_send->list, &cm_node->retrans_list);
-		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 	}
-	if (type == NES_TIMER_TYPE_RECV) {
-		new_send->seq_num = ntohl(tcp_hdr(skb)->seq);
-		new_send->timetosend = jiffies;
-		spin_lock_irqsave(&cm_node->recv_list_lock, flags);
-		list_add_tail(&new_send->list, &cm_node->recv_list);
-		spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
-	}
-	cm_core = cm_node->cm_core;
 
-	was_timer_set = timer_pending(&cm_core->tcp_timer);
-
-	if (!was_timer_set) {
-		cm_core->tcp_timer.expires = new_send->timetosend;
-		add_timer(&cm_core->tcp_timer);
-	}
+	if (!timer_pending(&cm_core->tcp_timer))
+		mod_timer(&cm_core->tcp_timer, new_send->timetosend);
 
 	return ret;
 }
 
+static void nes_retrans_expired(struct nes_cm_node *cm_node)
+{
+	struct iw_cm_id *cm_id = cm_node->cm_id;
+	enum nes_cm_node_state state = cm_node->state;
+	cm_node->state = NES_CM_STATE_CLOSED;
+
+	switch (state) {
+	case NES_CM_STATE_SYN_RCVD:
+	case NES_CM_STATE_CLOSING:
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		break;
+	case NES_CM_STATE_LAST_ACK:
+	case NES_CM_STATE_FIN_WAIT1:
+		if (cm_node->cm_id)
+			cm_id->rem_ref(cm_id);
+		send_reset(cm_node, NULL);
+		break;
+	default:
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, NULL);
+		create_event(cm_node, NES_CM_EVENT_ABORTED);
+	}
+}
+
+static void handle_recv_entry(struct nes_cm_node *cm_node, u32 rem_node)
+{
+	struct nes_timer_entry *recv_entry = cm_node->recv_entry;
+	struct iw_cm_id *cm_id = cm_node->cm_id;
+	struct nes_qp *nesqp;
+	unsigned long qplockflags;
+
+	if (!recv_entry)
+		return;
+	nesqp = (struct nes_qp *)recv_entry->skb;
+	if (nesqp) {
+		spin_lock_irqsave(&nesqp->lock, qplockflags);
+		if (nesqp->cm_id) {
+			nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, "
+				  "refcount = %d: HIT A "
+				  "NES_TIMER_TYPE_CLOSE with something "
+				  "to do!!!\n", nesqp->hwqp.qp_id, cm_id,
+				  atomic_read(&nesqp->refcount));
+			nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
+			nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
+			nesqp->ibqp_state = IB_QPS_ERR;
+			spin_unlock_irqrestore(&nesqp->lock, qplockflags);
+			nes_cm_disconn(nesqp);
+		} else {
+			spin_unlock_irqrestore(&nesqp->lock, qplockflags);
+			nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, "
+				  "refcount = %d: HIT A "
+				  "NES_TIMER_TYPE_CLOSE with nothing "
+				  "to do!!!\n", nesqp->hwqp.qp_id, cm_id,
+				  atomic_read(&nesqp->refcount));
+		}
+	} else if (rem_node) {
+		/* TIME_WAIT state */
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+	}
+	if (cm_node->cm_id)
+		cm_id->rem_ref(cm_id);
+	kfree(recv_entry);
+	cm_node->recv_entry = NULL;
+}
 
 /**
  * nes_cm_timer_tick
  */
 static void nes_cm_timer_tick(unsigned long pass)
 {
-	unsigned long flags, qplockflags;
+	unsigned long flags;
 	unsigned long nexttimeout = jiffies + NES_LONG_TIME;
-	struct iw_cm_id *cm_id;
 	struct nes_cm_node *cm_node;
 	struct nes_timer_entry *send_entry, *recv_entry;
-	struct list_head *list_core, *list_core_temp;
-	struct list_head *list_node, *list_node_temp;
+	struct list_head *list_core_temp;
+	struct list_head *list_node;
 	struct nes_cm_core *cm_core = g_cm_core;
-	struct nes_qp *nesqp;
-	struct sk_buff *skb;
 	u32 settimer = 0;
+	unsigned long timetosend;
 	int ret = NETDEV_TX_OK;
-	int    node_done;
 
+	struct list_head timer_list;
+
+	INIT_LIST_HEAD(&timer_list);
 	spin_lock_irqsave(&cm_core->ht_lock, flags);
 
-	list_for_each_safe(list_node, list_core_temp, &cm_core->connected_nodes) {
+	list_for_each_safe(list_node, list_core_temp,
+			   &cm_core->connected_nodes) {
 		cm_node = container_of(list_node, struct nes_cm_node, list);
-		add_ref_cm_node(cm_node);
-		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
-		spin_lock_irqsave(&cm_node->recv_list_lock, flags);
-		list_for_each_safe(list_core, list_node_temp, &cm_node->recv_list) {
-			recv_entry = container_of(list_core, struct nes_timer_entry, list);
-			if ((time_after(recv_entry->timetosend, jiffies)) &&
-					(recv_entry->type == NES_TIMER_TYPE_CLOSE)) {
-				if (nexttimeout > recv_entry->timetosend || !settimer) {
+		if ((cm_node->recv_entry) || (cm_node->send_entry)) {
+			add_ref_cm_node(cm_node);
+			list_add(&cm_node->timer_entry, &timer_list);
+		}
+	}
+	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
+
+	list_for_each_safe(list_node, list_core_temp, &timer_list) {
+		cm_node = container_of(list_node, struct nes_cm_node,
+				       timer_entry);
+		recv_entry = cm_node->recv_entry;
+
+		if (recv_entry) {
+			if (time_after(recv_entry->timetosend, jiffies)) {
+				if (nexttimeout > recv_entry->timetosend ||
+				    !settimer) {
 					nexttimeout = recv_entry->timetosend;
 					settimer = 1;
 				}
-				continue;
-			}
-			list_del(&recv_entry->list);
-			cm_id = cm_node->cm_id;
-			spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
-			if (recv_entry->type == NES_TIMER_TYPE_CLOSE) {
-				nesqp = (struct nes_qp *)recv_entry->skb;
-				spin_lock_irqsave(&nesqp->lock, qplockflags);
-				if (nesqp->cm_id) {
-					nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, refcount = %d: "
-							"****** HIT A NES_TIMER_TYPE_CLOSE"
-							" with something to do!!! ******\n",
-							nesqp->hwqp.qp_id, cm_id,
-							atomic_read(&nesqp->refcount));
-					nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-					nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
-					nesqp->ibqp_state = IB_QPS_ERR;
-					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_cm_disconn(nesqp);
-				} else {
-					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_debug(NES_DBG_CM, "QP%u: cm_id = %p, refcount = %d:"
-							" ****** HIT A NES_TIMER_TYPE_CLOSE"
-							" with nothing to do!!! ******\n",
-							nesqp->hwqp.qp_id, cm_id,
-							atomic_read(&nesqp->refcount));
-					nes_rem_ref(&nesqp->ibqp);
-				}
-				if (cm_id)
-					cm_id->rem_ref(cm_id);
+			} else {
+				handle_recv_entry(cm_node, 1);
 			}
-			kfree(recv_entry);
-			spin_lock_irqsave(&cm_node->recv_list_lock, flags);
 		}
-		spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
 
 		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-		node_done = 0;
-		list_for_each_safe(list_core, list_node_temp, &cm_node->retrans_list) {
-			if (node_done) {
+		do {
+			send_entry = cm_node->send_entry;
+			if (!send_entry)
 				break;
-			}
-			send_entry = container_of(list_core, struct nes_timer_entry, list);
 			if (time_after(send_entry->timetosend, jiffies)) {
 				if (cm_node->state != NES_CM_STATE_TSA) {
-					if ((nexttimeout > send_entry->timetosend) || !settimer) {
-						nexttimeout = send_entry->timetosend;
+					if ((nexttimeout >
+					     send_entry->timetosend) ||
+					    !settimer) {
+						nexttimeout =
+							send_entry->timetosend;
 						settimer = 1;
 					}
-					node_done = 1;
-					continue;
 				} else {
-					list_del(&send_entry->list);
-					skb = send_entry->skb;
-					spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-					dev_kfree_skb_any(skb);
-					kfree(send_entry);
-					spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-					continue;
+					free_retrans_entry(cm_node);
 				}
+				break;
 			}
-			if (send_entry->type == NES_TIMER_NODE_CLEANUP) {
-				list_del(&send_entry->list);
-				spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-				kfree(send_entry);
-				spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-				continue;
-			}
-			if ((send_entry->seq_num < cm_node->tcp_cntxt.rem_ack_num) ||
-					(cm_node->state == NES_CM_STATE_TSA) ||
-					(cm_node->state == NES_CM_STATE_CLOSED)) {
-				skb = send_entry->skb;
-				list_del(&send_entry->list);
-				spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-				kfree(send_entry);
-				dev_kfree_skb_any(skb);
-				spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-				continue;
+
+			if ((cm_node->state == NES_CM_STATE_TSA) ||
+			    (cm_node->state == NES_CM_STATE_CLOSED)) {
+				free_retrans_entry(cm_node);
+				break;
 			}
 
-			if (!send_entry->retranscount || !send_entry->retrycount) {
+			if (!send_entry->retranscount ||
+			    !send_entry->retrycount) {
 				cm_packets_dropped++;
-				skb = send_entry->skb;
-				list_del(&send_entry->list);
-				spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-				dev_kfree_skb_any(skb);
-				kfree(send_entry);
-				if (cm_node->state == NES_CM_STATE_SYN_RCVD) {
-					/* this node never even generated an indication up to the cm */
-					rem_ref_cm_node(cm_core, cm_node);
-				} else {
-					cm_node->state = NES_CM_STATE_CLOSED;
-					create_event(cm_node, NES_CM_EVENT_ABORTED);
-				}
-				spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-				continue;
+				free_retrans_entry(cm_node);
+
+				spin_unlock_irqrestore(
+					&cm_node->retrans_list_lock, flags);
+				nes_retrans_expired(cm_node);
+				cm_node->state = NES_CM_STATE_CLOSED;
+				spin_lock_irqsave(&cm_node->retrans_list_lock,
+						  flags);
+				break;
 			}
-			/* this seems like the correct place, but leave send entry unprotected */
-			/* spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags); */
 			atomic_inc(&send_entry->skb->users);
 			cm_packets_retrans++;
-			nes_debug(NES_DBG_CM, "Retransmitting send_entry %p for node %p,"
-					" jiffies = %lu, time to send =  %lu, retranscount = %u, "
-					"send_entry->seq_num = 0x%08X, cm_node->tcp_cntxt.rem_ack_num = 0x%08X\n",
-					send_entry, cm_node, jiffies, send_entry->timetosend, send_entry->retranscount,
-					send_entry->seq_num, cm_node->tcp_cntxt.rem_ack_num);
-
-			spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
+			nes_debug(NES_DBG_CM, "Retransmitting send_entry %p "
+				  "for node %p, jiffies = %lu, time to send = "
+				  "%lu, retranscount = %u, send_entry->seq_num = "
+				  "0x%08X, cm_node->tcp_cntxt.rem_ack_num = "
+				  "0x%08X\n", send_entry, cm_node, jiffies,
+				  send_entry->timetosend,
+				  send_entry->retranscount,
+				  send_entry->seq_num,
+				  cm_node->tcp_cntxt.rem_ack_num);
+
+			spin_unlock_irqrestore(&cm_node->retrans_list_lock,
+					       flags);
 			ret = nes_nic_cm_xmit(send_entry->skb, cm_node->netdev);
+			spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 			if (ret != NETDEV_TX_OK) {
+				nes_debug(NES_DBG_CM, "rexmit failed for "
+					  "node=%p\n", cm_node);
 				cm_packets_bounced++;
-				atomic_dec(&send_entry->skb->users);
 				send_entry->retrycount--;
 				nexttimeout = jiffies + NES_SHORT_TIME;
 				settimer = 1;
-				node_done = 1;
-				spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-				continue;
+				break;
 			} else {
 				cm_packets_sent++;
 			}
-			spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-			list_del(&send_entry->list);
-			nes_debug(NES_DBG_CM, "Packet Sent: retrans count = %u, retry count = %u.\n",
-					send_entry->retranscount, send_entry->retrycount);
+			nes_debug(NES_DBG_CM, "Packet Sent: retrans count = "
+				  "%u, retry count = %u.\n",
+				  send_entry->retranscount,
+				  send_entry->retrycount);
 			if (send_entry->send_retrans) {
 				send_entry->retranscount--;
-				send_entry->timetosend = jiffies + NES_RETRY_TIMEOUT;
-				if (nexttimeout > send_entry->timetosend || !settimer) {
+				timetosend = (NES_RETRY_TIMEOUT <<
+					      (NES_DEFAULT_RETRANS - send_entry->retranscount));
+
+				send_entry->timetosend = jiffies +
+							 min(timetosend, NES_MAX_TIMEOUT);
+				if (nexttimeout > send_entry->timetosend ||
+				    !settimer) {
 					nexttimeout = send_entry->timetosend;
 					settimer = 1;
 				}
-				list_add(&send_entry->list, &cm_node->retrans_list);
-				continue;
 			} else {
 				int close_when_complete;
-				skb = send_entry->skb;
-				close_when_complete = send_entry->close_when_complete;
-				spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-				if (close_when_complete) {
-					BUG_ON(atomic_read(&cm_node->ref_count) == 1);
-					rem_ref_cm_node(cm_core, cm_node);
-				}
-				dev_kfree_skb_any(skb);
-				kfree(send_entry);
-				spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-				continue;
+				close_when_complete =
+					send_entry->close_when_complete;
+				nes_debug(NES_DBG_CM, "cm_node=%p state=%d\n",
+					  cm_node, cm_node->state);
+				free_retrans_entry(cm_node);
+				if (close_when_complete)
+					rem_ref_cm_node(cm_node->cm_core,
+							cm_node);
 			}
-		}
-		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-
-		rem_ref_cm_node(cm_core, cm_node);
+		} while (0);
 
-		spin_lock_irqsave(&cm_core->ht_lock, flags);
-		if (ret != NETDEV_TX_OK)
-			break;
+		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
 	}
-	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 
 	if (settimer) {
-		if (!timer_pending(&cm_core->tcp_timer)) {
-			cm_core->tcp_timer.expires  = nexttimeout;
-			add_timer(&cm_core->tcp_timer);
-		}
+		if (!timer_pending(&cm_core->tcp_timer))
+			mod_timer(&cm_core->tcp_timer, nexttimeout);
 	}
 }
 
@@ -667,14 +1089,14 @@ static void nes_cm_timer_tick(unsigned long pass)
 /**
  * send_syn
  */
-static int send_syn(struct nes_cm_node *cm_node, u32 sendack)
+static int send_syn(struct nes_cm_node *cm_node, u32 sendack,
+		    struct sk_buff *skb)
 {
 	int ret;
 	int flags = SET_SYN;
-	struct sk_buff *skb;
 	char optionsbuffer[sizeof(struct option_mss) +
-			sizeof(struct option_windowscale) +
-			sizeof(struct option_base) + 1];
+			   sizeof(struct option_windowscale) + sizeof(struct option_base) +
+			   TCP_OPTIONS_PADDING];
 
 	int optionssize = 0;
 	/* Sending MSS option */
@@ -695,8 +1117,7 @@ static int send_syn(struct nes_cm_node *cm_node, u32 sendack)
 	options->as_windowscale.shiftcount = cm_node->tcp_cntxt.rcv_wscale;
 	optionssize += sizeof(struct option_windowscale);
 
-	if (sendack && !(NES_DRV_OPT_SUPRESS_OPTION_BC & nes_drv_opt)
-			) {
+	if (sendack && !(NES_DRV_OPT_SUPRESS_OPTION_BC & nes_drv_opt)) {
 		options = (union all_known_options *)&optionsbuffer[optionssize];
 		options->as_base.optionnum = OPTION_NUMBER_WRITE0;
 		options->as_base.length = sizeof(struct option_base);
@@ -714,7 +1135,8 @@ static int send_syn(struct nes_cm_node *cm_node, u32 sendack)
 	options->as_end = OPTION_NUMBER_END;
 	optionssize += 1;
 
-	skb = get_free_pkt(cm_node);
+	if (!skb)
+		skb = dev_alloc_skb(MAX_CM_BUFFER);
 	if (!skb) {
 		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
 		return -1;
@@ -733,18 +1155,18 @@ static int send_syn(struct nes_cm_node *cm_node, u32 sendack)
 /**
  * send_reset
  */
-static int send_reset(struct nes_cm_node *cm_node)
+static int send_reset(struct nes_cm_node *cm_node, struct sk_buff *skb)
 {
 	int ret;
-	struct sk_buff *skb = get_free_pkt(cm_node);
 	int flags = SET_RST | SET_ACK;
 
+	if (!skb)
+		skb = dev_alloc_skb(MAX_CM_BUFFER);
 	if (!skb) {
 		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
-		return -1;
+		return -ENOMEM;
 	}
 
-	add_ref_cm_node(cm_node);
 	form_cm_frame(skb, cm_node, NULL, 0, NULL, 0, flags);
 	ret = schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 0, 1);
 
@@ -755,10 +1177,12 @@ static int send_reset(struct nes_cm_node *cm_node)
 /**
  * send_ack
  */
-static int send_ack(struct nes_cm_node *cm_node)
+static int send_ack(struct nes_cm_node *cm_node, struct sk_buff *skb)
 {
 	int ret;
-	struct sk_buff *skb = get_free_pkt(cm_node);
+
+	if (!skb)
+		skb = dev_alloc_skb(MAX_CM_BUFFER);
 
 	if (!skb) {
 		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
@@ -781,7 +1205,7 @@ static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb)
 
 	/* if we didn't get a frame get one */
 	if (!skb)
-		skb = get_free_pkt(cm_node);
+		skb = dev_alloc_skb(MAX_CM_BUFFER);
 
 	if (!skb) {
 		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
@@ -796,76 +1220,32 @@ static int send_fin(struct nes_cm_node *cm_node, struct sk_buff *skb)
 
 
 /**
- * get_free_pkt
- */
-static struct sk_buff *get_free_pkt(struct nes_cm_node *cm_node)
-{
-	struct sk_buff *skb, *new_skb;
-
-	/* check to see if we need to repopulate the free tx pkt queue */
-	if (skb_queue_len(&cm_node->cm_core->tx_free_list) < NES_CM_FREE_PKT_LO_WATERMARK) {
-		while (skb_queue_len(&cm_node->cm_core->tx_free_list) <
-				cm_node->cm_core->free_tx_pkt_max) {
-			/* replace the frame we took, we won't get it back */
-			new_skb = dev_alloc_skb(cm_node->cm_core->mtu);
-			BUG_ON(!new_skb);
-			/* add a replacement frame to the free tx list head */
-			skb_queue_head(&cm_node->cm_core->tx_free_list, new_skb);
-		}
-	}
-
-	skb = skb_dequeue(&cm_node->cm_core->tx_free_list);
-
-	return skb;
-}
-
-
-/**
- * make_hashkey - generate hash key from node tuple
- */
-static inline int make_hashkey(u16 loc_port, nes_addr_t loc_addr, u16 rem_port,
-		nes_addr_t rem_addr)
-{
-	u32 hashkey = 0;
-
-	hashkey = loc_addr + rem_addr + loc_port + rem_port;
-	hashkey = (hashkey % NES_CM_HASHTABLE_SIZE);
-
-	return hashkey;
-}
-
-
-/**
  * find_node - find a cm node that matches the reference cm node
  */
 static struct nes_cm_node *find_node(struct nes_cm_core *cm_core,
-		u16 rem_port, nes_addr_t rem_addr, u16 loc_port, nes_addr_t loc_addr)
+				     u16 rem_port, nes_addr_t rem_addr, u16 loc_port, nes_addr_t loc_addr)
 {
 	unsigned long flags;
-	u32 hashkey;
 	struct list_head *hte;
 	struct nes_cm_node *cm_node;
 
-	/* make a hash index key for this packet */
-	hashkey = make_hashkey(loc_port, loc_addr, rem_port, rem_addr);
-
 	/* get a handle on the hte */
 	hte = &cm_core->connected_nodes;
 
-	nes_debug(NES_DBG_CM, "Searching for an owner node: " NIPQUAD_FMT ":%x from core %p->%p\n",
-		  HIPQUAD(loc_addr), loc_port, cm_core, hte);
-
 	/* walk list and find cm_node associated with this session ID */
 	spin_lock_irqsave(&cm_core->ht_lock, flags);
 	list_for_each_entry(cm_node, hte, list) {
 		/* compare quad, return node handle if a match */
 		nes_debug(NES_DBG_CM, "finding node %x:%x =? %x:%x ^ %x:%x =? %x:%x\n",
-				cm_node->loc_addr, cm_node->loc_port,
-				loc_addr, loc_port,
-				cm_node->rem_addr, cm_node->rem_port,
-				rem_addr, rem_port);
-		if ((cm_node->loc_addr == loc_addr) && (cm_node->loc_port == loc_port) &&
-				(cm_node->rem_addr == rem_addr) && (cm_node->rem_port == rem_port)) {
+			  cm_node->loc_addr, cm_node->loc_port,
+			  loc_addr, loc_port,
+			  cm_node->rem_addr, cm_node->rem_port,
+			  rem_addr, rem_port);
+		if ((cm_node->mapped_loc_addr == loc_addr) &&
+			(cm_node->mapped_loc_port == loc_port) &&
+			(cm_node->mapped_rem_addr == rem_addr) &&
+			(cm_node->mapped_rem_port == rem_port)) {
+
 			add_ref_cm_node(cm_node);
 			spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 			return cm_node;
@@ -882,19 +1262,29 @@ static struct nes_cm_node *find_node(struct nes_cm_core *cm_core,
  * find_listener - find a cm node listening on this addr-port pair
  */
 static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core,
-		nes_addr_t dst_addr, u16 dst_port, enum nes_cm_listener_state listener_state)
+					nes_addr_t dst_addr, u16 dst_port,
+					enum nes_cm_listener_state listener_state, int local)
 {
 	unsigned long flags;
 	struct nes_cm_listener *listen_node;
+	nes_addr_t listen_addr;
+	u16 listen_port;
 
 	/* walk list and find cm_node associated with this session ID */
 	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
 	list_for_each_entry(listen_node, &cm_core->listen_list.list, list) {
+		if (local) {
+			listen_addr = listen_node->loc_addr;
+			listen_port = listen_node->loc_port;
+		} else {
+			listen_addr = listen_node->mapped_loc_addr;
+			listen_port = listen_node->mapped_loc_port;
+		}
 		/* compare node pair, return node handle if a match */
-		if (((listen_node->loc_addr == dst_addr) ||
-				listen_node->loc_addr == 0x00000000) &&
-				(listen_node->loc_port == dst_port) &&
-				(listener_state & listen_node->listener_state)) {
+		if (((listen_addr == dst_addr) ||
+		     listen_addr == 0x00000000) &&
+		    (listen_port == dst_port) &&
+		    (listener_state & listen_node->listener_state)) {
 			atomic_inc(&listen_node->ref_count);
 			spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 			return listen_node;
@@ -902,32 +1292,23 @@ static struct nes_cm_listener *find_listener(struct nes_cm_core *cm_core,
 	}
 	spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 
-	nes_debug(NES_DBG_CM, "Unable to find listener for " NIPQUAD_FMT ":%x\n",
-		  HIPQUAD(dst_addr), dst_port);
-
 	/* no listener */
 	return NULL;
 }
 
-
 /**
  * add_hte_node - add a cm node to the hash table
  */
 static int add_hte_node(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node)
 {
 	unsigned long flags;
-	u32 hashkey;
 	struct list_head *hte;
 
 	if (!cm_node || !cm_core)
 		return -EINVAL;
 
-	nes_debug(NES_DBG_CM, "Adding Node to Active Connection HT\n");
-
-	/* first, make an index into our hash table */
-	hashkey = make_hashkey(cm_node->loc_port, cm_node->loc_addr,
-			cm_node->rem_port, cm_node->rem_addr);
-	cm_node->hashkey = hashkey;
+	nes_debug(NES_DBG_CM, "Adding Node %p to Active Connection HT\n",
+		  cm_node);
 
 	spin_lock_irqsave(&cm_core->ht_lock, flags);
 
@@ -946,10 +1327,86 @@ static int add_hte_node(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node
  * mini_cm_dec_refcnt_listen
  */
 static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
-		struct nes_cm_listener *listener, int free_hanging_nodes)
+				     struct nes_cm_listener *listener, int free_hanging_nodes)
 {
-	int ret = 1;
+	int ret = -EINVAL;
+	int err = 0;
 	unsigned long flags;
+	struct list_head *list_pos = NULL;
+	struct list_head *list_temp = NULL;
+	struct nes_cm_node *cm_node = NULL;
+	struct list_head reset_list;
+
+	nes_debug(NES_DBG_CM, "attempting listener= %p free_nodes= %d, "
+		  "refcnt=%d\n", listener, free_hanging_nodes,
+		  atomic_read(&listener->ref_count));
+	/* free non-accelerated child nodes for this listener */
+	INIT_LIST_HEAD(&reset_list);
+	if (free_hanging_nodes) {
+		spin_lock_irqsave(&cm_core->ht_lock, flags);
+		list_for_each_safe(list_pos, list_temp,
+				   &g_cm_core->connected_nodes) {
+			cm_node = container_of(list_pos, struct nes_cm_node,
+					       list);
+			if ((cm_node->listener == listener) &&
+			    (!cm_node->accelerated)) {
+				add_ref_cm_node(cm_node);
+				list_add(&cm_node->reset_entry, &reset_list);
+			}
+		}
+		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
+	}
+
+	list_for_each_safe(list_pos, list_temp, &reset_list) {
+		cm_node = container_of(list_pos, struct nes_cm_node,
+				       reset_entry);
+		{
+			struct nes_cm_node *loopback = cm_node->loopbackpartner;
+			enum nes_cm_node_state old_state;
+			if (NES_CM_STATE_FIN_WAIT1 <= cm_node->state) {
+				rem_ref_cm_node(cm_node->cm_core, cm_node);
+			} else {
+				if (!loopback) {
+					cleanup_retrans_entry(cm_node);
+					err = send_reset(cm_node, NULL);
+					if (err) {
+						cm_node->state =
+							NES_CM_STATE_CLOSED;
+						WARN_ON(1);
+					} else {
+						old_state = cm_node->state;
+						cm_node->state = NES_CM_STATE_LISTENER_DESTROYED;
+						if (old_state != NES_CM_STATE_MPAREQ_RCVD)
+							rem_ref_cm_node(
+								cm_node->cm_core,
+								cm_node);
+					}
+				} else {
+					struct nes_cm_event event;
+
+					event.cm_node = loopback;
+					event.cm_info.rem_addr =
+							loopback->rem_addr;
+					event.cm_info.loc_addr =
+							loopback->loc_addr;
+					event.cm_info.rem_port =
+							loopback->rem_port;
+					event.cm_info.loc_port =
+							 loopback->loc_port;
+					event.cm_info.cm_id = loopback->cm_id;
+					add_ref_cm_node(loopback);
+					loopback->state = NES_CM_STATE_CLOSED;
+					cm_event_connect_error(&event);
+					cm_node->state = NES_CM_STATE_LISTENER_DESTROYED;
+
+					rem_ref_cm_node(cm_node->cm_core,
+							 cm_node);
+
+				}
+			}
+		}
+	}
+
 	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
 	if (!atomic_dec_return(&listener->ref_count)) {
 		list_del(&listener->list);
@@ -960,8 +1417,18 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
 		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 
 		if (listener->nesvnic) {
-			nes_manage_apbvt(listener->nesvnic, listener->loc_port,
-					PCI_FUNC(listener->nesvnic->nesdev->pcidev->devfn), NES_MANAGE_APBVT_DEL);
+			nes_manage_apbvt(listener->nesvnic,
+				listener->mapped_loc_port,
+				PCI_FUNC(listener->nesvnic->nesdev->pcidev->devfn),
+				NES_MANAGE_APBVT_DEL);
+
+			nes_remove_mapinfo(listener->loc_addr,
+					listener->loc_port,
+					listener->mapped_loc_addr,
+					listener->mapped_loc_port);
+			nes_debug(NES_DBG_NLMSG,
+					"Delete APBVT mapped_loc_port = %04X\n",
+					listener->mapped_loc_port);
 		}
 
 		nes_debug(NES_DBG_CM, "destroying listener (%p)\n", listener);
@@ -969,15 +1436,15 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
 		kfree(listener);
 		listener = NULL;
 		ret = 0;
-		cm_listens_destroyed++;
+		atomic_inc(&cm_listens_destroyed);
 	} else {
 		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 	}
 	if (listener) {
 		if (atomic_read(&listener->pend_accepts_cnt) > 0)
 			nes_debug(NES_DBG_CM, "destroying listener (%p)"
-					" with non-zero pending accepts=%u\n",
-					listener, atomic_read(&listener->pend_accepts_cnt));
+				  " with non-zero pending accepts=%u\n",
+				  listener, atomic_read(&listener->pend_accepts_cnt));
 	}
 
 	return ret;
@@ -988,7 +1455,7 @@ static int mini_cm_dec_refcnt_listen(struct nes_cm_core *cm_core,
  * mini_cm_del_listen
  */
 static int mini_cm_del_listen(struct nes_cm_core *cm_core,
-		struct nes_cm_listener *listener)
+			      struct nes_cm_listener *listener)
 {
 	listener->listener_state = NES_CM_LISTENER_PASSIVE_STATE;
 	listener->cm_id = NULL; /* going to be destroyed pretty soon */
@@ -1000,61 +1467,98 @@ static int mini_cm_del_listen(struct nes_cm_core *cm_core,
  * mini_cm_accelerated
  */
 static inline int mini_cm_accelerated(struct nes_cm_core *cm_core,
-		struct nes_cm_node *cm_node)
+				      struct nes_cm_node *cm_node)
 {
-	u32 was_timer_set;
 	cm_node->accelerated = 1;
 
 	if (cm_node->accept_pend) {
 		BUG_ON(!cm_node->listener);
 		atomic_dec(&cm_node->listener->pend_accepts_cnt);
+		cm_node->accept_pend = 0;
 		BUG_ON(atomic_read(&cm_node->listener->pend_accepts_cnt) < 0);
 	}
 
-	was_timer_set = timer_pending(&cm_core->tcp_timer);
-	if (!was_timer_set) {
-		cm_core->tcp_timer.expires = jiffies + NES_SHORT_TIME;
-		add_timer(&cm_core->tcp_timer);
-	}
+	if (!timer_pending(&cm_core->tcp_timer))
+		mod_timer(&cm_core->tcp_timer, (jiffies + NES_SHORT_TIME));
 
 	return 0;
 }
 
 
 /**
- * nes_addr_send_arp
+ * nes_addr_resolve_neigh
  */
-static void nes_addr_send_arp(u32 dst_ip)
+static int nes_addr_resolve_neigh(struct nes_vnic *nesvnic, u32 dst_ip, int arpindex)
 {
 	struct rtable *rt;
-	struct flowi fl;
+	struct neighbour *neigh;
+	int rc = arpindex;
+	struct net_device *netdev;
+	struct nes_adapter *nesadapter = nesvnic->nesdev->nesadapter;
+
+	rt = ip_route_output(&init_net, htonl(dst_ip), 0, 0, 0);
+	if (IS_ERR(rt)) {
+		printk(KERN_ERR "%s: ip_route_output_key failed for 0x%08X\n",
+		       __func__, dst_ip);
+		return rc;
+	}
 
-	memset(&fl, 0, sizeof fl);
-	fl.nl_u.ip4_u.daddr = htonl(dst_ip);
-	if (ip_route_output_key(&init_net, &rt, &fl)) {
-		printk("%s: ip_route_output_key failed for 0x%08X\n",
-				__func__, dst_ip);
-		return;
+	if (netif_is_bond_slave(nesvnic->netdev))
+		netdev = netdev_master_upper_dev_get(nesvnic->netdev);
+	else
+		netdev = nesvnic->netdev;
+
+	neigh = neigh_lookup(&arp_tbl, &rt->rt_gateway, netdev);
+
+	rcu_read_lock();
+	if (neigh) {
+		if (neigh->nud_state & NUD_VALID) {
+			nes_debug(NES_DBG_CM, "Neighbor MAC address for 0x%08X"
+				  " is %pM, Gateway is 0x%08X \n", dst_ip,
+				  neigh->ha, ntohl(rt->rt_gateway));
+
+			if (arpindex >= 0) {
+				if (ether_addr_equal(nesadapter->arp_table[arpindex].mac_addr, neigh->ha)) {
+					/* Mac address same as in nes_arp_table */
+					goto out;
+				}
+
+				nes_manage_arp_cache(nesvnic->netdev,
+						     nesadapter->arp_table[arpindex].mac_addr,
+						     dst_ip, NES_ARP_DELETE);
+			}
+
+			nes_manage_arp_cache(nesvnic->netdev, neigh->ha,
+					     dst_ip, NES_ARP_ADD);
+			rc = nes_arp_table(nesvnic->nesdev, dst_ip, NULL,
+					   NES_ARP_RESOLVE);
+		} else {
+			neigh_event_send(neigh, NULL);
+		}
 	}
+out:
+	rcu_read_unlock();
+
+	if (neigh)
+		neigh_release(neigh);
 
-	neigh_event_send(rt->u.dst.neighbour, NULL);
 	ip_rt_put(rt);
+	return rc;
 }
 
-
 /**
  * make_cm_node - create a new instance of a cm node
  */
 static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
-		struct nes_vnic *nesvnic, struct nes_cm_info *cm_info,
-		struct nes_cm_listener *listener)
+					struct nes_vnic *nesvnic, struct nes_cm_info *cm_info,
+					struct nes_cm_listener *listener)
 {
 	struct nes_cm_node *cm_node;
 	struct timespec ts;
+	int oldarpindex = 0;
 	int arpindex = 0;
 	struct nes_device *nesdev;
 	struct nes_adapter *nesadapter;
-	DECLARE_MAC_BUF(mac);
 
 	/* create an hte and cm_node for this instance */
 	cm_node = kzalloc(sizeof(*cm_node), GFP_ATOMIC);
@@ -1066,22 +1570,30 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
 	cm_node->rem_addr = cm_info->rem_addr;
 	cm_node->loc_port = cm_info->loc_port;
 	cm_node->rem_port = cm_info->rem_port;
-	cm_node->send_write0 = send_first;
-	nes_debug(NES_DBG_CM, "Make node addresses : loc = " NIPQUAD_FMT ":%x, rem = " NIPQUAD_FMT ":%x\n",
-		  HIPQUAD(cm_node->loc_addr), cm_node->loc_port,
-		  HIPQUAD(cm_node->rem_addr), cm_node->rem_port);
+
+	cm_node->mapped_loc_addr = cm_info->mapped_loc_addr;
+	cm_node->mapped_rem_addr = cm_info->mapped_rem_addr;
+	cm_node->mapped_loc_port = cm_info->mapped_loc_port;
+	cm_node->mapped_rem_port = cm_info->mapped_rem_port;
+
+	cm_node->mpa_frame_rev = mpa_version;
+	cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
+	cm_node->mpav2_ird_ord = 0;
+	cm_node->ird_size = 0;
+	cm_node->ord_size = 0;
+
+	nes_debug(NES_DBG_CM, "Make node addresses : loc = %pI4:%x, rem = %pI4:%x\n",
+		  &cm_node->loc_addr, cm_node->loc_port,
+		  &cm_node->rem_addr, cm_node->rem_port);
 	cm_node->listener = listener;
 	cm_node->netdev = nesvnic->netdev;
 	cm_node->cm_id = cm_info->cm_id;
 	memcpy(cm_node->loc_mac, nesvnic->netdev->dev_addr, ETH_ALEN);
 
-	nes_debug(NES_DBG_CM, "listener=%p, cm_id=%p\n",
-			cm_node->listener, cm_node->cm_id);
+	nes_debug(NES_DBG_CM, "listener=%p, cm_id=%p\n", cm_node->listener,
+		  cm_node->cm_id);
 
-	INIT_LIST_HEAD(&cm_node->retrans_list);
 	spin_lock_init(&cm_node->retrans_list_lock);
-	INIT_LIST_HEAD(&cm_node->recv_list);
-	spin_lock_init(&cm_node->recv_list_lock);
 
 	cm_node->loopbackpartner = NULL;
 	atomic_set(&cm_node->ref_count, 1);
@@ -1090,11 +1602,11 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
 	cm_node->tcp_cntxt.loc_id = NES_CM_DEF_LOCAL_ID;
 	cm_node->tcp_cntxt.rcv_wscale = NES_CM_DEFAULT_RCV_WND_SCALE;
 	cm_node->tcp_cntxt.rcv_wnd = NES_CM_DEFAULT_RCV_WND_SCALED >>
-			NES_CM_DEFAULT_RCV_WND_SCALE;
+				     NES_CM_DEFAULT_RCV_WND_SCALE;
 	ts = current_kernel_time();
 	cm_node->tcp_cntxt.loc_seq_num = htonl(ts.tv_nsec);
 	cm_node->tcp_cntxt.mss = nesvnic->max_frame_size - sizeof(struct iphdr) -
-			sizeof(struct tcphdr) - ETH_HLEN - VLAN_HLEN;
+				 sizeof(struct tcphdr) - ETH_HLEN - VLAN_HLEN;
 	cm_node->tcp_cntxt.rcv_nxt = 0;
 	/* get a unique session ID , add thread_id to an upcounter to handle race */
 	atomic_inc(&cm_core->node_cnt);
@@ -1108,18 +1620,21 @@ static struct nes_cm_node *make_cm_node(struct nes_cm_core *cm_core,
 	nesadapter = nesdev->nesadapter;
 
 	cm_node->loopbackpartner = NULL;
+
 	/* get the mac addr for the remote node */
-	arpindex = nes_arp_table(nesdev, cm_node->rem_addr, NULL, NES_ARP_RESOLVE);
+	oldarpindex = nes_arp_table(nesdev, cm_node->mapped_rem_addr,
+				NULL, NES_ARP_RESOLVE);
+	arpindex = nes_addr_resolve_neigh(nesvnic,
+				cm_node->mapped_rem_addr, oldarpindex);
 	if (arpindex < 0) {
 		kfree(cm_node);
-		nes_addr_send_arp(cm_info->rem_addr);
 		return NULL;
 	}
 
 	/* copy the mac addr to node context */
 	memcpy(cm_node->rem_mac, nesadapter->arp_table[arpindex].mac_addr, ETH_ALEN);
-	nes_debug(NES_DBG_CM, "Remote mac addr from arp table: %s\n",
-		  print_mac(mac, cm_node->rem_mac));
+	nes_debug(NES_DBG_CM, "Remote mac addr from arp table: %pM\n",
+		  cm_node->rem_mac);
 
 	add_hte_node(cm_core, cm_node);
 	atomic_inc(&cm_nodes_created);
@@ -1142,13 +1657,9 @@ static int add_ref_cm_node(struct nes_cm_node *cm_node)
  * rem_ref_cm_node - destroy an instance of a cm node
  */
 static int rem_ref_cm_node(struct nes_cm_core *cm_core,
-		struct nes_cm_node *cm_node)
+			   struct nes_cm_node *cm_node)
 {
-	unsigned long flags, qplockflags;
-	struct nes_timer_entry *send_entry;
-	struct nes_timer_entry *recv_entry;
-	struct iw_cm_id *cm_id;
-	struct list_head *list_core, *list_node_temp;
+	unsigned long flags;
 	struct nes_qp *nesqp;
 
 	if (!cm_node)
@@ -1169,75 +1680,41 @@ static int rem_ref_cm_node(struct nes_cm_core *cm_core,
 		atomic_dec(&cm_node->listener->pend_accepts_cnt);
 		BUG_ON(atomic_read(&cm_node->listener->pend_accepts_cnt) < 0);
 	}
-
-	spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-	list_for_each_safe(list_core, list_node_temp, &cm_node->retrans_list) {
-		send_entry = container_of(list_core, struct nes_timer_entry, list);
-		list_del(&send_entry->list);
-		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-		dev_kfree_skb_any(send_entry->skb);
-		kfree(send_entry);
-		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
-		continue;
-	}
-	spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
-
-	spin_lock_irqsave(&cm_node->recv_list_lock, flags);
-	list_for_each_safe(list_core, list_node_temp, &cm_node->recv_list) {
-		recv_entry = container_of(list_core, struct nes_timer_entry, list);
-		list_del(&recv_entry->list);
-		cm_id = cm_node->cm_id;
-		spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
-		if (recv_entry->type == NES_TIMER_TYPE_CLOSE) {
-			nesqp = (struct nes_qp *)recv_entry->skb;
-			spin_lock_irqsave(&nesqp->lock, qplockflags);
-			if (nesqp->cm_id) {
-				nes_debug(NES_DBG_CM, "QP%u: cm_id = %p: ****** HIT A NES_TIMER_TYPE_CLOSE"
-						" with something to do!!! ******\n",
-						nesqp->hwqp.qp_id, cm_id);
-				nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-				nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
-				nesqp->ibqp_state = IB_QPS_ERR;
-				spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-				nes_cm_disconn(nesqp);
-			} else {
-				spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-				nes_debug(NES_DBG_CM, "QP%u: cm_id = %p: ****** HIT A NES_TIMER_TYPE_CLOSE"
-						" with nothing to do!!! ******\n",
-						nesqp->hwqp.qp_id, cm_id);
-				nes_rem_ref(&nesqp->ibqp);
-			}
-			cm_id->rem_ref(cm_id);
-		} else if (recv_entry->type == NES_TIMER_TYPE_RECV) {
-			dev_kfree_skb_any(recv_entry->skb);
-		}
-		kfree(recv_entry);
-		spin_lock_irqsave(&cm_node->recv_list_lock, flags);
-	}
-	spin_unlock_irqrestore(&cm_node->recv_list_lock, flags);
-
+	WARN_ON(cm_node->send_entry);
+	if (cm_node->recv_entry)
+		handle_recv_entry(cm_node, 0);
 	if (cm_node->listener) {
 		mini_cm_dec_refcnt_listen(cm_core, cm_node->listener, 0);
 	} else {
 		if (cm_node->apbvt_set && cm_node->nesvnic) {
-			nes_manage_apbvt(cm_node->nesvnic, cm_node->loc_port,
-					PCI_FUNC(cm_node->nesvnic->nesdev->pcidev->devfn),
-					NES_MANAGE_APBVT_DEL);
+			nes_manage_apbvt(cm_node->nesvnic, cm_node->mapped_loc_port,
+					 PCI_FUNC(cm_node->nesvnic->nesdev->pcidev->devfn),
+					 NES_MANAGE_APBVT_DEL);
 		}
+		nes_debug(NES_DBG_NLMSG, "Delete APBVT mapped_loc_port = %04X\n",
+					cm_node->mapped_loc_port);
+		nes_remove_mapinfo(cm_node->loc_addr, cm_node->loc_port,
+			cm_node->mapped_loc_addr, cm_node->mapped_loc_port);
 	}
 
-	kfree(cm_node);
 	atomic_dec(&cm_core->node_cnt);
 	atomic_inc(&cm_nodes_destroyed);
+	nesqp = cm_node->nesqp;
+	if (nesqp) {
+		nesqp->cm_node = NULL;
+		nes_rem_ref(&nesqp->ibqp);
+		cm_node->nesqp = NULL;
+	}
 
+	kfree(cm_node);
 	return 0;
 }
 
-
 /**
  * process_options
  */
-static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc, u32 optionsize, u32 syn_packet)
+static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc,
+			   u32 optionsize, u32 syn_packet)
 {
 	u32 tmp;
 	u32 offset = 0;
@@ -1247,35 +1724,34 @@ static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc, u32 opti
 	while (offset < optionsize) {
 		all_options = (union all_known_options *)(optionsloc + offset);
 		switch (all_options->as_base.optionnum) {
-			case OPTION_NUMBER_END:
-				offset = optionsize;
-				break;
-			case OPTION_NUMBER_NONE:
-				offset += 1;
-				continue;
-			case OPTION_NUMBER_MSS:
-				nes_debug(NES_DBG_CM, "%s: MSS Length: %d Offset: %d Size: %d\n",
-						__func__,
-						all_options->as_mss.length, offset, optionsize);
-				got_mss_option = 1;
-				if (all_options->as_mss.length != 4) {
-					return 1;
-				} else {
-					tmp = ntohs(all_options->as_mss.mss);
-					if (tmp > 0 && tmp < cm_node->tcp_cntxt.mss)
-						cm_node->tcp_cntxt.mss = tmp;
-				}
-				break;
-			case OPTION_NUMBER_WINDOW_SCALE:
-				cm_node->tcp_cntxt.snd_wscale = all_options->as_windowscale.shiftcount;
-				break;
-			case OPTION_NUMBER_WRITE0:
-				cm_node->send_write0 = 1;
-				break;
-			default:
-				nes_debug(NES_DBG_CM, "TCP Option not understood: %x\n",
-						all_options->as_base.optionnum);
-				break;
+		case OPTION_NUMBER_END:
+			offset = optionsize;
+			break;
+		case OPTION_NUMBER_NONE:
+			offset += 1;
+			continue;
+		case OPTION_NUMBER_MSS:
+			nes_debug(NES_DBG_CM, "%s: MSS Length: %d Offset: %d "
+				  "Size: %d\n", __func__,
+				  all_options->as_mss.length, offset, optionsize);
+			got_mss_option = 1;
+			if (all_options->as_mss.length != 4) {
+				return 1;
+			} else {
+				tmp = ntohs(all_options->as_mss.mss);
+				if (tmp > 0 && tmp <
+				    cm_node->tcp_cntxt.mss)
+					cm_node->tcp_cntxt.mss = tmp;
+			}
+			break;
+		case OPTION_NUMBER_WINDOW_SCALE:
+			cm_node->tcp_cntxt.snd_wscale =
+				all_options->as_windowscale.shiftcount;
+			break;
+		default:
+			nes_debug(NES_DBG_CM, "TCP Option not understood: %x\n",
+				  all_options->as_base.optionnum);
+			break;
 		}
 		offset += all_options->as_base.length;
 	}
@@ -1284,310 +1760,623 @@ static int process_options(struct nes_cm_node *cm_node, u8 *optionsloc, u32 opti
 	return 0;
 }
 
+static void drop_packet(struct sk_buff *skb)
+{
+	atomic_inc(&cm_accel_dropped_pkts);
+	dev_kfree_skb_any(skb);
+}
 
-/**
- * process_packet
- */
-static int process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb,
-			  struct nes_cm_core *cm_core)
+static void handle_fin_pkt(struct nes_cm_node *cm_node)
 {
-	int optionsize;
-	int datasize;
-	int ret = 0;
-	struct tcphdr *tcph = tcp_hdr(skb);
-	u32 inc_sequence;
-	if (cm_node->state == NES_CM_STATE_SYN_SENT && tcph->syn) {
-		inc_sequence = ntohl(tcph->seq);
-		cm_node->tcp_cntxt.rcv_nxt = inc_sequence;
+	nes_debug(NES_DBG_CM, "Received FIN, cm_node = %p, state = %u. "
+		  "refcnt=%d\n", cm_node, cm_node->state,
+		  atomic_read(&cm_node->ref_count));
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_RCVD:
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_MPAREJ_RCVD:
+		cm_node->tcp_cntxt.rcv_nxt++;
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_LAST_ACK;
+		send_fin(cm_node, NULL);
+		break;
+	case NES_CM_STATE_MPAREQ_SENT:
+		create_event(cm_node, NES_CM_EVENT_ABORTED);
+		cm_node->tcp_cntxt.rcv_nxt++;
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSED;
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, NULL);
+		break;
+	case NES_CM_STATE_FIN_WAIT1:
+		cm_node->tcp_cntxt.rcv_nxt++;
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSING;
+		send_ack(cm_node, NULL);
+		/* Wait for ACK as this is simultaneous close..
+		* After we receive ACK, do not send anything..
+		* Just rm the node.. Done.. */
+		break;
+	case NES_CM_STATE_FIN_WAIT2:
+		cm_node->tcp_cntxt.rcv_nxt++;
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_TIME_WAIT;
+		send_ack(cm_node, NULL);
+		schedule_nes_timer(cm_node, NULL,  NES_TIMER_TYPE_CLOSE, 1, 0);
+		break;
+	case NES_CM_STATE_TIME_WAIT:
+		cm_node->tcp_cntxt.rcv_nxt++;
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSED;
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		break;
+	case NES_CM_STATE_TSA:
+	default:
+		nes_debug(NES_DBG_CM, "Error Rcvd FIN for node-%p state = %d\n",
+			cm_node, cm_node->state);
+		break;
 	}
+}
 
-	if ((!tcph) || (cm_node->state == NES_CM_STATE_TSA)) {
-		BUG_ON(!tcph);
-		atomic_inc(&cm_accel_dropped_pkts);
-		return -1;
+
+static void handle_rst_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
+	struct tcphdr *tcph)
+{
+
+	int	reset = 0;	/* whether to send reset in case of err.. */
+	atomic_inc(&cm_resets_recvd);
+	nes_debug(NES_DBG_CM, "Received Reset, cm_node = %p, state = %u."
+			" refcnt=%d\n", cm_node, cm_node->state,
+			atomic_read(&cm_node->ref_count));
+	cleanup_retrans_entry(cm_node);
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_MPAREQ_SENT:
+		nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
+			"listener=%p state=%d\n", __func__, __LINE__, cm_node,
+			cm_node->listener, cm_node->state);
+		switch (cm_node->mpa_frame_rev) {
+		case IETF_MPA_V2:
+			cm_node->mpa_frame_rev = IETF_MPA_V1;
+			/* send a syn and goto syn sent state */
+			cm_node->state = NES_CM_STATE_SYN_SENT;
+			if (send_syn(cm_node, 0, NULL)) {
+				active_open_err(cm_node, skb, reset);
+			}
+			break;
+		case IETF_MPA_V1:
+		default:
+			active_open_err(cm_node, skb, reset);
+			break;
+		}
+		break;
+	case NES_CM_STATE_MPAREQ_RCVD:
+		atomic_inc(&cm_node->passive_state);
+		dev_kfree_skb_any(skb);
+		break;
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_SYN_RCVD:
+	case NES_CM_STATE_LISTENING:
+		nes_debug(NES_DBG_CM, "Bad state %s[%u]\n", __func__, __LINE__);
+		passive_open_err(cm_node, skb, reset);
+		break;
+	case NES_CM_STATE_TSA:
+		active_open_err(cm_node, skb, reset);
+		break;
+	case NES_CM_STATE_CLOSED:
+		drop_packet(skb);
+		break;
+	case NES_CM_STATE_FIN_WAIT2:
+	case NES_CM_STATE_FIN_WAIT1:
+	case NES_CM_STATE_LAST_ACK:
+		cm_node->cm_id->rem_ref(cm_node->cm_id);
+	case NES_CM_STATE_TIME_WAIT:
+		cm_node->state = NES_CM_STATE_CLOSED;
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		drop_packet(skb);
+		break;
+	default:
+		drop_packet(skb);
+		break;
 	}
+}
 
-	if (tcph->rst) {
-		atomic_inc(&cm_resets_recvd);
-		nes_debug(NES_DBG_CM, "Received Reset, cm_node = %p, state = %u. refcnt=%d\n",
-				cm_node, cm_node->state, atomic_read(&cm_node->ref_count));
-		switch (cm_node->state) {
-			case NES_CM_STATE_LISTENING:
-				rem_ref_cm_node(cm_core, cm_node);
-				break;
-			case NES_CM_STATE_TSA:
-			case NES_CM_STATE_CLOSED:
-				break;
-			case NES_CM_STATE_SYN_RCVD:
-					nes_debug(NES_DBG_CM, "Received a reset for local 0x%08X:%04X,"
-							" remote 0x%08X:%04X, node state = %u\n",
-							cm_node->loc_addr, cm_node->loc_port,
-							cm_node->rem_addr, cm_node->rem_port,
-							cm_node->state);
-				rem_ref_cm_node(cm_core, cm_node);
-				break;
-			case NES_CM_STATE_ONE_SIDE_ESTABLISHED:
-			case NES_CM_STATE_ESTABLISHED:
-			case NES_CM_STATE_MPAREQ_SENT:
-			default:
-					nes_debug(NES_DBG_CM, "Received a reset for local 0x%08X:%04X,"
-							" remote 0x%08X:%04X, node state = %u refcnt=%d\n",
-							cm_node->loc_addr, cm_node->loc_port,
-							cm_node->rem_addr, cm_node->rem_port,
-							cm_node->state, atomic_read(&cm_node->ref_count));
-				/* create event */
-				cm_node->state = NES_CM_STATE_CLOSED;
 
-				create_event(cm_node, NES_CM_EVENT_ABORTED);
-				break;
+static void handle_rcv_mpa(struct nes_cm_node *cm_node, struct sk_buff *skb)
+{
+	int ret = 0;
+	int datasize = skb->len;
+	u8 *dataloc = skb->data;
+
+	enum nes_cm_event_type type = NES_CM_EVENT_UNKNOWN;
+	u32 res_type;
+
+	ret = parse_mpa(cm_node, dataloc, &res_type, datasize);
+	if (ret) {
+		nes_debug(NES_DBG_CM, "didn't like MPA Request\n");
+		if (cm_node->state == NES_CM_STATE_MPAREQ_SENT) {
+			nes_debug(NES_DBG_CM, "%s[%u] create abort for "
+				  "cm_node=%p listener=%p state=%d\n", __func__,
+				  __LINE__, cm_node, cm_node->listener,
+				  cm_node->state);
+			active_open_err(cm_node, skb, 1);
+		} else {
+			passive_open_err(cm_node, skb, 1);
+		}
+		return;
+	}
 
+	switch (cm_node->state) {
+	case NES_CM_STATE_ESTABLISHED:
+		if (res_type == NES_MPA_REQUEST_REJECT)
+			/*BIG problem as we are receiving the MPA.. So should
+			 * not be REJECT.. This is Passive Open.. We can
+			 * only receive it Reject for Active Open...*/
+			WARN_ON(1);
+		cm_node->state = NES_CM_STATE_MPAREQ_RCVD;
+		type = NES_CM_EVENT_MPA_REQ;
+		atomic_set(&cm_node->passive_state,
+			   NES_PASSIVE_STATE_INDICATED);
+		break;
+	case NES_CM_STATE_MPAREQ_SENT:
+		cleanup_retrans_entry(cm_node);
+		if (res_type == NES_MPA_REQUEST_REJECT) {
+			type = NES_CM_EVENT_MPA_REJECT;
+			cm_node->state = NES_CM_STATE_MPAREJ_RCVD;
+		} else {
+			type = NES_CM_EVENT_CONNECTED;
+			cm_node->state = NES_CM_STATE_TSA;
 		}
-		return -1;
+
+		break;
+	default:
+		WARN_ON(1);
+		break;
 	}
+	dev_kfree_skb_any(skb);
+	create_event(cm_node, type);
+}
 
-	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
+static void indicate_pkt_err(struct nes_cm_node *cm_node, struct sk_buff *skb)
+{
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_MPAREQ_SENT:
+		nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
+			  "listener=%p state=%d\n", __func__, __LINE__, cm_node,
+			  cm_node->listener, cm_node->state);
+		active_open_err(cm_node, skb, 1);
+		break;
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_SYN_RCVD:
+		passive_open_err(cm_node, skb, 1);
+		break;
+	case NES_CM_STATE_TSA:
+	default:
+		drop_packet(skb);
+	}
+}
 
-	skb_pull(skb, ip_hdr(skb)->ihl << 2);
-	skb_pull(skb, tcph->doff << 2);
+static int check_syn(struct nes_cm_node *cm_node, struct tcphdr *tcph,
+		     struct sk_buff *skb)
+{
+	int err;
 
-	datasize = skb->len;
-	inc_sequence = ntohl(tcph->seq);
-	nes_debug(NES_DBG_CM, "datasize = %u, sequence = 0x%08X, ack_seq = 0x%08X,"
-			" rcv_nxt = 0x%08X Flags: %s %s.\n",
-			datasize, inc_sequence, ntohl(tcph->ack_seq),
-			cm_node->tcp_cntxt.rcv_nxt, (tcph->syn ? "SYN":""),
-			(tcph->ack ? "ACK":""));
-
-	if (!tcph->syn && (inc_sequence != cm_node->tcp_cntxt.rcv_nxt)
-		) {
-		nes_debug(NES_DBG_CM, "dropping packet, datasize = %u, sequence = 0x%08X,"
-				" ack_seq = 0x%08X, rcv_nxt = 0x%08X Flags: %s.\n",
-				datasize, inc_sequence, ntohl(tcph->ack_seq),
-				cm_node->tcp_cntxt.rcv_nxt, (tcph->ack ? "ACK":""));
-		if (cm_node->state == NES_CM_STATE_LISTENING) {
-			rem_ref_cm_node(cm_core, cm_node);
-		}
-		return -1;
+	err = ((ntohl(tcph->ack_seq) == cm_node->tcp_cntxt.loc_seq_num)) ? 0 : 1;
+	if (err)
+		active_open_err(cm_node, skb, 1);
+
+	return err;
+}
+
+static int check_seq(struct nes_cm_node *cm_node, struct tcphdr *tcph,
+		     struct sk_buff *skb)
+{
+	int err = 0;
+	u32 seq;
+	u32 ack_seq;
+	u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num;
+	u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
+	u32 rcv_wnd;
+
+	seq = ntohl(tcph->seq);
+	ack_seq = ntohl(tcph->ack_seq);
+	rcv_wnd = cm_node->tcp_cntxt.rcv_wnd;
+	if (ack_seq != loc_seq_num)
+		err = 1;
+	else if (!between(seq, rcv_nxt, (rcv_nxt + rcv_wnd)))
+		err = 1;
+	if (err) {
+		nes_debug(NES_DBG_CM, "%s[%u] create abort for cm_node=%p "
+			  "listener=%p state=%d\n", __func__, __LINE__, cm_node,
+			  cm_node->listener, cm_node->state);
+		indicate_pkt_err(cm_node, skb);
+		nes_debug(NES_DBG_CM, "seq ERROR cm_node =%p seq=0x%08X "
+			  "rcv_nxt=0x%08X rcv_wnd=0x%x\n", cm_node, seq, rcv_nxt,
+			  rcv_wnd);
 	}
+	return err;
+}
 
-		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
+/*
+ * handle_syn_pkt() is for Passive node. The syn packet is received when a node
+ * is created with a listener or it may comein as rexmitted packet which in
+ * that case will be just dropped.
+ */
+static void handle_syn_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			   struct tcphdr *tcph)
+{
+	int ret;
+	u32 inc_sequence;
+	int optionsize;
 
+	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
+	skb_trim(skb, 0);
+	inc_sequence = ntohl(tcph->seq);
 
-	if (optionsize) {
-		u8 *optionsloc = (u8 *)&tcph[1];
-		if (process_options(cm_node, optionsloc, optionsize, (u32)tcph->syn)) {
-			nes_debug(NES_DBG_CM, "%s: Node %p, Sending RESET\n", __func__, cm_node);
-			send_reset(cm_node);
-			if (cm_node->state != NES_CM_STATE_SYN_SENT)
-			rem_ref_cm_node(cm_core, cm_node);
-			return 0;
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_MPAREQ_SENT:
+		/* Rcvd syn on active open connection*/
+		active_open_err(cm_node, skb, 1);
+		break;
+	case NES_CM_STATE_LISTENING:
+		/* Passive OPEN */
+		if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
+		    cm_node->listener->backlog) {
+			nes_debug(NES_DBG_CM, "drop syn due to backlog "
+				  "pressure \n");
+			cm_backlog_drops++;
+			passive_open_err(cm_node, skb, 0);
+			break;
 		}
-	} else if (tcph->syn)
-		cm_node->tcp_cntxt.mss = NES_CM_DEFAULT_MSS;
+		ret = handle_tcp_options(cm_node, tcph, skb, optionsize,
+					 1);
+		if (ret) {
+			passive_open_err(cm_node, skb, 0);
+			/* drop pkt */
+			break;
+		}
+		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
+		BUG_ON(cm_node->send_entry);
+		cm_node->accept_pend = 1;
+		atomic_inc(&cm_node->listener->pend_accepts_cnt);
+
+		cm_node->state = NES_CM_STATE_SYN_RCVD;
+		send_syn(cm_node, 1, skb);
+		break;
+	case NES_CM_STATE_CLOSED:
+		cleanup_retrans_entry(cm_node);
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, skb);
+		break;
+	case NES_CM_STATE_TSA:
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_FIN_WAIT1:
+	case NES_CM_STATE_FIN_WAIT2:
+	case NES_CM_STATE_MPAREQ_RCVD:
+	case NES_CM_STATE_LAST_ACK:
+	case NES_CM_STATE_CLOSING:
+	case NES_CM_STATE_UNKNOWN:
+	default:
+		drop_packet(skb);
+		break;
+	}
+}
 
-	cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window) <<
-			cm_node->tcp_cntxt.snd_wscale;
+static void handle_synack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			      struct tcphdr *tcph)
+{
+	int ret;
+	u32 inc_sequence;
+	int optionsize;
 
-	if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd) {
-		cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
+	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
+	skb_trim(skb, 0);
+	inc_sequence = ntohl(tcph->seq);
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_SENT:
+		cleanup_retrans_entry(cm_node);
+		/* active open */
+		if (check_syn(cm_node, tcph, skb))
+			return;
+		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
+		/* setup options */
+		ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 0);
+		if (ret) {
+			nes_debug(NES_DBG_CM, "cm_node=%p tcp_options failed\n",
+				  cm_node);
+			break;
+		}
+		cleanup_retrans_entry(cm_node);
+		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
+		send_mpa_request(cm_node, skb);
+		cm_node->state = NES_CM_STATE_MPAREQ_SENT;
+		break;
+	case NES_CM_STATE_MPAREQ_RCVD:
+		/* passive open, so should not be here */
+		passive_open_err(cm_node, skb, 1);
+		break;
+	case NES_CM_STATE_LISTENING:
+		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSED;
+		send_reset(cm_node, skb);
+		break;
+	case NES_CM_STATE_CLOSED:
+		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
+		cleanup_retrans_entry(cm_node);
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, skb);
+		break;
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_FIN_WAIT1:
+	case NES_CM_STATE_FIN_WAIT2:
+	case NES_CM_STATE_LAST_ACK:
+	case NES_CM_STATE_TSA:
+	case NES_CM_STATE_CLOSING:
+	case NES_CM_STATE_UNKNOWN:
+	case NES_CM_STATE_MPAREQ_SENT:
+	default:
+		drop_packet(skb);
+		break;
 	}
+}
 
-	if (tcph->ack) {
+static int handle_ack_pkt(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			  struct tcphdr *tcph)
+{
+	int datasize = 0;
+	u32 inc_sequence;
+	int ret = 0;
+	int optionsize;
+
+	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
+
+	if (check_seq(cm_node, tcph, skb))
+		return -EINVAL;
+
+	skb_pull(skb, tcph->doff << 2);
+	inc_sequence = ntohl(tcph->seq);
+	datasize = skb->len;
+	switch (cm_node->state) {
+	case NES_CM_STATE_SYN_RCVD:
+		/* Passive OPEN */
+		cleanup_retrans_entry(cm_node);
+		ret = handle_tcp_options(cm_node, tcph, skb, optionsize, 1);
+		if (ret)
+			break;
 		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
-		switch (cm_node->state) {
-			case NES_CM_STATE_SYN_RCVD:
-			case NES_CM_STATE_SYN_SENT:
-				/* read and stash current sequence number */
-				if (cm_node->tcp_cntxt.rem_ack_num != cm_node->tcp_cntxt.loc_seq_num) {
-					nes_debug(NES_DBG_CM, "ERROR - cm_node->tcp_cntxt.rem_ack_num !="
-							" cm_node->tcp_cntxt.loc_seq_num\n");
-					send_reset(cm_node);
-					return 0;
-				}
-				if (cm_node->state == NES_CM_STATE_SYN_SENT)
-					cm_node->state = NES_CM_STATE_ONE_SIDE_ESTABLISHED;
-				else {
-						cm_node->state = NES_CM_STATE_ESTABLISHED;
-				}
-				break;
-			case NES_CM_STATE_LAST_ACK:
-				cm_node->state = NES_CM_STATE_CLOSED;
-				break;
-			case NES_CM_STATE_FIN_WAIT1:
-				cm_node->state = NES_CM_STATE_FIN_WAIT2;
-				break;
-			case NES_CM_STATE_CLOSING:
-				cm_node->state = NES_CM_STATE_TIME_WAIT;
-				/* need to schedule this to happen in 2MSL timeouts */
-				cm_node->state = NES_CM_STATE_CLOSED;
-				break;
-			case NES_CM_STATE_ONE_SIDE_ESTABLISHED:
-			case NES_CM_STATE_ESTABLISHED:
-			case NES_CM_STATE_MPAREQ_SENT:
-			case NES_CM_STATE_CLOSE_WAIT:
-			case NES_CM_STATE_TIME_WAIT:
-			case NES_CM_STATE_CLOSED:
-				break;
-			case NES_CM_STATE_LISTENING:
-				nes_debug(NES_DBG_CM, "Received an ACK on a listening port (SYN %d)\n", tcph->syn);
-				cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
-				send_reset(cm_node);
-				/* send_reset bumps refcount, this should have been a new node */
-				rem_ref_cm_node(cm_core, cm_node);
-				return -1;
-				break;
-			case NES_CM_STATE_TSA:
-				nes_debug(NES_DBG_CM, "Received a packet with the ack bit set while in TSA state\n");
-				break;
-			case NES_CM_STATE_UNKNOWN:
-			case NES_CM_STATE_INITED:
-			case NES_CM_STATE_ACCEPTING:
-			case NES_CM_STATE_FIN_WAIT2:
-			default:
-				nes_debug(NES_DBG_CM, "Received ack from unknown state: %x\n",
-						cm_node->state);
-				send_reset(cm_node);
-				break;
+		cm_node->state = NES_CM_STATE_ESTABLISHED;
+		if (datasize) {
+			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
+			handle_rcv_mpa(cm_node, skb);
+		} else { /* rcvd ACK only */
+			dev_kfree_skb_any(skb);
+		}
+		break;
+	case NES_CM_STATE_ESTABLISHED:
+		/* Passive OPEN */
+		cleanup_retrans_entry(cm_node);
+		if (datasize) {
+			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
+			handle_rcv_mpa(cm_node, skb);
+		} else {
+			drop_packet(skb);
 		}
+		break;
+	case NES_CM_STATE_MPAREQ_SENT:
+		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
+		if (datasize) {
+			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
+			handle_rcv_mpa(cm_node, skb);
+		} else { /* Could be just an ack pkt.. */
+			dev_kfree_skb_any(skb);
+		}
+		break;
+	case NES_CM_STATE_LISTENING:
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSED;
+		send_reset(cm_node, skb);
+		break;
+	case NES_CM_STATE_CLOSED:
+		cleanup_retrans_entry(cm_node);
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, skb);
+		break;
+	case NES_CM_STATE_LAST_ACK:
+	case NES_CM_STATE_CLOSING:
+		cleanup_retrans_entry(cm_node);
+		cm_node->state = NES_CM_STATE_CLOSED;
+		cm_node->cm_id->rem_ref(cm_node->cm_id);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		drop_packet(skb);
+		break;
+	case NES_CM_STATE_FIN_WAIT1:
+		cleanup_retrans_entry(cm_node);
+		drop_packet(skb);
+		cm_node->state = NES_CM_STATE_FIN_WAIT2;
+		break;
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_FIN_WAIT2:
+	case NES_CM_STATE_TSA:
+	case NES_CM_STATE_MPAREQ_RCVD:
+	case NES_CM_STATE_UNKNOWN:
+	default:
+		cleanup_retrans_entry(cm_node);
+		drop_packet(skb);
+		break;
 	}
+	return ret;
+}
 
-	if (tcph->syn) {
-		if (cm_node->state == NES_CM_STATE_LISTENING) {
-			/* do not exceed backlog */
-			atomic_inc(&cm_node->listener->pend_accepts_cnt);
-			if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
-					cm_node->listener->backlog) {
-				nes_debug(NES_DBG_CM, "drop syn due to backlog pressure \n");
-				cm_backlog_drops++;
-				atomic_dec(&cm_node->listener->pend_accepts_cnt);
-				rem_ref_cm_node(cm_core, cm_node);
-				return 0;
-			}
-			cm_node->accept_pend = 1;
 
-		}
-		if (datasize == 0)
-			cm_node->tcp_cntxt.rcv_nxt ++;
 
-		if (cm_node->state == NES_CM_STATE_LISTENING) {
-			cm_node->state = NES_CM_STATE_SYN_RCVD;
-			send_syn(cm_node, 1);
-		}
-		if (cm_node->state == NES_CM_STATE_ONE_SIDE_ESTABLISHED) {
-			cm_node->state = NES_CM_STATE_ESTABLISHED;
-			/* send final handshake ACK */
-			ret = send_ack(cm_node);
-			if (ret < 0)
-				return ret;
+static int handle_tcp_options(struct nes_cm_node *cm_node, struct tcphdr *tcph,
+			      struct sk_buff *skb, int optionsize, int passive)
+{
+	u8 *optionsloc = (u8 *)&tcph[1];
 
-				cm_node->state = NES_CM_STATE_MPAREQ_SENT;
-				ret = send_mpa_request(cm_node);
-				if (ret < 0)
-					return ret;
+	if (optionsize) {
+		if (process_options(cm_node, optionsloc, optionsize,
+				    (u32)tcph->syn)) {
+			nes_debug(NES_DBG_CM, "%s: Node %p, Sending RESET\n",
+				  __func__, cm_node);
+			if (passive)
+				passive_open_err(cm_node, skb, 1);
+			else
+				active_open_err(cm_node, skb, 1);
+			return 1;
 		}
 	}
 
-	if (tcph->fin) {
-		cm_node->tcp_cntxt.rcv_nxt++;
-		switch (cm_node->state) {
-			case NES_CM_STATE_SYN_RCVD:
-			case NES_CM_STATE_SYN_SENT:
-			case NES_CM_STATE_ONE_SIDE_ESTABLISHED:
-			case NES_CM_STATE_ESTABLISHED:
-			case NES_CM_STATE_ACCEPTING:
-			case NES_CM_STATE_MPAREQ_SENT:
-				cm_node->state = NES_CM_STATE_CLOSE_WAIT;
-				cm_node->state = NES_CM_STATE_LAST_ACK;
-				ret = send_fin(cm_node, NULL);
-				break;
-			case NES_CM_STATE_FIN_WAIT1:
-				cm_node->state = NES_CM_STATE_CLOSING;
-				ret = send_ack(cm_node);
-				break;
-			case NES_CM_STATE_FIN_WAIT2:
-				cm_node->state = NES_CM_STATE_TIME_WAIT;
-				cm_node->tcp_cntxt.loc_seq_num ++;
-				ret = send_ack(cm_node);
-				/* need to schedule this to happen in 2MSL timeouts */
-				cm_node->state = NES_CM_STATE_CLOSED;
-				break;
-			case NES_CM_STATE_CLOSE_WAIT:
-			case NES_CM_STATE_LAST_ACK:
-			case NES_CM_STATE_CLOSING:
-			case NES_CM_STATE_TSA:
-			default:
-				nes_debug(NES_DBG_CM, "Received a fin while in %x state\n",
-						cm_node->state);
-				ret = -EINVAL;
-				break;
-		}
+	cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window) <<
+				     cm_node->tcp_cntxt.snd_wscale;
+
+	if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd)
+		cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
+	return 0;
+}
+
+/*
+ * active_open_err() will send reset() if flag set..
+ * It will also send ABORT event.
+ */
+static void active_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			    int reset)
+{
+	cleanup_retrans_entry(cm_node);
+	if (reset) {
+		nes_debug(NES_DBG_CM, "ERROR active err called for cm_node=%p, "
+			  "state=%d\n", cm_node, cm_node->state);
+		add_ref_cm_node(cm_node);
+		send_reset(cm_node, skb);
+	} else {
+		dev_kfree_skb_any(skb);
 	}
 
-	if (datasize) {
-		u8 *dataloc = skb->data;
-		/* figure out what state we are in and handle transition to next state */
-		switch (cm_node->state) {
-			case NES_CM_STATE_LISTENING:
-			case NES_CM_STATE_SYN_RCVD:
-			case NES_CM_STATE_SYN_SENT:
-			case NES_CM_STATE_FIN_WAIT1:
-			case NES_CM_STATE_FIN_WAIT2:
-			case NES_CM_STATE_CLOSE_WAIT:
-			case NES_CM_STATE_LAST_ACK:
-			case NES_CM_STATE_CLOSING:
-				break;
-			case  NES_CM_STATE_MPAREQ_SENT:
-				/* recv the mpa res frame, ret=frame len (incl priv data) */
-				ret = parse_mpa(cm_node, dataloc, datasize);
-				if (ret < 0)
-					break;
-				/* set the req frame payload len in skb */
-				/* we are done handling this state, set node to a TSA state */
-				cm_node->state = NES_CM_STATE_TSA;
-				send_ack(cm_node);
-				create_event(cm_node, NES_CM_EVENT_CONNECTED);
-				break;
+	cm_node->state = NES_CM_STATE_CLOSED;
+	create_event(cm_node, NES_CM_EVENT_ABORTED);
+}
 
-			case  NES_CM_STATE_ESTABLISHED:
-				/* we are expecting an MPA req frame */
-				ret = parse_mpa(cm_node, dataloc, datasize);
-				if (ret < 0) {
-					break;
-				}
-				cm_node->state = NES_CM_STATE_TSA;
-				send_ack(cm_node);
-				/* we got a valid MPA request, create an event */
-				create_event(cm_node, NES_CM_EVENT_MPA_REQ);
-				break;
-			case  NES_CM_STATE_TSA:
-				handle_exception_pkt(cm_node, skb);
-				break;
-			case NES_CM_STATE_UNKNOWN:
-			case NES_CM_STATE_INITED:
-			default:
-				ret = -1;
-		}
+/*
+ * passive_open_err() will either do a reset() or will free up the skb and
+ * remove the cm_node.
+ */
+static void passive_open_err(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			     int reset)
+{
+	cleanup_retrans_entry(cm_node);
+	cm_node->state = NES_CM_STATE_CLOSED;
+	if (reset) {
+		nes_debug(NES_DBG_CM, "passive_open_err sending RST for "
+			  "cm_node=%p state =%d\n", cm_node, cm_node->state);
+		send_reset(cm_node, skb);
+	} else {
+		dev_kfree_skb_any(skb);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
 	}
+}
 
-	return ret;
+/*
+ * free_retrans_entry() routines assumes that the retrans_list_lock has
+ * been acquired before calling.
+ */
+static void free_retrans_entry(struct nes_cm_node *cm_node)
+{
+	struct nes_timer_entry *send_entry;
+
+	send_entry = cm_node->send_entry;
+	if (send_entry) {
+		cm_node->send_entry = NULL;
+		dev_kfree_skb_any(send_entry->skb);
+		kfree(send_entry);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+	}
+}
+
+static void cleanup_retrans_entry(struct nes_cm_node *cm_node)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
+	free_retrans_entry(cm_node);
+	spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 }
 
+/**
+ * process_packet
+ * Returns skb if to be freed, else it will return NULL if already used..
+ */
+static void process_packet(struct nes_cm_node *cm_node, struct sk_buff *skb,
+			   struct nes_cm_core *cm_core)
+{
+	enum nes_tcpip_pkt_type pkt_type = NES_PKT_TYPE_UNKNOWN;
+	struct tcphdr *tcph = tcp_hdr(skb);
+	u32 fin_set = 0;
+	int ret = 0;
+
+	skb_pull(skb, ip_hdr(skb)->ihl << 2);
+
+	nes_debug(NES_DBG_CM, "process_packet: cm_node=%p state =%d syn=%d "
+		  "ack=%d rst=%d fin=%d\n", cm_node, cm_node->state, tcph->syn,
+		  tcph->ack, tcph->rst, tcph->fin);
+
+	if (tcph->rst) {
+		pkt_type = NES_PKT_TYPE_RST;
+	} else if (tcph->syn) {
+		pkt_type = NES_PKT_TYPE_SYN;
+		if (tcph->ack)
+			pkt_type = NES_PKT_TYPE_SYNACK;
+	} else if (tcph->ack) {
+		pkt_type = NES_PKT_TYPE_ACK;
+	}
+	if (tcph->fin)
+		fin_set = 1;
+
+	switch (pkt_type) {
+	case NES_PKT_TYPE_SYN:
+		handle_syn_pkt(cm_node, skb, tcph);
+		break;
+	case NES_PKT_TYPE_SYNACK:
+		handle_synack_pkt(cm_node, skb, tcph);
+		break;
+	case NES_PKT_TYPE_ACK:
+		ret = handle_ack_pkt(cm_node, skb, tcph);
+		if (fin_set && !ret)
+			handle_fin_pkt(cm_node);
+		break;
+	case NES_PKT_TYPE_RST:
+		handle_rst_pkt(cm_node, skb, tcph);
+		break;
+	default:
+		if ((fin_set) && (!check_seq(cm_node, tcph, skb)))
+			handle_fin_pkt(cm_node);
+		drop_packet(skb);
+		break;
+	}
+}
 
 /**
  * mini_cm_listen - create a listen node with params
  */
 static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
-		struct nes_vnic *nesvnic, struct nes_cm_info *cm_info)
+			struct nes_vnic *nesvnic, struct nes_cm_info *cm_info)
 {
 	struct nes_cm_listener *listener;
+	struct iwpm_dev_data pm_reg_msg;
+	struct iwpm_sa_data pm_msg;
 	unsigned long flags;
+	int iwpm_err = 0;
 
 	nes_debug(NES_DBG_CM, "Search for 0x%08x : 0x%04x\n",
-		cm_info->loc_addr, cm_info->loc_port);
+		  cm_info->loc_addr, cm_info->loc_port);
 
 	/* cannot have multiple matching listeners */
-	listener = find_listener(cm_core, htonl(cm_info->loc_addr),
-			htons(cm_info->loc_port), NES_CM_LISTENER_EITHER_STATE);
+	listener = find_listener(cm_core, cm_info->loc_addr, cm_info->loc_port,
+				NES_CM_LISTENER_EITHER_STATE, 1);
+
 	if (listener && listener->listener_state == NES_CM_LISTENER_ACTIVE_STATE) {
 		/* find automatically incs ref count ??? */
 		atomic_dec(&listener->ref_count);
@@ -1596,6 +2385,22 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
 	}
 
 	if (!listener) {
+		nes_form_reg_msg(nesvnic, &pm_reg_msg);
+		iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_NES);
+		if (iwpm_err) {
+			nes_debug(NES_DBG_NLMSG,
+			"Port Mapper reg pid fail (err = %d).\n", iwpm_err);
+		}
+		if (iwpm_valid_pid() && !iwpm_err) {
+			nes_form_pm_msg(cm_info, &pm_msg);
+			iwpm_err = iwpm_add_mapping(&pm_msg, RDMA_NL_NES);
+			if (iwpm_err)
+				nes_debug(NES_DBG_NLMSG,
+				"Port Mapper query fail (err = %d).\n", iwpm_err);
+			else
+				nes_record_pm_msg(cm_info, &pm_msg);
+		}
+
 		/* create a CM listen node (1/2 node to compare incoming traffic to) */
 		listener = kzalloc(sizeof(*listener), GFP_ATOMIC);
 		if (!listener) {
@@ -1603,8 +2408,10 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
 			return NULL;
 		}
 
-		listener->loc_addr = htonl(cm_info->loc_addr);
-		listener->loc_port = htons(cm_info->loc_port);
+		listener->loc_addr = cm_info->loc_addr;
+		listener->loc_port = cm_info->loc_port;
+		listener->mapped_loc_addr = cm_info->mapped_loc_addr;
+		listener->mapped_loc_port = cm_info->mapped_loc_port;
 		listener->reused_node = 0;
 
 		atomic_set(&listener->ref_count, 1);
@@ -1633,9 +2440,9 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
 	}
 
 	nes_debug(NES_DBG_CM, "Api - listen(): addr=0x%08X, port=0x%04x,"
-			" listener = %p, backlog = %d, cm_id = %p.\n",
-			cm_info->loc_addr, cm_info->loc_port,
-			listener, listener->backlog, listener->cm_id);
+		  " listener = %p, backlog = %d, cm_id = %p.\n",
+		  cm_info->loc_addr, cm_info->loc_port,
+		  listener, listener->backlog, listener->cm_id);
 
 	return listener;
 }
@@ -1645,23 +2452,15 @@ static struct nes_cm_listener *mini_cm_listen(struct nes_cm_core *cm_core,
  * mini_cm_connect - make a connection node with params
  */
 static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
-					   struct nes_vnic *nesvnic,
-					   struct ietf_mpa_frame *mpa_frame,
-					   struct nes_cm_info *cm_info)
+					   struct nes_vnic *nesvnic, u16 private_data_len,
+					   void *private_data, struct nes_cm_info *cm_info)
 {
 	int ret = 0;
 	struct nes_cm_node *cm_node;
 	struct nes_cm_listener *loopbackremotelistener;
 	struct nes_cm_node *loopbackremotenode;
 	struct nes_cm_info loopback_cm_info;
-
-	u16 mpa_frame_size = sizeof(struct ietf_mpa_frame) +
-			ntohs(mpa_frame->priv_data_len);
-
-	cm_info->loc_addr = htonl(cm_info->loc_addr);
-	cm_info->rem_addr = htonl(cm_info->rem_addr);
-	cm_info->loc_port = htons(cm_info->loc_port);
-	cm_info->rem_port = htons(cm_info->rem_port);
+	u8 *start_buff;
 
 	/* create a CM connection node */
 	cm_node = make_cm_node(cm_core, nesvnic, cm_info, NULL);
@@ -1673,55 +2472,85 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
 	cm_node->tcp_cntxt.rcv_wscale = NES_CM_DEFAULT_RCV_WND_SCALE;
 
 	if (cm_info->loc_addr == cm_info->rem_addr) {
-		loopbackremotelistener = find_listener(cm_core, cm_node->rem_addr,
-				cm_node->rem_port, NES_CM_LISTENER_ACTIVE_STATE);
+		loopbackremotelistener = find_listener(cm_core,
+			cm_node->mapped_loc_addr, cm_node->mapped_rem_port,
+			NES_CM_LISTENER_ACTIVE_STATE, 0);
 		if (loopbackremotelistener == NULL) {
 			create_event(cm_node, NES_CM_EVENT_ABORTED);
 		} else {
-			atomic_inc(&cm_loopbacks);
 			loopback_cm_info = *cm_info;
 			loopback_cm_info.loc_port = cm_info->rem_port;
 			loopback_cm_info.rem_port = cm_info->loc_port;
+			loopback_cm_info.mapped_loc_port =
+				cm_info->mapped_rem_port;
+			loopback_cm_info.mapped_rem_port =
+				cm_info->mapped_loc_port;
 			loopback_cm_info.cm_id = loopbackremotelistener->cm_id;
-			loopbackremotenode = make_cm_node(cm_core, nesvnic, &loopback_cm_info,
-					loopbackremotelistener);
+			loopbackremotenode = make_cm_node(cm_core, nesvnic,
+							  &loopback_cm_info, loopbackremotelistener);
+			if (!loopbackremotenode) {
+				rem_ref_cm_node(cm_node->cm_core, cm_node);
+				return NULL;
+			}
+			atomic_inc(&cm_loopbacks);
 			loopbackremotenode->loopbackpartner = cm_node;
-			loopbackremotenode->tcp_cntxt.rcv_wscale = NES_CM_DEFAULT_RCV_WND_SCALE;
+			loopbackremotenode->tcp_cntxt.rcv_wscale =
+				NES_CM_DEFAULT_RCV_WND_SCALE;
 			cm_node->loopbackpartner = loopbackremotenode;
-			memcpy(loopbackremotenode->mpa_frame_buf, &mpa_frame->priv_data,
-					mpa_frame_size);
-			loopbackremotenode->mpa_frame_size = mpa_frame_size -
-					sizeof(struct ietf_mpa_frame);
+			memcpy(loopbackremotenode->mpa_frame_buf, private_data,
+			       private_data_len);
+			loopbackremotenode->mpa_frame_size = private_data_len;
 
-			/* we are done handling this state, set node to a TSA state */
+			/* we are done handling this state. */
+			/* set node to a TSA state */
 			cm_node->state = NES_CM_STATE_TSA;
-			cm_node->tcp_cntxt.rcv_nxt = loopbackremotenode->tcp_cntxt.loc_seq_num;
-			loopbackremotenode->tcp_cntxt.rcv_nxt = cm_node->tcp_cntxt.loc_seq_num;
-			cm_node->tcp_cntxt.max_snd_wnd = loopbackremotenode->tcp_cntxt.rcv_wnd;
-			loopbackremotenode->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.rcv_wnd;
-			cm_node->tcp_cntxt.snd_wnd = loopbackremotenode->tcp_cntxt.rcv_wnd;
-			loopbackremotenode->tcp_cntxt.snd_wnd = cm_node->tcp_cntxt.rcv_wnd;
-			cm_node->tcp_cntxt.snd_wscale = loopbackremotenode->tcp_cntxt.rcv_wscale;
-			loopbackremotenode->tcp_cntxt.snd_wscale = cm_node->tcp_cntxt.rcv_wscale;
-
+			cm_node->tcp_cntxt.rcv_nxt =
+				loopbackremotenode->tcp_cntxt.loc_seq_num;
+			loopbackremotenode->tcp_cntxt.rcv_nxt =
+				cm_node->tcp_cntxt.loc_seq_num;
+			cm_node->tcp_cntxt.max_snd_wnd =
+				loopbackremotenode->tcp_cntxt.rcv_wnd;
+			loopbackremotenode->tcp_cntxt.max_snd_wnd =
+				cm_node->tcp_cntxt.rcv_wnd;
+			cm_node->tcp_cntxt.snd_wnd =
+				loopbackremotenode->tcp_cntxt.rcv_wnd;
+			loopbackremotenode->tcp_cntxt.snd_wnd =
+				cm_node->tcp_cntxt.rcv_wnd;
+			cm_node->tcp_cntxt.snd_wscale =
+				loopbackremotenode->tcp_cntxt.rcv_wscale;
+			loopbackremotenode->tcp_cntxt.snd_wscale =
+				cm_node->tcp_cntxt.rcv_wscale;
+			loopbackremotenode->state = NES_CM_STATE_MPAREQ_RCVD;
 			create_event(loopbackremotenode, NES_CM_EVENT_MPA_REQ);
 		}
 		return cm_node;
 	}
 
-	/* set our node side to client (active) side */
-	cm_node->tcp_cntxt.client = 1;
-	/* init our MPA frame ptr */
-	memcpy(&cm_node->mpa_frame, mpa_frame, mpa_frame_size);
-	cm_node->mpa_frame_size = mpa_frame_size;
+	start_buff = &cm_node->mpa_frame_buf[0] + sizeof(struct ietf_mpa_v2);
+	cm_node->mpa_frame_size = private_data_len;
+
+	memcpy(start_buff, private_data, private_data_len);
 
 	/* send a syn and goto syn sent state */
 	cm_node->state = NES_CM_STATE_SYN_SENT;
-	ret = send_syn(cm_node, 0);
+	ret = send_syn(cm_node, 0, NULL);
+
+	if (ret) {
+		/* error in sending the syn free up the cm_node struct */
+		nes_debug(NES_DBG_CM, "Api - connect() FAILED: dest "
+			  "addr=0x%08X, port=0x%04x, cm_node=%p, cm_id = %p.\n",
+			  cm_node->rem_addr, cm_node->rem_port, cm_node,
+			  cm_node->cm_id);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		cm_node = NULL;
+	}
 
-	nes_debug(NES_DBG_CM, "Api - connect(): dest addr=0x%08X, port=0x%04x,"
-			" cm_node=%p, cm_id = %p.\n",
-			cm_node->rem_addr, cm_node->rem_port, cm_node, cm_node->cm_id);
+	if (cm_node) {
+		nes_debug(NES_DBG_CM, "Api - connect(): dest addr=0x%08X,"
+			  "port=0x%04x, cm_node=%p, cm_id = %p.\n",
+			  cm_node->rem_addr, cm_node->rem_port, cm_node,
+			  cm_node->cm_id);
+	}
 
 	return cm_node;
 }
@@ -1731,8 +2560,7 @@ static struct nes_cm_node *mini_cm_connect(struct nes_cm_core *cm_core,
  * mini_cm_accept - accept a connection
  * This function is never called
  */
-static int mini_cm_accept(struct nes_cm_core *cm_core, struct ietf_mpa_frame *mpa_frame,
-			  struct nes_cm_node *cm_node)
+static int mini_cm_accept(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node)
 {
 	return 0;
 }
@@ -1741,31 +2569,62 @@ static int mini_cm_accept(struct nes_cm_core *cm_core, struct ietf_mpa_frame *mp
 /**
  * mini_cm_reject - reject and teardown a connection
  */
-static int mini_cm_reject(struct nes_cm_core *cm_core,
-			  struct ietf_mpa_frame *mpa_frame,
-			  struct nes_cm_node *cm_node)
+static int mini_cm_reject(struct nes_cm_core *cm_core, struct nes_cm_node *cm_node)
 {
 	int ret = 0;
-	struct sk_buff *skb;
-	u16 mpa_frame_size = sizeof(struct ietf_mpa_frame) +
-			ntohs(mpa_frame->priv_data_len);
+	int err = 0;
+	int passive_state;
+	struct nes_cm_event event;
+	struct iw_cm_id *cm_id = cm_node->cm_id;
+	struct nes_cm_node *loopback = cm_node->loopbackpartner;
 
-	skb = get_free_pkt(cm_node);
-	if (!skb) {
-		nes_debug(NES_DBG_CM, "Failed to get a Free pkt\n");
-		return -1;
-	}
+	nes_debug(NES_DBG_CM, "%s cm_node=%p type=%d state=%d\n",
+		  __func__, cm_node, cm_node->tcp_cntxt.client, cm_node->state);
 
-	/* send an MPA Request frame */
-	form_cm_frame(skb, cm_node, NULL, 0, mpa_frame, mpa_frame_size, SET_ACK | SET_FIN);
-	ret = schedule_nes_timer(cm_node, skb, NES_TIMER_TYPE_SEND, 1, 0);
+	if (cm_node->tcp_cntxt.client)
+		return ret;
+	cleanup_retrans_entry(cm_node);
 
-	cm_node->state = NES_CM_STATE_CLOSED;
-	ret = send_fin(cm_node, NULL);
+	if (!loopback) {
+		passive_state = atomic_add_return(1, &cm_node->passive_state);
+		if (passive_state == NES_SEND_RESET_EVENT) {
+			cm_node->state = NES_CM_STATE_CLOSED;
+			rem_ref_cm_node(cm_core, cm_node);
+		} else {
+			if (cm_node->state == NES_CM_STATE_LISTENER_DESTROYED) {
+				rem_ref_cm_node(cm_core, cm_node);
+			} else {
+				ret = send_mpa_reject(cm_node);
+				if (ret) {
+					cm_node->state = NES_CM_STATE_CLOSED;
+					err = send_reset(cm_node, NULL);
+					if (err)
+						WARN_ON(1);
+				} else {
+					cm_id->add_ref(cm_id);
+				}
+			}
+		}
+	} else {
+		cm_node->cm_id = NULL;
+		if (cm_node->state == NES_CM_STATE_LISTENER_DESTROYED) {
+			rem_ref_cm_node(cm_core, cm_node);
+			rem_ref_cm_node(cm_core, loopback);
+		} else {
+			event.cm_node = loopback;
+			event.cm_info.rem_addr = loopback->rem_addr;
+			event.cm_info.loc_addr = loopback->loc_addr;
+			event.cm_info.rem_port = loopback->rem_port;
+			event.cm_info.loc_port = loopback->loc_port;
+			event.cm_info.cm_id = loopback->cm_id;
+			cm_event_mpa_reject(&event);
+			rem_ref_cm_node(cm_core, cm_node);
+			loopback->state = NES_CM_STATE_CLOSING;
 
-	if (ret < 0) {
-		printk(KERN_INFO PFX "failed to send MPA Reply (reject)\n");
-		return ret;
+			cm_id = loopback->cm_id;
+			rem_ref_cm_node(cm_core, loopback);
+			cm_id->rem_ref(cm_id);
+		}
 	}
 
 	return ret;
@@ -1783,37 +2642,45 @@ static int mini_cm_close(struct nes_cm_core *cm_core, struct nes_cm_node *cm_nod
 		return -EINVAL;
 
 	switch (cm_node->state) {
-		/* if passed in node is null, create a reference key node for node search */
-		/* check if we found an owner node for this pkt */
-		case NES_CM_STATE_SYN_RCVD:
-		case NES_CM_STATE_SYN_SENT:
-		case NES_CM_STATE_ONE_SIDE_ESTABLISHED:
-		case NES_CM_STATE_ESTABLISHED:
-		case NES_CM_STATE_ACCEPTING:
-		case NES_CM_STATE_MPAREQ_SENT:
-			cm_node->state = NES_CM_STATE_FIN_WAIT1;
-			send_fin(cm_node, NULL);
-			break;
-		case NES_CM_STATE_CLOSE_WAIT:
-			cm_node->state = NES_CM_STATE_LAST_ACK;
-			send_fin(cm_node, NULL);
-			break;
-		case NES_CM_STATE_FIN_WAIT1:
-		case NES_CM_STATE_FIN_WAIT2:
-		case NES_CM_STATE_LAST_ACK:
-		case NES_CM_STATE_TIME_WAIT:
-		case NES_CM_STATE_CLOSING:
-			ret = -1;
-			break;
-		case NES_CM_STATE_LISTENING:
-		case NES_CM_STATE_UNKNOWN:
-		case NES_CM_STATE_INITED:
-		case NES_CM_STATE_CLOSED:
-		case NES_CM_STATE_TSA:
-			ret = rem_ref_cm_node(cm_core, cm_node);
-			break;
+	case NES_CM_STATE_SYN_RCVD:
+	case NES_CM_STATE_SYN_SENT:
+	case NES_CM_STATE_ONE_SIDE_ESTABLISHED:
+	case NES_CM_STATE_ESTABLISHED:
+	case NES_CM_STATE_ACCEPTING:
+	case NES_CM_STATE_MPAREQ_SENT:
+	case NES_CM_STATE_MPAREQ_RCVD:
+		cleanup_retrans_entry(cm_node);
+		send_reset(cm_node, NULL);
+		break;
+	case NES_CM_STATE_CLOSE_WAIT:
+		cm_node->state = NES_CM_STATE_LAST_ACK;
+		send_fin(cm_node, NULL);
+		break;
+	case NES_CM_STATE_FIN_WAIT1:
+	case NES_CM_STATE_FIN_WAIT2:
+	case NES_CM_STATE_LAST_ACK:
+	case NES_CM_STATE_TIME_WAIT:
+	case NES_CM_STATE_CLOSING:
+		ret = -1;
+		break;
+	case NES_CM_STATE_LISTENING:
+		cleanup_retrans_entry(cm_node);
+		send_reset(cm_node, NULL);
+		break;
+	case NES_CM_STATE_MPAREJ_RCVD:
+	case NES_CM_STATE_UNKNOWN:
+	case NES_CM_STATE_INITED:
+	case NES_CM_STATE_CLOSED:
+	case NES_CM_STATE_LISTENER_DESTROYED:
+		ret = rem_ref_cm_node(cm_core, cm_node);
+		break;
+	case NES_CM_STATE_TSA:
+		if (cm_node->send_entry)
+			printk(KERN_ERR "ERROR Close got called from STATE_TSA "
+			       "send_entry=%p\n", cm_node->send_entry);
+		ret = rem_ref_cm_node(cm_core, cm_node);
+		break;
 	}
-	cm_node->cm_id = NULL;
 	return ret;
 }
 
@@ -1822,92 +2689,102 @@ static int mini_cm_close(struct nes_cm_core *cm_core, struct nes_cm_node *cm_nod
  * recv_pkt - recv an ETHERNET packet, and process it through CM
  * node state machine
  */
-static int mini_cm_recv_pkt(struct nes_cm_core *cm_core, struct nes_vnic *nesvnic,
-			    struct sk_buff *skb)
+static int mini_cm_recv_pkt(struct nes_cm_core *cm_core,
+			    struct nes_vnic *nesvnic, struct sk_buff *skb)
 {
 	struct nes_cm_node *cm_node = NULL;
 	struct nes_cm_listener *listener = NULL;
 	struct iphdr *iph;
 	struct tcphdr *tcph;
 	struct nes_cm_info nfo;
-	int ret = 0;
+	int skb_handled = 1;
+	__be32 tmp_daddr, tmp_saddr;
 
-	if (!skb || skb->len < sizeof(struct iphdr) + sizeof(struct tcphdr)) {
-		ret = -EINVAL;
-		goto out;
-	}
+	if (!skb)
+		return 0;
+	if (skb->len < sizeof(struct iphdr) + sizeof(struct tcphdr))
+		return 0;
 
 	iph = (struct iphdr *)skb->data;
 	tcph = (struct tcphdr *)(skb->data + sizeof(struct iphdr));
-	skb_reset_network_header(skb);
-	skb_set_transport_header(skb, sizeof(*tcph));
-	skb->len = ntohs(iph->tot_len);
 
 	nfo.loc_addr = ntohl(iph->daddr);
 	nfo.loc_port = ntohs(tcph->dest);
 	nfo.rem_addr = ntohl(iph->saddr);
 	nfo.rem_port = ntohs(tcph->source);
 
-	nes_debug(NES_DBG_CM, "Received packet: dest=" NIPQUAD_FMT
-		  ":0x%04X src=" NIPQUAD_FMT ":0x%04X\n",
-		  NIPQUAD(iph->daddr), tcph->dest,
-		  NIPQUAD(iph->saddr), tcph->source);
+	/* If port mapper is available these should be mapped address info */
+	nfo.mapped_loc_addr = ntohl(iph->daddr);
+	nfo.mapped_loc_port = ntohs(tcph->dest);
+	nfo.mapped_rem_addr = ntohl(iph->saddr);
+	nfo.mapped_rem_port = ntohs(tcph->source);
 
-	/* note: this call is going to increment cm_node ref count */
-	cm_node = find_node(cm_core,
-			nfo.rem_port, nfo.rem_addr,
-			nfo.loc_port, nfo.loc_addr);
+	tmp_daddr = cpu_to_be32(iph->daddr);
+	tmp_saddr = cpu_to_be32(iph->saddr);
 
-	if (!cm_node) {
-		listener = find_listener(cm_core, nfo.loc_addr, nfo.loc_port,
-				NES_CM_LISTENER_ACTIVE_STATE);
-		if (listener) {
-			nfo.cm_id = listener->cm_id;
-			nfo.conn_type = listener->conn_type;
-		} else {
-			nfo.cm_id = NULL;
-			nfo.conn_type = 0;
-		}
+	nes_debug(NES_DBG_CM, "Received packet: dest=%pI4:0x%04X src=%pI4:0x%04X\n",
+		  &tmp_daddr, tcph->dest, &tmp_saddr, tcph->source);
+
+	do {
+		cm_node = find_node(cm_core,
+				    nfo.mapped_rem_port, nfo.mapped_rem_addr,
+				    nfo.mapped_loc_port, nfo.mapped_loc_addr);
 
-		cm_node = make_cm_node(cm_core, nesvnic, &nfo, listener);
 		if (!cm_node) {
-			nes_debug(NES_DBG_CM, "Unable to allocate node\n");
-			if (listener) {
-				nes_debug(NES_DBG_CM, "unable to allocate node and decrementing listener refcount\n");
+			/* Only type of packet accepted are for */
+			/* the PASSIVE open (syn only) */
+			if ((!tcph->syn) || (tcph->ack)) {
+				skb_handled = 0;
+				break;
+			}
+			listener = find_listener(cm_core, nfo.mapped_loc_addr,
+					nfo.mapped_loc_port,
+					NES_CM_LISTENER_ACTIVE_STATE, 0);
+			if (!listener) {
+				nfo.cm_id = NULL;
+				nfo.conn_type = 0;
+				nes_debug(NES_DBG_CM, "Unable to find listener for the pkt\n");
+				skb_handled = 0;
+				break;
+			}
+			nfo.cm_id = listener->cm_id;
+			nfo.conn_type = listener->conn_type;
+			cm_node = make_cm_node(cm_core, nesvnic, &nfo,
+					       listener);
+			if (!cm_node) {
+				nes_debug(NES_DBG_CM, "Unable to allocate "
+					  "node\n");
+				cm_packets_dropped++;
 				atomic_dec(&listener->ref_count);
+				dev_kfree_skb_any(skb);
+				break;
 			}
-			ret = -1;
-			goto out;
-		}
-		if (!listener) {
-			nes_debug(NES_DBG_CM, "Packet found for unknown port %x refcnt=%d\n",
-					nfo.loc_port, atomic_read(&cm_node->ref_count));
-			if (!tcph->rst) {
-				nes_debug(NES_DBG_CM, "Packet found for unknown port=%d"
-						" rem_port=%d refcnt=%d\n",
-						nfo.loc_port, nfo.rem_port, atomic_read(&cm_node->ref_count));
-
-				cm_node->tcp_cntxt.rcv_nxt = ntohl(tcph->seq);
-				cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
-				send_reset(cm_node);
+			if (!tcph->rst && !tcph->fin) {
+				cm_node->state = NES_CM_STATE_LISTENING;
+			} else {
+				cm_packets_dropped++;
+				rem_ref_cm_node(cm_core, cm_node);
+				dev_kfree_skb_any(skb);
+				break;
 			}
-			rem_ref_cm_node(cm_core, cm_node);
-			ret = -1;
-			goto out;
+			add_ref_cm_node(cm_node);
+		} else if (cm_node->state == NES_CM_STATE_TSA) {
+			if (cm_node->nesqp->pau_mode)
+				nes_queue_mgt_skbs(skb, nesvnic, cm_node->nesqp);
+			else {
+				rem_ref_cm_node(cm_core, cm_node);
+				atomic_inc(&cm_accel_dropped_pkts);
+				dev_kfree_skb_any(skb);
+			}
+			break;
 		}
-		add_ref_cm_node(cm_node);
-		cm_node->state = NES_CM_STATE_LISTENING;
-	}
-
-	nes_debug(NES_DBG_CM, "Processing Packet for node %p, data = (%p):\n",
-			cm_node, skb->data);
-	process_packet(cm_node, skb, cm_core);
-
-	rem_ref_cm_node(cm_core, cm_node);
-	out:
-	if (skb)
-		dev_kfree_skb_any(skb);
-	return ret;
+		skb_reset_network_header(skb);
+		skb_set_transport_header(skb, sizeof(*tcph));
+		skb->len = ntohs(iph->tot_len);
+		process_packet(cm_node, skb, cm_core);
+		rem_ref_cm_node(cm_core, cm_node);
+	} while (0);
+	return skb_handled;
 }
 
 
@@ -1916,10 +2793,7 @@ static int mini_cm_recv_pkt(struct nes_cm_core *cm_core, struct nes_vnic *nesvni
  */
 static struct nes_cm_core *nes_cm_alloc_core(void)
 {
-	int i;
-
 	struct nes_cm_core *cm_core;
-	struct sk_buff *skb = NULL;
 
 	/* setup the CM core */
 	/* alloc top level core control structure */
@@ -1931,25 +2805,12 @@ static struct nes_cm_core *nes_cm_alloc_core(void)
 	init_timer(&cm_core->tcp_timer);
 	cm_core->tcp_timer.function = nes_cm_timer_tick;
 
-	cm_core->mtu   = NES_CM_DEFAULT_MTU;
+	cm_core->mtu = NES_CM_DEFAULT_MTU;
 	cm_core->state = NES_CM_STATE_INITED;
 	cm_core->free_tx_pkt_max = NES_CM_DEFAULT_FREE_PKTS;
 
 	atomic_set(&cm_core->events_posted, 0);
 
-	/* init the packet lists */
-	skb_queue_head_init(&cm_core->tx_free_list);
-
-	for (i = 0; i < NES_CM_DEFAULT_FRAME_CNT; i++) {
-		skb = dev_alloc_skb(cm_core->mtu);
-		if (!skb) {
-			kfree(cm_core);
-			return NULL;
-		}
-		/* add 'raw' skb to free frame list */
-		skb_queue_head(&cm_core->tx_free_list, skb);
-	}
-
 	cm_core->api = &nes_cm_api;
 
 	spin_lock_init(&cm_core->ht_lock);
@@ -1982,9 +2843,8 @@ static int mini_cm_dealloc_core(struct nes_cm_core *cm_core)
 
 	barrier();
 
-	if (timer_pending(&cm_core->tcp_timer)) {
+	if (timer_pending(&cm_core->tcp_timer))
 		del_timer(&cm_core->tcp_timer);
-	}
 
 	destroy_workqueue(cm_core->event_wq);
 	destroy_workqueue(cm_core->disconn_wq);
@@ -2012,15 +2872,15 @@ static int mini_cm_set(struct nes_cm_core *cm_core, u32 type, u32 value)
 	int ret = 0;
 
 	switch (type) {
-		case NES_CM_SET_PKT_SIZE:
-			cm_core->mtu = value;
-			break;
-		case NES_CM_SET_FREE_PKT_Q_SIZE:
-			cm_core->free_tx_pkt_max = value;
-			break;
-		default:
-			/* unknown set option */
-			ret = -EINVAL;
+	case NES_CM_SET_PKT_SIZE:
+		cm_core->mtu = value;
+		break;
+	case NES_CM_SET_FREE_PKT_Q_SIZE:
+		cm_core->free_tx_pkt_max = value;
+		break;
+	default:
+		/* unknown set option */
+		ret = -EINVAL;
 	}
 
 	return ret;
@@ -2039,8 +2899,8 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
 		return -EINVAL;
 
 	nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_IPV4 |
-			NES_QPCONTEXT_MISC_NO_NAGLE | NES_QPCONTEXT_MISC_DO_NOT_FRAG |
-			NES_QPCONTEXT_MISC_DROS);
+						  NES_QPCONTEXT_MISC_NO_NAGLE | NES_QPCONTEXT_MISC_DO_NOT_FRAG |
+						  NES_QPCONTEXT_MISC_DROS);
 
 	if (cm_node->tcp_cntxt.snd_wscale || cm_node->tcp_cntxt.rcv_wscale)
 		nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_WSCALE);
@@ -2050,15 +2910,15 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
 	nesqp->nesqp_context->mss |= cpu_to_le32(((u32)cm_node->tcp_cntxt.mss) << 16);
 
 	nesqp->nesqp_context->tcp_state_flow_label |= cpu_to_le32(
-			(u32)NES_QPCONTEXT_TCPSTATE_EST << NES_QPCONTEXT_TCPFLOW_TCP_STATE_SHIFT);
+		(u32)NES_QPCONTEXT_TCPSTATE_EST << NES_QPCONTEXT_TCPFLOW_TCP_STATE_SHIFT);
 
 	nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32(
-			(cm_node->tcp_cntxt.snd_wscale << NES_QPCONTEXT_PDWSCALE_SND_WSCALE_SHIFT) &
-			NES_QPCONTEXT_PDWSCALE_SND_WSCALE_MASK);
+		(cm_node->tcp_cntxt.snd_wscale << NES_QPCONTEXT_PDWSCALE_SND_WSCALE_SHIFT) &
+		NES_QPCONTEXT_PDWSCALE_SND_WSCALE_MASK);
 
 	nesqp->nesqp_context->pd_index_wscale |= cpu_to_le32(
-			(cm_node->tcp_cntxt.rcv_wscale << NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_SHIFT) &
-			NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_MASK);
+		(cm_node->tcp_cntxt.rcv_wscale << NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_SHIFT) &
+		NES_QPCONTEXT_PDWSCALE_RCV_WSCALE_MASK);
 
 	nesqp->nesqp_context->keepalive = cpu_to_le32(0x80);
 	nesqp->nesqp_context->ts_recent = 0;
@@ -2067,24 +2927,24 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
 	nesqp->nesqp_context->snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.snd_wnd);
 	nesqp->nesqp_context->rcv_nxt = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt);
 	nesqp->nesqp_context->rcv_wnd = cpu_to_le32(cm_node->tcp_cntxt.rcv_wnd <<
-			cm_node->tcp_cntxt.rcv_wscale);
+						    cm_node->tcp_cntxt.rcv_wscale);
 	nesqp->nesqp_context->snd_max = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
 	nesqp->nesqp_context->snd_una = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
 	nesqp->nesqp_context->srtt = 0;
 	nesqp->nesqp_context->rttvar = cpu_to_le32(0x6);
 	nesqp->nesqp_context->ssthresh = cpu_to_le32(0x3FFFC000);
-	nesqp->nesqp_context->cwnd = cpu_to_le32(2*cm_node->tcp_cntxt.mss);
+	nesqp->nesqp_context->cwnd = cpu_to_le32(2 * cm_node->tcp_cntxt.mss);
 	nesqp->nesqp_context->snd_wl1 = cpu_to_le32(cm_node->tcp_cntxt.rcv_nxt);
 	nesqp->nesqp_context->snd_wl2 = cpu_to_le32(cm_node->tcp_cntxt.loc_seq_num);
 	nesqp->nesqp_context->max_snd_wnd = cpu_to_le32(cm_node->tcp_cntxt.max_snd_wnd);
 
 	nes_debug(NES_DBG_CM, "QP%u: rcv_nxt = 0x%08X, snd_nxt = 0x%08X,"
-			" Setting MSS to %u, PDWscale = 0x%08X, rcv_wnd = %u, context misc = 0x%08X.\n",
-			nesqp->hwqp.qp_id, le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
-			le32_to_cpu(nesqp->nesqp_context->snd_nxt),
-			cm_node->tcp_cntxt.mss, le32_to_cpu(nesqp->nesqp_context->pd_index_wscale),
-			le32_to_cpu(nesqp->nesqp_context->rcv_wnd),
-			le32_to_cpu(nesqp->nesqp_context->misc));
+		  " Setting MSS to %u, PDWscale = 0x%08X, rcv_wnd = %u, context misc = 0x%08X.\n",
+		  nesqp->hwqp.qp_id, le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
+		  le32_to_cpu(nesqp->nesqp_context->snd_nxt),
+		  cm_node->tcp_cntxt.mss, le32_to_cpu(nesqp->nesqp_context->pd_index_wscale),
+		  le32_to_cpu(nesqp->nesqp_context->rcv_wnd),
+		  le32_to_cpu(nesqp->nesqp_context->misc));
 	nes_debug(NES_DBG_CM, "  snd_wnd  = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->snd_wnd));
 	nes_debug(NES_DBG_CM, "  snd_cwnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->cwnd));
 	nes_debug(NES_DBG_CM, "  max_swnd = 0x%08X.\n", le32_to_cpu(nesqp->nesqp_context->max_snd_wnd));
@@ -2101,22 +2961,16 @@ static int nes_cm_init_tsa_conn(struct nes_qp *nesqp, struct nes_cm_node *cm_nod
  */
 int nes_cm_disconn(struct nes_qp *nesqp)
 {
-	unsigned long flags;
+	struct disconn_work *work;
 
-	spin_lock_irqsave(&nesqp->lock, flags);
-	if (nesqp->disconn_pending == 0) {
-		nesqp->disconn_pending++;
-		spin_unlock_irqrestore(&nesqp->lock, flags);
-		/* nes_add_ref(&nesqp->ibqp); */
-		/* init our disconnect work element, to */
-		INIT_WORK(&nesqp->disconn_work, nes_disconnect_worker);
-
-		queue_work(g_cm_core->disconn_wq, &nesqp->disconn_work);
-	} else {
-		spin_unlock_irqrestore(&nesqp->lock, flags);
-		nes_rem_ref(&nesqp->ibqp);
-	}
+	work = kzalloc(sizeof *work, GFP_ATOMIC);
+	if (!work)
+		return -ENOMEM;  /* Timer will clean up */
 
+	nes_add_ref(&nesqp->ibqp);
+	work->nesqp = nesqp;
+	INIT_WORK(&work->work, nes_disconnect_worker);
+	queue_work(g_cm_core->disconn_wq, &work->work);
 	return 0;
 }
 
@@ -2126,11 +2980,14 @@ int nes_cm_disconn(struct nes_qp *nesqp)
  */
 static void nes_disconnect_worker(struct work_struct *work)
 {
-	struct nes_qp *nesqp = container_of(work, struct nes_qp, disconn_work);
+	struct disconn_work *dwork = container_of(work, struct disconn_work, work);
+	struct nes_qp *nesqp = dwork->nesqp;
 
+	kfree(dwork);
 	nes_debug(NES_DBG_CM, "processing AEQE id 0x%04X for QP%u.\n",
-			nesqp->last_aeq, nesqp->hwqp.qp_id);
+		  nesqp->last_aeq, nesqp->hwqp.qp_id);
 	nes_cm_disconn_true(nesqp);
+	nes_rem_ref(&nesqp->ibqp);
 }
 
 
@@ -2147,7 +3004,12 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
 	u16 last_ae;
 	u8 original_hw_tcp_state;
 	u8 original_ibqp_state;
-	u8 issued_disconnect_reset = 0;
+	int disconn_status = 0;
+	int issue_disconn = 0;
+	int issue_close = 0;
+	int issue_flush = 0;
+	u32 flush_q = NES_CQP_FLUSH_RQ;
+	struct ib_event ibevent;
 
 	if (!nesqp) {
 		nes_debug(NES_DBG_CM, "disconnect_worker nesqp is NULL\n");
@@ -2159,9 +3021,8 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
 	/* make sure we havent already closed this connection */
 	if (!cm_id) {
 		nes_debug(NES_DBG_CM, "QP%u disconnect_worker cmid is NULL\n",
-				nesqp->hwqp.qp_id);
+			  nesqp->hwqp.qp_id);
 		spin_unlock_irqrestore(&nesqp->lock, flags);
-		nes_rem_ref(&nesqp->ibqp);
 		return -1;
 	}
 
@@ -2169,67 +3030,86 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
 	nes_debug(NES_DBG_CM, "Disconnecting QP%u\n", nesqp->hwqp.qp_id);
 
 	original_hw_tcp_state = nesqp->hw_tcp_state;
-	original_ibqp_state   = nesqp->ibqp_state;
+	original_ibqp_state = nesqp->ibqp_state;
 	last_ae = nesqp->last_aeq;
 
+	if (nesqp->term_flags) {
+		issue_disconn = 1;
+		issue_close = 1;
+		nesqp->cm_id = NULL;
+		del_timer(&nesqp->terminate_timer);
+		if (nesqp->flush_issued == 0) {
+			nesqp->flush_issued = 1;
+			issue_flush = 1;
+		}
+	} else if ((original_hw_tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
+			((original_ibqp_state == IB_QPS_RTS) &&
+			(last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
+		issue_disconn = 1;
+		if (last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET)
+			disconn_status = -ECONNRESET;
+	}
 
-	nes_debug(NES_DBG_CM, "set ibqp_state=%u\n", nesqp->ibqp_state);
+	if (((original_hw_tcp_state == NES_AEQE_TCP_STATE_CLOSED) ||
+		 (original_hw_tcp_state == NES_AEQE_TCP_STATE_TIME_WAIT) ||
+		 (last_ae == NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) ||
+		 (last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
+		issue_close = 1;
+		nesqp->cm_id = NULL;
+		if (nesqp->flush_issued == 0) {
+			nesqp->flush_issued = 1;
+			issue_flush = 1;
+		}
+	}
+
+	spin_unlock_irqrestore(&nesqp->lock, flags);
+
+	if ((issue_flush) && (nesqp->destroyed == 0)) {
+		/* Flush the queue(s) */
+		if (nesqp->hw_iwarp_state >= NES_AEQE_IWARP_STATE_TERMINATE)
+			flush_q |= NES_CQP_FLUSH_SQ;
+		flush_wqes(nesvnic->nesdev, nesqp, flush_q, 1);
+
+		if (nesqp->term_flags) {
+			ibevent.device = nesqp->ibqp.device;
+			ibevent.event = nesqp->terminate_eventtype;
+			ibevent.element.qp = &nesqp->ibqp;
+			if (nesqp->ibqp.event_handler)
+				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
+		}
+	}
 
-	if ((nesqp->cm_id) && (cm_id->event_handler)) {
-		if ((original_hw_tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
-				((original_ibqp_state == IB_QPS_RTS) &&
-				(last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
+	if ((cm_id) && (cm_id->event_handler)) {
+		if (issue_disconn) {
 			atomic_inc(&cm_disconnects);
 			cm_event.event = IW_CM_EVENT_DISCONNECT;
-			if (last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET) {
-				issued_disconnect_reset = 1;
-				cm_event.status = IW_CM_EVENT_STATUS_RESET;
-				nes_debug(NES_DBG_CM, "Generating a CM Disconnect Event (status reset) for "
-						" QP%u, cm_id = %p. \n",
-						nesqp->hwqp.qp_id, cm_id);
-			} else {
-				cm_event.status = IW_CM_EVENT_STATUS_OK;
-			}
-
+			cm_event.status = disconn_status;
 			cm_event.local_addr = cm_id->local_addr;
 			cm_event.remote_addr = cm_id->remote_addr;
 			cm_event.private_data = NULL;
 			cm_event.private_data_len = 0;
 
-			nes_debug(NES_DBG_CM, "Generating a CM Disconnect Event for "
-					" QP%u, SQ Head = %u, SQ Tail = %u. cm_id = %p, refcount = %u.\n",
-					nesqp->hwqp.qp_id,
-					nesqp->hwqp.sq_head, nesqp->hwqp.sq_tail, cm_id,
-					atomic_read(&nesqp->refcount));
+			nes_debug(NES_DBG_CM, "Generating a CM Disconnect Event"
+				  " for  QP%u, SQ Head = %u, SQ Tail = %u. "
+				  "cm_id = %p, refcount = %u.\n",
+				  nesqp->hwqp.qp_id, nesqp->hwqp.sq_head,
+				  nesqp->hwqp.sq_tail, cm_id,
+				  atomic_read(&nesqp->refcount));
 
-			spin_unlock_irqrestore(&nesqp->lock, flags);
 			ret = cm_id->event_handler(cm_id, &cm_event);
 			if (ret)
-				nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
-			spin_lock_irqsave(&nesqp->lock, flags);
+				nes_debug(NES_DBG_CM, "OFA CM event_handler "
+					  "returned, ret=%d\n", ret);
 		}
 
-		nesqp->disconn_pending = 0;
-		/* There might have been another AE while the lock was released */
-		original_hw_tcp_state = nesqp->hw_tcp_state;
-		original_ibqp_state   = nesqp->ibqp_state;
-		last_ae = nesqp->last_aeq;
-
-		if ((issued_disconnect_reset == 0) && (nesqp->cm_id) &&
-				((original_hw_tcp_state == NES_AEQE_TCP_STATE_CLOSED) ||
-				 (original_hw_tcp_state == NES_AEQE_TCP_STATE_TIME_WAIT) ||
-				 (last_ae == NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) ||
-				 (last_ae == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
+		if (issue_close) {
 			atomic_inc(&cm_closes);
-			nesqp->cm_id = NULL;
-			nesqp->in_disconnect = 0;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
 			nes_disconnect(nesqp, 1);
 
 			cm_id->provider_data = nesqp;
 			/* Send up the close complete event */
 			cm_event.event = IW_CM_EVENT_CLOSE;
-			cm_event.status = IW_CM_EVENT_STATUS_OK;
+			cm_event.status = 0;
 			cm_event.provider_data = cm_id->provider_data;
 			cm_event.local_addr = cm_id->local_addr;
 			cm_event.remote_addr = cm_id->remote_addr;
@@ -2237,41 +3117,12 @@ static int nes_cm_disconn_true(struct nes_qp *nesqp)
 			cm_event.private_data_len = 0;
 
 			ret = cm_id->event_handler(cm_id, &cm_event);
-			if (ret) {
+			if (ret)
 				nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
-			}
 
 			cm_id->rem_ref(cm_id);
-
-			spin_lock_irqsave(&nesqp->lock, flags);
-			if (nesqp->flush_issued == 0) {
-				nesqp->flush_issued = 1;
-				spin_unlock_irqrestore(&nesqp->lock, flags);
-				flush_wqes(nesvnic->nesdev, nesqp, NES_CQP_FLUSH_RQ, 1);
-			} else {
-				spin_unlock_irqrestore(&nesqp->lock, flags);
-			}
-
-			/* This reference is from either ModifyQP or the AE processing,
-					there is still a race here with modifyqp */
-			nes_rem_ref(&nesqp->ibqp);
-
-		} else {
-			cm_id = nesqp->cm_id;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			/* check to see if the inbound reset beat the outbound reset */
-			if ((!cm_id) && (last_ae==NES_AEQE_AEID_RESET_SENT)) {
-				nes_debug(NES_DBG_CM, "QP%u: Decing refcount due to inbound reset"
-						" beating the outbound reset.\n",
-						nesqp->hwqp.qp_id);
-				nes_rem_ref(&nesqp->ibqp);
-			}
 		}
-	} else {
-		nesqp->disconn_pending = 0;
-		spin_unlock_irqrestore(&nesqp->lock, flags);
 	}
-	nes_rem_ref(&nesqp->ibqp);
 
 	return 0;
 }
@@ -2285,15 +3136,17 @@ static int nes_disconnect(struct nes_qp *nesqp, int abrupt)
 	int ret = 0;
 	struct nes_vnic *nesvnic;
 	struct nes_device *nesdev;
+	struct nes_ib_device *nesibdev;
 
 	nesvnic = to_nesvnic(nesqp->ibqp.device);
 	if (!nesvnic)
 		return -EINVAL;
 
 	nesdev = nesvnic->nesdev;
+	nesibdev = nesvnic->nesibdev;
 
 	nes_debug(NES_DBG_CM, "netdev refcnt = %u.\n",
-			atomic_read(&nesvnic->netdev->refcnt));
+			netdev_refcnt_read(nesvnic->netdev));
 
 	if (nesqp->active_conn) {
 
@@ -2302,9 +3155,11 @@ static int nes_disconnect(struct nes_qp *nesqp, int abrupt)
 	} else {
 		/* Need to free the Last Streaming Mode Message */
 		if (nesqp->ietf_frame) {
+			if (nesqp->lsmm_mr)
+				nesibdev->ibdev.dereg_mr(nesqp->lsmm_mr);
 			pci_free_consistent(nesdev->pcidev,
-					nesqp->private_data_len+sizeof(struct ietf_mpa_frame),
-					nesqp->ietf_frame, nesqp->ietf_frame_pbase);
+					    nesqp->private_data_len + nesqp->ietf_frame_size,
+					    nesqp->ietf_frame, nesqp->ietf_frame_pbase);
 		}
 	}
 
@@ -2313,7 +3168,6 @@ static int nes_disconnect(struct nes_qp *nesqp, int abrupt)
 		nes_debug(NES_DBG_CM, "Call close API\n");
 
 		g_cm_core->api->close(g_cm_core, nesqp->cm_node);
-		nesqp->cm_node = NULL;
 	}
 
 	return ret;
@@ -2338,6 +3192,20 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	struct nes_v4_quad nes_quad;
 	u32 crc_value;
 	int ret;
+	int passive_state;
+	struct nes_ib_device *nesibdev;
+	struct ib_mr *ibmr = NULL;
+	struct ib_phys_buf ibphysbuf;
+	struct nes_pd *nespd;
+	u64 tagged_offset;
+	u8 mpa_frame_offset = 0;
+	struct ietf_mpa_v2 *mpa_v2_frame;
+	u8 start_addr = 0;
+	u8 *start_ptr = &start_addr;
+	u8 **start_buff = &start_ptr;
+	u16 buff_len = 0;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
 
 	ibqp = nes_get_qp(cm_id->device, conn_param->qpn);
 	if (!ibqp)
@@ -2349,72 +3217,117 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	nesdev = nesvnic->nesdev;
 	adapter = nesdev->nesadapter;
 
-	nes_debug(NES_DBG_CM, "nesvnic=%p, netdev=%p, %s\n",
-			nesvnic, nesvnic->netdev, nesvnic->netdev->name);
-
-	/* since this is from a listen, we were able to put node handle into cm_id */
 	cm_node = (struct nes_cm_node *)cm_id->provider_data;
+	nes_debug(NES_DBG_CM, "nes_accept: cm_node= %p nesvnic=%p, netdev=%p,"
+		"%s\n", cm_node, nesvnic, nesvnic->netdev,
+		nesvnic->netdev->name);
+
+	if (NES_CM_STATE_LISTENER_DESTROYED == cm_node->state) {
+		if (cm_node->loopbackpartner)
+			rem_ref_cm_node(cm_node->cm_core, cm_node->loopbackpartner);
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		return -EINVAL;
+	}
 
+	passive_state = atomic_add_return(1, &cm_node->passive_state);
+	if (passive_state == NES_SEND_RESET_EVENT) {
+		rem_ref_cm_node(cm_node->cm_core, cm_node);
+		return -ECONNRESET;
+	}
 	/* associate the node with the QP */
 	nesqp->cm_node = (void *)cm_node;
+	cm_node->nesqp = nesqp;
 
-	nes_debug(NES_DBG_CM, "QP%u, cm_node=%p, jiffies = %lu\n",
-			nesqp->hwqp.qp_id, cm_node, jiffies);
+
+	nes_debug(NES_DBG_CM, "QP%u, cm_node=%p, jiffies = %lu listener = %p\n",
+		nesqp->hwqp.qp_id, cm_node, jiffies, cm_node->listener);
 	atomic_inc(&cm_accepts);
 
 	nes_debug(NES_DBG_CM, "netdev refcnt = %u.\n",
-			atomic_read(&nesvnic->netdev->refcnt));
+			netdev_refcnt_read(nesvnic->netdev));
+
+	nesqp->ietf_frame_size = sizeof(struct ietf_mpa_v2);
+	/* allocate the ietf frame and space for private data */
+	nesqp->ietf_frame = pci_alloc_consistent(nesdev->pcidev,
+						 nesqp->ietf_frame_size + conn_param->private_data_len,
+						 &nesqp->ietf_frame_pbase);
+
+	if (!nesqp->ietf_frame) {
+		nes_debug(NES_DBG_CM, "Unable to allocate memory for private data\n");
+		return -ENOMEM;
+	}
+	mpa_v2_frame = (struct ietf_mpa_v2 *)nesqp->ietf_frame;
 
-		/* allocate the ietf frame and space for private data */
-		nesqp->ietf_frame = pci_alloc_consistent(nesdev->pcidev,
-				sizeof(struct ietf_mpa_frame) + conn_param->private_data_len,
-				&nesqp->ietf_frame_pbase);
+	if (cm_node->mpa_frame_rev == IETF_MPA_V1)
+		mpa_frame_offset = 4;
+
+	if (cm_node->mpa_frame_rev == IETF_MPA_V1 ||
+			cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) {
+		record_ird_ord(cm_node, (u16)conn_param->ird, (u16)conn_param->ord);
+	}
+
+	memcpy(mpa_v2_frame->priv_data, conn_param->private_data,
+	       conn_param->private_data_len);
+
+	cm_build_mpa_frame(cm_node, start_buff, &buff_len, nesqp->ietf_frame, MPA_KEY_REPLY);
+	nesqp->private_data_len = conn_param->private_data_len;
 
-		if (!nesqp->ietf_frame) {
-			nes_debug(NES_DBG_CM, "Unable to allocate memory for private data\n");
+	/* setup our first outgoing iWarp send WQE (the IETF frame response) */
+	wqe = &nesqp->hwqp.sq_vbase[0];
+
+	if (raddr->sin_addr.s_addr != laddr->sin_addr.s_addr) {
+		u64temp = (unsigned long)nesqp;
+		nesibdev = nesvnic->nesibdev;
+		nespd = nesqp->nespd;
+		ibphysbuf.addr = nesqp->ietf_frame_pbase + mpa_frame_offset;
+		ibphysbuf.size = buff_len;
+		tagged_offset = (u64)(unsigned long)*start_buff;
+		ibmr = nesibdev->ibdev.reg_phys_mr((struct ib_pd *)nespd,
+						   &ibphysbuf, 1,
+						   IB_ACCESS_LOCAL_WRITE,
+						   &tagged_offset);
+		if (!ibmr) {
+			nes_debug(NES_DBG_CM, "Unable to register memory region"
+				  "for lSMM for cm_node = %p \n",
+				  cm_node);
+			pci_free_consistent(nesdev->pcidev,
+					    nesqp->private_data_len + nesqp->ietf_frame_size,
+					    nesqp->ietf_frame, nesqp->ietf_frame_pbase);
 			return -ENOMEM;
 		}
 
-
-		/* setup the MPA frame */
-		nesqp->private_data_len = conn_param->private_data_len;
-		memcpy(nesqp->ietf_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
-
-		memcpy(nesqp->ietf_frame->priv_data, conn_param->private_data,
-				conn_param->private_data_len);
-
-		nesqp->ietf_frame->priv_data_len = cpu_to_be16(conn_param->private_data_len);
-		nesqp->ietf_frame->rev = mpa_version;
-		nesqp->ietf_frame->flags = IETF_MPA_FLAGS_CRC;
-
-		/* setup our first outgoing iWarp send WQE (the IETF frame response) */
-		wqe = &nesqp->hwqp.sq_vbase[0];
-
-		if (cm_id->remote_addr.sin_addr.s_addr != cm_id->local_addr.sin_addr.s_addr) {
-			u64temp = (unsigned long)nesqp;
-			u64temp |= NES_SW_CONTEXT_ALIGN>>1;
-			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX,
-					    u64temp);
-			wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
-					cpu_to_le32(NES_IWARP_SQ_WQE_STREAMING | NES_IWARP_SQ_WQE_WRPDU);
-			wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] =
-					cpu_to_le32(conn_param->private_data_len + sizeof(struct ietf_mpa_frame));
-			wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_LOW_IDX] =
-					cpu_to_le32((u32)nesqp->ietf_frame_pbase);
-			wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_HIGH_IDX] =
-					cpu_to_le32((u32)((u64)nesqp->ietf_frame_pbase >> 32));
-			wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] =
-					cpu_to_le32(conn_param->private_data_len + sizeof(struct ietf_mpa_frame));
-			wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 0;
-
-			nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32(
-					NES_QPCONTEXT_ORDIRD_LSMM_PRESENT | NES_QPCONTEXT_ORDIRD_WRPDU);
-		} else {
-			nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32((NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
-					NES_QPCONTEXT_ORDIRD_WRPDU | NES_QPCONTEXT_ORDIRD_ALSMM));
+		ibmr->pd = &nespd->ibpd;
+		ibmr->device = nespd->ibpd.device;
+		nesqp->lsmm_mr = ibmr;
+
+		u64temp |= NES_SW_CONTEXT_ALIGN >> 1;
+		set_wqe_64bit_value(wqe->wqe_words,
+				    NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX,
+				    u64temp);
+		wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] =
+			cpu_to_le32(NES_IWARP_SQ_WQE_STREAMING |
+				    NES_IWARP_SQ_WQE_WRPDU);
+		wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] =
+			cpu_to_le32(buff_len);
+		set_wqe_64bit_value(wqe->wqe_words,
+				    NES_IWARP_SQ_WQE_FRAG0_LOW_IDX,
+				    (u64)(unsigned long)(*start_buff));
+		wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] =
+			cpu_to_le32(buff_len);
+		wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = ibmr->lkey;
+		if (nesqp->sq_kmapped) {
+			nesqp->sq_kmapped = 0;
+			kunmap(nesqp->page);
 		}
-		nesqp->skip_lsmm = 1;
 
+		nesqp->nesqp_context->ird_ord_sizes |=
+			cpu_to_le32(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
+				    NES_QPCONTEXT_ORDIRD_WRPDU);
+	} else {
+		nesqp->nesqp_context->ird_ord_sizes |=
+			cpu_to_le32(NES_QPCONTEXT_ORDIRD_WRPDU);
+	}
+	nesqp->skip_lsmm = 1;
 
 	/* Cache the cm_id in the qp */
 	nesqp->cm_id = cm_id;
@@ -2422,82 +3335,95 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 
 	/*  nesqp->cm_node = (void *)cm_id->provider_data; */
 	cm_id->provider_data = nesqp;
-	nesqp->active_conn   = 0;
+	nesqp->active_conn = 0;
+
+	if (cm_node->state == NES_CM_STATE_TSA)
+		nes_debug(NES_DBG_CM, "Already state = TSA for cm_node=%p\n",
+			  cm_node);
 
 	nes_cm_init_tsa_conn(nesqp, cm_node);
 
-	nesqp->nesqp_context->tcpPorts[0] = cpu_to_le16(ntohs(cm_id->local_addr.sin_port));
-	nesqp->nesqp_context->tcpPorts[1] = cpu_to_le16(ntohs(cm_id->remote_addr.sin_port));
-	nesqp->nesqp_context->ip0 = cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr));
+	nesqp->nesqp_context->tcpPorts[0] =
+				cpu_to_le16(cm_node->mapped_loc_port);
+	nesqp->nesqp_context->tcpPorts[1] =
+				cpu_to_le16(cm_node->mapped_rem_port);
+
+	nesqp->nesqp_context->ip0 = cpu_to_le32(cm_node->mapped_rem_addr);
 
 	nesqp->nesqp_context->misc2 |= cpu_to_le32(
-			(u32)PCI_FUNC(nesdev->pcidev->devfn) << NES_QPCONTEXT_MISC2_SRC_IP_SHIFT);
+		(u32)PCI_FUNC(nesdev->pcidev->devfn) <<
+		NES_QPCONTEXT_MISC2_SRC_IP_SHIFT);
 
-	nesqp->nesqp_context->arp_index_vlan |= cpu_to_le32(
-			nes_arp_table(nesdev, le32_to_cpu(nesqp->nesqp_context->ip0), NULL,
-			NES_ARP_RESOLVE) << 16);
+	nesqp->nesqp_context->arp_index_vlan |=
+		cpu_to_le32(nes_arp_table(nesdev,
+					  le32_to_cpu(nesqp->nesqp_context->ip0), NULL,
+					  NES_ARP_RESOLVE) << 16);
 
 	nesqp->nesqp_context->ts_val_delta = cpu_to_le32(
-			jiffies - nes_read_indexed(nesdev, NES_IDX_TCP_NOW));
+		jiffies - nes_read_indexed(nesdev, NES_IDX_TCP_NOW));
 
 	nesqp->nesqp_context->ird_index = cpu_to_le32(nesqp->hwqp.qp_id);
 
 	nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32(
-			((u32)1 << NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT));
-	nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32((u32)conn_param->ord);
+		((u32)1 << NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT));
+	nesqp->nesqp_context->ird_ord_sizes |=
+		cpu_to_le32((u32)cm_node->ord_size);
 
 	memset(&nes_quad, 0, sizeof(nes_quad));
-	nes_quad.DstIpAdrIndex = cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
-	nes_quad.SrcIpadr      = cm_id->remote_addr.sin_addr.s_addr;
-	nes_quad.TcpPorts[0]   = cm_id->remote_addr.sin_port;
-	nes_quad.TcpPorts[1]   = cm_id->local_addr.sin_port;
+	nes_quad.DstIpAdrIndex =
+		cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
+	nes_quad.SrcIpadr = htonl(cm_node->mapped_rem_addr);
+	nes_quad.TcpPorts[0] = htons(cm_node->mapped_rem_port);
+	nes_quad.TcpPorts[1] = htons(cm_node->mapped_loc_port);
 
 	/* Produce hash key */
 	crc_value = get_crc_value(&nes_quad);
 	nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
 	nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, CRC = 0x%08X\n",
-			nesqp->hte_index, nesqp->hte_index & adapter->hte_index_mask);
+		  nesqp->hte_index, nesqp->hte_index & adapter->hte_index_mask);
 
 	nesqp->hte_index &= adapter->hte_index_mask;
 	nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
 
 	cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node);
 
-	nes_debug(NES_DBG_CM, "QP%u, Destination IP = 0x%08X:0x%04X, local = 0x%08X:0x%04X,"
-			" rcv_nxt=0x%08X, snd_nxt=0x%08X, mpa + private data length=%zu.\n",
-			nesqp->hwqp.qp_id,
-			ntohl(cm_id->remote_addr.sin_addr.s_addr),
-			ntohs(cm_id->remote_addr.sin_port),
-			ntohl(cm_id->local_addr.sin_addr.s_addr),
-			ntohs(cm_id->local_addr.sin_port),
-			le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
-			le32_to_cpu(nesqp->nesqp_context->snd_nxt),
-			conn_param->private_data_len+sizeof(struct ietf_mpa_frame));
-
-	attr.qp_state = IB_QPS_RTS;
-	nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL);
+	nes_debug(NES_DBG_CM, "QP%u, Destination IP = 0x%08X:0x%04X, local = "
+		  "0x%08X:0x%04X, rcv_nxt=0x%08X, snd_nxt=0x%08X, mpa + "
+		  "private data length=%u.\n", nesqp->hwqp.qp_id,
+		  ntohl(raddr->sin_addr.s_addr), ntohs(raddr->sin_port),
+		  ntohl(laddr->sin_addr.s_addr), ntohs(laddr->sin_port),
+		  le32_to_cpu(nesqp->nesqp_context->rcv_nxt),
+		  le32_to_cpu(nesqp->nesqp_context->snd_nxt),
+		  buff_len);
 
-	/* notify OF layer that accept event was successfull */
+	/* notify OF layer that accept event was successful */
 	cm_id->add_ref(cm_id);
+	nes_add_ref(&nesqp->ibqp);
 
 	cm_event.event = IW_CM_EVENT_ESTABLISHED;
-	cm_event.status = IW_CM_EVENT_STATUS_ACCEPTED;
+	cm_event.status = 0;
 	cm_event.provider_data = (void *)nesqp;
 	cm_event.local_addr = cm_id->local_addr;
 	cm_event.remote_addr = cm_id->remote_addr;
 	cm_event.private_data = NULL;
 	cm_event.private_data_len = 0;
+	cm_event.ird = cm_node->ird_size;
+	cm_event.ord = cm_node->ord_size;
+
 	ret = cm_id->event_handler(cm_id, &cm_event);
+	attr.qp_state = IB_QPS_RTS;
+	nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL);
 	if (cm_node->loopbackpartner) {
-		cm_node->loopbackpartner->mpa_frame_size = nesqp->private_data_len;
+		cm_node->loopbackpartner->mpa_frame_size =
+			nesqp->private_data_len;
 		/* copy entire MPA frame to our cm_node's frame */
-		memcpy(cm_node->loopbackpartner->mpa_frame_buf, nesqp->ietf_frame->priv_data,
-			   nesqp->private_data_len);
+		memcpy(cm_node->loopbackpartner->mpa_frame_buf,
+		       conn_param->private_data, conn_param->private_data_len);
 		create_event(cm_node->loopbackpartner, NES_CM_EVENT_CONNECTED);
 	}
 	if (ret)
-		printk("%s[%u] OFA CM event_handler returned, ret=%d\n",
-				__func__, __LINE__, ret);
+		printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
+		       "ret=%d\n", __func__, __LINE__, ret);
 
 	return 0;
 }
@@ -2509,23 +3435,29 @@ int nes_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 int nes_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
 {
 	struct nes_cm_node *cm_node;
+	struct nes_cm_node *loopback;
 	struct nes_cm_core *cm_core;
+	u8 *start_buff;
 
 	atomic_inc(&cm_rejects);
-	cm_node = (struct nes_cm_node *) cm_id->provider_data;
+	cm_node = (struct nes_cm_node *)cm_id->provider_data;
+	loopback = cm_node->loopbackpartner;
 	cm_core = cm_node->cm_core;
-	cm_node->mpa_frame_size = sizeof(struct ietf_mpa_frame) + pdata_len;
-
-	strcpy(&cm_node->mpa_frame.key[0], IEFT_MPA_KEY_REP);
-	memcpy(&cm_node->mpa_frame.priv_data, pdata, pdata_len);
-
-	cm_node->mpa_frame.priv_data_len = cpu_to_be16(pdata_len);
-	cm_node->mpa_frame.rev = mpa_version;
-	cm_node->mpa_frame.flags = IETF_MPA_FLAGS_CRC | IETF_MPA_FLAGS_REJECT;
+	cm_node->cm_id = cm_id;
 
-	cm_core->api->reject(cm_core, &cm_node->mpa_frame, cm_node);
+	if (pdata_len + sizeof(struct ietf_mpa_v2) > MAX_CM_BUFFER)
+		return -EINVAL;
 
-	return 0;
+	if (loopback) {
+		memcpy(&loopback->mpa_frame.priv_data, pdata, pdata_len);
+		loopback->mpa_frame.priv_data_len = pdata_len;
+		loopback->mpa_frame_size = pdata_len;
+	} else {
+		start_buff = &cm_node->mpa_frame_buf[0] + sizeof(struct ietf_mpa_v2);
+		cm_node->mpa_frame_size = pdata_len;
+		memcpy(start_buff, pdata, pdata_len);
+	}
+	return cm_core->api->reject(cm_core, cm_node);
 }
 
 
@@ -2541,7 +3473,15 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	struct nes_device *nesdev;
 	struct nes_cm_node *cm_node;
 	struct nes_cm_info cm_info;
-
+	int apbvt_set = 0;
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
+	struct sockaddr_in *raddr = (struct sockaddr_in *)&cm_id->remote_addr;
+	struct iwpm_dev_data pm_reg_msg;
+	struct iwpm_sa_data pm_msg;
+	int iwpm_err = 0;
+
+	if (cm_id->remote_addr.ss_family != AF_INET)
+		return -ENOSYS;
 	ibqp = nes_get_qp(cm_id->device, conn_param->qpn);
 	if (!ibqp)
 		return -EINVAL;
@@ -2551,78 +3491,99 @@ int nes_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 	nesvnic = to_nesvnic(nesqp->ibqp.device);
 	if (!nesvnic)
 		return -EINVAL;
-	nesdev  = nesvnic->nesdev;
+	nesdev = nesvnic->nesdev;
 	if (!nesdev)
 		return -EINVAL;
 
-	atomic_inc(&cm_connects);
+	if (!laddr->sin_port || !raddr->sin_port)
+		return -EINVAL;
 
-	nesqp->ietf_frame = kzalloc(sizeof(struct ietf_mpa_frame) +
-			conn_param->private_data_len, GFP_KERNEL);
-	if (!nesqp->ietf_frame)
-		return -ENOMEM;
+	nes_debug(NES_DBG_CM, "QP%u, current IP = 0x%08X, Destination IP = "
+		  "0x%08X:0x%04X, local = 0x%08X:0x%04X.\n", nesqp->hwqp.qp_id,
+		  ntohl(nesvnic->local_ipaddr), ntohl(raddr->sin_addr.s_addr),
+		  ntohs(raddr->sin_port), ntohl(laddr->sin_addr.s_addr),
+		  ntohs(laddr->sin_port));
 
-	/* set qp as having an active connection */
+	atomic_inc(&cm_connects);
 	nesqp->active_conn = 1;
 
-	nes_debug(NES_DBG_CM, "QP%u, Destination IP = 0x%08X:0x%04X, local = 0x%08X:0x%04X.\n",
-			nesqp->hwqp.qp_id,
-			ntohl(cm_id->remote_addr.sin_addr.s_addr),
-			ntohs(cm_id->remote_addr.sin_port),
-			ntohl(cm_id->local_addr.sin_addr.s_addr),
-			ntohs(cm_id->local_addr.sin_port));
-
 	/* cache the cm_id in the qp */
 	nesqp->cm_id = cm_id;
-
 	cm_id->provider_data = nesqp;
-
-	/* copy the private data */
-	if (conn_param->private_data_len) {
-		memcpy(nesqp->ietf_frame->priv_data, conn_param->private_data,
-				conn_param->private_data_len);
-	}
-
 	nesqp->private_data_len = conn_param->private_data_len;
-	nesqp->nesqp_context->ird_ord_sizes |= cpu_to_le32((u32)conn_param->ord);
-	nes_debug(NES_DBG_CM, "requested ord = 0x%08X.\n", (u32)conn_param->ord);
-	nes_debug(NES_DBG_CM, "mpa private data len =%u\n", conn_param->private_data_len);
-
-	strcpy(&nesqp->ietf_frame->key[0], IEFT_MPA_KEY_REQ);
-	nesqp->ietf_frame->flags = IETF_MPA_FLAGS_CRC;
-	nesqp->ietf_frame->rev = IETF_MPA_VERSION;
-	nesqp->ietf_frame->priv_data_len = htons(conn_param->private_data_len);
 
-	if (cm_id->local_addr.sin_addr.s_addr != cm_id->remote_addr.sin_addr.s_addr)
-		nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port),
-				PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD);
+	nes_debug(NES_DBG_CM, "requested ord = 0x%08X.\n", (u32)conn_param->ord);
+	nes_debug(NES_DBG_CM, "mpa private data len =%u\n",
+		  conn_param->private_data_len);
 
 	/* set up the connection params for the node */
-	cm_info.loc_addr = (cm_id->local_addr.sin_addr.s_addr);
-	cm_info.loc_port = (cm_id->local_addr.sin_port);
-	cm_info.rem_addr = (cm_id->remote_addr.sin_addr.s_addr);
-	cm_info.rem_port = (cm_id->remote_addr.sin_port);
+	cm_info.loc_addr = ntohl(laddr->sin_addr.s_addr);
+	cm_info.loc_port = ntohs(laddr->sin_port);
+	cm_info.rem_addr = ntohl(raddr->sin_addr.s_addr);
+	cm_info.rem_port = ntohs(raddr->sin_port);
 	cm_info.cm_id = cm_id;
 	cm_info.conn_type = NES_CM_IWARP_CONN_TYPE;
 
+	/* No port mapper available, go with the specified peer information */
+	cm_info.mapped_loc_addr = cm_info.loc_addr;
+	cm_info.mapped_loc_port = cm_info.loc_port;
+	cm_info.mapped_rem_addr = cm_info.rem_addr;
+	cm_info.mapped_rem_port = cm_info.rem_port;
+
+	nes_form_reg_msg(nesvnic, &pm_reg_msg);
+	iwpm_err = iwpm_register_pid(&pm_reg_msg, RDMA_NL_NES);
+	if (iwpm_err) {
+		nes_debug(NES_DBG_NLMSG,
+			"Port Mapper reg pid fail (err = %d).\n", iwpm_err);
+	}
+	if (iwpm_valid_pid() && !iwpm_err) {
+		nes_form_pm_msg(&cm_info, &pm_msg);
+		iwpm_err = iwpm_add_and_query_mapping(&pm_msg, RDMA_NL_NES);
+		if (iwpm_err)
+			nes_debug(NES_DBG_NLMSG,
+			"Port Mapper query fail (err = %d).\n", iwpm_err);
+		else
+			nes_record_pm_msg(&cm_info, &pm_msg);
+	}
+
+	if (laddr->sin_addr.s_addr != raddr->sin_addr.s_addr) {
+		nes_manage_apbvt(nesvnic, cm_info.mapped_loc_port,
+			PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD);
+		apbvt_set = 1;
+	}
+
+	if (nes_create_mapinfo(&cm_info))
+		return -ENOMEM;
+
 	cm_id->add_ref(cm_id);
-	nes_add_ref(&nesqp->ibqp);
 
 	/* create a connect CM node connection */
-	cm_node = g_cm_core->api->connect(g_cm_core, nesvnic, nesqp->ietf_frame, &cm_info);
+	cm_node = g_cm_core->api->connect(g_cm_core, nesvnic,
+					  conn_param->private_data_len, (void *)conn_param->private_data,
+					  &cm_info);
 	if (!cm_node) {
-		if (cm_id->local_addr.sin_addr.s_addr != cm_id->remote_addr.sin_addr.s_addr)
-			nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port),
-					PCI_FUNC(nesdev->pcidev->devfn), NES_MANAGE_APBVT_DEL);
-		nes_rem_ref(&nesqp->ibqp);
-		kfree(nesqp->ietf_frame);
-		nesqp->ietf_frame = NULL;
+		if (apbvt_set)
+			nes_manage_apbvt(nesvnic, cm_info.mapped_loc_port,
+					 PCI_FUNC(nesdev->pcidev->devfn),
+					 NES_MANAGE_APBVT_DEL);
+
+		nes_debug(NES_DBG_NLMSG, "Delete mapped_loc_port = %04X\n",
+				cm_info.mapped_loc_port);
+		nes_remove_mapinfo(cm_info.loc_addr, cm_info.loc_port,
+			cm_info.mapped_loc_addr, cm_info.mapped_loc_port);
 		cm_id->rem_ref(cm_id);
 		return -ENOMEM;
 	}
 
-	cm_node->apbvt_set = 1;
+	record_ird_ord(cm_node, (u16)conn_param->ird, (u16)conn_param->ord);
+	if (cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO &&
+				cm_node->ord_size == 0)
+		cm_node->ord_size = 1;
+
+	cm_node->apbvt_set = apbvt_set;
 	nesqp->cm_node = cm_node;
+	cm_node->nesqp = nesqp;
+	nes_add_ref(&nesqp->ibqp);
 
 	return 0;
 }
@@ -2636,50 +3597,59 @@ int nes_create_listen(struct iw_cm_id *cm_id, int backlog)
 	struct nes_vnic *nesvnic;
 	struct nes_cm_listener *cm_node;
 	struct nes_cm_info cm_info;
-	struct nes_adapter *adapter;
 	int err;
-
+	struct sockaddr_in *laddr = (struct sockaddr_in *)&cm_id->local_addr;
 
 	nes_debug(NES_DBG_CM, "cm_id = %p, local port = 0x%04X.\n",
-			cm_id, ntohs(cm_id->local_addr.sin_port));
+		  cm_id, ntohs(laddr->sin_port));
 
+	if (cm_id->local_addr.ss_family != AF_INET)
+		return -ENOSYS;
 	nesvnic = to_nesvnic(cm_id->device);
 	if (!nesvnic)
 		return -EINVAL;
-	adapter = nesvnic->nesdev->nesadapter;
+
 	nes_debug(NES_DBG_CM, "nesvnic=%p, netdev=%p, %s\n",
 			nesvnic, nesvnic->netdev, nesvnic->netdev->name);
 
 	nes_debug(NES_DBG_CM, "nesvnic->local_ipaddr=0x%08x, sin_addr.s_addr=0x%08x\n",
-			nesvnic->local_ipaddr, cm_id->local_addr.sin_addr.s_addr);
+			nesvnic->local_ipaddr, laddr->sin_addr.s_addr);
 
 	/* setup listen params in our api call struct */
-	cm_info.loc_addr = nesvnic->local_ipaddr;
-	cm_info.loc_port = cm_id->local_addr.sin_port;
+	cm_info.loc_addr = ntohl(nesvnic->local_ipaddr);
+	cm_info.loc_port = ntohs(laddr->sin_port);
 	cm_info.backlog = backlog;
 	cm_info.cm_id = cm_id;
 
 	cm_info.conn_type = NES_CM_IWARP_CONN_TYPE;
 
+	/* No port mapper available, go with the specified info */
+	cm_info.mapped_loc_addr = cm_info.loc_addr;
+	cm_info.mapped_loc_port = cm_info.loc_port;
 
 	cm_node = g_cm_core->api->listen(g_cm_core, nesvnic, &cm_info);
 	if (!cm_node) {
-		printk("%s[%u] Error returned from listen API call\n",
-				__func__, __LINE__);
+		printk(KERN_ERR "%s[%u] Error returned from listen API call\n",
+		       __func__, __LINE__);
 		return -ENOMEM;
 	}
 
 	cm_id->provider_data = cm_node;
 
 	if (!cm_node->reused_node) {
-		err = nes_manage_apbvt(nesvnic, ntohs(cm_id->local_addr.sin_port),
-				PCI_FUNC(nesvnic->nesdev->pcidev->devfn), NES_MANAGE_APBVT_ADD);
+		if (nes_create_mapinfo(&cm_info))
+			return -ENOMEM;
+
+		err = nes_manage_apbvt(nesvnic, cm_node->mapped_loc_port,
+				       PCI_FUNC(nesvnic->nesdev->pcidev->devfn),
+				       NES_MANAGE_APBVT_ADD);
 		if (err) {
-			printk("nes_manage_apbvt call returned %d.\n", err);
+			printk(KERN_ERR "nes_manage_apbvt call returned %d.\n",
+			       err);
 			g_cm_core->api->stop_listener(g_cm_core, (void *)cm_node);
 			return err;
 		}
-		cm_listens_created++;
+		atomic_inc(&cm_listens_created);
 	}
 
 	cm_id->add_ref(cm_id);
@@ -2711,15 +3681,16 @@ int nes_destroy_listen(struct iw_cm_id *cm_id)
  */
 int nes_cm_recv(struct sk_buff *skb, struct net_device *netdevice)
 {
+	int rc = 0;
+
 	cm_packets_received++;
-	if ((g_cm_core) && (g_cm_core->api)) {
-		g_cm_core->api->recv_pkt(g_cm_core, netdev_priv(netdevice), skb);
-	} else {
+	if ((g_cm_core) && (g_cm_core->api))
+		rc = g_cm_core->api->recv_pkt(g_cm_core, netdev_priv(netdevice), skb);
+	else
 		nes_debug(NES_DBG_CM, "Unable to process packet for CM,"
-				" cm is not setup properly.\n");
-	}
+			  " cm is not setup properly.\n");
 
-	return 0;
+	return rc;
 }
 
 
@@ -2732,11 +3703,10 @@ int nes_cm_start(void)
 	nes_debug(NES_DBG_CM, "\n");
 	/* create the primary CM core, pass this handle to subsequent core inits */
 	g_cm_core = nes_cm_alloc_core();
-	if (g_cm_core) {
+	if (g_cm_core)
 		return 0;
-	} else {
+	else
 		return -ENOMEM;
-	}
 }
 
 
@@ -2757,7 +3727,6 @@ int nes_cm_stop(void)
  */
 static void cm_event_connected(struct nes_cm_event *event)
 {
-	u64 u64temp;
 	struct nes_qp *nesqp;
 	struct nes_vnic *nesvnic;
 	struct nes_device *nesdev;
@@ -2766,10 +3735,12 @@ static void cm_event_connected(struct nes_cm_event *event)
 	struct ib_qp_attr attr;
 	struct iw_cm_id *cm_id;
 	struct iw_cm_event cm_event;
-	struct nes_hw_qp_wqe *wqe;
 	struct nes_v4_quad nes_quad;
 	u32 crc_value;
 	int ret;
+	struct sockaddr_in *laddr;
+	struct sockaddr_in *raddr;
+	struct sockaddr_in *cm_event_laddr;
 
 	/* get all our handles */
 	cm_node = event->cm_node;
@@ -2779,111 +3750,97 @@ static void cm_event_connected(struct nes_cm_event *event)
 	nesvnic = to_nesvnic(nesqp->ibqp.device);
 	nesdev = nesvnic->nesdev;
 	nesadapter = nesdev->nesadapter;
+	laddr = (struct sockaddr_in *)&cm_id->local_addr;
+	raddr = (struct sockaddr_in *)&cm_id->remote_addr;
+	cm_event_laddr = (struct sockaddr_in *)&cm_event.local_addr;
 
-	if (nesqp->destroyed) {
+	if (nesqp->destroyed)
 		return;
-	}
 	atomic_inc(&cm_connecteds);
 	nes_debug(NES_DBG_CM, "QP%u attempting to connect to  0x%08X:0x%04X on"
-			" local port 0x%04X. jiffies = %lu.\n",
-			nesqp->hwqp.qp_id,
-			ntohl(cm_id->remote_addr.sin_addr.s_addr),
-			ntohs(cm_id->remote_addr.sin_port),
-			ntohs(cm_id->local_addr.sin_port),
-			jiffies);
+		  " local port 0x%04X. jiffies = %lu.\n",
+		  nesqp->hwqp.qp_id, ntohl(raddr->sin_addr.s_addr),
+		  ntohs(raddr->sin_port), ntohs(laddr->sin_port), jiffies);
 
 	nes_cm_init_tsa_conn(nesqp, cm_node);
 
 	/* set the QP tsa context */
-	nesqp->nesqp_context->tcpPorts[0] = cpu_to_le16(ntohs(cm_id->local_addr.sin_port));
-	nesqp->nesqp_context->tcpPorts[1] = cpu_to_le16(ntohs(cm_id->remote_addr.sin_port));
-	nesqp->nesqp_context->ip0 = cpu_to_le32(ntohl(cm_id->remote_addr.sin_addr.s_addr));
+	nesqp->nesqp_context->tcpPorts[0] =
+			cpu_to_le16(cm_node->mapped_loc_port);
+	nesqp->nesqp_context->tcpPorts[1] =
+			cpu_to_le16(cm_node->mapped_rem_port);
+	nesqp->nesqp_context->ip0 = cpu_to_le32(cm_node->mapped_rem_addr);
 
 	nesqp->nesqp_context->misc2 |= cpu_to_le32(
-			(u32)PCI_FUNC(nesdev->pcidev->devfn) << NES_QPCONTEXT_MISC2_SRC_IP_SHIFT);
+			(u32)PCI_FUNC(nesdev->pcidev->devfn) <<
+			NES_QPCONTEXT_MISC2_SRC_IP_SHIFT);
 	nesqp->nesqp_context->arp_index_vlan |= cpu_to_le32(
-			nes_arp_table(nesdev, le32_to_cpu(nesqp->nesqp_context->ip0),
+			nes_arp_table(nesdev,
+			le32_to_cpu(nesqp->nesqp_context->ip0),
 			NULL, NES_ARP_RESOLVE) << 16);
 	nesqp->nesqp_context->ts_val_delta = cpu_to_le32(
 			jiffies - nes_read_indexed(nesdev, NES_IDX_TCP_NOW));
 	nesqp->nesqp_context->ird_index = cpu_to_le32(nesqp->hwqp.qp_id);
 	nesqp->nesqp_context->ird_ord_sizes |=
-			cpu_to_le32((u32)1 << NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT);
+			cpu_to_le32((u32)1 <<
+			NES_QPCONTEXT_ORDIRD_IWARP_MODE_SHIFT);
+	nesqp->nesqp_context->ird_ord_sizes |=
+			cpu_to_le32((u32)cm_node->ord_size);
 
 	/* Adjust tail for not having a LSMM */
-	nesqp->hwqp.sq_tail = 1;
-
-#if defined(NES_SEND_FIRST_WRITE)
-		if (cm_node->send_write0) {
-			nes_debug(NES_DBG_CM, "Sending first write.\n");
-			wqe = &nesqp->hwqp.sq_vbase[0];
-			u64temp = (unsigned long)nesqp;
-			u64temp |= NES_SW_CONTEXT_ALIGN>>1;
-			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_COMP_CTX_LOW_IDX,
-					    u64temp);
-			wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] = cpu_to_le32(NES_IWARP_SQ_OP_RDMAW);
-			wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX] = 0;
-			wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_LOW_IDX] = 0;
-			wqe->wqe_words[NES_IWARP_SQ_WQE_FRAG0_HIGH_IDX] = 0;
-			wqe->wqe_words[NES_IWARP_SQ_WQE_LENGTH0_IDX] = 0;
-			wqe->wqe_words[NES_IWARP_SQ_WQE_STAG0_IDX] = 0;
-
-			/* use the reserved spot on the WQ for the extra first WQE */
-			nesqp->nesqp_context->ird_ord_sizes &= cpu_to_le32(~(NES_QPCONTEXT_ORDIRD_LSMM_PRESENT |
-					NES_QPCONTEXT_ORDIRD_WRPDU | NES_QPCONTEXT_ORDIRD_ALSMM));
-			nesqp->skip_lsmm = 1;
-			nesqp->hwqp.sq_tail = 0;
-			nes_write32(nesdev->regs + NES_WQE_ALLOC,
-					(1 << 24) | 0x00800000 | nesqp->hwqp.qp_id);
-		}
-#endif
+	/*nesqp->hwqp.sq_tail = 1;*/
+
+	build_rdma0_msg(cm_node, &nesqp);
+
+	nes_write32(nesdev->regs + NES_WQE_ALLOC,
+		    (1 << 24) | 0x00800000 | nesqp->hwqp.qp_id);
 
 	memset(&nes_quad, 0, sizeof(nes_quad));
 
-	nes_quad.DstIpAdrIndex = cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
-	nes_quad.SrcIpadr = cm_id->remote_addr.sin_addr.s_addr;
-	nes_quad.TcpPorts[0] = cm_id->remote_addr.sin_port;
-	nes_quad.TcpPorts[1] = cm_id->local_addr.sin_port;
+	nes_quad.DstIpAdrIndex =
+		cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
+	nes_quad.SrcIpadr = htonl(cm_node->mapped_rem_addr);
+	nes_quad.TcpPorts[0] = htons(cm_node->mapped_rem_port);
+	nes_quad.TcpPorts[1] = htons(cm_node->mapped_loc_port);
 
 	/* Produce hash key */
 	crc_value = get_crc_value(&nes_quad);
 	nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
 	nes_debug(NES_DBG_CM, "HTE Index = 0x%08X, After CRC = 0x%08X\n",
-			nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask);
+		  nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask);
 
 	nesqp->hte_index &= nesadapter->hte_index_mask;
 	nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
 
 	nesqp->ietf_frame = &cm_node->mpa_frame;
-	nesqp->private_data_len = (u8) cm_node->mpa_frame_size;
+	nesqp->private_data_len = (u8)cm_node->mpa_frame_size;
 	cm_node->cm_core->api->accelerated(cm_node->cm_core, cm_node);
 
-	/* modify QP state to rts */
-	attr.qp_state = IB_QPS_RTS;
-	nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL);
-
 	/* notify OF layer we successfully created the requested connection */
 	cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
-	cm_event.status = IW_CM_EVENT_STATUS_ACCEPTED;
+	cm_event.status = 0;
 	cm_event.provider_data = cm_id->provider_data;
-	cm_event.local_addr.sin_family = AF_INET;
-	cm_event.local_addr.sin_port = cm_id->local_addr.sin_port;
+	cm_event_laddr->sin_family = AF_INET;
+	cm_event_laddr->sin_port = laddr->sin_port;
 	cm_event.remote_addr = cm_id->remote_addr;
 
-		cm_event.private_data = (void *)event->cm_node->mpa_frame_buf;
-		cm_event.private_data_len = (u8) event->cm_node->mpa_frame_size;
+	cm_event.private_data = (void *)event->cm_node->mpa_frame_buf;
+	cm_event.private_data_len = (u8)event->cm_node->mpa_frame_size;
+	cm_event.ird = cm_node->ird_size;
+	cm_event.ord = cm_node->ord_size;
 
-	cm_event.local_addr.sin_addr.s_addr = event->cm_info.rem_addr;
+	cm_event_laddr->sin_addr.s_addr = htonl(event->cm_info.rem_addr);
 	ret = cm_id->event_handler(cm_id, &cm_event);
 	nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
 
 	if (ret)
-		printk("%s[%u] OFA CM event_handler returned, ret=%d\n",
-				__func__, __LINE__, ret);
-	nes_debug(NES_DBG_CM, "Exiting connect thread for QP%u. jiffies = %lu\n",
-			nesqp->hwqp.qp_id, jiffies );
+		printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
+		       "ret=%d\n", __func__, __LINE__, ret);
+	attr.qp_state = IB_QPS_RTS;
+	nes_modify_qp(&nesqp->ibqp, &attr, IB_QP_STATE, NULL);
 
-	nes_rem_ref(&nesqp->ibqp);
+	nes_debug(NES_DBG_CM, "Exiting connect thread for QP%u. jiffies = "
+		  "%lu\n", nesqp->hwqp.qp_id, jiffies);
 
 	return;
 }
@@ -2904,40 +3861,46 @@ static void cm_event_connect_error(struct nes_cm_event *event)
 		return;
 
 	cm_id = event->cm_node->cm_id;
-	if (!cm_id) {
+	if (!cm_id)
 		return;
-	}
 
 	nes_debug(NES_DBG_CM, "cm_node=%p, cm_id=%p\n", event->cm_node, cm_id);
 	nesqp = cm_id->provider_data;
 
-	if (!nesqp) {
+	if (!nesqp)
 		return;
-	}
 
 	/* notify OF layer about this connection error event */
 	/* cm_id->rem_ref(cm_id); */
 	nesqp->cm_id = NULL;
 	cm_id->provider_data = NULL;
 	cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
-	cm_event.status = IW_CM_EVENT_STATUS_REJECTED;
+	cm_event.status = -ECONNRESET;
 	cm_event.provider_data = cm_id->provider_data;
 	cm_event.local_addr = cm_id->local_addr;
 	cm_event.remote_addr = cm_id->remote_addr;
 	cm_event.private_data = NULL;
 	cm_event.private_data_len = 0;
 
-	nes_debug(NES_DBG_CM, "call CM_EVENT REJECTED, local_addr=%08x, remove_addr=%08x\n",
-			cm_event.local_addr.sin_addr.s_addr, cm_event.remote_addr.sin_addr.s_addr);
+#ifdef CONFIG_INFINIBAND_NES_DEBUG
+	{
+		struct sockaddr_in *cm_event_laddr = (struct sockaddr_in *)
+						     &cm_event.local_addr;
+		struct sockaddr_in *cm_event_raddr = (struct sockaddr_in *)
+						     &cm_event.remote_addr;
+		nes_debug(NES_DBG_CM, "call CM_EVENT REJECTED, local_addr=%08x, remote_addr=%08x\n",
+			  cm_event_laddr->sin_addr.s_addr, cm_event_raddr->sin_addr.s_addr);
+	}
+#endif
 
 	ret = cm_id->event_handler(cm_id, &cm_event);
 	nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
 	if (ret)
-		printk("%s[%u] OFA CM event_handler returned, ret=%d\n",
-				__func__, __LINE__, ret);
-	nes_rem_ref(&nesqp->ibqp);
-		cm_id->rem_ref(cm_id);
+		printk(KERN_ERR "%s[%u] OFA CM event_handler returned, "
+		       "ret=%d\n", __func__, __LINE__, ret);
+	cm_id->rem_ref(cm_id);
 
+	rem_ref_cm_node(event->cm_node->cm_core, event->cm_node);
 	return;
 }
 
@@ -2963,18 +3926,32 @@ static void cm_event_reset(struct nes_cm_event *event)
 
 	nes_debug(NES_DBG_CM, "%p - cm_id = %p\n", event->cm_node, cm_id);
 	nesqp = cm_id->provider_data;
+	if (!nesqp)
+		return;
 
 	nesqp->cm_id = NULL;
 	/* cm_id->provider_data = NULL; */
 	cm_event.event = IW_CM_EVENT_DISCONNECT;
-	cm_event.status = IW_CM_EVENT_STATUS_RESET;
+	cm_event.status = -ECONNRESET;
 	cm_event.provider_data = cm_id->provider_data;
 	cm_event.local_addr = cm_id->local_addr;
 	cm_event.remote_addr = cm_id->remote_addr;
 	cm_event.private_data = NULL;
 	cm_event.private_data_len = 0;
 
+	cm_id->add_ref(cm_id);
 	ret = cm_id->event_handler(cm_id, &cm_event);
+	atomic_inc(&cm_closes);
+	cm_event.event = IW_CM_EVENT_CLOSE;
+	cm_event.status = 0;
+	cm_event.provider_data = cm_id->provider_data;
+	cm_event.local_addr = cm_id->local_addr;
+	cm_event.remote_addr = cm_id->remote_addr;
+	cm_event.private_data = NULL;
+	cm_event.private_data_len = 0;
+	nes_debug(NES_DBG_CM, "NODE %p Generating CLOSE\n", event->cm_node);
+	ret = cm_id->event_handler(cm_id, &cm_event);
+
 	nes_debug(NES_DBG_CM, "OFA CM event_handler returned, ret=%d\n", ret);
 
 
@@ -2990,10 +3967,14 @@ static void cm_event_reset(struct nes_cm_event *event)
  */
 static void cm_event_mpa_req(struct nes_cm_event *event)
 {
-	struct iw_cm_id   *cm_id;
+	struct iw_cm_id *cm_id;
 	struct iw_cm_event cm_event;
 	int ret;
 	struct nes_cm_node *cm_node;
+	struct sockaddr_in *cm_event_laddr = (struct sockaddr_in *)
+					     &cm_event.local_addr;
+	struct sockaddr_in *cm_event_raddr = (struct sockaddr_in *)
+					     &cm_event.remote_addr;
 
 	cm_node = event->cm_node;
 	if (!cm_node)
@@ -3002,27 +3983,81 @@ static void cm_event_mpa_req(struct nes_cm_event *event)
 
 	atomic_inc(&cm_connect_reqs);
 	nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n",
-			cm_node, cm_id, jiffies);
+		  cm_node, cm_id, jiffies);
 
 	cm_event.event = IW_CM_EVENT_CONNECT_REQUEST;
-	cm_event.status = IW_CM_EVENT_STATUS_OK;
+	cm_event.status = 0;
 	cm_event.provider_data = (void *)cm_node;
 
-	cm_event.local_addr.sin_family = AF_INET;
-	cm_event.local_addr.sin_port = htons(event->cm_info.loc_port);
-	cm_event.local_addr.sin_addr.s_addr = htonl(event->cm_info.loc_addr);
+	cm_event_laddr->sin_family = AF_INET;
+	cm_event_laddr->sin_port = htons(event->cm_info.loc_port);
+	cm_event_laddr->sin_addr.s_addr = htonl(event->cm_info.loc_addr);
+
+	cm_event_raddr->sin_family = AF_INET;
+	cm_event_raddr->sin_port = htons(event->cm_info.rem_port);
+	cm_event_raddr->sin_addr.s_addr = htonl(event->cm_info.rem_addr);
+	cm_event.private_data = cm_node->mpa_frame_buf;
+	cm_event.private_data_len = (u8)cm_node->mpa_frame_size;
+	if (cm_node->mpa_frame_rev == IETF_MPA_V1) {
+		cm_event.ird = NES_MAX_IRD;
+		cm_event.ord = NES_MAX_ORD;
+	} else {
+	cm_event.ird = cm_node->ird_size;
+	cm_event.ord = cm_node->ord_size;
+	}
+
+	ret = cm_id->event_handler(cm_id, &cm_event);
+	if (ret)
+		printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n",
+		       __func__, __LINE__, ret);
+	return;
+}
+
+
+static void cm_event_mpa_reject(struct nes_cm_event *event)
+{
+	struct iw_cm_id *cm_id;
+	struct iw_cm_event cm_event;
+	struct nes_cm_node *cm_node;
+	int ret;
+	struct sockaddr_in *cm_event_laddr = (struct sockaddr_in *)
+					     &cm_event.local_addr;
+	struct sockaddr_in *cm_event_raddr = (struct sockaddr_in *)
+					     &cm_event.remote_addr;
+
+	cm_node = event->cm_node;
+	if (!cm_node)
+		return;
+	cm_id = cm_node->cm_id;
+
+	atomic_inc(&cm_connect_reqs);
+	nes_debug(NES_DBG_CM, "cm_node = %p - cm_id = %p, jiffies = %lu\n",
+		  cm_node, cm_id, jiffies);
+
+	cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
+	cm_event.status = -ECONNREFUSED;
+	cm_event.provider_data = cm_id->provider_data;
+
+	cm_event_laddr->sin_family = AF_INET;
+	cm_event_laddr->sin_port = htons(event->cm_info.loc_port);
+	cm_event_laddr->sin_addr.s_addr = htonl(event->cm_info.loc_addr);
+
+	cm_event_raddr->sin_family = AF_INET;
+	cm_event_raddr->sin_port = htons(event->cm_info.rem_port);
+	cm_event_raddr->sin_addr.s_addr = htonl(event->cm_info.rem_addr);
 
-	cm_event.remote_addr.sin_family = AF_INET;
-	cm_event.remote_addr.sin_port = htons(event->cm_info.rem_port);
-	cm_event.remote_addr.sin_addr.s_addr = htonl(event->cm_info.rem_addr);
+	cm_event.private_data = cm_node->mpa_frame_buf;
+	cm_event.private_data_len = (u8)cm_node->mpa_frame_size;
 
-		cm_event.private_data                = cm_node->mpa_frame_buf;
-		cm_event.private_data_len            = (u8) cm_node->mpa_frame_size;
+	nes_debug(NES_DBG_CM, "call CM_EVENT_MPA_REJECTED, local_addr=%08x, "
+		  "remove_addr=%08x\n",
+		  cm_event_laddr->sin_addr.s_addr,
+		  cm_event_raddr->sin_addr.s_addr);
 
 	ret = cm_id->event_handler(cm_id, &cm_event);
 	if (ret)
-		printk("%s[%u] OFA CM event_handler returned, ret=%d\n",
-				__func__, __LINE__, ret);
+		printk(KERN_ERR "%s[%u] OFA CM event_handler returned, ret=%d\n",
+		       __func__, __LINE__, ret);
 
 	return;
 }
@@ -3040,7 +4075,8 @@ static int nes_cm_post_event(struct nes_cm_event *event)
 	add_ref_cm_node(event->cm_node);
 	event->cm_info.cm_id->add_ref(event->cm_info.cm_id);
 	INIT_WORK(&event->event_work, nes_cm_event_handler);
-	nes_debug(NES_DBG_CM, "queue_work, event=%p\n", event);
+	nes_debug(NES_DBG_CM, "cm_node=%p queue_work, event=%p\n",
+		  event->cm_node, event);
 
 	queue_work(event->cm_node->cm_core->event_wq, &event->event_work);
 
@@ -3056,46 +4092,56 @@ static int nes_cm_post_event(struct nes_cm_event *event)
  */
 static void nes_cm_event_handler(struct work_struct *work)
 {
-	struct nes_cm_event *event = container_of(work, struct nes_cm_event, event_work);
+	struct nes_cm_event *event = container_of(work, struct nes_cm_event,
+						  event_work);
 	struct nes_cm_core *cm_core;
 
-	if ((!event) || (!event->cm_node) || (!event->cm_node->cm_core)) {
+	if ((!event) || (!event->cm_node) || (!event->cm_node->cm_core))
 		return;
-	}
+
 	cm_core = event->cm_node->cm_core;
 	nes_debug(NES_DBG_CM, "event=%p, event->type=%u, events posted=%u\n",
-			event, event->type, atomic_read(&cm_core->events_posted));
+		  event, event->type, atomic_read(&cm_core->events_posted));
 
 	switch (event->type) {
-		case NES_CM_EVENT_MPA_REQ:
-			cm_event_mpa_req(event);
-			nes_debug(NES_DBG_CM, "CM Event: MPA REQUEST\n");
+	case NES_CM_EVENT_MPA_REQ:
+		cm_event_mpa_req(event);
+		nes_debug(NES_DBG_CM, "cm_node=%p CM Event: MPA REQUEST\n",
+			  event->cm_node);
+		break;
+	case NES_CM_EVENT_RESET:
+		nes_debug(NES_DBG_CM, "cm_node = %p CM Event: RESET\n",
+			  event->cm_node);
+		cm_event_reset(event);
+		break;
+	case NES_CM_EVENT_CONNECTED:
+		if ((!event->cm_node->cm_id) ||
+		    (event->cm_node->state != NES_CM_STATE_TSA))
 			break;
-		case NES_CM_EVENT_RESET:
-			nes_debug(NES_DBG_CM, "CM Event: RESET\n");
-			cm_event_reset(event);
-			break;
-		case NES_CM_EVENT_CONNECTED:
-			if ((!event->cm_node->cm_id) ||
-				(event->cm_node->state != NES_CM_STATE_TSA)) {
-				break;
-			}
-			cm_event_connected(event);
-			nes_debug(NES_DBG_CM, "CM Event: CONNECTED\n");
+		cm_event_connected(event);
+		nes_debug(NES_DBG_CM, "CM Event: CONNECTED\n");
+		break;
+	case NES_CM_EVENT_MPA_REJECT:
+		if ((!event->cm_node->cm_id) ||
+		    (event->cm_node->state == NES_CM_STATE_TSA))
 			break;
-		case NES_CM_EVENT_ABORTED:
-			if ((!event->cm_node->cm_id) || (event->cm_node->state == NES_CM_STATE_TSA)) {
-				break;
-			}
-			cm_event_connect_error(event);
-			nes_debug(NES_DBG_CM, "CM Event: ABORTED\n");
-			break;
-		case NES_CM_EVENT_DROPPED_PKT:
-			nes_debug(NES_DBG_CM, "CM Event: DROPPED PKT\n");
-			break;
-		default:
-			nes_debug(NES_DBG_CM, "CM Event: UNKNOWN EVENT TYPE\n");
+		cm_event_mpa_reject(event);
+		nes_debug(NES_DBG_CM, "CM Event: REJECT\n");
+		break;
+
+	case NES_CM_EVENT_ABORTED:
+		if ((!event->cm_node->cm_id) ||
+		    (event->cm_node->state == NES_CM_STATE_TSA))
 			break;
+		cm_event_connect_error(event);
+		nes_debug(NES_DBG_CM, "CM Event: ABORTED\n");
+		break;
+	case NES_CM_EVENT_DROPPED_PKT:
+		nes_debug(NES_DBG_CM, "CM Event: DROPPED PKT\n");
+		break;
+	default:
+		nes_debug(NES_DBG_CM, "CM Event: UNKNOWN EVENT TYPE\n");
+		break;
 	}
 
 	atomic_dec(&cm_core->events_posted);
diff --git a/drivers/infiniband/hw/nes/nes_cm.h b/drivers/infiniband/hw/nes/nes_cm.h
index 7717cb2ab50..f522cf63978 100644
--- a/drivers/infiniband/hw/nes/nes_cm.h
+++ b/drivers/infiniband/hw/nes/nes_cm.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2014 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -39,11 +39,27 @@
 #define NES_MANAGE_APBVT_DEL 0
 #define NES_MANAGE_APBVT_ADD 1
 
+#define NES_MPA_REQUEST_ACCEPT  1
+#define NES_MPA_REQUEST_REJECT  2
+
 /* IETF MPA -- defines, enums, structs */
 #define IEFT_MPA_KEY_REQ  "MPA ID Req Frame"
 #define IEFT_MPA_KEY_REP  "MPA ID Rep Frame"
 #define IETF_MPA_KEY_SIZE 16
 #define IETF_MPA_VERSION  1
+#define IETF_MAX_PRIV_DATA_LEN 512
+#define IETF_MPA_FRAME_SIZE    20
+#define IETF_RTR_MSG_SIZE      4
+#define IETF_MPA_V2_FLAG       0x10
+
+/* IETF RTR MSG Fields               */
+#define IETF_PEER_TO_PEER       0x8000
+#define IETF_FLPDU_ZERO_LEN     0x4000
+#define IETF_RDMA0_WRITE        0x8000
+#define IETF_RDMA0_READ         0x4000
+#define IETF_NO_IRD_ORD         0x3FFF
+#define NES_MAX_IRD		 0x40
+#define NES_MAX_ORD		 0x7F
 
 enum ietf_mpa_flags {
 	IETF_MPA_FLAGS_MARKERS = 0x80,	/* receive Markers */
@@ -51,7 +67,7 @@ enum ietf_mpa_flags {
 	IETF_MPA_FLAGS_REJECT  = 0x20,	/* Reject */
 };
 
-struct ietf_mpa_frame {
+struct ietf_mpa_v1 {
 	u8 key[IETF_MPA_KEY_SIZE];
 	u8 flags;
 	u8 rev;
@@ -61,6 +77,20 @@ struct ietf_mpa_frame {
 
 #define ietf_mpa_req_resp_frame ietf_mpa_frame
 
+struct ietf_rtr_msg {
+	__be16 ctrl_ird;
+	__be16 ctrl_ord;
+};
+
+struct ietf_mpa_v2 {
+	u8 key[IETF_MPA_KEY_SIZE];
+	u8 flags;
+	u8 rev;
+	 __be16 priv_data_len;
+	struct ietf_rtr_msg rtr_msg;
+	u8 priv_data[0];
+};
+
 struct nes_v4_quad {
 	u32 rsvd0;
 	__le32 DstIpAdrIndex;	/* Only most significant 5 bits are valid */
@@ -76,6 +106,10 @@ enum nes_timer_type {
 	NES_TIMER_TYPE_CLOSE,
 };
 
+#define NES_PASSIVE_STATE_INDICATED	0
+#define NES_DO_NOT_SEND_RESET_EVENT	1
+#define NES_SEND_RESET_EVENT		2
+
 #define MAX_NES_IFS 4
 
 #define SET_ACK 1
@@ -83,6 +117,8 @@ enum nes_timer_type {
 #define SET_FIN 4
 #define SET_RST 8
 
+#define TCP_OPTIONS_PADDING	3
+
 struct option_base {
 	u8 optionnum;
 	u8 length;
@@ -140,6 +176,7 @@ struct nes_timer_entry {
 #endif
 #define NES_SHORT_TIME      (10)
 #define NES_LONG_TIME       (2000*HZ/1000)
+#define NES_MAX_TIMEOUT     ((unsigned long) (12*HZ))
 
 #define NES_CM_HASHTABLE_SIZE         1024
 #define NES_CM_TCP_TIMER_INTERVAL     3000
@@ -159,6 +196,7 @@ struct nes_timer_entry {
 
 #define NES_CM_DEF_SEQ2      0x18ed5740
 #define NES_CM_DEF_LOCAL_ID2 0xb807
+#define	MAX_CM_BUFFER	(IETF_MPA_FRAME_SIZE + IETF_RTR_MSG_SIZE + IETF_MAX_PRIV_DATA_LEN)
 
 typedef u32 nes_addr_t;
 
@@ -177,6 +215,8 @@ enum nes_cm_node_state {
 	NES_CM_STATE_ESTABLISHED,
 	NES_CM_STATE_ACCEPTING,
 	NES_CM_STATE_MPAREQ_SENT,
+	NES_CM_STATE_MPAREQ_RCVD,
+	NES_CM_STATE_MPAREJ_RCVD,
 	NES_CM_STATE_TSA,
 	NES_CM_STATE_FIN_WAIT1,
 	NES_CM_STATE_FIN_WAIT2,
@@ -184,9 +224,35 @@ enum nes_cm_node_state {
 	NES_CM_STATE_TIME_WAIT,
 	NES_CM_STATE_LAST_ACK,
 	NES_CM_STATE_CLOSING,
+	NES_CM_STATE_LISTENER_DESTROYED,
 	NES_CM_STATE_CLOSED
 };
 
+enum mpa_frame_version {
+	IETF_MPA_V1 = 1,
+	IETF_MPA_V2 = 2
+};
+
+enum mpa_frame_key {
+	MPA_KEY_REQUEST,
+	MPA_KEY_REPLY
+};
+
+enum send_rdma0 {
+	SEND_RDMA_READ_ZERO = 1,
+	SEND_RDMA_WRITE_ZERO = 2
+};
+
+enum nes_tcpip_pkt_type {
+	NES_PKT_TYPE_UNKNOWN,
+	NES_PKT_TYPE_SYN,
+	NES_PKT_TYPE_SYNACK,
+	NES_PKT_TYPE_ACK,
+	NES_PKT_TYPE_FIN,
+	NES_PKT_TYPE_RST
+};
+
+
 /* type of nes connection */
 enum nes_cm_conn_type {
 	NES_CM_IWARP_CONN_TYPE,
@@ -218,17 +284,17 @@ struct nes_cm_tcp_context {
 
 
 enum nes_cm_listener_state {
-	NES_CM_LISTENER_PASSIVE_STATE=1,
-	NES_CM_LISTENER_ACTIVE_STATE=2,
-	NES_CM_LISTENER_EITHER_STATE=3
+	NES_CM_LISTENER_PASSIVE_STATE = 1,
+	NES_CM_LISTENER_ACTIVE_STATE = 2,
+	NES_CM_LISTENER_EITHER_STATE = 3
 };
 
 struct nes_cm_listener {
 	struct list_head           list;
 	struct nes_cm_core         *cm_core;
 	u8                         loc_mac[ETH_ALEN];
-	nes_addr_t                 loc_addr;
-	u16                        loc_port;
+	nes_addr_t                 loc_addr, mapped_loc_addr;
+	u16                        loc_port, mapped_loc_port;
 	struct iw_cm_id            *cm_id;
 	enum nes_cm_conn_type      conn_type;
 	atomic_t                   ref_count;
@@ -241,10 +307,10 @@ struct nes_cm_listener {
 
 /* per connection node and node state information */
 struct nes_cm_node {
-	u32                       hashkey;
-
 	nes_addr_t                loc_addr, rem_addr;
+	nes_addr_t                mapped_loc_addr, mapped_rem_addr;
 	u16                       loc_port, rem_port;
+	u16                       mapped_loc_port, mapped_rem_port;
 
 	u8                        loc_mac[ETH_ALEN];
 	u8                        rem_mac[ETH_ALEN];
@@ -257,16 +323,22 @@ struct nes_cm_node {
 	struct net_device         *netdev;
 
 	struct nes_cm_node        *loopbackpartner;
-	struct list_head          retrans_list;
+
+	struct nes_timer_entry	  *send_entry;
+	struct nes_timer_entry    *recv_entry;
 	spinlock_t                retrans_list_lock;
-	struct list_head          recv_list;
-	spinlock_t                recv_list_lock;
+	enum send_rdma0           send_rdma0_op;
 
-	int                       send_write0;
 	union {
-		struct ietf_mpa_frame mpa_frame;
-		u8                    mpa_frame_buf[NES_CM_DEFAULT_MTU];
+		struct ietf_mpa_v1 mpa_frame;
+		struct ietf_mpa_v2 mpa_v2_frame;
+		u8                 mpa_frame_buf[MAX_CM_BUFFER];
 	};
+	enum mpa_frame_version    mpa_frame_rev;
+	u16			  ird_size;
+	u16                       ord_size;
+	u16			  mpav2_ird_ord;
+
 	u16                       mpa_frame_size;
 	struct iw_cm_id           *cm_id;
 	struct list_head          list;
@@ -276,6 +348,10 @@ struct nes_cm_node {
 	struct nes_vnic           *nesvnic;
 	int                       apbvt_set;
 	int                       accept_pend;
+	struct list_head	timer_entry;
+	struct list_head	reset_entry;
+	struct nes_qp		*nesqp;
+	atomic_t 		passive_state;
 };
 
 /* structure for client or CM to fill when making CM api calls. */
@@ -290,6 +366,10 @@ struct nes_cm_info {
 	u16 rem_port;
 	nes_addr_t loc_addr;
 	nes_addr_t rem_addr;
+	u16 mapped_loc_port;
+	u16 mapped_rem_port;
+	nes_addr_t mapped_loc_addr;
+	nes_addr_t mapped_rem_addr;
 
 	enum nes_cm_conn_type  conn_type;
 	int backlog;
@@ -302,6 +382,7 @@ enum  nes_cm_event_type {
 	NES_CM_EVENT_MPA_REQ,
 	NES_CM_EVENT_MPA_CONNECT,
 	NES_CM_EVENT_MPA_ACCEPT,
+	NES_CM_EVENT_MPA_REJECT,
 	NES_CM_EVENT_MPA_ESTABLISHED,
 	NES_CM_EVENT_CONNECTED,
 	NES_CM_EVENT_CLOSED,
@@ -333,7 +414,6 @@ struct nes_cm_core {
 	u32                     mtu;
 	u32                     free_tx_pkt_max;
 	u32                     rx_pkt_posted;
-	struct sk_buff_head     tx_free_list;
 	atomic_t                ht_node_cnt;
 	struct list_head        connected_nodes;
 	/* struct list_head hashtable[NES_CM_HASHTABLE_SIZE]; */
@@ -366,13 +446,11 @@ struct nes_cm_ops {
 			struct nes_cm_info *);
 	int (*stop_listener)(struct nes_cm_core *, struct nes_cm_listener *);
 	struct nes_cm_node * (*connect)(struct nes_cm_core *,
-			struct nes_vnic *, struct ietf_mpa_frame *,
+			struct nes_vnic *, u16, void *,
 			struct nes_cm_info *);
 	int (*close)(struct nes_cm_core *, struct nes_cm_node *);
-	int (*accept)(struct nes_cm_core *, struct ietf_mpa_frame *,
-			struct nes_cm_node *);
-	int (*reject)(struct nes_cm_core *, struct ietf_mpa_frame *,
-			struct nes_cm_node *);
+	int (*accept)(struct nes_cm_core *, struct nes_cm_node *);
+	int (*reject)(struct nes_cm_core *, struct nes_cm_node *);
 	int (*recv_pkt)(struct nes_cm_core *, struct nes_vnic *,
 			struct sk_buff *);
 	int (*destroy_cm_core)(struct nes_cm_core *);
@@ -383,8 +461,6 @@ struct nes_cm_ops {
 int schedule_nes_timer(struct nes_cm_node *, struct sk_buff *,
 		enum nes_timer_type, int, int);
 
-int nes_cm_disconn(struct nes_qp *);
-
 int nes_accept(struct iw_cm_id *, struct iw_cm_conn_param *);
 int nes_reject(struct iw_cm_id *, const void *, u8);
 int nes_connect(struct iw_cm_id *, struct iw_cm_conn_param *);
@@ -394,5 +470,7 @@ int nes_destroy_listen(struct iw_cm_id *);
 int nes_cm_recv(struct sk_buff *, struct net_device *);
 int nes_cm_start(void);
 int nes_cm_stop(void);
+int nes_add_ref_cm_node(struct nes_cm_node *cm_node);
+int nes_rem_ref_cm_node(struct nes_cm_node *cm_node);
 
 #endif			/* NES_CM_H */
diff --git a/drivers/infiniband/hw/nes/nes_context.h b/drivers/infiniband/hw/nes/nes_context.h
index da9daba8e66..a69eef16d72 100644
--- a/drivers/infiniband/hw/nes/nes_context.h
+++ b/drivers/infiniband/hw/nes/nes_context.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
diff --git a/drivers/infiniband/hw/nes/nes_hw.c b/drivers/infiniband/hw/nes/nes_hw.c
index 85f26d19a32..90200245c5e 100644
--- a/drivers/infiniband/hw/nes/nes_hw.c
+++ b/drivers/infiniband/hw/nes/nes_hw.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -39,6 +39,7 @@
 #include <linux/tcp.h>
 #include <linux/if_vlan.h>
 #include <linux/inet_lro.h>
+#include <linux/slab.h>
 
 #include "nes.h"
 
@@ -46,6 +47,10 @@ static unsigned int nes_lro_max_aggr = NES_LRO_MAX_AGGR;
 module_param(nes_lro_max_aggr, uint, 0444);
 MODULE_PARM_DESC(nes_lro_max_aggr, "NIC LRO max packet aggregation");
 
+static int wide_ppm_offset;
+module_param(wide_ppm_offset, int, 0644);
+MODULE_PARM_DESC(wide_ppm_offset, "Increase CX4 interface clock ppm offset, 0=100ppm (default), 1=300ppm");
+
 static u32 crit_err_count;
 u32 int_mod_timer_init;
 u32 int_mod_cq_depth_256;
@@ -55,24 +60,26 @@ u32 int_mod_cq_depth_24;
 u32 int_mod_cq_depth_16;
 u32 int_mod_cq_depth_4;
 u32 int_mod_cq_depth_1;
-
+static const u8 nes_max_critical_error_count = 100;
 #include "nes_cm.h"
 
 static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq);
 static void nes_init_csr_ne020(struct nes_device *nesdev, u8 hw_rev, u8 port_count);
 static int nes_init_serdes(struct nes_device *nesdev, u8 hw_rev, u8 port_count,
-			   u8 OneG_Mode);
+				struct nes_adapter *nesadapter, u8  OneG_Mode);
 static void nes_nic_napi_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq);
 static void nes_process_aeq(struct nes_device *nesdev, struct nes_hw_aeq *aeq);
 static void nes_process_ceq(struct nes_device *nesdev, struct nes_hw_ceq *ceq);
 static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				   struct nes_hw_aeqe *aeqe);
+static void process_critical_error(struct nes_device *nesdev);
 static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number);
 static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_Mode);
+static void nes_terminate_start_timer(struct nes_qp *nesqp);
 
 #ifdef CONFIG_INFINIBAND_NES_DEBUG
 static unsigned char *nes_iwarp_state_str[] = {
-	"Non-Existant",
+	"Non-Existent",
 	"Idle",
 	"RTS",
 	"Closing",
@@ -83,7 +90,7 @@ static unsigned char *nes_iwarp_state_str[] = {
 };
 
 static unsigned char *nes_tcp_state_str[] = {
-	"Non-Existant",
+	"Non-Existent",
 	"Closed",
 	"Listen",
 	"SYN Sent",
@@ -102,6 +109,14 @@ static unsigned char *nes_tcp_state_str[] = {
 };
 #endif
 
+static inline void print_ip(struct nes_cm_node *cm_node)
+{
+	unsigned char *rem_addr;
+	if (cm_node) {
+		rem_addr = (unsigned char *)&cm_node->rem_addr;
+		printk(KERN_ERR PFX "Remote IP addr: %pI4\n", rem_addr);
+	}
+}
 
 /**
  * nes_nic_init_timer_defaults
@@ -222,11 +237,10 @@ static void nes_nic_tune_timer(struct nes_device *nesdev)
 	}
 
 	/* boundary checking */
-	if (shared_timer->timer_in_use > NES_NIC_FAST_TIMER_HIGH)
-		shared_timer->timer_in_use = NES_NIC_FAST_TIMER_HIGH;
-	else if (shared_timer->timer_in_use < NES_NIC_FAST_TIMER_LOW) {
-		shared_timer->timer_in_use = NES_NIC_FAST_TIMER_LOW;
-	}
+	if (shared_timer->timer_in_use > shared_timer->threshold_high)
+		shared_timer->timer_in_use = shared_timer->threshold_high;
+	else if (shared_timer->timer_in_use < shared_timer->threshold_low)
+		shared_timer->timer_in_use = shared_timer->threshold_low;
 
 	nesdev->currcq_count = 0;
 
@@ -254,6 +268,7 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 	u32 adapter_size;
 	u32 arp_table_size;
 	u16 vendor_id;
+	u16 device_id;
 	u8  OneG_Mode;
 	u8  func_index;
 
@@ -292,9 +307,6 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 
 	if ((port_count = nes_reset_adapter_ne020(nesdev, &OneG_Mode)) == 0)
 		return NULL;
-	if (nes_init_serdes(nesdev, hw_rev, port_count, OneG_Mode))
-		return NULL;
-	nes_init_csr_ne020(nesdev, hw_rev, port_count);
 
 	max_qp = nes_read_indexed(nesdev, NES_IDX_QP_CTX_SIZE);
 	nes_debug(NES_DBG_INIT, "QP_CTX_SIZE=%u\n", max_qp);
@@ -353,6 +365,29 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 	nes_debug(NES_DBG_INIT, "Allocating new nesadapter @ %p, size = %u (actual size = %u).\n",
 			nesadapter, (u32)sizeof(struct nes_adapter), adapter_size);
 
+	if (nes_read_eeprom_values(nesdev, nesadapter)) {
+		printk(KERN_ERR PFX "Unable to read EEPROM data.\n");
+		kfree(nesadapter);
+		return NULL;
+	}
+
+	nesadapter->vendor_id = (((u32) nesadapter->mac_addr_high) << 8) |
+				(nesadapter->mac_addr_low >> 24);
+
+	pci_bus_read_config_word(nesdev->pcidev->bus, nesdev->pcidev->devfn,
+				 PCI_DEVICE_ID, &device_id);
+	nesadapter->vendor_part_id = device_id;
+
+	if (nes_init_serdes(nesdev, hw_rev, port_count, nesadapter,
+							OneG_Mode)) {
+		kfree(nesadapter);
+		return NULL;
+	}
+	nes_init_csr_ne020(nesdev, hw_rev, port_count);
+
+	memset(nesadapter->pft_mcast_map, 255,
+	       sizeof nesadapter->pft_mcast_map);
+
 	/* populate the new nesadapter */
 	nesadapter->devfn = nesdev->pcidev->devfn;
 	nesadapter->bus_number = nesdev->pcidev->bus->number;
@@ -397,8 +432,9 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 
 	nesadapter->base_pd = 1;
 
-	nesadapter->device_cap_flags =
-		IB_DEVICE_LOCAL_DMA_LKEY | IB_DEVICE_MEM_WINDOW;
+	nesadapter->device_cap_flags = IB_DEVICE_LOCAL_DMA_LKEY |
+				       IB_DEVICE_MEM_WINDOW |
+				       IB_DEVICE_MEM_MGT_EXTENSIONS;
 
 	nesadapter->allocated_qps = (unsigned long *)&(((unsigned char *)nesadapter)
 			[(sizeof(struct nes_adapter)+(sizeof(unsigned long)-1))&(~(sizeof(unsigned long)-1))]);
@@ -409,11 +445,12 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 	nesadapter->qp_table = (struct nes_qp **)(&nesadapter->allocated_arps[BITS_TO_LONGS(arp_table_size)]);
 
 
-	/* mark the usual suspect QPs and CQs as in use */
+	/* mark the usual suspect QPs, MR and CQs as in use */
 	for (u32temp = 0; u32temp < NES_FIRST_QPN; u32temp++) {
 		set_bit(u32temp, nesadapter->allocated_qps);
 		set_bit(u32temp, nesadapter->allocated_cqs);
 	}
+	set_bit(0, nesadapter->allocated_mrs);
 
 	for (u32temp = 0; u32temp < 20; u32temp++)
 		set_bit(u32temp, nesadapter->allocated_pds);
@@ -454,7 +491,7 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 	nesadapter->max_irrq_wr = (u32temp >> 16) & 3;
 
 	nesadapter->max_sge = 4;
-	nesadapter->max_cqe = 32767;
+	nesadapter->max_cqe = 32766;
 
 	if (nes_read_eeprom_values(nesdev, nesadapter)) {
 		printk(KERN_ERR PFX "Unable to read EEPROM data.\n");
@@ -468,20 +505,25 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 
 	/* setup port configuration */
 	if (nesadapter->port_count == 1) {
-		u32temp = 0x00000000;
+		nesadapter->log_port = 0x00000000;
 		if (nes_drv_opt & NES_DRV_OPT_DUAL_LOGICAL_PORT)
 			nes_write_indexed(nesdev, NES_IDX_TX_POOL_SIZE, 0x00000002);
 		else
 			nes_write_indexed(nesdev, NES_IDX_TX_POOL_SIZE, 0x00000003);
 	} else {
-		if (nesadapter->port_count == 2)
-			u32temp = 0x00000044;
-		else
-			u32temp = 0x000000e4;
+		if (nesadapter->phy_type[0] == NES_PHY_TYPE_PUMA_1G) {
+			nesadapter->log_port = 0x000000D8;
+		} else {
+			if (nesadapter->port_count == 2)
+				nesadapter->log_port = 0x00000044;
+			else
+				nesadapter->log_port = 0x000000e4;
+		}
 		nes_write_indexed(nesdev, NES_IDX_TX_POOL_SIZE, 0x00000003);
 	}
 
-	nes_write_indexed(nesdev, NES_IDX_NIC_LOGPORT_TO_PHYPORT, u32temp);
+	nes_write_indexed(nesdev, NES_IDX_NIC_LOGPORT_TO_PHYPORT,
+						nesadapter->log_port);
 	nes_debug(NES_DBG_INIT, "Probe time, LOG2PHY=%u\n",
 			nes_read_indexed(nesdev, NES_IDX_NIC_LOGPORT_TO_PHYPORT));
 
@@ -521,7 +563,7 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 		}
 		if (int_cnt > 1) {
 			spin_lock_irqsave(&nesadapter->phy_lock, flags);
-			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1, 0x0000F088);
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1, 0x0000F0C8);
 			mh_detected++;
 			reset_value = nes_read32(nesdev->regs+NES_SOFTWARE_RESET);
 			reset_value |= 0x0000003d;
@@ -546,7 +588,7 @@ struct nes_adapter *nes_init_adapter(struct nes_device *nesdev, u8 hw_rev) {
 					if (++ext_cnt > int_cnt) {
 						spin_lock_irqsave(&nesadapter->phy_lock, flags);
 						nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1,
-								0x0000F0C8);
+								0x0000F088);
 						mh_detected++;
 						reset_value = nes_read32(nesdev->regs+NES_SOFTWARE_RESET);
 						reset_value |= 0x0000003d;
@@ -637,7 +679,7 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
 		i = 0;
 		while (((nes_read32(nesdev->regs+NES_SOFTWARE_RESET) & 0x00000040) == 0) && i++ < 10000)
 			mdelay(1);
-		if (i >= 10000) {
+		if (i > 10000) {
 			nes_debug(NES_DBG_INIT, "Did not see full soft reset done.\n");
 			return 0;
 		}
@@ -645,7 +687,7 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
 		i = 0;
 		while ((nes_read_indexed(nesdev, NES_IDX_INT_CPU_STATUS) != 0x80) && i++ < 10000)
 			mdelay(1);
-		if (i >= 10000) {
+		if (i > 10000) {
 			printk(KERN_ERR PFX "Internal CPU not ready, status = %02X\n",
 			       nes_read_indexed(nesdev, NES_IDX_INT_CPU_STATUS));
 			return 0;
@@ -671,7 +713,7 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
 	i = 0;
 	while (((nes_read32(nesdev->regs+NES_SOFTWARE_RESET) & 0x00000040) == 0) && i++ < 10000)
 		mdelay(1);
-	if (i >= 10000) {
+	if (i > 10000) {
 		nes_debug(NES_DBG_INIT, "Did not see port soft reset done.\n");
 		return 0;
 	}
@@ -681,7 +723,7 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
 	while (((u32temp = (nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_STATUS0)
 			& 0x0000000f)) != 0x0000000f) && i++ < 5000)
 		mdelay(1);
-	if (i >= 5000) {
+	if (i > 5000) {
 		nes_debug(NES_DBG_INIT, "Serdes 0 not ready, status=%x\n", u32temp);
 		return 0;
 	}
@@ -692,7 +734,7 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
 		while (((u32temp = (nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_STATUS1)
 				& 0x0000000f)) != 0x0000000f) && i++ < 5000)
 			mdelay(1);
-		if (i >= 5000) {
+		if (i > 5000) {
 			nes_debug(NES_DBG_INIT, "Serdes 1 not ready, status=%x\n", u32temp);
 			return 0;
 		}
@@ -706,23 +748,70 @@ static unsigned int nes_reset_adapter_ne020(struct nes_device *nesdev, u8 *OneG_
  * nes_init_serdes
  */
 static int nes_init_serdes(struct nes_device *nesdev, u8 hw_rev, u8 port_count,
-			   u8 OneG_Mode)
+				struct nes_adapter *nesadapter, u8  OneG_Mode)
 {
 	int i;
 	u32 u32temp;
+	u32 sds;
 
 	if (hw_rev != NE020_REV) {
 		/* init serdes 0 */
+		switch (nesadapter->phy_type[0]) {
+		case NES_PHY_TYPE_CX4:
+			if (wide_ppm_offset)
+				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000FFFAA);
+			else
+				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000000FF);
+			break;
+		case NES_PHY_TYPE_KR:
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000000FF);
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_EMP0, 0x00000000);
+			break;
+		case NES_PHY_TYPE_PUMA_1G:
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000000FF);
+			sds = nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0);
+			sds |= 0x00000100;
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0, sds);
+			break;
+		default:
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000000FF);
+			break;
+		}
 
-		nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL0, 0x000000FF);
 		if (!OneG_Mode)
 			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_HIGHZ_LANE_MODE0, 0x11110000);
-		if (port_count > 1) {
-			/* init serdes 1 */
+
+		if (port_count < 2)
+			return 0;
+
+		/* init serdes 1 */
+		if (!(OneG_Mode && (nesadapter->phy_type[1] != NES_PHY_TYPE_PUMA_1G)))
 			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL1, 0x000000FF);
-			if (!OneG_Mode)
-				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_HIGHZ_LANE_MODE1, 0x11110000);
-			}
+
+		switch (nesadapter->phy_type[1]) {
+		case NES_PHY_TYPE_ARGUS:
+		case NES_PHY_TYPE_SFP_D:
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_EMP0, 0x00000000);
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_EMP1, 0x00000000);
+			break;
+		case NES_PHY_TYPE_CX4:
+			if (wide_ppm_offset)
+				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_CDR_CONTROL1, 0x000FFFAA);
+			break;
+		case NES_PHY_TYPE_KR:
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_EMP1, 0x00000000);
+			break;
+		case NES_PHY_TYPE_PUMA_1G:
+			sds = nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1);
+			sds |= 0x000000100;
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1, sds);
+		}
+		if (!OneG_Mode) {
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_TX_HIGHZ_LANE_MODE1, 0x11110000);
+			sds = nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1);
+			sds &= 0xFFFFFFBF;
+			nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL1, sds);
+		}
 	} else {
 		/* init serdes 0 */
 		nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0, 0x00000008);
@@ -730,7 +819,7 @@ static int nes_init_serdes(struct nes_device *nesdev, u8 hw_rev, u8 port_count,
 		while (((u32temp = (nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_STATUS0)
 				& 0x0000000f)) != 0x0000000f) && i++ < 5000)
 			mdelay(1);
-		if (i >= 5000) {
+		if (i > 5000) {
 			nes_debug(NES_DBG_PHY, "Init: serdes 0 not ready, status=%x\n", u32temp);
 			return 1;
 		}
@@ -753,7 +842,7 @@ static int nes_init_serdes(struct nes_device *nesdev, u8 hw_rev, u8 port_count,
 			while (((u32temp = (nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_STATUS1)
 				& 0x0000000f)) != 0x0000000f) && (i++ < 5000))
 				mdelay(1);
-			if (i >= 5000) {
+			if (i > 5000) {
 				printk("%s: Init: serdes 1 not ready, status=%x\n", __func__, u32temp);
 				/* return 1; */
 			}
@@ -826,7 +915,8 @@ static void nes_init_csr_ne020(struct nes_device *nesdev, u8 hw_rev, u8 port_cou
 
 	nes_write_indexed(nesdev, 0x00005000, 0x00018000);
 	/* nes_write_indexed(nesdev, 0x00005000, 0x00010000); */
-	nes_write_indexed(nesdev, 0x00005004, 0x00020001);
+	nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG1, (wqm_quanta << 1) |
+							 0x00000001);
 	nes_write_indexed(nesdev, 0x00005008, 0x1F1F1F1F);
 	nes_write_indexed(nesdev, 0x00005010, 0x1F1F1F1F);
 	nes_write_indexed(nesdev, 0x00005018, 0x1F1F1F1F);
@@ -849,6 +939,12 @@ static void nes_init_csr_ne020(struct nes_device *nesdev, u8 hw_rev, u8 port_cou
 		u32temp &= 0x7fffffff;
 		u32temp |= 0x7fff0010;
 		nes_write_indexed(nesdev, 0x000021f8, u32temp);
+		if (port_count > 1) {
+			u32temp = nes_read_indexed(nesdev, 0x000023f8);
+			u32temp &= 0x7fffffff;
+			u32temp |= 0x7fff0010;
+			nes_write_indexed(nesdev, 0x000023f8, u32temp);
+		}
 	}
 }
 
@@ -1206,209 +1302,252 @@ int nes_destroy_cqp(struct nes_device *nesdev)
 
 
 /**
- * nes_init_phy
+ * nes_init_1g_phy
  */
-int nes_init_phy(struct nes_device *nesdev)
+static int nes_init_1g_phy(struct nes_device *nesdev, u8 phy_type, u8 phy_index)
 {
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 counter = 0;
-	u32 sds_common_control0;
-	u32 mac_index = nesdev->mac_index;
-	u32 tx_config = 0;
 	u16 phy_data;
+	int ret = 0;
+
+	nes_read_1G_phy_reg(nesdev, 1, phy_index, &phy_data);
+	nes_write_1G_phy_reg(nesdev, 23, phy_index, 0xb000);
+
+	/* Reset the PHY */
+	nes_write_1G_phy_reg(nesdev, 0, phy_index, 0x8000);
+	udelay(100);
+	counter = 0;
+	do {
+		nes_read_1G_phy_reg(nesdev, 0, phy_index, &phy_data);
+		if (counter++ > 100) {
+			ret = -1;
+			break;
+		}
+	} while (phy_data & 0x8000);
+
+	/* Setting no phy loopback */
+	phy_data &= 0xbfff;
+	phy_data |= 0x1140;
+	nes_write_1G_phy_reg(nesdev, 0, phy_index,  phy_data);
+	nes_read_1G_phy_reg(nesdev, 0, phy_index, &phy_data);
+	nes_read_1G_phy_reg(nesdev, 0x17, phy_index, &phy_data);
+	nes_read_1G_phy_reg(nesdev, 0x1e, phy_index, &phy_data);
+
+	/* Setting the interrupt mask */
+	nes_read_1G_phy_reg(nesdev, 0x19, phy_index, &phy_data);
+	nes_write_1G_phy_reg(nesdev, 0x19, phy_index, 0xffee);
+	nes_read_1G_phy_reg(nesdev, 0x19, phy_index, &phy_data);
+
+	/* turning on flow control */
+	nes_read_1G_phy_reg(nesdev, 4, phy_index, &phy_data);
+	nes_write_1G_phy_reg(nesdev, 4, phy_index, (phy_data & ~(0x03E0)) | 0xc00);
+	nes_read_1G_phy_reg(nesdev, 4, phy_index, &phy_data);
+
+	/* Clear Half duplex */
+	nes_read_1G_phy_reg(nesdev, 9, phy_index, &phy_data);
+	nes_write_1G_phy_reg(nesdev, 9, phy_index, phy_data & ~(0x0100));
+	nes_read_1G_phy_reg(nesdev, 9, phy_index, &phy_data);
+
+	nes_read_1G_phy_reg(nesdev, 0, phy_index, &phy_data);
+	nes_write_1G_phy_reg(nesdev, 0, phy_index, phy_data | 0x0300);
+
+	return ret;
+}
+
+
+/**
+ * nes_init_2025_phy
+ */
+static int nes_init_2025_phy(struct nes_device *nesdev, u8 phy_type, u8 phy_index)
+{
 	u32 temp_phy_data = 0;
 	u32 temp_phy_data2 = 0;
-	u32 i = 0;
-
-	if ((nesadapter->OneG_Mode) &&
-	    (nesadapter->phy_type[mac_index] != NES_PHY_TYPE_PUMA_1G)) {
-		nes_debug(NES_DBG_PHY, "1G PHY, mac_index = %d.\n", mac_index);
-		if (nesadapter->phy_type[mac_index] == NES_PHY_TYPE_1G) {
-			printk(PFX "%s: Programming mdc config for 1G\n", __func__);
-			tx_config = nes_read_indexed(nesdev, NES_IDX_MAC_TX_CONFIG);
-			tx_config |= 0x04;
-			nes_write_indexed(nesdev, NES_IDX_MAC_TX_CONFIG, tx_config);
+	u32 counter = 0;
+	u32 sds;
+	u32 mac_index = nesdev->mac_index;
+	int ret = 0;
+	unsigned int first_attempt = 1;
+
+	/* Check firmware heartbeat */
+	nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7ee);
+	temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+	udelay(1500);
+	nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7ee);
+	temp_phy_data2 = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+
+	if (temp_phy_data != temp_phy_data2) {
+		nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7fd);
+		temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+		if ((temp_phy_data & 0xff) > 0x20)
+			return 0;
+		printk(PFX "Reinitialize external PHY\n");
+	}
+
+	/* no heartbeat, configure the PHY */
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0x0000, 0x8000);
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc300, 0x0000);
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc316, 0x000A);
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc318, 0x0052);
+
+	switch (phy_type) {
+	case NES_PHY_TYPE_ARGUS:
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc316, 0x000A);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc318, 0x0052);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc302, 0x000C);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc319, 0x0008);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0027, 0x0001);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc31a, 0x0098);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0026, 0x0E00);
+
+		/* setup LEDs */
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd006, 0x0007);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd007, 0x000A);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd008, 0x0009);
+		break;
+
+	case NES_PHY_TYPE_SFP_D:
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc316, 0x000A);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc318, 0x0052);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc302, 0x0004);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc319, 0x0038);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0027, 0x0013);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc31a, 0x0098);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0026, 0x0E00);
+
+		/* setup LEDs */
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd006, 0x0007);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd007, 0x000A);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd008, 0x0009);
+		break;
+
+	case NES_PHY_TYPE_KR:
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc316, 0x000A);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc318, 0x0052);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc302, 0x000C);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc319, 0x0010);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0027, 0x0013);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc31a, 0x0080);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0026, 0x0E00);
+
+		/* setup LEDs */
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd006, 0x000B);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd007, 0x0003);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd008, 0x0004);
+
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0022, 0x406D);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0023, 0x0020);
+		break;
+	}
+
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0x0028, 0xA528);
+
+	/* Bring PHY out of reset */
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc300, 0x0002);
+
+	/* Check for heartbeat */
+	counter = 0;
+	mdelay(690);
+	nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7ee);
+	temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+	do {
+		if (counter++ > 150) {
+			printk(PFX "No PHY heartbeat\n");
+			break;
 		}
+		mdelay(1);
+		nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7ee);
+		temp_phy_data2 = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+	} while ((temp_phy_data2 == temp_phy_data));
 
-		nes_read_1G_phy_reg(nesdev, 1, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 1 phy address %u = 0x%X.\n",
-				nesadapter->phy_index[mac_index], phy_data);
-		nes_write_1G_phy_reg(nesdev, 23, nesadapter->phy_index[mac_index], 0xb000);
-
-		/* Reset the PHY */
-		nes_write_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index], 0x8000);
-		udelay(100);
-		counter = 0;
-		do {
-			nes_read_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index], &phy_data);
-			nes_debug(NES_DBG_PHY, "Phy data from register 0 = 0x%X.\n", phy_data);
-			if (counter++ > 100) break;
-		} while (phy_data & 0x8000);
-
-		/* Setting no phy loopback */
-		phy_data &= 0xbfff;
-		phy_data |= 0x1140;
-		nes_write_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index],  phy_data);
-		nes_read_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0 = 0x%X.\n", phy_data);
-
-		nes_read_1G_phy_reg(nesdev, 0x17, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x17 = 0x%X.\n", phy_data);
-
-		nes_read_1G_phy_reg(nesdev, 0x1e, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x1e = 0x%X.\n", phy_data);
-
-		/* Setting the interrupt mask */
-		nes_read_1G_phy_reg(nesdev, 0x19, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x19 = 0x%X.\n", phy_data);
-		nes_write_1G_phy_reg(nesdev, 0x19, nesadapter->phy_index[mac_index], 0xffee);
-
-		nes_read_1G_phy_reg(nesdev, 0x19, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x19 = 0x%X.\n", phy_data);
-
-		/* turning on flow control */
-		nes_read_1G_phy_reg(nesdev, 4, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x4 = 0x%X.\n", phy_data);
-		nes_write_1G_phy_reg(nesdev, 4, nesadapter->phy_index[mac_index],
-				(phy_data & ~(0x03E0)) | 0xc00);
-		/* nes_write_1G_phy_reg(nesdev, 4, nesadapter->phy_index[mac_index],
-				phy_data | 0xc00); */
-		nes_read_1G_phy_reg(nesdev, 4, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x4 = 0x%X.\n", phy_data);
-
-		nes_read_1G_phy_reg(nesdev, 9, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x9 = 0x%X.\n", phy_data);
-		/* Clear Half duplex */
-		nes_write_1G_phy_reg(nesdev, 9, nesadapter->phy_index[mac_index],
-				phy_data & ~(0x0100));
-		nes_read_1G_phy_reg(nesdev, 9, nesadapter->phy_index[mac_index], &phy_data);
-		nes_debug(NES_DBG_PHY, "Phy data from register 0x9 = 0x%X.\n", phy_data);
-
-		nes_read_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index], &phy_data);
-		nes_write_1G_phy_reg(nesdev, 0, nesadapter->phy_index[mac_index], phy_data | 0x0300);
-	} else {
-		if ((nesadapter->phy_type[mac_index] == NES_PHY_TYPE_IRIS) ||
-		    (nesadapter->phy_type[mac_index] == NES_PHY_TYPE_ARGUS)) {
-			/* setup 10G MDIO operation */
-			tx_config = nes_read_indexed(nesdev, NES_IDX_MAC_TX_CONFIG);
-			tx_config |= 0x14;
-			nes_write_indexed(nesdev, NES_IDX_MAC_TX_CONFIG, tx_config);
+	/* wait for tracking */
+	counter = 0;
+	do {
+		nes_read_10G_phy_reg(nesdev, phy_index, 0x3, 0xd7fd);
+		temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+		if (counter++ > 300) {
+			if (((temp_phy_data & 0xff) == 0x0) && first_attempt) {
+				first_attempt = 0;
+				counter = 0;
+				/* reset AMCC PHY and try again */
+				nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0xe854, 0x00c0);
+				nes_write_10G_phy_reg(nesdev, phy_index, 0x3, 0xe854, 0x0040);
+				continue;
+			} else {
+				ret = 1;
+				break;
+			}
 		}
-		if ((nesadapter->phy_type[mac_index] == NES_PHY_TYPE_ARGUS)) {
-			nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0xd7ee);
-
-			temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-			mdelay(10);
-			nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0xd7ee);
-			temp_phy_data2 = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-
-			/*
-			 * if firmware is already running (like from a
-			 * driver un-load/load, don't do anything.
-			 */
-			if (temp_phy_data == temp_phy_data2) {
-				/* configure QT2505 AMCC PHY */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0x0000, 0x8000);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc300, 0x0000);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc302, 0x0044);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc318, 0x0052);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc319, 0x0008);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc31a, 0x0098);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0x0026, 0x0E00);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0x0027, 0x0000);
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0x0028, 0xA528);
+		mdelay(10);
+	} while ((temp_phy_data & 0xff) < 0x30);
+
+	/* setup signal integrity */
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xd003, 0x0000);
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xF00D, 0x00FE);
+	nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xF00E, 0x0032);
+	if (phy_type == NES_PHY_TYPE_KR) {
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xF00F, 0x000C);
+	} else {
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xF00F, 0x0002);
+		nes_write_10G_phy_reg(nesdev, phy_index, 0x1, 0xc314, 0x0063);
+	}
 
-				/*
-				 * remove micro from reset; chip boots from ROM,
-				 * uploads EEPROM f/w image, uC executes f/w
-				 */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc300, 0x0002);
+	/* reset serdes */
+	sds = nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0 + mac_index * 0x200);
+	sds |= 0x1;
+	nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0 + mac_index * 0x200, sds);
+	sds &= 0xfffffffe;
+	nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0 + mac_index * 0x200, sds);
 
-				/*
-				 * wait for heart beat to start to
-				 * know loading is done
-				 */
-				counter = 0;
-				do {
-					nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0xd7ee);
-					temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-					if (counter++ > 1000) {
-						nes_debug(NES_DBG_PHY, "AMCC PHY- breaking from heartbeat check <this is bad!!!> \n");
-						break;
-					}
-					mdelay(100);
-					nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0xd7ee);
-					temp_phy_data2 = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-				} while ((temp_phy_data2 == temp_phy_data));
+	counter = 0;
+	while (((nes_read32(nesdev->regs + NES_SOFTWARE_RESET) & 0x00000040) != 0x00000040)
+			&& (counter++ < 5000))
+		;
 
-				/*
-				 * wait for tracking to start to know
-				 * f/w is good to go
-				 */
-				counter = 0;
-				do {
-					nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x3, 0xd7fd);
-					temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-					if (counter++ > 1000) {
-						nes_debug(NES_DBG_PHY, "AMCC PHY- breaking from status check <this is bad!!!> \n");
-						break;
-					}
-					mdelay(1000);
-					/*
-					 * nes_debug(NES_DBG_PHY, "AMCC PHY- phy_status not ready yet = 0x%02X\n",
-					 *			temp_phy_data);
-					 */
-				} while (((temp_phy_data & 0xff) != 0x50) && ((temp_phy_data & 0xff) != 0x70));
-
-				/* set LOS Control invert RXLOSB_I_PADINV */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xd003, 0x0000);
-				/* set LOS Control to mask of RXLOSB_I */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xc314, 0x0042);
-				/* set LED1 to input mode (LED1 and LED2 share same LED) */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xd006, 0x0007);
-				/* set LED2 to RX link_status and activity */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xd007, 0x000A);
-				/* set LED3 to RX link_status */
-				nes_write_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 0x1, 0xd008, 0x0009);
+	return ret;
+}
 
-				/*
-				 * reset the res-calibration on t2
-				 * serdes; ensures it is stable after
-				 * the amcc phy is stable
-				 */
 
-				sds_common_control0  = nes_read_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0);
-				sds_common_control0 |= 0x1;
-				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0, sds_common_control0);
+/**
+ * nes_init_phy
+ */
+int nes_init_phy(struct nes_device *nesdev)
+{
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
+	u32 mac_index = nesdev->mac_index;
+	u32 tx_config = 0;
+	unsigned long flags;
+	u8  phy_type = nesadapter->phy_type[mac_index];
+	u8  phy_index = nesadapter->phy_index[mac_index];
+	int ret = 0;
 
-				/* release the res-calibration reset */
-				sds_common_control0 &= 0xfffffffe;
-				nes_write_indexed(nesdev, NES_IDX_ETH_SERDES_COMMON_CONTROL0, sds_common_control0);
+	tx_config = nes_read_indexed(nesdev, NES_IDX_MAC_TX_CONFIG);
+	if (phy_type == NES_PHY_TYPE_1G) {
+		/* setup 1G MDIO operation */
+		tx_config &= 0xFFFFFFE3;
+		tx_config |= 0x04;
+	} else {
+		/* setup 10G MDIO operation */
+		tx_config &= 0xFFFFFFE3;
+		tx_config |= 0x1D;
+	}
+	nes_write_indexed(nesdev, NES_IDX_MAC_TX_CONFIG, tx_config);
 
-				i = 0;
-				while (((nes_read32(nesdev->regs + NES_SOFTWARE_RESET) & 0x00000040) != 0x00000040)
-						&& (i++ < 5000)) {
-					/* mdelay(1); */
-				}
+	spin_lock_irqsave(&nesdev->nesadapter->phy_lock, flags);
 
-				/*
-				 * wait for link train done before moving on,
-				 * or will get an interupt storm
-				 */
-				counter = 0;
-				do {
-					temp_phy_data = nes_read_indexed(nesdev, NES_IDX_PHY_PCS_CONTROL_STATUS0 +
-								(0x200 * (nesdev->mac_index & 1)));
-					if (counter++ > 1000) {
-						nes_debug(NES_DBG_PHY, "AMCC PHY- breaking from link train wait <this is bad, link didnt train!!!>\n");
-						break;
-					}
-					mdelay(1);
-				} while (((temp_phy_data & 0x0f1f0000) != 0x0f0f0000));
-			}
-		}
+	switch (phy_type) {
+	case NES_PHY_TYPE_1G:
+		ret = nes_init_1g_phy(nesdev, phy_type, phy_index);
+		break;
+	case NES_PHY_TYPE_ARGUS:
+	case NES_PHY_TYPE_SFP_D:
+	case NES_PHY_TYPE_KR:
+		ret = nes_init_2025_phy(nesdev, phy_type, phy_index);
+		break;
 	}
-	return 0;
+
+	spin_unlock_irqrestore(&nesdev->nesadapter->phy_lock, flags);
+
+	return ret;
 }
 
 
@@ -1423,6 +1562,7 @@ static void nes_replenish_nic_rq(struct nes_vnic *nesvnic)
 	struct nes_hw_nic_rq_wqe *nic_rqe;
 	struct nes_hw_nic *nesnic;
 	struct nes_device *nesdev;
+	struct nes_rskb_cb *cb;
 	u32 rx_wqes_posted = 0;
 
 	nesnic = &nesvnic->nic;
@@ -1448,6 +1588,9 @@ static void nes_replenish_nic_rq(struct nes_vnic *nesvnic)
 
 			bus_address = pci_map_single(nesdev->pcidev,
 					skb->data, nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = bus_address;
+			cb->maplen = nesvnic->max_frame_size;
 
 			nic_rqe = &nesnic->rq_vbase[nesvnic->nic.rq_head];
 			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] =
@@ -1537,6 +1680,7 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 	u32 cqp_head;
 	u32 counter;
 	u32 wqe_count;
+	struct nes_rskb_cb *cb;
 	u8 jumbomode=0;
 
 	/* Allocate fragment, SQ, RQ, and CQ; Reuse CEQ based on the PCI function */
@@ -1595,7 +1739,6 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 	nesvnic->post_cqp_request = nes_post_cqp_request;
 	nesvnic->mcrq_mcast_filter = NULL;
 
-	spin_lock_init(&nesvnic->nic.sq_lock);
 	spin_lock_init(&nesvnic->nic.rq_lock);
 
 	/* setup the RQ */
@@ -1714,6 +1857,9 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 
 		pmem = pci_map_single(nesdev->pcidev, skb->data,
 				nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+		cb = (struct nes_rskb_cb *)&skb->cb[0];
+		cb->busaddr = pmem;
+		cb->maplen = nesvnic->max_frame_size;
 
 		nic_rqe = &nesvnic->nic.rq_vbase[counter];
 		nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] = cpu_to_le32(nesvnic->max_frame_size);
@@ -1742,6 +1888,13 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
 			jumbomode = 1;
 		nes_nic_init_timer_defaults(nesdev, jumbomode);
 	}
+	if ((nesdev->nesadapter->allow_unaligned_fpdus) &&
+		(nes_init_mgt_qp(nesdev, netdev, nesvnic))) {
+			nes_debug(NES_DBG_INIT, "%s: Out of memory for pau nic\n", netdev->name);
+			nes_destroy_nic_qp(nesvnic);
+		return -ENOMEM;
+	}
+
 	nesvnic->lro_mgr.max_aggr       = nes_lro_max_aggr;
 	nesvnic->lro_mgr.max_desc       = NES_MAX_LRO_DESCRIPTORS;
 	nesvnic->lro_mgr.lro_arr        = nesvnic->lro_desc;
@@ -1759,25 +1912,91 @@ int nes_init_nic_qp(struct nes_device *nesdev, struct net_device *netdev)
  */
 void nes_destroy_nic_qp(struct nes_vnic *nesvnic)
 {
+	u64 u64temp;
+	dma_addr_t bus_address;
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_hw_cqp_wqe *cqp_wqe;
-	struct nes_hw_nic_rq_wqe *nic_rqe;
-	u64 wqe_frag;
+	struct nes_hw_nic_sq_wqe *nic_sqe;
+	__le16 *wqe_fragment_length;
+	u16  wqe_fragment_index;
 	u32 cqp_head;
+	u32 wqm_cfg0;
 	unsigned long flags;
+	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
 	int ret;
 
+	if (nesdev->nesadapter->allow_unaligned_fpdus)
+		nes_destroy_mgt(nesvnic);
+
+	/* clear wqe stall before destroying NIC QP */
+	wqm_cfg0 = nes_read_indexed(nesdev, NES_IDX_WQM_CONFIG0);
+	nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG0, wqm_cfg0 & 0xFFFF7FFF);
+
 	/* Free remaining NIC receive buffers */
 	while (nesvnic->nic.rq_head != nesvnic->nic.rq_tail) {
-		nic_rqe   = &nesvnic->nic.rq_vbase[nesvnic->nic.rq_tail];
-		wqe_frag  = (u64)le32_to_cpu(nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX]);
-		wqe_frag |= ((u64)le32_to_cpu(nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX])) << 32;
-		pci_unmap_single(nesdev->pcidev, (dma_addr_t)wqe_frag,
-				nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+		rx_skb = nesvnic->nic.rx_skb[nesvnic->nic.rq_tail];
+		cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+		pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen,
+			PCI_DMA_FROMDEVICE);
+
 		dev_kfree_skb(nesvnic->nic.rx_skb[nesvnic->nic.rq_tail++]);
 		nesvnic->nic.rq_tail &= (nesvnic->nic.rq_size - 1);
 	}
 
+	/* Free remaining NIC transmit buffers */
+	while (nesvnic->nic.sq_head != nesvnic->nic.sq_tail) {
+		nic_sqe = &nesvnic->nic.sq_vbase[nesvnic->nic.sq_tail];
+		wqe_fragment_index = 1;
+		wqe_fragment_length = (__le16 *)
+			&nic_sqe->wqe_words[NES_NIC_SQ_WQE_LENGTH_0_TAG_IDX];
+		/* bump past the vlan tag */
+		wqe_fragment_length++;
+		if (le16_to_cpu(wqe_fragment_length[wqe_fragment_index]) != 0) {
+			u64temp = (u64)le32_to_cpu(
+				nic_sqe->wqe_words[NES_NIC_SQ_WQE_FRAG0_LOW_IDX+
+				wqe_fragment_index*2]);
+			u64temp += ((u64)le32_to_cpu(
+				nic_sqe->wqe_words[NES_NIC_SQ_WQE_FRAG0_HIGH_IDX
+				+ wqe_fragment_index*2]))<<32;
+			bus_address = (dma_addr_t)u64temp;
+			if (test_and_clear_bit(nesvnic->nic.sq_tail,
+					nesvnic->nic.first_frag_overflow)) {
+				pci_unmap_single(nesdev->pcidev,
+						bus_address,
+						le16_to_cpu(wqe_fragment_length[
+							wqe_fragment_index++]),
+						PCI_DMA_TODEVICE);
+			}
+			for (; wqe_fragment_index < 5; wqe_fragment_index++) {
+				if (wqe_fragment_length[wqe_fragment_index]) {
+					u64temp = le32_to_cpu(
+						nic_sqe->wqe_words[
+						NES_NIC_SQ_WQE_FRAG0_LOW_IDX+
+						wqe_fragment_index*2]);
+					u64temp += ((u64)le32_to_cpu(
+						nic_sqe->wqe_words[
+						NES_NIC_SQ_WQE_FRAG0_HIGH_IDX+
+						wqe_fragment_index*2]))<<32;
+					bus_address = (dma_addr_t)u64temp;
+					pci_unmap_page(nesdev->pcidev,
+							bus_address,
+							le16_to_cpu(
+							wqe_fragment_length[
+							wqe_fragment_index]),
+							PCI_DMA_TODEVICE);
+				} else
+					break;
+			}
+		}
+		if (nesvnic->nic.tx_skb[nesvnic->nic.sq_tail])
+			dev_kfree_skb(
+				nesvnic->nic.tx_skb[nesvnic->nic.sq_tail]);
+
+		nesvnic->nic.sq_tail = (nesvnic->nic.sq_tail + 1)
+					& (nesvnic->nic.sq_size - 1);
+	}
+
 	spin_lock_irqsave(&nesdev->cqp.lock, flags);
 
 	/* Destroy NIC QP */
@@ -1830,6 +2049,9 @@ void nes_destroy_nic_qp(struct nes_vnic *nesvnic)
 
 	pci_free_consistent(nesdev->pcidev, nesvnic->nic_mem_size, nesvnic->nic_vbase,
 			nesvnic->nic_pbase);
+
+	/* restore old wqm_cfg0 value */
+	nes_write_indexed(nesdev, NES_IDX_WQM_CONFIG0, wqm_cfg0);
 }
 
 /**
@@ -1894,7 +2116,30 @@ int nes_napi_isr(struct nes_device *nesdev)
 	}
 }
 
-
+static void process_critical_error(struct nes_device *nesdev)
+{
+	u32 debug_error;
+	u32 nes_idx_debug_error_masks0 = 0;
+	u16 error_module = 0;
+
+	debug_error = nes_read_indexed(nesdev, NES_IDX_DEBUG_ERROR_CONTROL_STATUS);
+	printk(KERN_ERR PFX "Critical Error reported by device!!! 0x%02X\n",
+			(u16)debug_error);
+	nes_write_indexed(nesdev, NES_IDX_DEBUG_ERROR_CONTROL_STATUS,
+			0x01010000 | (debug_error & 0x0000ffff));
+	if (crit_err_count++ > 10)
+		nes_write_indexed(nesdev, NES_IDX_DEBUG_ERROR_MASKS1, 1 << 0x17);
+	error_module = (u16) (debug_error & 0x1F00) >> 8;
+	if (++nesdev->nesadapter->crit_error_count[error_module-1] >=
+			nes_max_critical_error_count) {
+		printk(KERN_ERR PFX "Masking off critical error for module "
+			"0x%02X\n", (u16)error_module);
+		nes_idx_debug_error_masks0 = nes_read_indexed(nesdev,
+			NES_IDX_DEBUG_ERROR_MASKS0);
+		nes_write_indexed(nesdev, NES_IDX_DEBUG_ERROR_MASKS0,
+			nes_idx_debug_error_masks0 | (1 << error_module));
+	}
+}
 /**
  * nes_dpc
  */
@@ -1909,7 +2154,6 @@ void nes_dpc(unsigned long param)
 	u32 timer_stat;
 	u32 temp_int_stat;
 	u32 intf_int_stat;
-	u32 debug_error;
 	u32 processed_intf_int = 0;
 	u16 processed_timer_int = 0;
 	u16 completion_ints = 0;
@@ -1987,14 +2231,7 @@ void nes_dpc(unsigned long param)
 				intf_int_stat = nes_read32(nesdev->regs+NES_INTF_INT_STAT);
 				intf_int_stat &= nesdev->intf_int_req;
 				if (NES_INTF_INT_CRITERR & intf_int_stat) {
-					debug_error = nes_read_indexed(nesdev, NES_IDX_DEBUG_ERROR_CONTROL_STATUS);
-					printk(KERN_ERR PFX "Critical Error reported by device!!! 0x%02X\n",
-							(u16)debug_error);
-					nes_write_indexed(nesdev, NES_IDX_DEBUG_ERROR_CONTROL_STATUS,
-							0x01010000 | (debug_error & 0x0000ffff));
-					/* BUG(); */
-					if (crit_err_count++ > 10)
-						nes_write_indexed(nesdev, NES_IDX_DEBUG_ERROR_MASKS1, 1 << 0x17);
+					process_critical_error(nesdev);
 				}
 				if (NES_INTF_INT_PCIERR & intf_int_stat) {
 					printk(KERN_ERR PFX "PCI Error reported by device!!!\n");
@@ -2145,6 +2382,8 @@ static void nes_process_aeq(struct nes_device *nesdev, struct nes_hw_aeq *aeq)
 
 		if (++head >= aeq_size)
 			head = 0;
+
+		nes_write32(nesdev->regs + NES_AEQ_ALLOC, 1 << 16);
 	}
 	while (1);
 	aeq->aeq_head = head;
@@ -2234,6 +2473,7 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 	u16 temp_phy_data;
 	u32 pcs_val  = 0x0f0f0000;
 	u32 pcs_mask = 0x0f1f0000;
+	u32 cdr_ctrl;
 
 	spin_lock_irqsave(&nesadapter->phy_lock, flags);
 	if (nesadapter->mac_sw_state[mac_number] != NES_MAC_SW_IDLE) {
@@ -2241,7 +2481,6 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 		return;
 	}
 	nesadapter->mac_sw_state[mac_number] = NES_MAC_SW_INTERRUPT;
-	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 
 	/* ack the MAC interrupt */
 	mac_status = nes_read_indexed(nesdev, NES_IDX_MAC_INT_STATUS + (mac_index * 0x200));
@@ -2252,13 +2491,12 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 
 	if (mac_status & (NES_MAC_INT_LINK_STAT_CHG | NES_MAC_INT_XGMII_EXT)) {
 		nesdev->link_status_interrupts++;
-		if (0 == (++nesadapter->link_interrupt_count[mac_index] % ((u16)NES_MAX_LINK_INTERRUPTS))) {
-			spin_lock_irqsave(&nesadapter->phy_lock, flags);
+		if (0 == (++nesadapter->link_interrupt_count[mac_index] % ((u16)NES_MAX_LINK_INTERRUPTS)))
 			nes_reset_link(nesdev, mac_index);
-			spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
-		}
+
 		/* read the PHY interrupt status register */
-		if (nesadapter->OneG_Mode) {
+		if ((nesadapter->OneG_Mode) &&
+		(nesadapter->phy_type[mac_index] != NES_PHY_TYPE_PUMA_1G)) {
 			do {
 				nes_read_1G_phy_reg(nesdev, 0x1a,
 						nesadapter->phy_index[mac_index], &phy_data);
@@ -2331,22 +2569,9 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 			}
 		} else {
 			switch (nesadapter->phy_type[mac_index]) {
-			case NES_PHY_TYPE_IRIS:
-				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 1);
-				temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-				u32temp = 20;
-				do {
-					nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 1);
-					phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-					if ((phy_data == temp_phy_data) || (!(--u32temp)))
-						break;
-					temp_phy_data = phy_data;
-				} while (1);
-				nes_debug(NES_DBG_PHY, "%s: Phy data = 0x%04X, link was %s.\n",
-					__func__, phy_data, nesadapter->mac_link_down[mac_index] ? "DOWN" : "UP");
-				break;
-
 			case NES_PHY_TYPE_ARGUS:
+			case NES_PHY_TYPE_SFP_D:
+			case NES_PHY_TYPE_KR:
 				/* clear the alarms */
 				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 4, 0x0008);
 				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 4, 0xc001);
@@ -2357,19 +2582,18 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 0x9004);
 				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 0x9005);
 				/* check link status */
-				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 1);
+				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 0x9003);
 				temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-				u32temp = 100;
-				do {
-					nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 1);
 
-					phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
-					if ((phy_data == temp_phy_data) || (!(--u32temp)))
-						break;
-					temp_phy_data = phy_data;
-				} while (1);
+				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 3, 0x0021);
+				nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+				nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 3, 0x0021);
+				phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+
+				phy_data = (!temp_phy_data && (phy_data == 0x8000)) ? 0x4 : 0x0;
+
 				nes_debug(NES_DBG_PHY, "%s: Phy data = 0x%04X, link was %s.\n",
-					__func__, phy_data, nesadapter->mac_link_down ? "DOWN" : "UP");
+					__func__, phy_data, nesadapter->mac_link_down[mac_index] ? "DOWN" : "UP");
 				break;
 
 			case NES_PHY_TYPE_PUMA_1G:
@@ -2385,6 +2609,17 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 		}
 
 		if (phy_data & 0x0004) {
+			if (wide_ppm_offset &&
+			    (nesadapter->phy_type[mac_index] == NES_PHY_TYPE_CX4) &&
+			    (nesadapter->hw_rev != NE020_REV)) {
+				cdr_ctrl = nes_read_indexed(nesdev,
+							    NES_IDX_ETH_SERDES_CDR_CONTROL0 +
+							    mac_index * 0x200);
+				nes_write_indexed(nesdev,
+						  NES_IDX_ETH_SERDES_CDR_CONTROL0 +
+						  mac_index * 0x200,
+						  cdr_ctrl | 0x000F0000);
+			}
 			nesadapter->mac_link_down[mac_index] = 0;
 			list_for_each_entry(nesvnic, &nesadapter->nesvnic_list[mac_index], list) {
 				nes_debug(NES_DBG_PHY, "The Link is UP!!.  linkup was %d\n",
@@ -2396,9 +2631,29 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 						netif_start_queue(nesvnic->netdev);
 					nesvnic->linkup = 1;
 					netif_carrier_on(nesvnic->netdev);
+
+					spin_lock(&nesvnic->port_ibevent_lock);
+					if (nesvnic->of_device_registered) {
+						if (nesdev->iw_status == 0) {
+							nesdev->iw_status = 1;
+							nes_port_ibevent(nesvnic);
+						}
+					}
+					spin_unlock(&nesvnic->port_ibevent_lock);
 				}
 			}
 		} else {
+			if (wide_ppm_offset &&
+			    (nesadapter->phy_type[mac_index] == NES_PHY_TYPE_CX4) &&
+			    (nesadapter->hw_rev != NE020_REV)) {
+				cdr_ctrl = nes_read_indexed(nesdev,
+							    NES_IDX_ETH_SERDES_CDR_CONTROL0 +
+							    mac_index * 0x200);
+				nes_write_indexed(nesdev,
+						  NES_IDX_ETH_SERDES_CDR_CONTROL0 +
+						  mac_index * 0x200,
+						  cdr_ctrl & 0xFFF0FFFF);
+			}
 			nesadapter->mac_link_down[mac_index] = 1;
 			list_for_each_entry(nesvnic, &nesadapter->nesvnic_list[mac_index], list) {
 				nes_debug(NES_DBG_PHY, "The Link is Down!!. linkup was %d\n",
@@ -2410,21 +2665,115 @@ static void nes_process_mac_intr(struct nes_device *nesdev, u32 mac_number)
 						netif_stop_queue(nesvnic->netdev);
 					nesvnic->linkup = 0;
 					netif_carrier_off(nesvnic->netdev);
+
+					spin_lock(&nesvnic->port_ibevent_lock);
+					if (nesvnic->of_device_registered) {
+						if (nesdev->iw_status == 1) {
+							nesdev->iw_status = 0;
+							nes_port_ibevent(nesvnic);
+						}
+					}
+					spin_unlock(&nesvnic->port_ibevent_lock);
 				}
 			}
 		}
+		if (nesadapter->phy_type[mac_index] == NES_PHY_TYPE_SFP_D) {
+			nesdev->link_recheck = 1;
+			mod_delayed_work(system_wq, &nesdev->work,
+					 NES_LINK_RECHECK_DELAY);
+		}
 	}
 
+	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
+
 	nesadapter->mac_sw_state[mac_number] = NES_MAC_SW_IDLE;
 }
 
+void nes_recheck_link_status(struct work_struct *work)
+{
+	unsigned long flags;
+	struct nes_device *nesdev = container_of(work, struct nes_device, work.work);
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
+	struct nes_vnic *nesvnic;
+	u32 mac_index = nesdev->mac_index;
+	u16 phy_data;
+	u16 temp_phy_data;
+
+	spin_lock_irqsave(&nesadapter->phy_lock, flags);
+
+	/* check link status */
+	nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 1, 0x9003);
+	temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+
+	nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 3, 0x0021);
+	nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+	nes_read_10G_phy_reg(nesdev, nesadapter->phy_index[mac_index], 3, 0x0021);
+	phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+
+	phy_data = (!temp_phy_data && (phy_data == 0x8000)) ? 0x4 : 0x0;
+
+	nes_debug(NES_DBG_PHY, "%s: Phy data = 0x%04X, link was %s.\n",
+		__func__, phy_data,
+		nesadapter->mac_link_down[mac_index] ? "DOWN" : "UP");
+
+	if (phy_data & 0x0004) {
+		nesadapter->mac_link_down[mac_index] = 0;
+		list_for_each_entry(nesvnic, &nesadapter->nesvnic_list[mac_index], list) {
+			if (nesvnic->linkup == 0) {
+				printk(PFX "The Link is now up for port %s, netdev %p.\n",
+						nesvnic->netdev->name, nesvnic->netdev);
+				if (netif_queue_stopped(nesvnic->netdev))
+					netif_start_queue(nesvnic->netdev);
+				nesvnic->linkup = 1;
+				netif_carrier_on(nesvnic->netdev);
+
+				spin_lock(&nesvnic->port_ibevent_lock);
+				if (nesvnic->of_device_registered) {
+					if (nesdev->iw_status == 0) {
+						nesdev->iw_status = 1;
+						nes_port_ibevent(nesvnic);
+					}
+				}
+				spin_unlock(&nesvnic->port_ibevent_lock);
+			}
+		}
+
+	} else {
+		nesadapter->mac_link_down[mac_index] = 1;
+		list_for_each_entry(nesvnic, &nesadapter->nesvnic_list[mac_index], list) {
+			if (nesvnic->linkup == 1) {
+				printk(PFX "The Link is now down for port %s, netdev %p.\n",
+						nesvnic->netdev->name, nesvnic->netdev);
+				if (!(netif_queue_stopped(nesvnic->netdev)))
+					netif_stop_queue(nesvnic->netdev);
+				nesvnic->linkup = 0;
+				netif_carrier_off(nesvnic->netdev);
+
+				spin_lock(&nesvnic->port_ibevent_lock);
+				if (nesvnic->of_device_registered) {
+					if (nesdev->iw_status == 1) {
+						nesdev->iw_status = 0;
+						nes_port_ibevent(nesvnic);
+					}
+				}
+				spin_unlock(&nesvnic->port_ibevent_lock);
+			}
+		}
+	}
+	if (nesdev->link_recheck++ < NES_LINK_RECHECK_MAX)
+		schedule_delayed_work(&nesdev->work, NES_LINK_RECHECK_DELAY);
+	else
+		nesdev->link_recheck = 0;
+
+	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
+}
 
 
 static void nes_nic_napi_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 {
 	struct nes_vnic *nesvnic = container_of(cq, struct nes_vnic, nic_cq);
 
-	netif_rx_schedule(nesdev->netdev[nesvnic->netdev_index], &nesvnic->napi);
+	napi_schedule(&nesvnic->napi);
 }
 
 
@@ -2447,6 +2796,7 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 	struct nes_hw_nic_sq_wqe *nic_sqe;
 	struct sk_buff *skb;
 	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
 	__le16 *wqe_fragment_length;
 	u32 head;
 	u32 cq_size;
@@ -2505,16 +2855,16 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 						} else
 							break;
 					}
-					if (skb)
-						dev_kfree_skb_any(skb);
 				}
+				if (skb)
+					dev_kfree_skb_any(skb);
 				nesnic->sq_tail++;
 				nesnic->sq_tail &= nesnic->sq_size-1;
 				if (sq_cqes > 128) {
 					barrier();
-				/* restart the queue if it had been stopped */
-				if (netif_queue_stopped(nesvnic->netdev))
-					netif_wake_queue(nesvnic->netdev);
+					/* restart the queue if it had been stopped */
+					if (netif_queue_stopped(nesvnic->netdev))
+						netif_wake_queue(nesvnic->netdev);
 					sq_cqes = 0;
 				}
 			} else {
@@ -2531,6 +2881,8 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 				bus_address += ((u64)le32_to_cpu(nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX])) << 32;
 				pci_unmap_single(nesdev->pcidev, bus_address,
 						nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+				cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+				cb->busaddr = 0;
 				/* rx_skb->tail = rx_skb->data + rx_pkt_size; */
 				/* rx_skb->len = rx_pkt_size; */
 				rx_skb->len = 0;  /* TODO: see if this is necessary */
@@ -2557,9 +2909,8 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 					if ((cqe_errv &
 							(NES_NIC_ERRV_BITS_IPV4_CSUM_ERR | NES_NIC_ERRV_BITS_TCPUDP_CSUM_ERR |
 							NES_NIC_ERRV_BITS_IPH_ERR | NES_NIC_ERRV_BITS_WQE_OVERRUN)) == 0) {
-						if (nesvnic->rx_checksum_disabled == 0) {
+						if (nesvnic->netdev->features & NETIF_F_RXCSUM)
 							rx_skb->ip_summed = CHECKSUM_UNNECESSARY;
-						}
 					} else
 						nes_debug(NES_DBG_CQ, "%s: unsuccessfully checksummed TCP or UDP packet."
 								" errv = 0x%X, pkt_type = 0x%X.\n",
@@ -2569,7 +2920,7 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 					if ((cqe_errv &
 							(NES_NIC_ERRV_BITS_IPV4_CSUM_ERR | NES_NIC_ERRV_BITS_IPH_ERR |
 							NES_NIC_ERRV_BITS_WQE_OVERRUN)) == 0) {
-						if (nesvnic->rx_checksum_disabled == 0) {
+						if (nesvnic->netdev->features & NETIF_F_RXCSUM) {
 							rx_skb->ip_summed = CHECKSUM_UNNECESSARY;
 							/* nes_debug(NES_DBG_CQ, "%s: Reporting successfully checksummed IPv4 packet.\n",
 								  nesvnic->netdev->name); */
@@ -2583,28 +2934,29 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 							pkt_type, (pkt_type & NES_PKT_TYPE_APBVT_MASK)); */
 
 				if ((pkt_type & NES_PKT_TYPE_APBVT_MASK) == NES_PKT_TYPE_APBVT_BITS) {
-					nes_cm_recv(rx_skb, nesvnic->netdev);
-				} else {
-					if ((cqe_misc & NES_NIC_CQE_TAG_VALID) && (nesvnic->vlan_grp != NULL)) {
-						vlan_tag = (u16)(le32_to_cpu(
-								cq->cq_vbase[head].cqe_words[NES_NIC_CQE_TAG_PKT_TYPE_IDX])
-								>> 16);
-						nes_debug(NES_DBG_CQ, "%s: Reporting stripped VLAN packet. Tag = 0x%04X\n",
-								nesvnic->netdev->name, vlan_tag);
-						if (nes_use_lro)
-							lro_vlan_hwaccel_receive_skb(&nesvnic->lro_mgr, rx_skb,
-									nesvnic->vlan_grp, vlan_tag, NULL);
-						else
-							nes_vlan_rx(rx_skb, nesvnic->vlan_grp, vlan_tag);
-					} else {
-						if (nes_use_lro)
-							lro_receive_skb(&nesvnic->lro_mgr, rx_skb, NULL);
-						else
-							nes_netif_rx(rx_skb);
-					}
+					if (nes_cm_recv(rx_skb, nesvnic->netdev))
+						rx_skb = NULL;
 				}
+				if (rx_skb == NULL)
+					goto skip_rx_indicate0;
+
+
+				if (cqe_misc & NES_NIC_CQE_TAG_VALID) {
+					vlan_tag = (u16)(le32_to_cpu(
+							cq->cq_vbase[head].cqe_words[NES_NIC_CQE_TAG_PKT_TYPE_IDX])
+							>> 16);
+					nes_debug(NES_DBG_CQ, "%s: Reporting stripped VLAN packet. Tag = 0x%04X\n",
+							nesvnic->netdev->name, vlan_tag);
+
+					__vlan_hwaccel_put_tag(rx_skb, htons(ETH_P_8021Q), vlan_tag);
+				}
+				if (nes_use_lro)
+					lro_receive_skb(&nesvnic->lro_mgr, rx_skb, NULL);
+				else
+					netif_receive_skb(rx_skb);
 
-				nesvnic->netdev->last_rx = jiffies;
+skip_rx_indicate0:
+				;
 				/* nesvnic->netstats.rx_packets++; */
 				/* nesvnic->netstats.rx_bytes += rx_pkt_size; */
 			}
@@ -2655,6 +3007,7 @@ void nes_nic_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
 }
 
 
+
 /**
  * nes_cqp_ce_handler
  */
@@ -2669,6 +3022,8 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 	u32 cq_size;
 	u32 cqe_count=0;
 	u32 error_code;
+	u32 opcode;
+	u32 ctx_index;
 	/* u32 counter; */
 
 	head = cq->cq_head;
@@ -2679,12 +3034,9 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 		/* nes_debug(NES_DBG_CQP, "head=%u cqe_words=%08X\n", head,
 			  le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX])); */
 
-		if (le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]) & NES_CQE_VALID) {
-			u64temp = (((u64)(le32_to_cpu(cq->cq_vbase[head].
-					cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX]))) << 32) |
-					((u64)(le32_to_cpu(cq->cq_vbase[head].
-					cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX])));
-			cqp = *((struct nes_hw_cqp **)&u64temp);
+		opcode = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]);
+		if (opcode & NES_CQE_VALID) {
+			cqp = &nesdev->cqp;
 
 			error_code = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_ERROR_CODE_IDX]);
 			if (error_code) {
@@ -2693,15 +3045,14 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 						le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX])&0x3f,
 						(u16)(error_code >> 16),
 						(u16)error_code);
-				nes_debug(NES_DBG_CQP, "cqp: qp_id=%u, sq_head=%u, sq_tail=%u\n",
-						cqp->qp_id, cqp->sq_head, cqp->sq_tail);
 			}
 
-			u64temp = (((u64)(le32_to_cpu(nesdev->cqp.sq_vbase[cqp->sq_tail].
-					wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX]))) << 32) |
-					((u64)(le32_to_cpu(nesdev->cqp.sq_vbase[cqp->sq_tail].
-					wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX])));
-			cqp_request = *((struct nes_cqp_request **)&u64temp);
+			u64temp = (((u64)(le32_to_cpu(cq->cq_vbase[head].
+					cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX]))) << 32) |
+					((u64)(le32_to_cpu(cq->cq_vbase[head].
+					cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX])));
+
+			cqp_request = (struct nes_cqp_request *)(unsigned long)u64temp;
 			if (cqp_request) {
 				if (cqp_request->waiting) {
 					/* nes_debug(NES_DBG_CQP, "%s: Waking up requestor\n"); */
@@ -2747,9 +3098,15 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 		cqp_wqe = &nesdev->cqp.sq_vbase[head];
 		memcpy(cqp_wqe, &cqp_request->cqp_wqe, sizeof(*cqp_wqe));
 		barrier();
-		cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_LOW_IDX] =
+
+		opcode = cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX];
+		if ((opcode & NES_CQP_OPCODE_MASK) == NES_CQP_DOWNLOAD_SEGMENT)
+			ctx_index = NES_CQP_WQE_DL_COMP_CTX_LOW_IDX;
+		else
+			ctx_index = NES_CQP_WQE_COMP_CTX_LOW_IDX;
+		cqp_wqe->wqe_words[ctx_index] =
 			cpu_to_le32((u32)((unsigned long)cqp_request));
-		cqp_wqe->wqe_words[NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX] =
+		cqp_wqe->wqe_words[ctx_index + 1] =
 			cpu_to_le32((u32)(upper_32_bits((unsigned long)cqp_request)));
 		nes_debug(NES_DBG_CQP, "CQP request %p (opcode 0x%02X) put on CQPs SQ wqe%u.\n",
 				cqp_request, le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX])&0x3f, head);
@@ -2765,6 +3122,415 @@ static void nes_cqp_ce_handler(struct nes_device *nesdev, struct nes_hw_cq *cq)
 	nes_read32(nesdev->regs+NES_CQE_ALLOC);
 }
 
+static u8 *locate_mpa(u8 *pkt, u32 aeq_info)
+{
+	if (aeq_info & NES_AEQE_Q2_DATA_ETHERNET) {
+		/* skip over ethernet header */
+		pkt += ETH_HLEN;
+
+		/* Skip over IP and TCP headers */
+		pkt += 4 * (pkt[0] & 0x0f);
+		pkt += 4 * ((pkt[12] >> 4) & 0x0f);
+	}
+	return pkt;
+}
+
+/* Determine if incoming error pkt is rdma layer */
+static u32 iwarp_opcode(struct nes_qp *nesqp, u32 aeq_info)
+{
+	u8 *pkt;
+	u16 *mpa;
+	u32 opcode = 0xffffffff;
+
+	if (aeq_info & NES_AEQE_Q2_DATA_WRITTEN) {
+		pkt = nesqp->hwqp.q2_vbase + BAD_FRAME_OFFSET;
+		mpa = (u16 *)locate_mpa(pkt, aeq_info);
+		opcode = be16_to_cpu(mpa[1]) & 0xf;
+	}
+
+	return opcode;
+}
+
+/* Build iWARP terminate header */
+static int nes_bld_terminate_hdr(struct nes_qp *nesqp, u16 async_event_id, u32 aeq_info)
+{
+	u8 *pkt = nesqp->hwqp.q2_vbase + BAD_FRAME_OFFSET;
+	u16 ddp_seg_len;
+	int copy_len = 0;
+	u8 is_tagged = 0;
+	u8 flush_code = 0;
+	struct nes_terminate_hdr *termhdr;
+
+	termhdr = (struct nes_terminate_hdr *)nesqp->hwqp.q2_vbase;
+	memset(termhdr, 0, 64);
+
+	if (aeq_info & NES_AEQE_Q2_DATA_WRITTEN) {
+
+		/* Use data from offending packet to fill in ddp & rdma hdrs */
+		pkt = locate_mpa(pkt, aeq_info);
+		ddp_seg_len = be16_to_cpu(*(u16 *)pkt);
+		if (ddp_seg_len) {
+			copy_len = 2;
+			termhdr->hdrct = DDP_LEN_FLAG;
+			if (pkt[2] & 0x80) {
+				is_tagged = 1;
+				if (ddp_seg_len >= TERM_DDP_LEN_TAGGED) {
+					copy_len += TERM_DDP_LEN_TAGGED;
+					termhdr->hdrct |= DDP_HDR_FLAG;
+				}
+			} else {
+				if (ddp_seg_len >= TERM_DDP_LEN_UNTAGGED) {
+					copy_len += TERM_DDP_LEN_UNTAGGED;
+					termhdr->hdrct |= DDP_HDR_FLAG;
+				}
+
+				if (ddp_seg_len >= (TERM_DDP_LEN_UNTAGGED + TERM_RDMA_LEN)) {
+					if ((pkt[3] & RDMA_OPCODE_MASK) == RDMA_READ_REQ_OPCODE) {
+						copy_len += TERM_RDMA_LEN;
+						termhdr->hdrct |= RDMA_HDR_FLAG;
+					}
+				}
+			}
+		}
+	}
+
+	switch (async_event_id) {
+	case NES_AEQE_AEID_AMP_UNALLOCATED_STAG:
+		switch (iwarp_opcode(nesqp, aeq_info)) {
+		case IWARP_OPCODE_WRITE:
+			flush_code = IB_WC_LOC_PROT_ERR;
+			termhdr->layer_etype = (LAYER_DDP << 4) | DDP_TAGGED_BUFFER;
+			termhdr->error_code = DDP_TAGGED_INV_STAG;
+			break;
+		default:
+			flush_code = IB_WC_REM_ACCESS_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+			termhdr->error_code = RDMAP_INV_STAG;
+		}
+		break;
+	case NES_AEQE_AEID_AMP_INVALID_STAG:
+		flush_code = IB_WC_REM_ACCESS_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+		termhdr->error_code = RDMAP_INV_STAG;
+		break;
+	case NES_AEQE_AEID_AMP_BAD_QP:
+		flush_code = IB_WC_LOC_QP_OP_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_QN;
+		break;
+	case NES_AEQE_AEID_AMP_BAD_STAG_KEY:
+	case NES_AEQE_AEID_AMP_BAD_STAG_INDEX:
+		switch (iwarp_opcode(nesqp, aeq_info)) {
+		case IWARP_OPCODE_SEND_INV:
+		case IWARP_OPCODE_SEND_SE_INV:
+			flush_code = IB_WC_REM_OP_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_OP;
+			termhdr->error_code = RDMAP_CANT_INV_STAG;
+			break;
+		default:
+			flush_code = IB_WC_REM_ACCESS_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+			termhdr->error_code = RDMAP_INV_STAG;
+		}
+		break;
+	case NES_AEQE_AEID_AMP_BOUNDS_VIOLATION:
+		if (aeq_info & (NES_AEQE_Q2_DATA_ETHERNET | NES_AEQE_Q2_DATA_MPA)) {
+			flush_code = IB_WC_LOC_PROT_ERR;
+			termhdr->layer_etype = (LAYER_DDP << 4) | DDP_TAGGED_BUFFER;
+			termhdr->error_code = DDP_TAGGED_BOUNDS;
+		} else {
+			flush_code = IB_WC_REM_ACCESS_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+			termhdr->error_code = RDMAP_INV_BOUNDS;
+		}
+		break;
+	case NES_AEQE_AEID_AMP_RIGHTS_VIOLATION:
+	case NES_AEQE_AEID_AMP_INVALIDATE_NO_REMOTE_ACCESS_RIGHTS:
+	case NES_AEQE_AEID_PRIV_OPERATION_DENIED:
+		flush_code = IB_WC_REM_ACCESS_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+		termhdr->error_code = RDMAP_ACCESS;
+		break;
+	case NES_AEQE_AEID_AMP_TO_WRAP:
+		flush_code = IB_WC_REM_ACCESS_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+		termhdr->error_code = RDMAP_TO_WRAP;
+		break;
+	case NES_AEQE_AEID_AMP_BAD_PD:
+		switch (iwarp_opcode(nesqp, aeq_info)) {
+		case IWARP_OPCODE_WRITE:
+			flush_code = IB_WC_LOC_PROT_ERR;
+			termhdr->layer_etype = (LAYER_DDP << 4) | DDP_TAGGED_BUFFER;
+			termhdr->error_code = DDP_TAGGED_UNASSOC_STAG;
+			break;
+		case IWARP_OPCODE_SEND_INV:
+		case IWARP_OPCODE_SEND_SE_INV:
+			flush_code = IB_WC_REM_ACCESS_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+			termhdr->error_code = RDMAP_CANT_INV_STAG;
+			break;
+		default:
+			flush_code = IB_WC_REM_ACCESS_ERR;
+			termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_PROT;
+			termhdr->error_code = RDMAP_UNASSOC_STAG;
+		}
+		break;
+	case NES_AEQE_AEID_LLP_RECEIVED_MARKER_AND_LENGTH_FIELDS_DONT_MATCH:
+		flush_code = IB_WC_LOC_LEN_ERR;
+		termhdr->layer_etype = (LAYER_MPA << 4) | DDP_LLP;
+		termhdr->error_code = MPA_MARKER;
+		break;
+	case NES_AEQE_AEID_LLP_RECEIVED_MPA_CRC_ERROR:
+		flush_code = IB_WC_GENERAL_ERR;
+		termhdr->layer_etype = (LAYER_MPA << 4) | DDP_LLP;
+		termhdr->error_code = MPA_CRC;
+		break;
+	case NES_AEQE_AEID_LLP_SEGMENT_TOO_LARGE:
+	case NES_AEQE_AEID_LLP_SEGMENT_TOO_SMALL:
+		flush_code = IB_WC_LOC_LEN_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_CATASTROPHIC;
+		termhdr->error_code = DDP_CATASTROPHIC_LOCAL;
+		break;
+	case NES_AEQE_AEID_DDP_LCE_LOCAL_CATASTROPHIC:
+	case NES_AEQE_AEID_DDP_NO_L_BIT:
+		flush_code = IB_WC_FATAL_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_CATASTROPHIC;
+		termhdr->error_code = DDP_CATASTROPHIC_LOCAL;
+		break;
+	case NES_AEQE_AEID_DDP_INVALID_MSN_GAP_IN_MSN:
+	case NES_AEQE_AEID_DDP_INVALID_MSN_RANGE_IS_NOT_VALID:
+		flush_code = IB_WC_GENERAL_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_MSN_RANGE;
+		break;
+	case NES_AEQE_AEID_DDP_UBE_DDP_MESSAGE_TOO_LONG_FOR_AVAILABLE_BUFFER:
+		flush_code = IB_WC_LOC_LEN_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_TOO_LONG;
+		break;
+	case NES_AEQE_AEID_DDP_UBE_INVALID_DDP_VERSION:
+		flush_code = IB_WC_GENERAL_ERR;
+		if (is_tagged) {
+			termhdr->layer_etype = (LAYER_DDP << 4) | DDP_TAGGED_BUFFER;
+			termhdr->error_code = DDP_TAGGED_INV_DDP_VER;
+		} else {
+			termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+			termhdr->error_code = DDP_UNTAGGED_INV_DDP_VER;
+		}
+		break;
+	case NES_AEQE_AEID_DDP_UBE_INVALID_MO:
+		flush_code = IB_WC_GENERAL_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_MO;
+		break;
+	case NES_AEQE_AEID_DDP_UBE_INVALID_MSN_NO_BUFFER_AVAILABLE:
+		flush_code = IB_WC_REM_OP_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_MSN_NO_BUF;
+		break;
+	case NES_AEQE_AEID_DDP_UBE_INVALID_QN:
+		flush_code = IB_WC_GENERAL_ERR;
+		termhdr->layer_etype = (LAYER_DDP << 4) | DDP_UNTAGGED_BUFFER;
+		termhdr->error_code = DDP_UNTAGGED_INV_QN;
+		break;
+	case NES_AEQE_AEID_RDMAP_ROE_INVALID_RDMAP_VERSION:
+		flush_code = IB_WC_GENERAL_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_OP;
+		termhdr->error_code = RDMAP_INV_RDMAP_VER;
+		break;
+	case NES_AEQE_AEID_RDMAP_ROE_UNEXPECTED_OPCODE:
+		flush_code = IB_WC_LOC_QP_OP_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_OP;
+		termhdr->error_code = RDMAP_UNEXPECTED_OP;
+		break;
+	default:
+		flush_code = IB_WC_FATAL_ERR;
+		termhdr->layer_etype = (LAYER_RDMA << 4) | RDMAP_REMOTE_OP;
+		termhdr->error_code = RDMAP_UNSPECIFIED;
+		break;
+	}
+
+	if (copy_len)
+		memcpy(termhdr + 1, pkt, copy_len);
+
+	if ((flush_code) && ((NES_AEQE_INBOUND_RDMA & aeq_info) == 0)) {
+		if (aeq_info & NES_AEQE_SQ)
+			nesqp->term_sq_flush_code = flush_code;
+		else
+			nesqp->term_rq_flush_code = flush_code;
+	}
+
+	return sizeof(struct nes_terminate_hdr) + copy_len;
+}
+
+static void nes_terminate_connection(struct nes_device *nesdev, struct nes_qp *nesqp,
+		 struct nes_hw_aeqe *aeqe, enum ib_event_type eventtype)
+{
+	u64 context;
+	unsigned long flags;
+	u32 aeq_info;
+	u16 async_event_id;
+	u8 tcp_state;
+	u8 iwarp_state;
+	u32 termlen = 0;
+	u32 mod_qp_flags = NES_CQP_QP_IWARP_STATE_TERMINATE |
+			   NES_CQP_QP_TERM_DONT_SEND_FIN;
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
+
+	if (nesqp->term_flags & NES_TERM_SENT)
+		return; /* Sanity check */
+
+	aeq_info = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_MISC_IDX]);
+	tcp_state = (aeq_info & NES_AEQE_TCP_STATE_MASK) >> NES_AEQE_TCP_STATE_SHIFT;
+	iwarp_state = (aeq_info & NES_AEQE_IWARP_STATE_MASK) >> NES_AEQE_IWARP_STATE_SHIFT;
+	async_event_id = (u16)aeq_info;
+
+	context = (unsigned long)nesadapter->qp_table[le32_to_cpu(
+		aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]) - NES_FIRST_QPN];
+	if (!context) {
+		WARN_ON(!context);
+		return;
+	}
+
+	nesqp = (struct nes_qp *)(unsigned long)context;
+	spin_lock_irqsave(&nesqp->lock, flags);
+	nesqp->hw_iwarp_state = iwarp_state;
+	nesqp->hw_tcp_state = tcp_state;
+	nesqp->last_aeq = async_event_id;
+	nesqp->terminate_eventtype = eventtype;
+	spin_unlock_irqrestore(&nesqp->lock, flags);
+
+	if (nesadapter->send_term_ok)
+		termlen = nes_bld_terminate_hdr(nesqp, async_event_id, aeq_info);
+	else
+		mod_qp_flags |= NES_CQP_QP_TERM_DONT_SEND_TERM_MSG;
+
+	if (!nesdev->iw_status)  {
+		nesqp->term_flags = NES_TERM_DONE;
+		nes_hw_modify_qp(nesdev, nesqp, NES_CQP_QP_IWARP_STATE_ERROR, 0, 0);
+		nes_cm_disconn(nesqp);
+	} else {
+		nes_terminate_start_timer(nesqp);
+		nesqp->term_flags |= NES_TERM_SENT;
+		nes_hw_modify_qp(nesdev, nesqp, mod_qp_flags, termlen, 0);
+	}
+}
+
+static void nes_terminate_send_fin(struct nes_device *nesdev,
+			  struct nes_qp *nesqp, struct nes_hw_aeqe *aeqe)
+{
+	u32 aeq_info;
+	u16 async_event_id;
+	u8 tcp_state;
+	u8 iwarp_state;
+	unsigned long flags;
+
+	aeq_info = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_MISC_IDX]);
+	tcp_state = (aeq_info & NES_AEQE_TCP_STATE_MASK) >> NES_AEQE_TCP_STATE_SHIFT;
+	iwarp_state = (aeq_info & NES_AEQE_IWARP_STATE_MASK) >> NES_AEQE_IWARP_STATE_SHIFT;
+	async_event_id = (u16)aeq_info;
+
+	spin_lock_irqsave(&nesqp->lock, flags);
+	nesqp->hw_iwarp_state = iwarp_state;
+	nesqp->hw_tcp_state = tcp_state;
+	nesqp->last_aeq = async_event_id;
+	spin_unlock_irqrestore(&nesqp->lock, flags);
+
+	/* Send the fin only */
+	nes_hw_modify_qp(nesdev, nesqp, NES_CQP_QP_IWARP_STATE_TERMINATE |
+		NES_CQP_QP_TERM_DONT_SEND_TERM_MSG, 0, 0);
+}
+
+/* Cleanup after a terminate sent or received */
+static void nes_terminate_done(struct nes_qp *nesqp, int timeout_occurred)
+{
+	u32 next_iwarp_state = NES_CQP_QP_IWARP_STATE_ERROR;
+	unsigned long flags;
+	struct nes_vnic *nesvnic = to_nesvnic(nesqp->ibqp.device);
+	struct nes_device *nesdev = nesvnic->nesdev;
+	u8 first_time = 0;
+
+	spin_lock_irqsave(&nesqp->lock, flags);
+	if (nesqp->hte_added) {
+		nesqp->hte_added = 0;
+		next_iwarp_state |= NES_CQP_QP_DEL_HTE;
+	}
+
+	first_time = (nesqp->term_flags & NES_TERM_DONE) == 0;
+	nesqp->term_flags |= NES_TERM_DONE;
+	spin_unlock_irqrestore(&nesqp->lock, flags);
+
+	/* Make sure we go through this only once */
+	if (first_time) {
+		if (timeout_occurred == 0)
+			del_timer(&nesqp->terminate_timer);
+		else
+			next_iwarp_state |= NES_CQP_QP_RESET;
+
+		nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0, 0);
+		nes_cm_disconn(nesqp);
+	}
+}
+
+static void nes_terminate_received(struct nes_device *nesdev,
+				struct nes_qp *nesqp, struct nes_hw_aeqe *aeqe)
+{
+	u32 aeq_info;
+	u8 *pkt;
+	u32 *mpa;
+	u8 ddp_ctl;
+	u8 rdma_ctl;
+	u16 aeq_id = 0;
+
+	aeq_info = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_MISC_IDX]);
+	if (aeq_info & NES_AEQE_Q2_DATA_WRITTEN) {
+		/* Terminate is not a performance path so the silicon */
+		/* did not validate the frame - do it now */
+		pkt = nesqp->hwqp.q2_vbase + BAD_FRAME_OFFSET;
+		mpa = (u32 *)locate_mpa(pkt, aeq_info);
+		ddp_ctl = (be32_to_cpu(mpa[0]) >> 8) & 0xff;
+		rdma_ctl = be32_to_cpu(mpa[0]) & 0xff;
+		if ((ddp_ctl & 0xc0) != 0x40)
+			aeq_id = NES_AEQE_AEID_DDP_LCE_LOCAL_CATASTROPHIC;
+		else if ((ddp_ctl & 0x03) != 1)
+			aeq_id = NES_AEQE_AEID_DDP_UBE_INVALID_DDP_VERSION;
+		else if (be32_to_cpu(mpa[2]) != 2)
+			aeq_id = NES_AEQE_AEID_DDP_UBE_INVALID_QN;
+		else if (be32_to_cpu(mpa[3]) != 1)
+			aeq_id = NES_AEQE_AEID_DDP_INVALID_MSN_GAP_IN_MSN;
+		else if (be32_to_cpu(mpa[4]) != 0)
+			aeq_id = NES_AEQE_AEID_DDP_UBE_INVALID_MO;
+		else if ((rdma_ctl & 0xc0) != 0x40)
+			aeq_id = NES_AEQE_AEID_RDMAP_ROE_INVALID_RDMAP_VERSION;
+
+		if (aeq_id) {
+			/* Bad terminate recvd - send back a terminate */
+			aeq_info = (aeq_info & 0xffff0000) | aeq_id;
+			aeqe->aeqe_words[NES_AEQE_MISC_IDX] = cpu_to_le32(aeq_info);
+			nes_terminate_connection(nesdev, nesqp, aeqe, IB_EVENT_QP_FATAL);
+			return;
+		}
+	}
+
+	nesqp->term_flags |= NES_TERM_RCVD;
+	nesqp->terminate_eventtype = IB_EVENT_QP_FATAL;
+	nes_terminate_start_timer(nesqp);
+	nes_terminate_send_fin(nesdev, nesqp, aeqe);
+}
+
+/* Timeout routine in case terminate fails to complete */
+void nes_terminate_timeout(unsigned long context)
+{
+	struct nes_qp *nesqp = (struct nes_qp *)(unsigned long)context;
+
+	nes_terminate_done(nesqp, 1);
+}
+
+/* Set a timer in case hw cannot complete the terminate sequence */
+static void nes_terminate_start_timer(struct nes_qp *nesqp)
+{
+	mod_timer(&nesqp->terminate_timer, (jiffies + HZ));
+}
 
 /**
  * nes_process_iwarp_aeqe
@@ -2773,33 +3539,33 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 				   struct nes_hw_aeqe *aeqe)
 {
 	u64 context;
-	u64 aeqe_context = 0;
 	unsigned long flags;
 	struct nes_qp *nesqp;
+	struct nes_hw_cq *hw_cq;
+	struct nes_cq *nescq;
 	int resource_allocated;
-	/* struct iw_cm_id *cm_id; */
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
-	struct ib_event ibevent;
-	/* struct iw_cm_event cm_event; */
 	u32 aeq_info;
 	u32 next_iwarp_state = 0;
+	u32 aeqe_cq_id;
 	u16 async_event_id;
 	u8 tcp_state;
 	u8 iwarp_state;
+	struct ib_event ibevent;
 
 	nes_debug(NES_DBG_AEQ, "\n");
 	aeq_info = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_MISC_IDX]);
-	if ((NES_AEQE_INBOUND_RDMA&aeq_info) || (!(NES_AEQE_QP&aeq_info))) {
+	if ((NES_AEQE_INBOUND_RDMA & aeq_info) || (!(NES_AEQE_QP & aeq_info))) {
 		context  = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_CTXT_LOW_IDX]);
 		context += ((u64)le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_CTXT_HIGH_IDX])) << 32;
 	} else {
-		aeqe_context = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_CTXT_LOW_IDX]);
-		aeqe_context += ((u64)le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_CTXT_HIGH_IDX])) << 32;
 		context = (unsigned long)nesadapter->qp_table[le32_to_cpu(
 						aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]) - NES_FIRST_QPN];
 		BUG_ON(!context);
 	}
 
+	/* context is nesqp unless async_event_id == CQ ERROR */
+	nesqp = (struct nes_qp *)(unsigned long)context;
 	async_event_id = (u16)aeq_info;
 	tcp_state = (aeq_info & NES_AEQE_TCP_STATE_MASK) >> NES_AEQE_TCP_STATE_SHIFT;
 	iwarp_state = (aeq_info & NES_AEQE_IWARP_STATE_MASK) >> NES_AEQE_IWARP_STATE_SHIFT;
@@ -2809,12 +3575,33 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]), aeqe,
 			nes_tcp_state_str[tcp_state], nes_iwarp_state_str[iwarp_state]);
 
+	aeqe_cq_id = le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]);
+	if (aeq_info & NES_AEQE_QP) {
+		if (!nes_is_resource_allocated(nesadapter,
+				nesadapter->allocated_qps,
+				aeqe_cq_id))
+			return;
+	}
+
 	switch (async_event_id) {
 		case NES_AEQE_AEID_LLP_FIN_RECEIVED:
-			nesqp = *((struct nes_qp **)&context);
+			if (nesqp->term_flags)
+				return; /* Ignore it, wait for close complete */
+
 			if (atomic_inc_return(&nesqp->close_timer_started) == 1) {
+				if ((tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) &&
+					(nesqp->ibqp_state == IB_QPS_RTS)) {
+					spin_lock_irqsave(&nesqp->lock, flags);
+					nesqp->hw_iwarp_state = iwarp_state;
+					nesqp->hw_tcp_state = tcp_state;
+					nesqp->last_aeq = async_event_id;
+					next_iwarp_state = NES_CQP_QP_IWARP_STATE_CLOSING;
+					nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_CLOSING;
+					spin_unlock_irqrestore(&nesqp->lock, flags);
+					nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0, 0);
+					nes_cm_disconn(nesqp);
+				}
 				nesqp->cm_id->add_ref(nesqp->cm_id);
-				nes_add_ref(&nesqp->ibqp);
 				schedule_nes_timer(nesqp->cm_node, (struct sk_buff *)nesqp,
 						NES_TIMER_TYPE_CLOSE, 1, 0);
 				nes_debug(NES_DBG_AEQ, "QP%u Not decrementing QP refcount (%d),"
@@ -2823,196 +3610,115 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 						nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 						async_event_id, nesqp->last_aeq, tcp_state);
 			}
-			if ((tcp_state != NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
-					(nesqp->ibqp_state != IB_QPS_RTS)) {
-				/* FIN Received but tcp state or IB state moved on,
-						should expect a	close complete */
-				return;
-			}
-		case NES_AEQE_AEID_LLP_CLOSE_COMPLETE:
-		case NES_AEQE_AEID_LLP_CONNECTION_RESET:
-		case NES_AEQE_AEID_TERMINATE_SENT:
-		case NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE:
-		case NES_AEQE_AEID_RESET_SENT:
-			nesqp = *((struct nes_qp **)&context);
-			if (async_event_id == NES_AEQE_AEID_RESET_SENT) {
-				tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-			}
-			nes_add_ref(&nesqp->ibqp);
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = iwarp_state;
-			nesqp->hw_tcp_state = tcp_state;
-			nesqp->last_aeq = async_event_id;
-
-			if ((tcp_state == NES_AEQE_TCP_STATE_CLOSED) ||
-					(tcp_state == NES_AEQE_TCP_STATE_TIME_WAIT)) {
-				nesqp->hte_added = 0;
-				spin_unlock_irqrestore(&nesqp->lock, flags);
-				nes_debug(NES_DBG_AEQ, "issuing hw modifyqp for QP%u to remove hte\n",
-						nesqp->hwqp.qp_id);
-				nes_hw_modify_qp(nesdev, nesqp,
-						NES_CQP_QP_IWARP_STATE_ERROR | NES_CQP_QP_DEL_HTE, 0);
-				spin_lock_irqsave(&nesqp->lock, flags);
-			}
-
-			if ((nesqp->ibqp_state == IB_QPS_RTS) &&
-					((tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
-					(async_event_id == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
-				switch (nesqp->hw_iwarp_state) {
-					case NES_AEQE_IWARP_STATE_RTS:
-						next_iwarp_state = NES_CQP_QP_IWARP_STATE_CLOSING;
-						nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_CLOSING;
-						break;
-					case NES_AEQE_IWARP_STATE_TERMINATE:
-						next_iwarp_state = NES_CQP_QP_IWARP_STATE_TERMINATE;
-						nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_TERMINATE;
-						if (async_event_id == NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) {
-							next_iwarp_state |= 0x02000000;
-							nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-						}
-						break;
-					default:
-						next_iwarp_state = 0;
-				}
-				spin_unlock_irqrestore(&nesqp->lock, flags);
-				if (next_iwarp_state) {
-					nes_add_ref(&nesqp->ibqp);
-					nes_debug(NES_DBG_AEQ, "issuing hw modifyqp for QP%u. next state = 0x%08X,"
-							" also added another reference\n",
-							nesqp->hwqp.qp_id, next_iwarp_state);
-					nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0);
-				}
-				nes_cm_disconn(nesqp);
-			} else {
-				if (async_event_id ==  NES_AEQE_AEID_LLP_FIN_RECEIVED) {
-					/* FIN Received but ib state not RTS,
-							close complete will be on its way */
-					spin_unlock_irqrestore(&nesqp->lock, flags);
-					nes_rem_ref(&nesqp->ibqp);
-					return;
-				}
-				spin_unlock_irqrestore(&nesqp->lock, flags);
-				if (async_event_id == NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) {
-					next_iwarp_state = NES_CQP_QP_IWARP_STATE_TERMINATE | 0x02000000;
-					nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-					nes_debug(NES_DBG_AEQ, "issuing hw modifyqp for QP%u. next state = 0x%08X,"
-							" also added another reference\n",
-							nesqp->hwqp.qp_id, next_iwarp_state);
-					nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0);
-				}
-				nes_cm_disconn(nesqp);
-			}
 			break;
-		case NES_AEQE_AEID_LLP_TERMINATE_RECEIVED:
-			nesqp = *((struct nes_qp **)&context);
+		case NES_AEQE_AEID_LLP_CLOSE_COMPLETE:
 			spin_lock_irqsave(&nesqp->lock, flags);
 			nesqp->hw_iwarp_state = iwarp_state;
 			nesqp->hw_tcp_state = tcp_state;
 			nesqp->last_aeq = async_event_id;
 			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_LLP_TERMINATE_RECEIVED"
-					" event on QP%u \n  Q2 Data:\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_FATAL;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
-			if ((tcp_state == NES_AEQE_TCP_STATE_CLOSE_WAIT) ||
-					((nesqp->ibqp_state == IB_QPS_RTS)&&
-					(async_event_id == NES_AEQE_AEID_LLP_CONNECTION_RESET))) {
-				nes_add_ref(&nesqp->ibqp);
-				nes_cm_disconn(nesqp);
-			} else {
-				nesqp->in_disconnect = 0;
-				wake_up(&nesqp->kick_waitq);
-			}
-			break;
-		case NES_AEQE_AEID_LLP_TOO_MANY_RETRIES:
-			nesqp = *((struct nes_qp **)&context);
-			nes_add_ref(&nesqp->ibqp);
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_ERROR;
-			nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
-			nesqp->last_aeq = async_event_id;
-			if (nesqp->cm_id) {
-				nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_LLP_TOO_MANY_RETRIES"
-						" event on QP%u, remote IP = 0x%08X \n",
-						nesqp->hwqp.qp_id,
-						ntohl(nesqp->cm_id->remote_addr.sin_addr.s_addr));
-			} else {
-				nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_LLP_TOO_MANY_RETRIES"
-						" event on QP%u \n",
-						nesqp->hwqp.qp_id);
-			}
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			next_iwarp_state = NES_CQP_QP_IWARP_STATE_ERROR | NES_CQP_QP_RESET;
-			nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_FATAL;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
+			nes_cm_disconn(nesqp);
 			break;
-		case NES_AEQE_AEID_AMP_BAD_STAG_INDEX:
-			if (NES_AEQE_INBOUND_RDMA&aeq_info) {
-				nesqp = nesadapter->qp_table[le32_to_cpu(
-						aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX])-NES_FIRST_QPN];
-			} else {
-				/* TODO: get the actual WQE and mask off wqe index */
-				context &= ~((u64)511);
-				nesqp = *((struct nes_qp **)&context);
-			}
+
+		case NES_AEQE_AEID_RESET_SENT:
+			tcp_state = NES_AEQE_TCP_STATE_CLOSED;
 			spin_lock_irqsave(&nesqp->lock, flags);
 			nesqp->hw_iwarp_state = iwarp_state;
 			nesqp->hw_tcp_state = tcp_state;
 			nesqp->last_aeq = async_event_id;
+			nesqp->hte_added = 0;
 			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_AMP_BAD_STAG_INDEX event on QP%u\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_ACCESS_ERR;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
+			next_iwarp_state = NES_CQP_QP_IWARP_STATE_ERROR | NES_CQP_QP_DEL_HTE;
+			nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0, 0);
+			nes_cm_disconn(nesqp);
 			break;
-		case NES_AEQE_AEID_AMP_UNALLOCATED_STAG:
-			nesqp = *((struct nes_qp **)&context);
+
+		case NES_AEQE_AEID_LLP_CONNECTION_RESET:
+			if (atomic_read(&nesqp->close_timer_started))
+				return;
 			spin_lock_irqsave(&nesqp->lock, flags);
 			nesqp->hw_iwarp_state = iwarp_state;
 			nesqp->hw_tcp_state = tcp_state;
 			nesqp->last_aeq = async_event_id;
 			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_AMP_UNALLOCATED_STAG event on QP%u\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_ACCESS_ERR;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
+			nes_cm_disconn(nesqp);
+			break;
+
+		case NES_AEQE_AEID_TERMINATE_SENT:
+			nes_terminate_send_fin(nesdev, nesqp, aeqe);
 			break;
+
+		case NES_AEQE_AEID_LLP_TERMINATE_RECEIVED:
+			nes_terminate_received(nesdev, nesqp, aeqe);
+			break;
+
+		case NES_AEQE_AEID_AMP_BAD_STAG_KEY:
+		case NES_AEQE_AEID_AMP_BAD_STAG_INDEX:
+		case NES_AEQE_AEID_AMP_UNALLOCATED_STAG:
+		case NES_AEQE_AEID_AMP_INVALID_STAG:
+		case NES_AEQE_AEID_AMP_RIGHTS_VIOLATION:
+		case NES_AEQE_AEID_AMP_INVALIDATE_NO_REMOTE_ACCESS_RIGHTS:
 		case NES_AEQE_AEID_PRIV_OPERATION_DENIED:
-			nesqp = nesadapter->qp_table[le32_to_cpu(aeqe->aeqe_words
-					[NES_AEQE_COMP_QP_CQ_ID_IDX])-NES_FIRST_QPN];
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = iwarp_state;
-			nesqp->hw_tcp_state = tcp_state;
-			nesqp->last_aeq = async_event_id;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_PRIV_OPERATION_DENIED event on QP%u,"
-					" nesqp = %p, AE reported %p\n",
-					nesqp->hwqp.qp_id, nesqp, *((struct nes_qp **)&context));
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_ACCESS_ERR;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
+		case NES_AEQE_AEID_DDP_UBE_DDP_MESSAGE_TOO_LONG_FOR_AVAILABLE_BUFFER:
+		case NES_AEQE_AEID_AMP_BOUNDS_VIOLATION:
+		case NES_AEQE_AEID_AMP_TO_WRAP:
+			printk(KERN_ERR PFX "QP[%u] async_event_id=0x%04X IB_EVENT_QP_ACCESS_ERR\n",
+					nesqp->hwqp.qp_id, async_event_id);
+			nes_terminate_connection(nesdev, nesqp, aeqe, IB_EVENT_QP_ACCESS_ERR);
+			break;
+
+		case NES_AEQE_AEID_LLP_SEGMENT_TOO_LARGE:
+		case NES_AEQE_AEID_LLP_SEGMENT_TOO_SMALL:
+		case NES_AEQE_AEID_DDP_UBE_INVALID_MO:
+		case NES_AEQE_AEID_DDP_UBE_INVALID_QN:
+			if (iwarp_opcode(nesqp, aeq_info) > IWARP_OPCODE_TERM) {
+				aeq_info &= 0xffff0000;
+				aeq_info |= NES_AEQE_AEID_RDMAP_ROE_UNEXPECTED_OPCODE;
+				aeqe->aeqe_words[NES_AEQE_MISC_IDX] = cpu_to_le32(aeq_info);
 			}
+
+		case NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE:
+		case NES_AEQE_AEID_LLP_TOO_MANY_RETRIES:
+		case NES_AEQE_AEID_DDP_UBE_INVALID_MSN_NO_BUFFER_AVAILABLE:
+		case NES_AEQE_AEID_LLP_RECEIVED_MPA_CRC_ERROR:
+		case NES_AEQE_AEID_AMP_BAD_QP:
+		case NES_AEQE_AEID_LLP_RECEIVED_MARKER_AND_LENGTH_FIELDS_DONT_MATCH:
+		case NES_AEQE_AEID_DDP_LCE_LOCAL_CATASTROPHIC:
+		case NES_AEQE_AEID_DDP_NO_L_BIT:
+		case NES_AEQE_AEID_DDP_INVALID_MSN_GAP_IN_MSN:
+		case NES_AEQE_AEID_DDP_INVALID_MSN_RANGE_IS_NOT_VALID:
+		case NES_AEQE_AEID_DDP_UBE_INVALID_DDP_VERSION:
+		case NES_AEQE_AEID_RDMAP_ROE_INVALID_RDMAP_VERSION:
+		case NES_AEQE_AEID_RDMAP_ROE_UNEXPECTED_OPCODE:
+		case NES_AEQE_AEID_AMP_BAD_PD:
+		case NES_AEQE_AEID_AMP_FASTREG_SHARED:
+		case NES_AEQE_AEID_AMP_FASTREG_VALID_STAG:
+		case NES_AEQE_AEID_AMP_FASTREG_MW_STAG:
+		case NES_AEQE_AEID_AMP_FASTREG_INVALID_RIGHTS:
+		case NES_AEQE_AEID_AMP_FASTREG_PBL_TABLE_OVERFLOW:
+		case NES_AEQE_AEID_AMP_FASTREG_INVALID_LENGTH:
+		case NES_AEQE_AEID_AMP_INVALIDATE_SHARED:
+		case NES_AEQE_AEID_AMP_INVALIDATE_MR_WITH_BOUND_WINDOWS:
+		case NES_AEQE_AEID_AMP_MWBIND_VALID_STAG:
+		case NES_AEQE_AEID_AMP_MWBIND_OF_MR_STAG:
+		case NES_AEQE_AEID_AMP_MWBIND_TO_ZERO_BASED_STAG:
+		case NES_AEQE_AEID_AMP_MWBIND_TO_MW_STAG:
+		case NES_AEQE_AEID_AMP_MWBIND_INVALID_RIGHTS:
+		case NES_AEQE_AEID_AMP_MWBIND_INVALID_BOUNDS:
+		case NES_AEQE_AEID_AMP_MWBIND_TO_INVALID_PARENT:
+		case NES_AEQE_AEID_AMP_MWBIND_BIND_DISABLED:
+		case NES_AEQE_AEID_BAD_CLOSE:
+		case NES_AEQE_AEID_RDMA_READ_WHILE_ORD_ZERO:
+		case NES_AEQE_AEID_STAG_ZERO_INVALID:
+		case NES_AEQE_AEID_ROE_INVALID_RDMA_READ_REQUEST:
+		case NES_AEQE_AEID_ROE_INVALID_RDMA_WRITE_OR_READ_RESP:
+			printk(KERN_ERR PFX "QP[%u] async_event_id=0x%04X IB_EVENT_QP_FATAL\n",
+					nesqp->hwqp.qp_id, async_event_id);
+			print_ip(nesqp->cm_node);
+			if (!atomic_read(&nesqp->close_timer_started))
+				nes_terminate_connection(nesdev, nesqp, aeqe, IB_EVENT_QP_FATAL);
 			break;
+
 		case NES_AEQE_AEID_CQ_OPERATION_ERROR:
 			context <<= 1;
 			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_CQ_OPERATION_ERROR event on CQ%u, %p\n",
@@ -3022,70 +3728,19 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 			if (resource_allocated) {
 				printk(KERN_ERR PFX "%s: Processing an NES_AEQE_AEID_CQ_OPERATION_ERROR event on CQ%u\n",
 						__func__, le32_to_cpu(aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX]));
+				hw_cq = (struct nes_hw_cq *)(unsigned long)context;
+				if (hw_cq) {
+					nescq = container_of(hw_cq, struct nes_cq, hw_cq);
+					if (nescq->ibcq.event_handler) {
+						ibevent.device = nescq->ibcq.device;
+						ibevent.event = IB_EVENT_CQ_ERR;
+						ibevent.element.cq = &nescq->ibcq;
+						nescq->ibcq.event_handler(&ibevent, nescq->ibcq.cq_context);
+					}
+				}
 			}
 			break;
-		case NES_AEQE_AEID_DDP_UBE_DDP_MESSAGE_TOO_LONG_FOR_AVAILABLE_BUFFER:
-			nesqp = nesadapter->qp_table[le32_to_cpu(
-					aeqe->aeqe_words[NES_AEQE_COMP_QP_CQ_ID_IDX])-NES_FIRST_QPN];
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = iwarp_state;
-			nesqp->hw_tcp_state = tcp_state;
-			nesqp->last_aeq = async_event_id;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_DDP_UBE_DDP_MESSAGE_TOO_LONG"
-					"_FOR_AVAILABLE_BUFFER event on QP%u\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_ACCESS_ERR;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
-			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
-			nes_cm_disconn(nesqp);
-			break;
-		case NES_AEQE_AEID_DDP_UBE_INVALID_MSN_NO_BUFFER_AVAILABLE:
-			nesqp = *((struct nes_qp **)&context);
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = iwarp_state;
-			nesqp->hw_tcp_state = tcp_state;
-			nesqp->last_aeq = async_event_id;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_DDP_UBE_INVALID_MSN"
-					"_NO_BUFFER_AVAILABLE event on QP%u\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_FATAL;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
-			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
-			nes_cm_disconn(nesqp);
-			break;
-		case NES_AEQE_AEID_LLP_RECEIVED_MPA_CRC_ERROR:
-			nesqp = *((struct nes_qp **)&context);
-			spin_lock_irqsave(&nesqp->lock, flags);
-			nesqp->hw_iwarp_state = iwarp_state;
-			nesqp->hw_tcp_state = tcp_state;
-			nesqp->last_aeq = async_event_id;
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-			nes_debug(NES_DBG_AEQ, "Processing an NES_AEQE_AEID_LLP_RECEIVED_MPA_CRC_ERROR"
-					" event on QP%u \n  Q2 Data:\n",
-					nesqp->hwqp.qp_id);
-			if (nesqp->ibqp.event_handler) {
-				ibevent.device = nesqp->ibqp.device;
-				ibevent.element.qp = &nesqp->ibqp;
-				ibevent.event = IB_EVENT_QP_FATAL;
-				nesqp->ibqp.event_handler(&ibevent, nesqp->ibqp.qp_context);
-			}
-			/* tell cm to disconnect, cm will queue work to thread */
-			nes_add_ref(&nesqp->ibqp);
-			nes_cm_disconn(nesqp);
-			break;
-			/* TODO: additional AEs need to be here */
+
 		default:
 			nes_debug(NES_DBG_AEQ, "Processing an iWARP related AE for QP, misc = 0x%04X\n",
 					async_event_id);
@@ -3094,7 +3749,6 @@ static void nes_process_iwarp_aeqe(struct nes_device *nesdev,
 
 }
 
-
 /**
  * nes_iwarp_ce_handler
  */
@@ -3229,6 +3883,8 @@ void flush_wqes(struct nes_device *nesdev, struct nes_qp *nesqp,
 {
 	struct nes_cqp_request *cqp_request;
 	struct nes_hw_cqp_wqe *cqp_wqe;
+	u32 sq_code = (NES_IWARP_CQE_MAJOR_FLUSH << 16) | NES_IWARP_CQE_MINOR_FLUSH;
+	u32 rq_code = (NES_IWARP_CQE_MAJOR_FLUSH << 16) | NES_IWARP_CQE_MINOR_FLUSH;
 	int ret;
 
 	cqp_request = nes_get_cqp_request(nesdev);
@@ -3245,6 +3901,24 @@ void flush_wqes(struct nes_device *nesdev, struct nes_qp *nesqp,
 	cqp_wqe = &cqp_request->cqp_wqe;
 	nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
 
+	/* If wqe in error was identified, set code to be put into cqe */
+	if ((nesqp->term_sq_flush_code) && (which_wq & NES_CQP_FLUSH_SQ)) {
+		which_wq |= NES_CQP_FLUSH_MAJ_MIN;
+		sq_code = (CQE_MAJOR_DRV << 16) | nesqp->term_sq_flush_code;
+		nesqp->term_sq_flush_code = 0;
+	}
+
+	if ((nesqp->term_rq_flush_code) && (which_wq & NES_CQP_FLUSH_RQ)) {
+		which_wq |= NES_CQP_FLUSH_MAJ_MIN;
+		rq_code = (CQE_MAJOR_DRV << 16) | nesqp->term_rq_flush_code;
+		nesqp->term_rq_flush_code = 0;
+	}
+
+	if (which_wq & NES_CQP_FLUSH_MAJ_MIN) {
+		cqp_wqe->wqe_words[NES_CQP_QP_WQE_FLUSH_SQ_CODE] = cpu_to_le32(sq_code);
+		cqp_wqe->wqe_words[NES_CQP_QP_WQE_FLUSH_RQ_CODE] = cpu_to_le32(rq_code);
+	}
+
 	cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] =
 			cpu_to_le32(NES_CQP_FLUSH_WQES | which_wq);
 	cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX] = cpu_to_le32(nesqp->hwqp.qp_id);
diff --git a/drivers/infiniband/hw/nes/nes_hw.h b/drivers/infiniband/hw/nes/nes_hw.h
index 7b81e0ae007..d748e4b31b8 100644
--- a/drivers/infiniband/hw/nes/nes_hw.h
+++ b/drivers/infiniband/hw/nes/nes_hw.h
@@ -1,5 +1,5 @@
 /*
-* Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+* Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
 *
 * This software is available to you under a choice of one of two
 * licenses.  You may choose to be licensed under the terms of the GNU
@@ -35,13 +35,22 @@
 
 #include <linux/inet_lro.h>
 
+#define NES_PHY_TYPE_CX4       1
 #define NES_PHY_TYPE_1G        2
-#define NES_PHY_TYPE_IRIS      3
 #define NES_PHY_TYPE_ARGUS     4
 #define NES_PHY_TYPE_PUMA_1G   5
 #define NES_PHY_TYPE_PUMA_10G  6
+#define NES_PHY_TYPE_GLADIUS   7
+#define NES_PHY_TYPE_SFP_D     8
+#define NES_PHY_TYPE_KR	       9
 
 #define NES_MULTICAST_PF_MAX 8
+#define NES_A0 3
+
+#define NES_ENABLE_PAU 0x07000001
+#define NES_DISABLE_PAU 0x07000000
+#define NES_PAU_COUNTER 10
+#define NES_CQP_OPCODE_MASK 0x3f
 
 enum pci_regs {
 	NES_INT_STAT = 0x0000,
@@ -60,6 +69,7 @@ enum pci_regs {
 	NES_CQ_ACK = 0x0034,
 	NES_WQE_ALLOC = 0x0040,
 	NES_CQE_ALLOC = 0x0044,
+	NES_AEQ_ALLOC = 0x0048
 };
 
 enum indexed_regs {
@@ -68,8 +78,10 @@ enum indexed_regs {
 	NES_IDX_QP_CONTROL = 0x0040,
 	NES_IDX_FLM_CONTROL = 0x0080,
 	NES_IDX_INT_CPU_STATUS = 0x00a0,
+	NES_IDX_GPR_TRIGGER = 0x00bc,
 	NES_IDX_GPIO_CONTROL = 0x00f0,
 	NES_IDX_GPIO_DATA = 0x00f4,
+	NES_IDX_GPR2 = 0x010c,
 	NES_IDX_TCP_CONFIG0 = 0x01e4,
 	NES_IDX_TCP_TIMER_CONFIG = 0x01ec,
 	NES_IDX_TCP_NOW = 0x01f0,
@@ -156,6 +168,8 @@ enum indexed_regs {
 	NES_IDX_ENDNODE0_NSTAT_TX_OCTETS_HI = 0x7004,
 	NES_IDX_ENDNODE0_NSTAT_TX_FRAMES_LO = 0x7008,
 	NES_IDX_ENDNODE0_NSTAT_TX_FRAMES_HI = 0x700c,
+	NES_IDX_WQM_CONFIG0 = 0x5000,
+	NES_IDX_WQM_CONFIG1 = 0x5004,
 	NES_IDX_CM_CONFIG = 0x5100,
 	NES_IDX_NIC_LOGPORT_TO_PHYPORT = 0x6000,
 	NES_IDX_NIC_PHYPORT_TO_USW = 0x6008,
@@ -195,6 +209,7 @@ enum nes_cqp_opcodes {
 	NES_CQP_REGISTER_SHARED_STAG = 0x0c,
 	NES_CQP_DEALLOCATE_STAG = 0x0d,
 	NES_CQP_MANAGE_ARP_CACHE = 0x0f,
+	NES_CQP_DOWNLOAD_SEGMENT = 0x10,
 	NES_CQP_SUSPEND_QPS = 0x11,
 	NES_CQP_UPLOAD_CONTEXT = 0x13,
 	NES_CQP_CREATE_CEQ = 0x16,
@@ -203,7 +218,8 @@ enum nes_cqp_opcodes {
 	NES_CQP_DESTROY_AEQ = 0x1b,
 	NES_CQP_LMI_ACCESS = 0x20,
 	NES_CQP_FLUSH_WQES = 0x22,
-	NES_CQP_MANAGE_APBVT = 0x23
+	NES_CQP_MANAGE_APBVT = 0x23,
+	NES_CQP_MANAGE_QUAD_HASH = 0x25
 };
 
 enum nes_cqp_wqe_word_idx {
@@ -215,6 +231,14 @@ enum nes_cqp_wqe_word_idx {
 	NES_CQP_WQE_COMP_SCRATCH_HIGH_IDX = 5,
 };
 
+enum nes_cqp_wqe_word_download_idx { /* format differs from other cqp ops */
+	NES_CQP_WQE_DL_OPCODE_IDX = 0,
+	NES_CQP_WQE_DL_COMP_CTX_LOW_IDX = 1,
+	NES_CQP_WQE_DL_COMP_CTX_HIGH_IDX = 2,
+	NES_CQP_WQE_DL_LENGTH_0_TOTAL_IDX = 3
+	/* For index values 4-15 use NES_NIC_SQ_WQE_ values */
+};
+
 enum nes_cqp_cq_wqeword_idx {
 	NES_CQP_CQ_WQE_PBL_LOW_IDX = 6,
 	NES_CQP_CQ_WQE_PBL_HIGH_IDX = 7,
@@ -235,7 +259,9 @@ enum nes_cqp_stag_wqeword_idx {
 	NES_CQP_STAG_WQE_PBL_LEN_IDX = 14
 };
 
+#define NES_CQP_OP_LOGICAL_PORT_SHIFT 26
 #define NES_CQP_OP_IWARP_STATE_SHIFT 28
+#define NES_CQP_OP_TERMLEN_SHIFT     28
 
 enum nes_cqp_qp_bits {
 	NES_CQP_QP_ARP_VALID = (1<<8),
@@ -260,12 +286,16 @@ enum nes_cqp_qp_bits {
 	NES_CQP_QP_IWARP_STATE_TERMINATE = (5<<NES_CQP_OP_IWARP_STATE_SHIFT),
 	NES_CQP_QP_IWARP_STATE_ERROR = (6<<NES_CQP_OP_IWARP_STATE_SHIFT),
 	NES_CQP_QP_IWARP_STATE_MASK = (7<<NES_CQP_OP_IWARP_STATE_SHIFT),
+	NES_CQP_QP_TERM_DONT_SEND_FIN = (1<<24),
+	NES_CQP_QP_TERM_DONT_SEND_TERM_MSG = (1<<25),
 	NES_CQP_QP_RESET = (1<<31),
 };
 
 enum nes_cqp_qp_wqe_word_idx {
 	NES_CQP_QP_WQE_CONTEXT_LOW_IDX = 6,
 	NES_CQP_QP_WQE_CONTEXT_HIGH_IDX = 7,
+	NES_CQP_QP_WQE_FLUSH_SQ_CODE = 8,
+	NES_CQP_QP_WQE_FLUSH_RQ_CODE = 9,
 	NES_CQP_QP_WQE_NEW_MSS_IDX = 15,
 };
 
@@ -356,6 +386,7 @@ enum nes_cqp_arp_bits {
 enum nes_cqp_flush_bits {
 	NES_CQP_FLUSH_SQ = (1<<30),
 	NES_CQP_FLUSH_RQ = (1<<31),
+	NES_CQP_FLUSH_MAJ_MIN = (1<<28),
 };
 
 enum nes_cqe_opcode_bits {
@@ -535,11 +566,23 @@ enum nes_iwarp_sq_fmr_wqe_word_idx {
 	NES_IWARP_SQ_FMR_WQE_PBL_LENGTH_IDX = 14,
 };
 
+enum nes_iwarp_sq_fmr_opcodes {
+	NES_IWARP_SQ_FMR_WQE_ZERO_BASED			= (1<<6),
+	NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_4K		= (0<<7),
+	NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_2M		= (1<<7),
+	NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_LOCAL_READ	= (1<<16),
+	NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_LOCAL_WRITE 	= (1<<17),
+	NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_REMOTE_READ 	= (1<<18),
+	NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_REMOTE_WRITE = (1<<19),
+	NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_WINDOW_BIND 	= (1<<20),
+};
+
+#define NES_IWARP_SQ_FMR_WQE_MR_LENGTH_HIGH_MASK	0xFF;
+
 enum nes_iwarp_sq_locinv_wqe_word_idx {
 	NES_IWARP_SQ_LOCINV_WQE_INV_STAG_IDX = 6,
 };
 
-
 enum nes_iwarp_rq_wqe_word_idx {
 	NES_IWARP_RQ_WQE_TOTAL_PAYLOAD_IDX = 1,
 	NES_IWARP_RQ_WQE_COMP_CTX_LOW_IDX = 2,
@@ -574,6 +617,7 @@ enum nes_nic_sq_wqe_bits {
 
 enum nes_nic_cqe_word_idx {
 	NES_NIC_CQE_ACCQP_ID_IDX = 0,
+	NES_NIC_CQE_HASH_RCVNXT = 1,
 	NES_NIC_CQE_TAG_PKT_TYPE_IDX = 2,
 	NES_NIC_CQE_MISC_IDX = 3,
 };
@@ -628,11 +672,14 @@ enum nes_aeqe_bits {
 	NES_AEQE_INBOUND_RDMA = (1<<19),
 	NES_AEQE_IWARP_STATE_MASK = (7<<20),
 	NES_AEQE_TCP_STATE_MASK = (0xf<<24),
+	NES_AEQE_Q2_DATA_WRITTEN = (0x3<<28),
 	NES_AEQE_VALID = (1<<31),
 };
 
 #define NES_AEQE_IWARP_STATE_SHIFT	20
 #define NES_AEQE_TCP_STATE_SHIFT	24
+#define NES_AEQE_Q2_DATA_ETHERNET       (1<<28)
+#define NES_AEQE_Q2_DATA_MPA            (1<<29)
 
 enum nes_aeqe_iwarp_state {
 	NES_AEQE_IWARP_STATE_NON_EXISTANT = 0,
@@ -746,6 +793,15 @@ enum nes_iwarp_sq_wqe_bits {
 	NES_IWARP_SQ_OP_NOP = 12,
 };
 
+enum nes_iwarp_cqe_major_code {
+	NES_IWARP_CQE_MAJOR_FLUSH = 1,
+	NES_IWARP_CQE_MAJOR_DRV = 0x8000
+};
+
+enum nes_iwarp_cqe_minor_code {
+	NES_IWARP_CQE_MINOR_FLUSH = 1
+};
+
 #define NES_EEPROM_READ_REQUEST (1<<16)
 #define NES_MAC_ADDR_VALID      (1<<20)
 
@@ -873,7 +929,6 @@ struct nes_hw_nic {
 	u8 replenishing_rq;
 	u8 reserved;
 
-	spinlock_t sq_lock;
 	spinlock_t rq_lock;
 };
 
@@ -967,6 +1022,12 @@ struct nes_arp_entry {
 #define DEFAULT_JUMBO_NES_QL_TARGET 40
 #define DEFAULT_JUMBO_NES_QL_HIGH   128
 #define NES_NIC_CQ_DOWNWARD_TREND   16
+#define NES_PFT_SIZE		    48
+
+#define NES_MGT_WQ_COUNT 32
+#define NES_MGT_CTX_SIZE ((NES_NIC_CTX_RQ_SIZE_32) | (NES_NIC_CTX_SQ_SIZE_32))
+#define NES_MGT_QP_OFFSET 36
+#define NES_MGT_QP_COUNT 4
 
 struct nes_hw_tune_timer {
     /* u16 cq_count; */
@@ -1064,11 +1125,12 @@ struct nes_adapter {
 	u32 wqm_wat;
 	u32 core_clock;
 	u32 firmware_version;
+	u32 eeprom_version;
 
 	u32 nic_rx_eth_route_err;
 
 	u32 et_rx_coalesce_usecs;
-	u32	et_rx_max_coalesced_frames;
+	u32 et_rx_max_coalesced_frames;
 	u32 et_rx_coalesce_usecs_irq;
 	u32 et_rx_max_coalesced_frames_irq;
 	u32 et_pkt_rate_low;
@@ -1079,6 +1141,8 @@ struct nes_adapter {
 	u32 et_rx_max_coalesced_frames_high;
 	u32 et_rate_sample_interval;
 	u32 timer_int_limit;
+	u32 wqm_quanta;
+	u8 allow_unaligned_fpdus;
 
 	/* Adapter base MAC address */
 	u32 mac_addr_low;
@@ -1094,12 +1158,14 @@ struct nes_adapter {
 	u16 pd_config_base[4];
 
 	u16 link_interrupt_count[4];
+	u8 crit_error_count[32];
 
 	/* the phy index for each port */
 	u8  phy_index[4];
 	u8  mac_sw_state[4];
 	u8  mac_link_down[4];
 	u8  phy_type[4];
+	u8  log_port;
 
 	/* PCI information */
 	unsigned int  devfn;
@@ -1111,8 +1177,10 @@ struct nes_adapter {
 	u8            netdev_max;	/* from host nic address count in EEPROM */
 	u8            port_count;
 	u8            virtwq;
+	u8            send_term_ok;
 	u8            et_use_adaptive_rx_coalesce;
 	u8            adapter_fcn_count;
+	u8 pft_mcast_map[NES_PFT_SIZE];
 };
 
 struct nes_pbl {
@@ -1125,6 +1193,19 @@ struct nes_pbl {
 	/* TODO: need to add list for two level tables */
 };
 
+#define NES_4K_PBL_CHUNK_SIZE	4096
+
+struct nes_fast_mr_wqe_pbl {
+	u64		*kva;
+	dma_addr_t	paddr;
+};
+
+struct nes_ib_fast_reg_page_list {
+	struct ib_fast_reg_page_list	ibfrpl;
+	struct nes_fast_mr_wqe_pbl 	nes_wqe_pbl;
+	u64 				pbl;
+};
+
 struct nes_listener {
 	struct work_struct      work;
 	struct workqueue_struct *wq;
@@ -1137,10 +1218,11 @@ struct nes_listener {
 
 struct nes_ib_device;
 
+#define NES_EVENT_DELAY msecs_to_jiffies(100)
+
 struct nes_vnic {
 	struct nes_ib_device *nesibdev;
 	u64 sq_full;
-	u64 sq_locked;
 	u64 tso_requests;
 	u64 segmented_tso_requests;
 	u64 linearized_skbs;
@@ -1154,7 +1236,6 @@ struct nes_vnic {
 	/* void *mem; */
 	struct nes_device *nesdev;
 	struct net_device *netdev;
-	struct vlan_group *vlan_grp;
 	atomic_t          rx_skbs_needed;
 	atomic_t          rx_skb_timer_running;
 	int               budget;
@@ -1188,10 +1269,21 @@ struct nes_vnic {
 	u8  next_qp_nic_index;
 	u8  of_device_registered;
 	u8  rdma_enabled;
-	u8  rx_checksum_disabled;
 	u32 lro_max_aggr;
 	struct net_lro_mgr lro_mgr;
 	struct net_lro_desc lro_desc[NES_MAX_LRO_DESCRIPTORS];
+	struct timer_list event_timer;
+	enum ib_event_type delayed_event;
+	enum ib_event_type last_dispatched_event;
+	spinlock_t port_ibevent_lock;
+	u32 mgt_mem_size;
+	void *mgt_vbase;
+	dma_addr_t mgt_pbase;
+	struct nes_vnic_mgt *mgtvnic[NES_MGT_QP_COUNT];
+	struct task_struct *mgt_thread;
+	wait_queue_head_t mgt_wait_queue;
+	struct sk_buff_head mgt_skb_list;
+
 };
 
 struct nes_ib_device {
@@ -1209,7 +1301,92 @@ struct nes_ib_device {
 	u32 num_pd;
 };
 
-#define nes_vlan_rx vlan_hwaccel_receive_skb
-#define nes_netif_rx netif_receive_skb
+enum nes_hdrct_flags {
+	DDP_LEN_FLAG                    = 0x80,
+	DDP_HDR_FLAG                    = 0x40,
+	RDMA_HDR_FLAG                   = 0x20
+};
+
+enum nes_term_layers {
+	LAYER_RDMA			= 0,
+	LAYER_DDP			= 1,
+	LAYER_MPA			= 2
+};
+
+enum nes_term_error_types {
+	RDMAP_CATASTROPHIC		= 0,
+	RDMAP_REMOTE_PROT		= 1,
+	RDMAP_REMOTE_OP			= 2,
+	DDP_CATASTROPHIC		= 0,
+	DDP_TAGGED_BUFFER		= 1,
+	DDP_UNTAGGED_BUFFER		= 2,
+	DDP_LLP				= 3
+};
+
+enum nes_term_rdma_errors {
+	RDMAP_INV_STAG			= 0x00,
+	RDMAP_INV_BOUNDS		= 0x01,
+	RDMAP_ACCESS			= 0x02,
+	RDMAP_UNASSOC_STAG		= 0x03,
+	RDMAP_TO_WRAP			= 0x04,
+	RDMAP_INV_RDMAP_VER		= 0x05,
+	RDMAP_UNEXPECTED_OP		= 0x06,
+	RDMAP_CATASTROPHIC_LOCAL	= 0x07,
+	RDMAP_CATASTROPHIC_GLOBAL	= 0x08,
+	RDMAP_CANT_INV_STAG		= 0x09,
+	RDMAP_UNSPECIFIED		= 0xff
+};
+
+enum nes_term_ddp_errors {
+	DDP_CATASTROPHIC_LOCAL		= 0x00,
+	DDP_TAGGED_INV_STAG		= 0x00,
+	DDP_TAGGED_BOUNDS		= 0x01,
+	DDP_TAGGED_UNASSOC_STAG		= 0x02,
+	DDP_TAGGED_TO_WRAP		= 0x03,
+	DDP_TAGGED_INV_DDP_VER		= 0x04,
+	DDP_UNTAGGED_INV_QN		= 0x01,
+	DDP_UNTAGGED_INV_MSN_NO_BUF	= 0x02,
+	DDP_UNTAGGED_INV_MSN_RANGE	= 0x03,
+	DDP_UNTAGGED_INV_MO		= 0x04,
+	DDP_UNTAGGED_INV_TOO_LONG	= 0x05,
+	DDP_UNTAGGED_INV_DDP_VER	= 0x06
+};
+
+enum nes_term_mpa_errors {
+	MPA_CLOSED			= 0x01,
+	MPA_CRC				= 0x02,
+	MPA_MARKER			= 0x03,
+	MPA_REQ_RSP			= 0x04,
+};
+
+struct nes_terminate_hdr {
+	u8 layer_etype;
+	u8 error_code;
+	u8 hdrct;
+	u8 rsvd;
+};
+
+/* Used to determine how to fill in terminate error codes */
+#define IWARP_OPCODE_WRITE		0
+#define IWARP_OPCODE_READREQ		1
+#define IWARP_OPCODE_READRSP		2
+#define IWARP_OPCODE_SEND		3
+#define IWARP_OPCODE_SEND_INV		4
+#define IWARP_OPCODE_SEND_SE		5
+#define IWARP_OPCODE_SEND_SE_INV	6
+#define IWARP_OPCODE_TERM		7
+
+/* These values are used only during terminate processing */
+#define TERM_DDP_LEN_TAGGED	14
+#define TERM_DDP_LEN_UNTAGGED	18
+#define TERM_RDMA_LEN		28
+#define RDMA_OPCODE_MASK	0x0f
+#define RDMA_READ_REQ_OPCODE	1
+#define BAD_FRAME_OFFSET	64
+#define CQE_MAJOR_DRV		0x8000
+
+/* Used for link status recheck after interrupt processing */
+#define NES_LINK_RECHECK_DELAY	msecs_to_jiffies(50)
+#define NES_LINK_RECHECK_MAX	60
 
 #endif		/* __NES_HW_H */
diff --git a/drivers/infiniband/hw/nes/nes_mgt.c b/drivers/infiniband/hw/nes/nes_mgt.c
new file mode 100644
index 00000000000..416645259b0
--- /dev/null
+++ b/drivers/infiniband/hw/nes/nes_mgt.c
@@ -0,0 +1,1160 @@
+/*
+ * Copyright (c) 2006 - 2011 Intel-NE, Inc.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/skbuff.h>
+#include <linux/etherdevice.h>
+#include <linux/kthread.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
+#include <net/tcp.h>
+#include "nes.h"
+#include "nes_mgt.h"
+
+atomic_t pau_qps_created;
+atomic_t pau_qps_destroyed;
+
+static void nes_replenish_mgt_rq(struct nes_vnic_mgt *mgtvnic)
+{
+	unsigned long flags;
+	dma_addr_t bus_address;
+	struct sk_buff *skb;
+	struct nes_hw_nic_rq_wqe *nic_rqe;
+	struct nes_hw_mgt *nesmgt;
+	struct nes_device *nesdev;
+	struct nes_rskb_cb *cb;
+	u32 rx_wqes_posted = 0;
+
+	nesmgt = &mgtvnic->mgt;
+	nesdev = mgtvnic->nesvnic->nesdev;
+	spin_lock_irqsave(&nesmgt->rq_lock, flags);
+	if (nesmgt->replenishing_rq != 0) {
+		if (((nesmgt->rq_size - 1) == atomic_read(&mgtvnic->rx_skbs_needed)) &&
+		    (atomic_read(&mgtvnic->rx_skb_timer_running) == 0)) {
+			atomic_set(&mgtvnic->rx_skb_timer_running, 1);
+			spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+			mgtvnic->rq_wqes_timer.expires = jiffies + (HZ / 2);      /* 1/2 second */
+			add_timer(&mgtvnic->rq_wqes_timer);
+		} else {
+			spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+		}
+		return;
+	}
+	nesmgt->replenishing_rq = 1;
+	spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+	do {
+		skb = dev_alloc_skb(mgtvnic->nesvnic->max_frame_size);
+		if (skb) {
+			skb->dev = mgtvnic->nesvnic->netdev;
+
+			bus_address = pci_map_single(nesdev->pcidev,
+						     skb->data, mgtvnic->nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = bus_address;
+			cb->maplen = mgtvnic->nesvnic->max_frame_size;
+
+			nic_rqe = &nesmgt->rq_vbase[mgtvnic->mgt.rq_head];
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] =
+				cpu_to_le32(mgtvnic->nesvnic->max_frame_size);
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_3_2_IDX] = 0;
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX] =
+				cpu_to_le32((u32)bus_address);
+			nic_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX] =
+				cpu_to_le32((u32)((u64)bus_address >> 32));
+			nesmgt->rx_skb[nesmgt->rq_head] = skb;
+			nesmgt->rq_head++;
+			nesmgt->rq_head &= nesmgt->rq_size - 1;
+			atomic_dec(&mgtvnic->rx_skbs_needed);
+			barrier();
+			if (++rx_wqes_posted == 255) {
+				nes_write32(nesdev->regs + NES_WQE_ALLOC, (rx_wqes_posted << 24) | nesmgt->qp_id);
+				rx_wqes_posted = 0;
+			}
+		} else {
+			spin_lock_irqsave(&nesmgt->rq_lock, flags);
+			if (((nesmgt->rq_size - 1) == atomic_read(&mgtvnic->rx_skbs_needed)) &&
+			    (atomic_read(&mgtvnic->rx_skb_timer_running) == 0)) {
+				atomic_set(&mgtvnic->rx_skb_timer_running, 1);
+				spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+				mgtvnic->rq_wqes_timer.expires = jiffies + (HZ / 2);      /* 1/2 second */
+				add_timer(&mgtvnic->rq_wqes_timer);
+			} else {
+				spin_unlock_irqrestore(&nesmgt->rq_lock, flags);
+			}
+			break;
+		}
+	} while (atomic_read(&mgtvnic->rx_skbs_needed));
+	barrier();
+	if (rx_wqes_posted)
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, (rx_wqes_posted << 24) | nesmgt->qp_id);
+	nesmgt->replenishing_rq = 0;
+}
+
+/**
+ * nes_mgt_rq_wqes_timeout
+ */
+static void nes_mgt_rq_wqes_timeout(unsigned long parm)
+{
+	struct nes_vnic_mgt *mgtvnic = (struct nes_vnic_mgt *)parm;
+
+	atomic_set(&mgtvnic->rx_skb_timer_running, 0);
+	if (atomic_read(&mgtvnic->rx_skbs_needed))
+		nes_replenish_mgt_rq(mgtvnic);
+}
+
+/**
+ * nes_mgt_free_skb - unmap and free skb
+ */
+static void nes_mgt_free_skb(struct nes_device *nesdev, struct sk_buff *skb, u32 dir)
+{
+	struct nes_rskb_cb *cb;
+
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen, dir);
+	cb->busaddr = 0;
+	dev_kfree_skb_any(skb);
+}
+
+/**
+ * nes_download_callback - handle download completions
+ */
+static void nes_download_callback(struct nes_device *nesdev, struct nes_cqp_request *cqp_request)
+{
+	struct pau_fpdu_info *fpdu_info = cqp_request->cqp_callback_pointer;
+	struct nes_qp *nesqp = fpdu_info->nesqp;
+	struct sk_buff *skb;
+	int i;
+
+	for (i = 0; i < fpdu_info->frag_cnt; i++) {
+		skb = fpdu_info->frags[i].skb;
+		if (fpdu_info->frags[i].cmplt) {
+			nes_mgt_free_skb(nesdev, skb, PCI_DMA_TODEVICE);
+			nes_rem_ref_cm_node(nesqp->cm_node);
+		}
+	}
+
+	if (fpdu_info->hdr_vbase)
+		pci_free_consistent(nesdev->pcidev, fpdu_info->hdr_len,
+				    fpdu_info->hdr_vbase, fpdu_info->hdr_pbase);
+	kfree(fpdu_info);
+}
+
+/**
+ * nes_get_seq - Get the seq, ack_seq and window from the packet
+ */
+static u32 nes_get_seq(struct sk_buff *skb, u32 *ack, u16 *wnd, u32 *fin_rcvd, u32 *rst_rcvd)
+{
+	struct nes_rskb_cb *cb = (struct nes_rskb_cb *)&skb->cb[0];
+	struct iphdr *iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+	struct tcphdr *tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+
+	*ack = be32_to_cpu(tcph->ack_seq);
+	*wnd = be16_to_cpu(tcph->window);
+	*fin_rcvd = tcph->fin;
+	*rst_rcvd = tcph->rst;
+	return be32_to_cpu(tcph->seq);
+}
+
+/**
+ * nes_get_next_skb - Get the next skb based on where current skb is in the queue
+ */
+static struct sk_buff *nes_get_next_skb(struct nes_device *nesdev, struct nes_qp *nesqp,
+					struct sk_buff *skb, u32 nextseq, u32 *ack,
+					u16 *wnd, u32 *fin_rcvd, u32 *rst_rcvd)
+{
+	u32 seq;
+	bool processacks;
+	struct sk_buff *old_skb;
+
+	if (skb) {
+		/* Continue processing fpdu */
+		if (skb->next == (struct sk_buff *)&nesqp->pau_list)
+			goto out;
+		skb = skb->next;
+		processacks = false;
+	} else {
+		/* Starting a new one */
+		if (skb_queue_empty(&nesqp->pau_list))
+			goto out;
+		skb = skb_peek(&nesqp->pau_list);
+		processacks = true;
+	}
+
+	while (1) {
+		if (skb_queue_empty(&nesqp->pau_list))
+			goto out;
+
+		seq = nes_get_seq(skb, ack, wnd, fin_rcvd, rst_rcvd);
+		if (seq == nextseq) {
+			if (skb->len || processacks)
+				break;
+		} else if (after(seq, nextseq)) {
+			goto out;
+		}
+
+		old_skb = skb;
+		skb = skb->next;
+		skb_unlink(old_skb, &nesqp->pau_list);
+		nes_mgt_free_skb(nesdev, old_skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+		if (skb == (struct sk_buff *)&nesqp->pau_list)
+			goto out;
+	}
+	return skb;
+
+out:
+	return NULL;
+}
+
+/**
+ * get_fpdu_info - Find the next complete fpdu and return its fragments.
+ */
+static int get_fpdu_info(struct nes_device *nesdev, struct nes_qp *nesqp,
+			 struct pau_fpdu_info **pau_fpdu_info)
+{
+	struct sk_buff *skb;
+	struct iphdr *iph;
+	struct tcphdr *tcph;
+	struct nes_rskb_cb *cb;
+	struct pau_fpdu_info *fpdu_info = NULL;
+	struct pau_fpdu_frag frags[MAX_FPDU_FRAGS];
+	u32 fpdu_len = 0;
+	u32 tmp_len;
+	int frag_cnt = 0;
+	u32 tot_len;
+	u32 frag_tot;
+	u32 ack;
+	u32 fin_rcvd;
+	u32 rst_rcvd;
+	u16 wnd;
+	int i;
+	int rc = 0;
+
+	*pau_fpdu_info = NULL;
+
+	skb = nes_get_next_skb(nesdev, nesqp, NULL, nesqp->pau_rcv_nxt, &ack, &wnd, &fin_rcvd, &rst_rcvd);
+	if (!skb)
+		goto out;
+
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	if (skb->len) {
+		fpdu_len = be16_to_cpu(*(__be16 *) skb->data) + MPA_FRAMING;
+		fpdu_len = (fpdu_len + 3) & 0xfffffffc;
+		tmp_len = fpdu_len;
+
+		/* See if we have all of the fpdu */
+		frag_tot = 0;
+		memset(&frags, 0, sizeof frags);
+		for (i = 0; i < MAX_FPDU_FRAGS; i++) {
+			frags[i].physaddr = cb->busaddr;
+			frags[i].physaddr += skb->data - cb->data_start;
+			frags[i].frag_len = min(tmp_len, skb->len);
+			frags[i].skb = skb;
+			frags[i].cmplt = (skb->len == frags[i].frag_len);
+			frag_tot += frags[i].frag_len;
+			frag_cnt++;
+
+			tmp_len -= frags[i].frag_len;
+			if (tmp_len == 0)
+				break;
+
+			skb = nes_get_next_skb(nesdev, nesqp, skb,
+					       nesqp->pau_rcv_nxt + frag_tot, &ack, &wnd, &fin_rcvd, &rst_rcvd);
+			if (!skb)
+				goto out;
+			if (rst_rcvd) {
+				/* rst received in the middle of fpdu */
+				for (; i >= 0; i--) {
+					skb_unlink(frags[i].skb, &nesqp->pau_list);
+					nes_mgt_free_skb(nesdev, frags[i].skb, PCI_DMA_TODEVICE);
+				}
+				cb = (struct nes_rskb_cb *)&skb->cb[0];
+				frags[0].physaddr = cb->busaddr;
+				frags[0].physaddr += skb->data - cb->data_start;
+				frags[0].frag_len = skb->len;
+				frags[0].skb = skb;
+				frags[0].cmplt = true;
+				frag_cnt = 1;
+				break;
+			}
+
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+		}
+	} else {
+		/* no data */
+		frags[0].physaddr = cb->busaddr;
+		frags[0].frag_len = 0;
+		frags[0].skb = skb;
+		frags[0].cmplt = true;
+		frag_cnt = 1;
+	}
+
+	/* Found one */
+	fpdu_info = kzalloc(sizeof(*fpdu_info), GFP_ATOMIC);
+	if (fpdu_info == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to alloc a fpdu_info.\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	fpdu_info->cqp_request = nes_get_cqp_request(nesdev);
+	if (fpdu_info->cqp_request == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		rc = -ENOMEM;
+		goto out;
+	}
+
+	cb = (struct nes_rskb_cb *)&frags[0].skb->cb[0];
+	iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+	tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	fpdu_info->hdr_len = (((unsigned char *)tcph) + 4 * (tcph->doff)) - cb->data_start;
+	fpdu_info->data_len = fpdu_len;
+	tot_len = fpdu_info->hdr_len + fpdu_len - ETH_HLEN;
+
+	if (frags[0].cmplt) {
+		fpdu_info->hdr_pbase = cb->busaddr;
+		fpdu_info->hdr_vbase = NULL;
+	} else {
+		fpdu_info->hdr_vbase = pci_alloc_consistent(nesdev->pcidev,
+							    fpdu_info->hdr_len, &fpdu_info->hdr_pbase);
+		if (!fpdu_info->hdr_vbase) {
+			nes_debug(NES_DBG_PAU, "Unable to allocate memory for pau first frag\n");
+			rc = -ENOMEM;
+			goto out;
+		}
+
+		/* Copy hdrs, adjusting len and seqnum */
+		memcpy(fpdu_info->hdr_vbase, cb->data_start, fpdu_info->hdr_len);
+		iph = (struct iphdr *)(fpdu_info->hdr_vbase + ETH_HLEN);
+		tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	}
+
+	iph->tot_len = cpu_to_be16(tot_len);
+	iph->saddr = cpu_to_be32(0x7f000001);
+
+	tcph->seq = cpu_to_be32(nesqp->pau_rcv_nxt);
+	tcph->ack_seq = cpu_to_be32(ack);
+	tcph->window = cpu_to_be16(wnd);
+
+	nesqp->pau_rcv_nxt += fpdu_len + fin_rcvd;
+
+	memcpy(fpdu_info->frags, frags, sizeof(fpdu_info->frags));
+	fpdu_info->frag_cnt = frag_cnt;
+	fpdu_info->nesqp = nesqp;
+	*pau_fpdu_info = fpdu_info;
+
+	/* Update skb's for next pass */
+	for (i = 0; i < frag_cnt; i++) {
+		cb = (struct nes_rskb_cb *)&frags[i].skb->cb[0];
+		skb_pull(frags[i].skb, frags[i].frag_len);
+
+		if (frags[i].skb->len == 0) {
+			/* Pull skb off the list - it will be freed in the callback */
+			if (!skb_queue_empty(&nesqp->pau_list))
+				skb_unlink(frags[i].skb, &nesqp->pau_list);
+		} else {
+			/* Last skb still has data so update the seq */
+			iph = (struct iphdr *)(cb->data_start + ETH_HLEN);
+			tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+			tcph->seq = cpu_to_be32(nesqp->pau_rcv_nxt);
+		}
+	}
+
+out:
+	if (rc) {
+		if (fpdu_info) {
+			if (fpdu_info->cqp_request)
+				nes_put_cqp_request(nesdev, fpdu_info->cqp_request);
+			kfree(fpdu_info);
+		}
+	}
+	return rc;
+}
+
+/**
+ * forward_fpdu - send complete fpdus, one at a time
+ */
+static int forward_fpdus(struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct pau_fpdu_info *fpdu_info;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	struct nes_cqp_request *cqp_request;
+	unsigned long flags;
+	u64 u64tmp;
+	u32 u32tmp;
+	int rc;
+
+	while (1) {
+		spin_lock_irqsave(&nesqp->pau_lock, flags);
+		rc = get_fpdu_info(nesdev, nesqp, &fpdu_info);
+		if (rc || (fpdu_info == NULL)) {
+			spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+			return rc;
+		}
+
+		cqp_request = fpdu_info->cqp_request;
+		cqp_wqe = &cqp_request->cqp_wqe;
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_DL_OPCODE_IDX,
+				    NES_CQP_DOWNLOAD_SEGMENT |
+				    (((u32)nesvnic->logical_port) << NES_CQP_OP_LOGICAL_PORT_SHIFT));
+
+		u32tmp = fpdu_info->hdr_len << 16;
+		u32tmp |= fpdu_info->hdr_len + (u32)fpdu_info->data_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_DL_LENGTH_0_TOTAL_IDX,
+				    u32tmp);
+
+		u32tmp = (fpdu_info->frags[1].frag_len << 16) | fpdu_info->frags[0].frag_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_LENGTH_2_1_IDX,
+				    u32tmp);
+
+		u32tmp = (fpdu_info->frags[3].frag_len << 16) | fpdu_info->frags[2].frag_len;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_LENGTH_4_3_IDX,
+				    u32tmp);
+
+		u64tmp = (u64)fpdu_info->hdr_pbase;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_LOW_IDX,
+				    lower_32_bits(u64tmp));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_HIGH_IDX,
+				    upper_32_bits(u64tmp));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[0].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[0].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG2_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[1].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG2_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[1].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG3_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[2].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG3_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[2].physaddr));
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG4_LOW_IDX,
+				    lower_32_bits(fpdu_info->frags[3].physaddr));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_NIC_SQ_WQE_FRAG4_HIGH_IDX,
+				    upper_32_bits(fpdu_info->frags[3].physaddr));
+
+		cqp_request->cqp_callback_pointer = fpdu_info;
+		cqp_request->callback = 1;
+		cqp_request->cqp_callback = nes_download_callback;
+
+		atomic_set(&cqp_request->refcount, 1);
+		nes_post_cqp_request(nesdev, cqp_request);
+		spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+	}
+
+	return 0;
+}
+
+static void process_fpdus(struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	int again = 1;
+	unsigned long flags;
+
+	do {
+		/* Ignore rc - if it failed, tcp retries will cause it to try again */
+		forward_fpdus(nesvnic, nesqp);
+
+		spin_lock_irqsave(&nesqp->pau_lock, flags);
+		if (nesqp->pau_pending) {
+			nesqp->pau_pending = 0;
+		} else {
+			nesqp->pau_busy = 0;
+			again = 0;
+		}
+
+		spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+	} while (again);
+}
+
+/**
+ * queue_fpdus - Handle fpdu's that hw passed up to sw
+ */
+static void queue_fpdus(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct sk_buff *tmpskb;
+	struct nes_rskb_cb *cb;
+	struct iphdr *iph;
+	struct tcphdr *tcph;
+	unsigned char *tcph_end;
+	u32 rcv_nxt;
+	u32 rcv_wnd;
+	u32 seqnum;
+	u32 len;
+	bool process_it = false;
+	unsigned long flags;
+
+	/* Move data ptr to after tcp header */
+	iph = (struct iphdr *)skb->data;
+	tcph = (struct tcphdr *)(((char *)iph) + (4 * iph->ihl));
+	seqnum = be32_to_cpu(tcph->seq);
+	tcph_end = (((char *)tcph) + (4 * tcph->doff));
+
+	len = be16_to_cpu(iph->tot_len);
+	if (skb->len > len)
+		skb_trim(skb, len);
+	skb_pull(skb, tcph_end - skb->data);
+
+	/* Initialize tracking values */
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	cb->seqnum = seqnum;
+
+	/* Make sure data is in the receive window */
+	rcv_nxt = nesqp->pau_rcv_nxt;
+	rcv_wnd = le32_to_cpu(nesqp->nesqp_context->rcv_wnd);
+	if (!between(seqnum, rcv_nxt, (rcv_nxt + rcv_wnd))) {
+		nes_mgt_free_skb(nesvnic->nesdev, skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+		return;
+	}
+
+	spin_lock_irqsave(&nesqp->pau_lock, flags);
+
+	if (nesqp->pau_busy)
+		nesqp->pau_pending = 1;
+	else
+		nesqp->pau_busy = 1;
+
+	/* Queue skb by sequence number */
+	if (skb_queue_len(&nesqp->pau_list) == 0) {
+		skb_queue_head(&nesqp->pau_list, skb);
+	} else {
+		tmpskb = nesqp->pau_list.next;
+		while (tmpskb != (struct sk_buff *)&nesqp->pau_list) {
+			cb = (struct nes_rskb_cb *)&tmpskb->cb[0];
+			if (before(seqnum, cb->seqnum))
+				break;
+			tmpskb = tmpskb->next;
+		}
+		skb_insert(tmpskb, skb, &nesqp->pau_list);
+	}
+	if (nesqp->pau_state == PAU_READY)
+		process_it = true;
+	spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+
+	if (process_it)
+		process_fpdus(nesvnic, nesqp);
+
+	return;
+}
+
+/**
+ * mgt_thread - Handle mgt skbs in a safe context
+ */
+static int mgt_thread(void *context)
+{
+	struct nes_vnic *nesvnic = context;
+	struct sk_buff *skb;
+	struct nes_rskb_cb *cb;
+
+	while (!kthread_should_stop()) {
+		wait_event_interruptible(nesvnic->mgt_wait_queue,
+					 skb_queue_len(&nesvnic->mgt_skb_list) || kthread_should_stop());
+		while ((skb_queue_len(&nesvnic->mgt_skb_list)) && !kthread_should_stop()) {
+			skb = skb_dequeue(&nesvnic->mgt_skb_list);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->data_start = skb->data - ETH_HLEN;
+			cb->busaddr = pci_map_single(nesvnic->nesdev->pcidev, cb->data_start,
+						     nesvnic->max_frame_size, PCI_DMA_TODEVICE);
+			queue_fpdus(skb, nesvnic, cb->nesqp);
+		}
+	}
+
+	/* Closing down so delete any entries on the queue */
+	while (skb_queue_len(&nesvnic->mgt_skb_list)) {
+		skb = skb_dequeue(&nesvnic->mgt_skb_list);
+		cb = (struct nes_rskb_cb *)&skb->cb[0];
+		nes_rem_ref_cm_node(cb->nesqp->cm_node);
+		dev_kfree_skb_any(skb);
+	}
+	return 0;
+}
+
+/**
+ * nes_queue_skbs - Queue skb so it can be handled in a thread context
+ */
+void nes_queue_mgt_skbs(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_rskb_cb *cb;
+
+	cb = (struct nes_rskb_cb *)&skb->cb[0];
+	cb->nesqp = nesqp;
+	skb_queue_tail(&nesvnic->mgt_skb_list, skb);
+	wake_up_interruptible(&nesvnic->mgt_wait_queue);
+}
+
+void nes_destroy_pau_qp(struct nes_device *nesdev, struct nes_qp *nesqp)
+{
+	struct sk_buff *skb;
+	unsigned long flags;
+	atomic_inc(&pau_qps_destroyed);
+
+	/* Free packets that have not yet been forwarded */
+	/* Lock is acquired by skb_dequeue when removing the skb */
+	spin_lock_irqsave(&nesqp->pau_lock, flags);
+	while (skb_queue_len(&nesqp->pau_list)) {
+		skb = skb_dequeue(&nesqp->pau_list);
+		nes_mgt_free_skb(nesdev, skb, PCI_DMA_TODEVICE);
+		nes_rem_ref_cm_node(nesqp->cm_node);
+	}
+	spin_unlock_irqrestore(&nesqp->pau_lock, flags);
+}
+
+static void nes_chg_qh_handler(struct nes_device *nesdev, struct nes_cqp_request *cqp_request)
+{
+	struct pau_qh_chg *qh_chg = cqp_request->cqp_callback_pointer;
+	struct nes_cqp_request *new_request;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	struct nes_adapter *nesadapter;
+	struct nes_qp *nesqp;
+	struct nes_v4_quad nes_quad;
+	u32 crc_value;
+	u64 u64temp;
+
+	nesadapter = nesdev->nesadapter;
+	nesqp = qh_chg->nesqp;
+
+	/* Should we handle the bad completion */
+	if (cqp_request->major_code)
+		WARN(1, PFX "Invalid cqp_request major_code=0x%x\n",
+		       cqp_request->major_code);
+
+	switch (nesqp->pau_state) {
+	case PAU_DEL_QH:
+		/* Old hash code deleted, now set the new one */
+		nesqp->pau_state = PAU_ADD_LB_QH;
+		new_request = nes_get_cqp_request(nesdev);
+		if (new_request == NULL) {
+			nes_debug(NES_DBG_PAU, "Failed to get a new_request.\n");
+			WARN_ON(1);
+			return;
+		}
+
+		memset(&nes_quad, 0, sizeof(nes_quad));
+		nes_quad.DstIpAdrIndex =
+			cpu_to_le32((u32)PCI_FUNC(nesdev->pcidev->devfn) << 24);
+		nes_quad.SrcIpadr = cpu_to_be32(0x7f000001);
+		nes_quad.TcpPorts[0] = swab16(nesqp->nesqp_context->tcpPorts[1]);
+		nes_quad.TcpPorts[1] = swab16(nesqp->nesqp_context->tcpPorts[0]);
+
+		/* Produce hash key */
+		crc_value = get_crc_value(&nes_quad);
+		nesqp->hte_index = cpu_to_be32(crc_value ^ 0xffffffff);
+		nes_debug(NES_DBG_PAU, "new HTE Index = 0x%08X, CRC = 0x%08X\n",
+			  nesqp->hte_index, nesqp->hte_index & nesadapter->hte_index_mask);
+
+		nesqp->hte_index &= nesadapter->hte_index_mask;
+		nesqp->nesqp_context->hte_index = cpu_to_le32(nesqp->hte_index);
+		nesqp->nesqp_context->ip0 = cpu_to_le32(0x7f000001);
+		nesqp->nesqp_context->rcv_nxt = cpu_to_le32(nesqp->pau_rcv_nxt);
+
+		cqp_wqe = &new_request->cqp_wqe;
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words,
+				    NES_CQP_WQE_OPCODE_IDX, NES_CQP_MANAGE_QUAD_HASH |
+				    NES_CQP_QP_TYPE_IWARP | NES_CQP_QP_CONTEXT_VALID | NES_CQP_QP_IWARP_STATE_RTS);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX, nesqp->hwqp.qp_id);
+		u64temp = (u64)nesqp->nesqp_context_pbase;
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+		nes_debug(NES_DBG_PAU, "Waiting for CQP completion for adding the quad hash.\n");
+
+		new_request->cqp_callback_pointer = qh_chg;
+		new_request->callback = 1;
+		new_request->cqp_callback = nes_chg_qh_handler;
+		atomic_set(&new_request->refcount, 1);
+		nes_post_cqp_request(nesdev, new_request);
+		break;
+
+	case PAU_ADD_LB_QH:
+		/* Start processing the queued fpdu's */
+		nesqp->pau_state = PAU_READY;
+		process_fpdus(qh_chg->nesvnic, qh_chg->nesqp);
+		kfree(qh_chg);
+		break;
+	}
+}
+
+/**
+ * nes_change_quad_hash
+ */
+static int nes_change_quad_hash(struct nes_device *nesdev,
+				struct nes_vnic *nesvnic, struct nes_qp *nesqp)
+{
+	struct nes_cqp_request *cqp_request = NULL;
+	struct pau_qh_chg *qh_chg = NULL;
+	u64 u64temp;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	int ret = 0;
+
+	cqp_request = nes_get_cqp_request(nesdev);
+	if (cqp_request == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		ret = -ENOMEM;
+		goto chg_qh_err;
+	}
+
+	qh_chg = kmalloc(sizeof *qh_chg, GFP_ATOMIC);
+	if (qh_chg == NULL) {
+		nes_debug(NES_DBG_PAU, "Failed to get a cqp_request.\n");
+		ret = -ENOMEM;
+		goto chg_qh_err;
+	}
+	qh_chg->nesdev = nesdev;
+	qh_chg->nesvnic = nesvnic;
+	qh_chg->nesqp = nesqp;
+	nesqp->pau_state = PAU_DEL_QH;
+
+	cqp_wqe = &cqp_request->cqp_wqe;
+	nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+	set_wqe_32bit_value(cqp_wqe->wqe_words,
+			    NES_CQP_WQE_OPCODE_IDX, NES_CQP_MANAGE_QUAD_HASH | NES_CQP_QP_DEL_HTE |
+			    NES_CQP_QP_TYPE_IWARP | NES_CQP_QP_CONTEXT_VALID | NES_CQP_QP_IWARP_STATE_RTS);
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX, nesqp->hwqp.qp_id);
+	u64temp = (u64)nesqp->nesqp_context_pbase;
+	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+	nes_debug(NES_DBG_PAU, "Waiting for CQP completion for deleting the quad hash.\n");
+
+	cqp_request->cqp_callback_pointer = qh_chg;
+	cqp_request->callback = 1;
+	cqp_request->cqp_callback = nes_chg_qh_handler;
+	atomic_set(&cqp_request->refcount, 1);
+	nes_post_cqp_request(nesdev, cqp_request);
+
+	return ret;
+
+chg_qh_err:
+	kfree(qh_chg);
+	if (cqp_request)
+		nes_put_cqp_request(nesdev, cqp_request);
+	return ret;
+}
+
+/**
+ * nes_mgt_ce_handler
+ * This management code deals with any packed and unaligned (pau) fpdu's
+ * that the hardware cannot handle.
+ */
+static void nes_mgt_ce_handler(struct nes_device *nesdev, struct nes_hw_nic_cq *cq)
+{
+	struct nes_vnic_mgt *mgtvnic = container_of(cq, struct nes_vnic_mgt, mgt_cq);
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
+	u32 head;
+	u32 cq_size;
+	u32 cqe_count = 0;
+	u32 cqe_misc;
+	u32 qp_id = 0;
+	u32 skbs_needed;
+	unsigned long context;
+	struct nes_qp *nesqp;
+	struct sk_buff *rx_skb;
+	struct nes_rskb_cb *cb;
+
+	head = cq->cq_head;
+	cq_size = cq->cq_size;
+
+	while (1) {
+		cqe_misc = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_NIC_CQE_MISC_IDX]);
+		if (!(cqe_misc & NES_NIC_CQE_VALID))
+			break;
+
+		nesqp = NULL;
+		if (cqe_misc & NES_NIC_CQE_ACCQP_VALID) {
+			qp_id = le32_to_cpu(cq->cq_vbase[head].cqe_words[NES_NIC_CQE_ACCQP_ID_IDX]);
+			qp_id &= 0x001fffff;
+			if (qp_id < nesadapter->max_qp) {
+				context = (unsigned long)nesadapter->qp_table[qp_id - NES_FIRST_QPN];
+				nesqp = (struct nes_qp *)context;
+			}
+		}
+
+		if (nesqp) {
+			if (nesqp->pau_mode == false) {
+				nesqp->pau_mode = true; /* First time for this qp */
+				nesqp->pau_rcv_nxt = le32_to_cpu(
+					cq->cq_vbase[head].cqe_words[NES_NIC_CQE_HASH_RCVNXT]);
+				skb_queue_head_init(&nesqp->pau_list);
+				spin_lock_init(&nesqp->pau_lock);
+				atomic_inc(&pau_qps_created);
+				nes_change_quad_hash(nesdev, mgtvnic->nesvnic, nesqp);
+			}
+
+			rx_skb = mgtvnic->mgt.rx_skb[mgtvnic->mgt.rq_tail];
+			rx_skb->len = 0;
+			skb_put(rx_skb, cqe_misc & 0x0000ffff);
+			rx_skb->protocol = eth_type_trans(rx_skb, mgtvnic->nesvnic->netdev);
+			cb = (struct nes_rskb_cb *)&rx_skb->cb[0];
+			pci_unmap_single(nesdev->pcidev, cb->busaddr, cb->maplen, PCI_DMA_FROMDEVICE);
+			cb->busaddr = 0;
+			mgtvnic->mgt.rq_tail++;
+			mgtvnic->mgt.rq_tail &= mgtvnic->mgt.rq_size - 1;
+
+			nes_add_ref_cm_node(nesqp->cm_node);
+			nes_queue_mgt_skbs(rx_skb, mgtvnic->nesvnic, nesqp);
+		} else {
+			printk(KERN_ERR PFX "Invalid QP %d for packed/unaligned handling\n", qp_id);
+		}
+
+		cq->cq_vbase[head].cqe_words[NES_NIC_CQE_MISC_IDX] = 0;
+		cqe_count++;
+		if (++head >= cq_size)
+			head = 0;
+
+		if (cqe_count == 255) {
+			/* Replenish mgt CQ */
+			nes_write32(nesdev->regs + NES_CQE_ALLOC, cq->cq_number | (cqe_count << 16));
+			nesdev->currcq_count += cqe_count;
+			cqe_count = 0;
+		}
+
+		skbs_needed = atomic_inc_return(&mgtvnic->rx_skbs_needed);
+		if (skbs_needed > (mgtvnic->mgt.rq_size >> 1))
+			nes_replenish_mgt_rq(mgtvnic);
+	}
+
+	cq->cq_head = head;
+	nes_write32(nesdev->regs + NES_CQE_ALLOC, NES_CQE_ALLOC_NOTIFY_NEXT |
+		    cq->cq_number | (cqe_count << 16));
+	nes_read32(nesdev->regs + NES_CQE_ALLOC);
+	nesdev->currcq_count += cqe_count;
+}
+
+/**
+ * nes_init_mgt_qp
+ */
+int nes_init_mgt_qp(struct nes_device *nesdev, struct net_device *netdev, struct nes_vnic *nesvnic)
+{
+	struct nes_vnic_mgt *mgtvnic;
+	u32 counter;
+	void *vmem;
+	dma_addr_t pmem;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	u32 cqp_head;
+	unsigned long flags;
+	struct nes_hw_nic_qp_context *mgt_context;
+	u64 u64temp;
+	struct nes_hw_nic_rq_wqe *mgt_rqe;
+	struct sk_buff *skb;
+	u32 wqe_count;
+	struct nes_rskb_cb *cb;
+	u32 mgt_mem_size;
+	void *mgt_vbase;
+	dma_addr_t mgt_pbase;
+	int i;
+	int ret;
+
+	/* Allocate space the all mgt QPs once */
+	mgtvnic = kzalloc(NES_MGT_QP_COUNT * sizeof(struct nes_vnic_mgt), GFP_KERNEL);
+	if (mgtvnic == NULL) {
+		nes_debug(NES_DBG_INIT, "Unable to allocate memory for mgt structure\n");
+		return -ENOMEM;
+	}
+
+	/* Allocate fragment, RQ, and CQ; Reuse CEQ based on the PCI function */
+	/* We are not sending from this NIC so sq is not allocated */
+	mgt_mem_size = 256 +
+		       (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe)) +
+		       (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_cqe)) +
+		       sizeof(struct nes_hw_nic_qp_context);
+	mgt_mem_size = (mgt_mem_size + PAGE_SIZE - 1) & ~(PAGE_SIZE - 1);
+	mgt_vbase = pci_alloc_consistent(nesdev->pcidev, NES_MGT_QP_COUNT * mgt_mem_size, &mgt_pbase);
+	if (!mgt_vbase) {
+		kfree(mgtvnic);
+		nes_debug(NES_DBG_INIT, "Unable to allocate memory for mgt host descriptor rings\n");
+		return -ENOMEM;
+	}
+
+	nesvnic->mgt_mem_size = NES_MGT_QP_COUNT * mgt_mem_size;
+	nesvnic->mgt_vbase = mgt_vbase;
+	nesvnic->mgt_pbase = mgt_pbase;
+
+	skb_queue_head_init(&nesvnic->mgt_skb_list);
+	init_waitqueue_head(&nesvnic->mgt_wait_queue);
+	nesvnic->mgt_thread = kthread_run(mgt_thread, nesvnic, "nes_mgt_thread");
+
+	for (i = 0; i < NES_MGT_QP_COUNT; i++) {
+		mgtvnic->nesvnic = nesvnic;
+		mgtvnic->mgt.qp_id = nesdev->mac_index + NES_MGT_QP_OFFSET + i;
+		memset(mgt_vbase, 0, mgt_mem_size);
+		nes_debug(NES_DBG_INIT, "Allocated mgt QP structures at %p (phys = %016lX), size = %u.\n",
+			  mgt_vbase, (unsigned long)mgt_pbase, mgt_mem_size);
+
+		vmem = (void *)(((unsigned long)mgt_vbase + (256 - 1)) &
+				~(unsigned long)(256 - 1));
+		pmem = (dma_addr_t)(((unsigned long long)mgt_pbase + (256 - 1)) &
+				    ~(unsigned long long)(256 - 1));
+
+		spin_lock_init(&mgtvnic->mgt.rq_lock);
+
+		/* setup the RQ */
+		mgtvnic->mgt.rq_vbase = vmem;
+		mgtvnic->mgt.rq_pbase = pmem;
+		mgtvnic->mgt.rq_head = 0;
+		mgtvnic->mgt.rq_tail = 0;
+		mgtvnic->mgt.rq_size = NES_MGT_WQ_COUNT;
+
+		/* setup the CQ */
+		vmem += (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe));
+		pmem += (NES_MGT_WQ_COUNT * sizeof(struct nes_hw_nic_rq_wqe));
+
+		mgtvnic->mgt_cq.cq_number = mgtvnic->mgt.qp_id;
+		mgtvnic->mgt_cq.cq_vbase = vmem;
+		mgtvnic->mgt_cq.cq_pbase = pmem;
+		mgtvnic->mgt_cq.cq_head = 0;
+		mgtvnic->mgt_cq.cq_size = NES_MGT_WQ_COUNT;
+
+		mgtvnic->mgt_cq.ce_handler = nes_mgt_ce_handler;
+
+		/* Send CreateCQ request to CQP */
+		spin_lock_irqsave(&nesdev->cqp.lock, flags);
+		cqp_head = nesdev->cqp.sq_head;
+
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] = cpu_to_le32(
+			NES_CQP_CREATE_CQ | NES_CQP_CQ_CEQ_VALID |
+			((u32)mgtvnic->mgt_cq.cq_size << 16));
+		cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX] = cpu_to_le32(
+			mgtvnic->mgt_cq.cq_number | ((u32)nesdev->ceq_index << 16));
+		u64temp = (u64)mgtvnic->mgt_cq.cq_pbase;
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_CQ_WQE_PBL_LOW_IDX, u64temp);
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_HIGH_IDX] = 0;
+		u64temp = (unsigned long)&mgtvnic->mgt_cq;
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_LOW_IDX] = cpu_to_le32((u32)(u64temp >> 1));
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_CQ_CONTEXT_HIGH_IDX] =
+			cpu_to_le32(((u32)((u64temp) >> 33)) & 0x7FFFFFFF);
+		cqp_wqe->wqe_words[NES_CQP_CQ_WQE_DOORBELL_INDEX_HIGH_IDX] = 0;
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		/* Send CreateQP request to CQP */
+		mgt_context = (void *)(&mgtvnic->mgt_cq.cq_vbase[mgtvnic->mgt_cq.cq_size]);
+		mgt_context->context_words[NES_NIC_CTX_MISC_IDX] =
+			cpu_to_le32((u32)NES_MGT_CTX_SIZE |
+				    ((u32)PCI_FUNC(nesdev->pcidev->devfn) << 12));
+		nes_debug(NES_DBG_INIT, "RX_WINDOW_BUFFER_PAGE_TABLE_SIZE = 0x%08X, RX_WINDOW_BUFFER_SIZE = 0x%08X\n",
+			  nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_PAGE_TABLE_SIZE),
+			  nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_SIZE));
+		if (nes_read_indexed(nesdev, NES_IDX_RX_WINDOW_BUFFER_SIZE) != 0)
+			mgt_context->context_words[NES_NIC_CTX_MISC_IDX] |= cpu_to_le32(NES_NIC_BACK_STORE);
+
+		u64temp = (u64)mgtvnic->mgt.rq_pbase;
+		mgt_context->context_words[NES_NIC_CTX_SQ_LOW_IDX] = cpu_to_le32((u32)u64temp);
+		mgt_context->context_words[NES_NIC_CTX_SQ_HIGH_IDX] = cpu_to_le32((u32)(u64temp >> 32));
+		u64temp = (u64)mgtvnic->mgt.rq_pbase;
+		mgt_context->context_words[NES_NIC_CTX_RQ_LOW_IDX] = cpu_to_le32((u32)u64temp);
+		mgt_context->context_words[NES_NIC_CTX_RQ_HIGH_IDX] = cpu_to_le32((u32)(u64temp >> 32));
+
+		cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] = cpu_to_le32(NES_CQP_CREATE_QP |
+									 NES_CQP_QP_TYPE_NIC);
+		cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX] = cpu_to_le32(mgtvnic->mgt.qp_id);
+		u64temp = (u64)mgtvnic->mgt_cq.cq_pbase +
+			  (mgtvnic->mgt_cq.cq_size * sizeof(struct nes_hw_nic_cqe));
+		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, u64temp);
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+		nesdev->cqp.sq_head = cqp_head;
+
+		barrier();
+
+		/* Ring doorbell (2 WQEs) */
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, 0x02800000 | nesdev->cqp.qp_id);
+
+		spin_unlock_irqrestore(&nesdev->cqp.lock, flags);
+		nes_debug(NES_DBG_INIT, "Waiting for create MGT QP%u to complete.\n",
+			  mgtvnic->mgt.qp_id);
+
+		ret = wait_event_timeout(nesdev->cqp.waitq, (nesdev->cqp.sq_tail == cqp_head),
+					 NES_EVENT_TIMEOUT);
+		nes_debug(NES_DBG_INIT, "Create MGT QP%u completed, wait_event_timeout ret = %u.\n",
+			  mgtvnic->mgt.qp_id, ret);
+		if (!ret) {
+			nes_debug(NES_DBG_INIT, "MGT QP%u create timeout expired\n", mgtvnic->mgt.qp_id);
+			if (i == 0) {
+				pci_free_consistent(nesdev->pcidev, nesvnic->mgt_mem_size, nesvnic->mgt_vbase,
+						    nesvnic->mgt_pbase);
+				kfree(mgtvnic);
+			} else {
+				nes_destroy_mgt(nesvnic);
+			}
+			return -EIO;
+		}
+
+		/* Populate the RQ */
+		for (counter = 0; counter < (NES_MGT_WQ_COUNT - 1); counter++) {
+			skb = dev_alloc_skb(nesvnic->max_frame_size);
+			if (!skb) {
+				nes_debug(NES_DBG_INIT, "%s: out of memory for receive skb\n", netdev->name);
+				return -ENOMEM;
+			}
+
+			skb->dev = netdev;
+
+			pmem = pci_map_single(nesdev->pcidev, skb->data,
+					      nesvnic->max_frame_size, PCI_DMA_FROMDEVICE);
+			cb = (struct nes_rskb_cb *)&skb->cb[0];
+			cb->busaddr = pmem;
+			cb->maplen = nesvnic->max_frame_size;
+
+			mgt_rqe = &mgtvnic->mgt.rq_vbase[counter];
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_1_0_IDX] = cpu_to_le32((u32)nesvnic->max_frame_size);
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_LENGTH_3_2_IDX] = 0;
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_LOW_IDX] = cpu_to_le32((u32)pmem);
+			mgt_rqe->wqe_words[NES_NIC_RQ_WQE_FRAG0_HIGH_IDX] = cpu_to_le32((u32)((u64)pmem >> 32));
+			mgtvnic->mgt.rx_skb[counter] = skb;
+		}
+
+		init_timer(&mgtvnic->rq_wqes_timer);
+		mgtvnic->rq_wqes_timer.function = nes_mgt_rq_wqes_timeout;
+		mgtvnic->rq_wqes_timer.data = (unsigned long)mgtvnic;
+
+		wqe_count = NES_MGT_WQ_COUNT - 1;
+		mgtvnic->mgt.rq_head = wqe_count;
+		barrier();
+		do {
+			counter = min(wqe_count, ((u32)255));
+			wqe_count -= counter;
+			nes_write32(nesdev->regs + NES_WQE_ALLOC, (counter << 24) | mgtvnic->mgt.qp_id);
+		} while (wqe_count);
+
+		nes_write32(nesdev->regs + NES_CQE_ALLOC, NES_CQE_ALLOC_NOTIFY_NEXT |
+			    mgtvnic->mgt_cq.cq_number);
+		nes_read32(nesdev->regs + NES_CQE_ALLOC);
+
+		mgt_vbase += mgt_mem_size;
+		mgt_pbase += mgt_mem_size;
+		nesvnic->mgtvnic[i] = mgtvnic++;
+	}
+	return 0;
+}
+
+
+void nes_destroy_mgt(struct nes_vnic *nesvnic)
+{
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct nes_vnic_mgt *mgtvnic;
+	struct nes_vnic_mgt *first_mgtvnic;
+	unsigned long flags;
+	struct nes_hw_cqp_wqe *cqp_wqe;
+	u32 cqp_head;
+	struct sk_buff *rx_skb;
+	int i;
+	int ret;
+
+	kthread_stop(nesvnic->mgt_thread);
+
+	/* Free remaining NIC receive buffers */
+	first_mgtvnic = nesvnic->mgtvnic[0];
+	for (i = 0; i < NES_MGT_QP_COUNT; i++) {
+		mgtvnic = nesvnic->mgtvnic[i];
+		if (mgtvnic == NULL)
+			continue;
+
+		while (mgtvnic->mgt.rq_head != mgtvnic->mgt.rq_tail) {
+			rx_skb = mgtvnic->mgt.rx_skb[mgtvnic->mgt.rq_tail];
+			nes_mgt_free_skb(nesdev, rx_skb, PCI_DMA_FROMDEVICE);
+			mgtvnic->mgt.rq_tail++;
+			mgtvnic->mgt.rq_tail &= (mgtvnic->mgt.rq_size - 1);
+		}
+
+		spin_lock_irqsave(&nesdev->cqp.lock, flags);
+
+		/* Destroy NIC QP */
+		cqp_head = nesdev->cqp.sq_head;
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX,
+				    (NES_CQP_DESTROY_QP | NES_CQP_QP_TYPE_NIC));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX,
+				    mgtvnic->mgt.qp_id);
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+
+		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
+
+		/* Destroy NIC CQ */
+		nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX,
+				    (NES_CQP_DESTROY_CQ | ((u32)mgtvnic->mgt_cq.cq_size << 16)));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX,
+				    (mgtvnic->mgt_cq.cq_number | ((u32)nesdev->ceq_index << 16)));
+
+		if (++cqp_head >= nesdev->cqp.sq_size)
+			cqp_head = 0;
+
+		nesdev->cqp.sq_head = cqp_head;
+		barrier();
+
+		/* Ring doorbell (2 WQEs) */
+		nes_write32(nesdev->regs + NES_WQE_ALLOC, 0x02800000 | nesdev->cqp.qp_id);
+
+		spin_unlock_irqrestore(&nesdev->cqp.lock, flags);
+		nes_debug(NES_DBG_SHUTDOWN, "Waiting for CQP, cqp_head=%u, cqp.sq_head=%u,"
+			  " cqp.sq_tail=%u, cqp.sq_size=%u\n",
+			  cqp_head, nesdev->cqp.sq_head,
+			  nesdev->cqp.sq_tail, nesdev->cqp.sq_size);
+
+		ret = wait_event_timeout(nesdev->cqp.waitq, (nesdev->cqp.sq_tail == cqp_head),
+					 NES_EVENT_TIMEOUT);
+
+		nes_debug(NES_DBG_SHUTDOWN, "Destroy MGT QP returned, wait_event_timeout ret = %u, cqp_head=%u,"
+			  " cqp.sq_head=%u, cqp.sq_tail=%u\n",
+			  ret, cqp_head, nesdev->cqp.sq_head, nesdev->cqp.sq_tail);
+		if (!ret)
+			nes_debug(NES_DBG_SHUTDOWN, "MGT QP%u destroy timeout expired\n",
+				  mgtvnic->mgt.qp_id);
+
+		nesvnic->mgtvnic[i] = NULL;
+	}
+
+	if (nesvnic->mgt_vbase) {
+		pci_free_consistent(nesdev->pcidev, nesvnic->mgt_mem_size, nesvnic->mgt_vbase,
+				    nesvnic->mgt_pbase);
+		nesvnic->mgt_vbase = NULL;
+		nesvnic->mgt_pbase = 0;
+	}
+
+	kfree(first_mgtvnic);
+}
diff --git a/drivers/infiniband/hw/nes/nes_mgt.h b/drivers/infiniband/hw/nes/nes_mgt.h
new file mode 100644
index 00000000000..4f7f701c4a8
--- /dev/null
+++ b/drivers/infiniband/hw/nes/nes_mgt.h
@@ -0,0 +1,97 @@
+/*
+* Copyright (c) 2006 - 2011 Intel-NE, Inc.  All rights reserved.
+*
+* This software is available to you under a choice of one of two
+* licenses.  You may choose to be licensed under the terms of the GNU
+* General Public License (GPL) Version 2, available from the file
+* COPYING in the main directory of this source tree, or the
+* OpenIB.org BSD license below:
+*
+*     Redistribution and use in source and binary forms, with or
+*     without modification, are permitted provided that the following
+*     conditions are met:
+*
+*      - Redistributions of source code must retain the above
+*        copyright notice, this list of conditions and the following
+*        disclaimer.
+*
+*      - Redistributions in binary form must reproduce the above
+*        copyright notice, this list of conditions and the following
+*        disclaimer in the documentation and/or other materials
+*        provided with the distribution.
+*
+* THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+* EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+* MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+* NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+* BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+* ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+* CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+* SOFTWARE.
+*/
+
+#ifndef __NES_MGT_H
+#define __NES_MGT_H
+
+#define MPA_FRAMING 6	/* length is 2 bytes, crc is 4 bytes */
+
+int nes_init_mgt_qp(struct nes_device *nesdev, struct net_device *netdev, struct nes_vnic *nesvnic);
+void nes_queue_mgt_skbs(struct sk_buff *skb, struct nes_vnic *nesvnic, struct nes_qp *nesqp);
+void nes_destroy_mgt(struct nes_vnic *nesvnic);
+void nes_destroy_pau_qp(struct nes_device *nesdev, struct nes_qp *nesqp);
+
+struct nes_hw_mgt {
+	struct nes_hw_nic_rq_wqe *rq_vbase;	/* virtual address of rq */
+	dma_addr_t rq_pbase;			/* PCI memory for host rings */
+	struct sk_buff *rx_skb[NES_NIC_WQ_SIZE];
+	u16 qp_id;
+	u16 sq_head;
+	u16 rq_head;
+	u16 rq_tail;
+	u16 rq_size;
+	u8 replenishing_rq;
+	u8 reserved;
+	spinlock_t rq_lock;
+};
+
+struct nes_vnic_mgt {
+	struct nes_vnic        *nesvnic;
+	struct nes_hw_mgt      mgt;
+	struct nes_hw_nic_cq   mgt_cq;
+	atomic_t               rx_skbs_needed;
+	struct timer_list      rq_wqes_timer;
+	atomic_t               rx_skb_timer_running;
+};
+
+#define MAX_FPDU_FRAGS 4
+struct pau_fpdu_frag {
+	struct sk_buff         *skb;
+	u64                    physaddr;
+	u32                    frag_len;
+	bool                   cmplt;
+};
+
+struct pau_fpdu_info {
+	struct nes_qp          *nesqp;
+	struct nes_cqp_request *cqp_request;
+	void                   *hdr_vbase;
+	dma_addr_t             hdr_pbase;
+	int                    hdr_len;
+	u16                    data_len;
+	u16                    frag_cnt;
+	struct pau_fpdu_frag   frags[MAX_FPDU_FRAGS];
+};
+
+enum pau_qh_state {
+	PAU_DEL_QH,
+	PAU_ADD_LB_QH,
+	PAU_READY
+};
+
+struct pau_qh_chg {
+	struct nes_device *nesdev;
+	struct nes_vnic *nesvnic;
+	struct nes_qp *nesqp;
+};
+
+#endif          /* __NES_MGT_H */
diff --git a/drivers/infiniband/hw/nes/nes_nic.c b/drivers/infiniband/hw/nes/nes_nic.c
index 1b0938c8777..49eb5111d2c 100644
--- a/drivers/infiniband/hw/nes/nes_nic.c
+++ b/drivers/infiniband/hw/nes/nes_nic.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -40,6 +40,7 @@
 #include <linux/if_arp.h>
 #include <linux/if_vlan.h>
 #include <linux/ethtool.h>
+#include <linux/slab.h>
 #include <net/tcp.h>
 
 #include <net/inet_common.h>
@@ -91,6 +92,7 @@ static struct nic_qp_map *nic_qp_mapping_per_function[] = {
 static const u32 default_msg = NETIF_MSG_DRV | NETIF_MSG_PROBE | NETIF_MSG_LINK
 		| NETIF_MSG_IFUP | NETIF_MSG_IFDOWN;
 static int debug = -1;
+static int nics_per_function = 1;
 
 /**
  * nes_netdev_poll
@@ -98,7 +100,6 @@ static int debug = -1;
 static int nes_netdev_poll(struct napi_struct *napi, int budget)
 {
 	struct nes_vnic *nesvnic = container_of(napi, struct nes_vnic, napi);
-	struct net_device *netdev = nesvnic->netdev;
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_hw_nic_cq *nescq = &nesvnic->nic_cq;
 
@@ -111,7 +112,7 @@ static int nes_netdev_poll(struct napi_struct *napi, int budget)
 	nes_nic_ce_handler(nesdev, nescq);
 
 	if (nescq->cqes_pending == 0) {
-		netif_rx_complete(netdev, napi);
+		napi_complete(napi);
 		/* clear out completed cqes and arm */
 		nes_write32(nesdev->regs+NES_CQE_ALLOC, NES_CQE_ALLOC_NOTIFY_NEXT |
 				nescq->cq_number | (nescq->cqe_allocs_pending << 16));
@@ -143,6 +144,7 @@ static int nes_netdev_open(struct net_device *netdev)
 	u32 nic_active_bit;
 	u32 nic_active;
 	struct list_head *list_pos, *list_temp;
+	unsigned long flags;
 
 	assert(nesdev != NULL);
 
@@ -201,7 +203,8 @@ static int nes_netdev_open(struct net_device *netdev)
 		nes_debug(NES_DBG_NETDEV, "i=%d, perfect filter table index= %d, PERF FILTER LOW"
 				" (Addr:%08X) = %08X, HIGH = %08X.\n",
 				i, nesvnic->qp_nic_index[i],
-				NES_IDX_PERFECT_FILTER_LOW+((nesvnic->perfect_filter_index + i) * 8),
+				NES_IDX_PERFECT_FILTER_LOW+
+					(nesvnic->qp_nic_index[i] * 8),
 				macaddr_low,
 				(u32)macaddr_high | NES_MAC_ADDR_VALID |
 				((((u32)nesvnic->nic_index) << 16)));
@@ -230,12 +233,36 @@ static int nes_netdev_open(struct net_device *netdev)
 				NES_MAC_INT_TX_UNDERFLOW | NES_MAC_INT_TX_ERROR));
 		first_nesvnic = nesvnic;
 	}
+
 	if (first_nesvnic->linkup) {
 		/* Enable network packets */
 		nesvnic->linkup = 1;
 		netif_start_queue(netdev);
 		netif_carrier_on(netdev);
 	}
+
+	spin_lock_irqsave(&nesdev->nesadapter->phy_lock, flags);
+	if (nesdev->nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_SFP_D) {
+		nesdev->link_recheck = 1;
+		mod_delayed_work(system_wq, &nesdev->work,
+				 NES_LINK_RECHECK_DELAY);
+	}
+	spin_unlock_irqrestore(&nesdev->nesadapter->phy_lock, flags);
+
+	spin_lock_irqsave(&nesvnic->port_ibevent_lock, flags);
+	if (nesvnic->of_device_registered) {
+		nesdev->nesadapter->send_term_ok = 1;
+		if (nesvnic->linkup == 1) {
+			if (nesdev->iw_status == 0) {
+				nesdev->iw_status = 1;
+				nes_port_ibevent(nesvnic);
+			}
+		} else {
+			nesdev->iw_status = 0;
+		}
+	}
+	spin_unlock_irqrestore(&nesvnic->port_ibevent_lock, flags);
+
 	napi_enable(&nesvnic->napi);
 	nesvnic->netdev_open = 1;
 
@@ -254,6 +281,7 @@ static int nes_netdev_stop(struct net_device *netdev)
 	u32 nic_active;
 	struct nes_vnic *first_nesvnic = NULL;
 	struct list_head *list_pos, *list_temp;
+	unsigned long flags;
 
 	nes_debug(NES_DBG_SHUTDOWN, "nesvnic=%p, nesdev=%p, netdev=%p %s\n",
 			nesvnic, nesdev, netdev, netdev->name);
@@ -262,6 +290,7 @@ static int nes_netdev_stop(struct net_device *netdev)
 
 	if (netif_msg_ifdown(nesvnic))
 		printk(KERN_INFO PFX "%s: disabling interface\n", netdev->name);
+	netif_carrier_off(netdev);
 
 	/* Disable network packets */
 	napi_disable(&nesvnic->napi);
@@ -272,14 +301,18 @@ static int nes_netdev_stop(struct net_device *netdev)
 			break;
 	}
 
-	if (first_nesvnic->netdev_open == 0)
+	if ((first_nesvnic->netdev_open == 1) && (first_nesvnic != nesvnic)  &&
+		(PCI_FUNC(first_nesvnic->nesdev->pcidev->devfn) !=
+		PCI_FUNC(nesvnic->nesdev->pcidev->devfn))) {
+			nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK+
+				(0x200*nesdev->mac_index), 0xffffffff);
+			nes_write_indexed(first_nesvnic->nesdev,
+				NES_IDX_MAC_INT_MASK+
+				(0x200*first_nesvnic->nesdev->mac_index),
+			~(NES_MAC_INT_LINK_STAT_CHG | NES_MAC_INT_XGMII_EXT |
+			NES_MAC_INT_TX_UNDERFLOW | NES_MAC_INT_TX_ERROR));
+	} else {
 		nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK+(0x200*nesdev->mac_index), 0xffffffff);
-	else if ((first_nesvnic != nesvnic) &&
-		 (PCI_FUNC(first_nesvnic->nesdev->pcidev->devfn) != PCI_FUNC(nesvnic->nesdev->pcidev->devfn))) {
-		nes_write_indexed(nesdev, NES_IDX_MAC_INT_MASK + (0x200 * nesdev->mac_index), 0xffffffff);
-		nes_write_indexed(first_nesvnic->nesdev, NES_IDX_MAC_INT_MASK + (0x200 * first_nesvnic->nesdev->mac_index),
-				~(NES_MAC_INT_LINK_STAT_CHG | NES_MAC_INT_XGMII_EXT |
-				NES_MAC_INT_TX_UNDERFLOW | NES_MAC_INT_TX_ERROR));
 	}
 
 	nic_active_mask = ~((u32)(1 << nesvnic->nic_index));
@@ -301,12 +334,17 @@ static int nes_netdev_stop(struct net_device *netdev)
 	nic_active &= nic_active_mask;
 	nes_write_indexed(nesdev, NES_IDX_NIC_BROADCAST_ON, nic_active);
 
-
+	spin_lock_irqsave(&nesvnic->port_ibevent_lock, flags);
 	if (nesvnic->of_device_registered) {
-		nes_destroy_ofa_device(nesvnic->nesibdev);
-		nesvnic->nesibdev = NULL;
-		nesvnic->of_device_registered = 0;
+		nesdev->nesadapter->send_term_ok = 0;
+		nesdev->iw_status = 0;
+		if (nesvnic->linkup == 1)
+			nes_port_ibevent(nesvnic);
 	}
+	del_timer_sync(&nesvnic->event_timer);
+	nesvnic->event_timer.function = NULL;
+	spin_unlock_irqrestore(&nesvnic->port_ibevent_lock, flags);
+
 	nes_destroy_nic_qp(nesvnic);
 
 	nesvnic->netdev_open = 0;
@@ -346,24 +384,20 @@ static int nes_nic_send(struct sk_buff *skb, struct net_device *netdev)
 	/* bump past the vlan tag */
 	wqe_fragment_length++;
 	/*	wqe_fragment_address = (u64 *)&nic_sqe->wqe_words[NES_NIC_SQ_WQE_FRAG0_LOW_IDX]; */
+	wqe_misc |= NES_NIC_SQ_WQE_COMPLETION;
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
-		tcph = tcp_hdr(skb);
-		if (1) {
-			if (skb_is_gso(skb)) {
-				/* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... seg size = %u\n",
-						netdev->name, skb_is_gso(skb)); */
-				wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE |
-						NES_NIC_SQ_WQE_COMPLETION | (u16)skb_is_gso(skb);
-				set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_LSO_INFO_IDX,
-						((u32)tcph->doff) |
-						(((u32)(((unsigned char *)tcph) - skb->data)) << 4));
-			} else {
-				wqe_misc |= NES_NIC_SQ_WQE_COMPLETION;
-			}
+		if (skb_is_gso(skb)) {
+			tcph = tcp_hdr(skb);
+			/* nes_debug(NES_DBG_NIC_TX, "%s: TSO request... is_gso = %u seg size = %u\n",
+					netdev->name, skb_is_gso(skb), skb_shinfo(skb)->gso_size); */
+			wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE | (u16)skb_shinfo(skb)->gso_size;
+			set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_LSO_INFO_IDX,
+					((u32)tcph->doff) |
+					(((u32)(((unsigned char *)tcph) - skb->data)) << 4));
 		}
 	} else {	/* CHECKSUM_HW */
-		wqe_misc |= NES_NIC_SQ_WQE_DISABLE_CHKSUM | NES_NIC_SQ_WQE_COMPLETION;
+		wqe_misc |= NES_NIC_SQ_WQE_DISABLE_CHKSUM;
 	}
 
 	set_wqe_32bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_TOTAL_LENGTH_IDX,
@@ -395,21 +429,20 @@ static int nes_nic_send(struct sk_buff *skb, struct net_device *netdev)
 	if (skb_headlen(skb) == skb->len) {
 		if (skb_headlen(skb) <= NES_FIRST_FRAG_SIZE) {
 			nic_sqe->wqe_words[NES_NIC_SQ_WQE_LENGTH_2_1_IDX] = 0;
-			nesnic->tx_skb[nesnic->sq_head] = NULL;
-			dev_kfree_skb(skb);
+			nesnic->tx_skb[nesnic->sq_head] = skb;
 		}
 	} else {
 		/* Deal with Fragments */
 		nesnic->tx_skb[nesnic->sq_head] = skb;
 		for (skb_fragment_index = 0; skb_fragment_index < skb_shinfo(skb)->nr_frags;
 				skb_fragment_index++) {
-			bus_address = pci_map_page( nesdev->pcidev,
-					skb_shinfo(skb)->frags[skb_fragment_index].page,
-					skb_shinfo(skb)->frags[skb_fragment_index].page_offset,
-					skb_shinfo(skb)->frags[skb_fragment_index].size,
-					PCI_DMA_TODEVICE);
+			skb_frag_t *frag =
+				&skb_shinfo(skb)->frags[skb_fragment_index];
+			bus_address = skb_frag_dma_map(&nesdev->pcidev->dev,
+						       frag, 0, skb_frag_size(frag),
+						       DMA_TO_DEVICE);
 			wqe_fragment_length[wqe_fragment_index] =
-					cpu_to_le16(skb_shinfo(skb)->frags[skb_fragment_index].size);
+					cpu_to_le16(skb_frag_size(&skb_shinfo(skb)->frags[skb_fragment_index]));
 			set_wqe_64bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_LOW_IDX+(2*wqe_fragment_index),
 				bus_address);
 			wqe_fragment_index++;
@@ -437,7 +470,7 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 	struct nes_hw_nic_sq_wqe *nic_sqe;
 	struct tcphdr *tcph;
 	/* struct udphdr *udph; */
-#define NES_MAX_TSO_FRAGS 18
+#define NES_MAX_TSO_FRAGS MAX_SKB_FRAGS
 	/* 64K segment plus overflow on each side */
 	dma_addr_t tso_bus_address[NES_MAX_TSO_FRAGS];
 	dma_addr_t bus_address;
@@ -448,7 +481,6 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 	u32 wqe_count=1;
 	u32 send_rc;
 	struct iphdr *iph;
-	unsigned long flags;
 	__le16 *wqe_fragment_length;
 	u32 nr_frags;
 	u32 original_first_length;
@@ -459,7 +491,6 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 	u16 nhoffset;
 	u16 wqes_needed;
 	u16 wqes_available;
-	u32 old_head;
 	u32 wqe_misc;
 
 	/*
@@ -475,13 +506,6 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 	if (netif_queue_stopped(netdev))
 		return NETDEV_TX_BUSY;
 
-	local_irq_save(flags);
-	if (!spin_trylock(&nesnic->sq_lock)) {
-		local_irq_restore(flags);
-		nesvnic->sq_locked++;
-		return NETDEV_TX_LOCKED;
-	}
-
 	/* Check if SQ is full */
 	if ((((nesnic->sq_tail+(nesnic->sq_size*2))-nesnic->sq_head) & (nesnic->sq_size - 1)) == 1) {
 		if (!netif_queue_stopped(netdev)) {
@@ -493,7 +517,6 @@ static int nes_netdev_start_xmit(struct sk_buff *skb, struct net_device *netdev)
 			}
 		}
 		nesvnic->sq_full++;
-		spin_unlock_irqrestore(&nesnic->sq_lock, flags);
 		return NETDEV_TX_BUSY;
 	}
 
@@ -507,7 +530,6 @@ sq_no_longer_full:
 		if (skb_is_gso(skb)) {
 			nesvnic->segmented_tso_requests++;
 			nesvnic->tso_requests++;
-			old_head = nesnic->sq_head;
 			/* Basically 4 fragments available per WQE with extended fragments */
 			wqes_needed = nr_frags >> 2;
 			wqes_needed += (nr_frags&3)?1:0;
@@ -526,7 +548,6 @@ sq_no_longer_full:
 					}
 				}
 				nesvnic->sq_full++;
-				spin_unlock_irqrestore(&nesnic->sq_lock, flags);
 				nes_debug(NES_DBG_NIC_TX, "%s: HNIC SQ full- TSO request has too many frags!\n",
 						netdev->name);
 				return NETDEV_TX_BUSY;
@@ -535,11 +556,12 @@ tso_sq_no_longer_full:
 			/* Map all the buffers */
 			for (tso_frag_count=0; tso_frag_count < skb_shinfo(skb)->nr_frags;
 					tso_frag_count++) {
-				tso_bus_address[tso_frag_count] = pci_map_page( nesdev->pcidev,
-						skb_shinfo(skb)->frags[tso_frag_count].page,
-						skb_shinfo(skb)->frags[tso_frag_count].page_offset,
-						skb_shinfo(skb)->frags[tso_frag_count].size,
-						PCI_DMA_TODEVICE);
+				skb_frag_t *frag =
+					&skb_shinfo(skb)->frags[tso_frag_count];
+				tso_bus_address[tso_frag_count] =
+					skb_frag_dma_map(&nesdev->pcidev->dev,
+							 frag, 0, skb_frag_size(frag),
+							 DMA_TO_DEVICE);
 			}
 
 			tso_frag_index = 0;
@@ -570,10 +592,10 @@ tso_sq_no_longer_full:
 					nes_debug(NES_DBG_NIC_TX, "ERROR: SKB header too big, headlen=%u, FIRST_FRAG_SIZE=%u\n",
 							original_first_length, NES_FIRST_FRAG_SIZE);
 					nes_debug(NES_DBG_NIC_TX, "%s Request to tx NIC packet length %u, headlen %u,"
-							" (%u frags), tso_size=%u\n",
+							" (%u frags), is_gso = %u tso_size=%u\n",
 							netdev->name,
 							skb->len, skb_headlen(skb),
-							skb_shinfo(skb)->nr_frags, skb_is_gso(skb));
+							skb_shinfo(skb)->nr_frags, skb_is_gso(skb), skb_shinfo(skb)->gso_size);
 				}
 				memcpy(&nesnic->first_frag_vbase[nesnic->sq_head].buffer,
 						skb->data, min(((unsigned int)NES_FIRST_FRAG_SIZE),
@@ -605,14 +627,16 @@ tso_sq_no_longer_full:
 					wqe_fragment_length[wqe_fragment_index] = 0;
 					set_wqe_64bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_FRAG1_LOW_IDX,
 									bus_address);
+					tso_wqe_length += skb_headlen(skb) -
+							original_first_length;
 				}
 				while (wqe_fragment_index < 5) {
 					wqe_fragment_length[wqe_fragment_index] =
-							cpu_to_le16(skb_shinfo(skb)->frags[tso_frag_index].size);
+							cpu_to_le16(skb_frag_size(&skb_shinfo(skb)->frags[tso_frag_index]));
 					set_wqe_64bit_value(nic_sqe->wqe_words, NES_NIC_SQ_WQE_FRAG0_LOW_IDX+(2*wqe_fragment_index),
 						(u64)tso_bus_address[tso_frag_index]);
 					wqe_fragment_index++;
-					tso_wqe_length += skb_shinfo(skb)->frags[tso_frag_index++].size;
+					tso_wqe_length += skb_frag_size(&skb_shinfo(skb)->frags[tso_frag_index++]);
 					if (wqe_fragment_index < 5)
 						wqe_fragment_length[wqe_fragment_index] = 0;
 					if (tso_frag_index == tso_frag_count)
@@ -623,8 +647,8 @@ tso_sq_no_longer_full:
 				} else {
 					nesnic->tx_skb[nesnic->sq_head] = NULL;
 				}
-				wqe_misc |= NES_NIC_SQ_WQE_COMPLETION | (u16)skb_is_gso(skb);
-				if ((tso_wqe_length + original_first_length) > skb_is_gso(skb)) {
+				wqe_misc |= NES_NIC_SQ_WQE_COMPLETION | (u16)skb_shinfo(skb)->gso_size;
+				if ((tso_wqe_length + original_first_length) > skb_shinfo(skb)->gso_size) {
 					wqe_misc |= NES_NIC_SQ_WQE_LSO_ENABLE;
 				} else {
 					iph->tot_len = htons(tso_wqe_length + original_first_length - nhoffset);
@@ -649,17 +673,13 @@ tso_sq_no_longer_full:
 			skb_set_transport_header(skb, hoffset);
 			skb_set_network_header(skb, nhoffset);
 			send_rc = nes_nic_send(skb, netdev);
-			if (send_rc != NETDEV_TX_OK) {
-				spin_unlock_irqrestore(&nesnic->sq_lock, flags);
+			if (send_rc != NETDEV_TX_OK)
 				return NETDEV_TX_OK;
-			}
 		}
 	} else {
 		send_rc = nes_nic_send(skb, netdev);
-		if (send_rc != NETDEV_TX_OK) {
-			spin_unlock_irqrestore(&nesnic->sq_lock, flags);
+		if (send_rc != NETDEV_TX_OK)
 			return NETDEV_TX_OK;
-		}
 	}
 
 	barrier();
@@ -669,7 +689,6 @@ tso_sq_no_longer_full:
 				(wqe_count << 24) | (1 << 23) | nesvnic->nic.qp_id);
 
 	netdev->trans_start = jiffies;
-	spin_unlock_irqrestore(&nesnic->sq_lock, flags);
 
 	return NETDEV_TX_OK;
 }
@@ -789,14 +808,13 @@ static int nes_netdev_set_mac_address(struct net_device *netdev, void *p)
 	int i;
 	u32 macaddr_low;
 	u16 macaddr_high;
-	DECLARE_MAC_BUF(mac);
 
 	if (!is_valid_ether_addr(mac_addr->sa_data))
 		return -EADDRNOTAVAIL;
 
 	memcpy(netdev->dev_addr, mac_addr->sa_data, netdev->addr_len);
-	printk(PFX "%s: Address length = %d, Address = %s\n",
-	       __func__, netdev->addr_len, print_mac(mac, mac_addr->sa_data));
+	printk(PFX "%s: Address length = %d, Address = %pM\n",
+	       __func__, netdev->addr_len, mac_addr->sa_data);
 	macaddr_high  = ((u16)netdev->dev_addr[0]) << 8;
 	macaddr_high += (u16)netdev->dev_addr[1];
 	macaddr_low   = ((u32)netdev->dev_addr[2]) << 24;
@@ -820,6 +838,20 @@ static int nes_netdev_set_mac_address(struct net_device *netdev, void *p)
 }
 
 
+static void set_allmulti(struct nes_device *nesdev, u32 nic_active_bit)
+{
+	u32 nic_active;
+
+	nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL);
+	nic_active |= nic_active_bit;
+	nes_write_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL, nic_active);
+	nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL);
+	nic_active &= ~nic_active_bit;
+	nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
+}
+
+#define get_addr(addrs, index) ((addrs) + (index) * ETH_ALEN)
+
 /**
  * nes_netdev_set_multicast_list
  */
@@ -827,7 +859,7 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 {
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
 	struct nes_device *nesdev = nesvnic->nesdev;
-	struct dev_mc_list *multicast_addr;
+	struct nes_adapter *nesadapter = nesvnic->nesdev->nesadapter;
 	u32 nic_active_bit;
 	u32 nic_active;
 	u32 perfect_filter_register_address;
@@ -836,7 +868,13 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 	u8 mc_all_on = 0;
 	u8 mc_index;
 	int mc_nic_index = -1;
+	u8 pft_entries_preallocated = max(nesadapter->adapter_fcn_count *
+					nics_per_function, 4);
+	u8 max_pft_entries_avaiable = NES_PFT_SIZE - pft_entries_preallocated;
+	unsigned long flags;
+	int mc_count = netdev_mc_count(netdev);
 
+	spin_lock_irqsave(&nesadapter->resource_lock, flags);
 	nic_active_bit = 1 << nesvnic->nic_index;
 
 	if (netdev->flags & IFF_PROMISC) {
@@ -847,14 +885,9 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 		nic_active |= nic_active_bit;
 		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
 		mc_all_on = 1;
-	} else if ((netdev->flags & IFF_ALLMULTI) || (netdev->mc_count > NES_MULTICAST_PF_MAX) ||
+	} else if ((netdev->flags & IFF_ALLMULTI) ||
 			   (nesvnic->nic_index > 3)) {
-		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL);
-		nic_active |= nic_active_bit;
-		nes_write_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL, nic_active);
-		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL);
-		nic_active &= ~nic_active_bit;
-		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
+		set_allmulti(nesdev, nic_active_bit);
 		mc_all_on = 1;
 	} else {
 		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL);
@@ -865,30 +898,59 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
 	}
 
-	nes_debug(NES_DBG_NIC_RX, "Number of MC entries = %d, Promiscous = %d, All Multicast = %d.\n",
-			  netdev->mc_count, (netdev->flags & IFF_PROMISC)?1:0,
-			  (netdev->flags & IFF_ALLMULTI)?1:0);
+	nes_debug(NES_DBG_NIC_RX, "Number of MC entries = %d, Promiscuous = %d, All Multicast = %d.\n",
+		  mc_count, !!(netdev->flags & IFF_PROMISC),
+		  !!(netdev->flags & IFF_ALLMULTI));
 	if (!mc_all_on) {
-		multicast_addr = netdev->mc_list;
-		perfect_filter_register_address = NES_IDX_PERFECT_FILTER_LOW + 0x80;
-		perfect_filter_register_address += nesvnic->nic_index*0x40;
-		for (mc_index=0; mc_index < NES_MULTICAST_PF_MAX; mc_index++) {
-			while (multicast_addr && nesvnic->mcrq_mcast_filter && ((mc_nic_index = nesvnic->mcrq_mcast_filter(nesvnic, multicast_addr->dmi_addr)) == 0))
-				multicast_addr = multicast_addr->next;
+		char *addrs;
+		int i;
+		struct netdev_hw_addr *ha;
+
+		addrs = kmalloc(ETH_ALEN * mc_count, GFP_ATOMIC);
+		if (!addrs) {
+			set_allmulti(nesdev, nic_active_bit);
+			goto unlock;
+		}
+		i = 0;
+		netdev_for_each_mc_addr(ha, netdev)
+			memcpy(get_addr(addrs, i++), ha->addr, ETH_ALEN);
+
+		perfect_filter_register_address = NES_IDX_PERFECT_FILTER_LOW +
+						pft_entries_preallocated * 0x8;
+		for (i = 0, mc_index = 0; mc_index < max_pft_entries_avaiable;
+		     mc_index++) {
+			while (i < mc_count && nesvnic->mcrq_mcast_filter &&
+			((mc_nic_index = nesvnic->mcrq_mcast_filter(nesvnic,
+					get_addr(addrs, i++))) == 0));
 			if (mc_nic_index < 0)
 				mc_nic_index = nesvnic->nic_index;
-			if (multicast_addr) {
-				DECLARE_MAC_BUF(mac);
-				nes_debug(NES_DBG_NIC_RX, "Assigning MC Address %s to register 0x%04X nic_idx=%d\n",
-					  print_mac(mac, multicast_addr->dmi_addr),
+			while (nesadapter->pft_mcast_map[mc_index] < 16 &&
+				nesadapter->pft_mcast_map[mc_index] !=
+					nesvnic->nic_index &&
+					mc_index < max_pft_entries_avaiable) {
+						nes_debug(NES_DBG_NIC_RX,
+					"mc_index=%d skipping nic_index=%d, "
+					"used for=%d \n", mc_index,
+					nesvnic->nic_index,
+					nesadapter->pft_mcast_map[mc_index]);
+				mc_index++;
+			}
+			if (mc_index >= max_pft_entries_avaiable)
+				break;
+			if (i < mc_count) {
+				char *addr = get_addr(addrs, i++);
+
+				nes_debug(NES_DBG_NIC_RX, "Assigning MC Address %pM to register 0x%04X nic_idx=%d\n",
+					  addr,
 					  perfect_filter_register_address+(mc_index * 8),
 					  mc_nic_index);
-				macaddr_high  = ((u16)multicast_addr->dmi_addr[0]) << 8;
-				macaddr_high += (u16)multicast_addr->dmi_addr[1];
-				macaddr_low   = ((u32)multicast_addr->dmi_addr[2]) << 24;
-				macaddr_low  += ((u32)multicast_addr->dmi_addr[3]) << 16;
-				macaddr_low  += ((u32)multicast_addr->dmi_addr[4]) << 8;
-				macaddr_low  += (u32)multicast_addr->dmi_addr[5];
+				macaddr_high  = ((u8) addr[0]) << 8;
+				macaddr_high += (u8) addr[1];
+				macaddr_low   = ((u8) addr[2]) << 24;
+				macaddr_low  += ((u8) addr[3]) << 16;
+				macaddr_low  += ((u8) addr[4]) << 8;
+				macaddr_low  += (u8) addr[5];
+
 				nes_write_indexed(nesdev,
 						perfect_filter_register_address+(mc_index * 8),
 						macaddr_low);
@@ -896,16 +958,25 @@ static void nes_netdev_set_multicast_list(struct net_device *netdev)
 						perfect_filter_register_address+4+(mc_index * 8),
 						(u32)macaddr_high | NES_MAC_ADDR_VALID |
 						((((u32)(1<<mc_nic_index)) << 16)));
-				multicast_addr = multicast_addr->next;
+				nesadapter->pft_mcast_map[mc_index] =
+							nesvnic->nic_index;
 			} else {
 				nes_debug(NES_DBG_NIC_RX, "Clearing MC Address at register 0x%04X\n",
 						  perfect_filter_register_address+(mc_index * 8));
 				nes_write_indexed(nesdev,
 						perfect_filter_register_address+4+(mc_index * 8),
 						0);
+				nesadapter->pft_mcast_map[mc_index] = 255;
 			}
 		}
+		kfree(addrs);
+		/* PFT is not large enough */
+		if (i < mc_count)
+			set_allmulti(nesdev, nic_active_bit);
 	}
+
+unlock:
+	spin_unlock_irqrestore(&nesadapter->resource_lock, flags);
 }
 
 
@@ -918,6 +989,10 @@ static int nes_netdev_change_mtu(struct net_device *netdev, int new_mtu)
 	struct nes_device *nesdev = nesvnic->nesdev;
 	int ret = 0;
 	u8 jumbomode = 0;
+	u32 nic_active;
+	u32 nic_active_bit;
+	u32 uc_all_active;
+	u32 mc_all_active;
 
 	if ((new_mtu < ETH_ZLEN) || (new_mtu > max_mtu))
 		return -EINVAL;
@@ -931,13 +1006,30 @@ static int nes_netdev_change_mtu(struct net_device *netdev, int new_mtu)
 	nes_nic_init_timer_defaults(nesdev, jumbomode);
 
 	if (netif_running(netdev)) {
+		nic_active_bit = 1 << nesvnic->nic_index;
+		mc_all_active = nes_read_indexed(nesdev,
+				NES_IDX_NIC_MULTICAST_ALL) & nic_active_bit;
+		uc_all_active = nes_read_indexed(nesdev,
+				NES_IDX_NIC_UNICAST_ALL)  & nic_active_bit;
+
 		nes_netdev_stop(netdev);
 		nes_netdev_open(netdev);
+
+		nic_active = nes_read_indexed(nesdev,
+					NES_IDX_NIC_MULTICAST_ALL);
+		nic_active |= mc_all_active;
+		nes_write_indexed(nesdev, NES_IDX_NIC_MULTICAST_ALL,
+							nic_active);
+
+		nic_active = nes_read_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL);
+		nic_active |= uc_all_active;
+		nes_write_indexed(nesdev, NES_IDX_NIC_UNICAST_ALL, nic_active);
 	}
 
 	return ret;
 }
 
+
 static const char nes_ethtool_stringset[][ETH_GSTRING_LEN] = {
 	"Link Change Interrupts",
 	"Linearized SKBs",
@@ -946,18 +1038,21 @@ static const char nes_ethtool_stringset[][ETH_GSTRING_LEN] = {
 	"Pause Frames Received",
 	"Internal Routing Errors",
 	"SQ SW Dropped SKBs",
-	"SQ Locked",
 	"SQ Full",
 	"Segmented TSO Requests",
 	"Rx Symbol Errors",
 	"Rx Jabber Errors",
 	"Rx Oversized Frames",
 	"Rx Short Frames",
+	"Rx Length Errors",
+	"Rx CRC Errors",
+	"Rx Port Discard",
 	"Endnode Rx Discards",
 	"Endnode Rx Octets",
 	"Endnode Rx Frames",
 	"Endnode Tx Octets",
 	"Endnode Tx Frames",
+	"Tx Errors",
 	"mh detected",
 	"mh pauses",
 	"Retransmission Count",
@@ -986,56 +1081,27 @@ static const char nes_ethtool_stringset[][ETH_GSTRING_LEN] = {
 	"CM Nodes Destroyed",
 	"CM Accel Drops",
 	"CM Resets Received",
+	"Free 4Kpbls",
+	"Free 256pbls",
 	"Timer Inits",
-	"CQ Depth 1",
-	"CQ Depth 4",
-	"CQ Depth 16",
-	"CQ Depth 24",
-	"CQ Depth 32",
-	"CQ Depth 128",
-	"CQ Depth 256",
 	"LRO aggregated",
 	"LRO flushed",
 	"LRO no_desc",
+	"PAU CreateQPs",
+	"PAU DestroyQPs",
 };
-
 #define NES_ETHTOOL_STAT_COUNT  ARRAY_SIZE(nes_ethtool_stringset)
 
-/**
- * nes_netdev_get_rx_csum
- */
-static u32 nes_netdev_get_rx_csum (struct net_device *netdev)
-{
-	struct nes_vnic *nesvnic = netdev_priv(netdev);
-
-	if (nesvnic->rx_checksum_disabled)
-		return 0;
-	else
-		return 1;
-}
-
 
 /**
- * nes_netdev_set_rc_csum
+ * nes_netdev_get_sset_count
  */
-static int nes_netdev_set_rx_csum(struct net_device *netdev, u32 enable)
+static int nes_netdev_get_sset_count(struct net_device *netdev, int stringset)
 {
-	struct nes_vnic *nesvnic = netdev_priv(netdev);
-
-	if (enable)
-		nesvnic->rx_checksum_disabled = 0;
+	if (stringset == ETH_SS_STATS)
+		return NES_ETHTOOL_STAT_COUNT;
 	else
-		nesvnic->rx_checksum_disabled = 1;
-	return 0;
-}
-
-
-/**
- * nes_netdev_get_stats_count
- */
-static int nes_netdev_get_stats_count(struct net_device *netdev)
-{
-	return NES_ETHTOOL_STAT_COUNT;
+		return -EINVAL;
 }
 
 
@@ -1055,24 +1121,27 @@ static void nes_netdev_get_strings(struct net_device *netdev, u32 stringset,
 /**
  * nes_netdev_get_ethtool_stats
  */
+
 static void nes_netdev_get_ethtool_stats(struct net_device *netdev,
 		struct ethtool_stats *target_ethtool_stats, u64 *target_stat_values)
 {
 	u64 u64temp;
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
 	struct nes_device *nesdev = nesvnic->nesdev;
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 nic_count;
 	u32 u32temp;
+	u32 index = 0;
 
 	target_ethtool_stats->n_stats = NES_ETHTOOL_STAT_COUNT;
-	target_stat_values[0] = nesvnic->nesdev->link_status_interrupts;
-	target_stat_values[1] = nesvnic->linearized_skbs;
-	target_stat_values[2] = nesvnic->tso_requests;
+	target_stat_values[index] = nesvnic->nesdev->link_status_interrupts;
+	target_stat_values[++index] = nesvnic->linearized_skbs;
+	target_stat_values[++index] = nesvnic->tso_requests;
 
 	u32temp = nes_read_indexed(nesdev,
 			NES_IDX_MAC_TX_PAUSE_FRAMES + (nesvnic->nesdev->mac_index*0x200));
 	nesvnic->nesdev->mac_pause_frames_sent += u32temp;
-	target_stat_values[3] = nesvnic->nesdev->mac_pause_frames_sent;
+	target_stat_values[++index] = nesvnic->nesdev->mac_pause_frames_sent;
 
 	u32temp = nes_read_indexed(nesdev,
 			NES_IDX_MAC_RX_PAUSE_FRAMES + (nesvnic->nesdev->mac_index*0x200));
@@ -1088,6 +1157,46 @@ static void nes_netdev_get_ethtool_stats(struct net_device *netdev,
 	nesvnic->nesdev->port_tx_discards += u32temp;
 	nesvnic->netstats.tx_dropped += u32temp;
 
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_SHORT_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->netstats.rx_dropped += u32temp;
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+	nesvnic->nesdev->mac_rx_short_frames += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_OVERSIZED_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->netstats.rx_dropped += u32temp;
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+	nesvnic->nesdev->mac_rx_oversized_frames += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_JABBER_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->netstats.rx_dropped += u32temp;
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+	nesvnic->nesdev->mac_rx_jabber_frames += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_SYMBOL_ERR_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->netstats.rx_dropped += u32temp;
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+	nesvnic->nesdev->mac_rx_symbol_err_frames += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_LENGTH_ERR_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->netstats.rx_length_errors += u32temp;
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_RX_CRC_ERR_FRAMES + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->nesdev->mac_rx_errors += u32temp;
+	nesvnic->nesdev->mac_rx_crc_errors += u32temp;
+	nesvnic->netstats.rx_crc_errors += u32temp;
+
+	u32temp = nes_read_indexed(nesdev,
+			NES_IDX_MAC_TX_ERRORS + (nesvnic->nesdev->mac_index*0x200));
+	nesvnic->nesdev->mac_tx_errors += u32temp;
+	nesvnic->netstats.tx_errors += u32temp;
+
 	for (nic_count = 0; nic_count < NES_MAX_PORT_COUNT; nic_count++) {
 		if (nesvnic->qp_nic_index[nic_count] == 0xf)
 			break;
@@ -1143,64 +1252,62 @@ static void nes_netdev_get_ethtool_stats(struct net_device *netdev,
 		nesvnic->endnode_ipv4_tcp_retransmits += u32temp;
 	}
 
-	target_stat_values[4] = nesvnic->nesdev->mac_pause_frames_received;
-	target_stat_values[5] = nesdev->nesadapter->nic_rx_eth_route_err;
-	target_stat_values[6] = nesvnic->tx_sw_dropped;
-	target_stat_values[7] = nesvnic->sq_locked;
-	target_stat_values[8] = nesvnic->sq_full;
-	target_stat_values[9] = nesvnic->segmented_tso_requests;
-	target_stat_values[10] = nesvnic->nesdev->mac_rx_symbol_err_frames;
-	target_stat_values[11] = nesvnic->nesdev->mac_rx_jabber_frames;
-	target_stat_values[12] = nesvnic->nesdev->mac_rx_oversized_frames;
-	target_stat_values[13] = nesvnic->nesdev->mac_rx_short_frames;
-	target_stat_values[14] = nesvnic->endnode_nstat_rx_discard;
-	target_stat_values[15] = nesvnic->endnode_nstat_rx_octets;
-	target_stat_values[16] = nesvnic->endnode_nstat_rx_frames;
-	target_stat_values[17] = nesvnic->endnode_nstat_tx_octets;
-	target_stat_values[18] = nesvnic->endnode_nstat_tx_frames;
-	target_stat_values[19] = mh_detected;
-	target_stat_values[20] = mh_pauses_sent;
-	target_stat_values[21] = nesvnic->endnode_ipv4_tcp_retransmits;
-	target_stat_values[22] = atomic_read(&cm_connects);
-	target_stat_values[23] = atomic_read(&cm_accepts);
-	target_stat_values[24] = atomic_read(&cm_disconnects);
-	target_stat_values[25] = atomic_read(&cm_connecteds);
-	target_stat_values[26] = atomic_read(&cm_connect_reqs);
-	target_stat_values[27] = atomic_read(&cm_rejects);
-	target_stat_values[28] = atomic_read(&mod_qp_timouts);
-	target_stat_values[29] = atomic_read(&qps_created);
-	target_stat_values[30] = atomic_read(&sw_qps_destroyed);
-	target_stat_values[31] = atomic_read(&qps_destroyed);
-	target_stat_values[32] = atomic_read(&cm_closes);
-	target_stat_values[33] = cm_packets_sent;
-	target_stat_values[34] = cm_packets_bounced;
-	target_stat_values[35] = cm_packets_created;
-	target_stat_values[36] = cm_packets_received;
-	target_stat_values[37] = cm_packets_dropped;
-	target_stat_values[38] = cm_packets_retrans;
-	target_stat_values[39] = cm_listens_created;
-	target_stat_values[40] = cm_listens_destroyed;
-	target_stat_values[41] = cm_backlog_drops;
-	target_stat_values[42] = atomic_read(&cm_loopbacks);
-	target_stat_values[43] = atomic_read(&cm_nodes_created);
-	target_stat_values[44] = atomic_read(&cm_nodes_destroyed);
-	target_stat_values[45] = atomic_read(&cm_accel_dropped_pkts);
-	target_stat_values[46] = atomic_read(&cm_resets_recvd);
-	target_stat_values[47] = int_mod_timer_init;
-	target_stat_values[48] = int_mod_cq_depth_1;
-	target_stat_values[49] = int_mod_cq_depth_4;
-	target_stat_values[50] = int_mod_cq_depth_16;
-	target_stat_values[51] = int_mod_cq_depth_24;
-	target_stat_values[52] = int_mod_cq_depth_32;
-	target_stat_values[53] = int_mod_cq_depth_128;
-	target_stat_values[54] = int_mod_cq_depth_256;
-	target_stat_values[55] = nesvnic->lro_mgr.stats.aggregated;
-	target_stat_values[56] = nesvnic->lro_mgr.stats.flushed;
-	target_stat_values[57] = nesvnic->lro_mgr.stats.no_desc;
-
+	target_stat_values[++index] = nesvnic->nesdev->mac_pause_frames_received;
+	target_stat_values[++index] = nesdev->nesadapter->nic_rx_eth_route_err;
+	target_stat_values[++index] = nesvnic->tx_sw_dropped;
+	target_stat_values[++index] = nesvnic->sq_full;
+	target_stat_values[++index] = nesvnic->segmented_tso_requests;
+	target_stat_values[++index] = nesvnic->nesdev->mac_rx_symbol_err_frames;
+	target_stat_values[++index] = nesvnic->nesdev->mac_rx_jabber_frames;
+	target_stat_values[++index] = nesvnic->nesdev->mac_rx_oversized_frames;
+	target_stat_values[++index] = nesvnic->nesdev->mac_rx_short_frames;
+	target_stat_values[++index] = nesvnic->netstats.rx_length_errors;
+	target_stat_values[++index] = nesvnic->nesdev->mac_rx_crc_errors;
+	target_stat_values[++index] = nesvnic->nesdev->port_rx_discards;
+	target_stat_values[++index] = nesvnic->endnode_nstat_rx_discard;
+	target_stat_values[++index] = nesvnic->endnode_nstat_rx_octets;
+	target_stat_values[++index] = nesvnic->endnode_nstat_rx_frames;
+	target_stat_values[++index] = nesvnic->endnode_nstat_tx_octets;
+	target_stat_values[++index] = nesvnic->endnode_nstat_tx_frames;
+	target_stat_values[++index] = nesvnic->nesdev->mac_tx_errors;
+	target_stat_values[++index] = mh_detected;
+	target_stat_values[++index] = mh_pauses_sent;
+	target_stat_values[++index] = nesvnic->endnode_ipv4_tcp_retransmits;
+	target_stat_values[++index] = atomic_read(&cm_connects);
+	target_stat_values[++index] = atomic_read(&cm_accepts);
+	target_stat_values[++index] = atomic_read(&cm_disconnects);
+	target_stat_values[++index] = atomic_read(&cm_connecteds);
+	target_stat_values[++index] = atomic_read(&cm_connect_reqs);
+	target_stat_values[++index] = atomic_read(&cm_rejects);
+	target_stat_values[++index] = atomic_read(&mod_qp_timouts);
+	target_stat_values[++index] = atomic_read(&qps_created);
+	target_stat_values[++index] = atomic_read(&sw_qps_destroyed);
+	target_stat_values[++index] = atomic_read(&qps_destroyed);
+	target_stat_values[++index] = atomic_read(&cm_closes);
+	target_stat_values[++index] = cm_packets_sent;
+	target_stat_values[++index] = cm_packets_bounced;
+	target_stat_values[++index] = cm_packets_created;
+	target_stat_values[++index] = cm_packets_received;
+	target_stat_values[++index] = cm_packets_dropped;
+	target_stat_values[++index] = cm_packets_retrans;
+	target_stat_values[++index] = atomic_read(&cm_listens_created);
+	target_stat_values[++index] = atomic_read(&cm_listens_destroyed);
+	target_stat_values[++index] = cm_backlog_drops;
+	target_stat_values[++index] = atomic_read(&cm_loopbacks);
+	target_stat_values[++index] = atomic_read(&cm_nodes_created);
+	target_stat_values[++index] = atomic_read(&cm_nodes_destroyed);
+	target_stat_values[++index] = atomic_read(&cm_accel_dropped_pkts);
+	target_stat_values[++index] = atomic_read(&cm_resets_recvd);
+	target_stat_values[++index] = nesadapter->free_4kpbl;
+	target_stat_values[++index] = nesadapter->free_256pbl;
+	target_stat_values[++index] = int_mod_timer_init;
+	target_stat_values[++index] = nesvnic->lro_mgr.stats.aggregated;
+	target_stat_values[++index] = nesvnic->lro_mgr.stats.flushed;
+	target_stat_values[++index] = nesvnic->lro_mgr.stats.no_desc;
+	target_stat_values[++index] = atomic_read(&pau_qps_created);
+	target_stat_values[++index] = atomic_read(&pau_qps_destroyed);
 }
 
-
 /**
  * nes_netdev_get_drvinfo
  */
@@ -1208,12 +1315,15 @@ static void nes_netdev_get_drvinfo(struct net_device *netdev,
 		struct ethtool_drvinfo *drvinfo)
 {
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
-
-	strcpy(drvinfo->driver, DRV_NAME);
-	strcpy(drvinfo->bus_info, pci_name(nesvnic->nesdev->pcidev));
-	strcpy(drvinfo->fw_version, "TBD");
-	strcpy(drvinfo->version, DRV_VERSION);
-	drvinfo->n_stats = nes_netdev_get_stats_count(netdev);
+	struct nes_adapter *nesadapter = nesvnic->nesdev->nesadapter;
+
+	strlcpy(drvinfo->driver, DRV_NAME, sizeof(drvinfo->driver));
+	strlcpy(drvinfo->bus_info, pci_name(nesvnic->nesdev->pcidev),
+		sizeof(drvinfo->bus_info));
+	snprintf(drvinfo->fw_version, sizeof(drvinfo->fw_version),
+		 "%u.%u", nesadapter->firmware_version >> 16,
+		 nesadapter->firmware_version & 0x000000ff);
+	strlcpy(drvinfo->version, DRV_VERSION, sizeof(drvinfo->version));
 	drvinfo->testinfo_len = 0;
 	drvinfo->eedump_len = 0;
 	drvinfo->regdump_len = 0;
@@ -1337,14 +1447,14 @@ static int nes_netdev_set_pauseparam(struct net_device *netdev,
 				NES_IDX_MAC_TX_CONFIG + (nesdev->mac_index*0x200));
 		u32temp |= NES_IDX_MAC_TX_CONFIG_ENABLE_PAUSE;
 		nes_write_indexed(nesdev,
-				NES_IDX_MAC_TX_CONFIG_ENABLE_PAUSE + (nesdev->mac_index*0x200), u32temp);
+				NES_IDX_MAC_TX_CONFIG + (nesdev->mac_index*0x200), u32temp);
 		nesdev->disable_tx_flow_control = 0;
 	} else if ((et_pauseparam->tx_pause == 0) && (nesdev->disable_tx_flow_control == 0)) {
 		u32temp = nes_read_indexed(nesdev,
 				NES_IDX_MAC_TX_CONFIG + (nesdev->mac_index*0x200));
 		u32temp &= ~NES_IDX_MAC_TX_CONFIG_ENABLE_PAUSE;
 		nes_write_indexed(nesdev,
-				NES_IDX_MAC_TX_CONFIG_ENABLE_PAUSE + (nesdev->mac_index*0x200), u32temp);
+				NES_IDX_MAC_TX_CONFIG + (nesdev->mac_index*0x200), u32temp);
 		nesdev->disable_tx_flow_control = 1;
 	}
 	if ((et_pauseparam->rx_pause == 1) && (nesdev->disable_rx_flow_control == 1)) {
@@ -1375,49 +1485,58 @@ static int nes_netdev_get_settings(struct net_device *netdev, struct ethtool_cmd
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
+	u32 mac_index = nesdev->mac_index;
+	u8 phy_type = nesadapter->phy_type[mac_index];
+	u8 phy_index = nesadapter->phy_index[mac_index];
 	u16 phy_data;
 
 	et_cmd->duplex = DUPLEX_FULL;
 	et_cmd->port   = PORT_MII;
+	et_cmd->maxtxpkt = 511;
+	et_cmd->maxrxpkt = 511;
 
 	if (nesadapter->OneG_Mode) {
-		et_cmd->speed = SPEED_1000;
-		if (nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_PUMA_1G) {
+		ethtool_cmd_speed_set(et_cmd, SPEED_1000);
+		if (phy_type == NES_PHY_TYPE_PUMA_1G) {
 			et_cmd->supported   = SUPPORTED_1000baseT_Full;
 			et_cmd->advertising = ADVERTISED_1000baseT_Full;
 			et_cmd->autoneg     = AUTONEG_DISABLE;
 			et_cmd->transceiver = XCVR_INTERNAL;
-			et_cmd->phy_address = nesdev->mac_index;
+			et_cmd->phy_address = mac_index;
 		} else {
-			et_cmd->supported   = SUPPORTED_1000baseT_Full | SUPPORTED_Autoneg;
-			et_cmd->advertising = ADVERTISED_1000baseT_Full | ADVERTISED_Autoneg;
-			nes_read_1G_phy_reg(nesdev, 0, nesadapter->phy_index[nesdev->mac_index], &phy_data);
+			unsigned long flags;
+			et_cmd->supported   = SUPPORTED_1000baseT_Full
+					    | SUPPORTED_Autoneg;
+			et_cmd->advertising = ADVERTISED_1000baseT_Full
+					    | ADVERTISED_Autoneg;
+			spin_lock_irqsave(&nesadapter->phy_lock, flags);
+			nes_read_1G_phy_reg(nesdev, 0, phy_index, &phy_data);
+			spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 			if (phy_data & 0x1000)
 				et_cmd->autoneg = AUTONEG_ENABLE;
 			else
 				et_cmd->autoneg = AUTONEG_DISABLE;
 			et_cmd->transceiver = XCVR_EXTERNAL;
-			et_cmd->phy_address = nesadapter->phy_index[nesdev->mac_index];
+			et_cmd->phy_address = phy_index;
 		}
+		return 0;
+	}
+	if ((phy_type == NES_PHY_TYPE_ARGUS) ||
+	    (phy_type == NES_PHY_TYPE_SFP_D) ||
+	    (phy_type == NES_PHY_TYPE_KR)) {
+		et_cmd->transceiver = XCVR_EXTERNAL;
+		et_cmd->port        = PORT_FIBRE;
+		et_cmd->supported   = SUPPORTED_FIBRE;
+		et_cmd->advertising = ADVERTISED_FIBRE;
+		et_cmd->phy_address = phy_index;
 	} else {
-		if ((nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_IRIS) ||
-		    (nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_ARGUS)) {
-			et_cmd->transceiver = XCVR_EXTERNAL;
-			et_cmd->port        = PORT_FIBRE;
-			et_cmd->supported   = SUPPORTED_FIBRE;
-			et_cmd->advertising = ADVERTISED_FIBRE;
-			et_cmd->phy_address = nesadapter->phy_index[nesdev->mac_index];
-		} else {
-			et_cmd->transceiver = XCVR_INTERNAL;
-			et_cmd->supported   = SUPPORTED_10000baseT_Full;
-			et_cmd->advertising = ADVERTISED_10000baseT_Full;
-			et_cmd->phy_address = nesdev->mac_index;
-		}
-		et_cmd->speed = SPEED_10000;
-		et_cmd->autoneg = AUTONEG_DISABLE;
+		et_cmd->transceiver = XCVR_INTERNAL;
+		et_cmd->supported   = SUPPORTED_10000baseT_Full;
+		et_cmd->advertising = ADVERTISED_10000baseT_Full;
+		et_cmd->phy_address = mac_index;
 	}
-	et_cmd->maxtxpkt = 511;
-	et_cmd->maxrxpkt = 511;
+	ethtool_cmd_speed_set(et_cmd, SPEED_10000);
+	et_cmd->autoneg = AUTONEG_DISABLE;
 	return 0;
 }
 
@@ -1430,12 +1549,15 @@ static int nes_netdev_set_settings(struct net_device *netdev, struct ethtool_cmd
 	struct nes_vnic *nesvnic = netdev_priv(netdev);
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
-	u16 phy_data;
 
 	if ((nesadapter->OneG_Mode) &&
 	    (nesadapter->phy_type[nesdev->mac_index] != NES_PHY_TYPE_PUMA_1G)) {
-		nes_read_1G_phy_reg(nesdev, 0, nesadapter->phy_index[nesdev->mac_index],
-				&phy_data);
+		unsigned long flags;
+		u16 phy_data;
+		u8 phy_index = nesadapter->phy_index[nesdev->mac_index];
+
+		spin_lock_irqsave(&nesadapter->phy_lock, flags);
+		nes_read_1G_phy_reg(nesdev, 0, phy_index, &phy_data);
 		if (et_cmd->autoneg) {
 			/* Turn on Full duplex, Autoneg, and restart autonegotiation */
 			phy_data |= 0x1300;
@@ -1443,55 +1565,41 @@ static int nes_netdev_set_settings(struct net_device *netdev, struct ethtool_cmd
 			/* Turn off autoneg */
 			phy_data &= ~0x1000;
 		}
-		nes_write_1G_phy_reg(nesdev, 0, nesadapter->phy_index[nesdev->mac_index],
-				phy_data);
+		nes_write_1G_phy_reg(nesdev, 0, phy_index, phy_data);
+		spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 	}
 
 	return 0;
 }
 
 
-static struct ethtool_ops nes_ethtool_ops = {
+static const struct ethtool_ops nes_ethtool_ops = {
 	.get_link = ethtool_op_get_link,
 	.get_settings = nes_netdev_get_settings,
 	.set_settings = nes_netdev_set_settings,
-	.get_tx_csum = ethtool_op_get_tx_csum,
-	.get_rx_csum = nes_netdev_get_rx_csum,
-	.get_sg = ethtool_op_get_sg,
 	.get_strings = nes_netdev_get_strings,
-	.get_stats_count = nes_netdev_get_stats_count,
+	.get_sset_count = nes_netdev_get_sset_count,
 	.get_ethtool_stats = nes_netdev_get_ethtool_stats,
 	.get_drvinfo = nes_netdev_get_drvinfo,
 	.get_coalesce = nes_netdev_get_coalesce,
 	.set_coalesce = nes_netdev_set_coalesce,
 	.get_pauseparam = nes_netdev_get_pauseparam,
 	.set_pauseparam = nes_netdev_set_pauseparam,
-	.set_tx_csum = ethtool_op_set_tx_csum,
-	.set_rx_csum = nes_netdev_set_rx_csum,
-	.set_sg = ethtool_op_set_sg,
-	.get_tso = ethtool_op_get_tso,
-	.set_tso = ethtool_op_set_tso,
-	.get_flags = ethtool_op_get_flags,
-	.set_flags = ethtool_op_set_flags,
 };
 
-
-static void nes_netdev_vlan_rx_register(struct net_device *netdev, struct vlan_group *grp)
+static void nes_vlan_mode(struct net_device *netdev, struct nes_device *nesdev, netdev_features_t features)
 {
-	struct nes_vnic *nesvnic = netdev_priv(netdev);
-	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 u32temp;
 	unsigned long flags;
 
 	spin_lock_irqsave(&nesadapter->phy_lock, flags);
-	nesvnic->vlan_grp = grp;
 
 	nes_debug(NES_DBG_NETDEV, "%s: %s\n", __func__, netdev->name);
 
 	/* Enable/Disable VLAN Stripping */
 	u32temp = nes_read_indexed(nesdev, NES_IDX_PCIX_DIAG);
-	if (grp)
+	if (features & NETIF_F_HW_VLAN_CTAG_RX)
 		u32temp &= 0xfdffffff;
 	else
 		u32temp	|= 0x02000000;
@@ -1500,6 +1608,45 @@ static void nes_netdev_vlan_rx_register(struct net_device *netdev, struct vlan_g
 	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 }
 
+static netdev_features_t nes_fix_features(struct net_device *netdev, netdev_features_t features)
+{
+	/*
+	 * Since there is no support for separate rx/tx vlan accel
+	 * enable/disable make sure tx flag is always in same state as rx.
+	 */
+	if (features & NETIF_F_HW_VLAN_CTAG_RX)
+		features |= NETIF_F_HW_VLAN_CTAG_TX;
+	else
+		features &= ~NETIF_F_HW_VLAN_CTAG_TX;
+
+	return features;
+}
+
+static int nes_set_features(struct net_device *netdev, netdev_features_t features)
+{
+	struct nes_vnic *nesvnic = netdev_priv(netdev);
+	struct nes_device *nesdev = nesvnic->nesdev;
+	u32 changed = netdev->features ^ features;
+
+	if (changed & NETIF_F_HW_VLAN_CTAG_RX)
+		nes_vlan_mode(netdev, nesdev, features);
+
+	return 0;
+}
+
+static const struct net_device_ops nes_netdev_ops = {
+	.ndo_open		= nes_netdev_open,
+	.ndo_stop		= nes_netdev_stop,
+	.ndo_start_xmit		= nes_netdev_start_xmit,
+	.ndo_get_stats		= nes_netdev_get_stats,
+	.ndo_tx_timeout		= nes_netdev_tx_timeout,
+	.ndo_set_mac_address	= nes_netdev_set_mac_address,
+	.ndo_set_rx_mode	= nes_netdev_set_multicast_list,
+	.ndo_change_mtu		= nes_netdev_change_mtu,
+	.ndo_validate_addr	= eth_validate_addr,
+	.ndo_fix_features	= nes_fix_features,
+	.ndo_set_features	= nes_set_features,
+};
 
 /**
  * nes_netdev_init - initialize network device
@@ -1508,47 +1655,32 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 		void __iomem *mmio_addr)
 {
 	u64 u64temp;
-	struct nes_vnic *nesvnic = NULL;
+	struct nes_vnic *nesvnic;
 	struct net_device *netdev;
 	struct nic_qp_map *curr_qp_map;
-	u32 u32temp;
-	u16 phy_data;
-	u16 temp_phy_data;
+	u8 phy_type = nesdev->nesadapter->phy_type[nesdev->mac_index];
 
 	netdev = alloc_etherdev(sizeof(struct nes_vnic));
 	if (!netdev) {
 		printk(KERN_ERR PFX "nesvnic etherdev alloc failed");
 		return NULL;
 	}
+	nesvnic = netdev_priv(netdev);
 
 	nes_debug(NES_DBG_INIT, "netdev = %p, %s\n", netdev, netdev->name);
 
 	SET_NETDEV_DEV(netdev, &nesdev->pcidev->dev);
 
-	nesvnic = netdev_priv(netdev);
-	memset(nesvnic, 0, sizeof(*nesvnic));
-
-	netdev->open = nes_netdev_open;
-	netdev->stop = nes_netdev_stop;
-	netdev->hard_start_xmit = nes_netdev_start_xmit;
-	netdev->get_stats = nes_netdev_get_stats;
-	netdev->tx_timeout = nes_netdev_tx_timeout;
-	netdev->set_mac_address = nes_netdev_set_mac_address;
-	netdev->set_multicast_list = nes_netdev_set_multicast_list;
-	netdev->change_mtu = nes_netdev_change_mtu;
 	netdev->watchdog_timeo = NES_TX_TIMEOUT;
 	netdev->irq = nesdev->pcidev->irq;
 	netdev->mtu = ETH_DATA_LEN;
 	netdev->hard_header_len = ETH_HLEN;
 	netdev->addr_len = ETH_ALEN;
 	netdev->type = ARPHRD_ETHER;
-	netdev->features = NETIF_F_HIGHDMA;
+	netdev->netdev_ops = &nes_netdev_ops;
 	netdev->ethtool_ops = &nes_ethtool_ops;
 	netif_napi_add(netdev, &nesvnic->napi, nes_netdev_poll, 128);
 	nes_debug(NES_DBG_INIT, "Enabling VLAN Insert/Delete.\n");
-	netdev->features |= NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX;
-	netdev->vlan_rx_register = nes_netdev_vlan_rx_register;
-	netdev->features |= NETIF_F_LLTX;
 
 	/* Fill in the port structure */
 	nesvnic->netdev = netdev;
@@ -1573,21 +1705,22 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 	netdev->dev_addr[3] = (u8)(u64temp>>16);
 	netdev->dev_addr[4] = (u8)(u64temp>>8);
 	netdev->dev_addr[5] = (u8)u64temp;
-	memcpy(netdev->perm_addr, netdev->dev_addr, 6);
 
-	if ((nesvnic->logical_port < 2) || (nesdev->nesadapter->hw_rev != NE020_REV)) {
-		netdev->features |= NETIF_F_TSO | NETIF_F_SG | NETIF_F_IP_CSUM;
-		netdev->features |= NETIF_F_GSO | NETIF_F_TSO | NETIF_F_SG | NETIF_F_IP_CSUM;
-	} else {
-		netdev->features |= NETIF_F_SG | NETIF_F_IP_CSUM;
-	}
+	netdev->hw_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_RXCSUM | NETIF_F_HW_VLAN_CTAG_RX;
+	if ((nesvnic->logical_port < 2) || (nesdev->nesadapter->hw_rev != NE020_REV))
+		netdev->hw_features |= NETIF_F_TSO;
+
+	netdev->features = netdev->hw_features | NETIF_F_HIGHDMA | NETIF_F_HW_VLAN_CTAG_TX;
+	netdev->hw_features |= NETIF_F_LRO;
 
 	nes_debug(NES_DBG_INIT, "nesvnic = %p, reported features = 0x%lX, QPid = %d,"
 			" nic_index = %d, logical_port = %d, mac_index = %d.\n",
 			nesvnic, (unsigned long)netdev->features, nesvnic->nic.qp_id,
 			nesvnic->nic_index, nesvnic->logical_port,  nesdev->mac_index);
 
-	if (nesvnic->nesdev->nesadapter->port_count == 1) {
+	if (nesvnic->nesdev->nesadapter->port_count == 1 &&
+		nesvnic->nesdev->nesadapter->adapter_fcn_count == 1) {
+
 		nesvnic->qp_nic_index[0] = nesvnic->nic_index;
 		nesvnic->qp_nic_index[1] = nesvnic->nic_index + 1;
 		if (nes_drv_opt & NES_DRV_OPT_DUAL_LOGICAL_PORT) {
@@ -1598,11 +1731,14 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 			nesvnic->qp_nic_index[3] = nesvnic->nic_index + 3;
 		}
 	} else {
-		if (nesvnic->nesdev->nesadapter->port_count == 2) {
-			nesvnic->qp_nic_index[0] = nesvnic->nic_index;
-			nesvnic->qp_nic_index[1] = nesvnic->nic_index + 2;
-			nesvnic->qp_nic_index[2] = 0xf;
-			nesvnic->qp_nic_index[3] = 0xf;
+		if (nesvnic->nesdev->nesadapter->port_count == 2 ||
+			(nesvnic->nesdev->nesadapter->port_count == 1 &&
+			nesvnic->nesdev->nesadapter->adapter_fcn_count == 2)) {
+				nesvnic->qp_nic_index[0] = nesvnic->nic_index;
+				nesvnic->qp_nic_index[1] = nesvnic->nic_index
+									+ 2;
+				nesvnic->qp_nic_index[2] = 0xf;
+				nesvnic->qp_nic_index[3] = 0xf;
 		} else {
 			nesvnic->qp_nic_index[0] = nesvnic->nic_index;
 			nesvnic->qp_nic_index[1] = 0xf;
@@ -1618,7 +1754,10 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 		nesvnic->rdma_enabled = 0;
 	}
 	nesvnic->nic_cq.cq_number = nesvnic->nic.qp_id;
+	init_timer(&nesvnic->event_timer);
+	nesvnic->event_timer.function = NULL;
 	spin_lock_init(&nesvnic->tx_lock);
+	spin_lock_init(&nesvnic->port_ibevent_lock);
 	nesdev->netdev[nesdev->netdev_count] = netdev;
 
 	nes_debug(NES_DBG_INIT, "Adding nesvnic (%p) to the adapters nesvnic_list for MAC%d.\n",
@@ -1627,68 +1766,81 @@ struct net_device *nes_netdev_init(struct nes_device *nesdev,
 
 	if ((nesdev->netdev_count == 0) &&
 	    ((PCI_FUNC(nesdev->pcidev->devfn) == nesdev->mac_index) ||
-	     ((nesdev->nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_PUMA_1G) &&
+	     ((phy_type == NES_PHY_TYPE_PUMA_1G) &&
 	      (((PCI_FUNC(nesdev->pcidev->devfn) == 1) && (nesdev->mac_index == 2)) ||
 	       ((PCI_FUNC(nesdev->pcidev->devfn) == 2) && (nesdev->mac_index == 1)))))) {
-		/*
-		 * nes_debug(NES_DBG_INIT, "Setting up PHY interrupt mask. Using register index 0x%04X\n",
-		 *		NES_IDX_PHY_PCS_CONTROL_STATUS0 + (0x200 * (nesvnic->logical_port & 1)));
-		 */
+		u32 u32temp;
+		u32 link_mask = 0;
+		u32 link_val = 0;
+		u16 temp_phy_data;
+		u16 phy_data = 0;
+		unsigned long flags;
+
 		u32temp = nes_read_indexed(nesdev, NES_IDX_PHY_PCS_CONTROL_STATUS0 +
 				(0x200 * (nesdev->mac_index & 1)));
-		if (nesdev->nesadapter->phy_type[nesdev->mac_index] != NES_PHY_TYPE_PUMA_1G) {
+		if (phy_type != NES_PHY_TYPE_PUMA_1G) {
 			u32temp |= 0x00200000;
 			nes_write_indexed(nesdev, NES_IDX_PHY_PCS_CONTROL_STATUS0 +
 				(0x200 * (nesdev->mac_index & 1)), u32temp);
 		}
 
-		u32temp = nes_read_indexed(nesdev, NES_IDX_PHY_PCS_CONTROL_STATUS0 +
-				(0x200 * (nesdev->mac_index & 1)));
-
-		if ((u32temp&0x0f1f0000) == 0x0f0f0000) {
-			if (nesdev->nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_IRIS) {
-				nes_init_phy(nesdev);
-				nes_read_10G_phy_reg(nesdev, nesdev->nesadapter->phy_index[nesdev->mac_index], 1, 1);
-				temp_phy_data = (u16)nes_read_indexed(nesdev,
-									NES_IDX_MAC_MDIO_CONTROL);
-				u32temp = 20;
-				do {
-					nes_read_10G_phy_reg(nesdev, nesdev->nesadapter->phy_index[nesdev->mac_index], 1, 1);
-					phy_data = (u16)nes_read_indexed(nesdev,
-									NES_IDX_MAC_MDIO_CONTROL);
-					if ((phy_data == temp_phy_data) || (!(--u32temp)))
-						break;
-					temp_phy_data = phy_data;
-				} while (1);
-				if (phy_data & 4) {
-					nes_debug(NES_DBG_INIT, "The Link is UP!!.\n");
-					nesvnic->linkup = 1;
-				} else {
-					nes_debug(NES_DBG_INIT, "The Link is DOWN!!.\n");
-				}
+		/* Check and set linkup here.  This is for back to back */
+		/* configuration where second port won't get link interrupt */
+		switch (phy_type) {
+		case NES_PHY_TYPE_PUMA_1G:
+			if (nesdev->mac_index < 2) {
+				link_mask = 0x01010000;
+				link_val = 0x01010000;
 			} else {
-				nes_debug(NES_DBG_INIT, "The Link is UP!!.\n");
-				nesvnic->linkup = 1;
+				link_mask = 0x02020000;
+				link_val = 0x02020000;
 			}
-		} else if (nesdev->nesadapter->phy_type[nesdev->mac_index] == NES_PHY_TYPE_PUMA_1G) {
-			nes_debug(NES_DBG_INIT, "mac_index=%d, logical_port=%d, u32temp=0x%04X, PCI_FUNC=%d\n",
-				nesdev->mac_index, nesvnic->logical_port, u32temp, PCI_FUNC(nesdev->pcidev->devfn));
-			if (((nesdev->mac_index < 2) && ((u32temp&0x01010000) == 0x01010000)) ||
-			    ((nesdev->mac_index > 1) && ((u32temp&0x02020000) == 0x02020000)))  {
-				nes_debug(NES_DBG_INIT, "The Link is UP!!.\n");
+			break;
+		case NES_PHY_TYPE_SFP_D:
+			spin_lock_irqsave(&nesdev->nesadapter->phy_lock, flags);
+			nes_read_10G_phy_reg(nesdev,
+					     nesdev->nesadapter->phy_index[nesdev->mac_index],
+					     1, 0x9003);
+			temp_phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+			nes_read_10G_phy_reg(nesdev,
+					     nesdev->nesadapter->phy_index[nesdev->mac_index],
+					     3, 0x0021);
+			nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+			nes_read_10G_phy_reg(nesdev,
+					     nesdev->nesadapter->phy_index[nesdev->mac_index],
+					     3, 0x0021);
+			phy_data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
+			spin_unlock_irqrestore(&nesdev->nesadapter->phy_lock, flags);
+			phy_data = (!temp_phy_data && (phy_data == 0x8000)) ? 0x4 : 0x0;
+			break;
+		default:
+			link_mask = 0x0f1f0000;
+			link_val = 0x0f0f0000;
+			break;
+		}
+
+		u32temp = nes_read_indexed(nesdev,
+					   NES_IDX_PHY_PCS_CONTROL_STATUS0 +
+					   (0x200 * (nesdev->mac_index & 1)));
+
+		if (phy_type == NES_PHY_TYPE_SFP_D) {
+			if (phy_data & 0x0004)
+				nesvnic->linkup = 1;
+		} else {
+			if ((u32temp & link_mask) == link_val)
 				nesvnic->linkup = 1;
-			}
 		}
+
 		/* clear the MAC interrupt status, assumes direct logical to physical mapping */
 		u32temp = nes_read_indexed(nesdev, NES_IDX_MAC_INT_STATUS + (0x200 * nesdev->mac_index));
 		nes_debug(NES_DBG_INIT, "Phy interrupt status = 0x%X.\n", u32temp);
 		nes_write_indexed(nesdev, NES_IDX_MAC_INT_STATUS + (0x200 * nesdev->mac_index), u32temp);
 
-		if (nesdev->nesadapter->phy_type[nesdev->mac_index] != NES_PHY_TYPE_IRIS)
-			nes_init_phy(nesdev);
-
+		nes_init_phy(nesdev);
 	}
 
+	nes_vlan_mode(netdev, nesdev, netdev->features);
+
 	return netdev;
 }
 
diff --git a/drivers/infiniband/hw/nes/nes_user.h b/drivers/infiniband/hw/nes/nes_user.h
index e64306bce80..529c421bb15 100644
--- a/drivers/infiniband/hw/nes/nes_user.h
+++ b/drivers/infiniband/hw/nes/nes_user.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect.  All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  * Copyright (c) 2005 Topspin Communications.  All rights reserved.
  * Copyright (c) 2005 Cisco Systems.  All rights reserved.
  * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
@@ -39,8 +39,8 @@
 
 #include <linux/types.h>
 
-#define NES_ABI_USERSPACE_VER 1
-#define NES_ABI_KERNEL_VER    1
+#define NES_ABI_USERSPACE_VER 2
+#define NES_ABI_KERNEL_VER    2
 
 /*
  * Make sure that all structs defined in this file remain laid out so
@@ -78,6 +78,7 @@ struct nes_create_cq_req {
 
 struct nes_create_qp_req {
 	__u64 user_wqe_buffers;
+	__u64 user_qp_buffer;
 };
 
 enum iwnes_memreg_type {
@@ -86,6 +87,7 @@ enum iwnes_memreg_type {
 	IWNES_MEMREG_TYPE_CQ = 0x0002,
 	IWNES_MEMREG_TYPE_MW = 0x0003,
 	IWNES_MEMREG_TYPE_FMR = 0x0004,
+	IWNES_MEMREG_TYPE_FMEM = 0x0005,
 };
 
 struct nes_mem_reg_req {
diff --git a/drivers/infiniband/hw/nes/nes_utils.c b/drivers/infiniband/hw/nes/nes_utils.c
index fb8cbd71a2e..2042c0f2975 100644
--- a/drivers/infiniband/hw/nes/nes_utils.c
+++ b/drivers/infiniband/hw/nes/nes_utils.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -38,6 +38,7 @@
 #include <linux/ethtool.h>
 #include <linux/mii.h>
 #include <linux/if_vlan.h>
+#include <linux/slab.h>
 #include <linux/crc32.h>
 #include <linux/in.h>
 #include <linux/ip.h>
@@ -50,13 +51,34 @@
 
 #include "nes.h"
 
-
-
 static u16 nes_read16_eeprom(void __iomem *addr, u16 offset);
 
 u32 mh_detected;
 u32 mh_pauses_sent;
 
+static u32 nes_set_pau(struct nes_device *nesdev)
+{
+	u32 ret = 0;
+	u32 counter;
+
+	nes_write_indexed(nesdev, NES_IDX_GPR2, NES_ENABLE_PAU);
+	nes_write_indexed(nesdev, NES_IDX_GPR_TRIGGER, 1);
+
+	for (counter = 0; counter < NES_PAU_COUNTER; counter++) {
+		udelay(30);
+		if (!nes_read_indexed(nesdev, NES_IDX_GPR2)) {
+			printk(KERN_INFO PFX "PAU is supported.\n");
+			break;
+		}
+		nes_write_indexed(nesdev, NES_IDX_GPR_TRIGGER, 1);
+	}
+	if (counter == NES_PAU_COUNTER) {
+		printk(KERN_INFO PFX "PAU is not supported.\n");
+		return -EPERM;
+	}
+	return ret;
+}
+
 /**
  * nes_read_eeprom_values -
  */
@@ -183,9 +205,22 @@ int nes_read_eeprom_values(struct nes_device *nesdev, struct nes_adapter *nesada
 		} else if (((major_ver == 2) && (minor_ver > 21)) || ((major_ver > 2) && (major_ver != 255))) {
 			nesadapter->virtwq = 1;
 		}
+		if (((major_ver == 3) && (minor_ver >= 16)) || (major_ver > 3))
+			nesadapter->send_term_ok = 1;
+
+		if (nes_drv_opt & NES_DRV_OPT_ENABLE_PAU) {
+			if (!nes_set_pau(nesdev))
+				nesadapter->allow_unaligned_fpdus = 1;
+		}
+
 		nesadapter->firmware_version = (((u32)(u8)(eeprom_data>>8))  <<  16) +
 				(u32)((u8)eeprom_data);
 
+		eeprom_data = nes_read16_eeprom(nesdev->regs, next_section_address + 10);
+		printk(PFX "EEPROM version %u.%u\n", (u8)(eeprom_data>>8), (u8)eeprom_data);
+		nesadapter->eeprom_version = (((u32)(u8)(eeprom_data>>8)) << 16) +
+				(u32)((u8)eeprom_data);
+
 no_fw_rev:
 		/* eeprom is valid */
 		eeprom_offset = nesadapter->software_eeprom_offset;
@@ -377,12 +412,8 @@ static u16 nes_read16_eeprom(void __iomem *addr, u16 offset)
  */
 void nes_write_1G_phy_reg(struct nes_device *nesdev, u8 phy_reg, u8 phy_addr, u16 data)
 {
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 u32temp;
 	u32 counter;
-	unsigned long flags;
-
-	spin_lock_irqsave(&nesadapter->phy_lock, flags);
 
 	nes_write_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL,
 			0x50020000 | data | ((u32)phy_reg << 18) | ((u32)phy_addr << 23));
@@ -398,8 +429,6 @@ void nes_write_1G_phy_reg(struct nes_device *nesdev, u8 phy_reg, u8 phy_addr, u1
 	if (!(u32temp & 1))
 		nes_debug(NES_DBG_PHY, "Phy is not responding. interrupt status = 0x%X.\n",
 				u32temp);
-
-	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 }
 
 
@@ -410,14 +439,11 @@ void nes_write_1G_phy_reg(struct nes_device *nesdev, u8 phy_reg, u8 phy_addr, u1
  */
 void nes_read_1G_phy_reg(struct nes_device *nesdev, u8 phy_reg, u8 phy_addr, u16 *data)
 {
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 u32temp;
 	u32 counter;
-	unsigned long flags;
 
 	/* nes_debug(NES_DBG_PHY, "phy addr = %d, mac_index = %d\n",
 			phy_addr, nesdev->mac_index); */
-	spin_lock_irqsave(&nesadapter->phy_lock, flags);
 
 	nes_write_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL,
 			0x60020000 | ((u32)phy_reg << 18) | ((u32)phy_addr << 23));
@@ -437,7 +463,6 @@ void nes_read_1G_phy_reg(struct nes_device *nesdev, u8 phy_reg, u8 phy_addr, u16
 	} else {
 		*data = (u16)nes_read_indexed(nesdev, NES_IDX_MAC_MDIO_CONTROL);
 	}
-	spin_unlock_irqrestore(&nesadapter->phy_lock, flags);
 }
 
 
@@ -540,12 +565,15 @@ struct nes_cqp_request *nes_get_cqp_request(struct nes_device *nesdev)
 
 	if (!list_empty(&nesdev->cqp_avail_reqs)) {
 		spin_lock_irqsave(&nesdev->cqp.lock, flags);
-		cqp_request = list_entry(nesdev->cqp_avail_reqs.next,
+		if (!list_empty(&nesdev->cqp_avail_reqs)) {
+			cqp_request = list_entry(nesdev->cqp_avail_reqs.next,
 				struct nes_cqp_request, list);
-		list_del_init(&cqp_request->list);
+			list_del_init(&cqp_request->list);
+		}
 		spin_unlock_irqrestore(&nesdev->cqp.lock, flags);
-	} else {
-		cqp_request = kzalloc(sizeof(struct nes_cqp_request), GFP_KERNEL);
+	}
+	if (cqp_request == NULL) {
+		cqp_request = kzalloc(sizeof(struct nes_cqp_request), GFP_ATOMIC);
 		if (cqp_request) {
 			cqp_request->dynamic = 1;
 			INIT_LIST_HEAD(&cqp_request->list);
@@ -592,6 +620,7 @@ void nes_put_cqp_request(struct nes_device *nesdev,
 		nes_free_cqp_request(nesdev, cqp_request);
 }
 
+
 /**
  * nes_post_cqp_request
  */
@@ -602,6 +631,8 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 	unsigned long flags;
 	u32 cqp_head;
 	u64 u64temp;
+	u32 opcode;
+	int ctx_index = NES_CQP_WQE_COMP_CTX_LOW_IDX;
 
 	spin_lock_irqsave(&nesdev->cqp.lock, flags);
 
@@ -612,17 +643,20 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 		nesdev->cqp.sq_head &= nesdev->cqp.sq_size-1;
 		cqp_wqe = &nesdev->cqp.sq_vbase[cqp_head];
 		memcpy(cqp_wqe, &cqp_request->cqp_wqe, sizeof(*cqp_wqe));
+		opcode = le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX]);
+		if ((opcode & NES_CQP_OPCODE_MASK) == NES_CQP_DOWNLOAD_SEGMENT)
+			ctx_index = NES_CQP_WQE_DL_COMP_CTX_LOW_IDX;
 		barrier();
 		u64temp = (unsigned long)cqp_request;
-		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_COMP_SCRATCH_LOW_IDX,
-				    u64temp);
+		set_wqe_64bit_value(cqp_wqe->wqe_words, ctx_index, u64temp);
 		nes_debug(NES_DBG_CQP, "CQP request (opcode 0x%02X), line 1 = 0x%08X put on CQPs SQ,"
-				" request = %p, cqp_head = %u, cqp_tail = %u, cqp_size = %u,"
-				" waiting = %d, refcount = %d.\n",
-				le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX])&0x3f,
-				le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX]), cqp_request,
-				nesdev->cqp.sq_head, nesdev->cqp.sq_tail, nesdev->cqp.sq_size,
-				cqp_request->waiting, atomic_read(&cqp_request->refcount));
+			" request = %p, cqp_head = %u, cqp_tail = %u, cqp_size = %u,"
+			" waiting = %d, refcount = %d.\n",
+			opcode & NES_CQP_OPCODE_MASK,
+			le32_to_cpu(cqp_wqe->wqe_words[NES_CQP_WQE_ID_IDX]), cqp_request,
+			nesdev->cqp.sq_head, nesdev->cqp.sq_tail, nesdev->cqp.sq_size,
+			cqp_request->waiting, atomic_read(&cqp_request->refcount));
+
 		barrier();
 
 		/* Ring doorbell (1 WQEs) */
@@ -643,7 +677,6 @@ void nes_post_cqp_request(struct nes_device *nesdev,
 	return;
 }
 
-
 /**
  * nes_arp_table
  */
@@ -652,6 +685,7 @@ int nes_arp_table(struct nes_device *nesdev, u32 ip_addr, u8 *mac_addr, u32 acti
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	int arp_index;
 	int err = 0;
+	__be32 tmp_addr;
 
 	for (arp_index = 0; (u32) arp_index < nesadapter->arp_table_size; arp_index++) {
 		if (nesadapter->arp_table[arp_index].ip_addr == ip_addr)
@@ -665,7 +699,7 @@ int nes_arp_table(struct nes_device *nesdev, u32 ip_addr, u8 *mac_addr, u32 acti
 
 		arp_index = 0;
 		err = nes_alloc_resource(nesadapter, nesadapter->allocated_arps,
-				nesadapter->arp_table_size, (u32 *)&arp_index, &nesadapter->next_arp_index);
+				nesadapter->arp_table_size, (u32 *)&arp_index, &nesadapter->next_arp_index, NES_RESOURCE_ARP);
 		if (err) {
 			nes_debug(NES_DBG_NETDEV, "nes_alloc_resource returned error = %u\n", err);
 			return err;
@@ -679,9 +713,9 @@ int nes_arp_table(struct nes_device *nesdev, u32 ip_addr, u8 *mac_addr, u32 acti
 
 	/* DELETE or RESOLVE */
 	if (arp_index == nesadapter->arp_table_size) {
-		nes_debug(NES_DBG_NETDEV, "MAC for " NIPQUAD_FMT " not in ARP table - cannot %s\n",
-			  HIPQUAD(ip_addr),
-			  action == NES_ARP_RESOLVE ? "resolve" : "delete");
+		tmp_addr = cpu_to_be32(ip_addr);
+		nes_debug(NES_DBG_NETDEV, "MAC for %pI4 not in ARP table - cannot %s\n",
+			  &tmp_addr, action == NES_ARP_RESOLVE ? "resolve" : "delete");
 		return -1;
 	}
 
diff --git a/drivers/infiniband/hw/nes/nes_verbs.c b/drivers/infiniband/hw/nes/nes_verbs.c
index e3939d13484..218dd357428 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.c
+++ b/drivers/infiniband/hw/nes/nes_verbs.c
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -35,6 +35,7 @@
 #include <linux/moduleparam.h>
 #include <linux/random.h>
 #include <linux/highmem.h>
+#include <linux/slab.h>
 #include <asm/byteorder.h>
 
 #include <rdma/ib_verbs.h>
@@ -54,7 +55,8 @@ static void nes_unregister_ofa_device(struct nes_ib_device *nesibdev);
 /**
  * nes_alloc_mw
  */
-static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) {
+static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd, enum ib_mw_type type)
+{
 	struct nes_pd *nespd = to_nespd(ibpd);
 	struct nes_vnic *nesvnic = to_nesvnic(ibpd->device);
 	struct nes_device *nesdev = nesvnic->nesdev;
@@ -70,6 +72,9 @@ static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) {
 	u32 driver_key = 0;
 	u8 stag_key = 0;
 
+	if (type != IB_MW_TYPE_1)
+		return ERR_PTR(-EINVAL);
+
 	get_random_bytes(&next_stag_index, sizeof(next_stag_index));
 	stag_key = (u8)next_stag_index;
 
@@ -79,7 +84,7 @@ static struct ib_mw *nes_alloc_mw(struct ib_pd *ibpd) {
 	next_stag_index %= nesadapter->max_mr;
 
 	ret = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs,
-			nesadapter->max_mr, &stag_index, &next_stag_index);
+			nesadapter->max_mr, &stag_index, &next_stag_index, NES_RESOURCE_MW);
 	if (ret) {
 		return ERR_PTR(ret);
 	}
@@ -220,15 +225,15 @@ static int nes_bind_mw(struct ib_qp *ibqp, struct ib_mw *ibmw,
 	if (nesqp->ibqp_state > IB_QPS_RTS)
 		return -EINVAL;
 
-		spin_lock_irqsave(&nesqp->lock, flags);
+	spin_lock_irqsave(&nesqp->lock, flags);
 
 	head = nesqp->hwqp.sq_head;
 	qsize = nesqp->hwqp.sq_tail;
 
 	/* Check for SQ overflow */
 	if (((head + (2 * qsize) - nesqp->hwqp.sq_tail) % qsize) == (qsize - 1)) {
-			spin_unlock_irqrestore(&nesqp->lock, flags);
-		return -EINVAL;
+		spin_unlock_irqrestore(&nesqp->lock, flags);
+		return -ENOMEM;
 	}
 
 	wqe = &nesqp->hwqp.sq_vbase[head];
@@ -243,20 +248,19 @@ static int nes_bind_mw(struct ib_qp *ibqp, struct ib_mw *ibmw,
 	if (ibmw_bind->send_flags & IB_SEND_SIGNALED)
 		wqe_misc |= NES_IWARP_SQ_WQE_SIGNALED_COMPL;
 
-	if (ibmw_bind->mw_access_flags & IB_ACCESS_REMOTE_WRITE) {
+	if (ibmw_bind->bind_info.mw_access_flags & IB_ACCESS_REMOTE_WRITE)
 		wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_WRITE;
-	}
-	if (ibmw_bind->mw_access_flags & IB_ACCESS_REMOTE_READ) {
+	if (ibmw_bind->bind_info.mw_access_flags & IB_ACCESS_REMOTE_READ)
 		wqe_misc |= NES_CQP_STAG_RIGHTS_REMOTE_READ;
-	}
 
 	set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_MISC_IDX, wqe_misc);
-	set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MR_IDX, ibmw_bind->mr->lkey);
+	set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MR_IDX,
+			    ibmw_bind->bind_info.mr->lkey);
 	set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_MW_IDX, ibmw->rkey);
 	set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_LENGTH_LOW_IDX,
-			ibmw_bind->length);
+			ibmw_bind->bind_info.length);
 	wqe->wqe_words[NES_IWARP_SQ_BIND_WQE_LENGTH_HIGH_IDX] = 0;
-	u64temp = (u64)ibmw_bind->addr;
+	u64temp = (u64)ibmw_bind->bind_info.addr;
 	set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_BIND_WQE_VA_FBO_LOW_IDX, u64temp);
 
 	head++;
@@ -269,323 +273,242 @@ static int nes_bind_mw(struct ib_qp *ibqp, struct ib_mw *ibmw,
 	nes_write32(nesdev->regs+NES_WQE_ALLOC,
 			(1 << 24) | 0x00800000 | nesqp->hwqp.qp_id);
 
-		spin_unlock_irqrestore(&nesqp->lock, flags);
+	spin_unlock_irqrestore(&nesqp->lock, flags);
 
 	return 0;
 }
 
 
-/**
- * nes_alloc_fmr
+/*
+ * nes_alloc_fast_mr
  */
-static struct ib_fmr *nes_alloc_fmr(struct ib_pd *ibpd,
-		int ibmr_access_flags,
-		struct ib_fmr_attr *ibfmr_attr)
+static int alloc_fast_reg_mr(struct nes_device *nesdev, struct nes_pd *nespd,
+			     u32 stag, u32 page_count)
 {
-	unsigned long flags;
-	struct nes_pd *nespd = to_nespd(ibpd);
-	struct nes_vnic *nesvnic = to_nesvnic(ibpd->device);
-	struct nes_device *nesdev = nesvnic->nesdev;
-	struct nes_adapter *nesadapter = nesdev->nesadapter;
-	struct nes_fmr *nesfmr;
-	struct nes_cqp_request *cqp_request;
 	struct nes_hw_cqp_wqe *cqp_wqe;
+	struct nes_cqp_request *cqp_request;
+	unsigned long flags;
 	int ret;
-	u32 stag;
-	u32 stag_index = 0;
-	u32 next_stag_index = 0;
-	u32 driver_key = 0;
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	u32 opcode = 0;
-	u8 stag_key = 0;
-	int i=0;
-	struct nes_vpbl vpbl;
-
-	get_random_bytes(&next_stag_index, sizeof(next_stag_index));
-	stag_key = (u8)next_stag_index;
-
-	driver_key = 0;
-
-	next_stag_index >>= 8;
-	next_stag_index %= nesadapter->max_mr;
-
-	ret = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs,
-			nesadapter->max_mr, &stag_index, &next_stag_index);
-	if (ret) {
-		goto failed_resource_alloc;
-	}
-
-	nesfmr = kzalloc(sizeof(*nesfmr), GFP_KERNEL);
-	if (!nesfmr) {
-		ret = -ENOMEM;
-		goto failed_fmr_alloc;
-	}
-
-	nesfmr->nesmr.mode = IWNES_MEMREG_TYPE_FMR;
-	if (ibfmr_attr->max_pages == 1) {
-		/* use zero length PBL */
-		nesfmr->nesmr.pbl_4k = 0;
-		nesfmr->nesmr.pbls_used = 0;
-	} else if (ibfmr_attr->max_pages <= 32) {
-		/* use PBL 256 */
-		nesfmr->nesmr.pbl_4k = 0;
-		nesfmr->nesmr.pbls_used = 1;
-	} else if (ibfmr_attr->max_pages <= 512) {
-		/* use 4K PBLs */
-		nesfmr->nesmr.pbl_4k = 1;
-		nesfmr->nesmr.pbls_used = 1;
-	} else {
-		/* use two level 4K PBLs */
-		/* add support for two level 256B PBLs */
-		nesfmr->nesmr.pbl_4k = 1;
-		nesfmr->nesmr.pbls_used = 1 + (ibfmr_attr->max_pages >> 9) +
-				((ibfmr_attr->max_pages & 511) ? 1 : 0);
-	}
-	/* Register the region with the adapter */
-	spin_lock_irqsave(&nesadapter->pbl_lock, flags);
-
-	/* track PBL resources */
-	if (nesfmr->nesmr.pbls_used != 0) {
-		if (nesfmr->nesmr.pbl_4k) {
-			if (nesfmr->nesmr.pbls_used > nesadapter->free_4kpbl) {
-				spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				ret = -ENOMEM;
-				goto failed_vpbl_alloc;
-			} else {
-				nesadapter->free_4kpbl -= nesfmr->nesmr.pbls_used;
-			}
-		} else {
-			if (nesfmr->nesmr.pbls_used > nesadapter->free_256pbl) {
-				spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				ret = -ENOMEM;
-				goto failed_vpbl_alloc;
-			} else {
-				nesadapter->free_256pbl -= nesfmr->nesmr.pbls_used;
-			}
-		}
-	}
-
-	/* one level pbl */
-	if (nesfmr->nesmr.pbls_used == 0) {
-		nesfmr->root_vpbl.pbl_vbase = NULL;
-		nes_debug(NES_DBG_MR,  "zero level pbl \n");
-	} else if (nesfmr->nesmr.pbls_used == 1) {
-		/* can change it to kmalloc & dma_map_single */
-		nesfmr->root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
-				&nesfmr->root_vpbl.pbl_pbase);
-		if (!nesfmr->root_vpbl.pbl_vbase) {
-			spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-			ret = -ENOMEM;
-			goto failed_vpbl_alloc;
-		}
-		nesfmr->leaf_pbl_cnt = 0;
-		nes_debug(NES_DBG_MR, "one level pbl, root_vpbl.pbl_vbase=%p \n",
-				nesfmr->root_vpbl.pbl_vbase);
-	}
-	/* two level pbl */
-	else {
-		nesfmr->root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 8192,
-				&nesfmr->root_vpbl.pbl_pbase);
-		if (!nesfmr->root_vpbl.pbl_vbase) {
-			spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-			ret = -ENOMEM;
-			goto failed_vpbl_alloc;
-		}
-
-		nesfmr->root_vpbl.leaf_vpbl = kzalloc(sizeof(*nesfmr->root_vpbl.leaf_vpbl)*1024, GFP_KERNEL);
-		if (!nesfmr->root_vpbl.leaf_vpbl) {
-			spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-			ret = -ENOMEM;
-			goto failed_leaf_vpbl_alloc;
-		}
-
-		nesfmr->leaf_pbl_cnt = nesfmr->nesmr.pbls_used-1;
-		nes_debug(NES_DBG_MR, "two level pbl, root_vpbl.pbl_vbase=%p"
-				" leaf_pbl_cnt=%d root_vpbl.leaf_vpbl=%p\n",
-				nesfmr->root_vpbl.pbl_vbase, nesfmr->leaf_pbl_cnt, nesfmr->root_vpbl.leaf_vpbl);
-
-		for (i=0; i<nesfmr->leaf_pbl_cnt; i++)
-			nesfmr->root_vpbl.leaf_vpbl[i].pbl_vbase = NULL;
-
-		for (i=0; i<nesfmr->leaf_pbl_cnt; i++) {
-			vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
-					&vpbl.pbl_pbase);
-
-			if (!vpbl.pbl_vbase) {
-				ret = -ENOMEM;
-				spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				goto failed_leaf_vpbl_pages_alloc;
-			}
-
-			nesfmr->root_vpbl.pbl_vbase[i].pa_low = cpu_to_le32((u32)vpbl.pbl_pbase);
-			nesfmr->root_vpbl.pbl_vbase[i].pa_high = cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32)));
-			nesfmr->root_vpbl.leaf_vpbl[i] = vpbl;
-
-			nes_debug(NES_DBG_MR, "pbase_low=0x%x, pbase_high=0x%x, vpbl=%p\n",
-					nesfmr->root_vpbl.pbl_vbase[i].pa_low,
-					nesfmr->root_vpbl.pbl_vbase[i].pa_high,
-					&nesfmr->root_vpbl.leaf_vpbl[i]);
-		}
-	}
-	nesfmr->ib_qp = NULL;
-	nesfmr->access_rights =0;
+	u16 major_code;
+	u64 region_length = page_count * PAGE_SIZE;
 
-	stag = stag_index << 8;
-	stag |= driver_key;
-	stag += (u32)stag_key;
 
-	spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
 	cqp_request = nes_get_cqp_request(nesdev);
 	if (cqp_request == NULL) {
 		nes_debug(NES_DBG_MR, "Failed to get a cqp_request.\n");
-		ret = -ENOMEM;
-		goto failed_leaf_vpbl_pages_alloc;
+		return -ENOMEM;
 	}
+	nes_debug(NES_DBG_MR, "alloc_fast_reg_mr: page_count = %d, "
+			      "region_length = %llu\n",
+			      page_count, region_length);
 	cqp_request->waiting = 1;
 	cqp_wqe = &cqp_request->cqp_wqe;
 
-	nes_debug(NES_DBG_MR, "Registering STag 0x%08X, index = 0x%08X\n",
-			stag, stag_index);
-
-	opcode = NES_CQP_ALLOCATE_STAG | NES_CQP_STAG_VA_TO | NES_CQP_STAG_MR;
-
-	if (nesfmr->nesmr.pbl_4k == 1)
-		opcode |= NES_CQP_STAG_PBL_BLK_SIZE;
-
-	if (ibmr_access_flags & IB_ACCESS_REMOTE_WRITE) {
-		opcode |= NES_CQP_STAG_RIGHTS_REMOTE_WRITE |
-				NES_CQP_STAG_RIGHTS_LOCAL_WRITE | NES_CQP_STAG_REM_ACC_EN;
-		nesfmr->access_rights |=
-				NES_CQP_STAG_RIGHTS_REMOTE_WRITE | NES_CQP_STAG_RIGHTS_LOCAL_WRITE |
-				NES_CQP_STAG_REM_ACC_EN;
+	spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+	if (nesadapter->free_4kpbl > 0) {
+		nesadapter->free_4kpbl--;
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+	} else {
+		/* No 4kpbl's available: */
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+		nes_debug(NES_DBG_MR, "Out of Pbls\n");
+		nes_free_cqp_request(nesdev, cqp_request);
+		return -ENOMEM;
 	}
 
-	if (ibmr_access_flags & IB_ACCESS_REMOTE_READ) {
-		opcode |= NES_CQP_STAG_RIGHTS_REMOTE_READ |
-				NES_CQP_STAG_RIGHTS_LOCAL_READ | NES_CQP_STAG_REM_ACC_EN;
-		nesfmr->access_rights |=
-				NES_CQP_STAG_RIGHTS_REMOTE_READ | NES_CQP_STAG_RIGHTS_LOCAL_READ |
-				NES_CQP_STAG_REM_ACC_EN;
-	}
+	opcode = NES_CQP_ALLOCATE_STAG | NES_CQP_STAG_MR |
+		 NES_CQP_STAG_PBL_BLK_SIZE | NES_CQP_STAG_VA_TO |
+		 NES_CQP_STAG_REM_ACC_EN;
+	/*
+	 * The current OFED API does not support the zero based TO option.
+	 * If added then need to changed the NES_CQP_STAG_VA* option.  Also,
+	 * the API does not support that ability to have the MR set for local
+	 * access only when created and not allow the SQ op to override. Given
+	 * this the remote enable must be set here.
+	 */
 
 	nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
 	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX, opcode);
-	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_LEN_HIGH_PD_IDX, (nespd->pd_id & 0x00007fff));
-	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_STAG_IDX, stag);
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PBL_BLK_COUNT_IDX, 1);
 
-	cqp_wqe->wqe_words[NES_CQP_STAG_WQE_PBL_BLK_COUNT_IDX] =
-			cpu_to_le32((nesfmr->nesmr.pbls_used>1) ?
-			(nesfmr->nesmr.pbls_used-1) : nesfmr->nesmr.pbls_used);
+	cqp_wqe->wqe_words[NES_CQP_STAG_WQE_LEN_HIGH_PD_IDX] =
+			cpu_to_le32((u32)(region_length >> 8) & 0xff000000);
+	cqp_wqe->wqe_words[NES_CQP_STAG_WQE_LEN_HIGH_PD_IDX] |=
+			cpu_to_le32(nespd->pd_id & 0x00007fff);
+
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_STAG_IDX, stag);
+	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_VA_LOW_IDX, 0);
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_LEN_LOW_IDX, 0);
+	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PA_LOW_IDX, 0);
+	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PBL_LEN_IDX, (page_count * 8));
+	cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] |= cpu_to_le32(NES_CQP_STAG_PBL_BLK_SIZE);
+	barrier();
 
 	atomic_set(&cqp_request->refcount, 2);
 	nes_post_cqp_request(nesdev, cqp_request);
 
 	/* Wait for CQP */
-	ret = wait_event_timeout(cqp_request->waitq, (cqp_request->request_done != 0),
-			NES_EVENT_TIMEOUT);
-	nes_debug(NES_DBG_MR, "Register STag 0x%08X completed, wait_event_timeout ret = %u,"
-			" CQP Major:Minor codes = 0x%04X:0x%04X.\n",
-			stag, ret, cqp_request->major_code, cqp_request->minor_code);
-
-	if ((!ret) || (cqp_request->major_code)) {
-		nes_put_cqp_request(nesdev, cqp_request);
-		ret = (!ret) ? -ETIME : -EIO;
-		goto failed_leaf_vpbl_pages_alloc;
-	}
+	ret = wait_event_timeout(cqp_request->waitq,
+				 (0 != cqp_request->request_done),
+				 NES_EVENT_TIMEOUT);
+
+	nes_debug(NES_DBG_MR, "Allocate STag 0x%08X completed, "
+		  "wait_event_timeout ret = %u, CQP Major:Minor codes = "
+		  "0x%04X:0x%04X.\n", stag, ret, cqp_request->major_code,
+		  cqp_request->minor_code);
+	major_code = cqp_request->major_code;
 	nes_put_cqp_request(nesdev, cqp_request);
-	nesfmr->nesmr.ibfmr.lkey = stag;
-	nesfmr->nesmr.ibfmr.rkey = stag;
-	nesfmr->attr = *ibfmr_attr;
-
-	return &nesfmr->nesmr.ibfmr;
-
-	failed_leaf_vpbl_pages_alloc:
-	/* unroll all allocated pages */
-	for (i=0; i<nesfmr->leaf_pbl_cnt; i++) {
-		if (nesfmr->root_vpbl.leaf_vpbl[i].pbl_vbase) {
-			pci_free_consistent(nesdev->pcidev, 4096, nesfmr->root_vpbl.leaf_vpbl[i].pbl_vbase,
-					nesfmr->root_vpbl.leaf_vpbl[i].pbl_pbase);
-		}
-	}
-	if (nesfmr->root_vpbl.leaf_vpbl)
-		kfree(nesfmr->root_vpbl.leaf_vpbl);
-
-	failed_leaf_vpbl_alloc:
-	if (nesfmr->leaf_pbl_cnt == 0) {
-		if (nesfmr->root_vpbl.pbl_vbase)
-			pci_free_consistent(nesdev->pcidev, 4096, nesfmr->root_vpbl.pbl_vbase,
-					nesfmr->root_vpbl.pbl_pbase);
-	} else
-		pci_free_consistent(nesdev->pcidev, 8192, nesfmr->root_vpbl.pbl_vbase,
-				nesfmr->root_vpbl.pbl_pbase);
-
-	failed_vpbl_alloc:
-	kfree(nesfmr);
 
-	failed_fmr_alloc:
-	nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+	if (!ret || major_code) {
+		spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+		nesadapter->free_4kpbl++;
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+	}
 
-	failed_resource_alloc:
-	return ERR_PTR(ret);
+	if (!ret)
+		return -ETIME;
+	else if (major_code)
+		return -EIO;
+	return 0;
 }
 
-
-/**
- * nes_dealloc_fmr
+/*
+ * nes_alloc_fast_reg_mr
  */
-static int nes_dealloc_fmr(struct ib_fmr *ibfmr)
+static struct ib_mr *nes_alloc_fast_reg_mr(struct ib_pd *ibpd, int max_page_list_len)
 {
-	struct nes_mr *nesmr = to_nesmr_from_ibfmr(ibfmr);
-	struct nes_fmr *nesfmr = to_nesfmr(nesmr);
-	struct nes_vnic *nesvnic = to_nesvnic(ibfmr->device);
+	struct nes_pd *nespd = to_nespd(ibpd);
+	struct nes_vnic *nesvnic = to_nesvnic(ibpd->device);
 	struct nes_device *nesdev = nesvnic->nesdev;
-	struct nes_mr temp_nesmr = *nesmr;
-	int i = 0;
+	struct nes_adapter *nesadapter = nesdev->nesadapter;
 
-	temp_nesmr.ibmw.device = ibfmr->device;
-	temp_nesmr.ibmw.pd = ibfmr->pd;
-	temp_nesmr.ibmw.rkey = ibfmr->rkey;
-	temp_nesmr.ibmw.uobject = NULL;
+	u32 next_stag_index;
+	u8 stag_key = 0;
+	u32 driver_key = 0;
+	int err = 0;
+	u32 stag_index = 0;
+	struct nes_mr *nesmr;
+	u32 stag;
+	int ret;
+	struct ib_mr *ibmr;
+/*
+ * Note:  Set to always use a fixed length single page entry PBL.  This is to allow
+ *	 for the fast_reg_mr operation to always know the size of the PBL.
+ */
+	if (max_page_list_len > (NES_4K_PBL_CHUNK_SIZE / sizeof(u64)))
+		return ERR_PTR(-E2BIG);
 
-	/* free the resources */
-	if (nesfmr->leaf_pbl_cnt == 0) {
-		/* single PBL case */
-		if (nesfmr->root_vpbl.pbl_vbase)
-			pci_free_consistent(nesdev->pcidev, 4096, nesfmr->root_vpbl.pbl_vbase,
-					nesfmr->root_vpbl.pbl_pbase);
-	} else {
-		for (i = 0; i < nesfmr->leaf_pbl_cnt; i++) {
-			pci_free_consistent(nesdev->pcidev, 4096, nesfmr->root_vpbl.leaf_vpbl[i].pbl_vbase,
-					nesfmr->root_vpbl.leaf_vpbl[i].pbl_pbase);
-		}
-		kfree(nesfmr->root_vpbl.leaf_vpbl);
-		pci_free_consistent(nesdev->pcidev, 8192, nesfmr->root_vpbl.pbl_vbase,
-				nesfmr->root_vpbl.pbl_pbase);
+	get_random_bytes(&next_stag_index, sizeof(next_stag_index));
+	stag_key = (u8)next_stag_index;
+	next_stag_index >>= 8;
+	next_stag_index %= nesadapter->max_mr;
+
+	err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs,
+				 nesadapter->max_mr, &stag_index,
+				 &next_stag_index, NES_RESOURCE_FAST_MR);
+	if (err)
+		return ERR_PTR(err);
+
+	nesmr = kzalloc(sizeof(*nesmr), GFP_KERNEL);
+	if (!nesmr) {
+		nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+		return ERR_PTR(-ENOMEM);
 	}
 
-	return nes_dealloc_mw(&temp_nesmr.ibmw);
-}
+	stag = stag_index << 8;
+	stag |= driver_key;
+	stag += (u32)stag_key;
 
+	nes_debug(NES_DBG_MR, "Allocating STag 0x%08X index = 0x%08X\n",
+		  stag, stag_index);
 
-/**
- * nes_map_phys_fmr
+	ret = alloc_fast_reg_mr(nesdev, nespd, stag, max_page_list_len);
+
+	if (ret == 0) {
+		nesmr->ibmr.rkey = stag;
+		nesmr->ibmr.lkey = stag;
+		nesmr->mode = IWNES_MEMREG_TYPE_FMEM;
+		ibmr = &nesmr->ibmr;
+	} else {
+		kfree(nesmr);
+		nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+		ibmr = ERR_PTR(-ENOMEM);
+	}
+	return ibmr;
+}
+
+/*
+ * nes_alloc_fast_reg_page_list
  */
-static int nes_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
-		int list_len, u64 iova)
+static struct ib_fast_reg_page_list *nes_alloc_fast_reg_page_list(
+							struct ib_device *ibdev,
+							int page_list_len)
 {
-	return 0;
-}
+	struct nes_vnic *nesvnic = to_nesvnic(ibdev);
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct ib_fast_reg_page_list *pifrpl;
+	struct nes_ib_fast_reg_page_list *pnesfrpl;
 
+	if (page_list_len > (NES_4K_PBL_CHUNK_SIZE / sizeof(u64)))
+		return ERR_PTR(-E2BIG);
+	/*
+	 * Allocate the ib_fast_reg_page_list structure, the
+	 * nes_fast_bpl structure, and the PLB table.
+	 */
+	pnesfrpl = kmalloc(sizeof(struct nes_ib_fast_reg_page_list) +
+			   page_list_len * sizeof(u64), GFP_KERNEL);
+
+	if (!pnesfrpl)
+		return ERR_PTR(-ENOMEM);
 
-/**
- * nes_unmap_frm
+	pifrpl = &pnesfrpl->ibfrpl;
+	pifrpl->page_list = &pnesfrpl->pbl;
+	pifrpl->max_page_list_len = page_list_len;
+	/*
+	 * Allocate the WQE PBL
+	 */
+	pnesfrpl->nes_wqe_pbl.kva = pci_alloc_consistent(nesdev->pcidev,
+							 page_list_len * sizeof(u64),
+							 &pnesfrpl->nes_wqe_pbl.paddr);
+
+	if (!pnesfrpl->nes_wqe_pbl.kva) {
+		kfree(pnesfrpl);
+		return ERR_PTR(-ENOMEM);
+	}
+	nes_debug(NES_DBG_MR, "nes_alloc_fast_reg_pbl: nes_frpl = %p, "
+		  "ibfrpl = %p, ibfrpl.page_list = %p, pbl.kva = %p, "
+		  "pbl.paddr = %llx\n", pnesfrpl, &pnesfrpl->ibfrpl,
+		  pnesfrpl->ibfrpl.page_list, pnesfrpl->nes_wqe_pbl.kva,
+		  (unsigned long long) pnesfrpl->nes_wqe_pbl.paddr);
+
+	return pifrpl;
+}
+
+/*
+ * nes_free_fast_reg_page_list
  */
-static int nes_unmap_fmr(struct list_head *ibfmr_list)
+static void nes_free_fast_reg_page_list(struct ib_fast_reg_page_list *pifrpl)
 {
-	return 0;
+	struct nes_vnic *nesvnic = to_nesvnic(pifrpl->device);
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct nes_ib_fast_reg_page_list *pnesfrpl;
+
+	pnesfrpl = container_of(pifrpl, struct nes_ib_fast_reg_page_list, ibfrpl);
+	/*
+	 * Free the WQE PBL.
+	 */
+	pci_free_consistent(nesdev->pcidev,
+			    pifrpl->max_page_list_len * sizeof(u64),
+			    pnesfrpl->nes_wqe_pbl.kva,
+			    pnesfrpl->nes_wqe_pbl.paddr);
+	/*
+	 * Free the PBL structure
+	 */
+	kfree(pnesfrpl);
 }
 
-
-
 /**
  * nes_query_device
  */
@@ -598,7 +521,7 @@ static int nes_query_device(struct ib_device *ibdev, struct ib_device_attr *prop
 	memset(props, 0, sizeof(*props));
 	memcpy(&props->sys_image_guid, nesvnic->netdev->dev_addr, 6);
 
-	props->fw_ver = nesdev->nesadapter->fw_ver;
+	props->fw_ver = nesdev->nesadapter->firmware_version;
 	props->device_cap_flags = nesdev->nesadapter->device_cap_flags;
 	props->vendor_id = nesdev->nesadapter->vendor_id;
 	props->vendor_part_id = nesdev->nesadapter->vendor_part_id;
@@ -608,28 +531,28 @@ static int nes_query_device(struct ib_device *ibdev, struct ib_device_attr *prop
 	props->max_qp_wr = nesdev->nesadapter->max_qp_wr - 2;
 	props->max_sge = nesdev->nesadapter->max_sge;
 	props->max_cq = nesibdev->max_cq;
-	props->max_cqe = nesdev->nesadapter->max_cqe - 1;
+	props->max_cqe = nesdev->nesadapter->max_cqe;
 	props->max_mr = nesibdev->max_mr;
 	props->max_mw = nesibdev->max_mr;
 	props->max_pd = nesibdev->max_pd;
 	props->max_sge_rd = 1;
 	switch (nesdev->nesadapter->max_irrq_wr) {
 		case 0:
-			props->max_qp_rd_atom = 1;
+			props->max_qp_rd_atom = 2;
 			break;
 		case 1:
-			props->max_qp_rd_atom = 4;
+			props->max_qp_rd_atom = 8;
 			break;
 		case 2:
-			props->max_qp_rd_atom = 16;
+			props->max_qp_rd_atom = 32;
 			break;
 		case 3:
-			props->max_qp_rd_atom = 32;
+			props->max_qp_rd_atom = 64;
 			break;
 		default:
 			props->max_qp_rd_atom = 0;
 	}
-	props->max_qp_init_rd_atom = props->max_qp_wr;
+	props->max_qp_init_rd_atom = props->max_qp_rd_atom;
 	props->atomic_cap = IB_ATOMIC_NONE;
 	props->max_map_per_fmr = 1;
 
@@ -642,15 +565,34 @@ static int nes_query_device(struct ib_device *ibdev, struct ib_device_attr *prop
  */
 static int nes_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr *props)
 {
+	struct nes_vnic *nesvnic = to_nesvnic(ibdev);
+	struct net_device *netdev = nesvnic->netdev;
+
 	memset(props, 0, sizeof(*props));
 
-	props->max_mtu = IB_MTU_2048;
-	props->active_mtu = IB_MTU_2048;
+	props->max_mtu = IB_MTU_4096;
+
+	if (netdev->mtu  >= 4096)
+		props->active_mtu = IB_MTU_4096;
+	else if (netdev->mtu  >= 2048)
+		props->active_mtu = IB_MTU_2048;
+	else if (netdev->mtu  >= 1024)
+		props->active_mtu = IB_MTU_1024;
+	else if (netdev->mtu  >= 512)
+		props->active_mtu = IB_MTU_512;
+	else
+		props->active_mtu = IB_MTU_256;
+
 	props->lid = 1;
 	props->lmc = 0;
 	props->sm_lid = 0;
 	props->sm_sl = 0;
-	props->state = IB_PORT_ACTIVE;
+	if (netif_queue_stopped(netdev))
+		props->state = IB_PORT_DOWN;
+	else if (nesvnic->linkup)
+		props->state = IB_PORT_ACTIVE;
+	else
+		props->state = IB_PORT_DOWN;
 	props->phys_state = 0;
 	props->port_cap_flags = IB_PORT_CM_SUP | IB_PORT_REINIT_SUP |
 			IB_PORT_VENDOR_CLASS_SUP | IB_PORT_BOOT_MGMT_SUP;
@@ -658,7 +600,7 @@ static int nes_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr
 	props->pkey_tbl_len = 1;
 	props->qkey_viol_cntr = 0;
 	props->active_width = IB_WIDTH_4X;
-	props->active_speed = 1;
+	props->active_speed = IB_SPEED_SDR;
 	props->max_msg_sz = 0x80000000;
 
 	return 0;
@@ -666,16 +608,6 @@ static int nes_query_port(struct ib_device *ibdev, u8 port, struct ib_port_attr
 
 
 /**
- * nes_modify_port
- */
-static int nes_modify_port(struct ib_device *ibdev, u8 port,
-		int port_modify_mask, struct ib_port_modify *props)
-{
-	return 0;
-}
-
-
-/**
  * nes_query_pkey
  */
 static int nes_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
@@ -848,10 +780,10 @@ static struct ib_pd *nes_alloc_pd(struct ib_device *ibdev,
 
 	nes_debug(NES_DBG_PD, "nesvnic=%p, netdev=%p %s, ibdev=%p, context=%p, netdev refcnt=%u\n",
 			nesvnic, nesdev->netdev[0], nesdev->netdev[0]->name, ibdev, context,
-			atomic_read(&nesvnic->netdev->refcnt));
+			netdev_refcnt_read(nesvnic->netdev));
 
 	err = nes_alloc_resource(nesadapter, nesadapter->allocated_pds,
-			nesadapter->max_pd, &pd_num, &nesadapter->next_pd);
+			nesadapter->max_pd, &pd_num, &nesadapter->next_pd, NES_RESOURCE_PD);
 	if (err) {
 		return ERR_PTR(err);
 	}
@@ -1079,6 +1011,7 @@ static int nes_setup_virt_qp(struct nes_qp *nesqp, struct nes_pbl *nespbl,
 		kunmap(nesqp->page);
 		return -ENOMEM;
 	}
+	nesqp->sq_kmapped = 1;
 	nesqp->hwqp.q2_vbase = mem;
 	mem += 256;
 	memset(nesqp->hwqp.q2_vbase, 0, 256);
@@ -1156,7 +1089,10 @@ static inline void nes_free_qp_mem(struct nes_device *nesdev,
 		pci_free_consistent(nesdev->pcidev, nesqp->qp_mem_size, nesqp->hwqp.q2_vbase, nesqp->hwqp.q2_pbase);
 		pci_free_consistent(nesdev->pcidev, 256, nesqp->pbl_vbase, nesqp->pbl_pbase );
 		nesqp->pbl_vbase = NULL;
-		kunmap(nesqp->page);
+		if (nesqp->sq_kmapped) {
+			nesqp->sq_kmapped = 0;
+			kunmap(nesqp->page);
+		}
 	}
 }
 
@@ -1224,7 +1160,7 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 			nes_debug(NES_DBG_QP, "RQ size=%u, SQ Size=%u\n", rq_size, sq_size);
 
 			ret = nes_alloc_resource(nesadapter, nesadapter->allocated_qps,
-					nesadapter->max_qp, &qp_num, &nesadapter->next_qp);
+					nesadapter->max_qp, &qp_num, &nesadapter->next_qp, NES_RESOURCE_QP);
 			if (ret) {
 				return ERR_PTR(ret);
 			}
@@ -1250,11 +1186,13 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 					nes_free_resource(nesadapter, nesadapter->allocated_qps, qp_num);
 					kfree(nesqp->allocated_buffer);
 					nes_debug(NES_DBG_QP, "ib_copy_from_udata() Failed \n");
-					return NULL;
+					return ERR_PTR(-EFAULT);
 				}
 				if (req.user_wqe_buffers) {
 					virt_wqs = 1;
 				}
+				if (req.user_qp_buffer)
+					nesqp->nesuqp_addr = req.user_qp_buffer;
 				if ((ibpd->uobject) && (ibpd->uobject->context)) {
 					nesqp->user_mode = 1;
 					nes_ucontext = to_nesucontext(ibpd->uobject->context);
@@ -1338,8 +1276,10 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 					NES_QPCONTEXT_MISC_RQ_SIZE_SHIFT);
 			nesqp->nesqp_context->misc |= cpu_to_le32((u32)nesqp->hwqp.sq_encoded_size <<
 					NES_QPCONTEXT_MISC_SQ_SIZE_SHIFT);
+			if (!udata) {
 				nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_PRIV_EN);
 				nesqp->nesqp_context->misc |= cpu_to_le32(NES_QPCONTEXT_MISC_FAST_REGISTER_EN);
+			}
 			nesqp->nesqp_context->cqs = cpu_to_le32(nesqp->nesscq->hw_cq.cq_number +
 					((u32)nesqp->nesrcq->hw_cq.cq_number << 16));
 			u64temp = (u64)nesqp->hwqp.sq_pbase;
@@ -1381,6 +1321,7 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 			nesqp->nesqp_context->aeq_token_low =  cpu_to_le32((u32)((unsigned long)(nesqp)));
 			nesqp->nesqp_context->aeq_token_high =  cpu_to_le32((u32)(upper_32_bits((unsigned long)(nesqp))));
 			nesqp->nesqp_context->ird_ord_sizes = cpu_to_le32(NES_QPCONTEXT_ORDIRD_ALSMM |
+					NES_QPCONTEXT_ORDIRD_AAH |
 					((((u32)nesadapter->max_irrq_wr) <<
 					NES_QPCONTEXT_ORDIRD_IRDSIZE_SHIFT) & NES_QPCONTEXT_ORDIRD_IRDSIZE_MASK));
 			if (disable_mpa_crc) {
@@ -1445,6 +1386,7 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 
 			if (ibpd->uobject) {
 				uresp.mmap_sq_db_index = nesqp->mmap_sq_db_index;
+				uresp.mmap_rq_db_index = 0;
 				uresp.actual_sq_size = sq_size;
 				uresp.actual_rq_size = rq_size;
 				uresp.qp_id = nesqp->hwqp.qp_id;
@@ -1460,24 +1402,59 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 			nes_debug(NES_DBG_QP, "QP%u structure located @%p.Size = %u.\n",
 					nesqp->hwqp.qp_id, nesqp, (u32)sizeof(*nesqp));
 			spin_lock_init(&nesqp->lock);
-			init_waitqueue_head(&nesqp->state_waitq);
-			init_waitqueue_head(&nesqp->kick_waitq);
 			nes_add_ref(&nesqp->ibqp);
 			break;
 		default:
 			nes_debug(NES_DBG_QP, "Invalid QP type: %d\n", init_attr->qp_type);
 			return ERR_PTR(-EINVAL);
-			break;
 	}
 
+	nesqp->sig_all = (init_attr->sq_sig_type == IB_SIGNAL_ALL_WR);
+	init_timer(&nesqp->terminate_timer);
+	nesqp->terminate_timer.function = nes_terminate_timeout;
+	nesqp->terminate_timer.data = (unsigned long)nesqp;
+
 	/* update the QP table */
 	nesdev->nesadapter->qp_table[nesqp->hwqp.qp_id-NES_FIRST_QPN] = nesqp;
 	nes_debug(NES_DBG_QP, "netdev refcnt=%u\n",
-			atomic_read(&nesvnic->netdev->refcnt));
+			netdev_refcnt_read(nesvnic->netdev));
 
 	return &nesqp->ibqp;
 }
 
+/**
+ * nes_clean_cq
+ */
+static void nes_clean_cq(struct nes_qp *nesqp, struct nes_cq *nescq)
+{
+	u32 cq_head;
+	u32 lo;
+	u32 hi;
+	u64 u64temp;
+	unsigned long flags = 0;
+
+	spin_lock_irqsave(&nescq->lock, flags);
+
+	cq_head = nescq->hw_cq.cq_head;
+	while (le32_to_cpu(nescq->hw_cq.cq_vbase[cq_head].cqe_words[NES_CQE_OPCODE_IDX]) & NES_CQE_VALID) {
+		rmb();
+		lo = le32_to_cpu(nescq->hw_cq.cq_vbase[cq_head].cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX]);
+		hi = le32_to_cpu(nescq->hw_cq.cq_vbase[cq_head].cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX]);
+		u64temp = (((u64)hi) << 32) | ((u64)lo);
+		u64temp &= ~(NES_SW_CONTEXT_ALIGN-1);
+		if (u64temp == (u64)(unsigned long)nesqp) {
+			/* Zero the context value so cqe will be ignored */
+			nescq->hw_cq.cq_vbase[cq_head].cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX] = 0;
+			nescq->hw_cq.cq_vbase[cq_head].cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX] = 0;
+		}
+
+		if (++cq_head >= nescq->hw_cq.cq_size)
+			cq_head = 0;
+	}
+
+	spin_unlock_irqrestore(&nescq->lock, flags);
+}
+
 
 /**
  * nes_destroy_qp
@@ -1485,12 +1462,11 @@ static struct ib_qp *nes_create_qp(struct ib_pd *ibpd,
 static int nes_destroy_qp(struct ib_qp *ibqp)
 {
 	struct nes_qp *nesqp = to_nesqp(ibqp);
-	/* struct nes_vnic *nesvnic = to_nesvnic(ibqp->device); */
 	struct nes_ucontext *nes_ucontext;
 	struct ib_qp_attr attr;
 	struct iw_cm_id *cm_id;
 	struct iw_cm_event cm_event;
-	int ret;
+	int ret = 0;
 
 	atomic_inc(&sw_qps_destroyed);
 	nesqp->destroyed = 1;
@@ -1506,7 +1482,7 @@ static int nes_destroy_qp(struct ib_qp *ibqp)
 			(nesqp->ibqp_state == IB_QPS_RTR)) && (nesqp->cm_id)) {
 		cm_id = nesqp->cm_id;
 		cm_event.event = IW_CM_EVENT_CONNECT_REPLY;
-		cm_event.status = IW_CM_EVENT_STATUS_TIMEOUT;
+		cm_event.status = -ETIMEDOUT;
 		cm_event.local_addr = cm_id->local_addr;
 		cm_event.remote_addr = cm_id->remote_addr;
 		cm_event.private_data = NULL;
@@ -1522,7 +1498,6 @@ static int nes_destroy_qp(struct ib_qp *ibqp)
 			nes_debug(NES_DBG_QP, "OFA CM event_handler returned, ret=%d\n", ret);
 	}
 
-
 	if (nesqp->user_mode) {
 		if ((ibqp->uobject)&&(ibqp->uobject->context)) {
 			nes_ucontext = to_nesucontext(ibqp->uobject->context);
@@ -1532,10 +1507,18 @@ static int nes_destroy_qp(struct ib_qp *ibqp)
 				nes_ucontext->first_free_wq = nesqp->mmap_sq_db_index;
 			}
 		}
-		if (nesqp->pbl_pbase)
+		if (nesqp->pbl_pbase && nesqp->sq_kmapped) {
+			nesqp->sq_kmapped = 0;
 			kunmap(nesqp->page);
-	}
+		}
+	} else {
+		/* Clean any pending completions from the cq(s) */
+		if (nesqp->nesscq)
+			nes_clean_cq(nesqp, nesqp->nesscq);
 
+		if ((nesqp->nesrcq) && (nesqp->nesrcq != nesqp->nesscq))
+			nes_clean_cq(nesqp, nesqp->nesrcq);
+	}
 	nes_rem_ref(&nesqp->ibqp);
 	return 0;
 }
@@ -1567,8 +1550,11 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 	unsigned long flags;
 	int ret;
 
+	if (entries > nesadapter->max_cqe)
+		return ERR_PTR(-EINVAL);
+
 	err = nes_alloc_resource(nesadapter, nesadapter->allocated_cqs,
-			nesadapter->max_cq, &cq_num, &nesadapter->next_cq);
+			nesadapter->max_cq, &cq_num, &nesadapter->next_cq, NES_RESOURCE_CQ);
 	if (err) {
 		return ERR_PTR(err);
 	}
@@ -1596,11 +1582,12 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 		nes_ucontext->mcrqf = req.mcrqf;
 		if (nes_ucontext->mcrqf) {
 			if (nes_ucontext->mcrqf & 0x80000000)
-				nescq->hw_cq.cq_number = nesvnic->nic.qp_id + 12 + (nes_ucontext->mcrqf & 0xf) - 1;
+				nescq->hw_cq.cq_number = nesvnic->nic.qp_id + 28 + 2 * ((nes_ucontext->mcrqf & 0xf) - 1);
 			else if (nes_ucontext->mcrqf & 0x40000000)
 				nescq->hw_cq.cq_number = nes_ucontext->mcrqf & 0xffff;
 			else
 				nescq->hw_cq.cq_number = nesvnic->mcrq_qp_id + nes_ucontext->mcrqf-1;
+			nescq->mcrqf = nes_ucontext->mcrqf;
 			nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 		}
 		nes_debug(NES_DBG_CQ, "CQ Virtual Address = %08lX, size = %u.\n",
@@ -1656,6 +1643,12 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 		if (!context)
 			pci_free_consistent(nesdev->pcidev, nescq->cq_mem_size, mem,
 					nescq->hw_cq.cq_pbase);
+		else {
+			pci_free_consistent(nesdev->pcidev, nespbl->pbl_size,
+					    nespbl->pbl_vbase, nespbl->pbl_pbase);
+			kfree(nespbl);
+		}
+
 		nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 		kfree(nescq);
 		return ERR_PTR(-ENOMEM);
@@ -1674,16 +1667,16 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 			/* use 4k pbl */
 			nes_debug(NES_DBG_CQ, "pbl_entries=%u, use a 4k PBL\n", pbl_entries);
 			if (nesadapter->free_4kpbl == 0) {
-				if (cqp_request->dynamic) {
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-					kfree(cqp_request);
-				} else {
-					list_add_tail(&cqp_request->list, &nesdev->cqp_avail_reqs);
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				}
+				spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+				nes_free_cqp_request(nesdev, cqp_request);
 				if (!context)
 					pci_free_consistent(nesdev->pcidev, nescq->cq_mem_size, mem,
 							nescq->hw_cq.cq_pbase);
+				else {
+					pci_free_consistent(nesdev->pcidev, nespbl->pbl_size,
+							    nespbl->pbl_vbase, nespbl->pbl_pbase);
+					kfree(nespbl);
+				}
 				nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 				kfree(nescq);
 				return ERR_PTR(-ENOMEM);
@@ -1696,16 +1689,16 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 			/* use 256 byte pbl */
 			nes_debug(NES_DBG_CQ, "pbl_entries=%u, use a 256 byte PBL\n", pbl_entries);
 			if (nesadapter->free_256pbl == 0) {
-				if (cqp_request->dynamic) {
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-					kfree(cqp_request);
-				} else {
-					list_add_tail(&cqp_request->list, &nesdev->cqp_avail_reqs);
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				}
+				spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+				nes_free_cqp_request(nesdev, cqp_request);
 				if (!context)
 					pci_free_consistent(nesdev->pcidev, nescq->cq_mem_size, mem,
 							nescq->hw_cq.cq_pbase);
+				else {
+					pci_free_consistent(nesdev->pcidev, nespbl->pbl_size,
+							    nespbl->pbl_vbase, nespbl->pbl_pbase);
+					kfree(nespbl);
+				}
 				nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 				kfree(nescq);
 				return ERR_PTR(-ENOMEM);
@@ -1758,6 +1751,11 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 		if (!context)
 			pci_free_consistent(nesdev->pcidev, nescq->cq_mem_size, mem,
 					nescq->hw_cq.cq_pbase);
+		else {
+			pci_free_consistent(nesdev->pcidev, nespbl->pbl_size,
+					    nespbl->pbl_vbase, nespbl->pbl_pbase);
+			kfree(nespbl);
+		}
 		nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 		kfree(nescq);
 		return ERR_PTR(-EIO);
@@ -1772,7 +1770,7 @@ static struct ib_cq *nes_create_cq(struct ib_device *ibdev, int entries,
 		resp.cq_id = nescq->hw_cq.cq_number;
 		resp.cq_size = nescq->hw_cq.cq_size;
 		resp.mmap_db_index = 0;
-		if (ib_copy_to_udata(udata, &resp, sizeof resp)) {
+		if (ib_copy_to_udata(udata, &resp, sizeof resp - sizeof resp.reserved)) {
 			nes_free_resource(nesadapter, nesadapter->allocated_cqs, cq_num);
 			kfree(nescq);
 			return ERR_PTR(-EFAULT);
@@ -1839,7 +1837,9 @@ static int nes_destroy_cq(struct ib_cq *ib_cq)
 	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX, opcode);
 	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX,
 		(nescq->hw_cq.cq_number | ((u32)PCI_FUNC(nesdev->pcidev->devfn) << 16)));
-	nes_free_resource(nesadapter, nesadapter->allocated_cqs, nescq->hw_cq.cq_number);
+	if (!nescq->mcrqf)
+		nes_free_resource(nesadapter, nesadapter->allocated_cqs, nescq->hw_cq.cq_number);
+
 	atomic_set(&cqp_request->refcount, 2);
 	nes_post_cqp_request(nesdev, cqp_request);
 
@@ -1873,21 +1873,74 @@ static int nes_destroy_cq(struct ib_cq *ib_cq)
 	return ret;
 }
 
+/**
+ * root_256
+ */
+static u32 root_256(struct nes_device *nesdev,
+		    struct nes_root_vpbl *root_vpbl,
+		    struct nes_root_vpbl *new_root,
+		    u16 pbl_count_4k)
+{
+	u64 leaf_pbl;
+	int i, j, k;
+
+	if (pbl_count_4k == 1) {
+		new_root->pbl_vbase = pci_alloc_consistent(nesdev->pcidev,
+						512, &new_root->pbl_pbase);
+
+		if (new_root->pbl_vbase == NULL)
+			return 0;
+
+		leaf_pbl = (u64)root_vpbl->pbl_pbase;
+		for (i = 0; i < 16; i++) {
+			new_root->pbl_vbase[i].pa_low =
+				cpu_to_le32((u32)leaf_pbl);
+			new_root->pbl_vbase[i].pa_high =
+				cpu_to_le32((u32)((((u64)leaf_pbl) >> 32)));
+			leaf_pbl += 256;
+		}
+	} else {
+		for (i = 3; i >= 0; i--) {
+			j = i * 16;
+			root_vpbl->pbl_vbase[j] = root_vpbl->pbl_vbase[i];
+			leaf_pbl = le32_to_cpu(root_vpbl->pbl_vbase[j].pa_low) +
+			    (((u64)le32_to_cpu(root_vpbl->pbl_vbase[j].pa_high))
+				<< 32);
+			for (k = 1; k < 16; k++) {
+				leaf_pbl += 256;
+				root_vpbl->pbl_vbase[j + k].pa_low =
+						cpu_to_le32((u32)leaf_pbl);
+				root_vpbl->pbl_vbase[j + k].pa_high =
+				    cpu_to_le32((u32)((((u64)leaf_pbl) >> 32)));
+			}
+		}
+	}
+
+	return 1;
+}
+
 
 /**
  * nes_reg_mr
  */
 static int nes_reg_mr(struct nes_device *nesdev, struct nes_pd *nespd,
 		u32 stag, u64 region_length, struct nes_root_vpbl *root_vpbl,
-		dma_addr_t single_buffer, u16 pbl_count, u16 residual_page_count,
-		int acc, u64 *iova_start)
+		dma_addr_t single_buffer, u16 pbl_count_4k,
+		u16 residual_page_count_4k, int acc, u64 *iova_start,
+		u16 *actual_pbl_cnt, u8 *used_4k_pbls)
 {
 	struct nes_hw_cqp_wqe *cqp_wqe;
 	struct nes_cqp_request *cqp_request;
 	unsigned long flags;
 	int ret;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
-	/* int count; */
+	uint pg_cnt = 0;
+	u16 pbl_count_256 = 0;
+	u16 pbl_count = 0;
+	u8  use_256_pbls = 0;
+	u8  use_4k_pbls = 0;
+	u16 use_two_level = (pbl_count_4k > 1) ? 1 : 0;
+	struct nes_root_vpbl new_root = { 0, NULL, NULL };
 	u32 opcode = 0;
 	u16 major_code;
 
@@ -1900,56 +1953,70 @@ static int nes_reg_mr(struct nes_device *nesdev, struct nes_pd *nespd,
 	cqp_request->waiting = 1;
 	cqp_wqe = &cqp_request->cqp_wqe;
 
-	spin_lock_irqsave(&nesadapter->pbl_lock, flags);
-	/* track PBL resources */
-	if (pbl_count != 0) {
-		if (pbl_count > 1) {
-			/* Two level PBL */
-			if ((pbl_count+1) > nesadapter->free_4kpbl) {
-				nes_debug(NES_DBG_MR, "Out of 4KB Pbls for two level request.\n");
-				if (cqp_request->dynamic) {
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-					kfree(cqp_request);
-				} else {
-					list_add_tail(&cqp_request->list, &nesdev->cqp_avail_reqs);
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				}
-				return -ENOMEM;
-			} else {
-				nesadapter->free_4kpbl -= pbl_count+1;
-			}
-		} else if (residual_page_count > 32) {
-			if (pbl_count > nesadapter->free_4kpbl) {
-				nes_debug(NES_DBG_MR, "Out of 4KB Pbls.\n");
-				if (cqp_request->dynamic) {
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-					kfree(cqp_request);
-				} else {
-					list_add_tail(&cqp_request->list, &nesdev->cqp_avail_reqs);
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				}
-				return -ENOMEM;
-			} else {
-				nesadapter->free_4kpbl -= pbl_count;
+	if (pbl_count_4k) {
+		spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+
+		pg_cnt = ((pbl_count_4k - 1) * 512) + residual_page_count_4k;
+		pbl_count_256 = (pg_cnt + 31) / 32;
+		if (pg_cnt <= 32) {
+			if (pbl_count_256 <= nesadapter->free_256pbl)
+				use_256_pbls = 1;
+			else if (pbl_count_4k <= nesadapter->free_4kpbl)
+				use_4k_pbls = 1;
+		} else if (pg_cnt <= 2048) {
+			if (((pbl_count_4k + use_two_level) <= nesadapter->free_4kpbl) &&
+			    (nesadapter->free_4kpbl > (nesadapter->max_4kpbl >> 1))) {
+				use_4k_pbls = 1;
+			} else if ((pbl_count_256 + 1) <= nesadapter->free_256pbl) {
+				use_256_pbls = 1;
+				use_two_level = 1;
+			} else if ((pbl_count_4k + use_two_level) <= nesadapter->free_4kpbl) {
+				use_4k_pbls = 1;
 			}
 		} else {
-			if (pbl_count > nesadapter->free_256pbl) {
-				nes_debug(NES_DBG_MR, "Out of 256B Pbls.\n");
-				if (cqp_request->dynamic) {
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-					kfree(cqp_request);
-				} else {
-					list_add_tail(&cqp_request->list, &nesdev->cqp_avail_reqs);
-					spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
-				}
-				return -ENOMEM;
-			} else {
-				nesadapter->free_256pbl -= pbl_count;
-			}
+			if ((pbl_count_4k + 1) <= nesadapter->free_4kpbl)
+				use_4k_pbls = 1;
+		}
+
+		if (use_256_pbls) {
+			pbl_count = pbl_count_256;
+			nesadapter->free_256pbl -= pbl_count + use_two_level;
+		} else if (use_4k_pbls) {
+			pbl_count =  pbl_count_4k;
+			nesadapter->free_4kpbl -= pbl_count + use_two_level;
+		} else {
+			spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+			nes_debug(NES_DBG_MR, "Out of Pbls\n");
+			nes_free_cqp_request(nesdev, cqp_request);
+			return -ENOMEM;
 		}
+
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
 	}
 
-	spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+	if (use_256_pbls && use_two_level) {
+		if (root_256(nesdev, root_vpbl, &new_root, pbl_count_4k) == 1) {
+			if (new_root.pbl_pbase != 0)
+				root_vpbl = &new_root;
+		} else {
+			spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+			nesadapter->free_256pbl += pbl_count_256 + use_two_level;
+			use_256_pbls = 0;
+
+			if (pbl_count_4k == 1)
+				use_two_level = 0;
+			pbl_count = pbl_count_4k;
+
+			if ((pbl_count_4k + use_two_level) <= nesadapter->free_4kpbl) {
+				nesadapter->free_4kpbl -= pbl_count + use_two_level;
+				use_4k_pbls = 1;
+			}
+			spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+
+			if (use_4k_pbls == 0)
+				return -ENOMEM;
+		}
+	}
 
 	opcode = NES_CQP_REGISTER_STAG | NES_CQP_STAG_RIGHTS_LOCAL_READ |
 					NES_CQP_STAG_VA_TO | NES_CQP_STAG_MR;
@@ -1978,10 +2045,9 @@ static int nes_reg_mr(struct nes_device *nesdev, struct nes_pd *nespd,
 	} else {
 		set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PA_LOW_IDX, root_vpbl->pbl_pbase);
 		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PBL_BLK_COUNT_IDX, pbl_count);
-		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PBL_LEN_IDX,
-				(((pbl_count - 1) * 4096) + (residual_page_count*8)));
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_STAG_WQE_PBL_LEN_IDX, (pg_cnt * 8));
 
-		if ((pbl_count > 1) || (residual_page_count > 32))
+		if (use_4k_pbls)
 			cqp_wqe->wqe_words[NES_CQP_WQE_OPCODE_IDX] |= cpu_to_le32(NES_CQP_STAG_PBL_BLK_SIZE);
 	}
 	barrier();
@@ -1998,13 +2064,25 @@ static int nes_reg_mr(struct nes_device *nesdev, struct nes_pd *nespd,
 	major_code = cqp_request->major_code;
 	nes_put_cqp_request(nesdev, cqp_request);
 
+	if ((!ret || major_code) && pbl_count != 0) {
+		spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+		if (use_256_pbls)
+			nesadapter->free_256pbl += pbl_count + use_two_level;
+		else if (use_4k_pbls)
+			nesadapter->free_4kpbl += pbl_count + use_two_level;
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+	}
+	if (new_root.pbl_pbase)
+		pci_free_consistent(nesdev->pcidev, 512, new_root.pbl_vbase,
+				    new_root.pbl_pbase);
+
 	if (!ret)
 		return -ETIME;
 	else if (major_code)
 		return -EIO;
-	else
-		return 0;
 
+	*actual_pbl_cnt = pbl_count + use_two_level;
+	*used_4k_pbls = use_4k_pbls;
 	return 0;
 }
 
@@ -2027,18 +2105,18 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 	struct nes_root_vpbl root_vpbl;
 	u32 stag;
 	u32 i;
+	unsigned long mask;
 	u32 stag_index = 0;
 	u32 next_stag_index = 0;
 	u32 driver_key = 0;
 	u32 root_pbl_index = 0;
 	u32 cur_pbl_index = 0;
-	int err = 0, pbl_depth = 0;
+	int err = 0;
 	int ret = 0;
 	u16 pbl_count = 0;
 	u8 single_page = 1;
 	u8 stag_key = 0;
 
-	pbl_depth = 0;
 	region_length = 0;
 	vpbl.pbl_vbase = NULL;
 	root_vpbl.pbl_vbase = NULL;
@@ -2055,8 +2133,11 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 		return ERR_PTR(-E2BIG);
 	}
 
+	if ((buffer_list[0].addr ^ *iova_start) & ~PAGE_MASK)
+		return ERR_PTR(-EINVAL);
+
 	err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs, nesadapter->max_mr,
-			&stag_index, &next_stag_index);
+			&stag_index, &next_stag_index, NES_RESOURCE_PHYS_MR);
 	if (err) {
 		return ERR_PTR(err);
 	}
@@ -2120,19 +2201,16 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 			root_pbl_index++;
 			cur_pbl_index = 0;
 		}
-		if (buffer_list[i].addr & ~PAGE_MASK) {
-			/* TODO: Unwind allocated buffers */
-			nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-			nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n",
-					(unsigned int) buffer_list[i].addr);
-			ibmr = ERR_PTR(-EINVAL);
-			kfree(nesmr);
-			goto reg_phys_err;
-		}
 
-		if (!buffer_list[i].size) {
+		mask = !buffer_list[i].size;
+		if (i != 0)
+			mask |= buffer_list[i].addr;
+		if (i != num_phys_buf - 1)
+			mask |= buffer_list[i].addr + buffer_list[i].size;
+
+		if (mask & ~PAGE_MASK) {
 			nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-			nes_debug(NES_DBG_MR, "Invalid Buffer Size\n");
+			nes_debug(NES_DBG_MR, "Invalid buffer addr or size\n");
 			ibmr = ERR_PTR(-EINVAL);
 			kfree(nesmr);
 			goto reg_phys_err;
@@ -2143,7 +2221,7 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 			if ((buffer_list[i-1].addr+PAGE_SIZE) != buffer_list[i].addr)
 				single_page = 0;
 		}
-		vpbl.pbl_vbase[cur_pbl_index].pa_low = cpu_to_le32((u32)buffer_list[i].addr);
+		vpbl.pbl_vbase[cur_pbl_index].pa_low = cpu_to_le32((u32)buffer_list[i].addr & PAGE_MASK);
 		vpbl.pbl_vbase[cur_pbl_index++].pa_high =
 				cpu_to_le32((u32)((((u64)buffer_list[i].addr) >> 32)));
 	}
@@ -2156,8 +2234,6 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 			" length = 0x%016lX, index = 0x%08X\n",
 			stag, (unsigned long)*iova_start, (unsigned long)region_length, stag_index);
 
-	region_length -= (*iova_start)&PAGE_MASK;
-
 	/* Make the leaf PBL the root if only one PBL */
 	if (root_pbl_index == 1) {
 		root_vpbl.pbl_pbase = vpbl.pbl_pbase;
@@ -2169,18 +2245,14 @@ static struct ib_mr *nes_reg_phys_mr(struct ib_pd *ib_pd,
 		pbl_count = root_pbl_index;
 	}
 	ret = nes_reg_mr(nesdev, nespd, stag, region_length, &root_vpbl,
-			buffer_list[0].addr, pbl_count, (u16)cur_pbl_index, acc, iova_start);
+			buffer_list[0].addr, pbl_count, (u16)cur_pbl_index, acc, iova_start,
+			&nesmr->pbls_used, &nesmr->pbl_4k);
 
 	if (ret == 0) {
 		nesmr->ibmr.rkey = stag;
 		nesmr->ibmr.lkey = stag;
 		nesmr->mode = IWNES_MEMREG_TYPE_MEM;
 		ibmr = &nesmr->ibmr;
-		nesmr->pbl_4k = ((pbl_count > 1) || (cur_pbl_index > 32)) ? 1 : 0;
-		nesmr->pbls_used = pbl_count;
-		if (pbl_count > 1) {
-			nesmr->pbls_used++;
-		}
 	} else {
 		kfree(nesmr);
 		ibmr = ERR_PTR(-ENOMEM);
@@ -2237,7 +2309,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	struct ib_mr *ibmr = ERR_PTR(-EINVAL);
-	struct ib_umem_chunk *chunk;
+	struct scatterlist *sg;
 	struct nes_ucontext *nes_ucontext;
 	struct nes_pbl *nespbl;
 	struct nes_mr *nesmr;
@@ -2245,7 +2317,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	struct nes_mem_reg_req req;
 	struct nes_vpbl vpbl;
 	struct nes_root_vpbl root_vpbl;
-	int nmap_index, page_index;
+	int entry, page_index;
 	int page_count = 0;
 	int err, pbl_depth = 0;
 	int chunk_pages;
@@ -2260,6 +2332,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 	u16 pbl_count;
 	u8 single_page = 1;
 	u8 stag_key;
+	int first_page = 1;
 
 	region = ib_umem_get(pd->uobject->context, start, length, acc, 0);
 	if (IS_ERR(region)) {
@@ -2273,8 +2346,10 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 
 	skip_pages = ((u32)region->offset) >> 12;
 
-	if (ib_copy_from_udata(&req, udata, sizeof(req)))
+	if (ib_copy_from_udata(&req, udata, sizeof(req))) {
+		ib_umem_release(region);
 		return ERR_PTR(-EFAULT);
+	}
 	nes_debug(NES_DBG_MR, "Memory Registration type = %08X.\n", req.reg_type);
 
 	switch (req.reg_type) {
@@ -2294,7 +2369,7 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			next_stag_index %= nesadapter->max_mr;
 
 			err = nes_alloc_resource(nesadapter, nesadapter->allocated_mrs,
-					nesadapter->max_mr, &stag_index, &next_stag_index);
+					nesadapter->max_mr, &stag_index, &next_stag_index, NES_RESOURCE_USER_MR);
 			if (err) {
 				ib_umem_release(region);
 				return ERR_PTR(err);
@@ -2308,128 +2383,125 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			}
 			nesmr->region = region;
 
-			list_for_each_entry(chunk, &region->chunk_list, list) {
-				nes_debug(NES_DBG_MR, "Chunk: nents = %u, nmap = %u .\n",
-						chunk->nents, chunk->nmap);
-				for (nmap_index = 0; nmap_index < chunk->nmap; ++nmap_index) {
-					if (sg_dma_address(&chunk->page_list[nmap_index]) & ~PAGE_MASK) {
-						ib_umem_release(region);
-						nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-						nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n",
-								(unsigned int) sg_dma_address(&chunk->page_list[nmap_index]));
-						ibmr = ERR_PTR(-EINVAL);
-						kfree(nesmr);
-						goto reg_user_mr_err;
-					}
+			for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) {
+				if (sg_dma_address(sg) & ~PAGE_MASK) {
+					ib_umem_release(region);
+					nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+					nes_debug(NES_DBG_MR, "Unaligned Memory Buffer: 0x%x\n",
+						  (unsigned int) sg_dma_address(sg));
+					ibmr = ERR_PTR(-EINVAL);
+					kfree(nesmr);
+					goto reg_user_mr_err;
+				}
 
-					if (!sg_dma_len(&chunk->page_list[nmap_index])) {
-						ib_umem_release(region);
-						nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-								stag_index);
-						nes_debug(NES_DBG_MR, "Invalid Buffer Size\n");
-						ibmr = ERR_PTR(-EINVAL);
-						kfree(nesmr);
-						goto reg_user_mr_err;
-					}
+				if (!sg_dma_len(sg)) {
+					ib_umem_release(region);
+					nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+							  stag_index);
+					nes_debug(NES_DBG_MR, "Invalid Buffer Size\n");
+					ibmr = ERR_PTR(-EINVAL);
+					kfree(nesmr);
+					goto reg_user_mr_err;
+				}
 
-					region_length += sg_dma_len(&chunk->page_list[nmap_index]);
-					chunk_pages = sg_dma_len(&chunk->page_list[nmap_index]) >> 12;
-					region_length -= skip_pages << 12;
-					for (page_index=skip_pages; page_index < chunk_pages; page_index++) {
-						skip_pages = 0;
-						if ((page_count!=0)&&(page_count<<12)-(region->offset&(4096-1))>=region->length)
-							goto enough_pages;
-						if ((page_count&0x01FF) == 0) {
-							if (page_count >= 1024 * 512) {
+				region_length += sg_dma_len(sg);
+				chunk_pages = sg_dma_len(sg) >> 12;
+				region_length -= skip_pages << 12;
+				for (page_index = skip_pages; page_index < chunk_pages; page_index++) {
+					skip_pages = 0;
+					if ((page_count != 0) && (page_count<<12)-(region->offset&(4096-1)) >= region->length)
+						goto enough_pages;
+					if ((page_count&0x01FF) == 0) {
+						if (page_count >= 1024 * 512) {
+							ib_umem_release(region);
+							nes_free_resource(nesadapter,
+									  nesadapter->allocated_mrs, stag_index);
+							kfree(nesmr);
+							ibmr = ERR_PTR(-E2BIG);
+							goto reg_user_mr_err;
+						}
+						if (root_pbl_index == 1) {
+							root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev,
+									8192, &root_vpbl.pbl_pbase);
+							nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n",
+								  root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase);
+							if (!root_vpbl.pbl_vbase) {
 								ib_umem_release(region);
-								nes_free_resource(nesadapter,
-										nesadapter->allocated_mrs, stag_index);
+								pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
+										    vpbl.pbl_pbase);
+								nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+										  stag_index);
 								kfree(nesmr);
-								ibmr = ERR_PTR(-E2BIG);
+								ibmr = ERR_PTR(-ENOMEM);
 								goto reg_user_mr_err;
 							}
-							if (root_pbl_index == 1) {
-								root_vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev,
-										8192, &root_vpbl.pbl_pbase);
-								nes_debug(NES_DBG_MR, "Allocating root PBL, va = %p, pa = 0x%08X\n",
-										root_vpbl.pbl_vbase, (unsigned int)root_vpbl.pbl_pbase);
-								if (!root_vpbl.pbl_vbase) {
-									ib_umem_release(region);
-									pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
-											vpbl.pbl_pbase);
-									nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-											stag_index);
-									kfree(nesmr);
-									ibmr = ERR_PTR(-ENOMEM);
-									goto reg_user_mr_err;
-								}
-								root_vpbl.leaf_vpbl = kzalloc(sizeof(*root_vpbl.leaf_vpbl)*1024,
-										GFP_KERNEL);
-								if (!root_vpbl.leaf_vpbl) {
-									ib_umem_release(region);
-									pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase,
-											root_vpbl.pbl_pbase);
-									pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
-											vpbl.pbl_pbase);
-									nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-											stag_index);
-									kfree(nesmr);
-									ibmr = ERR_PTR(-ENOMEM);
-									goto reg_user_mr_err;
-								}
-								root_vpbl.pbl_vbase[0].pa_low =
-										cpu_to_le32((u32)vpbl.pbl_pbase);
-								root_vpbl.pbl_vbase[0].pa_high =
-										cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32)));
-								root_vpbl.leaf_vpbl[0] = vpbl;
-							}
-							vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
-									&vpbl.pbl_pbase);
-							nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n",
-									vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase);
-							if (!vpbl.pbl_vbase) {
+							root_vpbl.leaf_vpbl = kzalloc(sizeof(*root_vpbl.leaf_vpbl)*1024,
+									GFP_KERNEL);
+							if (!root_vpbl.leaf_vpbl) {
 								ib_umem_release(region);
-								nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
-								ibmr = ERR_PTR(-ENOMEM);
+								pci_free_consistent(nesdev->pcidev, 8192, root_vpbl.pbl_vbase,
+										    root_vpbl.pbl_pbase);
+								pci_free_consistent(nesdev->pcidev, 4096, vpbl.pbl_vbase,
+										    vpbl.pbl_pbase);
+								nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+										  stag_index);
 								kfree(nesmr);
+								ibmr = ERR_PTR(-ENOMEM);
 								goto reg_user_mr_err;
 							}
-							if (1 <= root_pbl_index) {
-								root_vpbl.pbl_vbase[root_pbl_index].pa_low =
-										cpu_to_le32((u32)vpbl.pbl_pbase);
-								root_vpbl.pbl_vbase[root_pbl_index].pa_high =
-										cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32)));
-								root_vpbl.leaf_vpbl[root_pbl_index] = vpbl;
-							}
-							root_pbl_index++;
-							cur_pbl_index = 0;
+							root_vpbl.pbl_vbase[0].pa_low =
+									cpu_to_le32((u32)vpbl.pbl_pbase);
+							root_vpbl.pbl_vbase[0].pa_high =
+									cpu_to_le32((u32)((((u64)vpbl.pbl_pbase) >> 32)));
+							root_vpbl.leaf_vpbl[0] = vpbl;
 						}
-						if (single_page) {
-							if (page_count != 0) {
-								if ((last_dma_addr+4096) !=
-										(sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096)))
-									single_page = 0;
-								last_dma_addr = sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096);
-							} else {
-								first_dma_addr = sg_dma_address(&chunk->page_list[nmap_index])+
-										(page_index*4096);
-								last_dma_addr = first_dma_addr;
-							}
+						vpbl.pbl_vbase = pci_alloc_consistent(nesdev->pcidev, 4096,
+								&vpbl.pbl_pbase);
+						nes_debug(NES_DBG_MR, "Allocating leaf PBL, va = %p, pa = 0x%08X\n",
+							  vpbl.pbl_vbase, (unsigned int)vpbl.pbl_pbase);
+						if (!vpbl.pbl_vbase) {
+							ib_umem_release(region);
+							nes_free_resource(nesadapter, nesadapter->allocated_mrs, stag_index);
+							ibmr = ERR_PTR(-ENOMEM);
+							kfree(nesmr);
+							goto reg_user_mr_err;
 						}
-
-						vpbl.pbl_vbase[cur_pbl_index].pa_low =
-								cpu_to_le32((u32)(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)));
-						vpbl.pbl_vbase[cur_pbl_index].pa_high =
-								cpu_to_le32((u32)((((u64)(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096))) >> 32)));
-						cur_pbl_index++;
-						page_count++;
+						if (1 <= root_pbl_index) {
+							root_vpbl.pbl_vbase[root_pbl_index].pa_low =
+									cpu_to_le32((u32)vpbl.pbl_pbase);
+							root_vpbl.pbl_vbase[root_pbl_index].pa_high =
+									cpu_to_le32((u32)((((u64)vpbl.pbl_pbase)>>32)));
+							root_vpbl.leaf_vpbl[root_pbl_index] = vpbl;
+						}
+						root_pbl_index++;
+						cur_pbl_index = 0;
 					}
+					if (single_page) {
+						if (page_count != 0) {
+							if ((last_dma_addr+4096) !=
+									(sg_dma_address(sg)+
+									(page_index*4096)))
+								single_page = 0;
+							last_dma_addr = sg_dma_address(sg)+
+									(page_index*4096);
+						} else {
+							first_dma_addr = sg_dma_address(sg)+
+									(page_index*4096);
+							last_dma_addr = first_dma_addr;
+						}
+					}
+
+					vpbl.pbl_vbase[cur_pbl_index].pa_low =
+							cpu_to_le32((u32)(sg_dma_address(sg)+
+							(page_index*4096)));
+					vpbl.pbl_vbase[cur_pbl_index].pa_high =
+							cpu_to_le32((u32)((((u64)(sg_dma_address(sg)+
+							(page_index*4096))) >> 32)));
+					cur_pbl_index++;
+					page_count++;
 				}
 			}
+
 			enough_pages:
 			nes_debug(NES_DBG_MR, "calculating stag, stag_index=0x%08x, driver_key=0x%08x,"
 					" stag_key=0x%08x\n",
@@ -2437,9 +2509,6 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			stag = stag_index << 8;
 			stag |= driver_key;
 			stag += (u32)stag_key;
-			if (stag == 0) {
-				stag = 1;
-			}
 
 			iova_start = virt;
 			/* Make the leaf PBL the root if only one PBL */
@@ -2458,8 +2527,9 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 					stag, (unsigned int)iova_start,
 					(unsigned int)region_length, stag_index,
 					(unsigned long long)region->length, pbl_count);
-			ret = nes_reg_mr( nesdev, nespd, stag, region->length, &root_vpbl,
-					first_dma_addr, pbl_count, (u16)cur_pbl_index, acc, &iova_start);
+			ret = nes_reg_mr(nesdev, nespd, stag, region->length, &root_vpbl,
+					 first_dma_addr, pbl_count, (u16)cur_pbl_index, acc,
+					 &iova_start, &nesmr->pbls_used, &nesmr->pbl_4k);
 
 			nes_debug(NES_DBG_MR, "ret=%d\n", ret);
 
@@ -2468,11 +2538,6 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				nesmr->ibmr.lkey = stag;
 				nesmr->mode = IWNES_MEMREG_TYPE_MEM;
 				ibmr = &nesmr->ibmr;
-				nesmr->pbl_4k = ((pbl_count > 1) || (cur_pbl_index > 32)) ? 1 : 0;
-				nesmr->pbls_used = pbl_count;
-				if (pbl_count > 1) {
-					nesmr->pbls_used++;
-				}
 			} else {
 				ib_umem_release(region);
 				kfree(nesmr);
@@ -2498,9 +2563,13 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			nes_debug(NES_DBG_MR, "Leaving, ibmr=%p", ibmr);
 
 			return ibmr;
-			break;
 		case IWNES_MEMREG_TYPE_QP:
 		case IWNES_MEMREG_TYPE_CQ:
+			if (!region->length) {
+				nes_debug(NES_DBG_MR, "Unable to register zero length region for CQ\n");
+				ib_umem_release(region);
+				return ERR_PTR(-EINVAL);
+			}
 			nespbl = kzalloc(sizeof(*nespbl), GFP_KERNEL);
 			if (!nespbl) {
 				nes_debug(NES_DBG_MR, "Unable to allocate PBL\n");
@@ -2544,25 +2613,28 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 				  nespbl->pbl_size, (unsigned long) nespbl->pbl_pbase,
 				  (void *) nespbl->pbl_vbase, nespbl->user_base);
 
-			list_for_each_entry(chunk, &region->chunk_list, list) {
-				for (nmap_index = 0; nmap_index < chunk->nmap; ++nmap_index) {
-					chunk_pages = sg_dma_len(&chunk->page_list[nmap_index]) >> 12;
-					chunk_pages += (sg_dma_len(&chunk->page_list[nmap_index]) & (4096-1)) ? 1 : 0;
-					nespbl->page = sg_page(&chunk->page_list[0]);
-					for (page_index=0; page_index<chunk_pages; page_index++) {
-						((__le32 *)pbl)[0] = cpu_to_le32((u32)
-								(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)));
-						((__le32 *)pbl)[1] = cpu_to_le32(((u64)
-								(sg_dma_address(&chunk->page_list[nmap_index])+
-								(page_index*4096)))>>32);
-						nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl,
-								(unsigned long long)*pbl,
-								le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0]));
-						pbl++;
-					}
+			for_each_sg(region->sg_head.sgl, sg, region->nmap, entry) {
+				chunk_pages = sg_dma_len(sg) >> 12;
+				chunk_pages += (sg_dma_len(sg) & (4096-1)) ? 1 : 0;
+				if (first_page) {
+					nespbl->page = sg_page(sg);
+					first_page = 0;
+				}
+
+				for (page_index = 0; page_index < chunk_pages; page_index++) {
+					((__le32 *)pbl)[0] = cpu_to_le32((u32)
+							(sg_dma_address(sg)+
+							(page_index*4096)));
+					((__le32 *)pbl)[1] = cpu_to_le32(((u64)
+							(sg_dma_address(sg)+
+							(page_index*4096)))>>32);
+					nes_debug(NES_DBG_MR, "pbl=%p, *pbl=0x%016llx, 0x%08x%08x\n", pbl,
+						  (unsigned long long)*pbl,
+						  le32_to_cpu(((__le32 *)pbl)[1]), le32_to_cpu(((__le32 *)pbl)[0]));
+					pbl++;
 				}
 			}
+
 			if (req.reg_type == IWNES_MEMREG_TYPE_QP) {
 				list_add_tail(&nespbl->list, &nes_ucontext->qp_reg_mem_list);
 			} else {
@@ -2572,9 +2644,9 @@ static struct ib_mr *nes_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
 			nesmr->ibmr.lkey = -1;
 			nesmr->mode = req.reg_type;
 			return &nesmr->ibmr;
-			break;
 	}
 
+	ib_umem_release(region);
 	return ERR_PTR(-ENOSYS);
 }
 
@@ -2613,24 +2685,6 @@ static int nes_dereg_mr(struct ib_mr *ib_mr)
 	cqp_request->waiting = 1;
 	cqp_wqe = &cqp_request->cqp_wqe;
 
-	spin_lock_irqsave(&nesadapter->pbl_lock, flags);
-	if (nesmr->pbls_used != 0) {
-		if (nesmr->pbl_4k) {
-			nesadapter->free_4kpbl += nesmr->pbls_used;
-			if (nesadapter->free_4kpbl > nesadapter->max_4kpbl) {
-				printk(KERN_ERR PFX "free 4KB PBLs(%u) has exceeded the max(%u)\n",
-						nesadapter->free_4kpbl, nesadapter->max_4kpbl);
-			}
-		} else {
-			nesadapter->free_256pbl += nesmr->pbls_used;
-			if (nesadapter->free_256pbl > nesadapter->max_256pbl) {
-				printk(KERN_ERR PFX "free 256B PBLs(%u) has exceeded the max(%u)\n",
-						nesadapter->free_256pbl, nesadapter->max_256pbl);
-			}
-		}
-	}
-
-	spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
 	nes_fill_init_cqp_wqe(cqp_wqe, nesdev);
 	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_OPCODE_IDX,
 			NES_CQP_DEALLOCATE_STAG | NES_CQP_STAG_VA_TO |
@@ -2648,11 +2702,6 @@ static int nes_dereg_mr(struct ib_mr *ib_mr)
 			" CQP Major:Minor codes = 0x%04X:0x%04X\n",
 			ib_mr->rkey, ret, cqp_request->major_code, cqp_request->minor_code);
 
-	nes_free_resource(nesadapter, nesadapter->allocated_mrs,
-			(ib_mr->rkey & 0x0fffff00) >> 8);
-
-	kfree(nesmr);
-
 	major_code = cqp_request->major_code;
 	minor_code = cqp_request->minor_code;
 
@@ -2668,8 +2717,33 @@ static int nes_dereg_mr(struct ib_mr *ib_mr)
 				" to destroy STag, ib_mr=%p, rkey = 0x%08X\n",
 				major_code, minor_code, ib_mr, ib_mr->rkey);
 		return -EIO;
-	} else
-		return 0;
+	}
+
+	if (nesmr->pbls_used != 0) {
+		spin_lock_irqsave(&nesadapter->pbl_lock, flags);
+		if (nesmr->pbl_4k) {
+			nesadapter->free_4kpbl += nesmr->pbls_used;
+			if (nesadapter->free_4kpbl > nesadapter->max_4kpbl)
+				printk(KERN_ERR PFX "free 4KB PBLs(%u) has "
+					"exceeded the max(%u)\n",
+					nesadapter->free_4kpbl,
+					nesadapter->max_4kpbl);
+		} else {
+			nesadapter->free_256pbl += nesmr->pbls_used;
+			if (nesadapter->free_256pbl > nesadapter->max_256pbl)
+				printk(KERN_ERR PFX "free 256B PBLs(%u) has "
+					"exceeded the max(%u)\n",
+					nesadapter->free_256pbl,
+					nesadapter->max_256pbl);
+		}
+		spin_unlock_irqrestore(&nesadapter->pbl_lock, flags);
+	}
+	nes_free_resource(nesadapter, nesadapter->allocated_mrs,
+			(ib_mr->rkey & 0x0fffff00) >> 8);
+
+	kfree(nesmr);
+
+	return 0;
 }
 
 
@@ -2699,10 +2773,9 @@ static ssize_t show_fw_ver(struct device *dev, struct device_attribute *attr,
 	struct nes_vnic *nesvnic = nesibdev->nesvnic;
 
 	nes_debug(NES_DBG_INIT, "\n");
-	return sprintf(buf, "%x.%x.%x\n",
-			(int)(nesvnic->nesdev->nesadapter->fw_ver >> 32),
-			(int)(nesvnic->nesdev->nesadapter->fw_ver >> 16) & 0xffff,
-			(int)(nesvnic->nesdev->nesadapter->fw_ver & 0xffff));
+	return sprintf(buf, "%u.%u\n",
+		(nesvnic->nesdev->nesadapter->firmware_version >> 16),
+		(nesvnic->nesdev->nesadapter->firmware_version & 0x000000ff));
 }
 
 
@@ -2755,17 +2828,16 @@ static int nes_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	attr->cap.max_send_wr = nesqp->hwqp.sq_size;
 	attr->cap.max_recv_wr = nesqp->hwqp.rq_size;
 	attr->cap.max_recv_sge = 1;
-	if (nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) {
-		init_attr->cap.max_inline_data = 0;
-	} else {
-		init_attr->cap.max_inline_data = 64;
-	}
+	if (nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA)
+		attr->cap.max_inline_data = 0;
+	else
+		attr->cap.max_inline_data = 64;
 
 	init_attr->event_handler = nesqp->ibqp.event_handler;
 	init_attr->qp_context = nesqp->ibqp.qp_context;
 	init_attr->send_cq = nesqp->ibqp.send_cq;
 	init_attr->recv_cq = nesqp->ibqp.recv_cq;
-	init_attr->srq = nesqp->ibqp.srq = nesqp->ibqp.srq;
+	init_attr->srq = nesqp->ibqp.srq;
 	init_attr->cap = attr->cap;
 
 	return 0;
@@ -2776,7 +2848,7 @@ static int nes_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
  * nes_hw_modify_qp
  */
 int nes_hw_modify_qp(struct nes_device *nesdev, struct nes_qp *nesqp,
-		u32 next_iwarp_state, u32 wait_completion)
+		u32 next_iwarp_state, u32 termlen, u32 wait_completion)
 {
 	struct nes_hw_cqp_wqe *cqp_wqe;
 	/* struct iw_cm_id *cm_id = nesqp->cm_id; */
@@ -2808,6 +2880,13 @@ int nes_hw_modify_qp(struct nes_device *nesdev, struct nes_qp *nesqp,
 	set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_WQE_ID_IDX, nesqp->hwqp.qp_id);
 	set_wqe_64bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_CONTEXT_LOW_IDX, (u64)nesqp->nesqp_context_pbase);
 
+	/* If sending a terminate message, fill in the length (in words) */
+	if (((next_iwarp_state & NES_CQP_QP_IWARP_STATE_MASK) == NES_CQP_QP_IWARP_STATE_TERMINATE) &&
+	    !(next_iwarp_state & NES_CQP_QP_TERM_DONT_SEND_TERM_MSG)) {
+		termlen = ((termlen + 3) >> 2) << NES_CQP_OP_TERMLEN_SHIFT;
+		set_wqe_32bit_value(cqp_wqe->wqe_words, NES_CQP_QP_WQE_NEW_MSS_IDX, termlen);
+	}
+
 	atomic_set(&cqp_request->refcount, 2);
 	nes_post_cqp_request(nesdev, cqp_request);
 
@@ -2859,7 +2938,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 	int ret;
 	u16 original_last_aeq;
 	u8 issue_modify_qp = 0;
-	u8 issue_disconnect = 0;
 	u8 dont_wait = 0;
 
 	nes_debug(NES_DBG_MOD_QP, "QP%u: QP State=%u, cur QP State=%u,"
@@ -2867,7 +2945,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 			nesqp->hwqp.qp_id, attr->qp_state, nesqp->ibqp_state,
 			nesqp->iwarp_state, atomic_read(&nesqp->refcount));
 
-	nes_add_ref(&nesqp->ibqp);
 	spin_lock_irqsave(&nesqp->lock, qplockflags);
 
 	nes_debug(NES_DBG_MOD_QP, "QP%u: hw_iwarp_state=0x%X, hw_tcp_state=0x%X,"
@@ -2882,7 +2959,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state > (u32)NES_CQP_QP_IWARP_STATE_IDLE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_IDLE;
@@ -2893,7 +2969,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>(u32)NES_CQP_QP_IWARP_STATE_IDLE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_IDLE;
@@ -2904,14 +2979,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>(u32)NES_CQP_QP_IWARP_STATE_RTS) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				if (nesqp->cm_id == NULL) {
 					nes_debug(NES_DBG_MOD_QP, "QP%u: Failing attempt to move QP to RTS without a CM_ID. \n",
 							nesqp->hwqp.qp_id );
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_RTS;
@@ -2929,7 +3002,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id, nesqp->hwqp.sq_head, nesqp->hwqp.sq_tail);
 				if (nesqp->iwarp_state == (u32)NES_CQP_QP_IWARP_STATE_CLOSING) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return 0;
 				} else {
 					if (nesqp->iwarp_state > (u32)NES_CQP_QP_IWARP_STATE_CLOSING) {
@@ -2937,7 +3009,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 								" ignored due to current iWARP state\n",
 								nesqp->hwqp.qp_id);
 						spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-						nes_rem_ref(&nesqp->ibqp);
 						return -EINVAL;
 					}
 					if (nesqp->hw_iwarp_state != NES_AEQE_IWARP_STATE_RTS) {
@@ -2945,11 +3016,11 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 								" already done based on hw state.\n",
 								nesqp->hwqp.qp_id);
 						issue_modify_qp = 0;
-						nesqp->in_disconnect = 0;
 					}
 					switch (nesqp->hw_iwarp_state) {
 						case NES_AEQE_IWARP_STATE_CLOSING:
 							next_iwarp_state = NES_CQP_QP_IWARP_STATE_CLOSING;
+							break;
 						case NES_AEQE_IWARP_STATE_TERMINATE:
 							next_iwarp_state = NES_CQP_QP_IWARP_STATE_TERMINATE;
 							break;
@@ -2958,7 +3029,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 							break;
 						default:
 							next_iwarp_state = NES_CQP_QP_IWARP_STATE_CLOSING;
-							nesqp->in_disconnect = 1;
 							nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_CLOSING;
 							break;
 					}
@@ -2969,24 +3039,24 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hwqp.qp_id);
 				if (nesqp->iwarp_state>=(u32)NES_CQP_QP_IWARP_STATE_TERMINATE) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				/* next_iwarp_state = (NES_CQP_QP_IWARP_STATE_TERMINATE | 0x02000000); */
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_TERMINATE;
 				nesqp->hw_iwarp_state = NES_AEQE_IWARP_STATE_TERMINATE;
 				issue_modify_qp = 1;
-				nesqp->in_disconnect = 1;
 				break;
 			case IB_QPS_ERR:
 			case IB_QPS_RESET:
 				if (nesqp->iwarp_state == (u32)NES_CQP_QP_IWARP_STATE_ERROR) {
 					spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-					nes_rem_ref(&nesqp->ibqp);
 					return -EINVAL;
 				}
 				nes_debug(NES_DBG_MOD_QP, "QP%u: new state = error\n",
 						nesqp->hwqp.qp_id);
+				if (nesqp->term_flags)
+					del_timer(&nesqp->terminate_timer);
+
 				next_iwarp_state = NES_CQP_QP_IWARP_STATE_ERROR;
 				/* next_iwarp_state = (NES_CQP_QP_IWARP_STATE_TERMINATE | 0x02000000); */
 					if (nesqp->hte_added) {
@@ -2995,9 +3065,9 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 						nesqp->hte_added = 0;
 					}
 				if ((nesqp->hw_tcp_state > NES_AEQE_TCP_STATE_CLOSED) &&
+						(nesdev->iw_status) &&
 						(nesqp->hw_tcp_state != NES_AEQE_TCP_STATE_TIME_WAIT)) {
 					next_iwarp_state |= NES_CQP_QP_RESET;
-					nesqp->in_disconnect = 1;
 				} else {
 					nes_debug(NES_DBG_MOD_QP, "QP%u NOT setting NES_CQP_QP_RESET since TCP state = %u\n",
 							nesqp->hwqp.qp_id, nesqp->hw_tcp_state);
@@ -3008,25 +3078,14 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 				break;
 			default:
 				spin_unlock_irqrestore(&nesqp->lock, qplockflags);
-				nes_rem_ref(&nesqp->ibqp);
 				return -EINVAL;
 				break;
 		}
 
 		nesqp->ibqp_state = attr->qp_state;
-		if (((nesqp->iwarp_state & NES_CQP_QP_IWARP_STATE_MASK) ==
-				(u32)NES_CQP_QP_IWARP_STATE_RTS) &&
-				((next_iwarp_state & NES_CQP_QP_IWARP_STATE_MASK) >
-				(u32)NES_CQP_QP_IWARP_STATE_RTS)) {
-			nesqp->iwarp_state = next_iwarp_state & NES_CQP_QP_IWARP_STATE_MASK;
-			nes_debug(NES_DBG_MOD_QP, "Change nesqp->iwarp_state=%08x\n",
-					nesqp->iwarp_state);
-			issue_disconnect = 1;
-		} else {
-			nesqp->iwarp_state = next_iwarp_state & NES_CQP_QP_IWARP_STATE_MASK;
-			nes_debug(NES_DBG_MOD_QP, "Change nesqp->iwarp_state=%08x\n",
-					nesqp->iwarp_state);
-		}
+		nesqp->iwarp_state = next_iwarp_state & NES_CQP_QP_IWARP_STATE_MASK;
+		nes_debug(NES_DBG_MOD_QP, "Change nesqp->iwarp_state=%08x\n",
+				nesqp->iwarp_state);
 	}
 
 	if (attr_mask & IB_QP_ACCESS_FLAGS) {
@@ -3065,7 +3124,7 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 
 	if (issue_modify_qp) {
 		nes_debug(NES_DBG_MOD_QP, "call nes_hw_modify_qp\n");
-		ret = nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 1);
+		ret = nes_hw_modify_qp(nesdev, nesqp, next_iwarp_state, 0, 1);
 		if (ret)
 			nes_debug(NES_DBG_MOD_QP, "nes_hw_modify_qp (next_iwarp_state = 0x%08X)"
 					" failed for QP%u.\n",
@@ -3078,9 +3137,7 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 				" original_last_aeq = 0x%04X. last_aeq = 0x%04X.\n",
 				nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 				original_last_aeq, nesqp->last_aeq);
-		if ((!ret) ||
-				((original_last_aeq != NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) &&
-				(ret))) {
+		if (!ret || original_last_aeq != NES_AEQE_AEID_RDMAP_ROE_BAD_LLP_CLOSE) {
 			if (dont_wait) {
 				if (nesqp->cm_id && nesqp->hw_tcp_state != 0) {
 					nes_debug(NES_DBG_MOD_QP, "QP%u Queuing fake disconnect for QP refcount (%d),"
@@ -3088,7 +3145,6 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 							nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 							original_last_aeq, nesqp->last_aeq);
 					/* this one is for the cm_disconnect thread */
-					nes_add_ref(&nesqp->ibqp);
 					spin_lock_irqsave(&nesqp->lock, qplockflags);
 					nesqp->hw_tcp_state = NES_AEQE_TCP_STATE_CLOSED;
 					nesqp->last_aeq = NES_AEQE_AEID_RESET_SENT;
@@ -3097,14 +3153,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 				} else {
 					nes_debug(NES_DBG_MOD_QP, "QP%u No fake disconnect, QP refcount=%d\n",
 							nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount));
-					nes_rem_ref(&nesqp->ibqp);
 				}
 			} else {
 				spin_lock_irqsave(&nesqp->lock, qplockflags);
 				if (nesqp->cm_id) {
 					/* These two are for the timer thread */
 					if (atomic_inc_return(&nesqp->close_timer_started) == 1) {
-						nes_add_ref(&nesqp->ibqp);
 						nesqp->cm_id->add_ref(nesqp->cm_id);
 						nes_debug(NES_DBG_MOD_QP, "QP%u Not decrementing QP refcount (%d),"
 								" need ae to finish up, original_last_aeq = 0x%04X."
@@ -3128,14 +3182,12 @@ int nes_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
 					" original_last_aeq = 0x%04X. last_aeq = 0x%04X.\n",
 					nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 					original_last_aeq, nesqp->last_aeq);
-			nes_rem_ref(&nesqp->ibqp);
 		}
 	} else {
 		nes_debug(NES_DBG_MOD_QP, "QP%u Decrementing QP refcount (%d), No ae to finish up,"
 				" original_last_aeq = 0x%04X. last_aeq = 0x%04X.\n",
 				nesqp->hwqp.qp_id, atomic_read(&nesqp->refcount),
 				original_last_aeq, nesqp->last_aeq);
-		nes_rem_ref(&nesqp->ibqp);
 	}
 
 	err = 0;
@@ -3214,30 +3266,32 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 	struct nes_device *nesdev = nesvnic->nesdev;
 	struct nes_qp *nesqp = to_nesqp(ibqp);
 	struct nes_hw_qp_wqe *wqe;
-	int err;
+	int err = 0;
 	u32 qsize = nesqp->hwqp.sq_size;
 	u32 head;
-	u32 wqe_misc;
-	u32 wqe_count;
+	u32 wqe_misc = 0;
+	u32 wqe_count = 0;
 	u32 counter;
-	u32 total_payload_length;
-
-	err = 0;
-	wqe_misc = 0;
-	wqe_count = 0;
-	total_payload_length = 0;
 
-	if (nesqp->ibqp_state > IB_QPS_RTS)
-		return -EINVAL;
+	if (nesqp->ibqp_state > IB_QPS_RTS) {
+		err = -EINVAL;
+		goto out;
+	}
 
-		spin_lock_irqsave(&nesqp->lock, flags);
+	spin_lock_irqsave(&nesqp->lock, flags);
 
 	head = nesqp->hwqp.sq_head;
 
 	while (ib_wr) {
+		/* Check for QP error */
+		if (nesqp->term_flags) {
+			err = -EINVAL;
+			break;
+		}
+
 		/* Check for SQ overflow */
 		if (((head + (2 * qsize) - nesqp->hwqp.sq_tail) % qsize) == (qsize - 1)) {
-			err = -EINVAL;
+			err = -ENOMEM;
 			break;
 		}
 
@@ -3248,94 +3302,210 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 		u64temp = (u64)(ib_wr->wr_id);
 		set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_COMP_SCRATCH_LOW_IDX,
 					u64temp);
-			switch (ib_wr->opcode) {
-				case IB_WR_SEND:
-					if (ib_wr->send_flags & IB_SEND_SOLICITED) {
-						wqe_misc = NES_IWARP_SQ_OP_SENDSE;
-					} else {
-						wqe_misc = NES_IWARP_SQ_OP_SEND;
-					}
-					if (ib_wr->num_sge > nesdev->nesadapter->max_sge) {
-						err = -EINVAL;
-						break;
-					}
-					if (ib_wr->send_flags & IB_SEND_FENCE) {
-						wqe_misc |= NES_IWARP_SQ_WQE_LOCAL_FENCE;
-					}
-					if ((ib_wr->send_flags & IB_SEND_INLINE) &&
-							((nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) == 0) &&
-							(ib_wr->sg_list[0].length <= 64)) {
-						memcpy(&wqe->wqe_words[NES_IWARP_SQ_WQE_IMM_DATA_START_IDX],
-							       (void *)(unsigned long)ib_wr->sg_list[0].addr, ib_wr->sg_list[0].length);
-						set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX,
-								ib_wr->sg_list[0].length);
-						wqe_misc |= NES_IWARP_SQ_WQE_IMM_DATA;
-					} else {
-						fill_wqe_sg_send(wqe, ib_wr, 1);
-					}
+		switch (ib_wr->opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_INV:
+			if (IB_WR_SEND == ib_wr->opcode) {
+				if (ib_wr->send_flags & IB_SEND_SOLICITED)
+					wqe_misc = NES_IWARP_SQ_OP_SENDSE;
+				else
+					wqe_misc = NES_IWARP_SQ_OP_SEND;
+			} else {
+				if (ib_wr->send_flags & IB_SEND_SOLICITED)
+					wqe_misc = NES_IWARP_SQ_OP_SENDSEINV;
+				else
+					wqe_misc = NES_IWARP_SQ_OP_SENDINV;
 
-					break;
-				case IB_WR_RDMA_WRITE:
-					wqe_misc = NES_IWARP_SQ_OP_RDMAW;
-					if (ib_wr->num_sge > nesdev->nesadapter->max_sge) {
-						nes_debug(NES_DBG_IW_TX, "Exceeded max sge, ib_wr=%u, max=%u\n",
-								ib_wr->num_sge,
-								nesdev->nesadapter->max_sge);
-						err = -EINVAL;
-						break;
-					}
-					if (ib_wr->send_flags & IB_SEND_FENCE) {
-						wqe_misc |= NES_IWARP_SQ_WQE_LOCAL_FENCE;
-					}
+				set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_INV_STAG_LOW_IDX,
+						    ib_wr->ex.invalidate_rkey);
+			}
 
-					set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
-							ib_wr->wr.rdma.rkey);
-					set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
-							ib_wr->wr.rdma.remote_addr);
-
-					if ((ib_wr->send_flags & IB_SEND_INLINE) &&
-							((nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) == 0) &&
-							(ib_wr->sg_list[0].length <= 64)) {
-						memcpy(&wqe->wqe_words[NES_IWARP_SQ_WQE_IMM_DATA_START_IDX],
-							       (void *)(unsigned long)ib_wr->sg_list[0].addr, ib_wr->sg_list[0].length);
-						set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX,
-								ib_wr->sg_list[0].length);
-						wqe_misc |= NES_IWARP_SQ_WQE_IMM_DATA;
-					} else {
-						fill_wqe_sg_send(wqe, ib_wr, 1);
-					}
-					wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX] =
-							wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX];
-					break;
-				case IB_WR_RDMA_READ:
-					/* iWARP only supports 1 sge for RDMA reads */
-					if (ib_wr->num_sge > 1) {
-						nes_debug(NES_DBG_IW_TX, "Exceeded max sge, ib_wr=%u, max=1\n",
-								ib_wr->num_sge);
-						err = -EINVAL;
-						break;
-					}
-					wqe_misc = NES_IWARP_SQ_OP_RDMAR;
-					set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
-							ib_wr->wr.rdma.remote_addr);
-					set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
-							ib_wr->wr.rdma.rkey);
-					set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX,
-							ib_wr->sg_list->length);
-					set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_FRAG0_LOW_IDX,
-							ib_wr->sg_list->addr);
-					set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_STAG0_IDX,
-							ib_wr->sg_list->lkey);
-					break;
-				default:
-					/* error */
-					err = -EINVAL;
-					break;
+			if (ib_wr->num_sge > nesdev->nesadapter->max_sge) {
+				err = -EINVAL;
+				break;
 			}
 
-		if (ib_wr->send_flags & IB_SEND_SIGNALED) {
-			wqe_misc |= NES_IWARP_SQ_WQE_SIGNALED_COMPL;
+			if (ib_wr->send_flags & IB_SEND_FENCE)
+				wqe_misc |= NES_IWARP_SQ_WQE_LOCAL_FENCE;
+
+			if ((ib_wr->send_flags & IB_SEND_INLINE) &&
+			    ((nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) == 0) &&
+			     (ib_wr->sg_list[0].length <= 64)) {
+				memcpy(&wqe->wqe_words[NES_IWARP_SQ_WQE_IMM_DATA_START_IDX],
+				       (void *)(unsigned long)ib_wr->sg_list[0].addr, ib_wr->sg_list[0].length);
+				set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX,
+						    ib_wr->sg_list[0].length);
+				wqe_misc |= NES_IWARP_SQ_WQE_IMM_DATA;
+			} else {
+				fill_wqe_sg_send(wqe, ib_wr, 1);
+			}
+
+			break;
+		case IB_WR_RDMA_WRITE:
+			wqe_misc = NES_IWARP_SQ_OP_RDMAW;
+			if (ib_wr->num_sge > nesdev->nesadapter->max_sge) {
+				nes_debug(NES_DBG_IW_TX, "Exceeded max sge, ib_wr=%u, max=%u\n",
+					  ib_wr->num_sge, nesdev->nesadapter->max_sge);
+				err = -EINVAL;
+				break;
+			}
+
+			if (ib_wr->send_flags & IB_SEND_FENCE)
+				wqe_misc |= NES_IWARP_SQ_WQE_LOCAL_FENCE;
+
+			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
+					    ib_wr->wr.rdma.rkey);
+			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
+					    ib_wr->wr.rdma.remote_addr);
+
+			if ((ib_wr->send_flags & IB_SEND_INLINE) &&
+			    ((nes_drv_opt & NES_DRV_OPT_NO_INLINE_DATA) == 0) &&
+			     (ib_wr->sg_list[0].length <= 64)) {
+				memcpy(&wqe->wqe_words[NES_IWARP_SQ_WQE_IMM_DATA_START_IDX],
+				       (void *)(unsigned long)ib_wr->sg_list[0].addr, ib_wr->sg_list[0].length);
+				set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX,
+						    ib_wr->sg_list[0].length);
+				wqe_misc |= NES_IWARP_SQ_WQE_IMM_DATA;
+			} else {
+				fill_wqe_sg_send(wqe, ib_wr, 1);
+			}
+
+			wqe->wqe_words[NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX] =
+				wqe->wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX];
+			break;
+		case IB_WR_RDMA_READ:
+		case IB_WR_RDMA_READ_WITH_INV:
+			/* iWARP only supports 1 sge for RDMA reads */
+			if (ib_wr->num_sge > 1) {
+				nes_debug(NES_DBG_IW_TX, "Exceeded max sge, ib_wr=%u, max=1\n",
+					  ib_wr->num_sge);
+				err = -EINVAL;
+				break;
+			}
+			if (ib_wr->opcode == IB_WR_RDMA_READ) {
+				wqe_misc = NES_IWARP_SQ_OP_RDMAR;
+			} else {
+				wqe_misc = NES_IWARP_SQ_OP_RDMAR_LOCINV;
+				set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_INV_STAG_LOW_IDX,
+						    ib_wr->ex.invalidate_rkey);
+			}
+
+			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_TO_LOW_IDX,
+					    ib_wr->wr.rdma.remote_addr);
+			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_STAG_IDX,
+					    ib_wr->wr.rdma.rkey);
+			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX,
+					    ib_wr->sg_list->length);
+			set_wqe_64bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_FRAG0_LOW_IDX,
+					    ib_wr->sg_list->addr);
+			set_wqe_32bit_value(wqe->wqe_words, NES_IWARP_SQ_WQE_STAG0_IDX,
+					    ib_wr->sg_list->lkey);
+			break;
+		case IB_WR_LOCAL_INV:
+			wqe_misc = NES_IWARP_SQ_OP_LOCINV;
+			set_wqe_32bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_LOCINV_WQE_INV_STAG_IDX,
+					    ib_wr->ex.invalidate_rkey);
+			break;
+		case IB_WR_FAST_REG_MR:
+		{
+			int i;
+			int flags = ib_wr->wr.fast_reg.access_flags;
+			struct nes_ib_fast_reg_page_list *pnesfrpl =
+				container_of(ib_wr->wr.fast_reg.page_list,
+					     struct nes_ib_fast_reg_page_list,
+					     ibfrpl);
+			u64 *src_page_list = pnesfrpl->ibfrpl.page_list;
+			u64 *dst_page_list = pnesfrpl->nes_wqe_pbl.kva;
+
+			if (ib_wr->wr.fast_reg.page_list_len >
+			    (NES_4K_PBL_CHUNK_SIZE / sizeof(u64))) {
+				nes_debug(NES_DBG_IW_TX, "SQ_FMR: bad page_list_len\n");
+				err = -EINVAL;
+				break;
+			}
+			wqe_misc = NES_IWARP_SQ_OP_FAST_REG;
+			set_wqe_64bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_VA_FBO_LOW_IDX,
+					    ib_wr->wr.fast_reg.iova_start);
+			set_wqe_32bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_LENGTH_LOW_IDX,
+					    ib_wr->wr.fast_reg.length);
+			set_wqe_32bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_LENGTH_HIGH_IDX, 0);
+			set_wqe_32bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_MR_STAG_IDX,
+					    ib_wr->wr.fast_reg.rkey);
+			/* Set page size: */
+			if (ib_wr->wr.fast_reg.page_shift == 12) {
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_4K;
+			} else if (ib_wr->wr.fast_reg.page_shift == 21) {
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_PAGE_SIZE_2M;
+			} else {
+				nes_debug(NES_DBG_IW_TX, "Invalid page shift,"
+					  " ib_wr=%u, max=1\n", ib_wr->num_sge);
+				err = -EINVAL;
+				break;
+			}
+			/* Set access_flags */
+			wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_LOCAL_READ;
+			if (flags & IB_ACCESS_LOCAL_WRITE)
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_LOCAL_WRITE;
+
+			if (flags & IB_ACCESS_REMOTE_WRITE)
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_REMOTE_WRITE;
+
+			if (flags & IB_ACCESS_REMOTE_READ)
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_REMOTE_READ;
+
+			if (flags & IB_ACCESS_MW_BIND)
+				wqe_misc |= NES_IWARP_SQ_FMR_WQE_RIGHTS_ENABLE_WINDOW_BIND;
+
+			/* Fill in PBL info: */
+			if (ib_wr->wr.fast_reg.page_list_len >
+			    pnesfrpl->ibfrpl.max_page_list_len) {
+				nes_debug(NES_DBG_IW_TX, "Invalid page list length,"
+					  " ib_wr=%p, value=%u, max=%u\n",
+					  ib_wr, ib_wr->wr.fast_reg.page_list_len,
+					  pnesfrpl->ibfrpl.max_page_list_len);
+				err = -EINVAL;
+				break;
+			}
+
+			set_wqe_64bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_PBL_ADDR_LOW_IDX,
+					    pnesfrpl->nes_wqe_pbl.paddr);
+
+			set_wqe_32bit_value(wqe->wqe_words,
+					    NES_IWARP_SQ_FMR_WQE_PBL_LENGTH_IDX,
+					    ib_wr->wr.fast_reg.page_list_len * 8);
+
+			for (i = 0; i < ib_wr->wr.fast_reg.page_list_len; i++)
+				dst_page_list[i] = cpu_to_le64(src_page_list[i]);
+
+			nes_debug(NES_DBG_IW_TX, "SQ_FMR: iova_start: %llx, "
+				  "length: %d, rkey: %0x, pgl_paddr: %llx, "
+				  "page_list_len: %u, wqe_misc: %x\n",
+				  (unsigned long long) ib_wr->wr.fast_reg.iova_start,
+				  ib_wr->wr.fast_reg.length,
+				  ib_wr->wr.fast_reg.rkey,
+				  (unsigned long long) pnesfrpl->nes_wqe_pbl.paddr,
+				  ib_wr->wr.fast_reg.page_list_len,
+				  wqe_misc);
+			break;
+		}
+		default:
+			/* error */
+			err = -EINVAL;
+			break;
 		}
+
+		if (err)
+			break;
+
+		if ((ib_wr->send_flags & IB_SEND_SIGNALED) || nesqp->sig_all)
+			wqe_misc |= NES_IWARP_SQ_WQE_SIGNALED_COMPL;
+
 		wqe->wqe_words[NES_IWARP_SQ_WQE_MISC_IDX] = cpu_to_le32(wqe_misc);
 
 		ib_wr = ib_wr->next;
@@ -3355,8 +3525,9 @@ static int nes_post_send(struct ib_qp *ibqp, struct ib_send_wr *ib_wr,
 				(counter << 24) | 0x00800000 | nesqp->hwqp.qp_id);
 	}
 
-		spin_unlock_irqrestore(&nesqp->lock, flags);
+	spin_unlock_irqrestore(&nesqp->lock, flags);
 
+out:
 	if (err)
 		*bad_wr = ib_wr;
 	return err;
@@ -3383,21 +3554,29 @@ static int nes_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr,
 	u32 counter;
 	u32 total_payload_length;
 
-	if (nesqp->ibqp_state > IB_QPS_RTS)
-		return -EINVAL;
+	if (nesqp->ibqp_state > IB_QPS_RTS) {
+		err = -EINVAL;
+		goto out;
+	}
 
-		spin_lock_irqsave(&nesqp->lock, flags);
+	spin_lock_irqsave(&nesqp->lock, flags);
 
 	head = nesqp->hwqp.rq_head;
 
 	while (ib_wr) {
+		/* Check for QP error */
+		if (nesqp->term_flags) {
+			err = -EINVAL;
+			break;
+		}
+
 		if (ib_wr->num_sge > nesdev->nesadapter->max_sge) {
 			err = -EINVAL;
 			break;
 		}
 		/* Check for RQ overflow */
 		if (((head + (2 * qsize) - nesqp->hwqp.rq_tail) % qsize) == (qsize - 1)) {
-			err = -EINVAL;
+			err = -ENOMEM;
 			break;
 		}
 
@@ -3439,8 +3618,9 @@ static int nes_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *ib_wr,
 		nes_write32(nesdev->regs+NES_WQE_ALLOC, (counter<<24) | nesqp->hwqp.qp_id);
 	}
 
-		spin_unlock_irqrestore(&nesqp->lock, flags);
+	spin_unlock_irqrestore(&nesqp->lock, flags);
 
+out:
 	if (err)
 		*bad_wr = ib_wr;
 	return err;
@@ -3454,7 +3634,6 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 {
 	u64 u64temp;
 	u64 wrid;
-	/* u64 u64temp; */
 	unsigned long flags = 0;
 	struct nes_vnic *nesvnic = to_nesvnic(ibcq->device);
 	struct nes_device *nesdev = nesvnic->nesdev;
@@ -3462,44 +3641,56 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 	struct nes_qp *nesqp;
 	struct nes_hw_cqe cqe;
 	u32 head;
-	u32 wq_tail;
+	u32 wq_tail = 0;
 	u32 cq_size;
 	u32 cqe_count = 0;
 	u32 wqe_index;
 	u32 u32temp;
-	/* u32 counter; */
+	u32 move_cq_head = 1;
+	u32 err_code;
 
 	nes_debug(NES_DBG_CQ, "\n");
 
-		spin_lock_irqsave(&nescq->lock, flags);
+	spin_lock_irqsave(&nescq->lock, flags);
 
 	head = nescq->hw_cq.cq_head;
 	cq_size = nescq->hw_cq.cq_size;
 
 	while (cqe_count < num_entries) {
-		if (le32_to_cpu(nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]) &
-				NES_CQE_VALID) {
-			/*
-			 * Make sure we read CQ entry contents *after*
-			 * we've checked the valid bit.
-			 */
-			rmb();
-
-			cqe = nescq->hw_cq.cq_vbase[head];
-			nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX] = 0;
-			u32temp = le32_to_cpu(cqe.cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX]);
-			wqe_index = u32temp &
-					(nesdev->nesadapter->max_qp_wr - 1);
-			u32temp &= ~(NES_SW_CONTEXT_ALIGN-1);
-			/* parse CQE, get completion context from WQE (either rq or sq */
-			u64temp = (((u64)(le32_to_cpu(cqe.cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX])))<<32) |
-					((u64)u32temp);
-			nesqp = *((struct nes_qp **)&u64temp);
+		if ((le32_to_cpu(nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX]) &
+				NES_CQE_VALID) == 0)
+			break;
+
+		/*
+		 * Make sure we read CQ entry contents *after*
+		 * we've checked the valid bit.
+		 */
+		rmb();
+
+		cqe = nescq->hw_cq.cq_vbase[head];
+		u32temp = le32_to_cpu(cqe.cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX]);
+		wqe_index = u32temp & (nesdev->nesadapter->max_qp_wr - 1);
+		u32temp &= ~(NES_SW_CONTEXT_ALIGN-1);
+		/* parse CQE, get completion context from WQE (either rq or sq) */
+		u64temp = (((u64)(le32_to_cpu(cqe.cqe_words[NES_CQE_COMP_COMP_CTX_HIGH_IDX])))<<32) |
+				((u64)u32temp);
+
+		if (u64temp) {
+			nesqp = (struct nes_qp *)(unsigned long)u64temp;
 			memset(entry, 0, sizeof *entry);
 			if (cqe.cqe_words[NES_CQE_ERROR_CODE_IDX] == 0) {
 				entry->status = IB_WC_SUCCESS;
 			} else {
-				entry->status = IB_WC_WR_FLUSH_ERR;
+				err_code = le32_to_cpu(cqe.cqe_words[NES_CQE_ERROR_CODE_IDX]);
+				if (NES_IWARP_CQE_MAJOR_DRV == (err_code >> 16)) {
+					entry->status = err_code & 0x0000ffff;
+
+					/* The rest of the cqe's will be marked as flushed */
+					nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_ERROR_CODE_IDX] =
+						cpu_to_le32((NES_IWARP_CQE_MAJOR_FLUSH << 16) |
+							    NES_IWARP_CQE_MINOR_FLUSH);
+				} else
+					entry->status = IB_WC_WR_FLUSH_ERR;
 			}
 
 			entry->qp = &nesqp->ibqp;
@@ -3508,20 +3699,18 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 			if (le32_to_cpu(cqe.cqe_words[NES_CQE_OPCODE_IDX]) & NES_CQE_SQ) {
 				if (nesqp->skip_lsmm) {
 					nesqp->skip_lsmm = 0;
-					wq_tail = nesqp->hwqp.sq_tail++;
+					nesqp->hwqp.sq_tail++;
 				}
 
 				/* Working on a SQ Completion*/
-				wq_tail = wqe_index;
-				nesqp->hwqp.sq_tail = (wqe_index+1)&(nesqp->hwqp.sq_size - 1);
-				wrid = (((u64)(cpu_to_le32((u32)nesqp->hwqp.sq_vbase[wq_tail].
+				wrid = (((u64)(cpu_to_le32((u32)nesqp->hwqp.sq_vbase[wqe_index].
 						wqe_words[NES_IWARP_SQ_WQE_COMP_SCRATCH_HIGH_IDX]))) << 32) |
-						((u64)(cpu_to_le32((u32)nesqp->hwqp.sq_vbase[wq_tail].
+						((u64)(cpu_to_le32((u32)nesqp->hwqp.sq_vbase[wqe_index].
 						wqe_words[NES_IWARP_SQ_WQE_COMP_SCRATCH_LOW_IDX])));
-				entry->byte_len = le32_to_cpu(nesqp->hwqp.sq_vbase[wq_tail].
+				entry->byte_len = le32_to_cpu(nesqp->hwqp.sq_vbase[wqe_index].
 						wqe_words[NES_IWARP_SQ_WQE_TOTAL_PAYLOAD_IDX]);
 
-				switch (le32_to_cpu(nesqp->hwqp.sq_vbase[wq_tail].
+				switch (le32_to_cpu(nesqp->hwqp.sq_vbase[wqe_index].
 						wqe_words[NES_IWARP_SQ_WQE_MISC_IDX]) & 0x3f) {
 					case NES_IWARP_SQ_OP_RDMAW:
 						nes_debug(NES_DBG_CQ, "Operation = RDMA WRITE.\n");
@@ -3530,7 +3719,7 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 					case NES_IWARP_SQ_OP_RDMAR:
 						nes_debug(NES_DBG_CQ, "Operation = RDMA READ.\n");
 						entry->opcode = IB_WC_RDMA_READ;
-						entry->byte_len = le32_to_cpu(nesqp->hwqp.sq_vbase[wq_tail].
+						entry->byte_len = le32_to_cpu(nesqp->hwqp.sq_vbase[wqe_index].
 								wqe_words[NES_IWARP_SQ_WQE_RDMA_LENGTH_IDX]);
 						break;
 					case NES_IWARP_SQ_OP_SENDINV:
@@ -3540,34 +3729,61 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 						nes_debug(NES_DBG_CQ, "Operation = Send.\n");
 						entry->opcode = IB_WC_SEND;
 						break;
+					case NES_IWARP_SQ_OP_LOCINV:
+						entry->opcode = IB_WC_LOCAL_INV;
+						break;
+					case NES_IWARP_SQ_OP_FAST_REG:
+						entry->opcode = IB_WC_FAST_REG_MR;
+						break;
+				}
+
+				nesqp->hwqp.sq_tail = (wqe_index+1)&(nesqp->hwqp.sq_size - 1);
+				if ((entry->status != IB_WC_SUCCESS) && (nesqp->hwqp.sq_tail != nesqp->hwqp.sq_head)) {
+					move_cq_head = 0;
+					wq_tail = nesqp->hwqp.sq_tail;
 				}
 			} else {
 				/* Working on a RQ Completion*/
-				wq_tail = wqe_index;
-					nesqp->hwqp.rq_tail = (wqe_index+1)&(nesqp->hwqp.rq_size - 1);
 				entry->byte_len = le32_to_cpu(cqe.cqe_words[NES_CQE_PAYLOAD_LENGTH_IDX]);
-				wrid = ((u64)(le32_to_cpu(nesqp->hwqp.rq_vbase[wq_tail].wqe_words[NES_IWARP_RQ_WQE_COMP_SCRATCH_LOW_IDX]))) |
-					((u64)(le32_to_cpu(nesqp->hwqp.rq_vbase[wq_tail].wqe_words[NES_IWARP_RQ_WQE_COMP_SCRATCH_HIGH_IDX]))<<32);
+				wrid = ((u64)(le32_to_cpu(nesqp->hwqp.rq_vbase[wqe_index].wqe_words[NES_IWARP_RQ_WQE_COMP_SCRATCH_LOW_IDX]))) |
+					((u64)(le32_to_cpu(nesqp->hwqp.rq_vbase[wqe_index].wqe_words[NES_IWARP_RQ_WQE_COMP_SCRATCH_HIGH_IDX]))<<32);
 					entry->opcode = IB_WC_RECV;
+
+				nesqp->hwqp.rq_tail = (wqe_index+1)&(nesqp->hwqp.rq_size - 1);
+				if ((entry->status != IB_WC_SUCCESS) && (nesqp->hwqp.rq_tail != nesqp->hwqp.rq_head)) {
+					move_cq_head = 0;
+					wq_tail = nesqp->hwqp.rq_tail;
+				}
 			}
+
 			entry->wr_id = wrid;
+			entry++;
+			cqe_count++;
+		}
 
+		if (move_cq_head) {
+			nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_OPCODE_IDX] = 0;
 			if (++head >= cq_size)
 				head = 0;
-			cqe_count++;
 			nescq->polled_completions++;
+
 			if ((nescq->polled_completions > (cq_size / 2)) ||
 					(nescq->polled_completions == 255)) {
 				nes_debug(NES_DBG_CQ, "CQ%u Issuing CQE Allocate since more than half of cqes"
-						" are pending %u of %u.\n",
-						nescq->hw_cq.cq_number, nescq->polled_completions, cq_size);
+					" are pending %u of %u.\n",
+					nescq->hw_cq.cq_number, nescq->polled_completions, cq_size);
 				nes_write32(nesdev->regs+NES_CQE_ALLOC,
-						nescq->hw_cq.cq_number | (nescq->polled_completions << 16));
+					nescq->hw_cq.cq_number | (nescq->polled_completions << 16));
 				nescq->polled_completions = 0;
 			}
-			entry++;
-		} else
-			break;
+		} else {
+			/* Update the wqe index and set status to flush */
+			wqe_index = le32_to_cpu(cqe.cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX]);
+			wqe_index = (wqe_index & (~(nesdev->nesadapter->max_qp_wr - 1))) | wq_tail;
+			nescq->hw_cq.cq_vbase[head].cqe_words[NES_CQE_COMP_COMP_CTX_LOW_IDX] =
+				cpu_to_le32(wqe_index);
+			move_cq_head = 1; /* ready for next pass */
+		}
 	}
 
 	if (nescq->polled_completions) {
@@ -3580,7 +3796,7 @@ static int nes_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
 	nes_debug(NES_DBG_CQ, "Reporting %u completions for CQ%u.\n",
 			cqe_count, nescq->hw_cq.cq_number);
 
-		spin_unlock_irqrestore(&nescq->lock, flags);
+	spin_unlock_irqrestore(&nescq->lock, flags);
 
 	return cqe_count;
 }
@@ -3664,7 +3880,6 @@ struct nes_ib_device *nes_init_ofa_device(struct net_device *netdev)
 	nesibdev->ibdev.dev.parent = &nesdev->pcidev->dev;
 	nesibdev->ibdev.query_device = nes_query_device;
 	nesibdev->ibdev.query_port = nes_query_port;
-	nesibdev->ibdev.modify_port = nes_modify_port;
 	nesibdev->ibdev.query_pkey = nes_query_pkey;
 	nesibdev->ibdev.query_gid = nes_query_gid;
 	nesibdev->ibdev.alloc_ucontext = nes_alloc_ucontext;
@@ -3689,10 +3904,9 @@ struct nes_ib_device *nes_init_ofa_device(struct net_device *netdev)
 	nesibdev->ibdev.dealloc_mw = nes_dealloc_mw;
 	nesibdev->ibdev.bind_mw = nes_bind_mw;
 
-	nesibdev->ibdev.alloc_fmr = nes_alloc_fmr;
-	nesibdev->ibdev.unmap_fmr = nes_unmap_fmr;
-	nesibdev->ibdev.dealloc_fmr = nes_dealloc_fmr;
-	nesibdev->ibdev.map_phys_fmr = nes_map_phys_fmr;
+	nesibdev->ibdev.alloc_fast_reg_mr = nes_alloc_fast_reg_mr;
+	nesibdev->ibdev.alloc_fast_reg_page_list = nes_alloc_fast_reg_page_list;
+	nesibdev->ibdev.free_fast_reg_page_list = nes_free_fast_reg_page_list;
 
 	nesibdev->ibdev.attach_mcast = nes_multicast_attach;
 	nesibdev->ibdev.detach_mcast = nes_multicast_detach;
@@ -3721,6 +3935,52 @@ struct nes_ib_device *nes_init_ofa_device(struct net_device *netdev)
 
 
 /**
+ * nes_handle_delayed_event
+ */
+static void nes_handle_delayed_event(unsigned long data)
+{
+	struct nes_vnic *nesvnic = (void *) data;
+
+	if (nesvnic->delayed_event != nesvnic->last_dispatched_event) {
+		struct ib_event event;
+
+		event.device = &nesvnic->nesibdev->ibdev;
+		if (!event.device)
+			goto stop_timer;
+		event.event = nesvnic->delayed_event;
+		event.element.port_num = nesvnic->logical_port + 1;
+		ib_dispatch_event(&event);
+	}
+
+stop_timer:
+	nesvnic->event_timer.function = NULL;
+}
+
+
+void  nes_port_ibevent(struct nes_vnic *nesvnic)
+{
+	struct nes_ib_device *nesibdev = nesvnic->nesibdev;
+	struct nes_device *nesdev = nesvnic->nesdev;
+	struct ib_event event;
+	event.device = &nesibdev->ibdev;
+	event.element.port_num = nesvnic->logical_port + 1;
+	event.event = nesdev->iw_status ? IB_EVENT_PORT_ACTIVE : IB_EVENT_PORT_ERR;
+
+	if (!nesvnic->event_timer.function) {
+		ib_dispatch_event(&event);
+		nesvnic->last_dispatched_event = event.event;
+		nesvnic->event_timer.function = nes_handle_delayed_event;
+		nesvnic->event_timer.data = (unsigned long) nesvnic;
+		nesvnic->event_timer.expires = jiffies + NES_EVENT_DELAY;
+		add_timer(&nesvnic->event_timer);
+	} else {
+		mod_timer(&nesvnic->event_timer, jiffies + NES_EVENT_DELAY);
+	}
+	nesvnic->delayed_event = event.event;
+}
+
+
+/**
  * nes_destroy_ofa_device
  */
 void nes_destroy_ofa_device(struct nes_ib_device *nesibdev)
@@ -3745,7 +4005,7 @@ int nes_register_ofa_device(struct nes_ib_device *nesibdev)
 	struct nes_adapter *nesadapter = nesdev->nesadapter;
 	int i, ret;
 
-	ret = ib_register_device(&nesvnic->nesibdev->ibdev);
+	ret = ib_register_device(&nesvnic->nesibdev->ibdev, NULL);
 	if (ret) {
 		return ret;
 	}
diff --git a/drivers/infiniband/hw/nes/nes_verbs.h b/drivers/infiniband/hw/nes/nes_verbs.h
index 6c6b4da5184..309b31c31ae 100644
--- a/drivers/infiniband/hw/nes/nes_verbs.h
+++ b/drivers/infiniband/hw/nes/nes_verbs.h
@@ -1,5 +1,5 @@
 /*
- * Copyright (c) 2006 - 2008 NetEffect, Inc. All rights reserved.
+ * Copyright (c) 2006 - 2011 Intel Corporation.  All rights reserved.
  * Copyright (c) 2005 Open Grid Computing, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -40,6 +40,10 @@ struct nes_device;
 #define NES_MAX_USER_DB_REGIONS  4096
 #define NES_MAX_USER_WQ_REGIONS  4096
 
+#define NES_TERM_SENT            0x01
+#define NES_TERM_RCVD            0x02
+#define NES_TERM_DONE            0x04
+
 struct nes_ucontext {
 	struct ib_ucontext ibucontext;
 	struct nes_device  *nesdev;
@@ -112,12 +116,18 @@ struct nes_cq {
 	spinlock_t       lock;
 	u8               virtual_cq;
 	u8               pad[3];
+	u32		 mcrqf;
 };
 
 struct nes_wq {
 	spinlock_t lock;
 };
 
+struct disconn_work {
+	struct work_struct    work;
+	struct nes_qp         *nesqp;
+};
+
 struct iw_cm_id;
 struct ietf_mpa_frame;
 
@@ -125,19 +135,16 @@ struct nes_qp {
 	struct ib_qp          ibqp;
 	void                  *allocated_buffer;
 	struct iw_cm_id       *cm_id;
-	struct workqueue_struct *wq;
-	struct work_struct    disconn_work;
 	struct nes_cq         *nesscq;
 	struct nes_cq         *nesrcq;
 	struct nes_pd         *nespd;
 	void *cm_node; /* handle of the node this QP is associated with */
-	struct ietf_mpa_frame *ietf_frame;
+	void                  *ietf_frame;
+	u8                    ietf_frame_size;
 	dma_addr_t            ietf_frame_pbase;
-	wait_queue_head_t     state_waitq;
-	unsigned long         socket;
+	struct ib_mr          *lsmm_mr;
 	struct nes_hw_qp      hwqp;
 	struct work_struct    work;
-	struct work_struct    ae_work;
 	enum ib_qp_state      ibqp_state;
 	u32                   iwarp_state;
 	u32                   hte_index;
@@ -148,22 +155,35 @@ struct nes_qp {
 	u32                   mmap_sq_db_index;
 	u32                   mmap_rq_db_index;
 	spinlock_t            lock;
+	spinlock_t            pau_lock;
 	struct nes_qp_context *nesqp_context;
 	dma_addr_t            nesqp_context_pbase;
 	void	              *pbl_vbase;
 	dma_addr_t            pbl_pbase;
 	struct page           *page;
-	wait_queue_head_t     kick_waitq;
-	u16                   in_disconnect;
+	struct timer_list     terminate_timer;
+	enum ib_event_type    terminate_eventtype;
+	struct sk_buff_head   pau_list;
+	u32                   pau_rcv_nxt;
+	u16                   active_conn:1;
+	u16                   skip_lsmm:1;
+	u16                   user_mode:1;
+	u16                   hte_added:1;
+	u16                   flush_issued:1;
+	u16                   destroyed:1;
+	u16                   sig_all:1;
+	u16                   pau_mode:1;
+	u16                   rsvd:8;
 	u16                   private_data_len;
-	u8                    active_conn;
-	u8                    skip_lsmm;
-	u8                    user_mode;
-	u8                    hte_added;
+	u16                   term_sq_flush_code;
+	u16                   term_rq_flush_code;
 	u8                    hw_iwarp_state;
-	u8                    flush_issued;
 	u8                    hw_tcp_state;
-	u8                    disconn_pending;
-	u8                    destroyed;
+	u8                    term_flags;
+	u8                    sq_kmapped;
+	u8                    pau_busy;
+	u8                    pau_pending;
+	u8                    pau_state;
+	__u64                 nesuqp_addr;
 };
 #endif			/* NES_VERBS_H */
diff --git a/drivers/infiniband/hw/ocrdma/Kconfig b/drivers/infiniband/hw/ocrdma/Kconfig
new file mode 100644
index 00000000000..c0cddc0192d
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/Kconfig
@@ -0,0 +1,8 @@
+config INFINIBAND_OCRDMA
+	tristate "Emulex One Connect HCA support"
+	depends on ETHERNET && NETDEVICES && PCI && INET && (IPV6 || IPV6=n)
+	select NET_VENDOR_EMULEX
+	select BE2NET
+	---help---
+	  This driver provides low-level InfiniBand over Ethernet
+	  support for Emulex One Connect host channel adapters (HCAs).
diff --git a/drivers/infiniband/hw/ocrdma/Makefile b/drivers/infiniband/hw/ocrdma/Makefile
new file mode 100644
index 00000000000..d1bfd4f4cdd
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/Makefile
@@ -0,0 +1,5 @@
+ccflags-y := -Idrivers/net/ethernet/emulex/benet
+
+obj-$(CONFIG_INFINIBAND_OCRDMA)	+= ocrdma.o
+
+ocrdma-y :=	ocrdma_main.o ocrdma_verbs.o ocrdma_hw.o ocrdma_ah.o ocrdma_stats.o
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma.h b/drivers/infiniband/hw/ocrdma/ocrdma.h
new file mode 100644
index 00000000000..19011dbb930
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma.h
@@ -0,0 +1,521 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_H__
+#define __OCRDMA_H__
+
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include <be_roce.h>
+#include "ocrdma_sli.h"
+
+#define OCRDMA_ROCE_DRV_VERSION "10.2.145.0u"
+
+#define OCRDMA_ROCE_DRV_DESC "Emulex OneConnect RoCE Driver"
+#define OCRDMA_NODE_DESC "Emulex OneConnect RoCE HCA"
+
+#define OC_NAME_SH	OCRDMA_NODE_DESC "(Skyhawk)"
+#define OC_NAME_UNKNOWN OCRDMA_NODE_DESC "(Unknown)"
+
+#define OC_SKH_DEVICE_PF 0x720
+#define OC_SKH_DEVICE_VF 0x728
+#define OCRDMA_MAX_AH 512
+
+#define OCRDMA_UVERBS(CMD_NAME) (1ull << IB_USER_VERBS_CMD_##CMD_NAME)
+
+#define convert_to_64bit(lo, hi) ((u64)hi << 32 | (u64)lo)
+
+struct ocrdma_dev_attr {
+	u8 fw_ver[32];
+	u32 vendor_id;
+	u32 device_id;
+	u16 max_pd;
+	u16 max_cq;
+	u16 max_cqe;
+	u16 max_qp;
+	u16 max_wqe;
+	u16 max_rqe;
+	u16 max_srq;
+	u32 max_inline_data;
+	int max_send_sge;
+	int max_recv_sge;
+	int max_srq_sge;
+	int max_rdma_sge;
+	int max_mr;
+	u64 max_mr_size;
+	u32 max_num_mr_pbl;
+	int max_mw;
+	int max_fmr;
+	int max_map_per_fmr;
+	int max_pages_per_frmr;
+	u16 max_ord_per_qp;
+	u16 max_ird_per_qp;
+
+	int device_cap_flags;
+	u8 cq_overflow_detect;
+	u8 srq_supported;
+
+	u32 wqe_size;
+	u32 rqe_size;
+	u32 ird_page_size;
+	u8 local_ca_ack_delay;
+	u8 ird;
+	u8 num_ird_pages;
+};
+
+struct ocrdma_dma_mem {
+	void *va;
+	dma_addr_t pa;
+	u32 size;
+};
+
+struct ocrdma_pbl {
+	void *va;
+	dma_addr_t pa;
+};
+
+struct ocrdma_queue_info {
+	void *va;
+	dma_addr_t dma;
+	u32 size;
+	u16 len;
+	u16 entry_size;		/* Size of an element in the queue */
+	u16 id;			/* qid, where to ring the doorbell. */
+	u16 head, tail;
+	bool created;
+};
+
+struct ocrdma_eq {
+	struct ocrdma_queue_info q;
+	u32 vector;
+	int cq_cnt;
+	struct ocrdma_dev *dev;
+	char irq_name[32];
+};
+
+struct ocrdma_mq {
+	struct ocrdma_queue_info sq;
+	struct ocrdma_queue_info cq;
+	bool rearm_cq;
+};
+
+struct mqe_ctx {
+	struct mutex lock; /* for serializing mailbox commands on MQ */
+	wait_queue_head_t cmd_wait;
+	u32 tag;
+	u16 cqe_status;
+	u16 ext_status;
+	bool cmd_done;
+};
+
+struct ocrdma_hw_mr {
+	u32 lkey;
+	u8 fr_mr;
+	u8 remote_atomic;
+	u8 remote_rd;
+	u8 remote_wr;
+	u8 local_rd;
+	u8 local_wr;
+	u8 mw_bind;
+	u8 rsvd;
+	u64 len;
+	struct ocrdma_pbl *pbl_table;
+	u32 num_pbls;
+	u32 num_pbes;
+	u32 pbl_size;
+	u32 pbe_size;
+	u64 fbo;
+	u64 va;
+};
+
+struct ocrdma_mr {
+	struct ib_mr ibmr;
+	struct ib_umem *umem;
+	struct ocrdma_hw_mr hwmr;
+};
+
+struct ocrdma_stats {
+	u8 type;
+	struct ocrdma_dev *dev;
+};
+
+struct stats_mem {
+	struct ocrdma_mqe mqe;
+	void *va;
+	dma_addr_t pa;
+	u32 size;
+	char *debugfs_mem;
+};
+
+struct phy_info {
+	u16 auto_speeds_supported;
+	u16 fixed_speeds_supported;
+	u16 phy_type;
+	u16 interface_type;
+};
+
+struct ocrdma_dev {
+	struct ib_device ibdev;
+	struct ocrdma_dev_attr attr;
+
+	struct mutex dev_lock; /* provides syncronise access to device data */
+	spinlock_t flush_q_lock ____cacheline_aligned;
+
+	struct ocrdma_cq **cq_tbl;
+	struct ocrdma_qp **qp_tbl;
+
+	struct ocrdma_eq *eq_tbl;
+	int eq_cnt;
+	u16 base_eqid;
+	u16 max_eq;
+
+	union ib_gid *sgid_tbl;
+	/* provided synchronization to sgid table for
+	 * updating gid entries triggered by notifier.
+	 */
+	spinlock_t sgid_lock;
+
+	int gsi_qp_created;
+	struct ocrdma_cq *gsi_sqcq;
+	struct ocrdma_cq *gsi_rqcq;
+
+	struct {
+		struct ocrdma_av *va;
+		dma_addr_t pa;
+		u32 size;
+		u32 num_ah;
+		/* provide synchronization for av
+		 * entry allocations.
+		 */
+		spinlock_t lock;
+		u32 ahid;
+		struct ocrdma_pbl pbl;
+	} av_tbl;
+
+	void *mbx_cmd;
+	struct ocrdma_mq mq;
+	struct mqe_ctx mqe_ctx;
+
+	struct be_dev_info nic_info;
+	struct phy_info phy;
+	char model_number[32];
+	u32 hba_port_num;
+
+	struct list_head entry;
+	struct rcu_head rcu;
+	int id;
+	u64 stag_arr[OCRDMA_MAX_STAG];
+	u16 pvid;
+	u32 asic_id;
+
+	ulong last_stats_time;
+	struct mutex stats_lock; /* provide synch for debugfs operations */
+	struct stats_mem stats_mem;
+	struct ocrdma_stats rsrc_stats;
+	struct ocrdma_stats rx_stats;
+	struct ocrdma_stats wqe_stats;
+	struct ocrdma_stats tx_stats;
+	struct ocrdma_stats db_err_stats;
+	struct ocrdma_stats tx_qp_err_stats;
+	struct ocrdma_stats rx_qp_err_stats;
+	struct ocrdma_stats tx_dbg_stats;
+	struct ocrdma_stats rx_dbg_stats;
+	struct dentry *dir;
+};
+
+struct ocrdma_cq {
+	struct ib_cq ibcq;
+	struct ocrdma_cqe *va;
+	u32 phase;
+	u32 getp;	/* pointer to pending wrs to
+			 * return to stack, wrap arounds
+			 * at max_hw_cqe
+			 */
+	u32 max_hw_cqe;
+	bool phase_change;
+	bool deferred_arm, deferred_sol;
+	bool first_arm;
+
+	spinlock_t cq_lock ____cacheline_aligned; /* provide synchronization
+						   * to cq polling
+						   */
+	/* syncronizes cq completion handler invoked from multiple context */
+	spinlock_t comp_handler_lock ____cacheline_aligned;
+	u16 id;
+	u16 eqn;
+
+	struct ocrdma_ucontext *ucontext;
+	dma_addr_t pa;
+	u32 len;
+	u32 cqe_cnt;
+
+	/* head of all qp's sq and rq for which cqes need to be flushed
+	 * by the software.
+	 */
+	struct list_head sq_head, rq_head;
+};
+
+struct ocrdma_pd {
+	struct ib_pd ibpd;
+	struct ocrdma_ucontext *uctx;
+	u32 id;
+	int num_dpp_qp;
+	u32 dpp_page;
+	bool dpp_enabled;
+};
+
+struct ocrdma_ah {
+	struct ib_ah ibah;
+	struct ocrdma_av *av;
+	u16 sgid_index;
+	u32 id;
+};
+
+struct ocrdma_qp_hwq_info {
+	u8 *va;			/* virtual address */
+	u32 max_sges;
+	u32 head, tail;
+	u32 entry_size;
+	u32 max_cnt;
+	u32 max_wqe_idx;
+	u16 dbid;		/* qid, where to ring the doorbell. */
+	u32 len;
+	dma_addr_t pa;
+};
+
+struct ocrdma_srq {
+	struct ib_srq ibsrq;
+	u8 __iomem *db;
+	struct ocrdma_qp_hwq_info rq;
+	u64 *rqe_wr_id_tbl;
+	u32 *idx_bit_fields;
+	u32 bit_fields_len;
+
+	/* provide synchronization to multiple context(s) posting rqe */
+	spinlock_t q_lock ____cacheline_aligned;
+
+	struct ocrdma_pd *pd;
+	u32 id;
+};
+
+struct ocrdma_qp {
+	struct ib_qp ibqp;
+	struct ocrdma_dev *dev;
+
+	u8 __iomem *sq_db;
+	struct ocrdma_qp_hwq_info sq;
+	struct {
+		uint64_t wrid;
+		uint16_t dpp_wqe_idx;
+		uint16_t dpp_wqe;
+		uint8_t  signaled;
+		uint8_t  rsvd[3];
+	} *wqe_wr_id_tbl;
+	u32 max_inline_data;
+
+	/* provide synchronization to multiple context(s) posting wqe, rqe */
+	spinlock_t q_lock ____cacheline_aligned;
+	struct ocrdma_cq *sq_cq;
+	/* list maintained per CQ to flush SQ errors */
+	struct list_head sq_entry;
+
+	u8 __iomem *rq_db;
+	struct ocrdma_qp_hwq_info rq;
+	u64 *rqe_wr_id_tbl;
+	struct ocrdma_cq *rq_cq;
+	struct ocrdma_srq *srq;
+	/* list maintained per CQ to flush RQ errors */
+	struct list_head rq_entry;
+
+	enum ocrdma_qp_state state;	/*  QP state */
+	int cap_flags;
+	u32 max_ord, max_ird;
+
+	u32 id;
+	struct ocrdma_pd *pd;
+
+	enum ib_qp_type qp_type;
+
+	int sgid_idx;
+	u32 qkey;
+	bool dpp_enabled;
+	u8 *ird_q_va;
+	bool signaled;
+};
+
+struct ocrdma_ucontext {
+	struct ib_ucontext ibucontext;
+
+	struct list_head mm_head;
+	struct mutex mm_list_lock; /* protects list entries of mm type */
+	struct ocrdma_pd *cntxt_pd;
+	int pd_in_use;
+
+	struct {
+		u32 *va;
+		dma_addr_t pa;
+		u32 len;
+	} ah_tbl;
+};
+
+struct ocrdma_mm {
+	struct {
+		u64 phy_addr;
+		unsigned long len;
+	} key;
+	struct list_head entry;
+};
+
+static inline struct ocrdma_dev *get_ocrdma_dev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct ocrdma_dev, ibdev);
+}
+
+static inline struct ocrdma_ucontext *get_ocrdma_ucontext(struct ib_ucontext
+							  *ibucontext)
+{
+	return container_of(ibucontext, struct ocrdma_ucontext, ibucontext);
+}
+
+static inline struct ocrdma_pd *get_ocrdma_pd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct ocrdma_pd, ibpd);
+}
+
+static inline struct ocrdma_cq *get_ocrdma_cq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct ocrdma_cq, ibcq);
+}
+
+static inline struct ocrdma_qp *get_ocrdma_qp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct ocrdma_qp, ibqp);
+}
+
+static inline struct ocrdma_mr *get_ocrdma_mr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct ocrdma_mr, ibmr);
+}
+
+static inline struct ocrdma_ah *get_ocrdma_ah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct ocrdma_ah, ibah);
+}
+
+static inline struct ocrdma_srq *get_ocrdma_srq(struct ib_srq *ibsrq)
+{
+	return container_of(ibsrq, struct ocrdma_srq, ibsrq);
+}
+
+static inline int is_cqe_valid(struct ocrdma_cq *cq, struct ocrdma_cqe *cqe)
+{
+	int cqe_valid;
+	cqe_valid = le32_to_cpu(cqe->flags_status_srcqpn) & OCRDMA_CQE_VALID;
+	return (cqe_valid == cq->phase);
+}
+
+static inline int is_cqe_for_sq(struct ocrdma_cqe *cqe)
+{
+	return (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_QTYPE) ? 0 : 1;
+}
+
+static inline int is_cqe_invalidated(struct ocrdma_cqe *cqe)
+{
+	return (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_INVALIDATE) ? 1 : 0;
+}
+
+static inline int is_cqe_imm(struct ocrdma_cqe *cqe)
+{
+	return (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_IMM) ? 1 : 0;
+}
+
+static inline int is_cqe_wr_imm(struct ocrdma_cqe *cqe)
+{
+	return (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_WRITE_IMM) ? 1 : 0;
+}
+
+static inline int ocrdma_resolve_dmac(struct ocrdma_dev *dev,
+		struct ib_ah_attr *ah_attr, u8 *mac_addr)
+{
+	struct in6_addr in6;
+
+	memcpy(&in6, ah_attr->grh.dgid.raw, sizeof(in6));
+	if (rdma_is_multicast_addr(&in6))
+		rdma_get_mcast_mac(&in6, mac_addr);
+	else
+		memcpy(mac_addr, ah_attr->dmac, ETH_ALEN);
+	return 0;
+}
+
+static inline char *hca_name(struct ocrdma_dev *dev)
+{
+	switch (dev->nic_info.pdev->device) {
+	case OC_SKH_DEVICE_PF:
+	case OC_SKH_DEVICE_VF:
+		return OC_NAME_SH;
+	default:
+		return OC_NAME_UNKNOWN;
+	}
+}
+
+static inline int ocrdma_get_eq_table_index(struct ocrdma_dev *dev,
+		int eqid)
+{
+	int indx;
+
+	for (indx = 0; indx < dev->eq_cnt; indx++) {
+		if (dev->eq_tbl[indx].q.id == eqid)
+			return indx;
+	}
+
+	return -EINVAL;
+}
+
+static inline u8 ocrdma_get_asic_type(struct ocrdma_dev *dev)
+{
+	if (dev->nic_info.dev_family == 0xF && !dev->asic_id) {
+		pci_read_config_dword(
+			dev->nic_info.pdev,
+			OCRDMA_SLI_ASIC_ID_OFFSET, &dev->asic_id);
+	}
+
+	return (dev->asic_id & OCRDMA_SLI_ASIC_GEN_NUM_MASK) >>
+				OCRDMA_SLI_ASIC_GEN_NUM_SHIFT;
+}
+
+#endif
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_abi.h b/drivers/infiniband/hw/ocrdma/ocrdma_abi.h
new file mode 100644
index 00000000000..1554cca5712
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_abi.h
@@ -0,0 +1,134 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_ABI_H__
+#define __OCRDMA_ABI_H__
+
+#define OCRDMA_ABI_VERSION 2
+#define OCRDMA_BE_ROCE_ABI_VERSION 1
+/* user kernel communication data structures. */
+
+struct ocrdma_alloc_ucontext_resp {
+	u32 dev_id;
+	u32 wqe_size;
+	u32 max_inline_data;
+	u32 dpp_wqe_size;
+	u64 ah_tbl_page;
+	u32 ah_tbl_len;
+	u32 rqe_size;
+	u8 fw_ver[32];
+	/* for future use/new features in progress */
+	u64 rsvd1;
+	u64 rsvd2;
+};
+
+struct ocrdma_alloc_pd_ureq {
+	u64 rsvd1;
+};
+
+struct ocrdma_alloc_pd_uresp {
+	u32 id;
+	u32 dpp_enabled;
+	u32 dpp_page_addr_hi;
+	u32 dpp_page_addr_lo;
+	u64 rsvd1;
+};
+
+struct ocrdma_create_cq_ureq {
+	u32 dpp_cq;
+	u32 rsvd; /* pad */
+};
+
+#define MAX_CQ_PAGES 8
+struct ocrdma_create_cq_uresp {
+	u32 cq_id;
+	u32 page_size;
+	u32 num_pages;
+	u32 max_hw_cqe;
+	u64 page_addr[MAX_CQ_PAGES];
+	u64 db_page_addr;
+	u32 db_page_size;
+	u32 phase_change;
+	/* for future use/new features in progress */
+	u64 rsvd1;
+	u64 rsvd2;
+};
+
+#define MAX_QP_PAGES 8
+#define MAX_UD_AV_PAGES 8
+
+struct ocrdma_create_qp_ureq {
+	u8 enable_dpp_cq;
+	u8 rsvd;
+	u16 dpp_cq_id;
+	u32 rsvd1;	/* pad */
+};
+
+struct ocrdma_create_qp_uresp {
+	u16 qp_id;
+	u16 sq_dbid;
+	u16 rq_dbid;
+	u16 resv0;	/* pad */
+	u32 sq_page_size;
+	u32 rq_page_size;
+	u32 num_sq_pages;
+	u32 num_rq_pages;
+	u64 sq_page_addr[MAX_QP_PAGES];
+	u64 rq_page_addr[MAX_QP_PAGES];
+	u64 db_page_addr;
+	u32 db_page_size;
+	u32 dpp_credit;
+	u32 dpp_offset;
+	u32 num_wqe_allocated;
+	u32 num_rqe_allocated;
+	u32 db_sq_offset;
+	u32 db_rq_offset;
+	u32 db_shift;
+	u64 rsvd[11];
+} __packed;
+
+struct ocrdma_create_srq_uresp {
+	u16 rq_dbid;
+	u16 resv0;	/* pad */
+	u32 resv1;
+
+	u32 rq_page_size;
+	u32 num_rq_pages;
+
+	u64 rq_page_addr[MAX_QP_PAGES];
+	u64 db_page_addr;
+
+	u32 db_page_size;
+	u32 num_rqe_allocated;
+	u32 db_rq_offset;
+	u32 db_shift;
+
+	u64 rsvd2;
+	u64 rsvd3;
+};
+
+#endif				/* __OCRDMA_ABI_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.c b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
new file mode 100644
index 00000000000..d4cc01f10c0
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.c
@@ -0,0 +1,175 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#include <net/neighbour.h>
+#include <net/netevent.h>
+
+#include <rdma/ib_addr.h>
+
+#include "ocrdma.h"
+#include "ocrdma_verbs.h"
+#include "ocrdma_ah.h"
+#include "ocrdma_hw.h"
+
+static inline int set_av_attr(struct ocrdma_dev *dev, struct ocrdma_ah *ah,
+				struct ib_ah_attr *attr, int pdid)
+{
+	int status = 0;
+	u16 vlan_tag; bool vlan_enabled = false;
+	struct ocrdma_eth_vlan eth;
+	struct ocrdma_grh grh;
+	int eth_sz;
+
+	memset(&eth, 0, sizeof(eth));
+	memset(&grh, 0, sizeof(grh));
+
+	ah->sgid_index = attr->grh.sgid_index;
+
+	vlan_tag = attr->vlan_id;
+	if (!vlan_tag || (vlan_tag > 0xFFF))
+		vlan_tag = dev->pvid;
+	if (vlan_tag && (vlan_tag < 0x1000)) {
+		eth.eth_type = cpu_to_be16(0x8100);
+		eth.roce_eth_type = cpu_to_be16(OCRDMA_ROCE_ETH_TYPE);
+		vlan_tag |= (attr->sl & 7) << 13;
+		eth.vlan_tag = cpu_to_be16(vlan_tag);
+		eth_sz = sizeof(struct ocrdma_eth_vlan);
+		vlan_enabled = true;
+	} else {
+		eth.eth_type = cpu_to_be16(OCRDMA_ROCE_ETH_TYPE);
+		eth_sz = sizeof(struct ocrdma_eth_basic);
+	}
+	memcpy(&eth.smac[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
+	memcpy(&eth.dmac[0], attr->dmac, ETH_ALEN);
+	status = ocrdma_resolve_dmac(dev, attr, &eth.dmac[0]);
+	if (status)
+		return status;
+	status = ocrdma_query_gid(&dev->ibdev, 1, attr->grh.sgid_index,
+			(union ib_gid *)&grh.sgid[0]);
+	if (status)
+		return status;
+
+	grh.tclass_flow = cpu_to_be32((6 << 28) |
+			(attr->grh.traffic_class << 24) |
+			attr->grh.flow_label);
+	/* 0x1b is next header value in GRH */
+	grh.pdid_hoplimit = cpu_to_be32((pdid << 16) |
+			(0x1b << 8) | attr->grh.hop_limit);
+
+	memcpy(&grh.dgid[0], attr->grh.dgid.raw, sizeof(attr->grh.dgid.raw));
+	memcpy(&ah->av->eth_hdr, &eth, eth_sz);
+	memcpy((u8 *)ah->av + eth_sz, &grh, sizeof(struct ocrdma_grh));
+	if (vlan_enabled)
+		ah->av->valid |= OCRDMA_AV_VLAN_VALID;
+	ah->av->valid = cpu_to_le32(ah->av->valid);
+	return status;
+}
+
+struct ib_ah *ocrdma_create_ah(struct ib_pd *ibpd, struct ib_ah_attr *attr)
+{
+	u32 *ahid_addr;
+	int status;
+	struct ocrdma_ah *ah;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+
+	if (!(attr->ah_flags & IB_AH_GRH))
+		return ERR_PTR(-EINVAL);
+
+	ah = kzalloc(sizeof(*ah), GFP_ATOMIC);
+	if (!ah)
+		return ERR_PTR(-ENOMEM);
+
+	status = ocrdma_alloc_av(dev, ah);
+	if (status)
+		goto av_err;
+	status = set_av_attr(dev, ah, attr, pd->id);
+	if (status)
+		goto av_conf_err;
+
+	/* if pd is for the user process, pass the ah_id to user space */
+	if ((pd->uctx) && (pd->uctx->ah_tbl.va)) {
+		ahid_addr = pd->uctx->ah_tbl.va + attr->dlid;
+		*ahid_addr = ah->id;
+	}
+	return &ah->ibah;
+
+av_conf_err:
+	ocrdma_free_av(dev, ah);
+av_err:
+	kfree(ah);
+	return ERR_PTR(status);
+}
+
+int ocrdma_destroy_ah(struct ib_ah *ibah)
+{
+	struct ocrdma_ah *ah = get_ocrdma_ah(ibah);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibah->device);
+
+	ocrdma_free_av(dev, ah);
+	kfree(ah);
+	return 0;
+}
+
+int ocrdma_query_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
+{
+	struct ocrdma_ah *ah = get_ocrdma_ah(ibah);
+	struct ocrdma_av *av = ah->av;
+	struct ocrdma_grh *grh;
+	attr->ah_flags |= IB_AH_GRH;
+	if (ah->av->valid & Bit(1)) {
+		grh = (struct ocrdma_grh *)((u8 *)ah->av +
+				sizeof(struct ocrdma_eth_vlan));
+		attr->sl = be16_to_cpu(av->eth_hdr.vlan_tag) >> 13;
+	} else {
+		grh = (struct ocrdma_grh *)((u8 *)ah->av +
+					sizeof(struct ocrdma_eth_basic));
+		attr->sl = 0;
+	}
+	memcpy(&attr->grh.dgid.raw[0], &grh->dgid[0], sizeof(grh->dgid));
+	attr->grh.sgid_index = ah->sgid_index;
+	attr->grh.hop_limit = be32_to_cpu(grh->pdid_hoplimit) & 0xff;
+	attr->grh.traffic_class = be32_to_cpu(grh->tclass_flow) >> 24;
+	attr->grh.flow_label = be32_to_cpu(grh->tclass_flow) & 0x00ffffffff;
+	return 0;
+}
+
+int ocrdma_modify_ah(struct ib_ah *ibah, struct ib_ah_attr *attr)
+{
+	/* modify_ah is unsupported */
+	return -ENOSYS;
+}
+
+int ocrdma_process_mad(struct ib_device *ibdev,
+		       int process_mad_flags,
+		       u8 port_num,
+		       struct ib_wc *in_wc,
+		       struct ib_grh *in_grh,
+		       struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	return IB_MAD_RESULT_SUCCESS;
+}
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_ah.h b/drivers/infiniband/hw/ocrdma/ocrdma_ah.h
new file mode 100644
index 00000000000..8ac49e7f96d
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_ah.h
@@ -0,0 +1,42 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_AH_H__
+#define __OCRDMA_AH_H__
+
+struct ib_ah *ocrdma_create_ah(struct ib_pd *, struct ib_ah_attr *);
+int ocrdma_destroy_ah(struct ib_ah *);
+int ocrdma_query_ah(struct ib_ah *, struct ib_ah_attr *);
+int ocrdma_modify_ah(struct ib_ah *, struct ib_ah_attr *);
+
+int ocrdma_process_mad(struct ib_device *,
+		       int process_mad_flags,
+		       u8 port_num,
+		       struct ib_wc *in_wc,
+		       struct ib_grh *in_grh,
+		       struct ib_mad *in_mad, struct ib_mad *out_mad);
+#endif				/* __OCRDMA_AH_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.c b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
new file mode 100644
index 00000000000..3bbf2010a82
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.c
@@ -0,0 +1,2737 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) CNA Adapters.              *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#include <linux/sched.h>
+#include <linux/interrupt.h>
+#include <linux/log2.h>
+#include <linux/dma-mapping.h>
+
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+
+#include "ocrdma.h"
+#include "ocrdma_hw.h"
+#include "ocrdma_verbs.h"
+#include "ocrdma_ah.h"
+
+enum mbx_status {
+	OCRDMA_MBX_STATUS_FAILED		= 1,
+	OCRDMA_MBX_STATUS_ILLEGAL_FIELD		= 3,
+	OCRDMA_MBX_STATUS_OOR			= 100,
+	OCRDMA_MBX_STATUS_INVALID_PD		= 101,
+	OCRDMA_MBX_STATUS_PD_INUSE		= 102,
+	OCRDMA_MBX_STATUS_INVALID_CQ		= 103,
+	OCRDMA_MBX_STATUS_INVALID_QP		= 104,
+	OCRDMA_MBX_STATUS_INVALID_LKEY		= 105,
+	OCRDMA_MBX_STATUS_ORD_EXCEEDS		= 106,
+	OCRDMA_MBX_STATUS_IRD_EXCEEDS		= 107,
+	OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS	= 108,
+	OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS	= 109,
+	OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS	= 110,
+	OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS	= 111,
+	OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS	= 112,
+	OCRDMA_MBX_STATUS_INVALID_STATE_CHANGE	= 113,
+	OCRDMA_MBX_STATUS_MW_BOUND		= 114,
+	OCRDMA_MBX_STATUS_INVALID_VA		= 115,
+	OCRDMA_MBX_STATUS_INVALID_LENGTH	= 116,
+	OCRDMA_MBX_STATUS_INVALID_FBO		= 117,
+	OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS	= 118,
+	OCRDMA_MBX_STATUS_INVALID_PBE_SIZE	= 119,
+	OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY	= 120,
+	OCRDMA_MBX_STATUS_INVALID_PBL_SHIFT	= 121,
+	OCRDMA_MBX_STATUS_INVALID_SRQ_ID	= 129,
+	OCRDMA_MBX_STATUS_SRQ_ERROR		= 133,
+	OCRDMA_MBX_STATUS_RQE_EXCEEDS		= 134,
+	OCRDMA_MBX_STATUS_MTU_EXCEEDS		= 135,
+	OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS	= 136,
+	OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS	= 137,
+	OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS	= 138,
+	OCRDMA_MBX_STATUS_QP_BOUND		= 130,
+	OCRDMA_MBX_STATUS_INVALID_CHANGE	= 139,
+	OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP	= 140,
+	OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER	= 141,
+	OCRDMA_MBX_STATUS_MW_STILL_BOUND	= 142,
+	OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID	= 143,
+	OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS	= 144
+};
+
+enum additional_status {
+	OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES = 22
+};
+
+enum cqe_status {
+	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES	= 1,
+	OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER		= 2,
+	OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES	= 3,
+	OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING		= 4,
+	OCRDMA_MBX_CQE_STATUS_DMA_FAILED		= 5
+};
+
+static inline void *ocrdma_get_eqe(struct ocrdma_eq *eq)
+{
+	return eq->q.va + (eq->q.tail * sizeof(struct ocrdma_eqe));
+}
+
+static inline void ocrdma_eq_inc_tail(struct ocrdma_eq *eq)
+{
+	eq->q.tail = (eq->q.tail + 1) & (OCRDMA_EQ_LEN - 1);
+}
+
+static inline void *ocrdma_get_mcqe(struct ocrdma_dev *dev)
+{
+	struct ocrdma_mcqe *cqe = (struct ocrdma_mcqe *)
+	    (dev->mq.cq.va + (dev->mq.cq.tail * sizeof(struct ocrdma_mcqe)));
+
+	if (!(le32_to_cpu(cqe->valid_ae_cmpl_cons) & OCRDMA_MCQE_VALID_MASK))
+		return NULL;
+	return cqe;
+}
+
+static inline void ocrdma_mcq_inc_tail(struct ocrdma_dev *dev)
+{
+	dev->mq.cq.tail = (dev->mq.cq.tail + 1) & (OCRDMA_MQ_CQ_LEN - 1);
+}
+
+static inline struct ocrdma_mqe *ocrdma_get_mqe(struct ocrdma_dev *dev)
+{
+	return dev->mq.sq.va + (dev->mq.sq.head * sizeof(struct ocrdma_mqe));
+}
+
+static inline void ocrdma_mq_inc_head(struct ocrdma_dev *dev)
+{
+	dev->mq.sq.head = (dev->mq.sq.head + 1) & (OCRDMA_MQ_LEN - 1);
+}
+
+static inline void *ocrdma_get_mqe_rsp(struct ocrdma_dev *dev)
+{
+	return dev->mq.sq.va + (dev->mqe_ctx.tag * sizeof(struct ocrdma_mqe));
+}
+
+enum ib_qp_state get_ibqp_state(enum ocrdma_qp_state qps)
+{
+	switch (qps) {
+	case OCRDMA_QPS_RST:
+		return IB_QPS_RESET;
+	case OCRDMA_QPS_INIT:
+		return IB_QPS_INIT;
+	case OCRDMA_QPS_RTR:
+		return IB_QPS_RTR;
+	case OCRDMA_QPS_RTS:
+		return IB_QPS_RTS;
+	case OCRDMA_QPS_SQD:
+	case OCRDMA_QPS_SQ_DRAINING:
+		return IB_QPS_SQD;
+	case OCRDMA_QPS_SQE:
+		return IB_QPS_SQE;
+	case OCRDMA_QPS_ERR:
+		return IB_QPS_ERR;
+	}
+	return IB_QPS_ERR;
+}
+
+static enum ocrdma_qp_state get_ocrdma_qp_state(enum ib_qp_state qps)
+{
+	switch (qps) {
+	case IB_QPS_RESET:
+		return OCRDMA_QPS_RST;
+	case IB_QPS_INIT:
+		return OCRDMA_QPS_INIT;
+	case IB_QPS_RTR:
+		return OCRDMA_QPS_RTR;
+	case IB_QPS_RTS:
+		return OCRDMA_QPS_RTS;
+	case IB_QPS_SQD:
+		return OCRDMA_QPS_SQD;
+	case IB_QPS_SQE:
+		return OCRDMA_QPS_SQE;
+	case IB_QPS_ERR:
+		return OCRDMA_QPS_ERR;
+	}
+	return OCRDMA_QPS_ERR;
+}
+
+static int ocrdma_get_mbx_errno(u32 status)
+{
+	int err_num;
+	u8 mbox_status = (status & OCRDMA_MBX_RSP_STATUS_MASK) >>
+					OCRDMA_MBX_RSP_STATUS_SHIFT;
+	u8 add_status = (status & OCRDMA_MBX_RSP_ASTATUS_MASK) >>
+					OCRDMA_MBX_RSP_ASTATUS_SHIFT;
+
+	switch (mbox_status) {
+	case OCRDMA_MBX_STATUS_OOR:
+	case OCRDMA_MBX_STATUS_MAX_QP_EXCEEDS:
+		err_num = -EAGAIN;
+		break;
+
+	case OCRDMA_MBX_STATUS_INVALID_PD:
+	case OCRDMA_MBX_STATUS_INVALID_CQ:
+	case OCRDMA_MBX_STATUS_INVALID_SRQ_ID:
+	case OCRDMA_MBX_STATUS_INVALID_QP:
+	case OCRDMA_MBX_STATUS_INVALID_CHANGE:
+	case OCRDMA_MBX_STATUS_MTU_EXCEEDS:
+	case OCRDMA_MBX_STATUS_INVALID_RNR_NAK_TIMER:
+	case OCRDMA_MBX_STATUS_PKEY_INDEX_INVALID:
+	case OCRDMA_MBX_STATUS_PKEY_INDEX_EXCEEDS:
+	case OCRDMA_MBX_STATUS_ILLEGAL_FIELD:
+	case OCRDMA_MBX_STATUS_INVALID_PBL_ENTRY:
+	case OCRDMA_MBX_STATUS_INVALID_LKEY:
+	case OCRDMA_MBX_STATUS_INVALID_VA:
+	case OCRDMA_MBX_STATUS_INVALID_LENGTH:
+	case OCRDMA_MBX_STATUS_INVALID_FBO:
+	case OCRDMA_MBX_STATUS_INVALID_ACC_RIGHTS:
+	case OCRDMA_MBX_STATUS_INVALID_PBE_SIZE:
+	case OCRDMA_MBX_STATUS_ATOMIC_OPS_UNSUP:
+	case OCRDMA_MBX_STATUS_SRQ_ERROR:
+	case OCRDMA_MBX_STATUS_SRQ_SIZE_UNDERUNS:
+		err_num = -EINVAL;
+		break;
+
+	case OCRDMA_MBX_STATUS_PD_INUSE:
+	case OCRDMA_MBX_STATUS_QP_BOUND:
+	case OCRDMA_MBX_STATUS_MW_STILL_BOUND:
+	case OCRDMA_MBX_STATUS_MW_BOUND:
+		err_num = -EBUSY;
+		break;
+
+	case OCRDMA_MBX_STATUS_RECVQ_RQE_EXCEEDS:
+	case OCRDMA_MBX_STATUS_SGE_RECV_EXCEEDS:
+	case OCRDMA_MBX_STATUS_RQE_EXCEEDS:
+	case OCRDMA_MBX_STATUS_SRQ_LIMIT_EXCEEDS:
+	case OCRDMA_MBX_STATUS_ORD_EXCEEDS:
+	case OCRDMA_MBX_STATUS_IRD_EXCEEDS:
+	case OCRDMA_MBX_STATUS_SENDQ_WQE_EXCEEDS:
+	case OCRDMA_MBX_STATUS_SGE_SEND_EXCEEDS:
+	case OCRDMA_MBX_STATUS_SGE_WRITE_EXCEEDS:
+		err_num = -ENOBUFS;
+		break;
+
+	case OCRDMA_MBX_STATUS_FAILED:
+		switch (add_status) {
+		case OCRDMA_MBX_ADDI_STATUS_INSUFFICIENT_RESOURCES:
+			err_num = -EAGAIN;
+			break;
+		}
+	default:
+		err_num = -EFAULT;
+	}
+	return err_num;
+}
+
+char *port_speed_string(struct ocrdma_dev *dev)
+{
+	char *str = "";
+	u16 speeds_supported;
+
+	speeds_supported = dev->phy.fixed_speeds_supported |
+				dev->phy.auto_speeds_supported;
+	if (speeds_supported & OCRDMA_PHY_SPEED_40GBPS)
+		str = "40Gbps ";
+	else if (speeds_supported & OCRDMA_PHY_SPEED_10GBPS)
+		str = "10Gbps ";
+	else if (speeds_supported & OCRDMA_PHY_SPEED_1GBPS)
+		str = "1Gbps ";
+
+	return str;
+}
+
+static int ocrdma_get_mbx_cqe_errno(u16 cqe_status)
+{
+	int err_num = -EINVAL;
+
+	switch (cqe_status) {
+	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_PRIVILEDGES:
+		err_num = -EPERM;
+		break;
+	case OCRDMA_MBX_CQE_STATUS_INVALID_PARAMETER:
+		err_num = -EINVAL;
+		break;
+	case OCRDMA_MBX_CQE_STATUS_INSUFFICIENT_RESOURCES:
+	case OCRDMA_MBX_CQE_STATUS_QUEUE_FLUSHING:
+		err_num = -EINVAL;
+		break;
+	case OCRDMA_MBX_CQE_STATUS_DMA_FAILED:
+	default:
+		err_num = -EINVAL;
+		break;
+	}
+	return err_num;
+}
+
+void ocrdma_ring_cq_db(struct ocrdma_dev *dev, u16 cq_id, bool armed,
+		       bool solicited, u16 cqe_popped)
+{
+	u32 val = cq_id & OCRDMA_DB_CQ_RING_ID_MASK;
+
+	val |= ((cq_id & OCRDMA_DB_CQ_RING_ID_EXT_MASK) <<
+	     OCRDMA_DB_CQ_RING_ID_EXT_MASK_SHIFT);
+
+	if (armed)
+		val |= (1 << OCRDMA_DB_CQ_REARM_SHIFT);
+	if (solicited)
+		val |= (1 << OCRDMA_DB_CQ_SOLICIT_SHIFT);
+	val |= (cqe_popped << OCRDMA_DB_CQ_NUM_POPPED_SHIFT);
+	iowrite32(val, dev->nic_info.db + OCRDMA_DB_CQ_OFFSET);
+}
+
+static void ocrdma_ring_mq_db(struct ocrdma_dev *dev)
+{
+	u32 val = 0;
+
+	val |= dev->mq.sq.id & OCRDMA_MQ_ID_MASK;
+	val |= 1 << OCRDMA_MQ_NUM_MQE_SHIFT;
+	iowrite32(val, dev->nic_info.db + OCRDMA_DB_MQ_OFFSET);
+}
+
+static void ocrdma_ring_eq_db(struct ocrdma_dev *dev, u16 eq_id,
+			      bool arm, bool clear_int, u16 num_eqe)
+{
+	u32 val = 0;
+
+	val |= eq_id & OCRDMA_EQ_ID_MASK;
+	val |= ((eq_id & OCRDMA_EQ_ID_EXT_MASK) << OCRDMA_EQ_ID_EXT_MASK_SHIFT);
+	if (arm)
+		val |= (1 << OCRDMA_REARM_SHIFT);
+	if (clear_int)
+		val |= (1 << OCRDMA_EQ_CLR_SHIFT);
+	val |= (1 << OCRDMA_EQ_TYPE_SHIFT);
+	val |= (num_eqe << OCRDMA_NUM_EQE_SHIFT);
+	iowrite32(val, dev->nic_info.db + OCRDMA_DB_EQ_OFFSET);
+}
+
+static void ocrdma_init_mch(struct ocrdma_mbx_hdr *cmd_hdr,
+			    u8 opcode, u8 subsys, u32 cmd_len)
+{
+	cmd_hdr->subsys_op = (opcode | (subsys << OCRDMA_MCH_SUBSYS_SHIFT));
+	cmd_hdr->timeout = 20; /* seconds */
+	cmd_hdr->cmd_len = cmd_len - sizeof(struct ocrdma_mbx_hdr);
+}
+
+static void *ocrdma_init_emb_mqe(u8 opcode, u32 cmd_len)
+{
+	struct ocrdma_mqe *mqe;
+
+	mqe = kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
+	if (!mqe)
+		return NULL;
+	mqe->hdr.spcl_sge_cnt_emb |=
+		(OCRDMA_MQE_EMBEDDED << OCRDMA_MQE_HDR_EMB_SHIFT) &
+					OCRDMA_MQE_HDR_EMB_MASK;
+	mqe->hdr.pyld_len = cmd_len - sizeof(struct ocrdma_mqe_hdr);
+
+	ocrdma_init_mch(&mqe->u.emb_req.mch, opcode, OCRDMA_SUBSYS_ROCE,
+			mqe->hdr.pyld_len);
+	return mqe;
+}
+
+static void *ocrdma_alloc_mqe(void)
+{
+	return kzalloc(sizeof(struct ocrdma_mqe), GFP_KERNEL);
+}
+
+static void ocrdma_free_q(struct ocrdma_dev *dev, struct ocrdma_queue_info *q)
+{
+	dma_free_coherent(&dev->nic_info.pdev->dev, q->size, q->va, q->dma);
+}
+
+static int ocrdma_alloc_q(struct ocrdma_dev *dev,
+			  struct ocrdma_queue_info *q, u16 len, u16 entry_size)
+{
+	memset(q, 0, sizeof(*q));
+	q->len = len;
+	q->entry_size = entry_size;
+	q->size = len * entry_size;
+	q->va = dma_alloc_coherent(&dev->nic_info.pdev->dev, q->size,
+				   &q->dma, GFP_KERNEL);
+	if (!q->va)
+		return -ENOMEM;
+	memset(q->va, 0, q->size);
+	return 0;
+}
+
+static void ocrdma_build_q_pages(struct ocrdma_pa *q_pa, int cnt,
+					dma_addr_t host_pa, int hw_page_size)
+{
+	int i;
+
+	for (i = 0; i < cnt; i++) {
+		q_pa[i].lo = (u32) (host_pa & 0xffffffff);
+		q_pa[i].hi = (u32) upper_32_bits(host_pa);
+		host_pa += hw_page_size;
+	}
+}
+
+static int ocrdma_mbx_delete_q(struct ocrdma_dev *dev,
+			       struct ocrdma_queue_info *q, int queue_type)
+{
+	u8 opcode = 0;
+	int status;
+	struct ocrdma_delete_q_req *cmd = dev->mbx_cmd;
+
+	switch (queue_type) {
+	case QTYPE_MCCQ:
+		opcode = OCRDMA_CMD_DELETE_MQ;
+		break;
+	case QTYPE_CQ:
+		opcode = OCRDMA_CMD_DELETE_CQ;
+		break;
+	case QTYPE_EQ:
+		opcode = OCRDMA_CMD_DELETE_EQ;
+		break;
+	default:
+		BUG();
+	}
+	memset(cmd, 0, sizeof(*cmd));
+	ocrdma_init_mch(&cmd->req, opcode, OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+	cmd->id = q->id;
+
+	status = be_roce_mcc_cmd(dev->nic_info.netdev,
+				 cmd, sizeof(*cmd), NULL, NULL);
+	if (!status)
+		q->created = false;
+	return status;
+}
+
+static int ocrdma_mbx_create_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
+{
+	int status;
+	struct ocrdma_create_eq_req *cmd = dev->mbx_cmd;
+	struct ocrdma_create_eq_rsp *rsp = dev->mbx_cmd;
+
+	memset(cmd, 0, sizeof(*cmd));
+	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_EQ, OCRDMA_SUBSYS_COMMON,
+			sizeof(*cmd));
+
+	cmd->req.rsvd_version = 2;
+	cmd->num_pages = 4;
+	cmd->valid = OCRDMA_CREATE_EQ_VALID;
+	cmd->cnt = 4 << OCRDMA_CREATE_EQ_CNT_SHIFT;
+
+	ocrdma_build_q_pages(&cmd->pa[0], cmd->num_pages, eq->q.dma,
+			     PAGE_SIZE_4K);
+	status = be_roce_mcc_cmd(dev->nic_info.netdev, cmd, sizeof(*cmd), NULL,
+				 NULL);
+	if (!status) {
+		eq->q.id = rsp->vector_eqid & 0xffff;
+		eq->vector = (rsp->vector_eqid >> 16) & 0xffff;
+		eq->q.created = true;
+	}
+	return status;
+}
+
+static int ocrdma_create_eq(struct ocrdma_dev *dev,
+			    struct ocrdma_eq *eq, u16 q_len)
+{
+	int status;
+
+	status = ocrdma_alloc_q(dev, &eq->q, OCRDMA_EQ_LEN,
+				sizeof(struct ocrdma_eqe));
+	if (status)
+		return status;
+
+	status = ocrdma_mbx_create_eq(dev, eq);
+	if (status)
+		goto mbx_err;
+	eq->dev = dev;
+	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
+
+	return 0;
+mbx_err:
+	ocrdma_free_q(dev, &eq->q);
+	return status;
+}
+
+int ocrdma_get_irq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
+{
+	int irq;
+
+	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX)
+		irq = dev->nic_info.pdev->irq;
+	else
+		irq = dev->nic_info.msix.vector_list[eq->vector];
+	return irq;
+}
+
+static void _ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
+{
+	if (eq->q.created) {
+		ocrdma_mbx_delete_q(dev, &eq->q, QTYPE_EQ);
+		ocrdma_free_q(dev, &eq->q);
+	}
+}
+
+static void ocrdma_destroy_eq(struct ocrdma_dev *dev, struct ocrdma_eq *eq)
+{
+	int irq;
+
+	/* disarm EQ so that interrupts are not generated
+	 * during freeing and EQ delete is in progress.
+	 */
+	ocrdma_ring_eq_db(dev, eq->q.id, false, false, 0);
+
+	irq = ocrdma_get_irq(dev, eq);
+	free_irq(irq, eq);
+	_ocrdma_destroy_eq(dev, eq);
+}
+
+static void ocrdma_destroy_eqs(struct ocrdma_dev *dev)
+{
+	int i;
+
+	for (i = 0; i < dev->eq_cnt; i++)
+		ocrdma_destroy_eq(dev, &dev->eq_tbl[i]);
+}
+
+static int ocrdma_mbx_mq_cq_create(struct ocrdma_dev *dev,
+				   struct ocrdma_queue_info *cq,
+				   struct ocrdma_queue_info *eq)
+{
+	struct ocrdma_create_cq_cmd *cmd = dev->mbx_cmd;
+	struct ocrdma_create_cq_cmd_rsp *rsp = dev->mbx_cmd;
+	int status;
+
+	memset(cmd, 0, sizeof(*cmd));
+	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_CQ,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+
+	cmd->req.rsvd_version = OCRDMA_CREATE_CQ_VER2;
+	cmd->pgsz_pgcnt = (cq->size / OCRDMA_MIN_Q_PAGE_SIZE) <<
+		OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
+	cmd->pgsz_pgcnt |= PAGES_4K_SPANNED(cq->va, cq->size);
+
+	cmd->ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
+	cmd->eqn = eq->id;
+	cmd->cqe_count = cq->size / sizeof(struct ocrdma_mcqe);
+
+	ocrdma_build_q_pages(&cmd->pa[0], cq->size / OCRDMA_MIN_Q_PAGE_SIZE,
+			     cq->dma, PAGE_SIZE_4K);
+	status = be_roce_mcc_cmd(dev->nic_info.netdev,
+				 cmd, sizeof(*cmd), NULL, NULL);
+	if (!status) {
+		cq->id = (u16) (rsp->cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
+		cq->created = true;
+	}
+	return status;
+}
+
+static u32 ocrdma_encoded_q_len(int q_len)
+{
+	u32 len_encoded = fls(q_len);	/* log2(len) + 1 */
+
+	if (len_encoded == 16)
+		len_encoded = 0;
+	return len_encoded;
+}
+
+static int ocrdma_mbx_create_mq(struct ocrdma_dev *dev,
+				struct ocrdma_queue_info *mq,
+				struct ocrdma_queue_info *cq)
+{
+	int num_pages, status;
+	struct ocrdma_create_mq_req *cmd = dev->mbx_cmd;
+	struct ocrdma_create_mq_rsp *rsp = dev->mbx_cmd;
+	struct ocrdma_pa *pa;
+
+	memset(cmd, 0, sizeof(*cmd));
+	num_pages = PAGES_4K_SPANNED(mq->va, mq->size);
+
+	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_CREATE_MQ_EXT,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+	cmd->req.rsvd_version = 1;
+	cmd->cqid_pages = num_pages;
+	cmd->cqid_pages |= (cq->id << OCRDMA_CREATE_MQ_CQ_ID_SHIFT);
+	cmd->async_cqid_valid = OCRDMA_CREATE_MQ_ASYNC_CQ_VALID;
+
+	cmd->async_event_bitmap = Bit(OCRDMA_ASYNC_GRP5_EVE_CODE);
+	cmd->async_event_bitmap |= Bit(OCRDMA_ASYNC_RDMA_EVE_CODE);
+
+	cmd->async_cqid_ringsize = cq->id;
+	cmd->async_cqid_ringsize |= (ocrdma_encoded_q_len(mq->len) <<
+				OCRDMA_CREATE_MQ_RING_SIZE_SHIFT);
+	cmd->valid = OCRDMA_CREATE_MQ_VALID;
+	pa = &cmd->pa[0];
+
+	ocrdma_build_q_pages(pa, num_pages, mq->dma, PAGE_SIZE_4K);
+	status = be_roce_mcc_cmd(dev->nic_info.netdev,
+				 cmd, sizeof(*cmd), NULL, NULL);
+	if (!status) {
+		mq->id = rsp->id;
+		mq->created = true;
+	}
+	return status;
+}
+
+static int ocrdma_create_mq(struct ocrdma_dev *dev)
+{
+	int status;
+
+	/* Alloc completion queue for Mailbox queue */
+	status = ocrdma_alloc_q(dev, &dev->mq.cq, OCRDMA_MQ_CQ_LEN,
+				sizeof(struct ocrdma_mcqe));
+	if (status)
+		goto alloc_err;
+
+	dev->eq_tbl[0].cq_cnt++;
+	status = ocrdma_mbx_mq_cq_create(dev, &dev->mq.cq, &dev->eq_tbl[0].q);
+	if (status)
+		goto mbx_cq_free;
+
+	memset(&dev->mqe_ctx, 0, sizeof(dev->mqe_ctx));
+	init_waitqueue_head(&dev->mqe_ctx.cmd_wait);
+	mutex_init(&dev->mqe_ctx.lock);
+
+	/* Alloc Mailbox queue */
+	status = ocrdma_alloc_q(dev, &dev->mq.sq, OCRDMA_MQ_LEN,
+				sizeof(struct ocrdma_mqe));
+	if (status)
+		goto mbx_cq_destroy;
+	status = ocrdma_mbx_create_mq(dev, &dev->mq.sq, &dev->mq.cq);
+	if (status)
+		goto mbx_q_free;
+	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, 0);
+	return 0;
+
+mbx_q_free:
+	ocrdma_free_q(dev, &dev->mq.sq);
+mbx_cq_destroy:
+	ocrdma_mbx_delete_q(dev, &dev->mq.cq, QTYPE_CQ);
+mbx_cq_free:
+	ocrdma_free_q(dev, &dev->mq.cq);
+alloc_err:
+	return status;
+}
+
+static void ocrdma_destroy_mq(struct ocrdma_dev *dev)
+{
+	struct ocrdma_queue_info *mbxq, *cq;
+
+	/* mqe_ctx lock synchronizes with any other pending cmds. */
+	mutex_lock(&dev->mqe_ctx.lock);
+	mbxq = &dev->mq.sq;
+	if (mbxq->created) {
+		ocrdma_mbx_delete_q(dev, mbxq, QTYPE_MCCQ);
+		ocrdma_free_q(dev, mbxq);
+	}
+	mutex_unlock(&dev->mqe_ctx.lock);
+
+	cq = &dev->mq.cq;
+	if (cq->created) {
+		ocrdma_mbx_delete_q(dev, cq, QTYPE_CQ);
+		ocrdma_free_q(dev, cq);
+	}
+}
+
+static void ocrdma_process_qpcat_error(struct ocrdma_dev *dev,
+				       struct ocrdma_qp *qp)
+{
+	enum ib_qp_state new_ib_qps = IB_QPS_ERR;
+	enum ib_qp_state old_ib_qps;
+
+	if (qp == NULL)
+		BUG();
+	ocrdma_qp_state_change(qp, new_ib_qps, &old_ib_qps);
+}
+
+static void ocrdma_dispatch_ibevent(struct ocrdma_dev *dev,
+				    struct ocrdma_ae_mcqe *cqe)
+{
+	struct ocrdma_qp *qp = NULL;
+	struct ocrdma_cq *cq = NULL;
+	struct ib_event ib_evt = { 0 };
+	int cq_event = 0;
+	int qp_event = 1;
+	int srq_event = 0;
+	int dev_event = 0;
+	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
+	    OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
+
+	if (cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPVALID)
+		qp = dev->qp_tbl[cqe->qpvalid_qpid & OCRDMA_AE_MCQE_QPID_MASK];
+	if (cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQVALID)
+		cq = dev->cq_tbl[cqe->cqvalid_cqid & OCRDMA_AE_MCQE_CQID_MASK];
+
+	ib_evt.device = &dev->ibdev;
+
+	switch (type) {
+	case OCRDMA_CQ_ERROR:
+		ib_evt.element.cq = &cq->ibcq;
+		ib_evt.event = IB_EVENT_CQ_ERR;
+		cq_event = 1;
+		qp_event = 0;
+		break;
+	case OCRDMA_CQ_OVERRUN_ERROR:
+		ib_evt.element.cq = &cq->ibcq;
+		ib_evt.event = IB_EVENT_CQ_ERR;
+		cq_event = 1;
+		qp_event = 0;
+		break;
+	case OCRDMA_CQ_QPCAT_ERROR:
+		ib_evt.element.qp = &qp->ibqp;
+		ib_evt.event = IB_EVENT_QP_FATAL;
+		ocrdma_process_qpcat_error(dev, qp);
+		break;
+	case OCRDMA_QP_ACCESS_ERROR:
+		ib_evt.element.qp = &qp->ibqp;
+		ib_evt.event = IB_EVENT_QP_ACCESS_ERR;
+		break;
+	case OCRDMA_QP_COMM_EST_EVENT:
+		ib_evt.element.qp = &qp->ibqp;
+		ib_evt.event = IB_EVENT_COMM_EST;
+		break;
+	case OCRDMA_SQ_DRAINED_EVENT:
+		ib_evt.element.qp = &qp->ibqp;
+		ib_evt.event = IB_EVENT_SQ_DRAINED;
+		break;
+	case OCRDMA_DEVICE_FATAL_EVENT:
+		ib_evt.element.port_num = 1;
+		ib_evt.event = IB_EVENT_DEVICE_FATAL;
+		qp_event = 0;
+		dev_event = 1;
+		break;
+	case OCRDMA_SRQCAT_ERROR:
+		ib_evt.element.srq = &qp->srq->ibsrq;
+		ib_evt.event = IB_EVENT_SRQ_ERR;
+		srq_event = 1;
+		qp_event = 0;
+		break;
+	case OCRDMA_SRQ_LIMIT_EVENT:
+		ib_evt.element.srq = &qp->srq->ibsrq;
+		ib_evt.event = IB_EVENT_SRQ_LIMIT_REACHED;
+		srq_event = 1;
+		qp_event = 0;
+		break;
+	case OCRDMA_QP_LAST_WQE_EVENT:
+		ib_evt.element.qp = &qp->ibqp;
+		ib_evt.event = IB_EVENT_QP_LAST_WQE_REACHED;
+		break;
+	default:
+		cq_event = 0;
+		qp_event = 0;
+		srq_event = 0;
+		dev_event = 0;
+		pr_err("%s() unknown type=0x%x\n", __func__, type);
+		break;
+	}
+
+	if (qp_event) {
+		if (qp->ibqp.event_handler)
+			qp->ibqp.event_handler(&ib_evt, qp->ibqp.qp_context);
+	} else if (cq_event) {
+		if (cq->ibcq.event_handler)
+			cq->ibcq.event_handler(&ib_evt, cq->ibcq.cq_context);
+	} else if (srq_event) {
+		if (qp->srq->ibsrq.event_handler)
+			qp->srq->ibsrq.event_handler(&ib_evt,
+						     qp->srq->ibsrq.
+						     srq_context);
+	} else if (dev_event) {
+		pr_err("%s: Fatal event received\n", dev->ibdev.name);
+		ib_dispatch_event(&ib_evt);
+	}
+
+}
+
+static void ocrdma_process_grp5_aync(struct ocrdma_dev *dev,
+					struct ocrdma_ae_mcqe *cqe)
+{
+	struct ocrdma_ae_pvid_mcqe *evt;
+	int type = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_TYPE_MASK) >>
+			OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT;
+
+	switch (type) {
+	case OCRDMA_ASYNC_EVENT_PVID_STATE:
+		evt = (struct ocrdma_ae_pvid_mcqe *)cqe;
+		if ((evt->tag_enabled & OCRDMA_AE_PVID_MCQE_ENABLED_MASK) >>
+			OCRDMA_AE_PVID_MCQE_ENABLED_SHIFT)
+			dev->pvid = ((evt->tag_enabled &
+					OCRDMA_AE_PVID_MCQE_TAG_MASK) >>
+					OCRDMA_AE_PVID_MCQE_TAG_SHIFT);
+		break;
+	default:
+		/* Not interested evts. */
+		break;
+	}
+}
+
+static void ocrdma_process_acqe(struct ocrdma_dev *dev, void *ae_cqe)
+{
+	/* async CQE processing */
+	struct ocrdma_ae_mcqe *cqe = ae_cqe;
+	u32 evt_code = (cqe->valid_ae_event & OCRDMA_AE_MCQE_EVENT_CODE_MASK) >>
+			OCRDMA_AE_MCQE_EVENT_CODE_SHIFT;
+
+	if (evt_code == OCRDMA_ASYNC_RDMA_EVE_CODE)
+		ocrdma_dispatch_ibevent(dev, cqe);
+	else if (evt_code == OCRDMA_ASYNC_GRP5_EVE_CODE)
+		ocrdma_process_grp5_aync(dev, cqe);
+	else
+		pr_err("%s(%d) invalid evt code=0x%x\n", __func__,
+		       dev->id, evt_code);
+}
+
+static void ocrdma_process_mcqe(struct ocrdma_dev *dev, struct ocrdma_mcqe *cqe)
+{
+	if (dev->mqe_ctx.tag == cqe->tag_lo && dev->mqe_ctx.cmd_done == false) {
+		dev->mqe_ctx.cqe_status = (cqe->status &
+		     OCRDMA_MCQE_STATUS_MASK) >> OCRDMA_MCQE_STATUS_SHIFT;
+		dev->mqe_ctx.ext_status =
+		    (cqe->status & OCRDMA_MCQE_ESTATUS_MASK)
+		    >> OCRDMA_MCQE_ESTATUS_SHIFT;
+		dev->mqe_ctx.cmd_done = true;
+		wake_up(&dev->mqe_ctx.cmd_wait);
+	} else
+		pr_err("%s() cqe for invalid tag0x%x.expected=0x%x\n",
+		       __func__, cqe->tag_lo, dev->mqe_ctx.tag);
+}
+
+static int ocrdma_mq_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
+{
+	u16 cqe_popped = 0;
+	struct ocrdma_mcqe *cqe;
+
+	while (1) {
+		cqe = ocrdma_get_mcqe(dev);
+		if (cqe == NULL)
+			break;
+		ocrdma_le32_to_cpu(cqe, sizeof(*cqe));
+		cqe_popped += 1;
+		if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_AE_MASK)
+			ocrdma_process_acqe(dev, cqe);
+		else if (cqe->valid_ae_cmpl_cons & OCRDMA_MCQE_CMPL_MASK)
+			ocrdma_process_mcqe(dev, cqe);
+		memset(cqe, 0, sizeof(struct ocrdma_mcqe));
+		ocrdma_mcq_inc_tail(dev);
+	}
+	ocrdma_ring_cq_db(dev, dev->mq.cq.id, true, false, cqe_popped);
+	return 0;
+}
+
+static void ocrdma_qp_buddy_cq_handler(struct ocrdma_dev *dev,
+				       struct ocrdma_cq *cq)
+{
+	unsigned long flags;
+	struct ocrdma_qp *qp;
+	bool buddy_cq_found = false;
+	/* Go through list of QPs in error state which are using this CQ
+	 * and invoke its callback handler to trigger CQE processing for
+	 * error/flushed CQE. It is rare to find more than few entries in
+	 * this list as most consumers stops after getting error CQE.
+	 * List is traversed only once when a matching buddy cq found for a QP.
+	 */
+	spin_lock_irqsave(&dev->flush_q_lock, flags);
+	list_for_each_entry(qp, &cq->sq_head, sq_entry) {
+		if (qp->srq)
+			continue;
+		/* if wq and rq share the same cq, than comp_handler
+		 * is already invoked.
+		 */
+		if (qp->sq_cq == qp->rq_cq)
+			continue;
+		/* if completion came on sq, rq's cq is buddy cq.
+		 * if completion came on rq, sq's cq is buddy cq.
+		 */
+		if (qp->sq_cq == cq)
+			cq = qp->rq_cq;
+		else
+			cq = qp->sq_cq;
+		buddy_cq_found = true;
+		break;
+	}
+	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
+	if (buddy_cq_found == false)
+		return;
+	if (cq->ibcq.comp_handler) {
+		spin_lock_irqsave(&cq->comp_handler_lock, flags);
+		(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+		spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+	}
+}
+
+static void ocrdma_qp_cq_handler(struct ocrdma_dev *dev, u16 cq_idx)
+{
+	unsigned long flags;
+	struct ocrdma_cq *cq;
+
+	if (cq_idx >= OCRDMA_MAX_CQ)
+		BUG();
+
+	cq = dev->cq_tbl[cq_idx];
+	if (cq == NULL)
+		return;
+
+	if (cq->ibcq.comp_handler) {
+		spin_lock_irqsave(&cq->comp_handler_lock, flags);
+		(*cq->ibcq.comp_handler) (&cq->ibcq, cq->ibcq.cq_context);
+		spin_unlock_irqrestore(&cq->comp_handler_lock, flags);
+	}
+	ocrdma_qp_buddy_cq_handler(dev, cq);
+}
+
+static void ocrdma_cq_handler(struct ocrdma_dev *dev, u16 cq_id)
+{
+	/* process the MQ-CQE. */
+	if (cq_id == dev->mq.cq.id)
+		ocrdma_mq_cq_handler(dev, cq_id);
+	else
+		ocrdma_qp_cq_handler(dev, cq_id);
+}
+
+static irqreturn_t ocrdma_irq_handler(int irq, void *handle)
+{
+	struct ocrdma_eq *eq = handle;
+	struct ocrdma_dev *dev = eq->dev;
+	struct ocrdma_eqe eqe;
+	struct ocrdma_eqe *ptr;
+	u16 cq_id;
+	int budget = eq->cq_cnt;
+
+	do {
+		ptr = ocrdma_get_eqe(eq);
+		eqe = *ptr;
+		ocrdma_le32_to_cpu(&eqe, sizeof(eqe));
+		if ((eqe.id_valid & OCRDMA_EQE_VALID_MASK) == 0)
+			break;
+
+		ptr->id_valid = 0;
+		/* ring eq doorbell as soon as its consumed. */
+		ocrdma_ring_eq_db(dev, eq->q.id, false, true, 1);
+		/* check whether its CQE or not. */
+		if ((eqe.id_valid & OCRDMA_EQE_FOR_CQE_MASK) == 0) {
+			cq_id = eqe.id_valid >> OCRDMA_EQE_RESOURCE_ID_SHIFT;
+			ocrdma_cq_handler(dev, cq_id);
+		}
+		ocrdma_eq_inc_tail(eq);
+
+		/* There can be a stale EQE after the last bound CQ is
+		 * destroyed. EQE valid and budget == 0 implies this.
+		 */
+		if (budget)
+			budget--;
+
+	} while (budget);
+
+	ocrdma_ring_eq_db(dev, eq->q.id, true, true, 0);
+	return IRQ_HANDLED;
+}
+
+static void ocrdma_post_mqe(struct ocrdma_dev *dev, struct ocrdma_mqe *cmd)
+{
+	struct ocrdma_mqe *mqe;
+
+	dev->mqe_ctx.tag = dev->mq.sq.head;
+	dev->mqe_ctx.cmd_done = false;
+	mqe = ocrdma_get_mqe(dev);
+	cmd->hdr.tag_lo = dev->mq.sq.head;
+	ocrdma_copy_cpu_to_le32(mqe, cmd, sizeof(*mqe));
+	/* make sure descriptor is written before ringing doorbell */
+	wmb();
+	ocrdma_mq_inc_head(dev);
+	ocrdma_ring_mq_db(dev);
+}
+
+static int ocrdma_wait_mqe_cmpl(struct ocrdma_dev *dev)
+{
+	long status;
+	/* 30 sec timeout */
+	status = wait_event_timeout(dev->mqe_ctx.cmd_wait,
+				    (dev->mqe_ctx.cmd_done != false),
+				    msecs_to_jiffies(30000));
+	if (status)
+		return 0;
+	else
+		return -1;
+}
+
+/* issue a mailbox command on the MQ */
+static int ocrdma_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe)
+{
+	int status = 0;
+	u16 cqe_status, ext_status;
+	struct ocrdma_mqe *rsp_mqe;
+	struct ocrdma_mbx_rsp *rsp = NULL;
+
+	mutex_lock(&dev->mqe_ctx.lock);
+	ocrdma_post_mqe(dev, mqe);
+	status = ocrdma_wait_mqe_cmpl(dev);
+	if (status)
+		goto mbx_err;
+	cqe_status = dev->mqe_ctx.cqe_status;
+	ext_status = dev->mqe_ctx.ext_status;
+	rsp_mqe = ocrdma_get_mqe_rsp(dev);
+	ocrdma_copy_le32_to_cpu(mqe, rsp_mqe, (sizeof(*mqe)));
+	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
+				OCRDMA_MQE_HDR_EMB_SHIFT)
+		rsp = &mqe->u.rsp;
+
+	if (cqe_status || ext_status) {
+		pr_err("%s() cqe_status=0x%x, ext_status=0x%x,",
+		       __func__, cqe_status, ext_status);
+		if (rsp) {
+			/* This is for embedded cmds. */
+			pr_err("opcode=0x%x, subsystem=0x%x\n",
+			       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
+				OCRDMA_MBX_RSP_OPCODE_SHIFT,
+				(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
+				OCRDMA_MBX_RSP_SUBSYS_SHIFT);
+		}
+		status = ocrdma_get_mbx_cqe_errno(cqe_status);
+		goto mbx_err;
+	}
+	/* For non embedded, rsp errors are handled in ocrdma_nonemb_mbx_cmd */
+	if (rsp && (mqe->u.rsp.status & OCRDMA_MBX_RSP_STATUS_MASK))
+		status = ocrdma_get_mbx_errno(mqe->u.rsp.status);
+mbx_err:
+	mutex_unlock(&dev->mqe_ctx.lock);
+	return status;
+}
+
+static int ocrdma_nonemb_mbx_cmd(struct ocrdma_dev *dev, struct ocrdma_mqe *mqe,
+				 void *payload_va)
+{
+	int status = 0;
+	struct ocrdma_mbx_rsp *rsp = payload_va;
+
+	if ((mqe->hdr.spcl_sge_cnt_emb & OCRDMA_MQE_HDR_EMB_MASK) >>
+				OCRDMA_MQE_HDR_EMB_SHIFT)
+		BUG();
+
+	status = ocrdma_mbx_cmd(dev, mqe);
+	if (!status)
+		/* For non embedded, only CQE failures are handled in
+		 * ocrdma_mbx_cmd. We need to check for RSP errors.
+		 */
+		if (rsp->status & OCRDMA_MBX_RSP_STATUS_MASK)
+			status = ocrdma_get_mbx_errno(rsp->status);
+
+	if (status)
+		pr_err("opcode=0x%x, subsystem=0x%x\n",
+		       (rsp->subsys_op & OCRDMA_MBX_RSP_OPCODE_MASK) >>
+			OCRDMA_MBX_RSP_OPCODE_SHIFT,
+			(rsp->subsys_op & OCRDMA_MBX_RSP_SUBSYS_MASK) >>
+			OCRDMA_MBX_RSP_SUBSYS_SHIFT);
+	return status;
+}
+
+static void ocrdma_get_attr(struct ocrdma_dev *dev,
+			      struct ocrdma_dev_attr *attr,
+			      struct ocrdma_mbx_query_config *rsp)
+{
+	attr->max_pd =
+	    (rsp->max_pd_ca_ack_delay & OCRDMA_MBX_QUERY_CFG_MAX_PD_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_PD_SHIFT;
+	attr->max_qp =
+	    (rsp->qp_srq_cq_ird_ord & OCRDMA_MBX_QUERY_CFG_MAX_QP_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_QP_SHIFT;
+	attr->max_srq =
+		(rsp->max_srq_rpir_qps & OCRDMA_MBX_QUERY_CFG_MAX_SRQ_MASK) >>
+		OCRDMA_MBX_QUERY_CFG_MAX_SRQ_OFFSET;
+	attr->max_send_sge = ((rsp->max_write_send_sge &
+			       OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK) >>
+			      OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT);
+	attr->max_recv_sge = (rsp->max_write_send_sge &
+			      OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT;
+	attr->max_srq_sge = (rsp->max_srq_rqe_sge &
+			      OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_OFFSET;
+	attr->max_rdma_sge = (rsp->max_write_send_sge &
+			      OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_SHIFT;
+	attr->max_ord_per_qp = (rsp->max_ird_ord_per_qp &
+				OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_SHIFT;
+	attr->max_ird_per_qp = (rsp->max_ird_ord_per_qp &
+				OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_SHIFT;
+	attr->cq_overflow_detect = (rsp->qp_srq_cq_ird_ord &
+				    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_SHIFT;
+	attr->srq_supported = (rsp->qp_srq_cq_ird_ord &
+			       OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_SHIFT;
+	attr->local_ca_ack_delay = (rsp->max_pd_ca_ack_delay &
+				    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_MASK) >>
+	    OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT;
+	attr->max_mw = rsp->max_mw;
+	attr->max_mr = rsp->max_mr;
+	attr->max_mr_size = ~0ull;
+	attr->max_fmr = 0;
+	attr->max_pages_per_frmr = rsp->max_pages_per_frmr;
+	attr->max_num_mr_pbl = rsp->max_num_mr_pbl;
+	attr->max_cqe = rsp->max_cq_cqes_per_cq &
+			OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_MASK;
+	attr->max_cq = (rsp->max_cq_cqes_per_cq &
+			OCRDMA_MBX_QUERY_CFG_MAX_CQ_MASK) >>
+			OCRDMA_MBX_QUERY_CFG_MAX_CQ_OFFSET;
+	attr->wqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
+		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_MASK) >>
+		OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_OFFSET) *
+		OCRDMA_WQE_STRIDE;
+	attr->rqe_size = ((rsp->wqe_rqe_stride_max_dpp_cqs &
+		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_MASK) >>
+		OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_OFFSET) *
+		OCRDMA_WQE_STRIDE;
+	attr->max_inline_data =
+	    attr->wqe_size - (sizeof(struct ocrdma_hdr_wqe) +
+			      sizeof(struct ocrdma_sge));
+	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
+		attr->ird = 1;
+		attr->ird_page_size = OCRDMA_MIN_Q_PAGE_SIZE;
+		attr->num_ird_pages = MAX_OCRDMA_IRD_PAGES;
+	}
+	dev->attr.max_wqe = rsp->max_wqes_rqes_per_q >>
+		 OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_OFFSET;
+	dev->attr.max_rqe = rsp->max_wqes_rqes_per_q &
+		OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_MASK;
+}
+
+static int ocrdma_check_fw_config(struct ocrdma_dev *dev,
+				   struct ocrdma_fw_conf_rsp *conf)
+{
+	u32 fn_mode;
+
+	fn_mode = conf->fn_mode & OCRDMA_FN_MODE_RDMA;
+	if (fn_mode != OCRDMA_FN_MODE_RDMA)
+		return -EINVAL;
+	dev->base_eqid = conf->base_eqid;
+	dev->max_eq = conf->max_eq;
+	return 0;
+}
+
+/* can be issued only during init time. */
+static int ocrdma_mbx_query_fw_ver(struct ocrdma_dev *dev)
+{
+	int status = -ENOMEM;
+	struct ocrdma_mqe *cmd;
+	struct ocrdma_fw_ver_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_VER, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
+			OCRDMA_CMD_GET_FW_VER,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_fw_ver_rsp *)cmd;
+	memset(&dev->attr.fw_ver[0], 0, sizeof(dev->attr.fw_ver));
+	memcpy(&dev->attr.fw_ver[0], &rsp->running_ver[0],
+	       sizeof(rsp->running_ver));
+	ocrdma_le32_to_cpu(dev->attr.fw_ver, sizeof(rsp->running_ver));
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+/* can be issued only during init time. */
+static int ocrdma_mbx_query_fw_config(struct ocrdma_dev *dev)
+{
+	int status = -ENOMEM;
+	struct ocrdma_mqe *cmd;
+	struct ocrdma_fw_conf_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_GET_FW_CONFIG, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
+			OCRDMA_CMD_GET_FW_CONFIG,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_fw_conf_rsp *)cmd;
+	status = ocrdma_check_fw_config(dev, rsp);
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_rdma_stats(struct ocrdma_dev *dev, bool reset)
+{
+	struct ocrdma_rdma_stats_req *req = dev->stats_mem.va;
+	struct ocrdma_mqe *mqe = &dev->stats_mem.mqe;
+	struct ocrdma_rdma_stats_resp *old_stats = NULL;
+	int status;
+
+	old_stats = kzalloc(sizeof(*old_stats), GFP_KERNEL);
+	if (old_stats == NULL)
+		return -ENOMEM;
+
+	memset(mqe, 0, sizeof(*mqe));
+	mqe->hdr.pyld_len = dev->stats_mem.size;
+	mqe->hdr.spcl_sge_cnt_emb |=
+			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
+				OCRDMA_MQE_HDR_SGE_CNT_MASK;
+	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dev->stats_mem.pa & 0xffffffff);
+	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dev->stats_mem.pa);
+	mqe->u.nonemb_req.sge[0].len = dev->stats_mem.size;
+
+	/* Cache the old stats */
+	memcpy(old_stats, req, sizeof(struct ocrdma_rdma_stats_resp));
+	memset(req, 0, dev->stats_mem.size);
+
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)req,
+			OCRDMA_CMD_GET_RDMA_STATS,
+			OCRDMA_SUBSYS_ROCE,
+			dev->stats_mem.size);
+	if (reset)
+		req->reset_stats = reset;
+
+	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dev->stats_mem.va);
+	if (status)
+		/* Copy from cache, if mbox fails */
+		memcpy(req, old_stats, sizeof(struct ocrdma_rdma_stats_resp));
+	else
+		ocrdma_le32_to_cpu(req, dev->stats_mem.size);
+
+	kfree(old_stats);
+	return status;
+}
+
+static int ocrdma_mbx_get_ctrl_attribs(struct ocrdma_dev *dev)
+{
+	int status = -ENOMEM;
+	struct ocrdma_dma_mem dma;
+	struct ocrdma_mqe *mqe;
+	struct ocrdma_get_ctrl_attribs_rsp *ctrl_attr_rsp;
+	struct mgmt_hba_attribs *hba_attribs;
+
+	mqe = ocrdma_alloc_mqe();
+	if (!mqe)
+		return status;
+	memset(mqe, 0, sizeof(*mqe));
+
+	dma.size = sizeof(struct ocrdma_get_ctrl_attribs_rsp);
+	dma.va	 = dma_alloc_coherent(&dev->nic_info.pdev->dev,
+					dma.size, &dma.pa, GFP_KERNEL);
+	if (!dma.va)
+		goto free_mqe;
+
+	mqe->hdr.pyld_len = dma.size;
+	mqe->hdr.spcl_sge_cnt_emb |=
+			(1 << OCRDMA_MQE_HDR_SGE_CNT_SHIFT) &
+			OCRDMA_MQE_HDR_SGE_CNT_MASK;
+	mqe->u.nonemb_req.sge[0].pa_lo = (u32) (dma.pa & 0xffffffff);
+	mqe->u.nonemb_req.sge[0].pa_hi = (u32) upper_32_bits(dma.pa);
+	mqe->u.nonemb_req.sge[0].len = dma.size;
+
+	memset(dma.va, 0, dma.size);
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)dma.va,
+			OCRDMA_CMD_GET_CTRL_ATTRIBUTES,
+			OCRDMA_SUBSYS_COMMON,
+			dma.size);
+
+	status = ocrdma_nonemb_mbx_cmd(dev, mqe, dma.va);
+	if (!status) {
+		ctrl_attr_rsp = (struct ocrdma_get_ctrl_attribs_rsp *)dma.va;
+		hba_attribs = &ctrl_attr_rsp->ctrl_attribs.hba_attribs;
+
+		dev->hba_port_num = hba_attribs->phy_port;
+		strncpy(dev->model_number,
+			hba_attribs->controller_model_number, 31);
+	}
+	dma_free_coherent(&dev->nic_info.pdev->dev, dma.size, dma.va, dma.pa);
+free_mqe:
+	kfree(mqe);
+	return status;
+}
+
+static int ocrdma_mbx_query_dev(struct ocrdma_dev *dev)
+{
+	int status = -ENOMEM;
+	struct ocrdma_mbx_query_config *rsp;
+	struct ocrdma_mqe *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_CONFIG, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_mbx_query_config *)cmd;
+	ocrdma_get_attr(dev, &dev->attr, rsp);
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_get_link_speed(struct ocrdma_dev *dev, u8 *lnk_speed)
+{
+	int status = -ENOMEM;
+	struct ocrdma_get_link_speed_rsp *rsp;
+	struct ocrdma_mqe *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
+				  sizeof(*cmd));
+	if (!cmd)
+		return status;
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
+			OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+
+	((struct ocrdma_mbx_hdr *)cmd->u.cmd)->rsvd_version = 0x1;
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+
+	rsp = (struct ocrdma_get_link_speed_rsp *)cmd;
+	*lnk_speed = rsp->phys_port_speed;
+
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+static int ocrdma_mbx_get_phy_info(struct ocrdma_dev *dev)
+{
+	int status = -ENOMEM;
+	struct ocrdma_mqe *cmd;
+	struct ocrdma_get_phy_info_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_PHY_DETAILS, sizeof(*cmd));
+	if (!cmd)
+		return status;
+
+	ocrdma_init_mch((struct ocrdma_mbx_hdr *)&cmd->u.cmd[0],
+			OCRDMA_CMD_PHY_DETAILS, OCRDMA_SUBSYS_COMMON,
+			sizeof(*cmd));
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+
+	rsp = (struct ocrdma_get_phy_info_rsp *)cmd;
+	dev->phy.phy_type = le16_to_cpu(rsp->phy_type);
+	dev->phy.auto_speeds_supported  =
+			le16_to_cpu(rsp->auto_speeds_supported);
+	dev->phy.fixed_speeds_supported =
+			le16_to_cpu(rsp->fixed_speeds_supported);
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_alloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
+{
+	int status = -ENOMEM;
+	struct ocrdma_alloc_pd *cmd;
+	struct ocrdma_alloc_pd_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_PD, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	if (pd->dpp_enabled)
+		cmd->enable_dpp_rsvd |= OCRDMA_ALLOC_PD_ENABLE_DPP;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_alloc_pd_rsp *)cmd;
+	pd->id = rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_PDID_MASK;
+	if (rsp->dpp_page_pdid & OCRDMA_ALLOC_PD_RSP_DPP) {
+		pd->dpp_enabled = true;
+		pd->dpp_page = rsp->dpp_page_pdid >>
+				OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT;
+	} else {
+		pd->dpp_enabled = false;
+		pd->num_dpp_qp = 0;
+	}
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_dealloc_pd(struct ocrdma_dev *dev, struct ocrdma_pd *pd)
+{
+	int status = -ENOMEM;
+	struct ocrdma_dealloc_pd *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_PD, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->id = pd->id;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	kfree(cmd);
+	return status;
+}
+
+static int ocrdma_build_q_conf(u32 *num_entries, int entry_size,
+			       int *num_pages, int *page_size)
+{
+	int i;
+	int mem_size;
+
+	*num_entries = roundup_pow_of_two(*num_entries);
+	mem_size = *num_entries * entry_size;
+	/* find the possible lowest possible multiplier */
+	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
+		if (mem_size <= (OCRDMA_Q_PAGE_BASE_SIZE << i))
+			break;
+	}
+	if (i >= OCRDMA_MAX_Q_PAGE_SIZE_CNT)
+		return -EINVAL;
+	mem_size = roundup(mem_size,
+		       ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES));
+	*num_pages =
+	    mem_size / ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
+	*page_size = ((OCRDMA_Q_PAGE_BASE_SIZE << i) / OCRDMA_MAX_Q_PAGES);
+	*num_entries = mem_size / entry_size;
+	return 0;
+}
+
+static int ocrdma_mbx_create_ah_tbl(struct ocrdma_dev *dev)
+{
+	int i;
+	int status = 0;
+	int max_ah;
+	struct ocrdma_create_ah_tbl *cmd;
+	struct ocrdma_create_ah_tbl_rsp *rsp;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	dma_addr_t pa;
+	struct ocrdma_pbe *pbes;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_AH_TBL, sizeof(*cmd));
+	if (!cmd)
+		return status;
+
+	max_ah = OCRDMA_MAX_AH;
+	dev->av_tbl.size = sizeof(struct ocrdma_av) * max_ah;
+
+	/* number of PBEs in PBL */
+	cmd->ah_conf = (OCRDMA_AH_TBL_PAGES <<
+				OCRDMA_CREATE_AH_NUM_PAGES_SHIFT) &
+				OCRDMA_CREATE_AH_NUM_PAGES_MASK;
+
+	/* page size */
+	for (i = 0; i < OCRDMA_MAX_Q_PAGE_SIZE_CNT; i++) {
+		if (PAGE_SIZE == (OCRDMA_MIN_Q_PAGE_SIZE << i))
+			break;
+	}
+	cmd->ah_conf |= (i << OCRDMA_CREATE_AH_PAGE_SIZE_SHIFT) &
+				OCRDMA_CREATE_AH_PAGE_SIZE_MASK;
+
+	/* ah_entry size */
+	cmd->ah_conf |= (sizeof(struct ocrdma_av) <<
+				OCRDMA_CREATE_AH_ENTRY_SIZE_SHIFT) &
+				OCRDMA_CREATE_AH_ENTRY_SIZE_MASK;
+
+	dev->av_tbl.pbl.va = dma_alloc_coherent(&pdev->dev, PAGE_SIZE,
+						&dev->av_tbl.pbl.pa,
+						GFP_KERNEL);
+	if (dev->av_tbl.pbl.va == NULL)
+		goto mem_err;
+
+	dev->av_tbl.va = dma_alloc_coherent(&pdev->dev, dev->av_tbl.size,
+					    &pa, GFP_KERNEL);
+	if (dev->av_tbl.va == NULL)
+		goto mem_err_ah;
+	dev->av_tbl.pa = pa;
+	dev->av_tbl.num_ah = max_ah;
+	memset(dev->av_tbl.va, 0, dev->av_tbl.size);
+
+	pbes = (struct ocrdma_pbe *)dev->av_tbl.pbl.va;
+	for (i = 0; i < dev->av_tbl.size / OCRDMA_MIN_Q_PAGE_SIZE; i++) {
+		pbes[i].pa_lo = (u32) (pa & 0xffffffff);
+		pbes[i].pa_hi = (u32) upper_32_bits(pa);
+		pa += PAGE_SIZE;
+	}
+	cmd->tbl_addr[0].lo = (u32)(dev->av_tbl.pbl.pa & 0xFFFFFFFF);
+	cmd->tbl_addr[0].hi = (u32)upper_32_bits(dev->av_tbl.pbl.pa);
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_create_ah_tbl_rsp *)cmd;
+	dev->av_tbl.ahid = rsp->ahid & 0xFFFF;
+	kfree(cmd);
+	return 0;
+
+mbx_err:
+	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
+			  dev->av_tbl.pa);
+	dev->av_tbl.va = NULL;
+mem_err_ah:
+	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
+			  dev->av_tbl.pbl.pa);
+	dev->av_tbl.pbl.va = NULL;
+	dev->av_tbl.size = 0;
+mem_err:
+	kfree(cmd);
+	return status;
+}
+
+static void ocrdma_mbx_delete_ah_tbl(struct ocrdma_dev *dev)
+{
+	struct ocrdma_delete_ah_tbl *cmd;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+
+	if (dev->av_tbl.va == NULL)
+		return;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_AH_TBL, sizeof(*cmd));
+	if (!cmd)
+		return;
+	cmd->ahid = dev->av_tbl.ahid;
+
+	ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	dma_free_coherent(&pdev->dev, dev->av_tbl.size, dev->av_tbl.va,
+			  dev->av_tbl.pa);
+	dma_free_coherent(&pdev->dev, PAGE_SIZE, dev->av_tbl.pbl.va,
+			  dev->av_tbl.pbl.pa);
+	kfree(cmd);
+}
+
+/* Multiple CQs uses the EQ. This routine returns least used
+ * EQ to associate with CQ. This will distributes the interrupt
+ * processing and CPU load to associated EQ, vector and so to that CPU.
+ */
+static u16 ocrdma_bind_eq(struct ocrdma_dev *dev)
+{
+	int i, selected_eq = 0, cq_cnt = 0;
+	u16 eq_id;
+
+	mutex_lock(&dev->dev_lock);
+	cq_cnt = dev->eq_tbl[0].cq_cnt;
+	eq_id = dev->eq_tbl[0].q.id;
+	/* find the EQ which is has the least number of
+	 * CQs associated with it.
+	 */
+	for (i = 0; i < dev->eq_cnt; i++) {
+		if (dev->eq_tbl[i].cq_cnt < cq_cnt) {
+			cq_cnt = dev->eq_tbl[i].cq_cnt;
+			eq_id = dev->eq_tbl[i].q.id;
+			selected_eq = i;
+		}
+	}
+	dev->eq_tbl[selected_eq].cq_cnt += 1;
+	mutex_unlock(&dev->dev_lock);
+	return eq_id;
+}
+
+static void ocrdma_unbind_eq(struct ocrdma_dev *dev, u16 eq_id)
+{
+	int i;
+
+	mutex_lock(&dev->dev_lock);
+	i = ocrdma_get_eq_table_index(dev, eq_id);
+	if (i == -EINVAL)
+		BUG();
+	dev->eq_tbl[i].cq_cnt -= 1;
+	mutex_unlock(&dev->dev_lock);
+}
+
+int ocrdma_mbx_create_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
+			 int entries, int dpp_cq, u16 pd_id)
+{
+	int status = -ENOMEM; int max_hw_cqe;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	struct ocrdma_create_cq *cmd;
+	struct ocrdma_create_cq_rsp *rsp;
+	u32 hw_pages, cqe_size, page_size, cqe_count;
+
+	if (entries > dev->attr.max_cqe) {
+		pr_err("%s(%d) max_cqe=0x%x, requester_cqe=0x%x\n",
+		       __func__, dev->id, dev->attr.max_cqe, entries);
+		return -EINVAL;
+	}
+	if (dpp_cq && (ocrdma_get_asic_type(dev) != OCRDMA_ASIC_GEN_SKH_R))
+		return -EINVAL;
+
+	if (dpp_cq) {
+		cq->max_hw_cqe = 1;
+		max_hw_cqe = 1;
+		cqe_size = OCRDMA_DPP_CQE_SIZE;
+		hw_pages = 1;
+	} else {
+		cq->max_hw_cqe = dev->attr.max_cqe;
+		max_hw_cqe = dev->attr.max_cqe;
+		cqe_size = sizeof(struct ocrdma_cqe);
+		hw_pages = OCRDMA_CREATE_CQ_MAX_PAGES;
+	}
+
+	cq->len = roundup(max_hw_cqe * cqe_size, OCRDMA_MIN_Q_PAGE_SIZE);
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_CQ, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	ocrdma_init_mch(&cmd->cmd.req, OCRDMA_CMD_CREATE_CQ,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+	cq->va = dma_alloc_coherent(&pdev->dev, cq->len, &cq->pa, GFP_KERNEL);
+	if (!cq->va) {
+		status = -ENOMEM;
+		goto mem_err;
+	}
+	memset(cq->va, 0, cq->len);
+	page_size = cq->len / hw_pages;
+	cmd->cmd.pgsz_pgcnt = (page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
+					OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT;
+	cmd->cmd.pgsz_pgcnt |= hw_pages;
+	cmd->cmd.ev_cnt_flags = OCRDMA_CREATE_CQ_DEF_FLAGS;
+
+	cq->eqn = ocrdma_bind_eq(dev);
+	cmd->cmd.req.rsvd_version = OCRDMA_CREATE_CQ_VER3;
+	cqe_count = cq->len / cqe_size;
+	cq->cqe_cnt = cqe_count;
+	if (cqe_count > 1024) {
+		/* Set cnt to 3 to indicate more than 1024 cq entries */
+		cmd->cmd.ev_cnt_flags |= (0x3 << OCRDMA_CREATE_CQ_CNT_SHIFT);
+	} else {
+		u8 count = 0;
+		switch (cqe_count) {
+		case 256:
+			count = 0;
+			break;
+		case 512:
+			count = 1;
+			break;
+		case 1024:
+			count = 2;
+			break;
+		default:
+			goto mbx_err;
+		}
+		cmd->cmd.ev_cnt_flags |= (count << OCRDMA_CREATE_CQ_CNT_SHIFT);
+	}
+	/* shared eq between all the consumer cqs. */
+	cmd->cmd.eqn = cq->eqn;
+	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
+		if (dpp_cq)
+			cmd->cmd.pgsz_pgcnt |= OCRDMA_CREATE_CQ_DPP <<
+				OCRDMA_CREATE_CQ_TYPE_SHIFT;
+		cq->phase_change = false;
+		cmd->cmd.cqe_count = (cq->len / cqe_size);
+	} else {
+		cmd->cmd.cqe_count = (cq->len / cqe_size) - 1;
+		cmd->cmd.ev_cnt_flags |= OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID;
+		cq->phase_change = true;
+	}
+
+	cmd->cmd.pd_id = pd_id; /* valid only for v3 */
+	ocrdma_build_q_pages(&cmd->cmd.pa[0], hw_pages, cq->pa, page_size);
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+
+	rsp = (struct ocrdma_create_cq_rsp *)cmd;
+	cq->id = (u16) (rsp->rsp.cq_id & OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK);
+	kfree(cmd);
+	return 0;
+mbx_err:
+	ocrdma_unbind_eq(dev, cq->eqn);
+	dma_free_coherent(&pdev->dev, cq->len, cq->va, cq->pa);
+mem_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_destroy_cq(struct ocrdma_dev *dev, struct ocrdma_cq *cq)
+{
+	int status = -ENOMEM;
+	struct ocrdma_destroy_cq *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_CQ, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	ocrdma_init_mch(&cmd->req, OCRDMA_CMD_DELETE_CQ,
+			OCRDMA_SUBSYS_COMMON, sizeof(*cmd));
+
+	cmd->bypass_flush_qid |=
+	    (cq->id << OCRDMA_DESTROY_CQ_QID_SHIFT) &
+	    OCRDMA_DESTROY_CQ_QID_MASK;
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	ocrdma_unbind_eq(dev, cq->eqn);
+	dma_free_coherent(&dev->nic_info.pdev->dev, cq->len, cq->va, cq->pa);
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
+			  u32 pdid, int addr_check)
+{
+	int status = -ENOMEM;
+	struct ocrdma_alloc_lkey *cmd;
+	struct ocrdma_alloc_lkey_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_ALLOC_LKEY, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->pdid = pdid;
+	cmd->pbl_sz_flags |= addr_check;
+	cmd->pbl_sz_flags |= (hwmr->fr_mr << OCRDMA_ALLOC_LKEY_FMR_SHIFT);
+	cmd->pbl_sz_flags |=
+	    (hwmr->remote_wr << OCRDMA_ALLOC_LKEY_REMOTE_WR_SHIFT);
+	cmd->pbl_sz_flags |=
+	    (hwmr->remote_rd << OCRDMA_ALLOC_LKEY_REMOTE_RD_SHIFT);
+	cmd->pbl_sz_flags |=
+	    (hwmr->local_wr << OCRDMA_ALLOC_LKEY_LOCAL_WR_SHIFT);
+	cmd->pbl_sz_flags |=
+	    (hwmr->remote_atomic << OCRDMA_ALLOC_LKEY_REMOTE_ATOMIC_SHIFT);
+	cmd->pbl_sz_flags |=
+	    (hwmr->num_pbls << OCRDMA_ALLOC_LKEY_PBL_SIZE_SHIFT);
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_alloc_lkey_rsp *)cmd;
+	hwmr->lkey = rsp->lrkey;
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_dealloc_lkey(struct ocrdma_dev *dev, int fr_mr, u32 lkey)
+{
+	int status = -ENOMEM;
+	struct ocrdma_dealloc_lkey *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DEALLOC_LKEY, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	cmd->lkey = lkey;
+	cmd->rsvd_frmr = fr_mr ? 1 : 0;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+static int ocrdma_mbx_reg_mr(struct ocrdma_dev *dev, struct ocrdma_hw_mr *hwmr,
+			     u32 pdid, u32 pbl_cnt, u32 pbe_size, u32 last)
+{
+	int status = -ENOMEM;
+	int i;
+	struct ocrdma_reg_nsmr *cmd;
+	struct ocrdma_reg_nsmr_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	cmd->num_pbl_pdid =
+	    pdid | (hwmr->num_pbls << OCRDMA_REG_NSMR_NUM_PBL_SHIFT);
+	cmd->fr_mr = hwmr->fr_mr;
+
+	cmd->flags_hpage_pbe_sz |= (hwmr->remote_wr <<
+				    OCRDMA_REG_NSMR_REMOTE_WR_SHIFT);
+	cmd->flags_hpage_pbe_sz |= (hwmr->remote_rd <<
+				    OCRDMA_REG_NSMR_REMOTE_RD_SHIFT);
+	cmd->flags_hpage_pbe_sz |= (hwmr->local_wr <<
+				    OCRDMA_REG_NSMR_LOCAL_WR_SHIFT);
+	cmd->flags_hpage_pbe_sz |= (hwmr->remote_atomic <<
+				    OCRDMA_REG_NSMR_REMOTE_ATOMIC_SHIFT);
+	cmd->flags_hpage_pbe_sz |= (hwmr->mw_bind <<
+				    OCRDMA_REG_NSMR_BIND_MEMWIN_SHIFT);
+	cmd->flags_hpage_pbe_sz |= (last << OCRDMA_REG_NSMR_LAST_SHIFT);
+
+	cmd->flags_hpage_pbe_sz |= (hwmr->pbe_size / OCRDMA_MIN_HPAGE_SIZE);
+	cmd->flags_hpage_pbe_sz |= (hwmr->pbl_size / OCRDMA_MIN_HPAGE_SIZE) <<
+					OCRDMA_REG_NSMR_HPAGE_SIZE_SHIFT;
+	cmd->totlen_low = hwmr->len;
+	cmd->totlen_high = upper_32_bits(hwmr->len);
+	cmd->fbo_low = (u32) (hwmr->fbo & 0xffffffff);
+	cmd->fbo_high = (u32) upper_32_bits(hwmr->fbo);
+	cmd->va_loaddr = (u32) hwmr->va;
+	cmd->va_hiaddr = (u32) upper_32_bits(hwmr->va);
+
+	for (i = 0; i < pbl_cnt; i++) {
+		cmd->pbl[i].lo = (u32) (hwmr->pbl_table[i].pa & 0xffffffff);
+		cmd->pbl[i].hi = upper_32_bits(hwmr->pbl_table[i].pa);
+	}
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_reg_nsmr_rsp *)cmd;
+	hwmr->lkey = rsp->lrkey;
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+static int ocrdma_mbx_reg_mr_cont(struct ocrdma_dev *dev,
+				  struct ocrdma_hw_mr *hwmr, u32 pbl_cnt,
+				  u32 pbl_offset, u32 last)
+{
+	int status = -ENOMEM;
+	int i;
+	struct ocrdma_reg_nsmr_cont *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_REGISTER_NSMR_CONT, sizeof(*cmd));
+	if (!cmd)
+		return -ENOMEM;
+	cmd->lrkey = hwmr->lkey;
+	cmd->num_pbl_offset = (pbl_cnt << OCRDMA_REG_NSMR_CONT_NUM_PBL_SHIFT) |
+	    (pbl_offset & OCRDMA_REG_NSMR_CONT_PBL_SHIFT_MASK);
+	cmd->last = last << OCRDMA_REG_NSMR_CONT_LAST_SHIFT;
+
+	for (i = 0; i < pbl_cnt; i++) {
+		cmd->pbl[i].lo =
+		    (u32) (hwmr->pbl_table[i + pbl_offset].pa & 0xffffffff);
+		cmd->pbl[i].hi =
+		    upper_32_bits(hwmr->pbl_table[i + pbl_offset].pa);
+	}
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_reg_mr(struct ocrdma_dev *dev,
+		  struct ocrdma_hw_mr *hwmr, u32 pdid, int acc)
+{
+	int status;
+	u32 last = 0;
+	u32 cur_pbl_cnt, pbl_offset;
+	u32 pending_pbl_cnt = hwmr->num_pbls;
+
+	pbl_offset = 0;
+	cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
+	if (cur_pbl_cnt == pending_pbl_cnt)
+		last = 1;
+
+	status = ocrdma_mbx_reg_mr(dev, hwmr, pdid,
+				   cur_pbl_cnt, hwmr->pbe_size, last);
+	if (status) {
+		pr_err("%s() status=%d\n", __func__, status);
+		return status;
+	}
+	/* if there is no more pbls to register then exit. */
+	if (last)
+		return 0;
+
+	while (!last) {
+		pbl_offset += cur_pbl_cnt;
+		pending_pbl_cnt -= cur_pbl_cnt;
+		cur_pbl_cnt = min(pending_pbl_cnt, MAX_OCRDMA_NSMR_PBL);
+		/* if we reach the end of the pbls, then need to set the last
+		 * bit, indicating no more pbls to register for this memory key.
+		 */
+		if (cur_pbl_cnt == pending_pbl_cnt)
+			last = 1;
+
+		status = ocrdma_mbx_reg_mr_cont(dev, hwmr, cur_pbl_cnt,
+						pbl_offset, last);
+		if (status)
+			break;
+	}
+	if (status)
+		pr_err("%s() err. status=%d\n", __func__, status);
+
+	return status;
+}
+
+bool ocrdma_is_qp_in_sq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
+{
+	struct ocrdma_qp *tmp;
+	bool found = false;
+	list_for_each_entry(tmp, &cq->sq_head, sq_entry) {
+		if (qp == tmp) {
+			found = true;
+			break;
+		}
+	}
+	return found;
+}
+
+bool ocrdma_is_qp_in_rq_flushlist(struct ocrdma_cq *cq, struct ocrdma_qp *qp)
+{
+	struct ocrdma_qp *tmp;
+	bool found = false;
+	list_for_each_entry(tmp, &cq->rq_head, rq_entry) {
+		if (qp == tmp) {
+			found = true;
+			break;
+		}
+	}
+	return found;
+}
+
+void ocrdma_flush_qp(struct ocrdma_qp *qp)
+{
+	bool found;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->dev->flush_q_lock, flags);
+	found = ocrdma_is_qp_in_sq_flushlist(qp->sq_cq, qp);
+	if (!found)
+		list_add_tail(&qp->sq_entry, &qp->sq_cq->sq_head);
+	if (!qp->srq) {
+		found = ocrdma_is_qp_in_rq_flushlist(qp->rq_cq, qp);
+		if (!found)
+			list_add_tail(&qp->rq_entry, &qp->rq_cq->rq_head);
+	}
+	spin_unlock_irqrestore(&qp->dev->flush_q_lock, flags);
+}
+
+static void ocrdma_init_hwq_ptr(struct ocrdma_qp *qp)
+{
+	qp->sq.head = 0;
+	qp->sq.tail = 0;
+	qp->rq.head = 0;
+	qp->rq.tail = 0;
+}
+
+int ocrdma_qp_state_change(struct ocrdma_qp *qp, enum ib_qp_state new_ib_state,
+			   enum ib_qp_state *old_ib_state)
+{
+	unsigned long flags;
+	int status = 0;
+	enum ocrdma_qp_state new_state;
+	new_state = get_ocrdma_qp_state(new_ib_state);
+
+	/* sync with wqe and rqe posting */
+	spin_lock_irqsave(&qp->q_lock, flags);
+
+	if (old_ib_state)
+		*old_ib_state = get_ibqp_state(qp->state);
+	if (new_state == qp->state) {
+		spin_unlock_irqrestore(&qp->q_lock, flags);
+		return 1;
+	}
+
+
+	if (new_state == OCRDMA_QPS_INIT) {
+		ocrdma_init_hwq_ptr(qp);
+		ocrdma_del_flush_qp(qp);
+	} else if (new_state == OCRDMA_QPS_ERR) {
+		ocrdma_flush_qp(qp);
+	}
+
+	qp->state = new_state;
+
+	spin_unlock_irqrestore(&qp->q_lock, flags);
+	return status;
+}
+
+static u32 ocrdma_set_create_qp_mbx_access_flags(struct ocrdma_qp *qp)
+{
+	u32 flags = 0;
+	if (qp->cap_flags & OCRDMA_QP_INB_RD)
+		flags |= OCRDMA_CREATE_QP_REQ_INB_RDEN_MASK;
+	if (qp->cap_flags & OCRDMA_QP_INB_WR)
+		flags |= OCRDMA_CREATE_QP_REQ_INB_WREN_MASK;
+	if (qp->cap_flags & OCRDMA_QP_MW_BIND)
+		flags |= OCRDMA_CREATE_QP_REQ_BIND_MEMWIN_MASK;
+	if (qp->cap_flags & OCRDMA_QP_LKEY0)
+		flags |= OCRDMA_CREATE_QP_REQ_ZERO_LKEYEN_MASK;
+	if (qp->cap_flags & OCRDMA_QP_FAST_REG)
+		flags |= OCRDMA_CREATE_QP_REQ_FMR_EN_MASK;
+	return flags;
+}
+
+static int ocrdma_set_create_qp_sq_cmd(struct ocrdma_create_qp_req *cmd,
+					struct ib_qp_init_attr *attrs,
+					struct ocrdma_qp *qp)
+{
+	int status;
+	u32 len, hw_pages, hw_page_size;
+	dma_addr_t pa;
+	struct ocrdma_dev *dev = qp->dev;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	u32 max_wqe_allocated;
+	u32 max_sges = attrs->cap.max_send_sge;
+
+	/* QP1 may exceed 127 */
+	max_wqe_allocated = min_t(u32, attrs->cap.max_send_wr + 1,
+				dev->attr.max_wqe);
+
+	status = ocrdma_build_q_conf(&max_wqe_allocated,
+		dev->attr.wqe_size, &hw_pages, &hw_page_size);
+	if (status) {
+		pr_err("%s() req. max_send_wr=0x%x\n", __func__,
+		       max_wqe_allocated);
+		return -EINVAL;
+	}
+	qp->sq.max_cnt = max_wqe_allocated;
+	len = (hw_pages * hw_page_size);
+
+	qp->sq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
+	if (!qp->sq.va)
+		return -EINVAL;
+	memset(qp->sq.va, 0, len);
+	qp->sq.len = len;
+	qp->sq.pa = pa;
+	qp->sq.entry_size = dev->attr.wqe_size;
+	ocrdma_build_q_pages(&cmd->wq_addr[0], hw_pages, pa, hw_page_size);
+
+	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
+				<< OCRDMA_CREATE_QP_REQ_SQ_PAGE_SIZE_SHIFT);
+	cmd->num_wq_rq_pages |= (hw_pages <<
+				 OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_SHIFT) &
+	    OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_MASK;
+	cmd->max_sge_send_write |= (max_sges <<
+				    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_SHIFT) &
+	    OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_MASK;
+	cmd->max_sge_send_write |= (max_sges <<
+				    OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_SHIFT) &
+					OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_MASK;
+	cmd->max_wqe_rqe |= (ilog2(qp->sq.max_cnt) <<
+			     OCRDMA_CREATE_QP_REQ_MAX_WQE_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_MAX_WQE_MASK;
+	cmd->wqe_rqe_size |= (dev->attr.wqe_size <<
+			      OCRDMA_CREATE_QP_REQ_WQE_SIZE_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_WQE_SIZE_MASK;
+	return 0;
+}
+
+static int ocrdma_set_create_qp_rq_cmd(struct ocrdma_create_qp_req *cmd,
+					struct ib_qp_init_attr *attrs,
+					struct ocrdma_qp *qp)
+{
+	int status;
+	u32 len, hw_pages, hw_page_size;
+	dma_addr_t pa = 0;
+	struct ocrdma_dev *dev = qp->dev;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	u32 max_rqe_allocated = attrs->cap.max_recv_wr + 1;
+
+	status = ocrdma_build_q_conf(&max_rqe_allocated, dev->attr.rqe_size,
+				     &hw_pages, &hw_page_size);
+	if (status) {
+		pr_err("%s() req. max_recv_wr=0x%x\n", __func__,
+		       attrs->cap.max_recv_wr + 1);
+		return status;
+	}
+	qp->rq.max_cnt = max_rqe_allocated;
+	len = (hw_pages * hw_page_size);
+
+	qp->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
+	if (!qp->rq.va)
+		return -ENOMEM;
+	memset(qp->rq.va, 0, len);
+	qp->rq.pa = pa;
+	qp->rq.len = len;
+	qp->rq.entry_size = dev->attr.rqe_size;
+
+	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
+	cmd->type_pgsz_pdn |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE) <<
+		OCRDMA_CREATE_QP_REQ_RQ_PAGE_SIZE_SHIFT);
+	cmd->num_wq_rq_pages |=
+	    (hw_pages << OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_SHIFT) &
+	    OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_MASK;
+	cmd->max_sge_recv_flags |= (attrs->cap.max_recv_sge <<
+				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_MASK;
+	cmd->max_wqe_rqe |= (ilog2(qp->rq.max_cnt) <<
+				OCRDMA_CREATE_QP_REQ_MAX_RQE_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_MAX_RQE_MASK;
+	cmd->wqe_rqe_size |= (dev->attr.rqe_size <<
+			OCRDMA_CREATE_QP_REQ_RQE_SIZE_SHIFT) &
+			OCRDMA_CREATE_QP_REQ_RQE_SIZE_MASK;
+	return 0;
+}
+
+static void ocrdma_set_create_qp_dpp_cmd(struct ocrdma_create_qp_req *cmd,
+					 struct ocrdma_pd *pd,
+					 struct ocrdma_qp *qp,
+					 u8 enable_dpp_cq, u16 dpp_cq_id)
+{
+	pd->num_dpp_qp--;
+	qp->dpp_enabled = true;
+	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
+	if (!enable_dpp_cq)
+		return;
+	cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK;
+	cmd->dpp_credits_cqid = dpp_cq_id;
+	cmd->dpp_credits_cqid |= OCRDMA_CREATE_QP_REQ_DPP_CREDIT_LIMIT <<
+					OCRDMA_CREATE_QP_REQ_DPP_CREDIT_SHIFT;
+}
+
+static int ocrdma_set_create_qp_ird_cmd(struct ocrdma_create_qp_req *cmd,
+					struct ocrdma_qp *qp)
+{
+	struct ocrdma_dev *dev = qp->dev;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	dma_addr_t pa = 0;
+	int ird_page_size = dev->attr.ird_page_size;
+	int ird_q_len = dev->attr.num_ird_pages * ird_page_size;
+	struct ocrdma_hdr_wqe *rqe;
+	int i  = 0;
+
+	if (dev->attr.ird == 0)
+		return 0;
+
+	qp->ird_q_va = dma_alloc_coherent(&pdev->dev, ird_q_len,
+					&pa, GFP_KERNEL);
+	if (!qp->ird_q_va)
+		return -ENOMEM;
+	memset(qp->ird_q_va, 0, ird_q_len);
+	ocrdma_build_q_pages(&cmd->ird_addr[0], dev->attr.num_ird_pages,
+			     pa, ird_page_size);
+	for (; i < ird_q_len / dev->attr.rqe_size; i++) {
+		rqe = (struct ocrdma_hdr_wqe *)(qp->ird_q_va +
+			(i * dev->attr.rqe_size));
+		rqe->cw = 0;
+		rqe->cw |= 2;
+		rqe->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
+		rqe->cw |= (8 << OCRDMA_WQE_SIZE_SHIFT);
+		rqe->cw |= (8 << OCRDMA_WQE_NXT_WQE_SIZE_SHIFT);
+	}
+	return 0;
+}
+
+static void ocrdma_get_create_qp_rsp(struct ocrdma_create_qp_rsp *rsp,
+				     struct ocrdma_qp *qp,
+				     struct ib_qp_init_attr *attrs,
+				     u16 *dpp_offset, u16 *dpp_credit_lmt)
+{
+	u32 max_wqe_allocated, max_rqe_allocated;
+	qp->id = rsp->qp_id & OCRDMA_CREATE_QP_RSP_QP_ID_MASK;
+	qp->rq.dbid = rsp->sq_rq_id & OCRDMA_CREATE_QP_RSP_RQ_ID_MASK;
+	qp->sq.dbid = rsp->sq_rq_id >> OCRDMA_CREATE_QP_RSP_SQ_ID_SHIFT;
+	qp->max_ird = rsp->max_ord_ird & OCRDMA_CREATE_QP_RSP_MAX_IRD_MASK;
+	qp->max_ord = (rsp->max_ord_ird >> OCRDMA_CREATE_QP_RSP_MAX_ORD_SHIFT);
+	qp->dpp_enabled = false;
+	if (rsp->dpp_response & OCRDMA_CREATE_QP_RSP_DPP_ENABLED_MASK) {
+		qp->dpp_enabled = true;
+		*dpp_credit_lmt = (rsp->dpp_response &
+				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_MASK) >>
+				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_SHIFT;
+		*dpp_offset = (rsp->dpp_response &
+				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_MASK) >>
+				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_SHIFT;
+	}
+	max_wqe_allocated =
+		rsp->max_wqe_rqe >> OCRDMA_CREATE_QP_RSP_MAX_WQE_SHIFT;
+	max_wqe_allocated = 1 << max_wqe_allocated;
+	max_rqe_allocated = 1 << ((u16)rsp->max_wqe_rqe);
+
+	qp->sq.max_cnt = max_wqe_allocated;
+	qp->sq.max_wqe_idx = max_wqe_allocated - 1;
+
+	if (!attrs->srq) {
+		qp->rq.max_cnt = max_rqe_allocated;
+		qp->rq.max_wqe_idx = max_rqe_allocated - 1;
+	}
+}
+
+int ocrdma_mbx_create_qp(struct ocrdma_qp *qp, struct ib_qp_init_attr *attrs,
+			 u8 enable_dpp_cq, u16 dpp_cq_id, u16 *dpp_offset,
+			 u16 *dpp_credit_lmt)
+{
+	int status = -ENOMEM;
+	u32 flags = 0;
+	struct ocrdma_dev *dev = qp->dev;
+	struct ocrdma_pd *pd = qp->pd;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	struct ocrdma_cq *cq;
+	struct ocrdma_create_qp_req *cmd;
+	struct ocrdma_create_qp_rsp *rsp;
+	int qptype;
+
+	switch (attrs->qp_type) {
+	case IB_QPT_GSI:
+		qptype = OCRDMA_QPT_GSI;
+		break;
+	case IB_QPT_RC:
+		qptype = OCRDMA_QPT_RC;
+		break;
+	case IB_QPT_UD:
+		qptype = OCRDMA_QPT_UD;
+		break;
+	default:
+		return -EINVAL;
+	}
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_QP, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->type_pgsz_pdn |= (qptype << OCRDMA_CREATE_QP_REQ_QPT_SHIFT) &
+						OCRDMA_CREATE_QP_REQ_QPT_MASK;
+	status = ocrdma_set_create_qp_sq_cmd(cmd, attrs, qp);
+	if (status)
+		goto sq_err;
+
+	if (attrs->srq) {
+		struct ocrdma_srq *srq = get_ocrdma_srq(attrs->srq);
+		cmd->max_sge_recv_flags |= OCRDMA_CREATE_QP_REQ_USE_SRQ_MASK;
+		cmd->rq_addr[0].lo = srq->id;
+		qp->srq = srq;
+	} else {
+		status = ocrdma_set_create_qp_rq_cmd(cmd, attrs, qp);
+		if (status)
+			goto rq_err;
+	}
+
+	status = ocrdma_set_create_qp_ird_cmd(cmd, qp);
+	if (status)
+		goto mbx_err;
+
+	cmd->type_pgsz_pdn |= (pd->id << OCRDMA_CREATE_QP_REQ_PD_ID_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_PD_ID_MASK;
+
+	flags = ocrdma_set_create_qp_mbx_access_flags(qp);
+
+	cmd->max_sge_recv_flags |= flags;
+	cmd->max_ord_ird |= (dev->attr.max_ord_per_qp <<
+			     OCRDMA_CREATE_QP_REQ_MAX_ORD_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_MAX_ORD_MASK;
+	cmd->max_ord_ird |= (dev->attr.max_ird_per_qp <<
+			     OCRDMA_CREATE_QP_REQ_MAX_IRD_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_MAX_IRD_MASK;
+	cq = get_ocrdma_cq(attrs->send_cq);
+	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_WQ_CQID_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_WQ_CQID_MASK;
+	qp->sq_cq = cq;
+	cq = get_ocrdma_cq(attrs->recv_cq);
+	cmd->wq_rq_cqid |= (cq->id << OCRDMA_CREATE_QP_REQ_RQ_CQID_SHIFT) &
+				OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK;
+	qp->rq_cq = cq;
+
+	if (pd->dpp_enabled && pd->num_dpp_qp) {
+		ocrdma_set_create_qp_dpp_cmd(cmd, pd, qp, enable_dpp_cq,
+					     dpp_cq_id);
+	}
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_create_qp_rsp *)cmd;
+	ocrdma_get_create_qp_rsp(rsp, qp, attrs, dpp_offset, dpp_credit_lmt);
+	qp->state = OCRDMA_QPS_RST;
+	kfree(cmd);
+	return 0;
+mbx_err:
+	if (qp->rq.va)
+		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
+rq_err:
+	pr_err("%s(%d) rq_err\n", __func__, dev->id);
+	dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
+sq_err:
+	pr_err("%s(%d) sq_err\n", __func__, dev->id);
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_query_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
+			struct ocrdma_qp_params *param)
+{
+	int status = -ENOMEM;
+	struct ocrdma_query_qp *cmd;
+	struct ocrdma_query_qp_rsp *rsp;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_QP, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->qp_id = qp->id;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_query_qp_rsp *)cmd;
+	memcpy(param, &rsp->params, sizeof(struct ocrdma_qp_params));
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+static int ocrdma_set_av_params(struct ocrdma_qp *qp,
+				struct ocrdma_modify_qp *cmd,
+				struct ib_qp_attr *attrs)
+{
+	int status;
+	struct ib_ah_attr *ah_attr = &attrs->ah_attr;
+	union ib_gid sgid, zgid;
+	u32 vlan_id;
+	u8 mac_addr[6];
+
+	if ((ah_attr->ah_flags & IB_AH_GRH) == 0)
+		return -EINVAL;
+	cmd->params.tclass_sq_psn |=
+	    (ah_attr->grh.traffic_class << OCRDMA_QP_PARAMS_TCLASS_SHIFT);
+	cmd->params.rnt_rc_sl_fl |=
+	    (ah_attr->grh.flow_label & OCRDMA_QP_PARAMS_FLOW_LABEL_MASK);
+	cmd->params.rnt_rc_sl_fl |= (ah_attr->sl << OCRDMA_QP_PARAMS_SL_SHIFT);
+	cmd->params.hop_lmt_rq_psn |=
+	    (ah_attr->grh.hop_limit << OCRDMA_QP_PARAMS_HOP_LMT_SHIFT);
+	cmd->flags |= OCRDMA_QP_PARA_FLOW_LBL_VALID;
+	memcpy(&cmd->params.dgid[0], &ah_attr->grh.dgid.raw[0],
+	       sizeof(cmd->params.dgid));
+	status = ocrdma_query_gid(&qp->dev->ibdev, 1,
+			ah_attr->grh.sgid_index, &sgid);
+	if (status)
+		return status;
+
+	memset(&zgid, 0, sizeof(zgid));
+	if (!memcmp(&sgid, &zgid, sizeof(zgid)))
+		return -EINVAL;
+
+	qp->sgid_idx = ah_attr->grh.sgid_index;
+	memcpy(&cmd->params.sgid[0], &sgid.raw[0], sizeof(cmd->params.sgid));
+	ocrdma_resolve_dmac(qp->dev, ah_attr, &mac_addr[0]);
+	cmd->params.dmac_b0_to_b3 = mac_addr[0] | (mac_addr[1] << 8) |
+				(mac_addr[2] << 16) | (mac_addr[3] << 24);
+	/* convert them to LE format. */
+	ocrdma_cpu_to_le32(&cmd->params.dgid[0], sizeof(cmd->params.dgid));
+	ocrdma_cpu_to_le32(&cmd->params.sgid[0], sizeof(cmd->params.sgid));
+	cmd->params.vlan_dmac_b4_to_b5 = mac_addr[4] | (mac_addr[5] << 8);
+	vlan_id = ah_attr->vlan_id;
+	if (vlan_id && (vlan_id < 0x1000)) {
+		cmd->params.vlan_dmac_b4_to_b5 |=
+		    vlan_id << OCRDMA_QP_PARAMS_VLAN_SHIFT;
+		cmd->flags |= OCRDMA_QP_PARA_VLAN_EN_VALID;
+	}
+	return 0;
+}
+
+static int ocrdma_set_qp_params(struct ocrdma_qp *qp,
+				struct ocrdma_modify_qp *cmd,
+				struct ib_qp_attr *attrs, int attr_mask)
+{
+	int status = 0;
+
+	if (attr_mask & IB_QP_PKEY_INDEX) {
+		cmd->params.path_mtu_pkey_indx |= (attrs->pkey_index &
+					    OCRDMA_QP_PARAMS_PKEY_INDEX_MASK);
+		cmd->flags |= OCRDMA_QP_PARA_PKEY_VALID;
+	}
+	if (attr_mask & IB_QP_QKEY) {
+		qp->qkey = attrs->qkey;
+		cmd->params.qkey = attrs->qkey;
+		cmd->flags |= OCRDMA_QP_PARA_QKEY_VALID;
+	}
+	if (attr_mask & IB_QP_AV) {
+		status = ocrdma_set_av_params(qp, cmd, attrs);
+		if (status)
+			return status;
+	} else if (qp->qp_type == IB_QPT_GSI || qp->qp_type == IB_QPT_UD) {
+		/* set the default mac address for UD, GSI QPs */
+		cmd->params.dmac_b0_to_b3 = qp->dev->nic_info.mac_addr[0] |
+			(qp->dev->nic_info.mac_addr[1] << 8) |
+			(qp->dev->nic_info.mac_addr[2] << 16) |
+			(qp->dev->nic_info.mac_addr[3] << 24);
+		cmd->params.vlan_dmac_b4_to_b5 = qp->dev->nic_info.mac_addr[4] |
+					(qp->dev->nic_info.mac_addr[5] << 8);
+	}
+	if ((attr_mask & IB_QP_EN_SQD_ASYNC_NOTIFY) &&
+	    attrs->en_sqd_async_notify) {
+		cmd->params.max_sge_recv_flags |=
+			OCRDMA_QP_PARAMS_FLAGS_SQD_ASYNC;
+		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
+	}
+	if (attr_mask & IB_QP_DEST_QPN) {
+		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->dest_qp_num &
+				OCRDMA_QP_PARAMS_DEST_QPN_MASK);
+		cmd->flags |= OCRDMA_QP_PARA_DST_QPN_VALID;
+	}
+	if (attr_mask & IB_QP_PATH_MTU) {
+		if (attrs->path_mtu < IB_MTU_256 ||
+		    attrs->path_mtu > IB_MTU_4096) {
+			status = -EINVAL;
+			goto pmtu_err;
+		}
+		cmd->params.path_mtu_pkey_indx |=
+		    (ib_mtu_enum_to_int(attrs->path_mtu) <<
+		     OCRDMA_QP_PARAMS_PATH_MTU_SHIFT) &
+		    OCRDMA_QP_PARAMS_PATH_MTU_MASK;
+		cmd->flags |= OCRDMA_QP_PARA_PMTU_VALID;
+	}
+	if (attr_mask & IB_QP_TIMEOUT) {
+		cmd->params.ack_to_rnr_rtc_dest_qpn |= attrs->timeout <<
+		    OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT;
+		cmd->flags |= OCRDMA_QP_PARA_ACK_TO_VALID;
+	}
+	if (attr_mask & IB_QP_RETRY_CNT) {
+		cmd->params.rnt_rc_sl_fl |= (attrs->retry_cnt <<
+				      OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT) &
+		    OCRDMA_QP_PARAMS_RETRY_CNT_MASK;
+		cmd->flags |= OCRDMA_QP_PARA_RETRY_CNT_VALID;
+	}
+	if (attr_mask & IB_QP_MIN_RNR_TIMER) {
+		cmd->params.rnt_rc_sl_fl |= (attrs->min_rnr_timer <<
+				      OCRDMA_QP_PARAMS_RNR_NAK_TIMER_SHIFT) &
+		    OCRDMA_QP_PARAMS_RNR_NAK_TIMER_MASK;
+		cmd->flags |= OCRDMA_QP_PARA_RNT_VALID;
+	}
+	if (attr_mask & IB_QP_RNR_RETRY) {
+		cmd->params.ack_to_rnr_rtc_dest_qpn |= (attrs->rnr_retry <<
+			OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT)
+			& OCRDMA_QP_PARAMS_RNR_RETRY_CNT_MASK;
+		cmd->flags |= OCRDMA_QP_PARA_RRC_VALID;
+	}
+	if (attr_mask & IB_QP_SQ_PSN) {
+		cmd->params.tclass_sq_psn |= (attrs->sq_psn & 0x00ffffff);
+		cmd->flags |= OCRDMA_QP_PARA_SQPSN_VALID;
+	}
+	if (attr_mask & IB_QP_RQ_PSN) {
+		cmd->params.hop_lmt_rq_psn |= (attrs->rq_psn & 0x00ffffff);
+		cmd->flags |= OCRDMA_QP_PARA_RQPSN_VALID;
+	}
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC) {
+		if (attrs->max_rd_atomic > qp->dev->attr.max_ord_per_qp) {
+			status = -EINVAL;
+			goto pmtu_err;
+		}
+		qp->max_ord = attrs->max_rd_atomic;
+		cmd->flags |= OCRDMA_QP_PARA_MAX_ORD_VALID;
+	}
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC) {
+		if (attrs->max_dest_rd_atomic > qp->dev->attr.max_ird_per_qp) {
+			status = -EINVAL;
+			goto pmtu_err;
+		}
+		qp->max_ird = attrs->max_dest_rd_atomic;
+		cmd->flags |= OCRDMA_QP_PARA_MAX_IRD_VALID;
+	}
+	cmd->params.max_ord_ird = (qp->max_ord <<
+				OCRDMA_QP_PARAMS_MAX_ORD_SHIFT) |
+				(qp->max_ird & OCRDMA_QP_PARAMS_MAX_IRD_MASK);
+pmtu_err:
+	return status;
+}
+
+int ocrdma_mbx_modify_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
+			 struct ib_qp_attr *attrs, int attr_mask)
+{
+	int status = -ENOMEM;
+	struct ocrdma_modify_qp *cmd;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_QP, sizeof(*cmd));
+	if (!cmd)
+		return status;
+
+	cmd->params.id = qp->id;
+	cmd->flags = 0;
+	if (attr_mask & IB_QP_STATE) {
+		cmd->params.max_sge_recv_flags |=
+		    (get_ocrdma_qp_state(attrs->qp_state) <<
+		     OCRDMA_QP_PARAMS_STATE_SHIFT) &
+		    OCRDMA_QP_PARAMS_STATE_MASK;
+		cmd->flags |= OCRDMA_QP_PARA_QPS_VALID;
+	} else {
+		cmd->params.max_sge_recv_flags |=
+		    (qp->state << OCRDMA_QP_PARAMS_STATE_SHIFT) &
+		    OCRDMA_QP_PARAMS_STATE_MASK;
+	}
+
+	status = ocrdma_set_qp_params(qp, cmd, attrs, attr_mask);
+	if (status)
+		goto mbx_err;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+
+mbx_err:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_destroy_qp(struct ocrdma_dev *dev, struct ocrdma_qp *qp)
+{
+	int status = -ENOMEM;
+	struct ocrdma_destroy_qp *cmd;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_QP, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->qp_id = qp->id;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+
+mbx_err:
+	kfree(cmd);
+	if (qp->sq.va)
+		dma_free_coherent(&pdev->dev, qp->sq.len, qp->sq.va, qp->sq.pa);
+	if (!qp->srq && qp->rq.va)
+		dma_free_coherent(&pdev->dev, qp->rq.len, qp->rq.va, qp->rq.pa);
+	if (qp->dpp_enabled)
+		qp->pd->num_dpp_qp++;
+	return status;
+}
+
+int ocrdma_mbx_create_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq,
+			  struct ib_srq_init_attr *srq_attr,
+			  struct ocrdma_pd *pd)
+{
+	int status = -ENOMEM;
+	int hw_pages, hw_page_size;
+	int len;
+	struct ocrdma_create_srq_rsp *rsp;
+	struct ocrdma_create_srq *cmd;
+	dma_addr_t pa;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	u32 max_rqe_allocated;
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_CREATE_SRQ, sizeof(*cmd));
+	if (!cmd)
+		return status;
+
+	cmd->pgsz_pdid = pd->id & OCRDMA_CREATE_SRQ_PD_ID_MASK;
+	max_rqe_allocated = srq_attr->attr.max_wr + 1;
+	status = ocrdma_build_q_conf(&max_rqe_allocated,
+				dev->attr.rqe_size,
+				&hw_pages, &hw_page_size);
+	if (status) {
+		pr_err("%s() req. max_wr=0x%x\n", __func__,
+		       srq_attr->attr.max_wr);
+		status = -EINVAL;
+		goto ret;
+	}
+	len = hw_pages * hw_page_size;
+	srq->rq.va = dma_alloc_coherent(&pdev->dev, len, &pa, GFP_KERNEL);
+	if (!srq->rq.va) {
+		status = -ENOMEM;
+		goto ret;
+	}
+	ocrdma_build_q_pages(&cmd->rq_addr[0], hw_pages, pa, hw_page_size);
+
+	srq->rq.entry_size = dev->attr.rqe_size;
+	srq->rq.pa = pa;
+	srq->rq.len = len;
+	srq->rq.max_cnt = max_rqe_allocated;
+
+	cmd->max_sge_rqe = ilog2(max_rqe_allocated);
+	cmd->max_sge_rqe |= srq_attr->attr.max_sge <<
+				OCRDMA_CREATE_SRQ_MAX_SGE_RECV_SHIFT;
+
+	cmd->pgsz_pdid |= (ilog2(hw_page_size / OCRDMA_MIN_Q_PAGE_SIZE)
+		<< OCRDMA_CREATE_SRQ_PG_SZ_SHIFT);
+	cmd->pages_rqe_sz |= (dev->attr.rqe_size
+		<< OCRDMA_CREATE_SRQ_RQE_SIZE_SHIFT)
+		& OCRDMA_CREATE_SRQ_RQE_SIZE_MASK;
+	cmd->pages_rqe_sz |= hw_pages << OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_SHIFT;
+
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status)
+		goto mbx_err;
+	rsp = (struct ocrdma_create_srq_rsp *)cmd;
+	srq->id = rsp->id;
+	srq->rq.dbid = rsp->id;
+	max_rqe_allocated = ((rsp->max_sge_rqe_allocated &
+		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_MASK) >>
+		OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_SHIFT);
+	max_rqe_allocated = (1 << max_rqe_allocated);
+	srq->rq.max_cnt = max_rqe_allocated;
+	srq->rq.max_wqe_idx = max_rqe_allocated - 1;
+	srq->rq.max_sges = (rsp->max_sge_rqe_allocated &
+		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_MASK) >>
+		OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_SHIFT;
+	goto ret;
+mbx_err:
+	dma_free_coherent(&pdev->dev, srq->rq.len, srq->rq.va, pa);
+ret:
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_modify_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
+{
+	int status = -ENOMEM;
+	struct ocrdma_modify_srq *cmd;
+	struct ocrdma_pd *pd = srq->pd;
+	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_MODIFY_SRQ, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->id = srq->id;
+	cmd->limit_max_rqe |= srq_attr->srq_limit <<
+	    OCRDMA_MODIFY_SRQ_LIMIT_SHIFT;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_query_srq(struct ocrdma_srq *srq, struct ib_srq_attr *srq_attr)
+{
+	int status = -ENOMEM;
+	struct ocrdma_query_srq *cmd;
+	struct ocrdma_dev *dev = get_ocrdma_dev(srq->ibsrq.device);
+
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_QUERY_SRQ, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->id = srq->rq.dbid;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (status == 0) {
+		struct ocrdma_query_srq_rsp *rsp =
+		    (struct ocrdma_query_srq_rsp *)cmd;
+		srq_attr->max_sge =
+		    rsp->srq_lmt_max_sge &
+		    OCRDMA_QUERY_SRQ_RSP_MAX_SGE_RECV_MASK;
+		srq_attr->max_wr =
+		    rsp->max_rqe_pdid >> OCRDMA_QUERY_SRQ_RSP_MAX_RQE_SHIFT;
+		srq_attr->srq_limit = rsp->srq_lmt_max_sge >>
+		    OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_SHIFT;
+	}
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_mbx_destroy_srq(struct ocrdma_dev *dev, struct ocrdma_srq *srq)
+{
+	int status = -ENOMEM;
+	struct ocrdma_destroy_srq *cmd;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	cmd = ocrdma_init_emb_mqe(OCRDMA_CMD_DELETE_SRQ, sizeof(*cmd));
+	if (!cmd)
+		return status;
+	cmd->id = srq->id;
+	status = ocrdma_mbx_cmd(dev, (struct ocrdma_mqe *)cmd);
+	if (srq->rq.va)
+		dma_free_coherent(&pdev->dev, srq->rq.len,
+				  srq->rq.va, srq->rq.pa);
+	kfree(cmd);
+	return status;
+}
+
+int ocrdma_alloc_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
+{
+	int i;
+	int status = -EINVAL;
+	struct ocrdma_av *av;
+	unsigned long flags;
+
+	av = dev->av_tbl.va;
+	spin_lock_irqsave(&dev->av_tbl.lock, flags);
+	for (i = 0; i < dev->av_tbl.num_ah; i++) {
+		if (av->valid == 0) {
+			av->valid = OCRDMA_AV_VALID;
+			ah->av = av;
+			ah->id = i;
+			status = 0;
+			break;
+		}
+		av++;
+	}
+	if (i == dev->av_tbl.num_ah)
+		status = -EAGAIN;
+	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
+	return status;
+}
+
+int ocrdma_free_av(struct ocrdma_dev *dev, struct ocrdma_ah *ah)
+{
+	unsigned long flags;
+	spin_lock_irqsave(&dev->av_tbl.lock, flags);
+	ah->av->valid = 0;
+	spin_unlock_irqrestore(&dev->av_tbl.lock, flags);
+	return 0;
+}
+
+static int ocrdma_create_eqs(struct ocrdma_dev *dev)
+{
+	int num_eq, i, status = 0;
+	int irq;
+	unsigned long flags = 0;
+
+	num_eq = dev->nic_info.msix.num_vectors -
+			dev->nic_info.msix.start_vector;
+	if (dev->nic_info.intr_mode == BE_INTERRUPT_MODE_INTX) {
+		num_eq = 1;
+		flags = IRQF_SHARED;
+	} else {
+		num_eq = min_t(u32, num_eq, num_online_cpus());
+	}
+
+	if (!num_eq)
+		return -EINVAL;
+
+	dev->eq_tbl = kzalloc(sizeof(struct ocrdma_eq) * num_eq, GFP_KERNEL);
+	if (!dev->eq_tbl)
+		return -ENOMEM;
+
+	for (i = 0; i < num_eq; i++) {
+		status = ocrdma_create_eq(dev, &dev->eq_tbl[i],
+					OCRDMA_EQ_LEN);
+		if (status) {
+			status = -EINVAL;
+			break;
+		}
+		sprintf(dev->eq_tbl[i].irq_name, "ocrdma%d-%d",
+			dev->id, i);
+		irq = ocrdma_get_irq(dev, &dev->eq_tbl[i]);
+		status = request_irq(irq, ocrdma_irq_handler, flags,
+				     dev->eq_tbl[i].irq_name,
+				     &dev->eq_tbl[i]);
+		if (status)
+			goto done;
+		dev->eq_cnt += 1;
+	}
+	/* one eq is sufficient for data path to work */
+	return 0;
+done:
+	ocrdma_destroy_eqs(dev);
+	return status;
+}
+
+int ocrdma_init_hw(struct ocrdma_dev *dev)
+{
+	int status;
+
+	/* create the eqs  */
+	status = ocrdma_create_eqs(dev);
+	if (status)
+		goto qpeq_err;
+	status = ocrdma_create_mq(dev);
+	if (status)
+		goto mq_err;
+	status = ocrdma_mbx_query_fw_config(dev);
+	if (status)
+		goto conf_err;
+	status = ocrdma_mbx_query_dev(dev);
+	if (status)
+		goto conf_err;
+	status = ocrdma_mbx_query_fw_ver(dev);
+	if (status)
+		goto conf_err;
+	status = ocrdma_mbx_create_ah_tbl(dev);
+	if (status)
+		goto conf_err;
+	status = ocrdma_mbx_get_phy_info(dev);
+	if (status)
+		goto conf_err;
+	status = ocrdma_mbx_get_ctrl_attribs(dev);
+	if (status)
+		goto conf_err;
+
+	return 0;
+
+conf_err:
+	ocrdma_destroy_mq(dev);
+mq_err:
+	ocrdma_destroy_eqs(dev);
+qpeq_err:
+	pr_err("%s() status=%d\n", __func__, status);
+	return status;
+}
+
+void ocrdma_cleanup_hw(struct ocrdma_dev *dev)
+{
+	ocrdma_mbx_delete_ah_tbl(dev);
+
+	/* cleanup the eqs */
+	ocrdma_destroy_eqs(dev);
+
+	/* cleanup the control path */
+	ocrdma_destroy_mq(dev);
+}
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_hw.h b/drivers/infiniband/hw/ocrdma/ocrdma_hw.h
new file mode 100644
index 00000000000..e513f729314
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_hw.h
@@ -0,0 +1,138 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) CNA Adapters.              *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_HW_H__
+#define __OCRDMA_HW_H__
+
+#include "ocrdma_sli.h"
+
+static inline void ocrdma_cpu_to_le32(void *dst, u32 len)
+{
+#ifdef __BIG_ENDIAN
+	int i = 0;
+	u32 *src_ptr = dst;
+	u32 *dst_ptr = dst;
+	for (; i < (len / 4); i++)
+		*(dst_ptr + i) = cpu_to_le32p(src_ptr + i);
+#endif
+}
+
+static inline void ocrdma_le32_to_cpu(void *dst, u32 len)
+{
+#ifdef __BIG_ENDIAN
+	int i = 0;
+	u32 *src_ptr = dst;
+	u32 *dst_ptr = dst;
+	for (; i < (len / sizeof(u32)); i++)
+		*(dst_ptr + i) = le32_to_cpu(*(src_ptr + i));
+#endif
+}
+
+static inline void ocrdma_copy_cpu_to_le32(void *dst, void *src, u32 len)
+{
+#ifdef __BIG_ENDIAN
+	int i = 0;
+	u32 *src_ptr = src;
+	u32 *dst_ptr = dst;
+	for (; i < (len / sizeof(u32)); i++)
+		*(dst_ptr + i) = cpu_to_le32p(src_ptr + i);
+#else
+	memcpy(dst, src, len);
+#endif
+}
+
+static inline void ocrdma_copy_le32_to_cpu(void *dst, void *src, u32 len)
+{
+#ifdef __BIG_ENDIAN
+	int i = 0;
+	u32 *src_ptr = src;
+	u32 *dst_ptr = dst;
+	for (; i < len / sizeof(u32); i++)
+		*(dst_ptr + i) = le32_to_cpu(*(src_ptr + i));
+#else
+	memcpy(dst, src, len);
+#endif
+}
+
+static inline u64 ocrdma_get_db_addr(struct ocrdma_dev *dev, u32 pdid)
+{
+	return dev->nic_info.unmapped_db + (pdid * dev->nic_info.db_page_size);
+}
+
+int ocrdma_init_hw(struct ocrdma_dev *);
+void ocrdma_cleanup_hw(struct ocrdma_dev *);
+
+enum ib_qp_state get_ibqp_state(enum ocrdma_qp_state qps);
+void ocrdma_ring_cq_db(struct ocrdma_dev *, u16 cq_id, bool armed,
+		       bool solicited, u16 cqe_popped);
+
+/* verbs specific mailbox commands */
+int ocrdma_mbx_get_link_speed(struct ocrdma_dev *dev, u8 *lnk_speed);
+int ocrdma_query_config(struct ocrdma_dev *,
+			struct ocrdma_mbx_query_config *config);
+
+int ocrdma_mbx_alloc_pd(struct ocrdma_dev *, struct ocrdma_pd *);
+int ocrdma_mbx_dealloc_pd(struct ocrdma_dev *, struct ocrdma_pd *);
+
+int ocrdma_mbx_alloc_lkey(struct ocrdma_dev *, struct ocrdma_hw_mr *hwmr,
+			  u32 pd_id, int addr_check);
+int ocrdma_mbx_dealloc_lkey(struct ocrdma_dev *, int fmr, u32 lkey);
+
+int ocrdma_reg_mr(struct ocrdma_dev *, struct ocrdma_hw_mr *hwmr,
+			u32 pd_id, int acc);
+int ocrdma_mbx_create_cq(struct ocrdma_dev *, struct ocrdma_cq *,
+				int entries, int dpp_cq, u16 pd_id);
+int ocrdma_mbx_destroy_cq(struct ocrdma_dev *, struct ocrdma_cq *);
+
+int ocrdma_mbx_create_qp(struct ocrdma_qp *, struct ib_qp_init_attr *attrs,
+			 u8 enable_dpp_cq, u16 dpp_cq_id, u16 *dpp_offset,
+			 u16 *dpp_credit_lmt);
+int ocrdma_mbx_modify_qp(struct ocrdma_dev *, struct ocrdma_qp *,
+			 struct ib_qp_attr *attrs, int attr_mask);
+int ocrdma_mbx_query_qp(struct ocrdma_dev *, struct ocrdma_qp *,
+			struct ocrdma_qp_params *param);
+int ocrdma_mbx_destroy_qp(struct ocrdma_dev *, struct ocrdma_qp *);
+int ocrdma_mbx_create_srq(struct ocrdma_dev *, struct ocrdma_srq *,
+			  struct ib_srq_init_attr *,
+			  struct ocrdma_pd *);
+int ocrdma_mbx_modify_srq(struct ocrdma_srq *, struct ib_srq_attr *);
+int ocrdma_mbx_query_srq(struct ocrdma_srq *, struct ib_srq_attr *);
+int ocrdma_mbx_destroy_srq(struct ocrdma_dev *, struct ocrdma_srq *);
+
+int ocrdma_alloc_av(struct ocrdma_dev *, struct ocrdma_ah *);
+int ocrdma_free_av(struct ocrdma_dev *, struct ocrdma_ah *);
+
+int ocrdma_qp_state_change(struct ocrdma_qp *, enum ib_qp_state new_state,
+			    enum ib_qp_state *old_ib_state);
+bool ocrdma_is_qp_in_sq_flushlist(struct ocrdma_cq *, struct ocrdma_qp *);
+bool ocrdma_is_qp_in_rq_flushlist(struct ocrdma_cq *, struct ocrdma_qp *);
+void ocrdma_flush_qp(struct ocrdma_qp *);
+int ocrdma_get_irq(struct ocrdma_dev *dev, struct ocrdma_eq *eq);
+
+int ocrdma_mbx_rdma_stats(struct ocrdma_dev *, bool reset);
+char *port_speed_string(struct ocrdma_dev *dev);
+#endif				/* __OCRDMA_HW_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_main.c b/drivers/infiniband/hw/ocrdma/ocrdma_main.c
new file mode 100644
index 00000000000..7c504e07974
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_main.c
@@ -0,0 +1,585 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#include <linux/module.h>
+#include <linux/idr.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include <linux/netdevice.h>
+#include <net/addrconf.h>
+
+#include "ocrdma.h"
+#include "ocrdma_verbs.h"
+#include "ocrdma_ah.h"
+#include "be_roce.h"
+#include "ocrdma_hw.h"
+#include "ocrdma_stats.h"
+#include "ocrdma_abi.h"
+
+MODULE_VERSION(OCRDMA_ROCE_DRV_VERSION);
+MODULE_DESCRIPTION(OCRDMA_ROCE_DRV_DESC " " OCRDMA_ROCE_DRV_VERSION);
+MODULE_AUTHOR("Emulex Corporation");
+MODULE_LICENSE("GPL");
+
+static LIST_HEAD(ocrdma_dev_list);
+static DEFINE_SPINLOCK(ocrdma_devlist_lock);
+static DEFINE_IDR(ocrdma_dev_id);
+
+static union ib_gid ocrdma_zero_sgid;
+
+void ocrdma_get_guid(struct ocrdma_dev *dev, u8 *guid)
+{
+	u8 mac_addr[6];
+
+	memcpy(&mac_addr[0], &dev->nic_info.mac_addr[0], ETH_ALEN);
+	guid[0] = mac_addr[0] ^ 2;
+	guid[1] = mac_addr[1];
+	guid[2] = mac_addr[2];
+	guid[3] = 0xff;
+	guid[4] = 0xfe;
+	guid[5] = mac_addr[3];
+	guid[6] = mac_addr[4];
+	guid[7] = mac_addr[5];
+}
+
+static bool ocrdma_add_sgid(struct ocrdma_dev *dev, union ib_gid *new_sgid)
+{
+	int i;
+	unsigned long flags;
+
+	memset(&ocrdma_zero_sgid, 0, sizeof(union ib_gid));
+
+
+	spin_lock_irqsave(&dev->sgid_lock, flags);
+	for (i = 0; i < OCRDMA_MAX_SGID; i++) {
+		if (!memcmp(&dev->sgid_tbl[i], &ocrdma_zero_sgid,
+			    sizeof(union ib_gid))) {
+			/* found free entry */
+			memcpy(&dev->sgid_tbl[i], new_sgid,
+			       sizeof(union ib_gid));
+			spin_unlock_irqrestore(&dev->sgid_lock, flags);
+			return true;
+		} else if (!memcmp(&dev->sgid_tbl[i], new_sgid,
+				   sizeof(union ib_gid))) {
+			/* entry already present, no addition is required. */
+			spin_unlock_irqrestore(&dev->sgid_lock, flags);
+			return false;
+		}
+	}
+	spin_unlock_irqrestore(&dev->sgid_lock, flags);
+	return false;
+}
+
+static bool ocrdma_del_sgid(struct ocrdma_dev *dev, union ib_gid *sgid)
+{
+	int found = false;
+	int i;
+	unsigned long flags;
+
+
+	spin_lock_irqsave(&dev->sgid_lock, flags);
+	/* first is default sgid, which cannot be deleted. */
+	for (i = 1; i < OCRDMA_MAX_SGID; i++) {
+		if (!memcmp(&dev->sgid_tbl[i], sgid, sizeof(union ib_gid))) {
+			/* found matching entry */
+			memset(&dev->sgid_tbl[i], 0, sizeof(union ib_gid));
+			found = true;
+			break;
+		}
+	}
+	spin_unlock_irqrestore(&dev->sgid_lock, flags);
+	return found;
+}
+
+static int ocrdma_addr_event(unsigned long event, struct net_device *netdev,
+			     union ib_gid *gid)
+{
+	struct ib_event gid_event;
+	struct ocrdma_dev *dev;
+	bool found = false;
+	bool updated = false;
+	bool is_vlan = false;
+
+	is_vlan = netdev->priv_flags & IFF_802_1Q_VLAN;
+	if (is_vlan)
+		netdev = rdma_vlan_dev_real_dev(netdev);
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(dev, &ocrdma_dev_list, entry) {
+		if (dev->nic_info.netdev == netdev) {
+			found = true;
+			break;
+		}
+	}
+	rcu_read_unlock();
+
+	if (!found)
+		return NOTIFY_DONE;
+
+	mutex_lock(&dev->dev_lock);
+	switch (event) {
+	case NETDEV_UP:
+		updated = ocrdma_add_sgid(dev, gid);
+		break;
+	case NETDEV_DOWN:
+		updated = ocrdma_del_sgid(dev, gid);
+		break;
+	default:
+		break;
+	}
+	if (updated) {
+		/* GID table updated, notify the consumers about it */
+		gid_event.device = &dev->ibdev;
+		gid_event.element.port_num = 1;
+		gid_event.event = IB_EVENT_GID_CHANGE;
+		ib_dispatch_event(&gid_event);
+	}
+	mutex_unlock(&dev->dev_lock);
+	return NOTIFY_OK;
+}
+
+static int ocrdma_inetaddr_event(struct notifier_block *notifier,
+				  unsigned long event, void *ptr)
+{
+	struct in_ifaddr *ifa = ptr;
+	union ib_gid gid;
+	struct net_device *netdev = ifa->ifa_dev->dev;
+
+	ipv6_addr_set_v4mapped(ifa->ifa_address, (struct in6_addr *)&gid);
+	return ocrdma_addr_event(event, netdev, &gid);
+}
+
+static struct notifier_block ocrdma_inetaddr_notifier = {
+	.notifier_call = ocrdma_inetaddr_event
+};
+
+#if IS_ENABLED(CONFIG_IPV6)
+
+static int ocrdma_inet6addr_event(struct notifier_block *notifier,
+				  unsigned long event, void *ptr)
+{
+	struct inet6_ifaddr *ifa = (struct inet6_ifaddr *)ptr;
+	union  ib_gid *gid = (union ib_gid *)&ifa->addr;
+	struct net_device *netdev = ifa->idev->dev;
+	return ocrdma_addr_event(event, netdev, gid);
+}
+
+static struct notifier_block ocrdma_inet6addr_notifier = {
+	.notifier_call = ocrdma_inet6addr_event
+};
+
+#endif /* IPV6 and VLAN */
+
+static enum rdma_link_layer ocrdma_link_layer(struct ib_device *device,
+					      u8 port_num)
+{
+	return IB_LINK_LAYER_ETHERNET;
+}
+
+static int ocrdma_register_device(struct ocrdma_dev *dev)
+{
+	strlcpy(dev->ibdev.name, "ocrdma%d", IB_DEVICE_NAME_MAX);
+	ocrdma_get_guid(dev, (u8 *)&dev->ibdev.node_guid);
+	memcpy(dev->ibdev.node_desc, OCRDMA_NODE_DESC,
+	       sizeof(OCRDMA_NODE_DESC));
+	dev->ibdev.owner = THIS_MODULE;
+	dev->ibdev.uverbs_abi_ver = OCRDMA_ABI_VERSION;
+	dev->ibdev.uverbs_cmd_mask =
+	    OCRDMA_UVERBS(GET_CONTEXT) |
+	    OCRDMA_UVERBS(QUERY_DEVICE) |
+	    OCRDMA_UVERBS(QUERY_PORT) |
+	    OCRDMA_UVERBS(ALLOC_PD) |
+	    OCRDMA_UVERBS(DEALLOC_PD) |
+	    OCRDMA_UVERBS(REG_MR) |
+	    OCRDMA_UVERBS(DEREG_MR) |
+	    OCRDMA_UVERBS(CREATE_COMP_CHANNEL) |
+	    OCRDMA_UVERBS(CREATE_CQ) |
+	    OCRDMA_UVERBS(RESIZE_CQ) |
+	    OCRDMA_UVERBS(DESTROY_CQ) |
+	    OCRDMA_UVERBS(REQ_NOTIFY_CQ) |
+	    OCRDMA_UVERBS(CREATE_QP) |
+	    OCRDMA_UVERBS(MODIFY_QP) |
+	    OCRDMA_UVERBS(QUERY_QP) |
+	    OCRDMA_UVERBS(DESTROY_QP) |
+	    OCRDMA_UVERBS(POLL_CQ) |
+	    OCRDMA_UVERBS(POST_SEND) |
+	    OCRDMA_UVERBS(POST_RECV);
+
+	dev->ibdev.uverbs_cmd_mask |=
+	    OCRDMA_UVERBS(CREATE_AH) |
+	     OCRDMA_UVERBS(MODIFY_AH) |
+	     OCRDMA_UVERBS(QUERY_AH) |
+	     OCRDMA_UVERBS(DESTROY_AH);
+
+	dev->ibdev.node_type = RDMA_NODE_IB_CA;
+	dev->ibdev.phys_port_cnt = 1;
+	dev->ibdev.num_comp_vectors = 1;
+
+	/* mandatory verbs. */
+	dev->ibdev.query_device = ocrdma_query_device;
+	dev->ibdev.query_port = ocrdma_query_port;
+	dev->ibdev.modify_port = ocrdma_modify_port;
+	dev->ibdev.query_gid = ocrdma_query_gid;
+	dev->ibdev.get_link_layer = ocrdma_link_layer;
+	dev->ibdev.alloc_pd = ocrdma_alloc_pd;
+	dev->ibdev.dealloc_pd = ocrdma_dealloc_pd;
+
+	dev->ibdev.create_cq = ocrdma_create_cq;
+	dev->ibdev.destroy_cq = ocrdma_destroy_cq;
+	dev->ibdev.resize_cq = ocrdma_resize_cq;
+
+	dev->ibdev.create_qp = ocrdma_create_qp;
+	dev->ibdev.modify_qp = ocrdma_modify_qp;
+	dev->ibdev.query_qp = ocrdma_query_qp;
+	dev->ibdev.destroy_qp = ocrdma_destroy_qp;
+
+	dev->ibdev.query_pkey = ocrdma_query_pkey;
+	dev->ibdev.create_ah = ocrdma_create_ah;
+	dev->ibdev.destroy_ah = ocrdma_destroy_ah;
+	dev->ibdev.query_ah = ocrdma_query_ah;
+	dev->ibdev.modify_ah = ocrdma_modify_ah;
+
+	dev->ibdev.poll_cq = ocrdma_poll_cq;
+	dev->ibdev.post_send = ocrdma_post_send;
+	dev->ibdev.post_recv = ocrdma_post_recv;
+	dev->ibdev.req_notify_cq = ocrdma_arm_cq;
+
+	dev->ibdev.get_dma_mr = ocrdma_get_dma_mr;
+	dev->ibdev.reg_phys_mr = ocrdma_reg_kernel_mr;
+	dev->ibdev.dereg_mr = ocrdma_dereg_mr;
+	dev->ibdev.reg_user_mr = ocrdma_reg_user_mr;
+
+	dev->ibdev.alloc_fast_reg_mr = ocrdma_alloc_frmr;
+	dev->ibdev.alloc_fast_reg_page_list = ocrdma_alloc_frmr_page_list;
+	dev->ibdev.free_fast_reg_page_list = ocrdma_free_frmr_page_list;
+
+	/* mandatory to support user space verbs consumer. */
+	dev->ibdev.alloc_ucontext = ocrdma_alloc_ucontext;
+	dev->ibdev.dealloc_ucontext = ocrdma_dealloc_ucontext;
+	dev->ibdev.mmap = ocrdma_mmap;
+	dev->ibdev.dma_device = &dev->nic_info.pdev->dev;
+
+	dev->ibdev.process_mad = ocrdma_process_mad;
+
+	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
+		dev->ibdev.uverbs_cmd_mask |=
+		     OCRDMA_UVERBS(CREATE_SRQ) |
+		     OCRDMA_UVERBS(MODIFY_SRQ) |
+		     OCRDMA_UVERBS(QUERY_SRQ) |
+		     OCRDMA_UVERBS(DESTROY_SRQ) |
+		     OCRDMA_UVERBS(POST_SRQ_RECV);
+
+		dev->ibdev.create_srq = ocrdma_create_srq;
+		dev->ibdev.modify_srq = ocrdma_modify_srq;
+		dev->ibdev.query_srq = ocrdma_query_srq;
+		dev->ibdev.destroy_srq = ocrdma_destroy_srq;
+		dev->ibdev.post_srq_recv = ocrdma_post_srq_recv;
+	}
+	return ib_register_device(&dev->ibdev, NULL);
+}
+
+static int ocrdma_alloc_resources(struct ocrdma_dev *dev)
+{
+	mutex_init(&dev->dev_lock);
+	dev->sgid_tbl = kzalloc(sizeof(union ib_gid) *
+				OCRDMA_MAX_SGID, GFP_KERNEL);
+	if (!dev->sgid_tbl)
+		goto alloc_err;
+	spin_lock_init(&dev->sgid_lock);
+
+	dev->cq_tbl = kzalloc(sizeof(struct ocrdma_cq *) *
+			      OCRDMA_MAX_CQ, GFP_KERNEL);
+	if (!dev->cq_tbl)
+		goto alloc_err;
+
+	if (dev->attr.max_qp) {
+		dev->qp_tbl = kzalloc(sizeof(struct ocrdma_qp *) *
+				      OCRDMA_MAX_QP, GFP_KERNEL);
+		if (!dev->qp_tbl)
+			goto alloc_err;
+	}
+	spin_lock_init(&dev->av_tbl.lock);
+	spin_lock_init(&dev->flush_q_lock);
+	return 0;
+alloc_err:
+	pr_err("%s(%d) error.\n", __func__, dev->id);
+	return -ENOMEM;
+}
+
+static void ocrdma_free_resources(struct ocrdma_dev *dev)
+{
+	kfree(dev->qp_tbl);
+	kfree(dev->cq_tbl);
+	kfree(dev->sgid_tbl);
+}
+
+/* OCRDMA sysfs interface */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct ocrdma_dev *dev = dev_get_drvdata(device);
+
+	return scnprintf(buf, PAGE_SIZE, "0x%x\n", dev->nic_info.pdev->vendor);
+}
+
+static ssize_t show_fw_ver(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct ocrdma_dev *dev = dev_get_drvdata(device);
+
+	return scnprintf(buf, PAGE_SIZE, "%s", &dev->attr.fw_ver[0]);
+}
+
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(fw_ver, S_IRUGO, show_fw_ver, NULL);
+
+static struct device_attribute *ocrdma_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_fw_ver
+};
+
+static void ocrdma_remove_sysfiles(struct ocrdma_dev *dev)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
+		device_remove_file(&dev->ibdev.dev, ocrdma_attributes[i]);
+}
+
+static struct ocrdma_dev *ocrdma_add(struct be_dev_info *dev_info)
+{
+	int status = 0, i;
+	struct ocrdma_dev *dev;
+
+	dev = (struct ocrdma_dev *)ib_alloc_device(sizeof(struct ocrdma_dev));
+	if (!dev) {
+		pr_err("Unable to allocate ib device\n");
+		return NULL;
+	}
+	dev->mbx_cmd = kzalloc(sizeof(struct ocrdma_mqe_emb_cmd), GFP_KERNEL);
+	if (!dev->mbx_cmd)
+		goto idr_err;
+
+	memcpy(&dev->nic_info, dev_info, sizeof(*dev_info));
+	dev->id = idr_alloc(&ocrdma_dev_id, NULL, 0, 0, GFP_KERNEL);
+	if (dev->id < 0)
+		goto idr_err;
+
+	status = ocrdma_init_hw(dev);
+	if (status)
+		goto init_err;
+
+	status = ocrdma_alloc_resources(dev);
+	if (status)
+		goto alloc_err;
+
+	status = ocrdma_register_device(dev);
+	if (status)
+		goto alloc_err;
+
+	for (i = 0; i < ARRAY_SIZE(ocrdma_attributes); i++)
+		if (device_create_file(&dev->ibdev.dev, ocrdma_attributes[i]))
+			goto sysfs_err;
+	spin_lock(&ocrdma_devlist_lock);
+	list_add_tail_rcu(&dev->entry, &ocrdma_dev_list);
+	spin_unlock(&ocrdma_devlist_lock);
+	/* Init stats */
+	ocrdma_add_port_stats(dev);
+
+	pr_info("%s %s: %s \"%s\" port %d\n",
+		dev_name(&dev->nic_info.pdev->dev), hca_name(dev),
+		port_speed_string(dev), dev->model_number,
+		dev->hba_port_num);
+	pr_info("%s ocrdma%d driver loaded successfully\n",
+		dev_name(&dev->nic_info.pdev->dev), dev->id);
+	return dev;
+
+sysfs_err:
+	ocrdma_remove_sysfiles(dev);
+alloc_err:
+	ocrdma_free_resources(dev);
+	ocrdma_cleanup_hw(dev);
+init_err:
+	idr_remove(&ocrdma_dev_id, dev->id);
+idr_err:
+	kfree(dev->mbx_cmd);
+	ib_dealloc_device(&dev->ibdev);
+	pr_err("%s() leaving. ret=%d\n", __func__, status);
+	return NULL;
+}
+
+static void ocrdma_remove_free(struct rcu_head *rcu)
+{
+	struct ocrdma_dev *dev = container_of(rcu, struct ocrdma_dev, rcu);
+
+	idr_remove(&ocrdma_dev_id, dev->id);
+	kfree(dev->mbx_cmd);
+	ib_dealloc_device(&dev->ibdev);
+}
+
+static void ocrdma_remove(struct ocrdma_dev *dev)
+{
+	/* first unregister with stack to stop all the active traffic
+	 * of the registered clients.
+	 */
+	ocrdma_rem_port_stats(dev);
+	ocrdma_remove_sysfiles(dev);
+
+	ib_unregister_device(&dev->ibdev);
+
+	spin_lock(&ocrdma_devlist_lock);
+	list_del_rcu(&dev->entry);
+	spin_unlock(&ocrdma_devlist_lock);
+
+	ocrdma_free_resources(dev);
+	ocrdma_cleanup_hw(dev);
+
+	call_rcu(&dev->rcu, ocrdma_remove_free);
+}
+
+static int ocrdma_open(struct ocrdma_dev *dev)
+{
+	struct ib_event port_event;
+
+	port_event.event = IB_EVENT_PORT_ACTIVE;
+	port_event.element.port_num = 1;
+	port_event.device = &dev->ibdev;
+	ib_dispatch_event(&port_event);
+	return 0;
+}
+
+static int ocrdma_close(struct ocrdma_dev *dev)
+{
+	int i;
+	struct ocrdma_qp *qp, **cur_qp;
+	struct ib_event err_event;
+	struct ib_qp_attr attrs;
+	int attr_mask = IB_QP_STATE;
+
+	attrs.qp_state = IB_QPS_ERR;
+	mutex_lock(&dev->dev_lock);
+	if (dev->qp_tbl) {
+		cur_qp = dev->qp_tbl;
+		for (i = 0; i < OCRDMA_MAX_QP; i++) {
+			qp = cur_qp[i];
+			if (qp && qp->ibqp.qp_type != IB_QPT_GSI) {
+				/* change the QP state to ERROR */
+				_ocrdma_modify_qp(&qp->ibqp, &attrs, attr_mask);
+
+				err_event.event = IB_EVENT_QP_FATAL;
+				err_event.element.qp = &qp->ibqp;
+				err_event.device = &dev->ibdev;
+				ib_dispatch_event(&err_event);
+			}
+		}
+	}
+	mutex_unlock(&dev->dev_lock);
+
+	err_event.event = IB_EVENT_PORT_ERR;
+	err_event.element.port_num = 1;
+	err_event.device = &dev->ibdev;
+	ib_dispatch_event(&err_event);
+	return 0;
+}
+
+/* event handling via NIC driver ensures that all the NIC specific
+ * initialization done before RoCE driver notifies
+ * event to stack.
+ */
+static void ocrdma_event_handler(struct ocrdma_dev *dev, u32 event)
+{
+	switch (event) {
+	case BE_DEV_UP:
+		ocrdma_open(dev);
+		break;
+	case BE_DEV_DOWN:
+		ocrdma_close(dev);
+		break;
+	}
+}
+
+static struct ocrdma_driver ocrdma_drv = {
+	.name			= "ocrdma_driver",
+	.add			= ocrdma_add,
+	.remove			= ocrdma_remove,
+	.state_change_handler	= ocrdma_event_handler,
+	.be_abi_version		= OCRDMA_BE_ROCE_ABI_VERSION,
+};
+
+static void ocrdma_unregister_inet6addr_notifier(void)
+{
+#if IS_ENABLED(CONFIG_IPV6)
+	unregister_inet6addr_notifier(&ocrdma_inet6addr_notifier);
+#endif
+}
+
+static void ocrdma_unregister_inetaddr_notifier(void)
+{
+	unregister_inetaddr_notifier(&ocrdma_inetaddr_notifier);
+}
+
+static int __init ocrdma_init_module(void)
+{
+	int status;
+
+	ocrdma_init_debugfs();
+
+	status = register_inetaddr_notifier(&ocrdma_inetaddr_notifier);
+	if (status)
+		return status;
+
+#if IS_ENABLED(CONFIG_IPV6)
+	status = register_inet6addr_notifier(&ocrdma_inet6addr_notifier);
+	if (status)
+		goto err_notifier6;
+#endif
+
+	status = be_roce_register_driver(&ocrdma_drv);
+	if (status)
+		goto err_be_reg;
+
+	return 0;
+
+err_be_reg:
+	ocrdma_unregister_inet6addr_notifier();
+err_notifier6:
+	ocrdma_unregister_inetaddr_notifier();
+	return status;
+}
+
+static void __exit ocrdma_exit_module(void)
+{
+	be_roce_unregister_driver(&ocrdma_drv);
+	ocrdma_unregister_inet6addr_notifier();
+	ocrdma_unregister_inetaddr_notifier();
+	ocrdma_rem_debugfs();
+}
+
+module_init(ocrdma_init_module);
+module_exit(ocrdma_exit_module);
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_sli.h b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
new file mode 100644
index 00000000000..96c9ee602ba
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_sli.h
@@ -0,0 +1,1953 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_SLI_H__
+#define __OCRDMA_SLI_H__
+
+#define Bit(_b) (1 << (_b))
+
+enum {
+	OCRDMA_ASIC_GEN_SKH_R = 0x04,
+	OCRDMA_ASIC_GEN_LANCER = 0x0B
+};
+
+enum {
+	OCRDMA_ASIC_REV_A0 = 0x00,
+	OCRDMA_ASIC_REV_B0 = 0x10,
+	OCRDMA_ASIC_REV_C0 = 0x20
+};
+
+#define OCRDMA_SUBSYS_ROCE 10
+enum {
+	OCRDMA_CMD_QUERY_CONFIG = 1,
+	OCRDMA_CMD_ALLOC_PD,
+	OCRDMA_CMD_DEALLOC_PD,
+
+	OCRDMA_CMD_CREATE_AH_TBL,
+	OCRDMA_CMD_DELETE_AH_TBL,
+
+	OCRDMA_CMD_CREATE_QP,
+	OCRDMA_CMD_QUERY_QP,
+	OCRDMA_CMD_MODIFY_QP,
+	OCRDMA_CMD_DELETE_QP,
+
+	OCRDMA_CMD_RSVD1,
+	OCRDMA_CMD_ALLOC_LKEY,
+	OCRDMA_CMD_DEALLOC_LKEY,
+	OCRDMA_CMD_REGISTER_NSMR,
+	OCRDMA_CMD_REREGISTER_NSMR,
+	OCRDMA_CMD_REGISTER_NSMR_CONT,
+	OCRDMA_CMD_QUERY_NSMR,
+	OCRDMA_CMD_ALLOC_MW,
+	OCRDMA_CMD_QUERY_MW,
+
+	OCRDMA_CMD_CREATE_SRQ,
+	OCRDMA_CMD_QUERY_SRQ,
+	OCRDMA_CMD_MODIFY_SRQ,
+	OCRDMA_CMD_DELETE_SRQ,
+
+	OCRDMA_CMD_ATTACH_MCAST,
+	OCRDMA_CMD_DETACH_MCAST,
+	OCRDMA_CMD_GET_RDMA_STATS,
+
+	OCRDMA_CMD_MAX
+};
+
+#define OCRDMA_SUBSYS_COMMON 1
+enum {
+	OCRDMA_CMD_QUERY_NTWK_LINK_CONFIG_V1 = 5,
+	OCRDMA_CMD_CREATE_CQ		= 12,
+	OCRDMA_CMD_CREATE_EQ		= 13,
+	OCRDMA_CMD_CREATE_MQ		= 21,
+	OCRDMA_CMD_GET_CTRL_ATTRIBUTES  = 32,
+	OCRDMA_CMD_GET_FW_VER		= 35,
+	OCRDMA_CMD_DELETE_MQ		= 53,
+	OCRDMA_CMD_DELETE_CQ		= 54,
+	OCRDMA_CMD_DELETE_EQ		= 55,
+	OCRDMA_CMD_GET_FW_CONFIG	= 58,
+	OCRDMA_CMD_CREATE_MQ_EXT	= 90,
+	OCRDMA_CMD_PHY_DETAILS		= 102
+};
+
+enum {
+	QTYPE_EQ	= 1,
+	QTYPE_CQ	= 2,
+	QTYPE_MCCQ	= 3
+};
+
+#define OCRDMA_MAX_SGID (8)
+
+#define OCRDMA_MAX_QP    2048
+#define OCRDMA_MAX_CQ    2048
+#define OCRDMA_MAX_STAG  8192
+
+enum {
+	OCRDMA_DB_RQ_OFFSET		= 0xE0,
+	OCRDMA_DB_GEN2_RQ_OFFSET        = 0x100,
+	OCRDMA_DB_SQ_OFFSET		= 0x60,
+	OCRDMA_DB_GEN2_SQ_OFFSET	= 0x1C0,
+	OCRDMA_DB_SRQ_OFFSET		= OCRDMA_DB_RQ_OFFSET,
+	OCRDMA_DB_GEN2_SRQ_OFFSET	= OCRDMA_DB_GEN2_RQ_OFFSET,
+	OCRDMA_DB_CQ_OFFSET		= 0x120,
+	OCRDMA_DB_EQ_OFFSET		= OCRDMA_DB_CQ_OFFSET,
+	OCRDMA_DB_MQ_OFFSET		= 0x140,
+
+	OCRDMA_DB_SQ_SHIFT		= 16,
+	OCRDMA_DB_RQ_SHIFT		= 24
+};
+
+#define OCRDMA_DB_CQ_RING_ID_MASK       0x3FF	/* bits 0 - 9 */
+#define OCRDMA_DB_CQ_RING_ID_EXT_MASK  0x0C00	/* bits 10-11 of qid at 12-11 */
+/* qid #2 msbits at 12-11 */
+#define OCRDMA_DB_CQ_RING_ID_EXT_MASK_SHIFT  0x1
+#define OCRDMA_DB_CQ_NUM_POPPED_SHIFT       (16)	/* bits 16 - 28 */
+/* Rearm bit */
+#define OCRDMA_DB_CQ_REARM_SHIFT        (29)	/* bit 29 */
+/* solicited bit */
+#define OCRDMA_DB_CQ_SOLICIT_SHIFT   (31)	/* bit 31 */
+
+#define OCRDMA_EQ_ID_MASK		0x1FF	/* bits 0 - 8 */
+#define OCRDMA_EQ_ID_EXT_MASK		0x3e00	/* bits 9-13 */
+#define OCRDMA_EQ_ID_EXT_MASK_SHIFT	(2)	/* qid bits 9-13 at 11-15 */
+
+/* Clear the interrupt for this eq */
+#define OCRDMA_EQ_CLR_SHIFT			(9)	/* bit 9 */
+/* Must be 1 */
+#define OCRDMA_EQ_TYPE_SHIFT		(10)	/* bit 10 */
+/* Number of event entries processed */
+#define OCRDMA_NUM_EQE_SHIFT		(16)	/* bits 16 - 28 */
+/* Rearm bit */
+#define OCRDMA_REARM_SHIFT		(29)	/* bit 29 */
+
+#define OCRDMA_MQ_ID_MASK		0x7FF	/* bits 0 - 10 */
+/* Number of entries posted */
+#define OCRDMA_MQ_NUM_MQE_SHIFT	(16)	/* bits 16 - 29 */
+
+#define OCRDMA_MIN_HPAGE_SIZE (4096)
+
+#define OCRDMA_MIN_Q_PAGE_SIZE (4096)
+#define OCRDMA_MAX_Q_PAGES     (8)
+
+#define OCRDMA_SLI_ASIC_ID_OFFSET	0x9C
+#define OCRDMA_SLI_ASIC_REV_MASK	0x000000FF
+#define OCRDMA_SLI_ASIC_GEN_NUM_MASK	0x0000FF00
+#define OCRDMA_SLI_ASIC_GEN_NUM_SHIFT	0x08
+/*
+# 0: 4K Bytes
+# 1: 8K Bytes
+# 2: 16K Bytes
+# 3: 32K Bytes
+# 4: 64K Bytes
+# 5: 128K Bytes
+# 6: 256K Bytes
+# 7: 512K Bytes
+*/
+#define OCRDMA_MAX_Q_PAGE_SIZE_CNT (8)
+#define OCRDMA_Q_PAGE_BASE_SIZE (OCRDMA_MIN_Q_PAGE_SIZE * OCRDMA_MAX_Q_PAGES)
+
+#define MAX_OCRDMA_QP_PAGES      (8)
+#define OCRDMA_MAX_WQE_MEM_SIZE (MAX_OCRDMA_QP_PAGES * OCRDMA_MIN_HQ_PAGE_SIZE)
+
+#define OCRDMA_CREATE_CQ_MAX_PAGES (4)
+#define OCRDMA_DPP_CQE_SIZE (4)
+
+#define OCRDMA_GEN2_MAX_CQE 1024
+#define OCRDMA_GEN2_CQ_PAGE_SIZE 4096
+#define OCRDMA_GEN2_WQE_SIZE 256
+#define OCRDMA_MAX_CQE  4095
+#define OCRDMA_CQ_PAGE_SIZE 16384
+#define OCRDMA_WQE_SIZE 128
+#define OCRDMA_WQE_STRIDE 8
+#define OCRDMA_WQE_ALIGN_BYTES 16
+
+#define MAX_OCRDMA_SRQ_PAGES MAX_OCRDMA_QP_PAGES
+
+enum {
+	OCRDMA_MCH_OPCODE_SHIFT	= 0,
+	OCRDMA_MCH_OPCODE_MASK	= 0xFF,
+	OCRDMA_MCH_SUBSYS_SHIFT	= 8,
+	OCRDMA_MCH_SUBSYS_MASK	= 0xFF00
+};
+
+/* mailbox cmd header */
+struct ocrdma_mbx_hdr {
+	u32 subsys_op;
+	u32 timeout;		/* in seconds */
+	u32 cmd_len;
+	u32 rsvd_version;
+};
+
+enum {
+	OCRDMA_MBX_RSP_OPCODE_SHIFT	= 0,
+	OCRDMA_MBX_RSP_OPCODE_MASK	= 0xFF,
+	OCRDMA_MBX_RSP_SUBSYS_SHIFT	= 8,
+	OCRDMA_MBX_RSP_SUBSYS_MASK	= 0xFF << OCRDMA_MBX_RSP_SUBSYS_SHIFT,
+
+	OCRDMA_MBX_RSP_STATUS_SHIFT	= 0,
+	OCRDMA_MBX_RSP_STATUS_MASK	= 0xFF,
+	OCRDMA_MBX_RSP_ASTATUS_SHIFT	= 8,
+	OCRDMA_MBX_RSP_ASTATUS_MASK	= 0xFF << OCRDMA_MBX_RSP_ASTATUS_SHIFT
+};
+
+/* mailbox cmd response */
+struct ocrdma_mbx_rsp {
+	u32 subsys_op;
+	u32 status;
+	u32 rsp_len;
+	u32 add_rsp_len;
+};
+
+enum {
+	OCRDMA_MQE_EMBEDDED	= 1,
+	OCRDMA_MQE_NONEMBEDDED	= 0
+};
+
+struct ocrdma_mqe_sge {
+	u32 pa_lo;
+	u32 pa_hi;
+	u32 len;
+};
+
+enum {
+	OCRDMA_MQE_HDR_EMB_SHIFT	= 0,
+	OCRDMA_MQE_HDR_EMB_MASK		= Bit(0),
+	OCRDMA_MQE_HDR_SGE_CNT_SHIFT	= 3,
+	OCRDMA_MQE_HDR_SGE_CNT_MASK	= 0x1F << OCRDMA_MQE_HDR_SGE_CNT_SHIFT,
+	OCRDMA_MQE_HDR_SPECIAL_SHIFT	= 24,
+	OCRDMA_MQE_HDR_SPECIAL_MASK	= 0xFF << OCRDMA_MQE_HDR_SPECIAL_SHIFT
+};
+
+struct ocrdma_mqe_hdr {
+	u32 spcl_sge_cnt_emb;
+	u32 pyld_len;
+	u32 tag_lo;
+	u32 tag_hi;
+	u32 rsvd3;
+};
+
+struct ocrdma_mqe_emb_cmd {
+	struct ocrdma_mbx_hdr mch;
+	u8 pyld[220];
+};
+
+struct ocrdma_mqe {
+	struct ocrdma_mqe_hdr hdr;
+	union {
+		struct ocrdma_mqe_emb_cmd emb_req;
+		struct {
+			struct ocrdma_mqe_sge sge[19];
+		} nonemb_req;
+		u8 cmd[236];
+		struct ocrdma_mbx_rsp rsp;
+	} u;
+};
+
+#define OCRDMA_EQ_LEN       4096
+#define OCRDMA_MQ_CQ_LEN    256
+#define OCRDMA_MQ_LEN       128
+
+#define PAGE_SHIFT_4K		12
+#define PAGE_SIZE_4K		(1 << PAGE_SHIFT_4K)
+
+/* Returns number of pages spanned by the data starting at the given addr */
+#define PAGES_4K_SPANNED(_address, size) \
+	((u32)((((size_t)(_address) & (PAGE_SIZE_4K - 1)) +	\
+			(size) + (PAGE_SIZE_4K - 1)) >> PAGE_SHIFT_4K))
+
+struct ocrdma_delete_q_req {
+	struct ocrdma_mbx_hdr req;
+	u32 id;
+};
+
+struct ocrdma_pa {
+	u32 lo;
+	u32 hi;
+};
+
+#define MAX_OCRDMA_EQ_PAGES (8)
+struct ocrdma_create_eq_req {
+	struct ocrdma_mbx_hdr req;
+	u32 num_pages;
+	u32 valid;
+	u32 cnt;
+	u32 delay;
+	u32 rsvd;
+	struct ocrdma_pa pa[MAX_OCRDMA_EQ_PAGES];
+};
+
+enum {
+	OCRDMA_CREATE_EQ_VALID	= Bit(29),
+	OCRDMA_CREATE_EQ_CNT_SHIFT	= 26,
+	OCRDMA_CREATE_CQ_DELAY_SHIFT	= 13,
+};
+
+struct ocrdma_create_eq_rsp {
+	struct ocrdma_mbx_rsp rsp;
+	u32 vector_eqid;
+};
+
+#define OCRDMA_EQ_MINOR_OTHER (0x1)
+
+enum {
+	OCRDMA_MCQE_STATUS_SHIFT	= 0,
+	OCRDMA_MCQE_STATUS_MASK		= 0xFFFF,
+	OCRDMA_MCQE_ESTATUS_SHIFT	= 16,
+	OCRDMA_MCQE_ESTATUS_MASK	= 0xFFFF << OCRDMA_MCQE_ESTATUS_SHIFT,
+	OCRDMA_MCQE_CONS_SHIFT		= 27,
+	OCRDMA_MCQE_CONS_MASK		= Bit(27),
+	OCRDMA_MCQE_CMPL_SHIFT		= 28,
+	OCRDMA_MCQE_CMPL_MASK		= Bit(28),
+	OCRDMA_MCQE_AE_SHIFT		= 30,
+	OCRDMA_MCQE_AE_MASK		= Bit(30),
+	OCRDMA_MCQE_VALID_SHIFT		= 31,
+	OCRDMA_MCQE_VALID_MASK		= Bit(31)
+};
+
+struct ocrdma_mcqe {
+	u32 status;
+	u32 tag_lo;
+	u32 tag_hi;
+	u32 valid_ae_cmpl_cons;
+};
+
+enum {
+	OCRDMA_AE_MCQE_QPVALID		= Bit(31),
+	OCRDMA_AE_MCQE_QPID_MASK	= 0xFFFF,
+
+	OCRDMA_AE_MCQE_CQVALID		= Bit(31),
+	OCRDMA_AE_MCQE_CQID_MASK	= 0xFFFF,
+	OCRDMA_AE_MCQE_VALID		= Bit(31),
+	OCRDMA_AE_MCQE_AE		= Bit(30),
+	OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT	= 16,
+	OCRDMA_AE_MCQE_EVENT_TYPE_MASK	=
+					0xFF << OCRDMA_AE_MCQE_EVENT_TYPE_SHIFT,
+	OCRDMA_AE_MCQE_EVENT_CODE_SHIFT	= 8,
+	OCRDMA_AE_MCQE_EVENT_CODE_MASK	=
+					0xFF << OCRDMA_AE_MCQE_EVENT_CODE_SHIFT
+};
+struct ocrdma_ae_mcqe {
+	u32 qpvalid_qpid;
+	u32 cqvalid_cqid;
+	u32 evt_tag;
+	u32 valid_ae_event;
+};
+
+enum {
+	OCRDMA_AE_PVID_MCQE_ENABLED_SHIFT = 0,
+	OCRDMA_AE_PVID_MCQE_ENABLED_MASK  = 0xFF,
+	OCRDMA_AE_PVID_MCQE_TAG_SHIFT = 16,
+	OCRDMA_AE_PVID_MCQE_TAG_MASK = 0xFFFF << OCRDMA_AE_PVID_MCQE_TAG_SHIFT
+};
+
+struct ocrdma_ae_pvid_mcqe {
+	u32 tag_enabled;
+	u32 event_tag;
+	u32 rsvd1;
+	u32 rsvd2;
+};
+
+enum {
+	OCRDMA_AE_MPA_MCQE_REQ_ID_SHIFT		= 16,
+	OCRDMA_AE_MPA_MCQE_REQ_ID_MASK		= 0xFFFF <<
+					OCRDMA_AE_MPA_MCQE_REQ_ID_SHIFT,
+
+	OCRDMA_AE_MPA_MCQE_EVENT_CODE_SHIFT	= 8,
+	OCRDMA_AE_MPA_MCQE_EVENT_CODE_MASK	= 0xFF <<
+					OCRDMA_AE_MPA_MCQE_EVENT_CODE_SHIFT,
+	OCRDMA_AE_MPA_MCQE_EVENT_TYPE_SHIFT	= 16,
+	OCRDMA_AE_MPA_MCQE_EVENT_TYPE_MASK	= 0xFF <<
+					OCRDMA_AE_MPA_MCQE_EVENT_TYPE_SHIFT,
+	OCRDMA_AE_MPA_MCQE_EVENT_AE_SHIFT	= 30,
+	OCRDMA_AE_MPA_MCQE_EVENT_AE_MASK	= Bit(30),
+	OCRDMA_AE_MPA_MCQE_EVENT_VALID_SHIFT	= 31,
+	OCRDMA_AE_MPA_MCQE_EVENT_VALID_MASK	= Bit(31)
+};
+
+struct ocrdma_ae_mpa_mcqe {
+	u32 req_id;
+	u32 w1;
+	u32 w2;
+	u32 valid_ae_event;
+};
+
+enum {
+	OCRDMA_AE_QP_MCQE_NEW_QP_STATE_SHIFT	= 0,
+	OCRDMA_AE_QP_MCQE_NEW_QP_STATE_MASK	= 0xFFFF,
+	OCRDMA_AE_QP_MCQE_QP_ID_SHIFT		= 16,
+	OCRDMA_AE_QP_MCQE_QP_ID_MASK		= 0xFFFF <<
+						OCRDMA_AE_QP_MCQE_QP_ID_SHIFT,
+
+	OCRDMA_AE_QP_MCQE_EVENT_CODE_SHIFT	= 8,
+	OCRDMA_AE_QP_MCQE_EVENT_CODE_MASK	= 0xFF <<
+				OCRDMA_AE_QP_MCQE_EVENT_CODE_SHIFT,
+	OCRDMA_AE_QP_MCQE_EVENT_TYPE_SHIFT	= 16,
+	OCRDMA_AE_QP_MCQE_EVENT_TYPE_MASK	= 0xFF <<
+				OCRDMA_AE_QP_MCQE_EVENT_TYPE_SHIFT,
+	OCRDMA_AE_QP_MCQE_EVENT_AE_SHIFT	= 30,
+	OCRDMA_AE_QP_MCQE_EVENT_AE_MASK		= Bit(30),
+	OCRDMA_AE_QP_MCQE_EVENT_VALID_SHIFT	= 31,
+	OCRDMA_AE_QP_MCQE_EVENT_VALID_MASK	= Bit(31)
+};
+
+struct ocrdma_ae_qp_mcqe {
+	u32 qp_id_state;
+	u32 w1;
+	u32 w2;
+	u32 valid_ae_event;
+};
+
+#define OCRDMA_ASYNC_RDMA_EVE_CODE 0x14
+#define OCRDMA_ASYNC_GRP5_EVE_CODE 0x5
+#define OCRDMA_ASYNC_EVENT_PVID_STATE 0x3
+
+enum OCRDMA_ASYNC_EVENT_TYPE {
+	OCRDMA_CQ_ERROR			= 0x00,
+	OCRDMA_CQ_OVERRUN_ERROR		= 0x01,
+	OCRDMA_CQ_QPCAT_ERROR		= 0x02,
+	OCRDMA_QP_ACCESS_ERROR		= 0x03,
+	OCRDMA_QP_COMM_EST_EVENT	= 0x04,
+	OCRDMA_SQ_DRAINED_EVENT		= 0x05,
+	OCRDMA_DEVICE_FATAL_EVENT	= 0x08,
+	OCRDMA_SRQCAT_ERROR		= 0x0E,
+	OCRDMA_SRQ_LIMIT_EVENT		= 0x0F,
+	OCRDMA_QP_LAST_WQE_EVENT	= 0x10
+};
+
+/* mailbox command request and responses */
+enum {
+	OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_SHIFT		= 2,
+	OCRDMA_MBX_QUERY_CFG_CQ_OVERFLOW_MASK		= Bit(2),
+	OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_SHIFT	= 3,
+	OCRDMA_MBX_QUERY_CFG_SRQ_SUPPORTED_MASK		= Bit(3),
+	OCRDMA_MBX_QUERY_CFG_MAX_QP_SHIFT		= 8,
+	OCRDMA_MBX_QUERY_CFG_MAX_QP_MASK		= 0xFFFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_QP_SHIFT,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_PD_SHIFT		= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_PD_MASK		= 0xFFFF <<
+					OCRDMA_MBX_QUERY_CFG_MAX_PD_SHIFT,
+	OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT		= 8,
+	OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_MASK		= 0xFF <<
+				OCRDMA_MBX_QUERY_CFG_CA_ACK_DELAY_SHIFT,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_SHIFT		= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_SEND_SGE_MASK		= 0xFFFF,
+	OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_SHIFT	= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_WRITE_SGE_SHIFT,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_SHIFT	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_ORD_PER_QP_MASK	= 0xFFFF,
+	OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_SHIFT	= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_MASK	= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_IRD_PER_QP_SHIFT,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_OFFSET	= 24,
+	OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_MASK		= 0xFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_WQE_SIZE_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_OFFSET	= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_MASK		= 0xFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_RQE_SIZE_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_CQES_OFFSET	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_CQES_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_DPP_CQES_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_OFFSET		= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_SRQ_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_RPIR_QPS_OFFSET	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_RPIR_QPS_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_RPIR_QPS_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_OFFSET		= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_DPP_PDS_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_CREDITS_OFFSET	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_CREDITS_MASK	= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_DPP_CREDITS_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_QPS_OFFSET		= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_DPP_QPS_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_DPP_QPS_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_OFFSET	= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_MASK	= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_WQES_PER_WQ_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_OFFSET	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_MASK	= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_RQES_PER_RQ_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_CQ_OFFSET		= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_CQ_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_CQ_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_OFFSET	= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_MASK	= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_CQES_PER_CQ_OFFSET,
+
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_RQE_OFFSET		= 16,
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_RQE_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_SRQ_RQE_OFFSET,
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_OFFSET		= 0,
+	OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_MASK		= 0xFFFF <<
+				OCRDMA_MBX_QUERY_CFG_MAX_SRQ_SGE_OFFSET,
+};
+
+struct ocrdma_mbx_query_config {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+	u32 qp_srq_cq_ird_ord;
+	u32 max_pd_ca_ack_delay;
+	u32 max_write_send_sge;
+	u32 max_ird_ord_per_qp;
+	u32 max_shared_ird_ord;
+	u32 max_mr;
+	u32 max_mr_size_lo;
+	u32 max_mr_size_hi;
+	u32 max_num_mr_pbl;
+	u32 max_mw;
+	u32 max_fmr;
+	u32 max_pages_per_frmr;
+	u32 max_mcast_group;
+	u32 max_mcast_qp_attach;
+	u32 max_total_mcast_qp_attach;
+	u32 wqe_rqe_stride_max_dpp_cqs;
+	u32 max_srq_rpir_qps;
+	u32 max_dpp_pds_credits;
+	u32 max_dpp_credits_pds_per_pd;
+	u32 max_wqes_rqes_per_q;
+	u32 max_cq_cqes_per_cq;
+	u32 max_srq_rqe_sge;
+};
+
+struct ocrdma_fw_ver_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u8 running_ver[32];
+};
+
+struct ocrdma_fw_conf_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 config_num;
+	u32 asic_revision;
+	u32 phy_port;
+	u32 fn_mode;
+	struct {
+		u32 mode;
+		u32 nic_wqid_base;
+		u32 nic_wq_tot;
+		u32 prot_wqid_base;
+		u32 prot_wq_tot;
+		u32 prot_rqid_base;
+		u32 prot_rqid_tot;
+		u32 rsvd[6];
+	} ulp[2];
+	u32 fn_capabilities;
+	u32 rsvd1;
+	u32 rsvd2;
+	u32 base_eqid;
+	u32 max_eq;
+
+};
+
+enum {
+	OCRDMA_FN_MODE_RDMA	= 0x4
+};
+
+struct ocrdma_get_phy_info_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u16 phy_type;
+	u16 interface_type;
+	u32 misc_params;
+	u16 ext_phy_details;
+	u16 rsvd;
+	u16 auto_speeds_supported;
+	u16 fixed_speeds_supported;
+	u32 future_use[2];
+};
+
+enum {
+	OCRDMA_PHY_SPEED_ZERO = 0x0,
+	OCRDMA_PHY_SPEED_10MBPS = 0x1,
+	OCRDMA_PHY_SPEED_100MBPS = 0x2,
+	OCRDMA_PHY_SPEED_1GBPS = 0x4,
+	OCRDMA_PHY_SPEED_10GBPS = 0x8,
+	OCRDMA_PHY_SPEED_40GBPS = 0x20
+};
+
+
+struct ocrdma_get_link_speed_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u8 pt_port_num;
+	u8 link_duplex;
+	u8 phys_port_speed;
+	u8 phys_port_fault;
+	u16 rsvd1;
+	u16 qos_lnk_speed;
+	u8 logical_lnk_status;
+	u8 rsvd2[3];
+};
+
+enum {
+	OCRDMA_PHYS_LINK_SPEED_ZERO = 0x0,
+	OCRDMA_PHYS_LINK_SPEED_10MBPS = 0x1,
+	OCRDMA_PHYS_LINK_SPEED_100MBPS = 0x2,
+	OCRDMA_PHYS_LINK_SPEED_1GBPS = 0x3,
+	OCRDMA_PHYS_LINK_SPEED_10GBPS = 0x4,
+	OCRDMA_PHYS_LINK_SPEED_20GBPS = 0x5,
+	OCRDMA_PHYS_LINK_SPEED_25GBPS = 0x6,
+	OCRDMA_PHYS_LINK_SPEED_40GBPS = 0x7,
+	OCRDMA_PHYS_LINK_SPEED_100GBPS = 0x8
+};
+
+enum {
+	OCRDMA_CREATE_CQ_VER2			= 2,
+	OCRDMA_CREATE_CQ_VER3			= 3,
+
+	OCRDMA_CREATE_CQ_PAGE_CNT_MASK		= 0xFFFF,
+	OCRDMA_CREATE_CQ_PAGE_SIZE_SHIFT	= 16,
+	OCRDMA_CREATE_CQ_PAGE_SIZE_MASK		= 0xFF,
+
+	OCRDMA_CREATE_CQ_COALESCWM_SHIFT	= 12,
+	OCRDMA_CREATE_CQ_COALESCWM_MASK		= Bit(13) | Bit(12),
+	OCRDMA_CREATE_CQ_FLAGS_NODELAY		= Bit(14),
+	OCRDMA_CREATE_CQ_FLAGS_AUTO_VALID	= Bit(15),
+
+	OCRDMA_CREATE_CQ_EQ_ID_MASK		= 0xFFFF,
+	OCRDMA_CREATE_CQ_CQE_COUNT_MASK		= 0xFFFF
+};
+
+enum {
+	OCRDMA_CREATE_CQ_VER0			= 0,
+	OCRDMA_CREATE_CQ_DPP			= 1,
+	OCRDMA_CREATE_CQ_TYPE_SHIFT		= 24,
+	OCRDMA_CREATE_CQ_EQID_SHIFT		= 22,
+
+	OCRDMA_CREATE_CQ_CNT_SHIFT		= 27,
+	OCRDMA_CREATE_CQ_FLAGS_VALID		= Bit(29),
+	OCRDMA_CREATE_CQ_FLAGS_EVENTABLE	= Bit(31),
+	OCRDMA_CREATE_CQ_DEF_FLAGS		= OCRDMA_CREATE_CQ_FLAGS_VALID |
+					OCRDMA_CREATE_CQ_FLAGS_EVENTABLE |
+					OCRDMA_CREATE_CQ_FLAGS_NODELAY
+};
+
+struct ocrdma_create_cq_cmd {
+	struct ocrdma_mbx_hdr req;
+	u32 pgsz_pgcnt;
+	u32 ev_cnt_flags;
+	u32 eqn;
+	u16 cqe_count;
+	u16 pd_id;
+	u32 rsvd6;
+	struct ocrdma_pa pa[OCRDMA_CREATE_CQ_MAX_PAGES];
+};
+
+struct ocrdma_create_cq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_create_cq_cmd cmd;
+};
+
+enum {
+	OCRDMA_CREATE_CQ_RSP_CQ_ID_MASK	= 0xFFFF
+};
+
+struct ocrdma_create_cq_cmd_rsp {
+	struct ocrdma_mbx_rsp rsp;
+	u32 cq_id;
+};
+
+struct ocrdma_create_cq_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_create_cq_cmd_rsp rsp;
+};
+
+enum {
+	OCRDMA_CREATE_MQ_V0_CQ_ID_SHIFT		= 22,
+	OCRDMA_CREATE_MQ_CQ_ID_SHIFT		= 16,
+	OCRDMA_CREATE_MQ_RING_SIZE_SHIFT	= 16,
+	OCRDMA_CREATE_MQ_VALID			= Bit(31),
+	OCRDMA_CREATE_MQ_ASYNC_CQ_VALID		= Bit(0)
+};
+
+struct ocrdma_create_mq_req {
+	struct ocrdma_mbx_hdr req;
+	u32 cqid_pages;
+	u32 async_event_bitmap;
+	u32 async_cqid_ringsize;
+	u32 valid;
+	u32 async_cqid_valid;
+	u32 rsvd;
+	struct ocrdma_pa pa[8];
+};
+
+struct ocrdma_create_mq_rsp {
+	struct ocrdma_mbx_rsp rsp;
+	u32 id;
+};
+
+enum {
+	OCRDMA_DESTROY_CQ_QID_SHIFT			= 0,
+	OCRDMA_DESTROY_CQ_QID_MASK			= 0xFFFF,
+	OCRDMA_DESTROY_CQ_QID_BYPASS_FLUSH_SHIFT	= 16,
+	OCRDMA_DESTROY_CQ_QID_BYPASS_FLUSH_MASK		= 0xFFFF <<
+				OCRDMA_DESTROY_CQ_QID_BYPASS_FLUSH_SHIFT
+};
+
+struct ocrdma_destroy_cq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 bypass_flush_qid;
+};
+
+struct ocrdma_destroy_cq_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+enum {
+	OCRDMA_QPT_GSI	= 1,
+	OCRDMA_QPT_RC	= 2,
+	OCRDMA_QPT_UD	= 4,
+};
+
+enum {
+	OCRDMA_CREATE_QP_REQ_PD_ID_SHIFT	= 0,
+	OCRDMA_CREATE_QP_REQ_PD_ID_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_SQ_PAGE_SIZE_SHIFT	= 16,
+	OCRDMA_CREATE_QP_REQ_RQ_PAGE_SIZE_SHIFT	= 19,
+	OCRDMA_CREATE_QP_REQ_QPT_SHIFT		= 29,
+	OCRDMA_CREATE_QP_REQ_QPT_MASK		= Bit(31) | Bit(30) | Bit(29),
+
+	OCRDMA_CREATE_QP_REQ_MAX_RQE_SHIFT	= 0,
+	OCRDMA_CREATE_QP_REQ_MAX_RQE_MASK	= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_MAX_WQE_SHIFT	= 16,
+	OCRDMA_CREATE_QP_REQ_MAX_WQE_MASK	= 0xFFFF <<
+					OCRDMA_CREATE_QP_REQ_MAX_WQE_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_SHIFT	= 0,
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_WRITE_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_MASK		= 0xFFFF <<
+					OCRDMA_CREATE_QP_REQ_MAX_SGE_SEND_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_FMR_EN_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_FMR_EN_MASK		= Bit(0),
+	OCRDMA_CREATE_QP_REQ_ZERO_LKEYEN_SHIFT		= 1,
+	OCRDMA_CREATE_QP_REQ_ZERO_LKEYEN_MASK		= Bit(1),
+	OCRDMA_CREATE_QP_REQ_BIND_MEMWIN_SHIFT		= 2,
+	OCRDMA_CREATE_QP_REQ_BIND_MEMWIN_MASK		= Bit(2),
+	OCRDMA_CREATE_QP_REQ_INB_WREN_SHIFT		= 3,
+	OCRDMA_CREATE_QP_REQ_INB_WREN_MASK		= Bit(3),
+	OCRDMA_CREATE_QP_REQ_INB_RDEN_SHIFT		= 4,
+	OCRDMA_CREATE_QP_REQ_INB_RDEN_MASK		= Bit(4),
+	OCRDMA_CREATE_QP_REQ_USE_SRQ_SHIFT		= 5,
+	OCRDMA_CREATE_QP_REQ_USE_SRQ_MASK		= Bit(5),
+	OCRDMA_CREATE_QP_REQ_ENABLE_RPIR_SHIFT		= 6,
+	OCRDMA_CREATE_QP_REQ_ENABLE_RPIR_MASK		= Bit(6),
+	OCRDMA_CREATE_QP_REQ_ENABLE_DPP_SHIFT		= 7,
+	OCRDMA_CREATE_QP_REQ_ENABLE_DPP_MASK		= Bit(7),
+	OCRDMA_CREATE_QP_REQ_ENABLE_DPP_CQ_SHIFT	= 8,
+	OCRDMA_CREATE_QP_REQ_ENABLE_DPP_CQ_MASK		= Bit(8),
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_MAX_SGE_RECV_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_MAX_IRD_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_MAX_IRD_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_MAX_ORD_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_MAX_ORD_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_MAX_ORD_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_NUM_RQ_PAGES_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_NUM_WQ_PAGES_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_RQE_SIZE_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_RQE_SIZE_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_WQE_SIZE_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_WQE_SIZE_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_WQE_SIZE_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_RQ_CQID_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_RQ_CQID_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_WQ_CQID_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_WQ_CQID_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_WQ_CQID_SHIFT,
+
+	OCRDMA_CREATE_QP_REQ_DPP_CQPID_SHIFT		= 0,
+	OCRDMA_CREATE_QP_REQ_DPP_CQPID_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_REQ_DPP_CREDIT_SHIFT		= 16,
+	OCRDMA_CREATE_QP_REQ_DPP_CREDIT_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_REQ_DPP_CREDIT_SHIFT
+};
+
+enum {
+	OCRDMA_CREATE_QP_REQ_DPP_CREDIT_LIMIT	= 16,
+	OCRDMA_CREATE_QP_RSP_DPP_PAGE_SHIFT	= 1
+};
+
+#define MAX_OCRDMA_IRD_PAGES 4
+
+enum ocrdma_qp_flags {
+	OCRDMA_QP_MW_BIND	= 1,
+	OCRDMA_QP_LKEY0		= (1 << 1),
+	OCRDMA_QP_FAST_REG	= (1 << 2),
+	OCRDMA_QP_INB_RD	= (1 << 6),
+	OCRDMA_QP_INB_WR	= (1 << 7),
+};
+
+enum ocrdma_qp_state {
+	OCRDMA_QPS_RST		= 0,
+	OCRDMA_QPS_INIT		= 1,
+	OCRDMA_QPS_RTR		= 2,
+	OCRDMA_QPS_RTS		= 3,
+	OCRDMA_QPS_SQE		= 4,
+	OCRDMA_QPS_SQ_DRAINING	= 5,
+	OCRDMA_QPS_ERR		= 6,
+	OCRDMA_QPS_SQD		= 7
+};
+
+struct ocrdma_create_qp_req {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 type_pgsz_pdn;
+	u32 max_wqe_rqe;
+	u32 max_sge_send_write;
+	u32 max_sge_recv_flags;
+	u32 max_ord_ird;
+	u32 num_wq_rq_pages;
+	u32 wqe_rqe_size;
+	u32 wq_rq_cqid;
+	struct ocrdma_pa wq_addr[MAX_OCRDMA_QP_PAGES];
+	struct ocrdma_pa rq_addr[MAX_OCRDMA_QP_PAGES];
+	u32 dpp_credits_cqid;
+	u32 rpir_lkey;
+	struct ocrdma_pa ird_addr[MAX_OCRDMA_IRD_PAGES];
+};
+
+enum {
+	OCRDMA_CREATE_QP_RSP_QP_ID_SHIFT		= 0,
+	OCRDMA_CREATE_QP_RSP_QP_ID_MASK			= 0xFFFF,
+
+	OCRDMA_CREATE_QP_RSP_MAX_RQE_SHIFT		= 0,
+	OCRDMA_CREATE_QP_RSP_MAX_RQE_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_RSP_MAX_WQE_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_MAX_WQE_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_MAX_WQE_SHIFT,
+
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_WRITE_SHIFT	= 0,
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_WRITE_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_SEND_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_SEND_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_MAX_SGE_SEND_SHIFT,
+
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_RECV_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_MAX_SGE_RECV_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_MAX_SGE_RECV_SHIFT,
+
+	OCRDMA_CREATE_QP_RSP_MAX_IRD_SHIFT		= 0,
+	OCRDMA_CREATE_QP_RSP_MAX_IRD_MASK		= 0xFFFF,
+	OCRDMA_CREATE_QP_RSP_MAX_ORD_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_MAX_ORD_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_MAX_ORD_SHIFT,
+
+	OCRDMA_CREATE_QP_RSP_RQ_ID_SHIFT		= 0,
+	OCRDMA_CREATE_QP_RSP_RQ_ID_MASK			= 0xFFFF,
+	OCRDMA_CREATE_QP_RSP_SQ_ID_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_SQ_ID_MASK			= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_SQ_ID_SHIFT,
+
+	OCRDMA_CREATE_QP_RSP_DPP_ENABLED_MASK		= Bit(0),
+	OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_SHIFT	= 1,
+	OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_MASK	= 0x7FFF <<
+				OCRDMA_CREATE_QP_RSP_DPP_PAGE_OFFSET_SHIFT,
+	OCRDMA_CREATE_QP_RSP_DPP_CREDITS_SHIFT		= 16,
+	OCRDMA_CREATE_QP_RSP_DPP_CREDITS_MASK		= 0xFFFF <<
+				OCRDMA_CREATE_QP_RSP_DPP_CREDITS_SHIFT,
+};
+
+struct ocrdma_create_qp_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 qp_id;
+	u32 max_wqe_rqe;
+	u32 max_sge_send_write;
+	u32 max_sge_recv;
+	u32 max_ord_ird;
+	u32 sq_rq_id;
+	u32 dpp_response;
+};
+
+struct ocrdma_destroy_qp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 qp_id;
+};
+
+struct ocrdma_destroy_qp_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+enum {
+	OCRDMA_MODIFY_QP_ID_SHIFT	= 0,
+	OCRDMA_MODIFY_QP_ID_MASK	= 0xFFFF,
+
+	OCRDMA_QP_PARA_QPS_VALID	= Bit(0),
+	OCRDMA_QP_PARA_SQD_ASYNC_VALID	= Bit(1),
+	OCRDMA_QP_PARA_PKEY_VALID	= Bit(2),
+	OCRDMA_QP_PARA_QKEY_VALID	= Bit(3),
+	OCRDMA_QP_PARA_PMTU_VALID	= Bit(4),
+	OCRDMA_QP_PARA_ACK_TO_VALID	= Bit(5),
+	OCRDMA_QP_PARA_RETRY_CNT_VALID	= Bit(6),
+	OCRDMA_QP_PARA_RRC_VALID	= Bit(7),
+	OCRDMA_QP_PARA_RQPSN_VALID	= Bit(8),
+	OCRDMA_QP_PARA_MAX_IRD_VALID	= Bit(9),
+	OCRDMA_QP_PARA_MAX_ORD_VALID	= Bit(10),
+	OCRDMA_QP_PARA_RNT_VALID	= Bit(11),
+	OCRDMA_QP_PARA_SQPSN_VALID	= Bit(12),
+	OCRDMA_QP_PARA_DST_QPN_VALID	= Bit(13),
+	OCRDMA_QP_PARA_MAX_WQE_VALID	= Bit(14),
+	OCRDMA_QP_PARA_MAX_RQE_VALID	= Bit(15),
+	OCRDMA_QP_PARA_SGE_SEND_VALID	= Bit(16),
+	OCRDMA_QP_PARA_SGE_RECV_VALID	= Bit(17),
+	OCRDMA_QP_PARA_SGE_WR_VALID	= Bit(18),
+	OCRDMA_QP_PARA_INB_RDEN_VALID	= Bit(19),
+	OCRDMA_QP_PARA_INB_WREN_VALID	= Bit(20),
+	OCRDMA_QP_PARA_FLOW_LBL_VALID	= Bit(21),
+	OCRDMA_QP_PARA_BIND_EN_VALID	= Bit(22),
+	OCRDMA_QP_PARA_ZLKEY_EN_VALID	= Bit(23),
+	OCRDMA_QP_PARA_FMR_EN_VALID	= Bit(24),
+	OCRDMA_QP_PARA_INBAT_EN_VALID	= Bit(25),
+	OCRDMA_QP_PARA_VLAN_EN_VALID	= Bit(26),
+
+	OCRDMA_MODIFY_QP_FLAGS_RD	= Bit(0),
+	OCRDMA_MODIFY_QP_FLAGS_WR	= Bit(1),
+	OCRDMA_MODIFY_QP_FLAGS_SEND	= Bit(2),
+	OCRDMA_MODIFY_QP_FLAGS_ATOMIC	= Bit(3)
+};
+
+enum {
+	OCRDMA_QP_PARAMS_SRQ_ID_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_SRQ_ID_MASK		= 0xFFFF,
+
+	OCRDMA_QP_PARAMS_MAX_RQE_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_MAX_RQE_MASK		= 0xFFFF,
+	OCRDMA_QP_PARAMS_MAX_WQE_SHIFT		= 16,
+	OCRDMA_QP_PARAMS_MAX_WQE_MASK		= 0xFFFF <<
+	    OCRDMA_QP_PARAMS_MAX_WQE_SHIFT,
+
+	OCRDMA_QP_PARAMS_MAX_SGE_WRITE_SHIFT	= 0,
+	OCRDMA_QP_PARAMS_MAX_SGE_WRITE_MASK	= 0xFFFF,
+	OCRDMA_QP_PARAMS_MAX_SGE_SEND_SHIFT	= 16,
+	OCRDMA_QP_PARAMS_MAX_SGE_SEND_MASK	= 0xFFFF <<
+					OCRDMA_QP_PARAMS_MAX_SGE_SEND_SHIFT,
+
+	OCRDMA_QP_PARAMS_FLAGS_FMR_EN		= Bit(0),
+	OCRDMA_QP_PARAMS_FLAGS_LKEY_0_EN	= Bit(1),
+	OCRDMA_QP_PARAMS_FLAGS_BIND_MW_EN	= Bit(2),
+	OCRDMA_QP_PARAMS_FLAGS_INBWR_EN		= Bit(3),
+	OCRDMA_QP_PARAMS_FLAGS_INBRD_EN		= Bit(4),
+	OCRDMA_QP_PARAMS_STATE_SHIFT		= 5,
+	OCRDMA_QP_PARAMS_STATE_MASK		= Bit(5) | Bit(6) | Bit(7),
+	OCRDMA_QP_PARAMS_FLAGS_SQD_ASYNC	= Bit(8),
+	OCRDMA_QP_PARAMS_FLAGS_INB_ATEN		= Bit(9),
+	OCRDMA_QP_PARAMS_MAX_SGE_RECV_SHIFT	= 16,
+	OCRDMA_QP_PARAMS_MAX_SGE_RECV_MASK	= 0xFFFF <<
+					OCRDMA_QP_PARAMS_MAX_SGE_RECV_SHIFT,
+
+	OCRDMA_QP_PARAMS_MAX_IRD_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_MAX_IRD_MASK		= 0xFFFF,
+	OCRDMA_QP_PARAMS_MAX_ORD_SHIFT		= 16,
+	OCRDMA_QP_PARAMS_MAX_ORD_MASK		= 0xFFFF <<
+					OCRDMA_QP_PARAMS_MAX_ORD_SHIFT,
+
+	OCRDMA_QP_PARAMS_RQ_CQID_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_RQ_CQID_MASK		= 0xFFFF,
+	OCRDMA_QP_PARAMS_WQ_CQID_SHIFT		= 16,
+	OCRDMA_QP_PARAMS_WQ_CQID_MASK		= 0xFFFF <<
+					OCRDMA_QP_PARAMS_WQ_CQID_SHIFT,
+
+	OCRDMA_QP_PARAMS_RQ_PSN_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_RQ_PSN_MASK		= 0xFFFFFF,
+	OCRDMA_QP_PARAMS_HOP_LMT_SHIFT		= 24,
+	OCRDMA_QP_PARAMS_HOP_LMT_MASK		= 0xFF <<
+					OCRDMA_QP_PARAMS_HOP_LMT_SHIFT,
+
+	OCRDMA_QP_PARAMS_SQ_PSN_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_SQ_PSN_MASK		= 0xFFFFFF,
+	OCRDMA_QP_PARAMS_TCLASS_SHIFT		= 24,
+	OCRDMA_QP_PARAMS_TCLASS_MASK		= 0xFF <<
+					OCRDMA_QP_PARAMS_TCLASS_SHIFT,
+
+	OCRDMA_QP_PARAMS_DEST_QPN_SHIFT		= 0,
+	OCRDMA_QP_PARAMS_DEST_QPN_MASK		= 0xFFFFFF,
+	OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT	= 24,
+	OCRDMA_QP_PARAMS_RNR_RETRY_CNT_MASK	= 0x7 <<
+					OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT,
+	OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT	= 27,
+	OCRDMA_QP_PARAMS_ACK_TIMEOUT_MASK	= 0x1F <<
+					OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT,
+
+	OCRDMA_QP_PARAMS_PKEY_IDNEX_SHIFT	= 0,
+	OCRDMA_QP_PARAMS_PKEY_INDEX_MASK	= 0xFFFF,
+	OCRDMA_QP_PARAMS_PATH_MTU_SHIFT		= 18,
+	OCRDMA_QP_PARAMS_PATH_MTU_MASK		= 0x3FFF <<
+					OCRDMA_QP_PARAMS_PATH_MTU_SHIFT,
+
+	OCRDMA_QP_PARAMS_FLOW_LABEL_SHIFT	= 0,
+	OCRDMA_QP_PARAMS_FLOW_LABEL_MASK	= 0xFFFFF,
+	OCRDMA_QP_PARAMS_SL_SHIFT		= 20,
+	OCRDMA_QP_PARAMS_SL_MASK		= 0xF <<
+					OCRDMA_QP_PARAMS_SL_SHIFT,
+	OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT	= 24,
+	OCRDMA_QP_PARAMS_RETRY_CNT_MASK		= 0x7 <<
+					OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT,
+	OCRDMA_QP_PARAMS_RNR_NAK_TIMER_SHIFT	= 27,
+	OCRDMA_QP_PARAMS_RNR_NAK_TIMER_MASK	= 0x1F <<
+					OCRDMA_QP_PARAMS_RNR_NAK_TIMER_SHIFT,
+
+	OCRDMA_QP_PARAMS_DMAC_B4_TO_B5_SHIFT	= 0,
+	OCRDMA_QP_PARAMS_DMAC_B4_TO_B5_MASK	= 0xFFFF,
+	OCRDMA_QP_PARAMS_VLAN_SHIFT		= 16,
+	OCRDMA_QP_PARAMS_VLAN_MASK		= 0xFFFF <<
+					OCRDMA_QP_PARAMS_VLAN_SHIFT
+};
+
+struct ocrdma_qp_params {
+	u32 id;
+	u32 max_wqe_rqe;
+	u32 max_sge_send_write;
+	u32 max_sge_recv_flags;
+	u32 max_ord_ird;
+	u32 wq_rq_cqid;
+	u32 hop_lmt_rq_psn;
+	u32 tclass_sq_psn;
+	u32 ack_to_rnr_rtc_dest_qpn;
+	u32 path_mtu_pkey_indx;
+	u32 rnt_rc_sl_fl;
+	u8 sgid[16];
+	u8 dgid[16];
+	u32 dmac_b0_to_b3;
+	u32 vlan_dmac_b4_to_b5;
+	u32 qkey;
+};
+
+
+struct ocrdma_modify_qp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	struct ocrdma_qp_params params;
+	u32 flags;
+	u32 rdma_flags;
+	u32 num_outstanding_atomic_rd;
+};
+
+enum {
+	OCRDMA_MODIFY_QP_RSP_MAX_RQE_SHIFT	= 0,
+	OCRDMA_MODIFY_QP_RSP_MAX_RQE_MASK	= 0xFFFF,
+	OCRDMA_MODIFY_QP_RSP_MAX_WQE_SHIFT	= 16,
+	OCRDMA_MODIFY_QP_RSP_MAX_WQE_MASK	= 0xFFFF <<
+					OCRDMA_MODIFY_QP_RSP_MAX_WQE_SHIFT,
+
+	OCRDMA_MODIFY_QP_RSP_MAX_IRD_SHIFT	= 0,
+	OCRDMA_MODIFY_QP_RSP_MAX_IRD_MASK	= 0xFFFF,
+	OCRDMA_MODIFY_QP_RSP_MAX_ORD_SHIFT	= 16,
+	OCRDMA_MODIFY_QP_RSP_MAX_ORD_MASK	= 0xFFFF <<
+					OCRDMA_MODIFY_QP_RSP_MAX_ORD_SHIFT
+};
+
+struct ocrdma_modify_qp_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 max_wqe_rqe;
+	u32 max_ord_ird;
+};
+
+struct ocrdma_query_qp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+#define OCRDMA_QUERY_UP_QP_ID_SHIFT	0
+#define OCRDMA_QUERY_UP_QP_ID_MASK	0xFFFFFF
+	u32 qp_id;
+};
+
+struct ocrdma_query_qp_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+	struct ocrdma_qp_params params;
+};
+
+enum {
+	OCRDMA_CREATE_SRQ_PD_ID_SHIFT		= 0,
+	OCRDMA_CREATE_SRQ_PD_ID_MASK		= 0xFFFF,
+	OCRDMA_CREATE_SRQ_PG_SZ_SHIFT		= 16,
+	OCRDMA_CREATE_SRQ_PG_SZ_MASK		= 0x3 <<
+					OCRDMA_CREATE_SRQ_PG_SZ_SHIFT,
+
+	OCRDMA_CREATE_SRQ_MAX_RQE_SHIFT		= 0,
+	OCRDMA_CREATE_SRQ_MAX_SGE_RECV_SHIFT	= 16,
+	OCRDMA_CREATE_SRQ_MAX_SGE_RECV_MASK	= 0xFFFF <<
+					OCRDMA_CREATE_SRQ_MAX_SGE_RECV_SHIFT,
+
+	OCRDMA_CREATE_SRQ_RQE_SIZE_SHIFT	= 0,
+	OCRDMA_CREATE_SRQ_RQE_SIZE_MASK		= 0xFFFF,
+	OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_SHIFT	= 16,
+	OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_MASK	= 0xFFFF <<
+					OCRDMA_CREATE_SRQ_NUM_RQ_PAGES_SHIFT
+};
+
+struct ocrdma_create_srq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 pgsz_pdid;
+	u32 max_sge_rqe;
+	u32 pages_rqe_sz;
+	struct ocrdma_pa rq_addr[MAX_OCRDMA_SRQ_PAGES];
+};
+
+enum {
+	OCRDMA_CREATE_SRQ_RSP_SRQ_ID_SHIFT			= 0,
+	OCRDMA_CREATE_SRQ_RSP_SRQ_ID_MASK			= 0xFFFFFF,
+
+	OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_SHIFT		= 0,
+	OCRDMA_CREATE_SRQ_RSP_MAX_RQE_ALLOCATED_MASK		= 0xFFFF,
+	OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_SHIFT	= 16,
+	OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_MASK	= 0xFFFF <<
+			OCRDMA_CREATE_SRQ_RSP_MAX_SGE_RECV_ALLOCATED_SHIFT
+};
+
+struct ocrdma_create_srq_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 id;
+	u32 max_sge_rqe_allocated;
+};
+
+enum {
+	OCRDMA_MODIFY_SRQ_ID_SHIFT	= 0,
+	OCRDMA_MODIFY_SRQ_ID_MASK	= 0xFFFFFF,
+
+	OCRDMA_MODIFY_SRQ_MAX_RQE_SHIFT	= 0,
+	OCRDMA_MODIFY_SRQ_MAX_RQE_MASK	= 0xFFFF,
+	OCRDMA_MODIFY_SRQ_LIMIT_SHIFT	= 16,
+	OCRDMA_MODIFY_SRQ__LIMIT_MASK	= 0xFFFF <<
+					OCRDMA_MODIFY_SRQ_LIMIT_SHIFT
+};
+
+struct ocrdma_modify_srq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rep;
+
+	u32 id;
+	u32 limit_max_rqe;
+};
+
+enum {
+	OCRDMA_QUERY_SRQ_ID_SHIFT	= 0,
+	OCRDMA_QUERY_SRQ_ID_MASK	= 0xFFFFFF
+};
+
+struct ocrdma_query_srq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp req;
+
+	u32 id;
+};
+
+enum {
+	OCRDMA_QUERY_SRQ_RSP_PD_ID_SHIFT	= 0,
+	OCRDMA_QUERY_SRQ_RSP_PD_ID_MASK		= 0xFFFF,
+	OCRDMA_QUERY_SRQ_RSP_MAX_RQE_SHIFT	= 16,
+	OCRDMA_QUERY_SRQ_RSP_MAX_RQE_MASK	= 0xFFFF <<
+					OCRDMA_QUERY_SRQ_RSP_MAX_RQE_SHIFT,
+
+	OCRDMA_QUERY_SRQ_RSP_MAX_SGE_RECV_SHIFT	= 0,
+	OCRDMA_QUERY_SRQ_RSP_MAX_SGE_RECV_MASK	= 0xFFFF,
+	OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_SHIFT	= 16,
+	OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_MASK	= 0xFFFF <<
+					OCRDMA_QUERY_SRQ_RSP_SRQ_LIMIT_SHIFT
+};
+
+struct ocrdma_query_srq_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp req;
+
+	u32 max_rqe_pdid;
+	u32 srq_lmt_max_sge;
+};
+
+enum {
+	OCRDMA_DESTROY_SRQ_ID_SHIFT	= 0,
+	OCRDMA_DESTROY_SRQ_ID_MASK	= 0xFFFFFF
+};
+
+struct ocrdma_destroy_srq {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp req;
+
+	u32 id;
+};
+
+enum {
+	OCRDMA_ALLOC_PD_ENABLE_DPP	= BIT(16),
+	OCRDMA_PD_MAX_DPP_ENABLED_QP	= 8,
+	OCRDMA_DPP_PAGE_SIZE		= 4096
+};
+
+struct ocrdma_alloc_pd {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 enable_dpp_rsvd;
+};
+
+enum {
+	OCRDMA_ALLOC_PD_RSP_DPP			= Bit(16),
+	OCRDMA_ALLOC_PD_RSP_DPP_PAGE_SHIFT	= 20,
+	OCRDMA_ALLOC_PD_RSP_PDID_MASK		= 0xFFFF,
+};
+
+struct ocrdma_alloc_pd_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+	u32 dpp_page_pdid;
+};
+
+struct ocrdma_dealloc_pd {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 id;
+};
+
+struct ocrdma_dealloc_pd_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+enum {
+	OCRDMA_ADDR_CHECK_ENABLE	= 1,
+	OCRDMA_ADDR_CHECK_DISABLE	= 0
+};
+
+enum {
+	OCRDMA_ALLOC_LKEY_PD_ID_SHIFT		= 0,
+	OCRDMA_ALLOC_LKEY_PD_ID_MASK		= 0xFFFF,
+
+	OCRDMA_ALLOC_LKEY_ADDR_CHECK_SHIFT	= 0,
+	OCRDMA_ALLOC_LKEY_ADDR_CHECK_MASK	= Bit(0),
+	OCRDMA_ALLOC_LKEY_FMR_SHIFT		= 1,
+	OCRDMA_ALLOC_LKEY_FMR_MASK		= Bit(1),
+	OCRDMA_ALLOC_LKEY_REMOTE_INV_SHIFT	= 2,
+	OCRDMA_ALLOC_LKEY_REMOTE_INV_MASK	= Bit(2),
+	OCRDMA_ALLOC_LKEY_REMOTE_WR_SHIFT	= 3,
+	OCRDMA_ALLOC_LKEY_REMOTE_WR_MASK	= Bit(3),
+	OCRDMA_ALLOC_LKEY_REMOTE_RD_SHIFT	= 4,
+	OCRDMA_ALLOC_LKEY_REMOTE_RD_MASK	= Bit(4),
+	OCRDMA_ALLOC_LKEY_LOCAL_WR_SHIFT	= 5,
+	OCRDMA_ALLOC_LKEY_LOCAL_WR_MASK		= Bit(5),
+	OCRDMA_ALLOC_LKEY_REMOTE_ATOMIC_MASK	= Bit(6),
+	OCRDMA_ALLOC_LKEY_REMOTE_ATOMIC_SHIFT	= 6,
+	OCRDMA_ALLOC_LKEY_PBL_SIZE_SHIFT	= 16,
+	OCRDMA_ALLOC_LKEY_PBL_SIZE_MASK		= 0xFFFF <<
+						OCRDMA_ALLOC_LKEY_PBL_SIZE_SHIFT
+};
+
+struct ocrdma_alloc_lkey {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 pdid;
+	u32 pbl_sz_flags;
+};
+
+struct ocrdma_alloc_lkey_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 lrkey;
+	u32 num_pbl_rsvd;
+};
+
+struct ocrdma_dealloc_lkey {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 lkey;
+	u32 rsvd_frmr;
+};
+
+struct ocrdma_dealloc_lkey_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+#define MAX_OCRDMA_NSMR_PBL    (u32)22
+#define MAX_OCRDMA_PBL_SIZE     65536
+#define MAX_OCRDMA_PBL_PER_LKEY	32767
+
+enum {
+	OCRDMA_REG_NSMR_LRKEY_INDEX_SHIFT	= 0,
+	OCRDMA_REG_NSMR_LRKEY_INDEX_MASK	= 0xFFFFFF,
+	OCRDMA_REG_NSMR_LRKEY_SHIFT		= 24,
+	OCRDMA_REG_NSMR_LRKEY_MASK		= 0xFF <<
+					OCRDMA_REG_NSMR_LRKEY_SHIFT,
+
+	OCRDMA_REG_NSMR_PD_ID_SHIFT		= 0,
+	OCRDMA_REG_NSMR_PD_ID_MASK		= 0xFFFF,
+	OCRDMA_REG_NSMR_NUM_PBL_SHIFT		= 16,
+	OCRDMA_REG_NSMR_NUM_PBL_MASK		= 0xFFFF <<
+					OCRDMA_REG_NSMR_NUM_PBL_SHIFT,
+
+	OCRDMA_REG_NSMR_PBE_SIZE_SHIFT		= 0,
+	OCRDMA_REG_NSMR_PBE_SIZE_MASK		= 0xFFFF,
+	OCRDMA_REG_NSMR_HPAGE_SIZE_SHIFT	= 16,
+	OCRDMA_REG_NSMR_HPAGE_SIZE_MASK		= 0xFF <<
+					OCRDMA_REG_NSMR_HPAGE_SIZE_SHIFT,
+	OCRDMA_REG_NSMR_BIND_MEMWIN_SHIFT	= 24,
+	OCRDMA_REG_NSMR_BIND_MEMWIN_MASK	= Bit(24),
+	OCRDMA_REG_NSMR_ZB_SHIFT		= 25,
+	OCRDMA_REG_NSMR_ZB_SHIFT_MASK		= Bit(25),
+	OCRDMA_REG_NSMR_REMOTE_INV_SHIFT	= 26,
+	OCRDMA_REG_NSMR_REMOTE_INV_MASK		= Bit(26),
+	OCRDMA_REG_NSMR_REMOTE_WR_SHIFT		= 27,
+	OCRDMA_REG_NSMR_REMOTE_WR_MASK		= Bit(27),
+	OCRDMA_REG_NSMR_REMOTE_RD_SHIFT		= 28,
+	OCRDMA_REG_NSMR_REMOTE_RD_MASK		= Bit(28),
+	OCRDMA_REG_NSMR_LOCAL_WR_SHIFT		= 29,
+	OCRDMA_REG_NSMR_LOCAL_WR_MASK		= Bit(29),
+	OCRDMA_REG_NSMR_REMOTE_ATOMIC_SHIFT	= 30,
+	OCRDMA_REG_NSMR_REMOTE_ATOMIC_MASK	= Bit(30),
+	OCRDMA_REG_NSMR_LAST_SHIFT		= 31,
+	OCRDMA_REG_NSMR_LAST_MASK		= Bit(31)
+};
+
+struct ocrdma_reg_nsmr {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr cmd;
+
+	u32 fr_mr;
+	u32 num_pbl_pdid;
+	u32 flags_hpage_pbe_sz;
+	u32 totlen_low;
+	u32 totlen_high;
+	u32 fbo_low;
+	u32 fbo_high;
+	u32 va_loaddr;
+	u32 va_hiaddr;
+	struct ocrdma_pa pbl[MAX_OCRDMA_NSMR_PBL];
+};
+
+enum {
+	OCRDMA_REG_NSMR_CONT_PBL_SHIFT		= 0,
+	OCRDMA_REG_NSMR_CONT_PBL_SHIFT_MASK	= 0xFFFF,
+	OCRDMA_REG_NSMR_CONT_NUM_PBL_SHIFT	= 16,
+	OCRDMA_REG_NSMR_CONT_NUM_PBL_MASK	= 0xFFFF <<
+					OCRDMA_REG_NSMR_CONT_NUM_PBL_SHIFT,
+
+	OCRDMA_REG_NSMR_CONT_LAST_SHIFT		= 31,
+	OCRDMA_REG_NSMR_CONT_LAST_MASK		= Bit(31)
+};
+
+struct ocrdma_reg_nsmr_cont {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr cmd;
+
+	u32 lrkey;
+	u32 num_pbl_offset;
+	u32 last;
+
+	struct ocrdma_pa pbl[MAX_OCRDMA_NSMR_PBL];
+};
+
+struct ocrdma_pbe {
+	u32 pa_hi;
+	u32 pa_lo;
+};
+
+enum {
+	OCRDMA_REG_NSMR_RSP_NUM_PBL_SHIFT	= 16,
+	OCRDMA_REG_NSMR_RSP_NUM_PBL_MASK	= 0xFFFF0000
+};
+struct ocrdma_reg_nsmr_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 lrkey;
+	u32 num_pbl;
+};
+
+enum {
+	OCRDMA_REG_NSMR_CONT_RSP_LRKEY_INDEX_SHIFT	= 0,
+	OCRDMA_REG_NSMR_CONT_RSP_LRKEY_INDEX_MASK	= 0xFFFFFF,
+	OCRDMA_REG_NSMR_CONT_RSP_LRKEY_SHIFT		= 24,
+	OCRDMA_REG_NSMR_CONT_RSP_LRKEY_MASK		= 0xFF <<
+					OCRDMA_REG_NSMR_CONT_RSP_LRKEY_SHIFT,
+
+	OCRDMA_REG_NSMR_CONT_RSP_NUM_PBL_SHIFT		= 16,
+	OCRDMA_REG_NSMR_CONT_RSP_NUM_PBL_MASK		= 0xFFFF <<
+					OCRDMA_REG_NSMR_CONT_RSP_NUM_PBL_SHIFT
+};
+
+struct ocrdma_reg_nsmr_cont_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 lrkey_key_index;
+	u32 num_pbl;
+};
+
+enum {
+	OCRDMA_ALLOC_MW_PD_ID_SHIFT	= 0,
+	OCRDMA_ALLOC_MW_PD_ID_MASK	= 0xFFFF
+};
+
+struct ocrdma_alloc_mw {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 pdid;
+};
+
+enum {
+	OCRDMA_ALLOC_MW_RSP_LRKEY_INDEX_SHIFT	= 0,
+	OCRDMA_ALLOC_MW_RSP_LRKEY_INDEX_MASK	= 0xFFFFFF
+};
+
+struct ocrdma_alloc_mw_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+
+	u32 lrkey_index;
+};
+
+struct ocrdma_attach_mcast {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 qp_id;
+	u8 mgid[16];
+	u32 mac_b0_to_b3;
+	u32 vlan_mac_b4_to_b5;
+};
+
+struct ocrdma_attach_mcast_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+struct ocrdma_detach_mcast {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 qp_id;
+	u8 mgid[16];
+	u32 mac_b0_to_b3;
+	u32 vlan_mac_b4_to_b5;
+};
+
+struct ocrdma_detach_mcast_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+enum {
+	OCRDMA_CREATE_AH_NUM_PAGES_SHIFT	= 19,
+	OCRDMA_CREATE_AH_NUM_PAGES_MASK		= 0xF <<
+					OCRDMA_CREATE_AH_NUM_PAGES_SHIFT,
+
+	OCRDMA_CREATE_AH_PAGE_SIZE_SHIFT	= 16,
+	OCRDMA_CREATE_AH_PAGE_SIZE_MASK		= 0x7 <<
+					OCRDMA_CREATE_AH_PAGE_SIZE_SHIFT,
+
+	OCRDMA_CREATE_AH_ENTRY_SIZE_SHIFT	= 23,
+	OCRDMA_CREATE_AH_ENTRY_SIZE_MASK	= 0x1FF <<
+					OCRDMA_CREATE_AH_ENTRY_SIZE_SHIFT,
+};
+
+#define OCRDMA_AH_TBL_PAGES 8
+
+struct ocrdma_create_ah_tbl {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+
+	u32 ah_conf;
+	struct ocrdma_pa tbl_addr[8];
+};
+
+struct ocrdma_create_ah_tbl_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+	u32 ahid;
+};
+
+struct ocrdma_delete_ah_tbl {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_hdr req;
+	u32 ahid;
+};
+
+struct ocrdma_delete_ah_tbl_rsp {
+	struct ocrdma_mqe_hdr hdr;
+	struct ocrdma_mbx_rsp rsp;
+};
+
+enum {
+	OCRDMA_EQE_VALID_SHIFT		= 0,
+	OCRDMA_EQE_VALID_MASK		= Bit(0),
+	OCRDMA_EQE_FOR_CQE_MASK		= 0xFFFE,
+	OCRDMA_EQE_RESOURCE_ID_SHIFT	= 16,
+	OCRDMA_EQE_RESOURCE_ID_MASK	= 0xFFFF <<
+				OCRDMA_EQE_RESOURCE_ID_SHIFT,
+};
+
+struct ocrdma_eqe {
+	u32 id_valid;
+};
+
+enum OCRDMA_CQE_STATUS {
+	OCRDMA_CQE_SUCCESS = 0,
+	OCRDMA_CQE_LOC_LEN_ERR,
+	OCRDMA_CQE_LOC_QP_OP_ERR,
+	OCRDMA_CQE_LOC_EEC_OP_ERR,
+	OCRDMA_CQE_LOC_PROT_ERR,
+	OCRDMA_CQE_WR_FLUSH_ERR,
+	OCRDMA_CQE_MW_BIND_ERR,
+	OCRDMA_CQE_BAD_RESP_ERR,
+	OCRDMA_CQE_LOC_ACCESS_ERR,
+	OCRDMA_CQE_REM_INV_REQ_ERR,
+	OCRDMA_CQE_REM_ACCESS_ERR,
+	OCRDMA_CQE_REM_OP_ERR,
+	OCRDMA_CQE_RETRY_EXC_ERR,
+	OCRDMA_CQE_RNR_RETRY_EXC_ERR,
+	OCRDMA_CQE_LOC_RDD_VIOL_ERR,
+	OCRDMA_CQE_REM_INV_RD_REQ_ERR,
+	OCRDMA_CQE_REM_ABORT_ERR,
+	OCRDMA_CQE_INV_EECN_ERR,
+	OCRDMA_CQE_INV_EEC_STATE_ERR,
+	OCRDMA_CQE_FATAL_ERR,
+	OCRDMA_CQE_RESP_TIMEOUT_ERR,
+	OCRDMA_CQE_GENERAL_ERR
+};
+
+enum {
+	/* w0 */
+	OCRDMA_CQE_WQEIDX_SHIFT		= 0,
+	OCRDMA_CQE_WQEIDX_MASK		= 0xFFFF,
+
+	/* w1 */
+	OCRDMA_CQE_UD_XFER_LEN_SHIFT	= 16,
+	OCRDMA_CQE_PKEY_SHIFT		= 0,
+	OCRDMA_CQE_PKEY_MASK		= 0xFFFF,
+
+	/* w2 */
+	OCRDMA_CQE_QPN_SHIFT		= 0,
+	OCRDMA_CQE_QPN_MASK		= 0x0000FFFF,
+
+	OCRDMA_CQE_BUFTAG_SHIFT		= 16,
+	OCRDMA_CQE_BUFTAG_MASK		= 0xFFFF << OCRDMA_CQE_BUFTAG_SHIFT,
+
+	/* w3 */
+	OCRDMA_CQE_UD_STATUS_SHIFT	= 24,
+	OCRDMA_CQE_UD_STATUS_MASK	= 0x7 << OCRDMA_CQE_UD_STATUS_SHIFT,
+	OCRDMA_CQE_STATUS_SHIFT		= 16,
+	OCRDMA_CQE_STATUS_MASK		= 0xFF << OCRDMA_CQE_STATUS_SHIFT,
+	OCRDMA_CQE_VALID		= Bit(31),
+	OCRDMA_CQE_INVALIDATE		= Bit(30),
+	OCRDMA_CQE_QTYPE		= Bit(29),
+	OCRDMA_CQE_IMM			= Bit(28),
+	OCRDMA_CQE_WRITE_IMM		= Bit(27),
+	OCRDMA_CQE_QTYPE_SQ		= 0,
+	OCRDMA_CQE_QTYPE_RQ		= 1,
+	OCRDMA_CQE_SRCQP_MASK		= 0xFFFFFF
+};
+
+struct ocrdma_cqe {
+	union {
+		/* w0 to w2 */
+		struct {
+			u32 wqeidx;
+			u32 bytes_xfered;
+			u32 qpn;
+		} wq;
+		struct {
+			u32 lkey_immdt;
+			u32 rxlen;
+			u32 buftag_qpn;
+		} rq;
+		struct {
+			u32 lkey_immdt;
+			u32 rxlen_pkey;
+			u32 buftag_qpn;
+		} ud;
+		struct {
+			u32 word_0;
+			u32 word_1;
+			u32 qpn;
+		} cmn;
+	};
+	u32 flags_status_srcqpn;	/* w3 */
+};
+
+struct ocrdma_sge {
+	u32 addr_hi;
+	u32 addr_lo;
+	u32 lrkey;
+	u32 len;
+};
+
+enum {
+	OCRDMA_FLAG_SIG		= 0x1,
+	OCRDMA_FLAG_INV		= 0x2,
+	OCRDMA_FLAG_FENCE_L	= 0x4,
+	OCRDMA_FLAG_FENCE_R	= 0x8,
+	OCRDMA_FLAG_SOLICIT	= 0x10,
+	OCRDMA_FLAG_IMM		= 0x20,
+
+	/* Stag flags */
+	OCRDMA_LKEY_FLAG_LOCAL_WR	= 0x1,
+	OCRDMA_LKEY_FLAG_REMOTE_RD	= 0x2,
+	OCRDMA_LKEY_FLAG_REMOTE_WR	= 0x4,
+	OCRDMA_LKEY_FLAG_VATO		= 0x8,
+};
+
+enum OCRDMA_WQE_OPCODE {
+	OCRDMA_WRITE		= 0x06,
+	OCRDMA_READ		= 0x0C,
+	OCRDMA_RESV0		= 0x02,
+	OCRDMA_SEND		= 0x00,
+	OCRDMA_CMP_SWP		= 0x14,
+	OCRDMA_BIND_MW		= 0x10,
+	OCRDMA_FR_MR            = 0x11,
+	OCRDMA_RESV1		= 0x0A,
+	OCRDMA_LKEY_INV		= 0x15,
+	OCRDMA_FETCH_ADD	= 0x13,
+	OCRDMA_POST_RQ		= 0x12
+};
+
+enum {
+	OCRDMA_TYPE_INLINE	= 0x0,
+	OCRDMA_TYPE_LKEY	= 0x1,
+};
+
+enum {
+	OCRDMA_WQE_OPCODE_SHIFT		= 0,
+	OCRDMA_WQE_OPCODE_MASK		= 0x0000001F,
+	OCRDMA_WQE_FLAGS_SHIFT		= 5,
+	OCRDMA_WQE_TYPE_SHIFT		= 16,
+	OCRDMA_WQE_TYPE_MASK		= 0x00030000,
+	OCRDMA_WQE_SIZE_SHIFT		= 18,
+	OCRDMA_WQE_SIZE_MASK		= 0xFF,
+	OCRDMA_WQE_NXT_WQE_SIZE_SHIFT	= 25,
+
+	OCRDMA_WQE_LKEY_FLAGS_SHIFT	= 0,
+	OCRDMA_WQE_LKEY_FLAGS_MASK	= 0xF
+};
+
+/* header WQE for all the SQ and RQ operations */
+struct ocrdma_hdr_wqe {
+	u32 cw;
+	union {
+		u32 rsvd_tag;
+		u32 rsvd_lkey_flags;
+	};
+	union {
+		u32 immdt;
+		u32 lkey;
+	};
+	u32 total_len;
+};
+
+struct ocrdma_ewqe_ud_hdr {
+	u32 rsvd_dest_qpn;
+	u32 qkey;
+	u32 rsvd_ahid;
+	u32 rsvd;
+};
+
+/* extended wqe followed by hdr_wqe for Fast Memory register */
+struct ocrdma_ewqe_fr {
+	u32 va_hi;
+	u32 va_lo;
+	u32 fbo_hi;
+	u32 fbo_lo;
+	u32 size_sge;
+	u32 num_sges;
+	u32 rsvd;
+	u32 rsvd2;
+};
+
+struct ocrdma_eth_basic {
+	u8 dmac[6];
+	u8 smac[6];
+	__be16 eth_type;
+} __packed;
+
+struct ocrdma_eth_vlan {
+	u8 dmac[6];
+	u8 smac[6];
+	__be16 eth_type;
+	__be16 vlan_tag;
+#define OCRDMA_ROCE_ETH_TYPE 0x8915
+	__be16 roce_eth_type;
+} __packed;
+
+struct ocrdma_grh {
+	__be32	tclass_flow;
+	__be32	pdid_hoplimit;
+	u8	sgid[16];
+	u8	dgid[16];
+	u16	rsvd;
+} __packed;
+
+#define OCRDMA_AV_VALID		Bit(7)
+#define OCRDMA_AV_VLAN_VALID	Bit(1)
+
+struct ocrdma_av {
+	struct ocrdma_eth_vlan eth_hdr;
+	struct ocrdma_grh grh;
+	u32 valid;
+} __packed;
+
+struct ocrdma_rsrc_stats {
+	u32 dpp_pds;
+	u32 non_dpp_pds;
+	u32 rc_dpp_qps;
+	u32 uc_dpp_qps;
+	u32 ud_dpp_qps;
+	u32 rc_non_dpp_qps;
+	u32 rsvd;
+	u32 uc_non_dpp_qps;
+	u32 ud_non_dpp_qps;
+	u32 rsvd1;
+	u32 srqs;
+	u32 rbqs;
+	u32 r64K_nsmr;
+	u32 r64K_to_2M_nsmr;
+	u32 r2M_to_44M_nsmr;
+	u32 r44M_to_1G_nsmr;
+	u32 r1G_to_4G_nsmr;
+	u32 nsmr_count_4G_to_32G;
+	u32 r32G_to_64G_nsmr;
+	u32 r64G_to_128G_nsmr;
+	u32 r128G_to_higher_nsmr;
+	u32 embedded_nsmr;
+	u32 frmr;
+	u32 prefetch_qps;
+	u32 ondemand_qps;
+	u32 phy_mr;
+	u32 mw;
+	u32 rsvd2[7];
+};
+
+struct ocrdma_db_err_stats {
+	u32 sq_doorbell_errors;
+	u32 cq_doorbell_errors;
+	u32 rq_srq_doorbell_errors;
+	u32 cq_overflow_errors;
+	u32 rsvd[4];
+};
+
+struct ocrdma_wqe_stats {
+	u32 large_send_rc_wqes_lo;
+	u32 large_send_rc_wqes_hi;
+	u32 large_write_rc_wqes_lo;
+	u32 large_write_rc_wqes_hi;
+	u32 rsvd[4];
+	u32 read_wqes_lo;
+	u32 read_wqes_hi;
+	u32 frmr_wqes_lo;
+	u32 frmr_wqes_hi;
+	u32 mw_bind_wqes_lo;
+	u32 mw_bind_wqes_hi;
+	u32 invalidate_wqes_lo;
+	u32 invalidate_wqes_hi;
+	u32 rsvd1[2];
+	u32 dpp_wqe_drops;
+	u32 rsvd2[5];
+};
+
+struct ocrdma_tx_stats {
+	u32 send_pkts_lo;
+	u32 send_pkts_hi;
+	u32 write_pkts_lo;
+	u32 write_pkts_hi;
+	u32 read_pkts_lo;
+	u32 read_pkts_hi;
+	u32 read_rsp_pkts_lo;
+	u32 read_rsp_pkts_hi;
+	u32 ack_pkts_lo;
+	u32 ack_pkts_hi;
+	u32 send_bytes_lo;
+	u32 send_bytes_hi;
+	u32 write_bytes_lo;
+	u32 write_bytes_hi;
+	u32 read_req_bytes_lo;
+	u32 read_req_bytes_hi;
+	u32 read_rsp_bytes_lo;
+	u32 read_rsp_bytes_hi;
+	u32 ack_timeouts;
+	u32 rsvd[5];
+};
+
+
+struct ocrdma_tx_qp_err_stats {
+	u32 local_length_errors;
+	u32 local_protection_errors;
+	u32 local_qp_operation_errors;
+	u32 retry_count_exceeded_errors;
+	u32 rnr_retry_count_exceeded_errors;
+	u32 rsvd[3];
+};
+
+struct ocrdma_rx_stats {
+	u32 roce_frame_bytes_lo;
+	u32 roce_frame_bytes_hi;
+	u32 roce_frame_icrc_drops;
+	u32 roce_frame_payload_len_drops;
+	u32 ud_drops;
+	u32 qp1_drops;
+	u32 psn_error_request_packets;
+	u32 psn_error_resp_packets;
+	u32 rnr_nak_timeouts;
+	u32 rnr_nak_receives;
+	u32 roce_frame_rxmt_drops;
+	u32 nak_count_psn_sequence_errors;
+	u32 rc_drop_count_lookup_errors;
+	u32 rq_rnr_naks;
+	u32 srq_rnr_naks;
+	u32 roce_frames_lo;
+	u32 roce_frames_hi;
+	u32 rsvd;
+};
+
+struct ocrdma_rx_qp_err_stats {
+	u32 nak_invalid_requst_errors;
+	u32 nak_remote_operation_errors;
+	u32 nak_count_remote_access_errors;
+	u32 local_length_errors;
+	u32 local_protection_errors;
+	u32 local_qp_operation_errors;
+	u32 rsvd[2];
+};
+
+struct ocrdma_tx_dbg_stats {
+	u32 data[100];
+};
+
+struct ocrdma_rx_dbg_stats {
+	u32 data[200];
+};
+
+struct ocrdma_rdma_stats_req {
+	struct ocrdma_mbx_hdr hdr;
+	u8 reset_stats;
+	u8 rsvd[3];
+} __packed;
+
+struct ocrdma_rdma_stats_resp {
+	struct ocrdma_mbx_hdr hdr;
+	struct ocrdma_rsrc_stats act_rsrc_stats;
+	struct ocrdma_rsrc_stats th_rsrc_stats;
+	struct ocrdma_db_err_stats	db_err_stats;
+	struct ocrdma_wqe_stats		wqe_stats;
+	struct ocrdma_tx_stats		tx_stats;
+	struct ocrdma_tx_qp_err_stats	tx_qp_err_stats;
+	struct ocrdma_rx_stats		rx_stats;
+	struct ocrdma_rx_qp_err_stats	rx_qp_err_stats;
+	struct ocrdma_tx_dbg_stats	tx_dbg_stats;
+	struct ocrdma_rx_dbg_stats	rx_dbg_stats;
+} __packed;
+
+
+struct mgmt_hba_attribs {
+	u8 flashrom_version_string[32];
+	u8 manufacturer_name[32];
+	u32 supported_modes;
+	u32 rsvd0[3];
+	u8 ncsi_ver_string[12];
+	u32 default_extended_timeout;
+	u8 controller_model_number[32];
+	u8 controller_description[64];
+	u8 controller_serial_number[32];
+	u8 ip_version_string[32];
+	u8 firmware_version_string[32];
+	u8 bios_version_string[32];
+	u8 redboot_version_string[32];
+	u8 driver_version_string[32];
+	u8 fw_on_flash_version_string[32];
+	u32 functionalities_supported;
+	u16 max_cdblength;
+	u8 asic_revision;
+	u8 generational_guid[16];
+	u8 hba_port_count;
+	u16 default_link_down_timeout;
+	u8 iscsi_ver_min_max;
+	u8 multifunction_device;
+	u8 cache_valid;
+	u8 hba_status;
+	u8 max_domains_supported;
+	u8 phy_port;
+	u32 firmware_post_status;
+	u32 hba_mtu[8];
+	u32 rsvd1[4];
+};
+
+struct mgmt_controller_attrib {
+	struct mgmt_hba_attribs hba_attribs;
+	u16 pci_vendor_id;
+	u16 pci_device_id;
+	u16 pci_sub_vendor_id;
+	u16 pci_sub_system_id;
+	u8 pci_bus_number;
+	u8 pci_device_number;
+	u8 pci_function_number;
+	u8 interface_type;
+	u64 unique_identifier;
+	u32 rsvd0[5];
+};
+
+struct ocrdma_get_ctrl_attribs_rsp {
+	struct ocrdma_mbx_hdr hdr;
+	struct mgmt_controller_attrib ctrl_attribs;
+};
+
+
+#endif				/* __OCRDMA_SLI_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_stats.c b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
new file mode 100644
index 00000000000..41a9aec9998
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_stats.c
@@ -0,0 +1,616 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2014 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#include <rdma/ib_addr.h>
+#include "ocrdma_stats.h"
+
+static struct dentry *ocrdma_dbgfs_dir;
+
+static int ocrdma_add_stat(char *start, char *pcur,
+				char *name, u64 count)
+{
+	char buff[128] = {0};
+	int cpy_len = 0;
+
+	snprintf(buff, 128, "%s: %llu\n", name, count);
+	cpy_len = strlen(buff);
+
+	if (pcur + cpy_len > start + OCRDMA_MAX_DBGFS_MEM) {
+		pr_err("%s: No space in stats buff\n", __func__);
+		return 0;
+	}
+
+	memcpy(pcur, buff, cpy_len);
+	return cpy_len;
+}
+
+static bool ocrdma_alloc_stats_mem(struct ocrdma_dev *dev)
+{
+	struct stats_mem *mem = &dev->stats_mem;
+
+	/* Alloc mbox command mem*/
+	mem->size = max_t(u32, sizeof(struct ocrdma_rdma_stats_req),
+			sizeof(struct ocrdma_rdma_stats_resp));
+
+	mem->va   = dma_alloc_coherent(&dev->nic_info.pdev->dev, mem->size,
+					 &mem->pa, GFP_KERNEL);
+	if (!mem->va) {
+		pr_err("%s: stats mbox allocation failed\n", __func__);
+		return false;
+	}
+
+	memset(mem->va, 0, mem->size);
+
+	/* Alloc debugfs mem */
+	mem->debugfs_mem = kzalloc(OCRDMA_MAX_DBGFS_MEM, GFP_KERNEL);
+	if (!mem->debugfs_mem) {
+		pr_err("%s: stats debugfs mem allocation failed\n", __func__);
+		return false;
+	}
+
+	return true;
+}
+
+static void ocrdma_release_stats_mem(struct ocrdma_dev *dev)
+{
+	struct stats_mem *mem = &dev->stats_mem;
+
+	if (mem->va)
+		dma_free_coherent(&dev->nic_info.pdev->dev, mem->size,
+				  mem->va, mem->pa);
+	kfree(mem->debugfs_mem);
+}
+
+static char *ocrdma_resource_stats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+			(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_rsrc_stats *rsrc_stats = &rdma_stats->act_rsrc_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "active_dpp_pds",
+				(u64)rsrc_stats->dpp_pds);
+	pcur += ocrdma_add_stat(stats, pcur, "active_non_dpp_pds",
+				(u64)rsrc_stats->non_dpp_pds);
+	pcur += ocrdma_add_stat(stats, pcur, "active_rc_dpp_qps",
+				(u64)rsrc_stats->rc_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_uc_dpp_qps",
+				(u64)rsrc_stats->uc_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_ud_dpp_qps",
+				(u64)rsrc_stats->ud_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_rc_non_dpp_qps",
+				(u64)rsrc_stats->rc_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_uc_non_dpp_qps",
+				(u64)rsrc_stats->uc_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_ud_non_dpp_qps",
+				(u64)rsrc_stats->ud_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_srqs",
+				(u64)rsrc_stats->srqs);
+	pcur += ocrdma_add_stat(stats, pcur, "active_rbqs",
+				(u64)rsrc_stats->rbqs);
+	pcur += ocrdma_add_stat(stats, pcur, "active_64K_nsmr",
+				(u64)rsrc_stats->r64K_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_64K_to_2M_nsmr",
+				(u64)rsrc_stats->r64K_to_2M_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_2M_to_44M_nsmr",
+				(u64)rsrc_stats->r2M_to_44M_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_44M_to_1G_nsmr",
+				(u64)rsrc_stats->r44M_to_1G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_1G_to_4G_nsmr",
+				(u64)rsrc_stats->r1G_to_4G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_nsmr_count_4G_to_32G",
+				(u64)rsrc_stats->nsmr_count_4G_to_32G);
+	pcur += ocrdma_add_stat(stats, pcur, "active_32G_to_64G_nsmr",
+				(u64)rsrc_stats->r32G_to_64G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_64G_to_128G_nsmr",
+				(u64)rsrc_stats->r64G_to_128G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_128G_to_higher_nsmr",
+				(u64)rsrc_stats->r128G_to_higher_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_embedded_nsmr",
+				(u64)rsrc_stats->embedded_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_frmr",
+				(u64)rsrc_stats->frmr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_prefetch_qps",
+				(u64)rsrc_stats->prefetch_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_ondemand_qps",
+				(u64)rsrc_stats->ondemand_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "active_phy_mr",
+				(u64)rsrc_stats->phy_mr);
+	pcur += ocrdma_add_stat(stats, pcur, "active_mw",
+				(u64)rsrc_stats->mw);
+
+	/* Print the threshold stats */
+	rsrc_stats = &rdma_stats->th_rsrc_stats;
+
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_dpp_pds",
+				(u64)rsrc_stats->dpp_pds);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_non_dpp_pds",
+				(u64)rsrc_stats->non_dpp_pds);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_rc_dpp_qps",
+				(u64)rsrc_stats->rc_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_uc_dpp_qps",
+				(u64)rsrc_stats->uc_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_ud_dpp_qps",
+				(u64)rsrc_stats->ud_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_rc_non_dpp_qps",
+				(u64)rsrc_stats->rc_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_uc_non_dpp_qps",
+				(u64)rsrc_stats->uc_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_ud_non_dpp_qps",
+				(u64)rsrc_stats->ud_non_dpp_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_srqs",
+				(u64)rsrc_stats->srqs);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_rbqs",
+				(u64)rsrc_stats->rbqs);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_64K_nsmr",
+				(u64)rsrc_stats->r64K_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_64K_to_2M_nsmr",
+				(u64)rsrc_stats->r64K_to_2M_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_2M_to_44M_nsmr",
+				(u64)rsrc_stats->r2M_to_44M_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_44M_to_1G_nsmr",
+				(u64)rsrc_stats->r44M_to_1G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_1G_to_4G_nsmr",
+				(u64)rsrc_stats->r1G_to_4G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_nsmr_count_4G_to_32G",
+				(u64)rsrc_stats->nsmr_count_4G_to_32G);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_32G_to_64G_nsmr",
+				(u64)rsrc_stats->r32G_to_64G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_64G_to_128G_nsmr",
+				(u64)rsrc_stats->r64G_to_128G_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_128G_to_higher_nsmr",
+				(u64)rsrc_stats->r128G_to_higher_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_embedded_nsmr",
+				(u64)rsrc_stats->embedded_nsmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_frmr",
+				(u64)rsrc_stats->frmr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_prefetch_qps",
+				(u64)rsrc_stats->prefetch_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_ondemand_qps",
+				(u64)rsrc_stats->ondemand_qps);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_phy_mr",
+				(u64)rsrc_stats->phy_mr);
+	pcur += ocrdma_add_stat(stats, pcur, "threshold_mw",
+				(u64)rsrc_stats->mw);
+	return stats;
+}
+
+static char *ocrdma_rx_stats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_rx_stats *rx_stats = &rdma_stats->rx_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat
+		(stats, pcur, "roce_frame_bytes",
+		 convert_to_64bit(rx_stats->roce_frame_bytes_lo,
+		 rx_stats->roce_frame_bytes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "roce_frame_icrc_drops",
+				(u64)rx_stats->roce_frame_icrc_drops);
+	pcur += ocrdma_add_stat(stats, pcur, "roce_frame_payload_len_drops",
+				(u64)rx_stats->roce_frame_payload_len_drops);
+	pcur += ocrdma_add_stat(stats, pcur, "ud_drops",
+				(u64)rx_stats->ud_drops);
+	pcur += ocrdma_add_stat(stats, pcur, "qp1_drops",
+				(u64)rx_stats->qp1_drops);
+	pcur += ocrdma_add_stat(stats, pcur, "psn_error_request_packets",
+				(u64)rx_stats->psn_error_request_packets);
+	pcur += ocrdma_add_stat(stats, pcur, "psn_error_resp_packets",
+				(u64)rx_stats->psn_error_resp_packets);
+	pcur += ocrdma_add_stat(stats, pcur, "rnr_nak_timeouts",
+				(u64)rx_stats->rnr_nak_timeouts);
+	pcur += ocrdma_add_stat(stats, pcur, "rnr_nak_receives",
+				(u64)rx_stats->rnr_nak_receives);
+	pcur += ocrdma_add_stat(stats, pcur, "roce_frame_rxmt_drops",
+				(u64)rx_stats->roce_frame_rxmt_drops);
+	pcur += ocrdma_add_stat(stats, pcur, "nak_count_psn_sequence_errors",
+				(u64)rx_stats->nak_count_psn_sequence_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "rc_drop_count_lookup_errors",
+				(u64)rx_stats->rc_drop_count_lookup_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "rq_rnr_naks",
+				(u64)rx_stats->rq_rnr_naks);
+	pcur += ocrdma_add_stat(stats, pcur, "srq_rnr_naks",
+				(u64)rx_stats->srq_rnr_naks);
+	pcur += ocrdma_add_stat(stats, pcur, "roce_frames",
+				convert_to_64bit(rx_stats->roce_frames_lo,
+						 rx_stats->roce_frames_hi));
+
+	return stats;
+}
+
+static char *ocrdma_tx_stats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_tx_stats *tx_stats = &rdma_stats->tx_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "send_pkts",
+				convert_to_64bit(tx_stats->send_pkts_lo,
+						 tx_stats->send_pkts_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "write_pkts",
+				convert_to_64bit(tx_stats->write_pkts_lo,
+						 tx_stats->write_pkts_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "read_pkts",
+				convert_to_64bit(tx_stats->read_pkts_lo,
+						 tx_stats->read_pkts_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "read_rsp_pkts",
+				convert_to_64bit(tx_stats->read_rsp_pkts_lo,
+						 tx_stats->read_rsp_pkts_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "ack_pkts",
+				convert_to_64bit(tx_stats->ack_pkts_lo,
+						 tx_stats->ack_pkts_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "send_bytes",
+				convert_to_64bit(tx_stats->send_bytes_lo,
+						 tx_stats->send_bytes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "write_bytes",
+				convert_to_64bit(tx_stats->write_bytes_lo,
+						 tx_stats->write_bytes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "read_req_bytes",
+				convert_to_64bit(tx_stats->read_req_bytes_lo,
+						 tx_stats->read_req_bytes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "read_rsp_bytes",
+				convert_to_64bit(tx_stats->read_rsp_bytes_lo,
+						 tx_stats->read_rsp_bytes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "ack_timeouts",
+				(u64)tx_stats->ack_timeouts);
+
+	return stats;
+}
+
+static char *ocrdma_wqe_stats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_wqe_stats	*wqe_stats = &rdma_stats->wqe_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "large_send_rc_wqes",
+		convert_to_64bit(wqe_stats->large_send_rc_wqes_lo,
+				 wqe_stats->large_send_rc_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "large_write_rc_wqes",
+		convert_to_64bit(wqe_stats->large_write_rc_wqes_lo,
+				 wqe_stats->large_write_rc_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "read_wqes",
+				convert_to_64bit(wqe_stats->read_wqes_lo,
+						 wqe_stats->read_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "frmr_wqes",
+				convert_to_64bit(wqe_stats->frmr_wqes_lo,
+						 wqe_stats->frmr_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "mw_bind_wqes",
+				convert_to_64bit(wqe_stats->mw_bind_wqes_lo,
+						 wqe_stats->mw_bind_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "invalidate_wqes",
+		convert_to_64bit(wqe_stats->invalidate_wqes_lo,
+				 wqe_stats->invalidate_wqes_hi));
+	pcur += ocrdma_add_stat(stats, pcur, "dpp_wqe_drops",
+				(u64)wqe_stats->dpp_wqe_drops);
+	return stats;
+}
+
+static char *ocrdma_db_errstats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_db_err_stats *db_err_stats = &rdma_stats->db_err_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "sq_doorbell_errors",
+				(u64)db_err_stats->sq_doorbell_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "cq_doorbell_errors",
+				(u64)db_err_stats->cq_doorbell_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "rq_srq_doorbell_errors",
+				(u64)db_err_stats->rq_srq_doorbell_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "cq_overflow_errors",
+				(u64)db_err_stats->cq_overflow_errors);
+	return stats;
+}
+
+static char *ocrdma_rxqp_errstats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_rx_qp_err_stats *rx_qp_err_stats =
+		 &rdma_stats->rx_qp_err_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "nak_invalid_requst_errors",
+			(u64)rx_qp_err_stats->nak_invalid_requst_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "nak_remote_operation_errors",
+			(u64)rx_qp_err_stats->nak_remote_operation_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "nak_count_remote_access_errors",
+			(u64)rx_qp_err_stats->nak_count_remote_access_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "local_length_errors",
+			(u64)rx_qp_err_stats->local_length_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "local_protection_errors",
+			(u64)rx_qp_err_stats->local_protection_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "local_qp_operation_errors",
+			(u64)rx_qp_err_stats->local_qp_operation_errors);
+	return stats;
+}
+
+static char *ocrdma_txqp_errstats(struct ocrdma_dev *dev)
+{
+	char *stats = dev->stats_mem.debugfs_mem, *pcur;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_tx_qp_err_stats *tx_qp_err_stats =
+		&rdma_stats->tx_qp_err_stats;
+
+	memset(stats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	pcur = stats;
+	pcur += ocrdma_add_stat(stats, pcur, "local_length_errors",
+			(u64)tx_qp_err_stats->local_length_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "local_protection_errors",
+			(u64)tx_qp_err_stats->local_protection_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "local_qp_operation_errors",
+			(u64)tx_qp_err_stats->local_qp_operation_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "retry_count_exceeded_errors",
+			(u64)tx_qp_err_stats->retry_count_exceeded_errors);
+	pcur += ocrdma_add_stat(stats, pcur, "rnr_retry_count_exceeded_errors",
+			(u64)tx_qp_err_stats->rnr_retry_count_exceeded_errors);
+	return stats;
+}
+
+static char *ocrdma_tx_dbg_stats(struct ocrdma_dev *dev)
+{
+	int i;
+	char *pstats = dev->stats_mem.debugfs_mem;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_tx_dbg_stats *tx_dbg_stats =
+		&rdma_stats->tx_dbg_stats;
+
+	memset(pstats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	for (i = 0; i < 100; i++)
+		pstats += snprintf(pstats, 80, "DW[%d] = 0x%x\n", i,
+				 tx_dbg_stats->data[i]);
+
+	return dev->stats_mem.debugfs_mem;
+}
+
+static char *ocrdma_rx_dbg_stats(struct ocrdma_dev *dev)
+{
+	int i;
+	char *pstats = dev->stats_mem.debugfs_mem;
+	struct ocrdma_rdma_stats_resp *rdma_stats =
+		(struct ocrdma_rdma_stats_resp *)dev->stats_mem.va;
+	struct ocrdma_rx_dbg_stats *rx_dbg_stats =
+		&rdma_stats->rx_dbg_stats;
+
+	memset(pstats, 0, (OCRDMA_MAX_DBGFS_MEM));
+
+	for (i = 0; i < 200; i++)
+		pstats += snprintf(pstats, 80, "DW[%d] = 0x%x\n", i,
+				 rx_dbg_stats->data[i]);
+
+	return dev->stats_mem.debugfs_mem;
+}
+
+static void ocrdma_update_stats(struct ocrdma_dev *dev)
+{
+	ulong now = jiffies, secs;
+	int status = 0;
+
+	secs = jiffies_to_msecs(now - dev->last_stats_time) / 1000U;
+	if (secs) {
+		/* update */
+		status = ocrdma_mbx_rdma_stats(dev, false);
+		if (status)
+			pr_err("%s: stats mbox failed with status = %d\n",
+			       __func__, status);
+		dev->last_stats_time = jiffies;
+	}
+}
+
+static ssize_t ocrdma_dbgfs_ops_read(struct file *filp, char __user *buffer,
+					size_t usr_buf_len, loff_t *ppos)
+{
+	struct ocrdma_stats *pstats = filp->private_data;
+	struct ocrdma_dev *dev = pstats->dev;
+	ssize_t status = 0;
+	char *data = NULL;
+
+	/* No partial reads */
+	if (*ppos != 0)
+		return 0;
+
+	mutex_lock(&dev->stats_lock);
+
+	ocrdma_update_stats(dev);
+
+	switch (pstats->type) {
+	case OCRDMA_RSRC_STATS:
+		data = ocrdma_resource_stats(dev);
+		break;
+	case OCRDMA_RXSTATS:
+		data = ocrdma_rx_stats(dev);
+		break;
+	case OCRDMA_WQESTATS:
+		data = ocrdma_wqe_stats(dev);
+		break;
+	case OCRDMA_TXSTATS:
+		data = ocrdma_tx_stats(dev);
+		break;
+	case OCRDMA_DB_ERRSTATS:
+		data = ocrdma_db_errstats(dev);
+		break;
+	case OCRDMA_RXQP_ERRSTATS:
+		data = ocrdma_rxqp_errstats(dev);
+		break;
+	case OCRDMA_TXQP_ERRSTATS:
+		data = ocrdma_txqp_errstats(dev);
+		break;
+	case OCRDMA_TX_DBG_STATS:
+		data = ocrdma_tx_dbg_stats(dev);
+		break;
+	case OCRDMA_RX_DBG_STATS:
+		data = ocrdma_rx_dbg_stats(dev);
+		break;
+
+	default:
+		status = -EFAULT;
+		goto exit;
+	}
+
+	if (usr_buf_len < strlen(data)) {
+		status = -ENOSPC;
+		goto exit;
+	}
+
+	status = simple_read_from_buffer(buffer, usr_buf_len, ppos, data,
+					 strlen(data));
+exit:
+	mutex_unlock(&dev->stats_lock);
+	return status;
+}
+
+static const struct file_operations ocrdma_dbg_ops = {
+	.owner = THIS_MODULE,
+	.open = simple_open,
+	.read = ocrdma_dbgfs_ops_read,
+};
+
+void ocrdma_add_port_stats(struct ocrdma_dev *dev)
+{
+	if (!ocrdma_dbgfs_dir)
+		return;
+
+	/* Create post stats base dir */
+	dev->dir = debugfs_create_dir(dev->ibdev.name, ocrdma_dbgfs_dir);
+	if (!dev->dir)
+		goto err;
+
+	dev->rsrc_stats.type = OCRDMA_RSRC_STATS;
+	dev->rsrc_stats.dev = dev;
+	if (!debugfs_create_file("resource_stats", S_IRUSR, dev->dir,
+				 &dev->rsrc_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->rx_stats.type = OCRDMA_RXSTATS;
+	dev->rx_stats.dev = dev;
+	if (!debugfs_create_file("rx_stats", S_IRUSR, dev->dir,
+				 &dev->rx_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->wqe_stats.type = OCRDMA_WQESTATS;
+	dev->wqe_stats.dev = dev;
+	if (!debugfs_create_file("wqe_stats", S_IRUSR, dev->dir,
+				 &dev->wqe_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->tx_stats.type = OCRDMA_TXSTATS;
+	dev->tx_stats.dev = dev;
+	if (!debugfs_create_file("tx_stats", S_IRUSR, dev->dir,
+				 &dev->tx_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->db_err_stats.type = OCRDMA_DB_ERRSTATS;
+	dev->db_err_stats.dev = dev;
+	if (!debugfs_create_file("db_err_stats", S_IRUSR, dev->dir,
+				 &dev->db_err_stats, &ocrdma_dbg_ops))
+		goto err;
+
+
+	dev->tx_qp_err_stats.type = OCRDMA_TXQP_ERRSTATS;
+	dev->tx_qp_err_stats.dev = dev;
+	if (!debugfs_create_file("tx_qp_err_stats", S_IRUSR, dev->dir,
+				 &dev->tx_qp_err_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->rx_qp_err_stats.type = OCRDMA_RXQP_ERRSTATS;
+	dev->rx_qp_err_stats.dev = dev;
+	if (!debugfs_create_file("rx_qp_err_stats", S_IRUSR, dev->dir,
+				 &dev->rx_qp_err_stats, &ocrdma_dbg_ops))
+		goto err;
+
+
+	dev->tx_dbg_stats.type = OCRDMA_TX_DBG_STATS;
+	dev->tx_dbg_stats.dev = dev;
+	if (!debugfs_create_file("tx_dbg_stats", S_IRUSR, dev->dir,
+				 &dev->tx_dbg_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	dev->rx_dbg_stats.type = OCRDMA_RX_DBG_STATS;
+	dev->rx_dbg_stats.dev = dev;
+	if (!debugfs_create_file("rx_dbg_stats", S_IRUSR, dev->dir,
+				 &dev->rx_dbg_stats, &ocrdma_dbg_ops))
+		goto err;
+
+	/* Now create dma_mem for stats mbx command */
+	if (!ocrdma_alloc_stats_mem(dev))
+		goto err;
+
+	mutex_init(&dev->stats_lock);
+
+	return;
+err:
+	ocrdma_release_stats_mem(dev);
+	debugfs_remove_recursive(dev->dir);
+	dev->dir = NULL;
+}
+
+void ocrdma_rem_port_stats(struct ocrdma_dev *dev)
+{
+	if (!dev->dir)
+		return;
+	mutex_destroy(&dev->stats_lock);
+	ocrdma_release_stats_mem(dev);
+	debugfs_remove(dev->dir);
+}
+
+void ocrdma_init_debugfs(void)
+{
+	/* Create base dir in debugfs root dir */
+	ocrdma_dbgfs_dir = debugfs_create_dir("ocrdma", NULL);
+}
+
+void ocrdma_rem_debugfs(void)
+{
+	debugfs_remove_recursive(ocrdma_dbgfs_dir);
+}
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_stats.h b/drivers/infiniband/hw/ocrdma/ocrdma_stats.h
new file mode 100644
index 00000000000..5f5e20c46d7
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_stats.h
@@ -0,0 +1,54 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2014 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_STATS_H__
+#define __OCRDMA_STATS_H__
+
+#include <linux/debugfs.h>
+#include "ocrdma.h"
+#include "ocrdma_hw.h"
+
+#define OCRDMA_MAX_DBGFS_MEM 4096
+
+enum OCRDMA_STATS_TYPE {
+	OCRDMA_RSRC_STATS,
+	OCRDMA_RXSTATS,
+	OCRDMA_WQESTATS,
+	OCRDMA_TXSTATS,
+	OCRDMA_DB_ERRSTATS,
+	OCRDMA_RXQP_ERRSTATS,
+	OCRDMA_TXQP_ERRSTATS,
+	OCRDMA_TX_DBG_STATS,
+	OCRDMA_RX_DBG_STATS
+};
+
+void ocrdma_rem_debugfs(void);
+void ocrdma_init_debugfs(void);
+void ocrdma_rem_port_stats(struct ocrdma_dev *dev);
+void ocrdma_add_port_stats(struct ocrdma_dev *dev);
+
+#endif	/* __OCRDMA_STATS_H__ */
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
new file mode 100644
index 00000000000..edf6211d84b
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.c
@@ -0,0 +1,3054 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#include <linux/dma-mapping.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_user_verbs.h>
+#include <rdma/iw_cm.h>
+#include <rdma/ib_umem.h>
+#include <rdma/ib_addr.h>
+
+#include "ocrdma.h"
+#include "ocrdma_hw.h"
+#include "ocrdma_verbs.h"
+#include "ocrdma_abi.h"
+
+int ocrdma_query_pkey(struct ib_device *ibdev, u8 port, u16 index, u16 *pkey)
+{
+	if (index > 1)
+		return -EINVAL;
+
+	*pkey = 0xffff;
+	return 0;
+}
+
+int ocrdma_query_gid(struct ib_device *ibdev, u8 port,
+		     int index, union ib_gid *sgid)
+{
+	struct ocrdma_dev *dev;
+
+	dev = get_ocrdma_dev(ibdev);
+	memset(sgid, 0, sizeof(*sgid));
+	if (index > OCRDMA_MAX_SGID)
+		return -EINVAL;
+
+	memcpy(sgid, &dev->sgid_tbl[index], sizeof(*sgid));
+
+	return 0;
+}
+
+int ocrdma_query_device(struct ib_device *ibdev, struct ib_device_attr *attr)
+{
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibdev);
+
+	memset(attr, 0, sizeof *attr);
+	memcpy(&attr->fw_ver, &dev->attr.fw_ver[0],
+	       min(sizeof(dev->attr.fw_ver), sizeof(attr->fw_ver)));
+	ocrdma_get_guid(dev, (u8 *)&attr->sys_image_guid);
+	attr->max_mr_size = ~0ull;
+	attr->page_size_cap = 0xffff000;
+	attr->vendor_id = dev->nic_info.pdev->vendor;
+	attr->vendor_part_id = dev->nic_info.pdev->device;
+	attr->hw_ver = 0;
+	attr->max_qp = dev->attr.max_qp;
+	attr->max_ah = OCRDMA_MAX_AH;
+	attr->max_qp_wr = dev->attr.max_wqe;
+
+	attr->device_cap_flags = IB_DEVICE_CURR_QP_STATE_MOD |
+					IB_DEVICE_RC_RNR_NAK_GEN |
+					IB_DEVICE_SHUTDOWN_PORT |
+					IB_DEVICE_SYS_IMAGE_GUID |
+					IB_DEVICE_LOCAL_DMA_LKEY |
+					IB_DEVICE_MEM_MGT_EXTENSIONS;
+	attr->max_sge = min(dev->attr.max_send_sge, dev->attr.max_srq_sge);
+	attr->max_sge_rd = 0;
+	attr->max_cq = dev->attr.max_cq;
+	attr->max_cqe = dev->attr.max_cqe;
+	attr->max_mr = dev->attr.max_mr;
+	attr->max_mw = dev->attr.max_mw;
+	attr->max_pd = dev->attr.max_pd;
+	attr->atomic_cap = 0;
+	attr->max_fmr = 0;
+	attr->max_map_per_fmr = 0;
+	attr->max_qp_rd_atom =
+	    min(dev->attr.max_ord_per_qp, dev->attr.max_ird_per_qp);
+	attr->max_qp_init_rd_atom = dev->attr.max_ord_per_qp;
+	attr->max_srq = dev->attr.max_srq;
+	attr->max_srq_sge = dev->attr.max_srq_sge;
+	attr->max_srq_wr = dev->attr.max_rqe;
+	attr->local_ca_ack_delay = dev->attr.local_ca_ack_delay;
+	attr->max_fast_reg_page_list_len = 0;
+	attr->max_pkeys = 1;
+	return 0;
+}
+
+static inline void get_link_speed_and_width(struct ocrdma_dev *dev,
+					    u8 *ib_speed, u8 *ib_width)
+{
+	int status;
+	u8 speed;
+
+	status = ocrdma_mbx_get_link_speed(dev, &speed);
+	if (status)
+		speed = OCRDMA_PHYS_LINK_SPEED_ZERO;
+
+	switch (speed) {
+	case OCRDMA_PHYS_LINK_SPEED_1GBPS:
+		*ib_speed = IB_SPEED_SDR;
+		*ib_width = IB_WIDTH_1X;
+		break;
+
+	case OCRDMA_PHYS_LINK_SPEED_10GBPS:
+		*ib_speed = IB_SPEED_QDR;
+		*ib_width = IB_WIDTH_1X;
+		break;
+
+	case OCRDMA_PHYS_LINK_SPEED_20GBPS:
+		*ib_speed = IB_SPEED_DDR;
+		*ib_width = IB_WIDTH_4X;
+		break;
+
+	case OCRDMA_PHYS_LINK_SPEED_40GBPS:
+		*ib_speed = IB_SPEED_QDR;
+		*ib_width = IB_WIDTH_4X;
+		break;
+
+	default:
+		/* Unsupported */
+		*ib_speed = IB_SPEED_SDR;
+		*ib_width = IB_WIDTH_1X;
+	}
+}
+
+int ocrdma_query_port(struct ib_device *ibdev,
+		      u8 port, struct ib_port_attr *props)
+{
+	enum ib_port_state port_state;
+	struct ocrdma_dev *dev;
+	struct net_device *netdev;
+
+	dev = get_ocrdma_dev(ibdev);
+	if (port > 1) {
+		pr_err("%s(%d) invalid_port=0x%x\n", __func__,
+		       dev->id, port);
+		return -EINVAL;
+	}
+	netdev = dev->nic_info.netdev;
+	if (netif_running(netdev) && netif_oper_up(netdev)) {
+		port_state = IB_PORT_ACTIVE;
+		props->phys_state = 5;
+	} else {
+		port_state = IB_PORT_DOWN;
+		props->phys_state = 3;
+	}
+	props->max_mtu = IB_MTU_4096;
+	props->active_mtu = iboe_get_mtu(netdev->mtu);
+	props->lid = 0;
+	props->lmc = 0;
+	props->sm_lid = 0;
+	props->sm_sl = 0;
+	props->state = port_state;
+	props->port_cap_flags =
+	    IB_PORT_CM_SUP |
+	    IB_PORT_REINIT_SUP |
+	    IB_PORT_DEVICE_MGMT_SUP | IB_PORT_VENDOR_CLASS_SUP | IB_PORT_IP_BASED_GIDS;
+	props->gid_tbl_len = OCRDMA_MAX_SGID;
+	props->pkey_tbl_len = 1;
+	props->bad_pkey_cntr = 0;
+	props->qkey_viol_cntr = 0;
+	get_link_speed_and_width(dev, &props->active_speed,
+				 &props->active_width);
+	props->max_msg_sz = 0x80000000;
+	props->max_vl_num = 4;
+	return 0;
+}
+
+int ocrdma_modify_port(struct ib_device *ibdev, u8 port, int mask,
+		       struct ib_port_modify *props)
+{
+	struct ocrdma_dev *dev;
+
+	dev = get_ocrdma_dev(ibdev);
+	if (port > 1) {
+		pr_err("%s(%d) invalid_port=0x%x\n", __func__, dev->id, port);
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int ocrdma_add_mmap(struct ocrdma_ucontext *uctx, u64 phy_addr,
+			   unsigned long len)
+{
+	struct ocrdma_mm *mm;
+
+	mm = kzalloc(sizeof(*mm), GFP_KERNEL);
+	if (mm == NULL)
+		return -ENOMEM;
+	mm->key.phy_addr = phy_addr;
+	mm->key.len = len;
+	INIT_LIST_HEAD(&mm->entry);
+
+	mutex_lock(&uctx->mm_list_lock);
+	list_add_tail(&mm->entry, &uctx->mm_head);
+	mutex_unlock(&uctx->mm_list_lock);
+	return 0;
+}
+
+static void ocrdma_del_mmap(struct ocrdma_ucontext *uctx, u64 phy_addr,
+			    unsigned long len)
+{
+	struct ocrdma_mm *mm, *tmp;
+
+	mutex_lock(&uctx->mm_list_lock);
+	list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
+		if (len != mm->key.len && phy_addr != mm->key.phy_addr)
+			continue;
+
+		list_del(&mm->entry);
+		kfree(mm);
+		break;
+	}
+	mutex_unlock(&uctx->mm_list_lock);
+}
+
+static bool ocrdma_search_mmap(struct ocrdma_ucontext *uctx, u64 phy_addr,
+			      unsigned long len)
+{
+	bool found = false;
+	struct ocrdma_mm *mm;
+
+	mutex_lock(&uctx->mm_list_lock);
+	list_for_each_entry(mm, &uctx->mm_head, entry) {
+		if (len != mm->key.len && phy_addr != mm->key.phy_addr)
+			continue;
+
+		found = true;
+		break;
+	}
+	mutex_unlock(&uctx->mm_list_lock);
+	return found;
+}
+
+static struct ocrdma_pd *_ocrdma_alloc_pd(struct ocrdma_dev *dev,
+					  struct ocrdma_ucontext *uctx,
+					  struct ib_udata *udata)
+{
+	struct ocrdma_pd *pd = NULL;
+	int status = 0;
+
+	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	if (udata && uctx) {
+		pd->dpp_enabled =
+			ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R;
+		pd->num_dpp_qp =
+			pd->dpp_enabled ? OCRDMA_PD_MAX_DPP_ENABLED_QP : 0;
+	}
+
+retry:
+	status = ocrdma_mbx_alloc_pd(dev, pd);
+	if (status) {
+		if (pd->dpp_enabled) {
+			pd->dpp_enabled = false;
+			pd->num_dpp_qp = 0;
+			goto retry;
+		} else {
+			kfree(pd);
+			return ERR_PTR(status);
+		}
+	}
+
+	return pd;
+}
+
+static inline int is_ucontext_pd(struct ocrdma_ucontext *uctx,
+				 struct ocrdma_pd *pd)
+{
+	return (uctx->cntxt_pd == pd ? true : false);
+}
+
+static int _ocrdma_dealloc_pd(struct ocrdma_dev *dev,
+			      struct ocrdma_pd *pd)
+{
+	int status = 0;
+
+	status = ocrdma_mbx_dealloc_pd(dev, pd);
+	kfree(pd);
+	return status;
+}
+
+static int ocrdma_alloc_ucontext_pd(struct ocrdma_dev *dev,
+				    struct ocrdma_ucontext *uctx,
+				    struct ib_udata *udata)
+{
+	int status = 0;
+
+	uctx->cntxt_pd = _ocrdma_alloc_pd(dev, uctx, udata);
+	if (IS_ERR(uctx->cntxt_pd)) {
+		status = PTR_ERR(uctx->cntxt_pd);
+		uctx->cntxt_pd = NULL;
+		goto err;
+	}
+
+	uctx->cntxt_pd->uctx = uctx;
+	uctx->cntxt_pd->ibpd.device = &dev->ibdev;
+err:
+	return status;
+}
+
+static int ocrdma_dealloc_ucontext_pd(struct ocrdma_ucontext *uctx)
+{
+	int status = 0;
+	struct ocrdma_pd *pd = uctx->cntxt_pd;
+	struct ocrdma_dev *dev = get_ocrdma_dev(pd->ibpd.device);
+
+	BUG_ON(uctx->pd_in_use);
+	uctx->cntxt_pd = NULL;
+	status = _ocrdma_dealloc_pd(dev, pd);
+	return status;
+}
+
+static struct ocrdma_pd *ocrdma_get_ucontext_pd(struct ocrdma_ucontext *uctx)
+{
+	struct ocrdma_pd *pd = NULL;
+
+	mutex_lock(&uctx->mm_list_lock);
+	if (!uctx->pd_in_use) {
+		uctx->pd_in_use = true;
+		pd = uctx->cntxt_pd;
+	}
+	mutex_unlock(&uctx->mm_list_lock);
+
+	return pd;
+}
+
+static void ocrdma_release_ucontext_pd(struct ocrdma_ucontext *uctx)
+{
+	mutex_lock(&uctx->mm_list_lock);
+	uctx->pd_in_use = false;
+	mutex_unlock(&uctx->mm_list_lock);
+}
+
+struct ib_ucontext *ocrdma_alloc_ucontext(struct ib_device *ibdev,
+					  struct ib_udata *udata)
+{
+	int status;
+	struct ocrdma_ucontext *ctx;
+	struct ocrdma_alloc_ucontext_resp resp;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibdev);
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	u32 map_len = roundup(sizeof(u32) * 2048, PAGE_SIZE);
+
+	if (!udata)
+		return ERR_PTR(-EFAULT);
+	ctx = kzalloc(sizeof(*ctx), GFP_KERNEL);
+	if (!ctx)
+		return ERR_PTR(-ENOMEM);
+	INIT_LIST_HEAD(&ctx->mm_head);
+	mutex_init(&ctx->mm_list_lock);
+
+	ctx->ah_tbl.va = dma_alloc_coherent(&pdev->dev, map_len,
+					    &ctx->ah_tbl.pa, GFP_KERNEL);
+	if (!ctx->ah_tbl.va) {
+		kfree(ctx);
+		return ERR_PTR(-ENOMEM);
+	}
+	memset(ctx->ah_tbl.va, 0, map_len);
+	ctx->ah_tbl.len = map_len;
+
+	memset(&resp, 0, sizeof(resp));
+	resp.ah_tbl_len = ctx->ah_tbl.len;
+	resp.ah_tbl_page = ctx->ah_tbl.pa;
+
+	status = ocrdma_add_mmap(ctx, resp.ah_tbl_page, resp.ah_tbl_len);
+	if (status)
+		goto map_err;
+
+	status = ocrdma_alloc_ucontext_pd(dev, ctx, udata);
+	if (status)
+		goto pd_err;
+
+	resp.dev_id = dev->id;
+	resp.max_inline_data = dev->attr.max_inline_data;
+	resp.wqe_size = dev->attr.wqe_size;
+	resp.rqe_size = dev->attr.rqe_size;
+	resp.dpp_wqe_size = dev->attr.wqe_size;
+
+	memcpy(resp.fw_ver, dev->attr.fw_ver, sizeof(resp.fw_ver));
+	status = ib_copy_to_udata(udata, &resp, sizeof(resp));
+	if (status)
+		goto cpy_err;
+	return &ctx->ibucontext;
+
+cpy_err:
+pd_err:
+	ocrdma_del_mmap(ctx, ctx->ah_tbl.pa, ctx->ah_tbl.len);
+map_err:
+	dma_free_coherent(&pdev->dev, ctx->ah_tbl.len, ctx->ah_tbl.va,
+			  ctx->ah_tbl.pa);
+	kfree(ctx);
+	return ERR_PTR(status);
+}
+
+int ocrdma_dealloc_ucontext(struct ib_ucontext *ibctx)
+{
+	int status = 0;
+	struct ocrdma_mm *mm, *tmp;
+	struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ibctx);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibctx->device);
+	struct pci_dev *pdev = dev->nic_info.pdev;
+
+	status = ocrdma_dealloc_ucontext_pd(uctx);
+
+	ocrdma_del_mmap(uctx, uctx->ah_tbl.pa, uctx->ah_tbl.len);
+	dma_free_coherent(&pdev->dev, uctx->ah_tbl.len, uctx->ah_tbl.va,
+			  uctx->ah_tbl.pa);
+
+	list_for_each_entry_safe(mm, tmp, &uctx->mm_head, entry) {
+		list_del(&mm->entry);
+		kfree(mm);
+	}
+	kfree(uctx);
+	return status;
+}
+
+int ocrdma_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+	struct ocrdma_ucontext *ucontext = get_ocrdma_ucontext(context);
+	struct ocrdma_dev *dev = get_ocrdma_dev(context->device);
+	unsigned long vm_page = vma->vm_pgoff << PAGE_SHIFT;
+	u64 unmapped_db = (u64) dev->nic_info.unmapped_db;
+	unsigned long len = (vma->vm_end - vma->vm_start);
+	int status = 0;
+	bool found;
+
+	if (vma->vm_start & (PAGE_SIZE - 1))
+		return -EINVAL;
+	found = ocrdma_search_mmap(ucontext, vma->vm_pgoff << PAGE_SHIFT, len);
+	if (!found)
+		return -EINVAL;
+
+	if ((vm_page >= unmapped_db) && (vm_page <= (unmapped_db +
+		dev->nic_info.db_total_size)) &&
+		(len <=	dev->nic_info.db_page_size)) {
+		if (vma->vm_flags & VM_READ)
+			return -EPERM;
+
+		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+		status = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+					    len, vma->vm_page_prot);
+	} else if (dev->nic_info.dpp_unmapped_len &&
+		(vm_page >= (u64) dev->nic_info.dpp_unmapped_addr) &&
+		(vm_page <= (u64) (dev->nic_info.dpp_unmapped_addr +
+			dev->nic_info.dpp_unmapped_len)) &&
+		(len <= dev->nic_info.dpp_unmapped_len)) {
+		if (vma->vm_flags & VM_READ)
+			return -EPERM;
+
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+		status = io_remap_pfn_range(vma, vma->vm_start, vma->vm_pgoff,
+					    len, vma->vm_page_prot);
+	} else {
+		status = remap_pfn_range(vma, vma->vm_start,
+					 vma->vm_pgoff, len, vma->vm_page_prot);
+	}
+	return status;
+}
+
+static int ocrdma_copy_pd_uresp(struct ocrdma_dev *dev, struct ocrdma_pd *pd,
+				struct ib_ucontext *ib_ctx,
+				struct ib_udata *udata)
+{
+	int status;
+	u64 db_page_addr;
+	u64 dpp_page_addr = 0;
+	u32 db_page_size;
+	struct ocrdma_alloc_pd_uresp rsp;
+	struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ib_ctx);
+
+	memset(&rsp, 0, sizeof(rsp));
+	rsp.id = pd->id;
+	rsp.dpp_enabled = pd->dpp_enabled;
+	db_page_addr = ocrdma_get_db_addr(dev, pd->id);
+	db_page_size = dev->nic_info.db_page_size;
+
+	status = ocrdma_add_mmap(uctx, db_page_addr, db_page_size);
+	if (status)
+		return status;
+
+	if (pd->dpp_enabled) {
+		dpp_page_addr = dev->nic_info.dpp_unmapped_addr +
+				(pd->id * PAGE_SIZE);
+		status = ocrdma_add_mmap(uctx, dpp_page_addr,
+				 PAGE_SIZE);
+		if (status)
+			goto dpp_map_err;
+		rsp.dpp_page_addr_hi = upper_32_bits(dpp_page_addr);
+		rsp.dpp_page_addr_lo = dpp_page_addr;
+	}
+
+	status = ib_copy_to_udata(udata, &rsp, sizeof(rsp));
+	if (status)
+		goto ucopy_err;
+
+	pd->uctx = uctx;
+	return 0;
+
+ucopy_err:
+	if (pd->dpp_enabled)
+		ocrdma_del_mmap(pd->uctx, dpp_page_addr, PAGE_SIZE);
+dpp_map_err:
+	ocrdma_del_mmap(pd->uctx, db_page_addr, db_page_size);
+	return status;
+}
+
+struct ib_pd *ocrdma_alloc_pd(struct ib_device *ibdev,
+			      struct ib_ucontext *context,
+			      struct ib_udata *udata)
+{
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibdev);
+	struct ocrdma_pd *pd;
+	struct ocrdma_ucontext *uctx = NULL;
+	int status;
+	u8 is_uctx_pd = false;
+
+	if (udata && context) {
+		uctx = get_ocrdma_ucontext(context);
+		pd = ocrdma_get_ucontext_pd(uctx);
+		if (pd) {
+			is_uctx_pd = true;
+			goto pd_mapping;
+		}
+	}
+
+	pd = _ocrdma_alloc_pd(dev, uctx, udata);
+	if (IS_ERR(pd)) {
+		status = PTR_ERR(pd);
+		goto exit;
+	}
+
+pd_mapping:
+	if (udata && context) {
+		status = ocrdma_copy_pd_uresp(dev, pd, context, udata);
+		if (status)
+			goto err;
+	}
+	return &pd->ibpd;
+
+err:
+	if (is_uctx_pd) {
+		ocrdma_release_ucontext_pd(uctx);
+	} else {
+		status = ocrdma_mbx_dealloc_pd(dev, pd);
+		kfree(pd);
+	}
+exit:
+	return ERR_PTR(status);
+}
+
+int ocrdma_dealloc_pd(struct ib_pd *ibpd)
+{
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+	struct ocrdma_ucontext *uctx = NULL;
+	int status = 0;
+	u64 usr_db;
+
+	uctx = pd->uctx;
+	if (uctx) {
+		u64 dpp_db = dev->nic_info.dpp_unmapped_addr +
+			(pd->id * PAGE_SIZE);
+		if (pd->dpp_enabled)
+			ocrdma_del_mmap(pd->uctx, dpp_db, PAGE_SIZE);
+		usr_db = ocrdma_get_db_addr(dev, pd->id);
+		ocrdma_del_mmap(pd->uctx, usr_db, dev->nic_info.db_page_size);
+
+		if (is_ucontext_pd(uctx, pd)) {
+			ocrdma_release_ucontext_pd(uctx);
+			return status;
+		}
+	}
+	status = _ocrdma_dealloc_pd(dev, pd);
+	return status;
+}
+
+static int ocrdma_alloc_lkey(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
+			    u32 pdid, int acc, u32 num_pbls, u32 addr_check)
+{
+	int status;
+
+	mr->hwmr.fr_mr = 0;
+	mr->hwmr.local_rd = 1;
+	mr->hwmr.remote_rd = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+	mr->hwmr.remote_wr = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+	mr->hwmr.local_wr = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+	mr->hwmr.mw_bind = (acc & IB_ACCESS_MW_BIND) ? 1 : 0;
+	mr->hwmr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+	mr->hwmr.num_pbls = num_pbls;
+
+	status = ocrdma_mbx_alloc_lkey(dev, &mr->hwmr, pdid, addr_check);
+	if (status)
+		return status;
+
+	mr->ibmr.lkey = mr->hwmr.lkey;
+	if (mr->hwmr.remote_wr || mr->hwmr.remote_rd)
+		mr->ibmr.rkey = mr->hwmr.lkey;
+	return 0;
+}
+
+struct ib_mr *ocrdma_get_dma_mr(struct ib_pd *ibpd, int acc)
+{
+	int status;
+	struct ocrdma_mr *mr;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+
+	if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE)) {
+		pr_err("%s err, invalid access rights\n", __func__);
+		return ERR_PTR(-EINVAL);
+	}
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	status = ocrdma_alloc_lkey(dev, mr, pd->id, acc, 0,
+				   OCRDMA_ADDR_CHECK_DISABLE);
+	if (status) {
+		kfree(mr);
+		return ERR_PTR(status);
+	}
+
+	return &mr->ibmr;
+}
+
+static void ocrdma_free_mr_pbl_tbl(struct ocrdma_dev *dev,
+				   struct ocrdma_hw_mr *mr)
+{
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	int i = 0;
+
+	if (mr->pbl_table) {
+		for (i = 0; i < mr->num_pbls; i++) {
+			if (!mr->pbl_table[i].va)
+				continue;
+			dma_free_coherent(&pdev->dev, mr->pbl_size,
+					  mr->pbl_table[i].va,
+					  mr->pbl_table[i].pa);
+		}
+		kfree(mr->pbl_table);
+		mr->pbl_table = NULL;
+	}
+}
+
+static int ocrdma_get_pbl_info(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
+			      u32 num_pbes)
+{
+	u32 num_pbls = 0;
+	u32 idx = 0;
+	int status = 0;
+	u32 pbl_size;
+
+	do {
+		pbl_size = OCRDMA_MIN_HPAGE_SIZE * (1 << idx);
+		if (pbl_size > MAX_OCRDMA_PBL_SIZE) {
+			status = -EFAULT;
+			break;
+		}
+		num_pbls = roundup(num_pbes, (pbl_size / sizeof(u64)));
+		num_pbls = num_pbls / (pbl_size / sizeof(u64));
+		idx++;
+	} while (num_pbls >= dev->attr.max_num_mr_pbl);
+
+	mr->hwmr.num_pbes = num_pbes;
+	mr->hwmr.num_pbls = num_pbls;
+	mr->hwmr.pbl_size = pbl_size;
+	return status;
+}
+
+static int ocrdma_build_pbl_tbl(struct ocrdma_dev *dev, struct ocrdma_hw_mr *mr)
+{
+	int status = 0;
+	int i;
+	u32 dma_len = mr->pbl_size;
+	struct pci_dev *pdev = dev->nic_info.pdev;
+	void *va;
+	dma_addr_t pa;
+
+	mr->pbl_table = kzalloc(sizeof(struct ocrdma_pbl) *
+				mr->num_pbls, GFP_KERNEL);
+
+	if (!mr->pbl_table)
+		return -ENOMEM;
+
+	for (i = 0; i < mr->num_pbls; i++) {
+		va = dma_alloc_coherent(&pdev->dev, dma_len, &pa, GFP_KERNEL);
+		if (!va) {
+			ocrdma_free_mr_pbl_tbl(dev, mr);
+			status = -ENOMEM;
+			break;
+		}
+		memset(va, 0, dma_len);
+		mr->pbl_table[i].va = va;
+		mr->pbl_table[i].pa = pa;
+	}
+	return status;
+}
+
+static void build_user_pbes(struct ocrdma_dev *dev, struct ocrdma_mr *mr,
+			    u32 num_pbes)
+{
+	struct ocrdma_pbe *pbe;
+	struct scatterlist *sg;
+	struct ocrdma_pbl *pbl_tbl = mr->hwmr.pbl_table;
+	struct ib_umem *umem = mr->umem;
+	int shift, pg_cnt, pages, pbe_cnt, entry, total_num_pbes = 0;
+
+	if (!mr->hwmr.num_pbes)
+		return;
+
+	pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+	pbe_cnt = 0;
+
+	shift = ilog2(umem->page_size);
+
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+		pages = sg_dma_len(sg) >> shift;
+		for (pg_cnt = 0; pg_cnt < pages; pg_cnt++) {
+			/* store the page address in pbe */
+			pbe->pa_lo =
+			    cpu_to_le32(sg_dma_address
+					(sg) +
+					(umem->page_size * pg_cnt));
+			pbe->pa_hi =
+			    cpu_to_le32(upper_32_bits
+					((sg_dma_address
+					  (sg) +
+					  umem->page_size * pg_cnt)));
+			pbe_cnt += 1;
+			total_num_pbes += 1;
+			pbe++;
+
+			/* if done building pbes, issue the mbx cmd. */
+			if (total_num_pbes == num_pbes)
+				return;
+
+			/* if the given pbl is full storing the pbes,
+			 * move to next pbl.
+			 */
+			if (pbe_cnt ==
+				(mr->hwmr.pbl_size / sizeof(u64))) {
+				pbl_tbl++;
+				pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+				pbe_cnt = 0;
+			}
+
+		}
+	}
+}
+
+struct ib_mr *ocrdma_reg_user_mr(struct ib_pd *ibpd, u64 start, u64 len,
+				 u64 usr_addr, int acc, struct ib_udata *udata)
+{
+	int status = -ENOMEM;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+	struct ocrdma_mr *mr;
+	struct ocrdma_pd *pd;
+	u32 num_pbes;
+
+	pd = get_ocrdma_pd(ibpd);
+
+	if (acc & IB_ACCESS_REMOTE_WRITE && !(acc & IB_ACCESS_LOCAL_WRITE))
+		return ERR_PTR(-EINVAL);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(status);
+	mr->umem = ib_umem_get(ibpd->uobject->context, start, len, acc, 0);
+	if (IS_ERR(mr->umem)) {
+		status = -EFAULT;
+		goto umem_err;
+	}
+	num_pbes = ib_umem_page_count(mr->umem);
+	status = ocrdma_get_pbl_info(dev, mr, num_pbes);
+	if (status)
+		goto umem_err;
+
+	mr->hwmr.pbe_size = mr->umem->page_size;
+	mr->hwmr.fbo = mr->umem->offset;
+	mr->hwmr.va = usr_addr;
+	mr->hwmr.len = len;
+	mr->hwmr.remote_wr = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+	mr->hwmr.remote_rd = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+	mr->hwmr.local_wr = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+	mr->hwmr.local_rd = 1;
+	mr->hwmr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+	status = ocrdma_build_pbl_tbl(dev, &mr->hwmr);
+	if (status)
+		goto umem_err;
+	build_user_pbes(dev, mr, num_pbes);
+	status = ocrdma_reg_mr(dev, &mr->hwmr, pd->id, acc);
+	if (status)
+		goto mbx_err;
+	mr->ibmr.lkey = mr->hwmr.lkey;
+	if (mr->hwmr.remote_wr || mr->hwmr.remote_rd)
+		mr->ibmr.rkey = mr->hwmr.lkey;
+
+	return &mr->ibmr;
+
+mbx_err:
+	ocrdma_free_mr_pbl_tbl(dev, &mr->hwmr);
+umem_err:
+	kfree(mr);
+	return ERR_PTR(status);
+}
+
+int ocrdma_dereg_mr(struct ib_mr *ib_mr)
+{
+	struct ocrdma_mr *mr = get_ocrdma_mr(ib_mr);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ib_mr->device);
+	int status;
+
+	status = ocrdma_mbx_dealloc_lkey(dev, mr->hwmr.fr_mr, mr->hwmr.lkey);
+
+	ocrdma_free_mr_pbl_tbl(dev, &mr->hwmr);
+
+	/* it could be user registered memory. */
+	if (mr->umem)
+		ib_umem_release(mr->umem);
+	kfree(mr);
+	return status;
+}
+
+static int ocrdma_copy_cq_uresp(struct ocrdma_dev *dev, struct ocrdma_cq *cq,
+				struct ib_udata *udata,
+				struct ib_ucontext *ib_ctx)
+{
+	int status;
+	struct ocrdma_ucontext *uctx = get_ocrdma_ucontext(ib_ctx);
+	struct ocrdma_create_cq_uresp uresp;
+
+	memset(&uresp, 0, sizeof(uresp));
+	uresp.cq_id = cq->id;
+	uresp.page_size = PAGE_ALIGN(cq->len);
+	uresp.num_pages = 1;
+	uresp.max_hw_cqe = cq->max_hw_cqe;
+	uresp.page_addr[0] = cq->pa;
+	uresp.db_page_addr =  ocrdma_get_db_addr(dev, uctx->cntxt_pd->id);
+	uresp.db_page_size = dev->nic_info.db_page_size;
+	uresp.phase_change = cq->phase_change ? 1 : 0;
+	status = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+	if (status) {
+		pr_err("%s(%d) copy error cqid=0x%x.\n",
+		       __func__, dev->id, cq->id);
+		goto err;
+	}
+	status = ocrdma_add_mmap(uctx, uresp.db_page_addr, uresp.db_page_size);
+	if (status)
+		goto err;
+	status = ocrdma_add_mmap(uctx, uresp.page_addr[0], uresp.page_size);
+	if (status) {
+		ocrdma_del_mmap(uctx, uresp.db_page_addr, uresp.db_page_size);
+		goto err;
+	}
+	cq->ucontext = uctx;
+err:
+	return status;
+}
+
+struct ib_cq *ocrdma_create_cq(struct ib_device *ibdev, int entries, int vector,
+			       struct ib_ucontext *ib_ctx,
+			       struct ib_udata *udata)
+{
+	struct ocrdma_cq *cq;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibdev);
+	struct ocrdma_ucontext *uctx = NULL;
+	u16 pd_id = 0;
+	int status;
+	struct ocrdma_create_cq_ureq ureq;
+
+	if (udata) {
+		if (ib_copy_from_udata(&ureq, udata, sizeof(ureq)))
+			return ERR_PTR(-EFAULT);
+	} else
+		ureq.dpp_cq = 0;
+	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-ENOMEM);
+
+	spin_lock_init(&cq->cq_lock);
+	spin_lock_init(&cq->comp_handler_lock);
+	INIT_LIST_HEAD(&cq->sq_head);
+	INIT_LIST_HEAD(&cq->rq_head);
+	cq->first_arm = true;
+
+	if (ib_ctx) {
+		uctx = get_ocrdma_ucontext(ib_ctx);
+		pd_id = uctx->cntxt_pd->id;
+	}
+
+	status = ocrdma_mbx_create_cq(dev, cq, entries, ureq.dpp_cq, pd_id);
+	if (status) {
+		kfree(cq);
+		return ERR_PTR(status);
+	}
+	if (ib_ctx) {
+		status = ocrdma_copy_cq_uresp(dev, cq, udata, ib_ctx);
+		if (status)
+			goto ctx_err;
+	}
+	cq->phase = OCRDMA_CQE_VALID;
+	dev->cq_tbl[cq->id] = cq;
+	return &cq->ibcq;
+
+ctx_err:
+	ocrdma_mbx_destroy_cq(dev, cq);
+	kfree(cq);
+	return ERR_PTR(status);
+}
+
+int ocrdma_resize_cq(struct ib_cq *ibcq, int new_cnt,
+		     struct ib_udata *udata)
+{
+	int status = 0;
+	struct ocrdma_cq *cq = get_ocrdma_cq(ibcq);
+
+	if (new_cnt < 1 || new_cnt > cq->max_hw_cqe) {
+		status = -EINVAL;
+		return status;
+	}
+	ibcq->cqe = new_cnt;
+	return status;
+}
+
+static void ocrdma_flush_cq(struct ocrdma_cq *cq)
+{
+	int cqe_cnt;
+	int valid_count = 0;
+	unsigned long flags;
+
+	struct ocrdma_dev *dev = get_ocrdma_dev(cq->ibcq.device);
+	struct ocrdma_cqe *cqe = NULL;
+
+	cqe = cq->va;
+	cqe_cnt = cq->cqe_cnt;
+
+	/* Last irq might have scheduled a polling thread
+	 * sync-up with it before hard flushing.
+	 */
+	spin_lock_irqsave(&cq->cq_lock, flags);
+	while (cqe_cnt) {
+		if (is_cqe_valid(cq, cqe))
+			valid_count++;
+		cqe++;
+		cqe_cnt--;
+	}
+	ocrdma_ring_cq_db(dev, cq->id, false, false, valid_count);
+	spin_unlock_irqrestore(&cq->cq_lock, flags);
+}
+
+int ocrdma_destroy_cq(struct ib_cq *ibcq)
+{
+	int status;
+	struct ocrdma_cq *cq = get_ocrdma_cq(ibcq);
+	struct ocrdma_eq *eq = NULL;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibcq->device);
+	int pdid = 0;
+	u32 irq, indx;
+
+	dev->cq_tbl[cq->id] = NULL;
+	indx = ocrdma_get_eq_table_index(dev, cq->eqn);
+	if (indx == -EINVAL)
+		BUG();
+
+	eq = &dev->eq_tbl[indx];
+	irq = ocrdma_get_irq(dev, eq);
+	synchronize_irq(irq);
+	ocrdma_flush_cq(cq);
+
+	status = ocrdma_mbx_destroy_cq(dev, cq);
+	if (cq->ucontext) {
+		pdid = cq->ucontext->cntxt_pd->id;
+		ocrdma_del_mmap(cq->ucontext, (u64) cq->pa,
+				PAGE_ALIGN(cq->len));
+		ocrdma_del_mmap(cq->ucontext,
+				ocrdma_get_db_addr(dev, pdid),
+				dev->nic_info.db_page_size);
+	}
+
+	kfree(cq);
+	return status;
+}
+
+static int ocrdma_add_qpn_map(struct ocrdma_dev *dev, struct ocrdma_qp *qp)
+{
+	int status = -EINVAL;
+
+	if (qp->id < OCRDMA_MAX_QP && dev->qp_tbl[qp->id] == NULL) {
+		dev->qp_tbl[qp->id] = qp;
+		status = 0;
+	}
+	return status;
+}
+
+static void ocrdma_del_qpn_map(struct ocrdma_dev *dev, struct ocrdma_qp *qp)
+{
+	dev->qp_tbl[qp->id] = NULL;
+}
+
+static int ocrdma_check_qp_params(struct ib_pd *ibpd, struct ocrdma_dev *dev,
+				  struct ib_qp_init_attr *attrs)
+{
+	if ((attrs->qp_type != IB_QPT_GSI) &&
+	    (attrs->qp_type != IB_QPT_RC) &&
+	    (attrs->qp_type != IB_QPT_UC) &&
+	    (attrs->qp_type != IB_QPT_UD)) {
+		pr_err("%s(%d) unsupported qp type=0x%x requested\n",
+		       __func__, dev->id, attrs->qp_type);
+		return -EINVAL;
+	}
+	/* Skip the check for QP1 to support CM size of 128 */
+	if ((attrs->qp_type != IB_QPT_GSI) &&
+	    (attrs->cap.max_send_wr > dev->attr.max_wqe)) {
+		pr_err("%s(%d) unsupported send_wr=0x%x requested\n",
+		       __func__, dev->id, attrs->cap.max_send_wr);
+		pr_err("%s(%d) supported send_wr=0x%x\n",
+		       __func__, dev->id, dev->attr.max_wqe);
+		return -EINVAL;
+	}
+	if (!attrs->srq && (attrs->cap.max_recv_wr > dev->attr.max_rqe)) {
+		pr_err("%s(%d) unsupported recv_wr=0x%x requested\n",
+		       __func__, dev->id, attrs->cap.max_recv_wr);
+		pr_err("%s(%d) supported recv_wr=0x%x\n",
+		       __func__, dev->id, dev->attr.max_rqe);
+		return -EINVAL;
+	}
+	if (attrs->cap.max_inline_data > dev->attr.max_inline_data) {
+		pr_err("%s(%d) unsupported inline data size=0x%x requested\n",
+		       __func__, dev->id, attrs->cap.max_inline_data);
+		pr_err("%s(%d) supported inline data size=0x%x\n",
+		       __func__, dev->id, dev->attr.max_inline_data);
+		return -EINVAL;
+	}
+	if (attrs->cap.max_send_sge > dev->attr.max_send_sge) {
+		pr_err("%s(%d) unsupported send_sge=0x%x requested\n",
+		       __func__, dev->id, attrs->cap.max_send_sge);
+		pr_err("%s(%d) supported send_sge=0x%x\n",
+		       __func__, dev->id, dev->attr.max_send_sge);
+		return -EINVAL;
+	}
+	if (attrs->cap.max_recv_sge > dev->attr.max_recv_sge) {
+		pr_err("%s(%d) unsupported recv_sge=0x%x requested\n",
+		       __func__, dev->id, attrs->cap.max_recv_sge);
+		pr_err("%s(%d) supported recv_sge=0x%x\n",
+		       __func__, dev->id, dev->attr.max_recv_sge);
+		return -EINVAL;
+	}
+	/* unprivileged user space cannot create special QP */
+	if (ibpd->uobject && attrs->qp_type == IB_QPT_GSI) {
+		pr_err
+		    ("%s(%d) Userspace can't create special QPs of type=0x%x\n",
+		     __func__, dev->id, attrs->qp_type);
+		return -EINVAL;
+	}
+	/* allow creating only one GSI type of QP */
+	if (attrs->qp_type == IB_QPT_GSI && dev->gsi_qp_created) {
+		pr_err("%s(%d) GSI special QPs already created.\n",
+		       __func__, dev->id);
+		return -EINVAL;
+	}
+	/* verify consumer QPs are not trying to use GSI QP's CQ */
+	if ((attrs->qp_type != IB_QPT_GSI) && (dev->gsi_qp_created)) {
+		if ((dev->gsi_sqcq == get_ocrdma_cq(attrs->send_cq)) ||
+			(dev->gsi_rqcq == get_ocrdma_cq(attrs->recv_cq))) {
+			pr_err("%s(%d) Consumer QP cannot use GSI CQs.\n",
+				__func__, dev->id);
+			return -EINVAL;
+		}
+	}
+	return 0;
+}
+
+static int ocrdma_copy_qp_uresp(struct ocrdma_qp *qp,
+				struct ib_udata *udata, int dpp_offset,
+				int dpp_credit_lmt, int srq)
+{
+	int status = 0;
+	u64 usr_db;
+	struct ocrdma_create_qp_uresp uresp;
+	struct ocrdma_dev *dev = qp->dev;
+	struct ocrdma_pd *pd = qp->pd;
+
+	memset(&uresp, 0, sizeof(uresp));
+	usr_db = dev->nic_info.unmapped_db +
+			(pd->id * dev->nic_info.db_page_size);
+	uresp.qp_id = qp->id;
+	uresp.sq_dbid = qp->sq.dbid;
+	uresp.num_sq_pages = 1;
+	uresp.sq_page_size = PAGE_ALIGN(qp->sq.len);
+	uresp.sq_page_addr[0] = qp->sq.pa;
+	uresp.num_wqe_allocated = qp->sq.max_cnt;
+	if (!srq) {
+		uresp.rq_dbid = qp->rq.dbid;
+		uresp.num_rq_pages = 1;
+		uresp.rq_page_size = PAGE_ALIGN(qp->rq.len);
+		uresp.rq_page_addr[0] = qp->rq.pa;
+		uresp.num_rqe_allocated = qp->rq.max_cnt;
+	}
+	uresp.db_page_addr = usr_db;
+	uresp.db_page_size = dev->nic_info.db_page_size;
+	uresp.db_sq_offset = OCRDMA_DB_GEN2_SQ_OFFSET;
+	uresp.db_rq_offset = OCRDMA_DB_GEN2_RQ_OFFSET;
+	uresp.db_shift = OCRDMA_DB_RQ_SHIFT;
+
+	if (qp->dpp_enabled) {
+		uresp.dpp_credit = dpp_credit_lmt;
+		uresp.dpp_offset = dpp_offset;
+	}
+	status = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+	if (status) {
+		pr_err("%s(%d) user copy error.\n", __func__, dev->id);
+		goto err;
+	}
+	status = ocrdma_add_mmap(pd->uctx, uresp.sq_page_addr[0],
+				 uresp.sq_page_size);
+	if (status)
+		goto err;
+
+	if (!srq) {
+		status = ocrdma_add_mmap(pd->uctx, uresp.rq_page_addr[0],
+					 uresp.rq_page_size);
+		if (status)
+			goto rq_map_err;
+	}
+	return status;
+rq_map_err:
+	ocrdma_del_mmap(pd->uctx, uresp.sq_page_addr[0], uresp.sq_page_size);
+err:
+	return status;
+}
+
+static void ocrdma_set_qp_db(struct ocrdma_dev *dev, struct ocrdma_qp *qp,
+			     struct ocrdma_pd *pd)
+{
+	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
+		qp->sq_db = dev->nic_info.db +
+			(pd->id * dev->nic_info.db_page_size) +
+			OCRDMA_DB_GEN2_SQ_OFFSET;
+		qp->rq_db = dev->nic_info.db +
+			(pd->id * dev->nic_info.db_page_size) +
+			OCRDMA_DB_GEN2_RQ_OFFSET;
+	} else {
+		qp->sq_db = dev->nic_info.db +
+			(pd->id * dev->nic_info.db_page_size) +
+			OCRDMA_DB_SQ_OFFSET;
+		qp->rq_db = dev->nic_info.db +
+			(pd->id * dev->nic_info.db_page_size) +
+			OCRDMA_DB_RQ_OFFSET;
+	}
+}
+
+static int ocrdma_alloc_wr_id_tbl(struct ocrdma_qp *qp)
+{
+	qp->wqe_wr_id_tbl =
+	    kzalloc(sizeof(*(qp->wqe_wr_id_tbl)) * qp->sq.max_cnt,
+		    GFP_KERNEL);
+	if (qp->wqe_wr_id_tbl == NULL)
+		return -ENOMEM;
+	qp->rqe_wr_id_tbl =
+	    kzalloc(sizeof(u64) * qp->rq.max_cnt, GFP_KERNEL);
+	if (qp->rqe_wr_id_tbl == NULL)
+		return -ENOMEM;
+
+	return 0;
+}
+
+static void ocrdma_set_qp_init_params(struct ocrdma_qp *qp,
+				      struct ocrdma_pd *pd,
+				      struct ib_qp_init_attr *attrs)
+{
+	qp->pd = pd;
+	spin_lock_init(&qp->q_lock);
+	INIT_LIST_HEAD(&qp->sq_entry);
+	INIT_LIST_HEAD(&qp->rq_entry);
+
+	qp->qp_type = attrs->qp_type;
+	qp->cap_flags = OCRDMA_QP_INB_RD | OCRDMA_QP_INB_WR;
+	qp->max_inline_data = attrs->cap.max_inline_data;
+	qp->sq.max_sges = attrs->cap.max_send_sge;
+	qp->rq.max_sges = attrs->cap.max_recv_sge;
+	qp->state = OCRDMA_QPS_RST;
+	qp->signaled = (attrs->sq_sig_type == IB_SIGNAL_ALL_WR) ? true : false;
+}
+
+static void ocrdma_store_gsi_qp_cq(struct ocrdma_dev *dev,
+				   struct ib_qp_init_attr *attrs)
+{
+	if (attrs->qp_type == IB_QPT_GSI) {
+		dev->gsi_qp_created = 1;
+		dev->gsi_sqcq = get_ocrdma_cq(attrs->send_cq);
+		dev->gsi_rqcq = get_ocrdma_cq(attrs->recv_cq);
+	}
+}
+
+struct ib_qp *ocrdma_create_qp(struct ib_pd *ibpd,
+			       struct ib_qp_init_attr *attrs,
+			       struct ib_udata *udata)
+{
+	int status;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_qp *qp;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+	struct ocrdma_create_qp_ureq ureq;
+	u16 dpp_credit_lmt, dpp_offset;
+
+	status = ocrdma_check_qp_params(ibpd, dev, attrs);
+	if (status)
+		goto gen_err;
+
+	memset(&ureq, 0, sizeof(ureq));
+	if (udata) {
+		if (ib_copy_from_udata(&ureq, udata, sizeof(ureq)))
+			return ERR_PTR(-EFAULT);
+	}
+	qp = kzalloc(sizeof(*qp), GFP_KERNEL);
+	if (!qp) {
+		status = -ENOMEM;
+		goto gen_err;
+	}
+	qp->dev = dev;
+	ocrdma_set_qp_init_params(qp, pd, attrs);
+	if (udata == NULL)
+		qp->cap_flags |= (OCRDMA_QP_MW_BIND | OCRDMA_QP_LKEY0 |
+					OCRDMA_QP_FAST_REG);
+
+	mutex_lock(&dev->dev_lock);
+	status = ocrdma_mbx_create_qp(qp, attrs, ureq.enable_dpp_cq,
+					ureq.dpp_cq_id,
+					&dpp_offset, &dpp_credit_lmt);
+	if (status)
+		goto mbx_err;
+
+	/* user space QP's wr_id table are managed in library */
+	if (udata == NULL) {
+		status = ocrdma_alloc_wr_id_tbl(qp);
+		if (status)
+			goto map_err;
+	}
+
+	status = ocrdma_add_qpn_map(dev, qp);
+	if (status)
+		goto map_err;
+	ocrdma_set_qp_db(dev, qp, pd);
+	if (udata) {
+		status = ocrdma_copy_qp_uresp(qp, udata, dpp_offset,
+					      dpp_credit_lmt,
+					      (attrs->srq != NULL));
+		if (status)
+			goto cpy_err;
+	}
+	ocrdma_store_gsi_qp_cq(dev, attrs);
+	qp->ibqp.qp_num = qp->id;
+	mutex_unlock(&dev->dev_lock);
+	return &qp->ibqp;
+
+cpy_err:
+	ocrdma_del_qpn_map(dev, qp);
+map_err:
+	ocrdma_mbx_destroy_qp(dev, qp);
+mbx_err:
+	mutex_unlock(&dev->dev_lock);
+	kfree(qp->wqe_wr_id_tbl);
+	kfree(qp->rqe_wr_id_tbl);
+	kfree(qp);
+	pr_err("%s(%d) error=%d\n", __func__, dev->id, status);
+gen_err:
+	return ERR_PTR(status);
+}
+
+int _ocrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		      int attr_mask)
+{
+	int status = 0;
+	struct ocrdma_qp *qp;
+	struct ocrdma_dev *dev;
+	enum ib_qp_state old_qps;
+
+	qp = get_ocrdma_qp(ibqp);
+	dev = qp->dev;
+	if (attr_mask & IB_QP_STATE)
+		status = ocrdma_qp_state_change(qp, attr->qp_state, &old_qps);
+	/* if new and previous states are same hw doesn't need to
+	 * know about it.
+	 */
+	if (status < 0)
+		return status;
+	status = ocrdma_mbx_modify_qp(dev, qp, attr, attr_mask);
+
+	return status;
+}
+
+int ocrdma_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_udata *udata)
+{
+	unsigned long flags;
+	int status = -EINVAL;
+	struct ocrdma_qp *qp;
+	struct ocrdma_dev *dev;
+	enum ib_qp_state old_qps, new_qps;
+
+	qp = get_ocrdma_qp(ibqp);
+	dev = qp->dev;
+
+	/* syncronize with multiple context trying to change, retrive qps */
+	mutex_lock(&dev->dev_lock);
+	/* syncronize with wqe, rqe posting and cqe processing contexts */
+	spin_lock_irqsave(&qp->q_lock, flags);
+	old_qps = get_ibqp_state(qp->state);
+	if (attr_mask & IB_QP_STATE)
+		new_qps = attr->qp_state;
+	else
+		new_qps = old_qps;
+	spin_unlock_irqrestore(&qp->q_lock, flags);
+
+	if (!ib_modify_qp_is_ok(old_qps, new_qps, ibqp->qp_type, attr_mask,
+				IB_LINK_LAYER_ETHERNET)) {
+		pr_err("%s(%d) invalid attribute mask=0x%x specified for\n"
+		       "qpn=0x%x of type=0x%x old_qps=0x%x, new_qps=0x%x\n",
+		       __func__, dev->id, attr_mask, qp->id, ibqp->qp_type,
+		       old_qps, new_qps);
+		goto param_err;
+	}
+
+	status = _ocrdma_modify_qp(ibqp, attr, attr_mask);
+	if (status > 0)
+		status = 0;
+param_err:
+	mutex_unlock(&dev->dev_lock);
+	return status;
+}
+
+static enum ib_mtu ocrdma_mtu_int_to_enum(u16 mtu)
+{
+	switch (mtu) {
+	case 256:
+		return IB_MTU_256;
+	case 512:
+		return IB_MTU_512;
+	case 1024:
+		return IB_MTU_1024;
+	case 2048:
+		return IB_MTU_2048;
+	case 4096:
+		return IB_MTU_4096;
+	default:
+		return IB_MTU_1024;
+	}
+}
+
+static int ocrdma_to_ib_qp_acc_flags(int qp_cap_flags)
+{
+	int ib_qp_acc_flags = 0;
+
+	if (qp_cap_flags & OCRDMA_QP_INB_WR)
+		ib_qp_acc_flags |= IB_ACCESS_REMOTE_WRITE;
+	if (qp_cap_flags & OCRDMA_QP_INB_RD)
+		ib_qp_acc_flags |= IB_ACCESS_LOCAL_WRITE;
+	return ib_qp_acc_flags;
+}
+
+int ocrdma_query_qp(struct ib_qp *ibqp,
+		    struct ib_qp_attr *qp_attr,
+		    int attr_mask, struct ib_qp_init_attr *qp_init_attr)
+{
+	int status;
+	u32 qp_state;
+	struct ocrdma_qp_params params;
+	struct ocrdma_qp *qp = get_ocrdma_qp(ibqp);
+	struct ocrdma_dev *dev = qp->dev;
+
+	memset(&params, 0, sizeof(params));
+	mutex_lock(&dev->dev_lock);
+	status = ocrdma_mbx_query_qp(dev, qp, &params);
+	mutex_unlock(&dev->dev_lock);
+	if (status)
+		goto mbx_err;
+	qp_attr->qp_state = get_ibqp_state(IB_QPS_INIT);
+	qp_attr->cur_qp_state = get_ibqp_state(IB_QPS_INIT);
+	qp_attr->path_mtu =
+		ocrdma_mtu_int_to_enum(params.path_mtu_pkey_indx &
+				OCRDMA_QP_PARAMS_PATH_MTU_MASK) >>
+				OCRDMA_QP_PARAMS_PATH_MTU_SHIFT;
+	qp_attr->path_mig_state = IB_MIG_MIGRATED;
+	qp_attr->rq_psn = params.hop_lmt_rq_psn & OCRDMA_QP_PARAMS_RQ_PSN_MASK;
+	qp_attr->sq_psn = params.tclass_sq_psn & OCRDMA_QP_PARAMS_SQ_PSN_MASK;
+	qp_attr->dest_qp_num =
+	    params.ack_to_rnr_rtc_dest_qpn & OCRDMA_QP_PARAMS_DEST_QPN_MASK;
+
+	qp_attr->qp_access_flags = ocrdma_to_ib_qp_acc_flags(qp->cap_flags);
+	qp_attr->cap.max_send_wr = qp->sq.max_cnt - 1;
+	qp_attr->cap.max_recv_wr = qp->rq.max_cnt - 1;
+	qp_attr->cap.max_send_sge = qp->sq.max_sges;
+	qp_attr->cap.max_recv_sge = qp->rq.max_sges;
+	qp_attr->cap.max_inline_data = qp->max_inline_data;
+	qp_init_attr->cap = qp_attr->cap;
+	memcpy(&qp_attr->ah_attr.grh.dgid, &params.dgid[0],
+	       sizeof(params.dgid));
+	qp_attr->ah_attr.grh.flow_label = params.rnt_rc_sl_fl &
+	    OCRDMA_QP_PARAMS_FLOW_LABEL_MASK;
+	qp_attr->ah_attr.grh.sgid_index = qp->sgid_idx;
+	qp_attr->ah_attr.grh.hop_limit = (params.hop_lmt_rq_psn &
+					  OCRDMA_QP_PARAMS_HOP_LMT_MASK) >>
+						OCRDMA_QP_PARAMS_HOP_LMT_SHIFT;
+	qp_attr->ah_attr.grh.traffic_class = (params.tclass_sq_psn &
+					      OCRDMA_QP_PARAMS_TCLASS_MASK) >>
+						OCRDMA_QP_PARAMS_TCLASS_SHIFT;
+
+	qp_attr->ah_attr.ah_flags = IB_AH_GRH;
+	qp_attr->ah_attr.port_num = 1;
+	qp_attr->ah_attr.sl = (params.rnt_rc_sl_fl &
+			       OCRDMA_QP_PARAMS_SL_MASK) >>
+				OCRDMA_QP_PARAMS_SL_SHIFT;
+	qp_attr->timeout = (params.ack_to_rnr_rtc_dest_qpn &
+			    OCRDMA_QP_PARAMS_ACK_TIMEOUT_MASK) >>
+				OCRDMA_QP_PARAMS_ACK_TIMEOUT_SHIFT;
+	qp_attr->rnr_retry = (params.ack_to_rnr_rtc_dest_qpn &
+			      OCRDMA_QP_PARAMS_RNR_RETRY_CNT_MASK) >>
+				OCRDMA_QP_PARAMS_RNR_RETRY_CNT_SHIFT;
+	qp_attr->retry_cnt =
+	    (params.rnt_rc_sl_fl & OCRDMA_QP_PARAMS_RETRY_CNT_MASK) >>
+		OCRDMA_QP_PARAMS_RETRY_CNT_SHIFT;
+	qp_attr->min_rnr_timer = 0;
+	qp_attr->pkey_index = 0;
+	qp_attr->port_num = 1;
+	qp_attr->ah_attr.src_path_bits = 0;
+	qp_attr->ah_attr.static_rate = 0;
+	qp_attr->alt_pkey_index = 0;
+	qp_attr->alt_port_num = 0;
+	qp_attr->alt_timeout = 0;
+	memset(&qp_attr->alt_ah_attr, 0, sizeof(qp_attr->alt_ah_attr));
+	qp_state = (params.max_sge_recv_flags & OCRDMA_QP_PARAMS_STATE_MASK) >>
+		    OCRDMA_QP_PARAMS_STATE_SHIFT;
+	qp_attr->sq_draining = (qp_state == OCRDMA_QPS_SQ_DRAINING) ? 1 : 0;
+	qp_attr->max_dest_rd_atomic =
+	    params.max_ord_ird >> OCRDMA_QP_PARAMS_MAX_ORD_SHIFT;
+	qp_attr->max_rd_atomic =
+	    params.max_ord_ird & OCRDMA_QP_PARAMS_MAX_IRD_MASK;
+	qp_attr->en_sqd_async_notify = (params.max_sge_recv_flags &
+				OCRDMA_QP_PARAMS_FLAGS_SQD_ASYNC) ? 1 : 0;
+mbx_err:
+	return status;
+}
+
+static void ocrdma_srq_toggle_bit(struct ocrdma_srq *srq, int idx)
+{
+	int i = idx / 32;
+	unsigned int mask = (1 << (idx % 32));
+
+	if (srq->idx_bit_fields[i] & mask)
+		srq->idx_bit_fields[i] &= ~mask;
+	else
+		srq->idx_bit_fields[i] |= mask;
+}
+
+static int ocrdma_hwq_free_cnt(struct ocrdma_qp_hwq_info *q)
+{
+	return ((q->max_wqe_idx - q->head) + q->tail) % q->max_cnt;
+}
+
+static int is_hw_sq_empty(struct ocrdma_qp *qp)
+{
+	return (qp->sq.tail == qp->sq.head);
+}
+
+static int is_hw_rq_empty(struct ocrdma_qp *qp)
+{
+	return (qp->rq.tail == qp->rq.head);
+}
+
+static void *ocrdma_hwq_head(struct ocrdma_qp_hwq_info *q)
+{
+	return q->va + (q->head * q->entry_size);
+}
+
+static void *ocrdma_hwq_head_from_idx(struct ocrdma_qp_hwq_info *q,
+				      u32 idx)
+{
+	return q->va + (idx * q->entry_size);
+}
+
+static void ocrdma_hwq_inc_head(struct ocrdma_qp_hwq_info *q)
+{
+	q->head = (q->head + 1) & q->max_wqe_idx;
+}
+
+static void ocrdma_hwq_inc_tail(struct ocrdma_qp_hwq_info *q)
+{
+	q->tail = (q->tail + 1) & q->max_wqe_idx;
+}
+
+/* discard the cqe for a given QP */
+static void ocrdma_discard_cqes(struct ocrdma_qp *qp, struct ocrdma_cq *cq)
+{
+	unsigned long cq_flags;
+	unsigned long flags;
+	int discard_cnt = 0;
+	u32 cur_getp, stop_getp;
+	struct ocrdma_cqe *cqe;
+	u32 qpn = 0, wqe_idx = 0;
+
+	spin_lock_irqsave(&cq->cq_lock, cq_flags);
+
+	/* traverse through the CQEs in the hw CQ,
+	 * find the matching CQE for a given qp,
+	 * mark the matching one discarded by clearing qpn.
+	 * ring the doorbell in the poll_cq() as
+	 * we don't complete out of order cqe.
+	 */
+
+	cur_getp = cq->getp;
+	/* find upto when do we reap the cq. */
+	stop_getp = cur_getp;
+	do {
+		if (is_hw_sq_empty(qp) && (!qp->srq && is_hw_rq_empty(qp)))
+			break;
+
+		cqe = cq->va + cur_getp;
+		/* if (a) done reaping whole hw cq, or
+		 *    (b) qp_xq becomes empty.
+		 * then exit
+		 */
+		qpn = cqe->cmn.qpn & OCRDMA_CQE_QPN_MASK;
+		/* if previously discarded cqe found, skip that too. */
+		/* check for matching qp */
+		if (qpn == 0 || qpn != qp->id)
+			goto skip_cqe;
+
+		if (is_cqe_for_sq(cqe)) {
+			ocrdma_hwq_inc_tail(&qp->sq);
+		} else {
+			if (qp->srq) {
+				wqe_idx = (le32_to_cpu(cqe->rq.buftag_qpn) >>
+					OCRDMA_CQE_BUFTAG_SHIFT) &
+					qp->srq->rq.max_wqe_idx;
+				if (wqe_idx < 1)
+					BUG();
+				spin_lock_irqsave(&qp->srq->q_lock, flags);
+				ocrdma_hwq_inc_tail(&qp->srq->rq);
+				ocrdma_srq_toggle_bit(qp->srq, wqe_idx - 1);
+				spin_unlock_irqrestore(&qp->srq->q_lock, flags);
+
+			} else {
+				ocrdma_hwq_inc_tail(&qp->rq);
+			}
+		}
+		/* mark cqe discarded so that it is not picked up later
+		 * in the poll_cq().
+		 */
+		discard_cnt += 1;
+		cqe->cmn.qpn = 0;
+skip_cqe:
+		cur_getp = (cur_getp + 1) % cq->max_hw_cqe;
+	} while (cur_getp != stop_getp);
+	spin_unlock_irqrestore(&cq->cq_lock, cq_flags);
+}
+
+void ocrdma_del_flush_qp(struct ocrdma_qp *qp)
+{
+	int found = false;
+	unsigned long flags;
+	struct ocrdma_dev *dev = qp->dev;
+	/* sync with any active CQ poll */
+
+	spin_lock_irqsave(&dev->flush_q_lock, flags);
+	found = ocrdma_is_qp_in_sq_flushlist(qp->sq_cq, qp);
+	if (found)
+		list_del(&qp->sq_entry);
+	if (!qp->srq) {
+		found = ocrdma_is_qp_in_rq_flushlist(qp->rq_cq, qp);
+		if (found)
+			list_del(&qp->rq_entry);
+	}
+	spin_unlock_irqrestore(&dev->flush_q_lock, flags);
+}
+
+int ocrdma_destroy_qp(struct ib_qp *ibqp)
+{
+	int status;
+	struct ocrdma_pd *pd;
+	struct ocrdma_qp *qp;
+	struct ocrdma_dev *dev;
+	struct ib_qp_attr attrs;
+	int attr_mask = IB_QP_STATE;
+	unsigned long flags;
+
+	qp = get_ocrdma_qp(ibqp);
+	dev = qp->dev;
+
+	attrs.qp_state = IB_QPS_ERR;
+	pd = qp->pd;
+
+	/* change the QP state to ERROR */
+	_ocrdma_modify_qp(ibqp, &attrs, attr_mask);
+
+	/* ensure that CQEs for newly created QP (whose id may be same with
+	 * one which just getting destroyed are same), dont get
+	 * discarded until the old CQEs are discarded.
+	 */
+	mutex_lock(&dev->dev_lock);
+	status = ocrdma_mbx_destroy_qp(dev, qp);
+
+	/*
+	 * acquire CQ lock while destroy is in progress, in order to
+	 * protect against proessing in-flight CQEs for this QP.
+	 */
+	spin_lock_irqsave(&qp->sq_cq->cq_lock, flags);
+	if (qp->rq_cq && (qp->rq_cq != qp->sq_cq))
+		spin_lock(&qp->rq_cq->cq_lock);
+
+	ocrdma_del_qpn_map(dev, qp);
+
+	if (qp->rq_cq && (qp->rq_cq != qp->sq_cq))
+		spin_unlock(&qp->rq_cq->cq_lock);
+	spin_unlock_irqrestore(&qp->sq_cq->cq_lock, flags);
+
+	if (!pd->uctx) {
+		ocrdma_discard_cqes(qp, qp->sq_cq);
+		ocrdma_discard_cqes(qp, qp->rq_cq);
+	}
+	mutex_unlock(&dev->dev_lock);
+
+	if (pd->uctx) {
+		ocrdma_del_mmap(pd->uctx, (u64) qp->sq.pa,
+				PAGE_ALIGN(qp->sq.len));
+		if (!qp->srq)
+			ocrdma_del_mmap(pd->uctx, (u64) qp->rq.pa,
+					PAGE_ALIGN(qp->rq.len));
+	}
+
+	ocrdma_del_flush_qp(qp);
+
+	kfree(qp->wqe_wr_id_tbl);
+	kfree(qp->rqe_wr_id_tbl);
+	kfree(qp);
+	return status;
+}
+
+static int ocrdma_copy_srq_uresp(struct ocrdma_dev *dev, struct ocrdma_srq *srq,
+				struct ib_udata *udata)
+{
+	int status;
+	struct ocrdma_create_srq_uresp uresp;
+
+	memset(&uresp, 0, sizeof(uresp));
+	uresp.rq_dbid = srq->rq.dbid;
+	uresp.num_rq_pages = 1;
+	uresp.rq_page_addr[0] = srq->rq.pa;
+	uresp.rq_page_size = srq->rq.len;
+	uresp.db_page_addr = dev->nic_info.unmapped_db +
+	    (srq->pd->id * dev->nic_info.db_page_size);
+	uresp.db_page_size = dev->nic_info.db_page_size;
+	uresp.num_rqe_allocated = srq->rq.max_cnt;
+	if (ocrdma_get_asic_type(dev) == OCRDMA_ASIC_GEN_SKH_R) {
+		uresp.db_rq_offset = OCRDMA_DB_GEN2_RQ_OFFSET;
+		uresp.db_shift = 24;
+	} else {
+		uresp.db_rq_offset = OCRDMA_DB_RQ_OFFSET;
+		uresp.db_shift = 16;
+	}
+
+	status = ib_copy_to_udata(udata, &uresp, sizeof(uresp));
+	if (status)
+		return status;
+	status = ocrdma_add_mmap(srq->pd->uctx, uresp.rq_page_addr[0],
+				 uresp.rq_page_size);
+	if (status)
+		return status;
+	return status;
+}
+
+struct ib_srq *ocrdma_create_srq(struct ib_pd *ibpd,
+				 struct ib_srq_init_attr *init_attr,
+				 struct ib_udata *udata)
+{
+	int status = -ENOMEM;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+	struct ocrdma_srq *srq;
+
+	if (init_attr->attr.max_sge > dev->attr.max_recv_sge)
+		return ERR_PTR(-EINVAL);
+	if (init_attr->attr.max_wr > dev->attr.max_rqe)
+		return ERR_PTR(-EINVAL);
+
+	srq = kzalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq)
+		return ERR_PTR(status);
+
+	spin_lock_init(&srq->q_lock);
+	srq->pd = pd;
+	srq->db = dev->nic_info.db + (pd->id * dev->nic_info.db_page_size);
+	status = ocrdma_mbx_create_srq(dev, srq, init_attr, pd);
+	if (status)
+		goto err;
+
+	if (udata == NULL) {
+		srq->rqe_wr_id_tbl = kzalloc(sizeof(u64) * srq->rq.max_cnt,
+			    GFP_KERNEL);
+		if (srq->rqe_wr_id_tbl == NULL)
+			goto arm_err;
+
+		srq->bit_fields_len = (srq->rq.max_cnt / 32) +
+		    (srq->rq.max_cnt % 32 ? 1 : 0);
+		srq->idx_bit_fields =
+		    kmalloc(srq->bit_fields_len * sizeof(u32), GFP_KERNEL);
+		if (srq->idx_bit_fields == NULL)
+			goto arm_err;
+		memset(srq->idx_bit_fields, 0xff,
+		       srq->bit_fields_len * sizeof(u32));
+	}
+
+	if (init_attr->attr.srq_limit) {
+		status = ocrdma_mbx_modify_srq(srq, &init_attr->attr);
+		if (status)
+			goto arm_err;
+	}
+
+	if (udata) {
+		status = ocrdma_copy_srq_uresp(dev, srq, udata);
+		if (status)
+			goto arm_err;
+	}
+
+	return &srq->ibsrq;
+
+arm_err:
+	ocrdma_mbx_destroy_srq(dev, srq);
+err:
+	kfree(srq->rqe_wr_id_tbl);
+	kfree(srq->idx_bit_fields);
+	kfree(srq);
+	return ERR_PTR(status);
+}
+
+int ocrdma_modify_srq(struct ib_srq *ibsrq,
+		      struct ib_srq_attr *srq_attr,
+		      enum ib_srq_attr_mask srq_attr_mask,
+		      struct ib_udata *udata)
+{
+	int status = 0;
+	struct ocrdma_srq *srq;
+
+	srq = get_ocrdma_srq(ibsrq);
+	if (srq_attr_mask & IB_SRQ_MAX_WR)
+		status = -EINVAL;
+	else
+		status = ocrdma_mbx_modify_srq(srq, srq_attr);
+	return status;
+}
+
+int ocrdma_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *srq_attr)
+{
+	int status;
+	struct ocrdma_srq *srq;
+
+	srq = get_ocrdma_srq(ibsrq);
+	status = ocrdma_mbx_query_srq(srq, srq_attr);
+	return status;
+}
+
+int ocrdma_destroy_srq(struct ib_srq *ibsrq)
+{
+	int status;
+	struct ocrdma_srq *srq;
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibsrq->device);
+
+	srq = get_ocrdma_srq(ibsrq);
+
+	status = ocrdma_mbx_destroy_srq(dev, srq);
+
+	if (srq->pd->uctx)
+		ocrdma_del_mmap(srq->pd->uctx, (u64) srq->rq.pa,
+				PAGE_ALIGN(srq->rq.len));
+
+	kfree(srq->idx_bit_fields);
+	kfree(srq->rqe_wr_id_tbl);
+	kfree(srq);
+	return status;
+}
+
+/* unprivileged verbs and their support functions. */
+static void ocrdma_build_ud_hdr(struct ocrdma_qp *qp,
+				struct ocrdma_hdr_wqe *hdr,
+				struct ib_send_wr *wr)
+{
+	struct ocrdma_ewqe_ud_hdr *ud_hdr =
+		(struct ocrdma_ewqe_ud_hdr *)(hdr + 1);
+	struct ocrdma_ah *ah = get_ocrdma_ah(wr->wr.ud.ah);
+
+	ud_hdr->rsvd_dest_qpn = wr->wr.ud.remote_qpn;
+	if (qp->qp_type == IB_QPT_GSI)
+		ud_hdr->qkey = qp->qkey;
+	else
+		ud_hdr->qkey = wr->wr.ud.remote_qkey;
+	ud_hdr->rsvd_ahid = ah->id;
+}
+
+static void ocrdma_build_sges(struct ocrdma_hdr_wqe *hdr,
+			      struct ocrdma_sge *sge, int num_sge,
+			      struct ib_sge *sg_list)
+{
+	int i;
+
+	for (i = 0; i < num_sge; i++) {
+		sge[i].lrkey = sg_list[i].lkey;
+		sge[i].addr_lo = sg_list[i].addr;
+		sge[i].addr_hi = upper_32_bits(sg_list[i].addr);
+		sge[i].len = sg_list[i].length;
+		hdr->total_len += sg_list[i].length;
+	}
+	if (num_sge == 0)
+		memset(sge, 0, sizeof(*sge));
+}
+
+static inline uint32_t ocrdma_sglist_len(struct ib_sge *sg_list, int num_sge)
+{
+	uint32_t total_len = 0, i;
+
+	for (i = 0; i < num_sge; i++)
+		total_len += sg_list[i].length;
+	return total_len;
+}
+
+
+static int ocrdma_build_inline_sges(struct ocrdma_qp *qp,
+				    struct ocrdma_hdr_wqe *hdr,
+				    struct ocrdma_sge *sge,
+				    struct ib_send_wr *wr, u32 wqe_size)
+{
+	int i;
+	char *dpp_addr;
+
+	if (wr->send_flags & IB_SEND_INLINE && qp->qp_type != IB_QPT_UD) {
+		hdr->total_len = ocrdma_sglist_len(wr->sg_list, wr->num_sge);
+		if (unlikely(hdr->total_len > qp->max_inline_data)) {
+			pr_err("%s() supported_len=0x%x,\n"
+			       " unspported len req=0x%x\n", __func__,
+				qp->max_inline_data, hdr->total_len);
+			return -EINVAL;
+		}
+		dpp_addr = (char *)sge;
+		for (i = 0; i < wr->num_sge; i++) {
+			memcpy(dpp_addr,
+			       (void *)(unsigned long)wr->sg_list[i].addr,
+			       wr->sg_list[i].length);
+			dpp_addr += wr->sg_list[i].length;
+		}
+
+		wqe_size += roundup(hdr->total_len, OCRDMA_WQE_ALIGN_BYTES);
+		if (0 == hdr->total_len)
+			wqe_size += sizeof(struct ocrdma_sge);
+		hdr->cw |= (OCRDMA_TYPE_INLINE << OCRDMA_WQE_TYPE_SHIFT);
+	} else {
+		ocrdma_build_sges(hdr, sge, wr->num_sge, wr->sg_list);
+		if (wr->num_sge)
+			wqe_size += (wr->num_sge * sizeof(struct ocrdma_sge));
+		else
+			wqe_size += sizeof(struct ocrdma_sge);
+		hdr->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
+	}
+	hdr->cw |= ((wqe_size / OCRDMA_WQE_STRIDE) << OCRDMA_WQE_SIZE_SHIFT);
+	return 0;
+}
+
+static int ocrdma_build_send(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
+			     struct ib_send_wr *wr)
+{
+	int status;
+	struct ocrdma_sge *sge;
+	u32 wqe_size = sizeof(*hdr);
+
+	if (qp->qp_type == IB_QPT_UD || qp->qp_type == IB_QPT_GSI) {
+		ocrdma_build_ud_hdr(qp, hdr, wr);
+		sge = (struct ocrdma_sge *)(hdr + 2);
+		wqe_size += sizeof(struct ocrdma_ewqe_ud_hdr);
+	} else {
+		sge = (struct ocrdma_sge *)(hdr + 1);
+	}
+
+	status = ocrdma_build_inline_sges(qp, hdr, sge, wr, wqe_size);
+	return status;
+}
+
+static int ocrdma_build_write(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
+			      struct ib_send_wr *wr)
+{
+	int status;
+	struct ocrdma_sge *ext_rw = (struct ocrdma_sge *)(hdr + 1);
+	struct ocrdma_sge *sge = ext_rw + 1;
+	u32 wqe_size = sizeof(*hdr) + sizeof(*ext_rw);
+
+	status = ocrdma_build_inline_sges(qp, hdr, sge, wr, wqe_size);
+	if (status)
+		return status;
+	ext_rw->addr_lo = wr->wr.rdma.remote_addr;
+	ext_rw->addr_hi = upper_32_bits(wr->wr.rdma.remote_addr);
+	ext_rw->lrkey = wr->wr.rdma.rkey;
+	ext_rw->len = hdr->total_len;
+	return 0;
+}
+
+static void ocrdma_build_read(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
+			      struct ib_send_wr *wr)
+{
+	struct ocrdma_sge *ext_rw = (struct ocrdma_sge *)(hdr + 1);
+	struct ocrdma_sge *sge = ext_rw + 1;
+	u32 wqe_size = ((wr->num_sge + 1) * sizeof(struct ocrdma_sge)) +
+	    sizeof(struct ocrdma_hdr_wqe);
+
+	ocrdma_build_sges(hdr, sge, wr->num_sge, wr->sg_list);
+	hdr->cw |= ((wqe_size / OCRDMA_WQE_STRIDE) << OCRDMA_WQE_SIZE_SHIFT);
+	hdr->cw |= (OCRDMA_READ << OCRDMA_WQE_OPCODE_SHIFT);
+	hdr->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
+
+	ext_rw->addr_lo = wr->wr.rdma.remote_addr;
+	ext_rw->addr_hi = upper_32_bits(wr->wr.rdma.remote_addr);
+	ext_rw->lrkey = wr->wr.rdma.rkey;
+	ext_rw->len = hdr->total_len;
+}
+
+static void build_frmr_pbes(struct ib_send_wr *wr, struct ocrdma_pbl *pbl_tbl,
+			    struct ocrdma_hw_mr *hwmr)
+{
+	int i;
+	u64 buf_addr = 0;
+	int num_pbes;
+	struct ocrdma_pbe *pbe;
+
+	pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+	num_pbes = 0;
+
+	/* go through the OS phy regions & fill hw pbe entries into pbls. */
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+		/* number of pbes can be more for one OS buf, when
+		 * buffers are of different sizes.
+		 * split the ib_buf to one or more pbes.
+		 */
+		buf_addr = wr->wr.fast_reg.page_list->page_list[i];
+		pbe->pa_lo = cpu_to_le32((u32) (buf_addr & PAGE_MASK));
+		pbe->pa_hi = cpu_to_le32((u32) upper_32_bits(buf_addr));
+		num_pbes += 1;
+		pbe++;
+
+		/* if the pbl is full storing the pbes,
+		 * move to next pbl.
+		*/
+		if (num_pbes == (hwmr->pbl_size/sizeof(u64))) {
+			pbl_tbl++;
+			pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+		}
+	}
+	return;
+}
+
+static int get_encoded_page_size(int pg_sz)
+{
+	/* Max size is 256M 4096 << 16 */
+	int i = 0;
+	for (; i < 17; i++)
+		if (pg_sz == (4096 << i))
+			break;
+	return i;
+}
+
+
+static int ocrdma_build_fr(struct ocrdma_qp *qp, struct ocrdma_hdr_wqe *hdr,
+			   struct ib_send_wr *wr)
+{
+	u64 fbo;
+	struct ocrdma_ewqe_fr *fast_reg = (struct ocrdma_ewqe_fr *)(hdr + 1);
+	struct ocrdma_mr *mr;
+	u32 wqe_size = sizeof(*fast_reg) + sizeof(*hdr);
+
+	wqe_size = roundup(wqe_size, OCRDMA_WQE_ALIGN_BYTES);
+
+	if (wr->wr.fast_reg.page_list_len > qp->dev->attr.max_pages_per_frmr)
+		return -EINVAL;
+
+	hdr->cw |= (OCRDMA_FR_MR << OCRDMA_WQE_OPCODE_SHIFT);
+	hdr->cw |= ((wqe_size / OCRDMA_WQE_STRIDE) << OCRDMA_WQE_SIZE_SHIFT);
+
+	if (wr->wr.fast_reg.page_list_len == 0)
+		BUG();
+	if (wr->wr.fast_reg.access_flags & IB_ACCESS_LOCAL_WRITE)
+		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_LOCAL_WR;
+	if (wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_WRITE)
+		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_REMOTE_WR;
+	if (wr->wr.fast_reg.access_flags & IB_ACCESS_REMOTE_READ)
+		hdr->rsvd_lkey_flags |= OCRDMA_LKEY_FLAG_REMOTE_RD;
+	hdr->lkey = wr->wr.fast_reg.rkey;
+	hdr->total_len = wr->wr.fast_reg.length;
+
+	fbo = wr->wr.fast_reg.iova_start -
+	    (wr->wr.fast_reg.page_list->page_list[0] & PAGE_MASK);
+
+	fast_reg->va_hi = upper_32_bits(wr->wr.fast_reg.iova_start);
+	fast_reg->va_lo = (u32) (wr->wr.fast_reg.iova_start & 0xffffffff);
+	fast_reg->fbo_hi = upper_32_bits(fbo);
+	fast_reg->fbo_lo = (u32) fbo & 0xffffffff;
+	fast_reg->num_sges = wr->wr.fast_reg.page_list_len;
+	fast_reg->size_sge =
+		get_encoded_page_size(1 << wr->wr.fast_reg.page_shift);
+	mr = (struct ocrdma_mr *) (unsigned long)
+		qp->dev->stag_arr[(hdr->lkey >> 8) & (OCRDMA_MAX_STAG - 1)];
+	build_frmr_pbes(wr, mr->hwmr.pbl_table, &mr->hwmr);
+	return 0;
+}
+
+static void ocrdma_ring_sq_db(struct ocrdma_qp *qp)
+{
+	u32 val = qp->sq.dbid | (1 << OCRDMA_DB_SQ_SHIFT);
+
+	iowrite32(val, qp->sq_db);
+}
+
+int ocrdma_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+		     struct ib_send_wr **bad_wr)
+{
+	int status = 0;
+	struct ocrdma_qp *qp = get_ocrdma_qp(ibqp);
+	struct ocrdma_hdr_wqe *hdr;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->q_lock, flags);
+	if (qp->state != OCRDMA_QPS_RTS && qp->state != OCRDMA_QPS_SQD) {
+		spin_unlock_irqrestore(&qp->q_lock, flags);
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+
+	while (wr) {
+		if (ocrdma_hwq_free_cnt(&qp->sq) == 0 ||
+		    wr->num_sge > qp->sq.max_sges) {
+			*bad_wr = wr;
+			status = -ENOMEM;
+			break;
+		}
+		hdr = ocrdma_hwq_head(&qp->sq);
+		hdr->cw = 0;
+		if (wr->send_flags & IB_SEND_SIGNALED || qp->signaled)
+			hdr->cw |= (OCRDMA_FLAG_SIG << OCRDMA_WQE_FLAGS_SHIFT);
+		if (wr->send_flags & IB_SEND_FENCE)
+			hdr->cw |=
+			    (OCRDMA_FLAG_FENCE_L << OCRDMA_WQE_FLAGS_SHIFT);
+		if (wr->send_flags & IB_SEND_SOLICITED)
+			hdr->cw |=
+			    (OCRDMA_FLAG_SOLICIT << OCRDMA_WQE_FLAGS_SHIFT);
+		hdr->total_len = 0;
+		switch (wr->opcode) {
+		case IB_WR_SEND_WITH_IMM:
+			hdr->cw |= (OCRDMA_FLAG_IMM << OCRDMA_WQE_FLAGS_SHIFT);
+			hdr->immdt = ntohl(wr->ex.imm_data);
+		case IB_WR_SEND:
+			hdr->cw |= (OCRDMA_SEND << OCRDMA_WQE_OPCODE_SHIFT);
+			ocrdma_build_send(qp, hdr, wr);
+			break;
+		case IB_WR_SEND_WITH_INV:
+			hdr->cw |= (OCRDMA_FLAG_INV << OCRDMA_WQE_FLAGS_SHIFT);
+			hdr->cw |= (OCRDMA_SEND << OCRDMA_WQE_OPCODE_SHIFT);
+			hdr->lkey = wr->ex.invalidate_rkey;
+			status = ocrdma_build_send(qp, hdr, wr);
+			break;
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			hdr->cw |= (OCRDMA_FLAG_IMM << OCRDMA_WQE_FLAGS_SHIFT);
+			hdr->immdt = ntohl(wr->ex.imm_data);
+		case IB_WR_RDMA_WRITE:
+			hdr->cw |= (OCRDMA_WRITE << OCRDMA_WQE_OPCODE_SHIFT);
+			status = ocrdma_build_write(qp, hdr, wr);
+			break;
+		case IB_WR_RDMA_READ_WITH_INV:
+			hdr->cw |= (OCRDMA_FLAG_INV << OCRDMA_WQE_FLAGS_SHIFT);
+		case IB_WR_RDMA_READ:
+			ocrdma_build_read(qp, hdr, wr);
+			break;
+		case IB_WR_LOCAL_INV:
+			hdr->cw |=
+			    (OCRDMA_LKEY_INV << OCRDMA_WQE_OPCODE_SHIFT);
+			hdr->cw |= ((sizeof(struct ocrdma_hdr_wqe) +
+					sizeof(struct ocrdma_sge)) /
+				OCRDMA_WQE_STRIDE) << OCRDMA_WQE_SIZE_SHIFT;
+			hdr->lkey = wr->ex.invalidate_rkey;
+			break;
+		case IB_WR_FAST_REG_MR:
+			status = ocrdma_build_fr(qp, hdr, wr);
+			break;
+		default:
+			status = -EINVAL;
+			break;
+		}
+		if (status) {
+			*bad_wr = wr;
+			break;
+		}
+		if (wr->send_flags & IB_SEND_SIGNALED || qp->signaled)
+			qp->wqe_wr_id_tbl[qp->sq.head].signaled = 1;
+		else
+			qp->wqe_wr_id_tbl[qp->sq.head].signaled = 0;
+		qp->wqe_wr_id_tbl[qp->sq.head].wrid = wr->wr_id;
+		ocrdma_cpu_to_le32(hdr, ((hdr->cw >> OCRDMA_WQE_SIZE_SHIFT) &
+				   OCRDMA_WQE_SIZE_MASK) * OCRDMA_WQE_STRIDE);
+		/* make sure wqe is written before adapter can access it */
+		wmb();
+		/* inform hw to start processing it */
+		ocrdma_ring_sq_db(qp);
+
+		/* update pointer, counter for next wr */
+		ocrdma_hwq_inc_head(&qp->sq);
+		wr = wr->next;
+	}
+	spin_unlock_irqrestore(&qp->q_lock, flags);
+	return status;
+}
+
+static void ocrdma_ring_rq_db(struct ocrdma_qp *qp)
+{
+	u32 val = qp->rq.dbid | (1 << OCRDMA_DB_RQ_SHIFT);
+
+	iowrite32(val, qp->rq_db);
+}
+
+static void ocrdma_build_rqe(struct ocrdma_hdr_wqe *rqe, struct ib_recv_wr *wr,
+			     u16 tag)
+{
+	u32 wqe_size = 0;
+	struct ocrdma_sge *sge;
+	if (wr->num_sge)
+		wqe_size = (wr->num_sge * sizeof(*sge)) + sizeof(*rqe);
+	else
+		wqe_size = sizeof(*sge) + sizeof(*rqe);
+
+	rqe->cw = ((wqe_size / OCRDMA_WQE_STRIDE) <<
+				OCRDMA_WQE_SIZE_SHIFT);
+	rqe->cw |= (OCRDMA_FLAG_SIG << OCRDMA_WQE_FLAGS_SHIFT);
+	rqe->cw |= (OCRDMA_TYPE_LKEY << OCRDMA_WQE_TYPE_SHIFT);
+	rqe->total_len = 0;
+	rqe->rsvd_tag = tag;
+	sge = (struct ocrdma_sge *)(rqe + 1);
+	ocrdma_build_sges(rqe, sge, wr->num_sge, wr->sg_list);
+	ocrdma_cpu_to_le32(rqe, wqe_size);
+}
+
+int ocrdma_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+		     struct ib_recv_wr **bad_wr)
+{
+	int status = 0;
+	unsigned long flags;
+	struct ocrdma_qp *qp = get_ocrdma_qp(ibqp);
+	struct ocrdma_hdr_wqe *rqe;
+
+	spin_lock_irqsave(&qp->q_lock, flags);
+	if (qp->state == OCRDMA_QPS_RST || qp->state == OCRDMA_QPS_ERR) {
+		spin_unlock_irqrestore(&qp->q_lock, flags);
+		*bad_wr = wr;
+		return -EINVAL;
+	}
+	while (wr) {
+		if (ocrdma_hwq_free_cnt(&qp->rq) == 0 ||
+		    wr->num_sge > qp->rq.max_sges) {
+			*bad_wr = wr;
+			status = -ENOMEM;
+			break;
+		}
+		rqe = ocrdma_hwq_head(&qp->rq);
+		ocrdma_build_rqe(rqe, wr, 0);
+
+		qp->rqe_wr_id_tbl[qp->rq.head] = wr->wr_id;
+		/* make sure rqe is written before adapter can access it */
+		wmb();
+
+		/* inform hw to start processing it */
+		ocrdma_ring_rq_db(qp);
+
+		/* update pointer, counter for next wr */
+		ocrdma_hwq_inc_head(&qp->rq);
+		wr = wr->next;
+	}
+	spin_unlock_irqrestore(&qp->q_lock, flags);
+	return status;
+}
+
+/* cqe for srq's rqe can potentially arrive out of order.
+ * index gives the entry in the shadow table where to store
+ * the wr_id. tag/index is returned in cqe to reference back
+ * for a given rqe.
+ */
+static int ocrdma_srq_get_idx(struct ocrdma_srq *srq)
+{
+	int row = 0;
+	int indx = 0;
+
+	for (row = 0; row < srq->bit_fields_len; row++) {
+		if (srq->idx_bit_fields[row]) {
+			indx = ffs(srq->idx_bit_fields[row]);
+			indx = (row * 32) + (indx - 1);
+			if (indx >= srq->rq.max_cnt)
+				BUG();
+			ocrdma_srq_toggle_bit(srq, indx);
+			break;
+		}
+	}
+
+	if (row == srq->bit_fields_len)
+		BUG();
+	return indx + 1; /* Use from index 1 */
+}
+
+static void ocrdma_ring_srq_db(struct ocrdma_srq *srq)
+{
+	u32 val = srq->rq.dbid | (1 << 16);
+
+	iowrite32(val, srq->db + OCRDMA_DB_GEN2_SRQ_OFFSET);
+}
+
+int ocrdma_post_srq_recv(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr)
+{
+	int status = 0;
+	unsigned long flags;
+	struct ocrdma_srq *srq;
+	struct ocrdma_hdr_wqe *rqe;
+	u16 tag;
+
+	srq = get_ocrdma_srq(ibsrq);
+
+	spin_lock_irqsave(&srq->q_lock, flags);
+	while (wr) {
+		if (ocrdma_hwq_free_cnt(&srq->rq) == 0 ||
+		    wr->num_sge > srq->rq.max_sges) {
+			status = -ENOMEM;
+			*bad_wr = wr;
+			break;
+		}
+		tag = ocrdma_srq_get_idx(srq);
+		rqe = ocrdma_hwq_head(&srq->rq);
+		ocrdma_build_rqe(rqe, wr, tag);
+
+		srq->rqe_wr_id_tbl[tag] = wr->wr_id;
+		/* make sure rqe is written before adapter can perform DMA */
+		wmb();
+		/* inform hw to start processing it */
+		ocrdma_ring_srq_db(srq);
+		/* update pointer, counter for next wr */
+		ocrdma_hwq_inc_head(&srq->rq);
+		wr = wr->next;
+	}
+	spin_unlock_irqrestore(&srq->q_lock, flags);
+	return status;
+}
+
+static enum ib_wc_status ocrdma_to_ibwc_err(u16 status)
+{
+	enum ib_wc_status ibwc_status;
+
+	switch (status) {
+	case OCRDMA_CQE_GENERAL_ERR:
+		ibwc_status = IB_WC_GENERAL_ERR;
+		break;
+	case OCRDMA_CQE_LOC_LEN_ERR:
+		ibwc_status = IB_WC_LOC_LEN_ERR;
+		break;
+	case OCRDMA_CQE_LOC_QP_OP_ERR:
+		ibwc_status = IB_WC_LOC_QP_OP_ERR;
+		break;
+	case OCRDMA_CQE_LOC_EEC_OP_ERR:
+		ibwc_status = IB_WC_LOC_EEC_OP_ERR;
+		break;
+	case OCRDMA_CQE_LOC_PROT_ERR:
+		ibwc_status = IB_WC_LOC_PROT_ERR;
+		break;
+	case OCRDMA_CQE_WR_FLUSH_ERR:
+		ibwc_status = IB_WC_WR_FLUSH_ERR;
+		break;
+	case OCRDMA_CQE_MW_BIND_ERR:
+		ibwc_status = IB_WC_MW_BIND_ERR;
+		break;
+	case OCRDMA_CQE_BAD_RESP_ERR:
+		ibwc_status = IB_WC_BAD_RESP_ERR;
+		break;
+	case OCRDMA_CQE_LOC_ACCESS_ERR:
+		ibwc_status = IB_WC_LOC_ACCESS_ERR;
+		break;
+	case OCRDMA_CQE_REM_INV_REQ_ERR:
+		ibwc_status = IB_WC_REM_INV_REQ_ERR;
+		break;
+	case OCRDMA_CQE_REM_ACCESS_ERR:
+		ibwc_status = IB_WC_REM_ACCESS_ERR;
+		break;
+	case OCRDMA_CQE_REM_OP_ERR:
+		ibwc_status = IB_WC_REM_OP_ERR;
+		break;
+	case OCRDMA_CQE_RETRY_EXC_ERR:
+		ibwc_status = IB_WC_RETRY_EXC_ERR;
+		break;
+	case OCRDMA_CQE_RNR_RETRY_EXC_ERR:
+		ibwc_status = IB_WC_RNR_RETRY_EXC_ERR;
+		break;
+	case OCRDMA_CQE_LOC_RDD_VIOL_ERR:
+		ibwc_status = IB_WC_LOC_RDD_VIOL_ERR;
+		break;
+	case OCRDMA_CQE_REM_INV_RD_REQ_ERR:
+		ibwc_status = IB_WC_REM_INV_RD_REQ_ERR;
+		break;
+	case OCRDMA_CQE_REM_ABORT_ERR:
+		ibwc_status = IB_WC_REM_ABORT_ERR;
+		break;
+	case OCRDMA_CQE_INV_EECN_ERR:
+		ibwc_status = IB_WC_INV_EECN_ERR;
+		break;
+	case OCRDMA_CQE_INV_EEC_STATE_ERR:
+		ibwc_status = IB_WC_INV_EEC_STATE_ERR;
+		break;
+	case OCRDMA_CQE_FATAL_ERR:
+		ibwc_status = IB_WC_FATAL_ERR;
+		break;
+	case OCRDMA_CQE_RESP_TIMEOUT_ERR:
+		ibwc_status = IB_WC_RESP_TIMEOUT_ERR;
+		break;
+	default:
+		ibwc_status = IB_WC_GENERAL_ERR;
+		break;
+	}
+	return ibwc_status;
+}
+
+static void ocrdma_update_wc(struct ocrdma_qp *qp, struct ib_wc *ibwc,
+		      u32 wqe_idx)
+{
+	struct ocrdma_hdr_wqe *hdr;
+	struct ocrdma_sge *rw;
+	int opcode;
+
+	hdr = ocrdma_hwq_head_from_idx(&qp->sq, wqe_idx);
+
+	ibwc->wr_id = qp->wqe_wr_id_tbl[wqe_idx].wrid;
+	/* Undo the hdr->cw swap */
+	opcode = le32_to_cpu(hdr->cw) & OCRDMA_WQE_OPCODE_MASK;
+	switch (opcode) {
+	case OCRDMA_WRITE:
+		ibwc->opcode = IB_WC_RDMA_WRITE;
+		break;
+	case OCRDMA_READ:
+		rw = (struct ocrdma_sge *)(hdr + 1);
+		ibwc->opcode = IB_WC_RDMA_READ;
+		ibwc->byte_len = rw->len;
+		break;
+	case OCRDMA_SEND:
+		ibwc->opcode = IB_WC_SEND;
+		break;
+	case OCRDMA_FR_MR:
+		ibwc->opcode = IB_WC_FAST_REG_MR;
+		break;
+	case OCRDMA_LKEY_INV:
+		ibwc->opcode = IB_WC_LOCAL_INV;
+		break;
+	default:
+		ibwc->status = IB_WC_GENERAL_ERR;
+		pr_err("%s() invalid opcode received = 0x%x\n",
+		       __func__, hdr->cw & OCRDMA_WQE_OPCODE_MASK);
+		break;
+	}
+}
+
+static void ocrdma_set_cqe_status_flushed(struct ocrdma_qp *qp,
+						struct ocrdma_cqe *cqe)
+{
+	if (is_cqe_for_sq(cqe)) {
+		cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+				cqe->flags_status_srcqpn) &
+					~OCRDMA_CQE_STATUS_MASK);
+		cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+				cqe->flags_status_srcqpn) |
+				(OCRDMA_CQE_WR_FLUSH_ERR <<
+					OCRDMA_CQE_STATUS_SHIFT));
+	} else {
+		if (qp->qp_type == IB_QPT_UD || qp->qp_type == IB_QPT_GSI) {
+			cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+					cqe->flags_status_srcqpn) &
+						~OCRDMA_CQE_UD_STATUS_MASK);
+			cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+					cqe->flags_status_srcqpn) |
+					(OCRDMA_CQE_WR_FLUSH_ERR <<
+						OCRDMA_CQE_UD_STATUS_SHIFT));
+		} else {
+			cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+					cqe->flags_status_srcqpn) &
+						~OCRDMA_CQE_STATUS_MASK);
+			cqe->flags_status_srcqpn = cpu_to_le32(le32_to_cpu(
+					cqe->flags_status_srcqpn) |
+					(OCRDMA_CQE_WR_FLUSH_ERR <<
+						OCRDMA_CQE_STATUS_SHIFT));
+		}
+	}
+}
+
+static bool ocrdma_update_err_cqe(struct ib_wc *ibwc, struct ocrdma_cqe *cqe,
+				  struct ocrdma_qp *qp, int status)
+{
+	bool expand = false;
+
+	ibwc->byte_len = 0;
+	ibwc->qp = &qp->ibqp;
+	ibwc->status = ocrdma_to_ibwc_err(status);
+
+	ocrdma_flush_qp(qp);
+	ocrdma_qp_state_change(qp, IB_QPS_ERR, NULL);
+
+	/* if wqe/rqe pending for which cqe needs to be returned,
+	 * trigger inflating it.
+	 */
+	if (!is_hw_rq_empty(qp) || !is_hw_sq_empty(qp)) {
+		expand = true;
+		ocrdma_set_cqe_status_flushed(qp, cqe);
+	}
+	return expand;
+}
+
+static int ocrdma_update_err_rcqe(struct ib_wc *ibwc, struct ocrdma_cqe *cqe,
+				  struct ocrdma_qp *qp, int status)
+{
+	ibwc->opcode = IB_WC_RECV;
+	ibwc->wr_id = qp->rqe_wr_id_tbl[qp->rq.tail];
+	ocrdma_hwq_inc_tail(&qp->rq);
+
+	return ocrdma_update_err_cqe(ibwc, cqe, qp, status);
+}
+
+static int ocrdma_update_err_scqe(struct ib_wc *ibwc, struct ocrdma_cqe *cqe,
+				  struct ocrdma_qp *qp, int status)
+{
+	ocrdma_update_wc(qp, ibwc, qp->sq.tail);
+	ocrdma_hwq_inc_tail(&qp->sq);
+
+	return ocrdma_update_err_cqe(ibwc, cqe, qp, status);
+}
+
+
+static bool ocrdma_poll_err_scqe(struct ocrdma_qp *qp,
+				 struct ocrdma_cqe *cqe, struct ib_wc *ibwc,
+				 bool *polled, bool *stop)
+{
+	bool expand;
+	int status = (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_STATUS_MASK) >> OCRDMA_CQE_STATUS_SHIFT;
+
+	/* when hw sq is empty, but rq is not empty, so we continue
+	 * to keep the cqe in order to get the cq event again.
+	 */
+	if (is_hw_sq_empty(qp) && !is_hw_rq_empty(qp)) {
+		/* when cq for rq and sq is same, it is safe to return
+		 * flush cqe for RQEs.
+		 */
+		if (!qp->srq && (qp->sq_cq == qp->rq_cq)) {
+			*polled = true;
+			status = OCRDMA_CQE_WR_FLUSH_ERR;
+			expand = ocrdma_update_err_rcqe(ibwc, cqe, qp, status);
+		} else {
+			/* stop processing further cqe as this cqe is used for
+			 * triggering cq event on buddy cq of RQ.
+			 * When QP is destroyed, this cqe will be removed
+			 * from the cq's hardware q.
+			 */
+			*polled = false;
+			*stop = true;
+			expand = false;
+		}
+	} else {
+		*polled = true;
+		expand = ocrdma_update_err_scqe(ibwc, cqe, qp, status);
+	}
+	return expand;
+}
+
+static bool ocrdma_poll_success_scqe(struct ocrdma_qp *qp,
+				     struct ocrdma_cqe *cqe,
+				     struct ib_wc *ibwc, bool *polled)
+{
+	bool expand = false;
+	int tail = qp->sq.tail;
+	u32 wqe_idx;
+
+	if (!qp->wqe_wr_id_tbl[tail].signaled) {
+		*polled = false;    /* WC cannot be consumed yet */
+	} else {
+		ibwc->status = IB_WC_SUCCESS;
+		ibwc->wc_flags = 0;
+		ibwc->qp = &qp->ibqp;
+		ocrdma_update_wc(qp, ibwc, tail);
+		*polled = true;
+	}
+	wqe_idx = (le32_to_cpu(cqe->wq.wqeidx) &
+			OCRDMA_CQE_WQEIDX_MASK) & qp->sq.max_wqe_idx;
+	if (tail != wqe_idx)
+		expand = true; /* Coalesced CQE can't be consumed yet */
+
+	ocrdma_hwq_inc_tail(&qp->sq);
+	return expand;
+}
+
+static bool ocrdma_poll_scqe(struct ocrdma_qp *qp, struct ocrdma_cqe *cqe,
+			     struct ib_wc *ibwc, bool *polled, bool *stop)
+{
+	int status;
+	bool expand;
+
+	status = (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_STATUS_MASK) >> OCRDMA_CQE_STATUS_SHIFT;
+
+	if (status == OCRDMA_CQE_SUCCESS)
+		expand = ocrdma_poll_success_scqe(qp, cqe, ibwc, polled);
+	else
+		expand = ocrdma_poll_err_scqe(qp, cqe, ibwc, polled, stop);
+	return expand;
+}
+
+static int ocrdma_update_ud_rcqe(struct ib_wc *ibwc, struct ocrdma_cqe *cqe)
+{
+	int status;
+
+	status = (le32_to_cpu(cqe->flags_status_srcqpn) &
+		OCRDMA_CQE_UD_STATUS_MASK) >> OCRDMA_CQE_UD_STATUS_SHIFT;
+	ibwc->src_qp = le32_to_cpu(cqe->flags_status_srcqpn) &
+						OCRDMA_CQE_SRCQP_MASK;
+	ibwc->pkey_index = le32_to_cpu(cqe->ud.rxlen_pkey) &
+						OCRDMA_CQE_PKEY_MASK;
+	ibwc->wc_flags = IB_WC_GRH;
+	ibwc->byte_len = (le32_to_cpu(cqe->ud.rxlen_pkey) >>
+					OCRDMA_CQE_UD_XFER_LEN_SHIFT);
+	return status;
+}
+
+static void ocrdma_update_free_srq_cqe(struct ib_wc *ibwc,
+				       struct ocrdma_cqe *cqe,
+				       struct ocrdma_qp *qp)
+{
+	unsigned long flags;
+	struct ocrdma_srq *srq;
+	u32 wqe_idx;
+
+	srq = get_ocrdma_srq(qp->ibqp.srq);
+	wqe_idx = (le32_to_cpu(cqe->rq.buftag_qpn) >>
+		OCRDMA_CQE_BUFTAG_SHIFT) & srq->rq.max_wqe_idx;
+	if (wqe_idx < 1)
+		BUG();
+
+	ibwc->wr_id = srq->rqe_wr_id_tbl[wqe_idx];
+	spin_lock_irqsave(&srq->q_lock, flags);
+	ocrdma_srq_toggle_bit(srq, wqe_idx - 1);
+	spin_unlock_irqrestore(&srq->q_lock, flags);
+	ocrdma_hwq_inc_tail(&srq->rq);
+}
+
+static bool ocrdma_poll_err_rcqe(struct ocrdma_qp *qp, struct ocrdma_cqe *cqe,
+				struct ib_wc *ibwc, bool *polled, bool *stop,
+				int status)
+{
+	bool expand;
+
+	/* when hw_rq is empty, but wq is not empty, so continue
+	 * to keep the cqe to get the cq event again.
+	 */
+	if (is_hw_rq_empty(qp) && !is_hw_sq_empty(qp)) {
+		if (!qp->srq && (qp->sq_cq == qp->rq_cq)) {
+			*polled = true;
+			status = OCRDMA_CQE_WR_FLUSH_ERR;
+			expand = ocrdma_update_err_scqe(ibwc, cqe, qp, status);
+		} else {
+			*polled = false;
+			*stop = true;
+			expand = false;
+		}
+	} else {
+		*polled = true;
+		expand = ocrdma_update_err_rcqe(ibwc, cqe, qp, status);
+	}
+	return expand;
+}
+
+static void ocrdma_poll_success_rcqe(struct ocrdma_qp *qp,
+				     struct ocrdma_cqe *cqe, struct ib_wc *ibwc)
+{
+	ibwc->opcode = IB_WC_RECV;
+	ibwc->qp = &qp->ibqp;
+	ibwc->status = IB_WC_SUCCESS;
+
+	if (qp->qp_type == IB_QPT_UD || qp->qp_type == IB_QPT_GSI)
+		ocrdma_update_ud_rcqe(ibwc, cqe);
+	else
+		ibwc->byte_len = le32_to_cpu(cqe->rq.rxlen);
+
+	if (is_cqe_imm(cqe)) {
+		ibwc->ex.imm_data = htonl(le32_to_cpu(cqe->rq.lkey_immdt));
+		ibwc->wc_flags |= IB_WC_WITH_IMM;
+	} else if (is_cqe_wr_imm(cqe)) {
+		ibwc->opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		ibwc->ex.imm_data = htonl(le32_to_cpu(cqe->rq.lkey_immdt));
+		ibwc->wc_flags |= IB_WC_WITH_IMM;
+	} else if (is_cqe_invalidated(cqe)) {
+		ibwc->ex.invalidate_rkey = le32_to_cpu(cqe->rq.lkey_immdt);
+		ibwc->wc_flags |= IB_WC_WITH_INVALIDATE;
+	}
+	if (qp->ibqp.srq) {
+		ocrdma_update_free_srq_cqe(ibwc, cqe, qp);
+	} else {
+		ibwc->wr_id = qp->rqe_wr_id_tbl[qp->rq.tail];
+		ocrdma_hwq_inc_tail(&qp->rq);
+	}
+}
+
+static bool ocrdma_poll_rcqe(struct ocrdma_qp *qp, struct ocrdma_cqe *cqe,
+			     struct ib_wc *ibwc, bool *polled, bool *stop)
+{
+	int status;
+	bool expand = false;
+
+	ibwc->wc_flags = 0;
+	if (qp->qp_type == IB_QPT_UD || qp->qp_type == IB_QPT_GSI) {
+		status = (le32_to_cpu(cqe->flags_status_srcqpn) &
+					OCRDMA_CQE_UD_STATUS_MASK) >>
+					OCRDMA_CQE_UD_STATUS_SHIFT;
+	} else {
+		status = (le32_to_cpu(cqe->flags_status_srcqpn) &
+			     OCRDMA_CQE_STATUS_MASK) >> OCRDMA_CQE_STATUS_SHIFT;
+	}
+
+	if (status == OCRDMA_CQE_SUCCESS) {
+		*polled = true;
+		ocrdma_poll_success_rcqe(qp, cqe, ibwc);
+	} else {
+		expand = ocrdma_poll_err_rcqe(qp, cqe, ibwc, polled, stop,
+					      status);
+	}
+	return expand;
+}
+
+static void ocrdma_change_cq_phase(struct ocrdma_cq *cq, struct ocrdma_cqe *cqe,
+				   u16 cur_getp)
+{
+	if (cq->phase_change) {
+		if (cur_getp == 0)
+			cq->phase = (~cq->phase & OCRDMA_CQE_VALID);
+	} else {
+		/* clear valid bit */
+		cqe->flags_status_srcqpn = 0;
+	}
+}
+
+static int ocrdma_poll_hwcq(struct ocrdma_cq *cq, int num_entries,
+			    struct ib_wc *ibwc)
+{
+	u16 qpn = 0;
+	int i = 0;
+	bool expand = false;
+	int polled_hw_cqes = 0;
+	struct ocrdma_qp *qp = NULL;
+	struct ocrdma_dev *dev = get_ocrdma_dev(cq->ibcq.device);
+	struct ocrdma_cqe *cqe;
+	u16 cur_getp; bool polled = false; bool stop = false;
+
+	cur_getp = cq->getp;
+	while (num_entries) {
+		cqe = cq->va + cur_getp;
+		/* check whether valid cqe or not */
+		if (!is_cqe_valid(cq, cqe))
+			break;
+		qpn = (le32_to_cpu(cqe->cmn.qpn) & OCRDMA_CQE_QPN_MASK);
+		/* ignore discarded cqe */
+		if (qpn == 0)
+			goto skip_cqe;
+		qp = dev->qp_tbl[qpn];
+		BUG_ON(qp == NULL);
+
+		if (is_cqe_for_sq(cqe)) {
+			expand = ocrdma_poll_scqe(qp, cqe, ibwc, &polled,
+						  &stop);
+		} else {
+			expand = ocrdma_poll_rcqe(qp, cqe, ibwc, &polled,
+						  &stop);
+		}
+		if (expand)
+			goto expand_cqe;
+		if (stop)
+			goto stop_cqe;
+		/* clear qpn to avoid duplicate processing by discard_cqe() */
+		cqe->cmn.qpn = 0;
+skip_cqe:
+		polled_hw_cqes += 1;
+		cur_getp = (cur_getp + 1) % cq->max_hw_cqe;
+		ocrdma_change_cq_phase(cq, cqe, cur_getp);
+expand_cqe:
+		if (polled) {
+			num_entries -= 1;
+			i += 1;
+			ibwc = ibwc + 1;
+			polled = false;
+		}
+	}
+stop_cqe:
+	cq->getp = cur_getp;
+	if (cq->deferred_arm) {
+		ocrdma_ring_cq_db(dev, cq->id, true, cq->deferred_sol,
+				  polled_hw_cqes);
+		cq->deferred_arm = false;
+		cq->deferred_sol = false;
+	} else {
+		/* We need to pop the CQE. No need to arm */
+		ocrdma_ring_cq_db(dev, cq->id, false, cq->deferred_sol,
+				  polled_hw_cqes);
+		cq->deferred_sol = false;
+	}
+
+	return i;
+}
+
+/* insert error cqe if the QP's SQ or RQ's CQ matches the CQ under poll. */
+static int ocrdma_add_err_cqe(struct ocrdma_cq *cq, int num_entries,
+			      struct ocrdma_qp *qp, struct ib_wc *ibwc)
+{
+	int err_cqes = 0;
+
+	while (num_entries) {
+		if (is_hw_sq_empty(qp) && is_hw_rq_empty(qp))
+			break;
+		if (!is_hw_sq_empty(qp) && qp->sq_cq == cq) {
+			ocrdma_update_wc(qp, ibwc, qp->sq.tail);
+			ocrdma_hwq_inc_tail(&qp->sq);
+		} else if (!is_hw_rq_empty(qp) && qp->rq_cq == cq) {
+			ibwc->wr_id = qp->rqe_wr_id_tbl[qp->rq.tail];
+			ocrdma_hwq_inc_tail(&qp->rq);
+		} else {
+			return err_cqes;
+		}
+		ibwc->byte_len = 0;
+		ibwc->status = IB_WC_WR_FLUSH_ERR;
+		ibwc = ibwc + 1;
+		err_cqes += 1;
+		num_entries -= 1;
+	}
+	return err_cqes;
+}
+
+int ocrdma_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *wc)
+{
+	int cqes_to_poll = num_entries;
+	struct ocrdma_cq *cq = get_ocrdma_cq(ibcq);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibcq->device);
+	int num_os_cqe = 0, err_cqes = 0;
+	struct ocrdma_qp *qp;
+	unsigned long flags;
+
+	/* poll cqes from adapter CQ */
+	spin_lock_irqsave(&cq->cq_lock, flags);
+	num_os_cqe = ocrdma_poll_hwcq(cq, cqes_to_poll, wc);
+	spin_unlock_irqrestore(&cq->cq_lock, flags);
+	cqes_to_poll -= num_os_cqe;
+
+	if (cqes_to_poll) {
+		wc = wc + num_os_cqe;
+		/* adapter returns single error cqe when qp moves to
+		 * error state. So insert error cqes with wc_status as
+		 * FLUSHED for pending WQEs and RQEs of QP's SQ and RQ
+		 * respectively which uses this CQ.
+		 */
+		spin_lock_irqsave(&dev->flush_q_lock, flags);
+		list_for_each_entry(qp, &cq->sq_head, sq_entry) {
+			if (cqes_to_poll == 0)
+				break;
+			err_cqes = ocrdma_add_err_cqe(cq, cqes_to_poll, qp, wc);
+			cqes_to_poll -= err_cqes;
+			num_os_cqe += err_cqes;
+			wc = wc + err_cqes;
+		}
+		spin_unlock_irqrestore(&dev->flush_q_lock, flags);
+	}
+	return num_os_cqe;
+}
+
+int ocrdma_arm_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags cq_flags)
+{
+	struct ocrdma_cq *cq = get_ocrdma_cq(ibcq);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibcq->device);
+	u16 cq_id;
+	unsigned long flags;
+	bool arm_needed = false, sol_needed = false;
+
+	cq_id = cq->id;
+
+	spin_lock_irqsave(&cq->cq_lock, flags);
+	if (cq_flags & IB_CQ_NEXT_COMP || cq_flags & IB_CQ_SOLICITED)
+		arm_needed = true;
+	if (cq_flags & IB_CQ_SOLICITED)
+		sol_needed = true;
+
+	if (cq->first_arm) {
+		ocrdma_ring_cq_db(dev, cq_id, arm_needed, sol_needed, 0);
+		cq->first_arm = false;
+		goto skip_defer;
+	}
+	cq->deferred_arm = true;
+
+skip_defer:
+	cq->deferred_sol = sol_needed;
+	spin_unlock_irqrestore(&cq->cq_lock, flags);
+
+	return 0;
+}
+
+struct ib_mr *ocrdma_alloc_frmr(struct ib_pd *ibpd, int max_page_list_len)
+{
+	int status;
+	struct ocrdma_mr *mr;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+
+	if (max_page_list_len > dev->attr.max_pages_per_frmr)
+		return ERR_PTR(-EINVAL);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(-ENOMEM);
+
+	status = ocrdma_get_pbl_info(dev, mr, max_page_list_len);
+	if (status)
+		goto pbl_err;
+	mr->hwmr.fr_mr = 1;
+	mr->hwmr.remote_rd = 0;
+	mr->hwmr.remote_wr = 0;
+	mr->hwmr.local_rd = 0;
+	mr->hwmr.local_wr = 0;
+	mr->hwmr.mw_bind = 0;
+	status = ocrdma_build_pbl_tbl(dev, &mr->hwmr);
+	if (status)
+		goto pbl_err;
+	status = ocrdma_reg_mr(dev, &mr->hwmr, pd->id, 0);
+	if (status)
+		goto mbx_err;
+	mr->ibmr.rkey = mr->hwmr.lkey;
+	mr->ibmr.lkey = mr->hwmr.lkey;
+	dev->stag_arr[(mr->hwmr.lkey >> 8) & (OCRDMA_MAX_STAG - 1)] =
+		(unsigned long) mr;
+	return &mr->ibmr;
+mbx_err:
+	ocrdma_free_mr_pbl_tbl(dev, &mr->hwmr);
+pbl_err:
+	kfree(mr);
+	return ERR_PTR(-ENOMEM);
+}
+
+struct ib_fast_reg_page_list *ocrdma_alloc_frmr_page_list(struct ib_device
+							  *ibdev,
+							  int page_list_len)
+{
+	struct ib_fast_reg_page_list *frmr_list;
+	int size;
+
+	size = sizeof(*frmr_list) + (page_list_len * sizeof(u64));
+	frmr_list = kzalloc(size, GFP_KERNEL);
+	if (!frmr_list)
+		return ERR_PTR(-ENOMEM);
+	frmr_list->page_list = (u64 *)(frmr_list + 1);
+	return frmr_list;
+}
+
+void ocrdma_free_frmr_page_list(struct ib_fast_reg_page_list *page_list)
+{
+	kfree(page_list);
+}
+
+#define MAX_KERNEL_PBE_SIZE 65536
+static inline int count_kernel_pbes(struct ib_phys_buf *buf_list,
+				    int buf_cnt, u32 *pbe_size)
+{
+	u64 total_size = 0;
+	u64 buf_size = 0;
+	int i;
+	*pbe_size = roundup(buf_list[0].size, PAGE_SIZE);
+	*pbe_size = roundup_pow_of_two(*pbe_size);
+
+	/* find the smallest PBE size that we can have */
+	for (i = 0; i < buf_cnt; i++) {
+		/* first addr may not be page aligned, so ignore checking */
+		if ((i != 0) && ((buf_list[i].addr & ~PAGE_MASK) ||
+				 (buf_list[i].size & ~PAGE_MASK))) {
+			return 0;
+		}
+
+		/* if configured PBE size is greater then the chosen one,
+		 * reduce the PBE size.
+		 */
+		buf_size = roundup(buf_list[i].size, PAGE_SIZE);
+		/* pbe_size has to be even multiple of 4K 1,2,4,8...*/
+		buf_size = roundup_pow_of_two(buf_size);
+		if (*pbe_size > buf_size)
+			*pbe_size = buf_size;
+
+		total_size += buf_size;
+	}
+	*pbe_size = *pbe_size > MAX_KERNEL_PBE_SIZE ?
+	    (MAX_KERNEL_PBE_SIZE) : (*pbe_size);
+
+	/* num_pbes = total_size / (*pbe_size);  this is implemented below. */
+
+	return total_size >> ilog2(*pbe_size);
+}
+
+static void build_kernel_pbes(struct ib_phys_buf *buf_list, int ib_buf_cnt,
+			      u32 pbe_size, struct ocrdma_pbl *pbl_tbl,
+			      struct ocrdma_hw_mr *hwmr)
+{
+	int i;
+	int idx;
+	int pbes_per_buf = 0;
+	u64 buf_addr = 0;
+	int num_pbes;
+	struct ocrdma_pbe *pbe;
+	int total_num_pbes = 0;
+
+	if (!hwmr->num_pbes)
+		return;
+
+	pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+	num_pbes = 0;
+
+	/* go through the OS phy regions & fill hw pbe entries into pbls. */
+	for (i = 0; i < ib_buf_cnt; i++) {
+		buf_addr = buf_list[i].addr;
+		pbes_per_buf =
+		    roundup_pow_of_two(roundup(buf_list[i].size, PAGE_SIZE)) /
+		    pbe_size;
+		hwmr->len += buf_list[i].size;
+		/* number of pbes can be more for one OS buf, when
+		 * buffers are of different sizes.
+		 * split the ib_buf to one or more pbes.
+		 */
+		for (idx = 0; idx < pbes_per_buf; idx++) {
+			/* we program always page aligned addresses,
+			 * first unaligned address is taken care by fbo.
+			 */
+			if (i == 0) {
+				/* for non zero fbo, assign the
+				 * start of the page.
+				 */
+				pbe->pa_lo =
+				    cpu_to_le32((u32) (buf_addr & PAGE_MASK));
+				pbe->pa_hi =
+				    cpu_to_le32((u32) upper_32_bits(buf_addr));
+			} else {
+				pbe->pa_lo =
+				    cpu_to_le32((u32) (buf_addr & 0xffffffff));
+				pbe->pa_hi =
+				    cpu_to_le32((u32) upper_32_bits(buf_addr));
+			}
+			buf_addr += pbe_size;
+			num_pbes += 1;
+			total_num_pbes += 1;
+			pbe++;
+
+			if (total_num_pbes == hwmr->num_pbes)
+				goto mr_tbl_done;
+			/* if the pbl is full storing the pbes,
+			 * move to next pbl.
+			 */
+			if (num_pbes == (hwmr->pbl_size/sizeof(u64))) {
+				pbl_tbl++;
+				pbe = (struct ocrdma_pbe *)pbl_tbl->va;
+				num_pbes = 0;
+			}
+		}
+	}
+mr_tbl_done:
+	return;
+}
+
+struct ib_mr *ocrdma_reg_kernel_mr(struct ib_pd *ibpd,
+				   struct ib_phys_buf *buf_list,
+				   int buf_cnt, int acc, u64 *iova_start)
+{
+	int status = -ENOMEM;
+	struct ocrdma_mr *mr;
+	struct ocrdma_pd *pd = get_ocrdma_pd(ibpd);
+	struct ocrdma_dev *dev = get_ocrdma_dev(ibpd->device);
+	u32 num_pbes;
+	u32 pbe_size = 0;
+
+	if ((acc & IB_ACCESS_REMOTE_WRITE) && !(acc & IB_ACCESS_LOCAL_WRITE))
+		return ERR_PTR(-EINVAL);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (!mr)
+		return ERR_PTR(status);
+
+	num_pbes = count_kernel_pbes(buf_list, buf_cnt, &pbe_size);
+	if (num_pbes == 0) {
+		status = -EINVAL;
+		goto pbl_err;
+	}
+	status = ocrdma_get_pbl_info(dev, mr, num_pbes);
+	if (status)
+		goto pbl_err;
+
+	mr->hwmr.pbe_size = pbe_size;
+	mr->hwmr.fbo = *iova_start - (buf_list[0].addr & PAGE_MASK);
+	mr->hwmr.va = *iova_start;
+	mr->hwmr.local_rd = 1;
+	mr->hwmr.remote_wr = (acc & IB_ACCESS_REMOTE_WRITE) ? 1 : 0;
+	mr->hwmr.remote_rd = (acc & IB_ACCESS_REMOTE_READ) ? 1 : 0;
+	mr->hwmr.local_wr = (acc & IB_ACCESS_LOCAL_WRITE) ? 1 : 0;
+	mr->hwmr.remote_atomic = (acc & IB_ACCESS_REMOTE_ATOMIC) ? 1 : 0;
+	mr->hwmr.mw_bind = (acc & IB_ACCESS_MW_BIND) ? 1 : 0;
+
+	status = ocrdma_build_pbl_tbl(dev, &mr->hwmr);
+	if (status)
+		goto pbl_err;
+	build_kernel_pbes(buf_list, buf_cnt, pbe_size, mr->hwmr.pbl_table,
+			  &mr->hwmr);
+	status = ocrdma_reg_mr(dev, &mr->hwmr, pd->id, acc);
+	if (status)
+		goto mbx_err;
+
+	mr->ibmr.lkey = mr->hwmr.lkey;
+	if (mr->hwmr.remote_wr || mr->hwmr.remote_rd)
+		mr->ibmr.rkey = mr->hwmr.lkey;
+	return &mr->ibmr;
+
+mbx_err:
+	ocrdma_free_mr_pbl_tbl(dev, &mr->hwmr);
+pbl_err:
+	kfree(mr);
+	return ERR_PTR(status);
+}
diff --git a/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h
new file mode 100644
index 00000000000..b8f7853fd36
--- /dev/null
+++ b/drivers/infiniband/hw/ocrdma/ocrdma_verbs.h
@@ -0,0 +1,99 @@
+/*******************************************************************
+ * This file is part of the Emulex RoCE Device Driver for          *
+ * RoCE (RDMA over Converged Ethernet) adapters.                   *
+ * Copyright (C) 2008-2012 Emulex. All rights reserved.            *
+ * EMULEX and SLI are trademarks of Emulex.                        *
+ * www.emulex.com                                                  *
+ *                                                                 *
+ * This program is free software; you can redistribute it and/or   *
+ * modify it under the terms of version 2 of the GNU General       *
+ * Public License as published by the Free Software Foundation.    *
+ * This program is distributed in the hope that it will be useful. *
+ * ALL EXPRESS OR IMPLIED CONDITIONS, REPRESENTATIONS AND          *
+ * WARRANTIES, INCLUDING ANY IMPLIED WARRANTY OF MERCHANTABILITY,  *
+ * FITNESS FOR A PARTICULAR PURPOSE, OR NON-INFRINGEMENT, ARE      *
+ * DISCLAIMED, EXCEPT TO THE EXTENT THAT SUCH DISCLAIMERS ARE HELD *
+ * TO BE LEGALLY INVALID.  See the GNU General Public License for  *
+ * more details, a copy of which can be found in the file COPYING  *
+ * included with this package.                                     *
+ *
+ * Contact Information:
+ * linux-drivers@emulex.com
+ *
+ * Emulex
+ * 3333 Susan Street
+ * Costa Mesa, CA 92626
+ *******************************************************************/
+
+#ifndef __OCRDMA_VERBS_H__
+#define __OCRDMA_VERBS_H__
+
+int ocrdma_post_send(struct ib_qp *, struct ib_send_wr *,
+		     struct ib_send_wr **bad_wr);
+int ocrdma_post_recv(struct ib_qp *, struct ib_recv_wr *,
+		     struct ib_recv_wr **bad_wr);
+
+int ocrdma_poll_cq(struct ib_cq *, int num_entries, struct ib_wc *wc);
+int ocrdma_arm_cq(struct ib_cq *, enum ib_cq_notify_flags flags);
+
+int ocrdma_query_device(struct ib_device *, struct ib_device_attr *props);
+int ocrdma_query_port(struct ib_device *, u8 port, struct ib_port_attr *props);
+int ocrdma_modify_port(struct ib_device *, u8 port, int mask,
+		       struct ib_port_modify *props);
+
+void ocrdma_get_guid(struct ocrdma_dev *, u8 *guid);
+int ocrdma_query_gid(struct ib_device *, u8 port,
+		     int index, union ib_gid *gid);
+int ocrdma_query_pkey(struct ib_device *, u8 port, u16 index, u16 *pkey);
+
+struct ib_ucontext *ocrdma_alloc_ucontext(struct ib_device *,
+					  struct ib_udata *);
+int ocrdma_dealloc_ucontext(struct ib_ucontext *);
+
+int ocrdma_mmap(struct ib_ucontext *, struct vm_area_struct *vma);
+
+struct ib_pd *ocrdma_alloc_pd(struct ib_device *,
+			      struct ib_ucontext *, struct ib_udata *);
+int ocrdma_dealloc_pd(struct ib_pd *pd);
+
+struct ib_cq *ocrdma_create_cq(struct ib_device *, int entries, int vector,
+			       struct ib_ucontext *, struct ib_udata *);
+int ocrdma_resize_cq(struct ib_cq *, int cqe, struct ib_udata *);
+int ocrdma_destroy_cq(struct ib_cq *);
+
+struct ib_qp *ocrdma_create_qp(struct ib_pd *,
+			       struct ib_qp_init_attr *attrs,
+			       struct ib_udata *);
+int _ocrdma_modify_qp(struct ib_qp *, struct ib_qp_attr *attr,
+		      int attr_mask);
+int ocrdma_modify_qp(struct ib_qp *, struct ib_qp_attr *attr,
+		     int attr_mask, struct ib_udata *udata);
+int ocrdma_query_qp(struct ib_qp *,
+		    struct ib_qp_attr *qp_attr,
+		    int qp_attr_mask, struct ib_qp_init_attr *);
+int ocrdma_destroy_qp(struct ib_qp *);
+void ocrdma_del_flush_qp(struct ocrdma_qp *qp);
+
+struct ib_srq *ocrdma_create_srq(struct ib_pd *, struct ib_srq_init_attr *,
+				 struct ib_udata *);
+int ocrdma_modify_srq(struct ib_srq *, struct ib_srq_attr *,
+		      enum ib_srq_attr_mask, struct ib_udata *);
+int ocrdma_query_srq(struct ib_srq *, struct ib_srq_attr *);
+int ocrdma_destroy_srq(struct ib_srq *);
+int ocrdma_post_srq_recv(struct ib_srq *, struct ib_recv_wr *,
+			 struct ib_recv_wr **bad_recv_wr);
+
+int ocrdma_dereg_mr(struct ib_mr *);
+struct ib_mr *ocrdma_get_dma_mr(struct ib_pd *, int acc);
+struct ib_mr *ocrdma_reg_kernel_mr(struct ib_pd *,
+				   struct ib_phys_buf *buffer_list,
+				   int num_phys_buf, int acc, u64 *iova_start);
+struct ib_mr *ocrdma_reg_user_mr(struct ib_pd *, u64 start, u64 length,
+				 u64 virt, int acc, struct ib_udata *);
+struct ib_mr *ocrdma_alloc_frmr(struct ib_pd *pd, int max_page_list_len);
+struct ib_fast_reg_page_list *ocrdma_alloc_frmr_page_list(struct ib_device
+							*ibdev,
+							int page_list_len);
+void ocrdma_free_frmr_page_list(struct ib_fast_reg_page_list *page_list);
+
+#endif				/* __OCRDMA_VERBS_H__ */
diff --git a/drivers/infiniband/hw/qib/Kconfig b/drivers/infiniband/hw/qib/Kconfig
new file mode 100644
index 00000000000..495be09781b
--- /dev/null
+++ b/drivers/infiniband/hw/qib/Kconfig
@@ -0,0 +1,15 @@
+config INFINIBAND_QIB
+	tristate "Intel PCIe HCA support"
+	depends on 64BIT
+	---help---
+	This is a low-level driver for Intel PCIe QLE InfiniBand host
+	channel adapters.  This driver does not support the Intel
+	HyperTransport card (model QHT7140).
+
+config INFINIBAND_QIB_DCA
+	bool "QIB DCA support"
+	depends on INFINIBAND_QIB && DCA && SMP && !(INFINIBAND_QIB=y && DCA=m)
+	default y
+	---help---
+	Setting this enables DCA support on some Intel chip sets
+	with the iba7322 HCA.
diff --git a/drivers/infiniband/hw/qib/Makefile b/drivers/infiniband/hw/qib/Makefile
new file mode 100644
index 00000000000..57f8103e51f
--- /dev/null
+++ b/drivers/infiniband/hw/qib/Makefile
@@ -0,0 +1,16 @@
+obj-$(CONFIG_INFINIBAND_QIB) += ib_qib.o
+
+ib_qib-y := qib_cq.o qib_diag.o qib_dma.o qib_driver.o qib_eeprom.o \
+	qib_file_ops.o qib_fs.o qib_init.o qib_intr.o qib_keys.o \
+	qib_mad.o qib_mmap.o qib_mr.o qib_pcie.o qib_pio_copy.o \
+	qib_qp.o qib_qsfp.o qib_rc.o qib_ruc.o qib_sdma.o qib_srq.o \
+	qib_sysfs.o qib_twsi.o qib_tx.o qib_uc.o qib_ud.o \
+	qib_user_pages.o qib_user_sdma.o qib_verbs_mcast.o qib_iba7220.o \
+	qib_sd7220.o qib_iba7322.o qib_verbs.o
+
+# 6120 has no fallback if no MSI interrupts, others can do INTx
+ib_qib-$(CONFIG_PCI_MSI) += qib_iba6120.o
+
+ib_qib-$(CONFIG_X86_64) += qib_wc_x86_64.o
+ib_qib-$(CONFIG_PPC64) += qib_wc_ppc64.o
+ib_qib-$(CONFIG_DEBUG_FS) += qib_debugfs.o
diff --git a/drivers/infiniband/hw/qib/qib.h b/drivers/infiniband/hw/qib/qib.h
new file mode 100644
index 00000000000..c00ae093b6f
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib.h
@@ -0,0 +1,1548 @@
+#ifndef _QIB_KERNEL_H
+#define _QIB_KERNEL_H
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This header file is the base header file for qlogic_ib kernel code
+ * qib_user.h serves a similar purpose for user code.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/dma-mapping.h>
+#include <linux/mutex.h>
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+#include <linux/slab.h>
+#include <linux/io.h>
+#include <linux/fs.h>
+#include <linux/completion.h>
+#include <linux/kref.h>
+#include <linux/sched.h>
+#include <linux/kthread.h>
+
+#include "qib_common.h"
+#include "qib_verbs.h"
+
+/* only s/w major version of QLogic_IB we can handle */
+#define QIB_CHIP_VERS_MAJ 2U
+
+/* don't care about this except printing */
+#define QIB_CHIP_VERS_MIN 0U
+
+/* The Organization Unique Identifier (Mfg code), and its position in GUID */
+#define QIB_OUI 0x001175
+#define QIB_OUI_LSB 40
+
+/*
+ * per driver stats, either not device nor port-specific, or
+ * summed over all of the devices and ports.
+ * They are described by name via ipathfs filesystem, so layout
+ * and number of elements can change without breaking compatibility.
+ * If members are added or deleted qib_statnames[] in qib_fs.c must
+ * change to match.
+ */
+struct qlogic_ib_stats {
+	__u64 sps_ints; /* number of interrupts handled */
+	__u64 sps_errints; /* number of error interrupts */
+	__u64 sps_txerrs; /* tx-related packet errors */
+	__u64 sps_rcverrs; /* non-crc rcv packet errors */
+	__u64 sps_hwerrs; /* hardware errors reported (parity, etc.) */
+	__u64 sps_nopiobufs; /* no pio bufs avail from kernel */
+	__u64 sps_ctxts; /* number of contexts currently open */
+	__u64 sps_lenerrs; /* number of kernel packets where RHF != LRH len */
+	__u64 sps_buffull;
+	__u64 sps_hdrfull;
+};
+
+extern struct qlogic_ib_stats qib_stats;
+extern const struct pci_error_handlers qib_pci_err_handler;
+
+#define QIB_CHIP_SWVERSION QIB_CHIP_VERS_MAJ
+/*
+ * First-cut critierion for "device is active" is
+ * two thousand dwords combined Tx, Rx traffic per
+ * 5-second interval. SMA packets are 64 dwords,
+ * and occur "a few per second", presumably each way.
+ */
+#define QIB_TRAFFIC_ACTIVE_THRESHOLD (2000)
+
+/*
+ * Struct used to indicate which errors are logged in each of the
+ * error-counters that are logged to EEPROM. A counter is incremented
+ * _once_ (saturating at 255) for each event with any bits set in
+ * the error or hwerror register masks below.
+ */
+#define QIB_EEP_LOG_CNT (4)
+struct qib_eep_log_mask {
+	u64 errs_to_log;
+	u64 hwerrs_to_log;
+};
+
+/*
+ * Below contains all data related to a single context (formerly called port).
+ */
+
+#ifdef CONFIG_DEBUG_FS
+struct qib_opcode_stats_perctx;
+#endif
+
+struct qib_ctxtdata {
+	void **rcvegrbuf;
+	dma_addr_t *rcvegrbuf_phys;
+	/* rcvhdrq base, needs mmap before useful */
+	void *rcvhdrq;
+	/* kernel virtual address where hdrqtail is updated */
+	void *rcvhdrtail_kvaddr;
+	/*
+	 * temp buffer for expected send setup, allocated at open, instead
+	 * of each setup call
+	 */
+	void *tid_pg_list;
+	/*
+	 * Shared page for kernel to signal user processes that send buffers
+	 * need disarming.  The process should call QIB_CMD_DISARM_BUFS
+	 * or QIB_CMD_ACK_EVENT with IPATH_EVENT_DISARM_BUFS set.
+	 */
+	unsigned long *user_event_mask;
+	/* when waiting for rcv or pioavail */
+	wait_queue_head_t wait;
+	/*
+	 * rcvegr bufs base, physical, must fit
+	 * in 44 bits so 32 bit programs mmap64 44 bit works)
+	 */
+	dma_addr_t rcvegr_phys;
+	/* mmap of hdrq, must fit in 44 bits */
+	dma_addr_t rcvhdrq_phys;
+	dma_addr_t rcvhdrqtailaddr_phys;
+
+	/*
+	 * number of opens (including slave sub-contexts) on this instance
+	 * (ignoring forks, dup, etc. for now)
+	 */
+	int cnt;
+	/*
+	 * how much space to leave at start of eager TID entries for
+	 * protocol use, on each TID
+	 */
+	/* instead of calculating it */
+	unsigned ctxt;
+	/* local node of context */
+	int node_id;
+	/* non-zero if ctxt is being shared. */
+	u16 subctxt_cnt;
+	/* non-zero if ctxt is being shared. */
+	u16 subctxt_id;
+	/* number of eager TID entries. */
+	u16 rcvegrcnt;
+	/* index of first eager TID entry. */
+	u16 rcvegr_tid_base;
+	/* number of pio bufs for this ctxt (all procs, if shared) */
+	u32 piocnt;
+	/* first pio buffer for this ctxt */
+	u32 pio_base;
+	/* chip offset of PIO buffers for this ctxt */
+	u32 piobufs;
+	/* how many alloc_pages() chunks in rcvegrbuf_pages */
+	u32 rcvegrbuf_chunks;
+	/* how many egrbufs per chunk */
+	u16 rcvegrbufs_perchunk;
+	/* ilog2 of above */
+	u16 rcvegrbufs_perchunk_shift;
+	/* order for rcvegrbuf_pages */
+	size_t rcvegrbuf_size;
+	/* rcvhdrq size (for freeing) */
+	size_t rcvhdrq_size;
+	/* per-context flags for fileops/intr communication */
+	unsigned long flag;
+	/* next expected TID to check when looking for free */
+	u32 tidcursor;
+	/* WAIT_RCV that timed out, no interrupt */
+	u32 rcvwait_to;
+	/* WAIT_PIO that timed out, no interrupt */
+	u32 piowait_to;
+	/* WAIT_RCV already happened, no wait */
+	u32 rcvnowait;
+	/* WAIT_PIO already happened, no wait */
+	u32 pionowait;
+	/* total number of polled urgent packets */
+	u32 urgent;
+	/* saved total number of polled urgent packets for poll edge trigger */
+	u32 urgent_poll;
+	/* pid of process using this ctxt */
+	pid_t pid;
+	pid_t subpid[QLOGIC_IB_MAX_SUBCTXT];
+	/* same size as task_struct .comm[], command that opened context */
+	char comm[16];
+	/* pkeys set by this use of this ctxt */
+	u16 pkeys[4];
+	/* so file ops can get at unit */
+	struct qib_devdata *dd;
+	/* so funcs that need physical port can get it easily */
+	struct qib_pportdata *ppd;
+	/* A page of memory for rcvhdrhead, rcvegrhead, rcvegrtail * N */
+	void *subctxt_uregbase;
+	/* An array of pages for the eager receive buffers * N */
+	void *subctxt_rcvegrbuf;
+	/* An array of pages for the eager header queue entries * N */
+	void *subctxt_rcvhdr_base;
+	/* The version of the library which opened this ctxt */
+	u32 userversion;
+	/* Bitmask of active slaves */
+	u32 active_slaves;
+	/* Type of packets or conditions we want to poll for */
+	u16 poll_type;
+	/* receive packet sequence counter */
+	u8 seq_cnt;
+	u8 redirect_seq_cnt;
+	/* ctxt rcvhdrq head offset */
+	u32 head;
+	/* lookaside fields */
+	struct qib_qp *lookaside_qp;
+	u32 lookaside_qpn;
+	/* QPs waiting for context processing */
+	struct list_head qp_wait_list;
+#ifdef CONFIG_DEBUG_FS
+	/* verbs stats per CTX */
+	struct qib_opcode_stats_perctx *opstats;
+#endif
+};
+
+struct qib_sge_state;
+
+struct qib_sdma_txreq {
+	int                 flags;
+	int                 sg_count;
+	dma_addr_t          addr;
+	void              (*callback)(struct qib_sdma_txreq *, int);
+	u16                 start_idx;  /* sdma private */
+	u16                 next_descq_idx;  /* sdma private */
+	struct list_head    list;       /* sdma private */
+};
+
+struct qib_sdma_desc {
+	__le64 qw[2];
+};
+
+struct qib_verbs_txreq {
+	struct qib_sdma_txreq   txreq;
+	struct qib_qp           *qp;
+	struct qib_swqe         *wqe;
+	u32                     dwords;
+	u16                     hdr_dwords;
+	u16                     hdr_inx;
+	struct qib_pio_header	*align_buf;
+	struct qib_mregion	*mr;
+	struct qib_sge_state    *ss;
+};
+
+#define QIB_SDMA_TXREQ_F_USELARGEBUF  0x1
+#define QIB_SDMA_TXREQ_F_HEADTOHOST   0x2
+#define QIB_SDMA_TXREQ_F_INTREQ       0x4
+#define QIB_SDMA_TXREQ_F_FREEBUF      0x8
+#define QIB_SDMA_TXREQ_F_FREEDESC     0x10
+
+#define QIB_SDMA_TXREQ_S_OK        0
+#define QIB_SDMA_TXREQ_S_SENDERROR 1
+#define QIB_SDMA_TXREQ_S_ABORTED   2
+#define QIB_SDMA_TXREQ_S_SHUTDOWN  3
+
+/*
+ * Get/Set IB link-level config parameters for f_get/set_ib_cfg()
+ * Mostly for MADs that set or query link parameters, also ipath
+ * config interfaces
+ */
+#define QIB_IB_CFG_LIDLMC 0 /* LID (LS16b) and Mask (MS16b) */
+#define QIB_IB_CFG_LWID_ENB 2 /* allowed Link-width */
+#define QIB_IB_CFG_LWID 3 /* currently active Link-width */
+#define QIB_IB_CFG_SPD_ENB 4 /* allowed Link speeds */
+#define QIB_IB_CFG_SPD 5 /* current Link spd */
+#define QIB_IB_CFG_RXPOL_ENB 6 /* Auto-RX-polarity enable */
+#define QIB_IB_CFG_LREV_ENB 7 /* Auto-Lane-reversal enable */
+#define QIB_IB_CFG_LINKLATENCY 8 /* Link Latency (IB1.2 only) */
+#define QIB_IB_CFG_HRTBT 9 /* IB heartbeat off/enable/auto; DDR/QDR only */
+#define QIB_IB_CFG_OP_VLS 10 /* operational VLs */
+#define QIB_IB_CFG_VL_HIGH_CAP 11 /* num of VL high priority weights */
+#define QIB_IB_CFG_VL_LOW_CAP 12 /* num of VL low priority weights */
+#define QIB_IB_CFG_OVERRUN_THRESH 13 /* IB overrun threshold */
+#define QIB_IB_CFG_PHYERR_THRESH 14 /* IB PHY error threshold */
+#define QIB_IB_CFG_LINKDEFAULT 15 /* IB link default (sleep/poll) */
+#define QIB_IB_CFG_PKEYS 16 /* update partition keys */
+#define QIB_IB_CFG_MTU 17 /* update MTU in IBC */
+#define QIB_IB_CFG_LSTATE 18 /* update linkcmd and linkinitcmd in IBC */
+#define QIB_IB_CFG_VL_HIGH_LIMIT 19
+#define QIB_IB_CFG_PMA_TICKS 20 /* PMA sample tick resolution */
+#define QIB_IB_CFG_PORT 21 /* switch port we are connected to */
+
+/*
+ * for CFG_LSTATE: LINKCMD in upper 16 bits, LINKINITCMD in lower 16
+ * IB_LINKINITCMD_POLL and SLEEP are also used as set/get values for
+ * QIB_IB_CFG_LINKDEFAULT cmd
+ */
+#define   IB_LINKCMD_DOWN   (0 << 16)
+#define   IB_LINKCMD_ARMED  (1 << 16)
+#define   IB_LINKCMD_ACTIVE (2 << 16)
+#define   IB_LINKINITCMD_NOP     0
+#define   IB_LINKINITCMD_POLL    1
+#define   IB_LINKINITCMD_SLEEP   2
+#define   IB_LINKINITCMD_DISABLE 3
+
+/*
+ * valid states passed to qib_set_linkstate() user call
+ */
+#define QIB_IB_LINKDOWN         0
+#define QIB_IB_LINKARM          1
+#define QIB_IB_LINKACTIVE       2
+#define QIB_IB_LINKDOWN_ONLY    3
+#define QIB_IB_LINKDOWN_SLEEP   4
+#define QIB_IB_LINKDOWN_DISABLE 5
+
+/*
+ * These 7 values (SDR, DDR, and QDR may be ORed for auto-speed
+ * negotiation) are used for the 3rd argument to path_f_set_ib_cfg
+ * with cmd QIB_IB_CFG_SPD_ENB, by direct calls or via sysfs.  They
+ * are also the the possible values for qib_link_speed_enabled and active
+ * The values were chosen to match values used within the IB spec.
+ */
+#define QIB_IB_SDR 1
+#define QIB_IB_DDR 2
+#define QIB_IB_QDR 4
+
+#define QIB_DEFAULT_MTU 4096
+
+/* max number of IB ports supported per HCA */
+#define QIB_MAX_IB_PORTS 2
+
+/*
+ * Possible IB config parameters for f_get/set_ib_table()
+ */
+#define QIB_IB_TBL_VL_HIGH_ARB 1 /* Get/set VL high priority weights */
+#define QIB_IB_TBL_VL_LOW_ARB 2 /* Get/set VL low priority weights */
+
+/*
+ * Possible "operations" for f_rcvctrl(ppd, op, ctxt)
+ * these are bits so they can be combined, e.g.
+ * QIB_RCVCTRL_INTRAVAIL_ENB | QIB_RCVCTRL_CTXT_ENB
+ */
+#define QIB_RCVCTRL_TAILUPD_ENB 0x01
+#define QIB_RCVCTRL_TAILUPD_DIS 0x02
+#define QIB_RCVCTRL_CTXT_ENB 0x04
+#define QIB_RCVCTRL_CTXT_DIS 0x08
+#define QIB_RCVCTRL_INTRAVAIL_ENB 0x10
+#define QIB_RCVCTRL_INTRAVAIL_DIS 0x20
+#define QIB_RCVCTRL_PKEY_ENB 0x40  /* Note, default is enabled */
+#define QIB_RCVCTRL_PKEY_DIS 0x80
+#define QIB_RCVCTRL_BP_ENB 0x0100
+#define QIB_RCVCTRL_BP_DIS 0x0200
+#define QIB_RCVCTRL_TIDFLOW_ENB 0x0400
+#define QIB_RCVCTRL_TIDFLOW_DIS 0x0800
+
+/*
+ * Possible "operations" for f_sendctrl(ppd, op, var)
+ * these are bits so they can be combined, e.g.
+ * QIB_SENDCTRL_BUFAVAIL_ENB | QIB_SENDCTRL_ENB
+ * Some operations (e.g. DISARM, ABORT) are known to
+ * be "one-shot", so do not modify shadow.
+ */
+#define QIB_SENDCTRL_DISARM       (0x1000)
+#define QIB_SENDCTRL_DISARM_BUF(bufn) ((bufn) | QIB_SENDCTRL_DISARM)
+	/* available (0x2000) */
+#define QIB_SENDCTRL_AVAIL_DIS    (0x4000)
+#define QIB_SENDCTRL_AVAIL_ENB    (0x8000)
+#define QIB_SENDCTRL_AVAIL_BLIP  (0x10000)
+#define QIB_SENDCTRL_SEND_DIS    (0x20000)
+#define QIB_SENDCTRL_SEND_ENB    (0x40000)
+#define QIB_SENDCTRL_FLUSH       (0x80000)
+#define QIB_SENDCTRL_CLEAR      (0x100000)
+#define QIB_SENDCTRL_DISARM_ALL (0x200000)
+
+/*
+ * These are the generic indices for requesting per-port
+ * counter values via the f_portcntr function.  They
+ * are always returned as 64 bit values, although most
+ * are 32 bit counters.
+ */
+/* send-related counters */
+#define QIBPORTCNTR_PKTSEND         0U
+#define QIBPORTCNTR_WORDSEND        1U
+#define QIBPORTCNTR_PSXMITDATA      2U
+#define QIBPORTCNTR_PSXMITPKTS      3U
+#define QIBPORTCNTR_PSXMITWAIT      4U
+#define QIBPORTCNTR_SENDSTALL       5U
+/* receive-related counters */
+#define QIBPORTCNTR_PKTRCV          6U
+#define QIBPORTCNTR_PSRCVDATA       7U
+#define QIBPORTCNTR_PSRCVPKTS       8U
+#define QIBPORTCNTR_RCVEBP          9U
+#define QIBPORTCNTR_RCVOVFL         10U
+#define QIBPORTCNTR_WORDRCV         11U
+/* IB link related error counters */
+#define QIBPORTCNTR_RXLOCALPHYERR   12U
+#define QIBPORTCNTR_RXVLERR         13U
+#define QIBPORTCNTR_ERRICRC         14U
+#define QIBPORTCNTR_ERRVCRC         15U
+#define QIBPORTCNTR_ERRLPCRC        16U
+#define QIBPORTCNTR_BADFORMAT       17U
+#define QIBPORTCNTR_ERR_RLEN        18U
+#define QIBPORTCNTR_IBSYMBOLERR     19U
+#define QIBPORTCNTR_INVALIDRLEN     20U
+#define QIBPORTCNTR_UNSUPVL         21U
+#define QIBPORTCNTR_EXCESSBUFOVFL   22U
+#define QIBPORTCNTR_ERRLINK         23U
+#define QIBPORTCNTR_IBLINKDOWN      24U
+#define QIBPORTCNTR_IBLINKERRRECOV  25U
+#define QIBPORTCNTR_LLI             26U
+/* other error counters */
+#define QIBPORTCNTR_RXDROPPKT       27U
+#define QIBPORTCNTR_VL15PKTDROP     28U
+#define QIBPORTCNTR_ERRPKEY         29U
+#define QIBPORTCNTR_KHDROVFL        30U
+/* sampling counters (these are actually control registers) */
+#define QIBPORTCNTR_PSINTERVAL      31U
+#define QIBPORTCNTR_PSSTART         32U
+#define QIBPORTCNTR_PSSTAT          33U
+
+/* how often we check for packet activity for "power on hours (in seconds) */
+#define ACTIVITY_TIMER 5
+
+#define MAX_NAME_SIZE 64
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+struct qib_irq_notify;
+#endif
+
+struct qib_msix_entry {
+	struct msix_entry msix;
+	void *arg;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	int dca;
+	int rcv;
+	struct qib_irq_notify *notifier;
+#endif
+	char name[MAX_NAME_SIZE];
+	cpumask_var_t mask;
+};
+
+/* Below is an opaque struct. Each chip (device) can maintain
+ * private data needed for its operation, but not germane to the
+ * rest of the driver.  For convenience, we define another that
+ * is chip-specific, per-port
+ */
+struct qib_chip_specific;
+struct qib_chipport_specific;
+
+enum qib_sdma_states {
+	qib_sdma_state_s00_hw_down,
+	qib_sdma_state_s10_hw_start_up_wait,
+	qib_sdma_state_s20_idle,
+	qib_sdma_state_s30_sw_clean_up_wait,
+	qib_sdma_state_s40_hw_clean_up_wait,
+	qib_sdma_state_s50_hw_halt_wait,
+	qib_sdma_state_s99_running,
+};
+
+enum qib_sdma_events {
+	qib_sdma_event_e00_go_hw_down,
+	qib_sdma_event_e10_go_hw_start,
+	qib_sdma_event_e20_hw_started,
+	qib_sdma_event_e30_go_running,
+	qib_sdma_event_e40_sw_cleaned,
+	qib_sdma_event_e50_hw_cleaned,
+	qib_sdma_event_e60_hw_halted,
+	qib_sdma_event_e70_go_idle,
+	qib_sdma_event_e7220_err_halted,
+	qib_sdma_event_e7322_err_halted,
+	qib_sdma_event_e90_timer_tick,
+};
+
+extern char *qib_sdma_state_names[];
+extern char *qib_sdma_event_names[];
+
+struct sdma_set_state_action {
+	unsigned op_enable:1;
+	unsigned op_intenable:1;
+	unsigned op_halt:1;
+	unsigned op_drain:1;
+	unsigned go_s99_running_tofalse:1;
+	unsigned go_s99_running_totrue:1;
+};
+
+struct qib_sdma_state {
+	struct kref          kref;
+	struct completion    comp;
+	enum qib_sdma_states current_state;
+	struct sdma_set_state_action *set_state_action;
+	unsigned             current_op;
+	unsigned             go_s99_running;
+	unsigned             first_sendbuf;
+	unsigned             last_sendbuf; /* really last +1 */
+	/* debugging/devel */
+	enum qib_sdma_states previous_state;
+	unsigned             previous_op;
+	enum qib_sdma_events last_event;
+};
+
+struct xmit_wait {
+	struct timer_list timer;
+	u64 counter;
+	u8 flags;
+	struct cache {
+		u64 psxmitdata;
+		u64 psrcvdata;
+		u64 psxmitpkts;
+		u64 psrcvpkts;
+		u64 psxmitwait;
+	} counter_cache;
+};
+
+/*
+ * The structure below encapsulates data relevant to a physical IB Port.
+ * Current chips support only one such port, but the separation
+ * clarifies things a bit. Note that to conform to IB conventions,
+ * port-numbers are one-based. The first or only port is port1.
+ */
+struct qib_pportdata {
+	struct qib_ibport ibport_data;
+
+	struct qib_devdata *dd;
+	struct qib_chippport_specific *cpspec; /* chip-specific per-port */
+	struct kobject pport_kobj;
+	struct kobject pport_cc_kobj;
+	struct kobject sl2vl_kobj;
+	struct kobject diagc_kobj;
+
+	/* GUID for this interface, in network order */
+	__be64 guid;
+
+	/* QIB_POLL, etc. link-state specific flags, per port */
+	u32 lflags;
+	/* qib_lflags driver is waiting for */
+	u32 state_wanted;
+	spinlock_t lflags_lock;
+
+	/* ref count for each pkey */
+	atomic_t pkeyrefs[4];
+
+	/*
+	 * this address is mapped readonly into user processes so they can
+	 * get status cheaply, whenever they want.  One qword of status per port
+	 */
+	u64 *statusp;
+
+	/* SendDMA related entries */
+
+	/* read mostly */
+	struct qib_sdma_desc *sdma_descq;
+	struct workqueue_struct *qib_wq;
+	struct qib_sdma_state sdma_state;
+	dma_addr_t       sdma_descq_phys;
+	volatile __le64 *sdma_head_dma; /* DMA'ed by chip */
+	dma_addr_t       sdma_head_phys;
+	u16                   sdma_descq_cnt;
+
+	/* read/write using lock */
+	spinlock_t            sdma_lock ____cacheline_aligned_in_smp;
+	struct list_head      sdma_activelist;
+	struct list_head      sdma_userpending;
+	u64                   sdma_descq_added;
+	u64                   sdma_descq_removed;
+	u16                   sdma_descq_tail;
+	u16                   sdma_descq_head;
+	u8                    sdma_generation;
+	u8                    sdma_intrequest;
+
+	struct tasklet_struct sdma_sw_clean_up_task
+		____cacheline_aligned_in_smp;
+
+	wait_queue_head_t state_wait; /* for state_wanted */
+
+	/* HoL blocking for SMP replies */
+	unsigned          hol_state;
+	struct timer_list hol_timer;
+
+	/*
+	 * Shadow copies of registers; size indicates read access size.
+	 * Most of them are readonly, but some are write-only register,
+	 * where we manipulate the bits in the shadow copy, and then write
+	 * the shadow copy to qlogic_ib.
+	 *
+	 * We deliberately make most of these 32 bits, since they have
+	 * restricted range.  For any that we read, we won't to generate 32
+	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
+	 * transactions for a 64 bit read, and we want to avoid unnecessary
+	 * bus transactions.
+	 */
+
+	/* This is the 64 bit group */
+	/* last ibcstatus.  opaque outside chip-specific code */
+	u64 lastibcstat;
+
+	/* these are the "32 bit" regs */
+
+	/*
+	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
+	 * all expect bit fields to be "unsigned long"
+	 */
+	unsigned long p_rcvctrl; /* shadow per-port rcvctrl */
+	unsigned long p_sendctrl; /* shadow per-port sendctrl */
+
+	u32 ibmtu; /* The MTU programmed for this unit */
+	/*
+	 * Current max size IB packet (in bytes) including IB headers, that
+	 * we can send. Changes when ibmtu changes.
+	 */
+	u32 ibmaxlen;
+	/*
+	 * ibmaxlen at init time, limited by chip and by receive buffer
+	 * size.  Not changed after init.
+	 */
+	u32 init_ibmaxlen;
+	/* LID programmed for this instance */
+	u16 lid;
+	/* list of pkeys programmed; 0 if not set */
+	u16 pkeys[4];
+	/* LID mask control */
+	u8 lmc;
+	u8 link_width_supported;
+	u8 link_speed_supported;
+	u8 link_width_enabled;
+	u8 link_speed_enabled;
+	u8 link_width_active;
+	u8 link_speed_active;
+	u8 vls_supported;
+	u8 vls_operational;
+	/* Rx Polarity inversion (compensate for ~tx on partner) */
+	u8 rx_pol_inv;
+
+	u8 hw_pidx;     /* physical port index */
+	u8 port;        /* IB port number and index into dd->pports - 1 */
+
+	u8 delay_mult;
+
+	/* used to override LED behavior */
+	u8 led_override;  /* Substituted for normal value, if non-zero */
+	u16 led_override_timeoff; /* delta to next timer event */
+	u8 led_override_vals[2]; /* Alternates per blink-frame */
+	u8 led_override_phase; /* Just counts, LSB picks from vals[] */
+	atomic_t led_override_timer_active;
+	/* Used to flash LEDs in override mode */
+	struct timer_list led_override_timer;
+	struct xmit_wait cong_stats;
+	struct timer_list symerr_clear_timer;
+
+	/* Synchronize access between driver writes and sysfs reads */
+	spinlock_t cc_shadow_lock
+		____cacheline_aligned_in_smp;
+
+	/* Shadow copy of the congestion control table */
+	struct cc_table_shadow *ccti_entries_shadow;
+
+	/* Shadow copy of the congestion control entries */
+	struct ib_cc_congestion_setting_attr_shadow *congestion_entries_shadow;
+
+	/* List of congestion control table entries */
+	struct ib_cc_table_entry_shadow *ccti_entries;
+
+	/* 16 congestion entries with each entry corresponding to a SL */
+	struct ib_cc_congestion_entry_shadow *congestion_entries;
+
+	/* Maximum number of congestion control entries that the agent expects
+	 * the manager to send.
+	 */
+	u16 cc_supported_table_entries;
+
+	/* Total number of congestion control table entries */
+	u16 total_cct_entry;
+
+	/* Bit map identifying service level */
+	u16 cc_sl_control_map;
+
+	/* maximum congestion control table index */
+	u16 ccti_limit;
+
+	/* CA's max number of 64 entry units in the congestion control table */
+	u8 cc_max_table_entries;
+};
+
+/* Observers. Not to be taken lightly, possibly not to ship. */
+/*
+ * If a diag read or write is to (bottom <= offset <= top),
+ * the "hoook" is called, allowing, e.g. shadows to be
+ * updated in sync with the driver. struct diag_observer
+ * is the "visible" part.
+ */
+struct diag_observer;
+
+typedef int (*diag_hook) (struct qib_devdata *dd,
+	const struct diag_observer *op,
+	u32 offs, u64 *data, u64 mask, int only_32);
+
+struct diag_observer {
+	diag_hook hook;
+	u32 bottom;
+	u32 top;
+};
+
+extern int qib_register_observer(struct qib_devdata *dd,
+	const struct diag_observer *op);
+
+/* Only declared here, not defined. Private to diags */
+struct diag_observer_list_elt;
+
+/* device data struct now contains only "general per-device" info.
+ * fields related to a physical IB port are in a qib_pportdata struct,
+ * described above) while fields only used by a particular chip-type are in
+ * a qib_chipdata struct, whose contents are opaque to this file.
+ */
+struct qib_devdata {
+	struct qib_ibdev verbs_dev;     /* must be first */
+	struct list_head list;
+	/* pointers to related structs for this device */
+	/* pci access data structure */
+	struct pci_dev *pcidev;
+	struct cdev *user_cdev;
+	struct cdev *diag_cdev;
+	struct device *user_device;
+	struct device *diag_device;
+
+	/* mem-mapped pointer to base of chip regs */
+	u64 __iomem *kregbase;
+	/* end of mem-mapped chip space excluding sendbuf and user regs */
+	u64 __iomem *kregend;
+	/* physical address of chip for io_remap, etc. */
+	resource_size_t physaddr;
+	/* qib_cfgctxts pointers */
+	struct qib_ctxtdata **rcd; /* Receive Context Data */
+
+	/* qib_pportdata, points to array of (physical) port-specific
+	 * data structs, indexed by pidx (0..n-1)
+	 */
+	struct qib_pportdata *pport;
+	struct qib_chip_specific *cspec; /* chip-specific */
+
+	/* kvirt address of 1st 2k pio buffer */
+	void __iomem *pio2kbase;
+	/* kvirt address of 1st 4k pio buffer */
+	void __iomem *pio4kbase;
+	/* mem-mapped pointer to base of PIO buffers (if using WC PAT) */
+	void __iomem *piobase;
+	/* mem-mapped pointer to base of user chip regs (if using WC PAT) */
+	u64 __iomem *userbase;
+	void __iomem *piovl15base; /* base of VL15 buffers, if not WC */
+	/*
+	 * points to area where PIOavail registers will be DMA'ed.
+	 * Has to be on a page of it's own, because the page will be
+	 * mapped into user program space.  This copy is *ONLY* ever
+	 * written by DMA, not by the driver!  Need a copy per device
+	 * when we get to multiple devices
+	 */
+	volatile __le64 *pioavailregs_dma; /* DMA'ed by chip */
+	/* physical address where updates occur */
+	dma_addr_t pioavailregs_phys;
+
+	/* device-specific implementations of functions needed by
+	 * common code. Contrary to previous consensus, we can't
+	 * really just point to a device-specific table, because we
+	 * may need to "bend", e.g. *_f_put_tid
+	 */
+	/* fallback to alternate interrupt type if possible */
+	int (*f_intr_fallback)(struct qib_devdata *);
+	/* hard reset chip */
+	int (*f_reset)(struct qib_devdata *);
+	void (*f_quiet_serdes)(struct qib_pportdata *);
+	int (*f_bringup_serdes)(struct qib_pportdata *);
+	int (*f_early_init)(struct qib_devdata *);
+	void (*f_clear_tids)(struct qib_devdata *, struct qib_ctxtdata *);
+	void (*f_put_tid)(struct qib_devdata *, u64 __iomem*,
+				u32, unsigned long);
+	void (*f_cleanup)(struct qib_devdata *);
+	void (*f_setextled)(struct qib_pportdata *, u32);
+	/* fill out chip-specific fields */
+	int (*f_get_base_info)(struct qib_ctxtdata *, struct qib_base_info *);
+	/* free irq */
+	void (*f_free_irq)(struct qib_devdata *);
+	struct qib_message_header *(*f_get_msgheader)
+					(struct qib_devdata *, __le32 *);
+	void (*f_config_ctxts)(struct qib_devdata *);
+	int (*f_get_ib_cfg)(struct qib_pportdata *, int);
+	int (*f_set_ib_cfg)(struct qib_pportdata *, int, u32);
+	int (*f_set_ib_loopback)(struct qib_pportdata *, const char *);
+	int (*f_get_ib_table)(struct qib_pportdata *, int, void *);
+	int (*f_set_ib_table)(struct qib_pportdata *, int, void *);
+	u32 (*f_iblink_state)(u64);
+	u8 (*f_ibphys_portstate)(u64);
+	void (*f_xgxs_reset)(struct qib_pportdata *);
+	/* per chip actions needed for IB Link up/down changes */
+	int (*f_ib_updown)(struct qib_pportdata *, int, u64);
+	u32 __iomem *(*f_getsendbuf)(struct qib_pportdata *, u64, u32 *);
+	/* Read/modify/write of GPIO pins (potentially chip-specific */
+	int (*f_gpio_mod)(struct qib_devdata *dd, u32 out, u32 dir,
+		u32 mask);
+	/* Enable writes to config EEPROM (if supported) */
+	int (*f_eeprom_wen)(struct qib_devdata *dd, int wen);
+	/*
+	 * modify rcvctrl shadow[s] and write to appropriate chip-regs.
+	 * see above QIB_RCVCTRL_xxx_ENB/DIS for operations.
+	 * (ctxt == -1) means "all contexts", only meaningful for
+	 * clearing. Could remove if chip_spec shutdown properly done.
+	 */
+	void (*f_rcvctrl)(struct qib_pportdata *, unsigned int op,
+		int ctxt);
+	/* Read/modify/write sendctrl appropriately for op and port. */
+	void (*f_sendctrl)(struct qib_pportdata *, u32 op);
+	void (*f_set_intr_state)(struct qib_devdata *, u32);
+	void (*f_set_armlaunch)(struct qib_devdata *, u32);
+	void (*f_wantpiobuf_intr)(struct qib_devdata *, u32);
+	int (*f_late_initreg)(struct qib_devdata *);
+	int (*f_init_sdma_regs)(struct qib_pportdata *);
+	u16 (*f_sdma_gethead)(struct qib_pportdata *);
+	int (*f_sdma_busy)(struct qib_pportdata *);
+	void (*f_sdma_update_tail)(struct qib_pportdata *, u16);
+	void (*f_sdma_set_desc_cnt)(struct qib_pportdata *, unsigned);
+	void (*f_sdma_sendctrl)(struct qib_pportdata *, unsigned);
+	void (*f_sdma_hw_clean_up)(struct qib_pportdata *);
+	void (*f_sdma_hw_start_up)(struct qib_pportdata *);
+	void (*f_sdma_init_early)(struct qib_pportdata *);
+	void (*f_set_cntr_sample)(struct qib_pportdata *, u32, u32);
+	void (*f_update_usrhead)(struct qib_ctxtdata *, u64, u32, u32, u32);
+	u32 (*f_hdrqempty)(struct qib_ctxtdata *);
+	u64 (*f_portcntr)(struct qib_pportdata *, u32);
+	u32 (*f_read_cntrs)(struct qib_devdata *, loff_t, char **,
+		u64 **);
+	u32 (*f_read_portcntrs)(struct qib_devdata *, loff_t, u32,
+		char **, u64 **);
+	u32 (*f_setpbc_control)(struct qib_pportdata *, u32, u8, u8);
+	void (*f_initvl15_bufs)(struct qib_devdata *);
+	void (*f_init_ctxt)(struct qib_ctxtdata *);
+	void (*f_txchk_change)(struct qib_devdata *, u32, u32, u32,
+		struct qib_ctxtdata *);
+	void (*f_writescratch)(struct qib_devdata *, u32);
+	int (*f_tempsense_rd)(struct qib_devdata *, int regnum);
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	int (*f_notify_dca)(struct qib_devdata *, unsigned long event);
+#endif
+
+	char *boardname; /* human readable board info */
+
+	/* template for writing TIDs  */
+	u64 tidtemplate;
+	/* value to write to free TIDs */
+	u64 tidinvalid;
+
+	/* number of registers used for pioavail */
+	u32 pioavregs;
+	/* device (not port) flags, basically device capabilities */
+	u32 flags;
+	/* last buffer for user use */
+	u32 lastctxt_piobuf;
+
+	/* reset value */
+	u64 z_int_counter;
+	/* percpu intcounter */
+	u64 __percpu *int_counter;
+
+	/* pio bufs allocated per ctxt */
+	u32 pbufsctxt;
+	/* if remainder on bufs/ctxt, ctxts < extrabuf get 1 extra */
+	u32 ctxts_extrabuf;
+	/*
+	 * number of ctxts configured as max; zero is set to number chip
+	 * supports, less gives more pio bufs/ctxt, etc.
+	 */
+	u32 cfgctxts;
+	/*
+	 * number of ctxts available for PSM open
+	 */
+	u32 freectxts;
+
+	/*
+	 * hint that we should update pioavailshadow before
+	 * looking for a PIO buffer
+	 */
+	u32 upd_pio_shadow;
+
+	/* internal debugging stats */
+	u32 maxpkts_call;
+	u32 avgpkts_call;
+	u64 nopiobufs;
+
+	/* PCI Vendor ID (here for NodeInfo) */
+	u16 vendorid;
+	/* PCI Device ID (here for NodeInfo) */
+	u16 deviceid;
+	/* for write combining settings */
+	unsigned long wc_cookie;
+	unsigned long wc_base;
+	unsigned long wc_len;
+
+	/* shadow copy of struct page *'s for exp tid pages */
+	struct page **pageshadow;
+	/* shadow copy of dma handles for exp tid pages */
+	dma_addr_t *physshadow;
+	u64 __iomem *egrtidbase;
+	spinlock_t sendctrl_lock; /* protect changes to sendctrl shadow */
+	/* around rcd and (user ctxts) ctxt_cnt use (intr vs free) */
+	spinlock_t uctxt_lock; /* rcd and user context changes */
+	/*
+	 * per unit status, see also portdata statusp
+	 * mapped readonly into user processes so they can get unit and
+	 * IB link status cheaply
+	 */
+	u64 *devstatusp;
+	char *freezemsg; /* freeze msg if hw error put chip in freeze */
+	u32 freezelen; /* max length of freezemsg */
+	/* timer used to prevent stats overflow, error throttling, etc. */
+	struct timer_list stats_timer;
+
+	/* timer to verify interrupts work, and fallback if possible */
+	struct timer_list intrchk_timer;
+	unsigned long ureg_align; /* user register alignment */
+
+	/*
+	 * Protects pioavailshadow, pioavailkernel, pio_need_disarm, and
+	 * pio_writing.
+	 */
+	spinlock_t pioavail_lock;
+	/*
+	 * index of last buffer to optimize search for next
+	 */
+	u32 last_pio;
+	/*
+	 * min kernel pio buffer to optimize search
+	 */
+	u32 min_kernel_pio;
+	/*
+	 * Shadow copies of registers; size indicates read access size.
+	 * Most of them are readonly, but some are write-only register,
+	 * where we manipulate the bits in the shadow copy, and then write
+	 * the shadow copy to qlogic_ib.
+	 *
+	 * We deliberately make most of these 32 bits, since they have
+	 * restricted range.  For any that we read, we won't to generate 32
+	 * bit accesses, since Opteron will generate 2 separate 32 bit HT
+	 * transactions for a 64 bit read, and we want to avoid unnecessary
+	 * bus transactions.
+	 */
+
+	/* This is the 64 bit group */
+
+	unsigned long pioavailshadow[6];
+	/* bitmap of send buffers available for the kernel to use with PIO. */
+	unsigned long pioavailkernel[6];
+	/* bitmap of send buffers which need to be disarmed. */
+	unsigned long pio_need_disarm[3];
+	/* bitmap of send buffers which are being written to. */
+	unsigned long pio_writing[3];
+	/* kr_revision shadow */
+	u64 revision;
+	/* Base GUID for device (from eeprom, network order) */
+	__be64 base_guid;
+
+	/*
+	 * kr_sendpiobufbase value (chip offset of pio buffers), and the
+	 * base of the 2KB buffer s(user processes only use 2K)
+	 */
+	u64 piobufbase;
+	u32 pio2k_bufbase;
+
+	/* these are the "32 bit" regs */
+
+	/* number of GUIDs in the flash for this interface */
+	u32 nguid;
+	/*
+	 * the following two are 32-bit bitmasks, but {test,clear,set}_bit
+	 * all expect bit fields to be "unsigned long"
+	 */
+	unsigned long rcvctrl; /* shadow per device rcvctrl */
+	unsigned long sendctrl; /* shadow per device sendctrl */
+
+	/* value we put in kr_rcvhdrcnt */
+	u32 rcvhdrcnt;
+	/* value we put in kr_rcvhdrsize */
+	u32 rcvhdrsize;
+	/* value we put in kr_rcvhdrentsize */
+	u32 rcvhdrentsize;
+	/* kr_ctxtcnt value */
+	u32 ctxtcnt;
+	/* kr_pagealign value */
+	u32 palign;
+	/* number of "2KB" PIO buffers */
+	u32 piobcnt2k;
+	/* size in bytes of "2KB" PIO buffers */
+	u32 piosize2k;
+	/* max usable size in dwords of a "2KB" PIO buffer before going "4KB" */
+	u32 piosize2kmax_dwords;
+	/* number of "4KB" PIO buffers */
+	u32 piobcnt4k;
+	/* size in bytes of "4KB" PIO buffers */
+	u32 piosize4k;
+	/* kr_rcvegrbase value */
+	u32 rcvegrbase;
+	/* kr_rcvtidbase value */
+	u32 rcvtidbase;
+	/* kr_rcvtidcnt value */
+	u32 rcvtidcnt;
+	/* kr_userregbase */
+	u32 uregbase;
+	/* shadow the control register contents */
+	u32 control;
+
+	/* chip address space used by 4k pio buffers */
+	u32 align4k;
+	/* size of each rcvegrbuffer */
+	u16 rcvegrbufsize;
+	/* log2 of above */
+	u16 rcvegrbufsize_shift;
+	/* localbus width (1, 2,4,8,16,32) from config space  */
+	u32 lbus_width;
+	/* localbus speed in MHz */
+	u32 lbus_speed;
+	int unit; /* unit # of this chip */
+
+	/* start of CHIP_SPEC move to chipspec, but need code changes */
+	/* low and high portions of MSI capability/vector */
+	u32 msi_lo;
+	/* saved after PCIe init for restore after reset */
+	u32 msi_hi;
+	/* MSI data (vector) saved for restore */
+	u16 msi_data;
+	/* so we can rewrite it after a chip reset */
+	u32 pcibar0;
+	/* so we can rewrite it after a chip reset */
+	u32 pcibar1;
+	u64 rhdrhead_intr_off;
+
+	/*
+	 * ASCII serial number, from flash, large enough for original
+	 * all digit strings, and longer QLogic serial number format
+	 */
+	u8 serial[16];
+	/* human readable board version */
+	u8 boardversion[96];
+	u8 lbus_info[32]; /* human readable localbus info */
+	/* chip major rev, from qib_revision */
+	u8 majrev;
+	/* chip minor rev, from qib_revision */
+	u8 minrev;
+
+	/* Misc small ints */
+	/* Number of physical ports available */
+	u8 num_pports;
+	/* Lowest context number which can be used by user processes */
+	u8 first_user_ctxt;
+	u8 n_krcv_queues;
+	u8 qpn_mask;
+	u8 skip_kctxt_mask;
+
+	u16 rhf_offset; /* offset of RHF within receive header entry */
+
+	/*
+	 * GPIO pins for twsi-connected devices, and device code for eeprom
+	 */
+	u8 gpio_sda_num;
+	u8 gpio_scl_num;
+	u8 twsi_eeprom_dev;
+	u8 board_atten;
+
+	/* Support (including locks) for EEPROM logging of errors and time */
+	/* control access to actual counters, timer */
+	spinlock_t eep_st_lock;
+	/* control high-level access to EEPROM */
+	struct mutex eep_lock;
+	uint64_t traffic_wds;
+	/* active time is kept in seconds, but logged in hours */
+	atomic_t active_time;
+	/* Below are nominal shadow of EEPROM, new since last EEPROM update */
+	uint8_t eep_st_errs[QIB_EEP_LOG_CNT];
+	uint8_t eep_st_new_errs[QIB_EEP_LOG_CNT];
+	uint16_t eep_hrs;
+	/*
+	 * masks for which bits of errs, hwerrs that cause
+	 * each of the counters to increment.
+	 */
+	struct qib_eep_log_mask eep_st_masks[QIB_EEP_LOG_CNT];
+	struct qib_diag_client *diag_client;
+	spinlock_t qib_diag_trans_lock; /* protect diag observer ops */
+	struct diag_observer_list_elt *diag_observer_list;
+
+	u8 psxmitwait_supported;
+	/* cycle length of PS* counters in HW (in picoseconds) */
+	u16 psxmitwait_check_rate;
+	/* high volume overflow errors defered to tasklet */
+	struct tasklet_struct error_tasklet;
+	/* per device cq worker */
+	struct kthread_worker *worker;
+
+	int assigned_node_id; /* NUMA node closest to HCA */
+};
+
+/* hol_state values */
+#define QIB_HOL_UP       0
+#define QIB_HOL_INIT     1
+
+#define QIB_SDMA_SENDCTRL_OP_ENABLE    (1U << 0)
+#define QIB_SDMA_SENDCTRL_OP_INTENABLE (1U << 1)
+#define QIB_SDMA_SENDCTRL_OP_HALT      (1U << 2)
+#define QIB_SDMA_SENDCTRL_OP_CLEANUP   (1U << 3)
+#define QIB_SDMA_SENDCTRL_OP_DRAIN     (1U << 4)
+
+/* operation types for f_txchk_change() */
+#define TXCHK_CHG_TYPE_DIS1  3
+#define TXCHK_CHG_TYPE_ENAB1 2
+#define TXCHK_CHG_TYPE_KERN  1
+#define TXCHK_CHG_TYPE_USER  0
+
+#define QIB_CHASE_TIME msecs_to_jiffies(145)
+#define QIB_CHASE_DIS_TIME msecs_to_jiffies(160)
+
+/* Private data for file operations */
+struct qib_filedata {
+	struct qib_ctxtdata *rcd;
+	unsigned subctxt;
+	unsigned tidcursor;
+	struct qib_user_sdma_queue *pq;
+	int rec_cpu_num; /* for cpu affinity; -1 if none */
+};
+
+extern struct list_head qib_dev_list;
+extern spinlock_t qib_devs_lock;
+extern struct qib_devdata *qib_lookup(int unit);
+extern u32 qib_cpulist_count;
+extern unsigned long *qib_cpulist;
+
+extern unsigned qib_wc_pat;
+extern unsigned qib_cc_table_size;
+int qib_init(struct qib_devdata *, int);
+int init_chip_wc_pat(struct qib_devdata *dd, u32);
+int qib_enable_wc(struct qib_devdata *dd);
+void qib_disable_wc(struct qib_devdata *dd);
+int qib_count_units(int *npresentp, int *nupp);
+int qib_count_active_units(void);
+
+int qib_cdev_init(int minor, const char *name,
+		  const struct file_operations *fops,
+		  struct cdev **cdevp, struct device **devp);
+void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp);
+int qib_dev_init(void);
+void qib_dev_cleanup(void);
+
+int qib_diag_add(struct qib_devdata *);
+void qib_diag_remove(struct qib_devdata *);
+void qib_handle_e_ibstatuschanged(struct qib_pportdata *, u64);
+void qib_sdma_update_tail(struct qib_pportdata *, u16); /* hold sdma_lock */
+
+int qib_decode_err(struct qib_devdata *dd, char *buf, size_t blen, u64 err);
+void qib_bad_intrstatus(struct qib_devdata *);
+void qib_handle_urcv(struct qib_devdata *, u64);
+
+/* clean up any per-chip chip-specific stuff */
+void qib_chip_cleanup(struct qib_devdata *);
+/* clean up any chip type-specific stuff */
+void qib_chip_done(void);
+
+/* check to see if we have to force ordering for write combining */
+int qib_unordered_wc(void);
+void qib_pio_copy(void __iomem *to, const void *from, size_t count);
+
+void qib_disarm_piobufs(struct qib_devdata *, unsigned, unsigned);
+int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *);
+void qib_disarm_piobufs_set(struct qib_devdata *, unsigned long *, unsigned);
+void qib_cancel_sends(struct qib_pportdata *);
+
+int qib_create_rcvhdrq(struct qib_devdata *, struct qib_ctxtdata *);
+int qib_setup_eagerbufs(struct qib_ctxtdata *);
+void qib_set_ctxtcnt(struct qib_devdata *);
+int qib_create_ctxts(struct qib_devdata *dd);
+struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *, u32, int);
+int qib_init_pportdata(struct qib_pportdata *, struct qib_devdata *, u8, u8);
+void qib_free_ctxtdata(struct qib_devdata *, struct qib_ctxtdata *);
+
+u32 qib_kreceive(struct qib_ctxtdata *, u32 *, u32 *);
+int qib_reset_device(int);
+int qib_wait_linkstate(struct qib_pportdata *, u32, int);
+int qib_set_linkstate(struct qib_pportdata *, u8);
+int qib_set_mtu(struct qib_pportdata *, u16);
+int qib_set_lid(struct qib_pportdata *, u32, u8);
+void qib_hol_down(struct qib_pportdata *);
+void qib_hol_init(struct qib_pportdata *);
+void qib_hol_up(struct qib_pportdata *);
+void qib_hol_event(unsigned long);
+void qib_disable_after_error(struct qib_devdata *);
+int qib_set_uevent_bits(struct qib_pportdata *, const int);
+
+/* for use in system calls, where we want to know device type, etc. */
+#define ctxt_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->rcd)
+#define subctxt_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->subctxt)
+#define tidcursor_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->tidcursor)
+#define user_sdma_queue_fp(fp) \
+	(((struct qib_filedata *)(fp)->private_data)->pq)
+
+static inline struct qib_devdata *dd_from_ppd(struct qib_pportdata *ppd)
+{
+	return ppd->dd;
+}
+
+static inline struct qib_devdata *dd_from_dev(struct qib_ibdev *dev)
+{
+	return container_of(dev, struct qib_devdata, verbs_dev);
+}
+
+static inline struct qib_devdata *dd_from_ibdev(struct ib_device *ibdev)
+{
+	return dd_from_dev(to_idev(ibdev));
+}
+
+static inline struct qib_pportdata *ppd_from_ibp(struct qib_ibport *ibp)
+{
+	return container_of(ibp, struct qib_pportdata, ibport_data);
+}
+
+static inline struct qib_ibport *to_iport(struct ib_device *ibdev, u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	WARN_ON(pidx >= dd->num_pports);
+	return &dd->pport[pidx].ibport_data;
+}
+
+/*
+ * values for dd->flags (_device_ related flags) and
+ */
+#define QIB_HAS_LINK_LATENCY  0x1 /* supports link latency (IB 1.2) */
+#define QIB_INITTED           0x2 /* chip and driver up and initted */
+#define QIB_DOING_RESET       0x4  /* in the middle of doing chip reset */
+#define QIB_PRESENT           0x8  /* chip accesses can be done */
+#define QIB_PIO_FLUSH_WC      0x10 /* Needs Write combining flush for PIO */
+#define QIB_HAS_THRESH_UPDATE 0x40
+#define QIB_HAS_SDMA_TIMEOUT  0x80
+#define QIB_USE_SPCL_TRIG     0x100 /* SpecialTrigger launch enabled */
+#define QIB_NODMA_RTAIL       0x200 /* rcvhdrtail register DMA enabled */
+#define QIB_HAS_INTX          0x800 /* Supports INTx interrupts */
+#define QIB_HAS_SEND_DMA      0x1000 /* Supports Send DMA */
+#define QIB_HAS_VLSUPP        0x2000 /* Supports multiple VLs; PBC different */
+#define QIB_HAS_HDRSUPP       0x4000 /* Supports header suppression */
+#define QIB_BADINTR           0x8000 /* severe interrupt problems */
+#define QIB_DCA_ENABLED       0x10000 /* Direct Cache Access enabled */
+#define QIB_HAS_QSFP          0x20000 /* device (card instance) has QSFP */
+
+/*
+ * values for ppd->lflags (_ib_port_ related flags)
+ */
+#define QIBL_LINKV             0x1 /* IB link state valid */
+#define QIBL_LINKDOWN          0x8 /* IB link is down */
+#define QIBL_LINKINIT          0x10 /* IB link level is up */
+#define QIBL_LINKARMED         0x20 /* IB link is ARMED */
+#define QIBL_LINKACTIVE        0x40 /* IB link is ACTIVE */
+/* leave a gap for more IB-link state */
+#define QIBL_IB_AUTONEG_INPROG 0x1000 /* non-IBTA DDR/QDR neg active */
+#define QIBL_IB_AUTONEG_FAILED 0x2000 /* non-IBTA DDR/QDR neg failed */
+#define QIBL_IB_LINK_DISABLED  0x4000 /* Linkdown-disable forced,
+				       * Do not try to bring up */
+#define QIBL_IB_FORCE_NOTIFY   0x8000 /* force notify on next ib change */
+
+/* IB dword length mask in PBC (lower 11 bits); same for all chips */
+#define QIB_PBC_LENGTH_MASK                     ((1 << 11) - 1)
+
+
+/* ctxt_flag bit offsets */
+		/* waiting for a packet to arrive */
+#define QIB_CTXT_WAITING_RCV   2
+		/* master has not finished initializing */
+#define QIB_CTXT_MASTER_UNINIT 4
+		/* waiting for an urgent packet to arrive */
+#define QIB_CTXT_WAITING_URG 5
+
+/* free up any allocated data at closes */
+void qib_free_data(struct qib_ctxtdata *dd);
+void qib_chg_pioavailkernel(struct qib_devdata *, unsigned, unsigned,
+			    u32, struct qib_ctxtdata *);
+struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *,
+					   const struct pci_device_id *);
+void qib_free_devdata(struct qib_devdata *);
+struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra);
+
+#define QIB_TWSI_NO_DEV 0xFF
+/* Below qib_twsi_ functions must be called with eep_lock held */
+int qib_twsi_reset(struct qib_devdata *dd);
+int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr, void *buffer,
+		    int len);
+int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
+		    const void *buffer, int len);
+void qib_get_eeprom_info(struct qib_devdata *);
+int qib_update_eeprom_log(struct qib_devdata *dd);
+void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr);
+void qib_dump_lookup_output_queue(struct qib_devdata *);
+void qib_force_pio_avail_update(struct qib_devdata *);
+void qib_clear_symerror_on_linkup(unsigned long opaque);
+
+/*
+ * Set LED override, only the two LSBs have "public" meaning, but
+ * any non-zero value substitutes them for the Link and LinkTrain
+ * LED states.
+ */
+#define QIB_LED_PHYS 1 /* Physical (linktraining) GREEN LED */
+#define QIB_LED_LOG 2  /* Logical (link) YELLOW LED */
+void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val);
+
+/* send dma routines */
+int qib_setup_sdma(struct qib_pportdata *);
+void qib_teardown_sdma(struct qib_pportdata *);
+void __qib_sdma_intr(struct qib_pportdata *);
+void qib_sdma_intr(struct qib_pportdata *);
+void qib_user_sdma_send_desc(struct qib_pportdata *dd,
+			struct list_head *pktlist);
+int qib_sdma_verbs_send(struct qib_pportdata *, struct qib_sge_state *,
+			u32, struct qib_verbs_txreq *);
+/* ppd->sdma_lock should be locked before calling this. */
+int qib_sdma_make_progress(struct qib_pportdata *dd);
+
+static inline int qib_sdma_empty(const struct qib_pportdata *ppd)
+{
+	return ppd->sdma_descq_added == ppd->sdma_descq_removed;
+}
+
+/* must be called under qib_sdma_lock */
+static inline u16 qib_sdma_descq_freecnt(const struct qib_pportdata *ppd)
+{
+	return ppd->sdma_descq_cnt -
+		(ppd->sdma_descq_added - ppd->sdma_descq_removed) - 1;
+}
+
+static inline int __qib_sdma_running(struct qib_pportdata *ppd)
+{
+	return ppd->sdma_state.current_state == qib_sdma_state_s99_running;
+}
+int qib_sdma_running(struct qib_pportdata *);
+void dump_sdma_state(struct qib_pportdata *ppd);
+void __qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
+void qib_sdma_process_event(struct qib_pportdata *, enum qib_sdma_events);
+
+/*
+ * number of words used for protocol header if not set by qib_userinit();
+ */
+#define QIB_DFLT_RCVHDRSIZE 9
+
+/*
+ * We need to be able to handle an IB header of at least 24 dwords.
+ * We need the rcvhdrq large enough to handle largest IB header, but
+ * still have room for a 2KB MTU standard IB packet.
+ * Additionally, some processor/memory controller combinations
+ * benefit quite strongly from having the DMA'ed data be cacheline
+ * aligned and a cacheline multiple, so we set the size to 32 dwords
+ * (2 64-byte primary cachelines for pretty much all processors of
+ * interest).  The alignment hurts nothing, other than using somewhat
+ * more memory.
+ */
+#define QIB_RCVHDR_ENTSIZE 32
+
+int qib_get_user_pages(unsigned long, size_t, struct page **);
+void qib_release_user_pages(struct page **, size_t);
+int qib_eeprom_read(struct qib_devdata *, u8, void *, int);
+int qib_eeprom_write(struct qib_devdata *, u8, const void *, int);
+u32 __iomem *qib_getsendbuf_range(struct qib_devdata *, u32 *, u32, u32);
+void qib_sendbuf_done(struct qib_devdata *, unsigned);
+
+static inline void qib_clear_rcvhdrtail(const struct qib_ctxtdata *rcd)
+{
+	*((u64 *) rcd->rcvhdrtail_kvaddr) = 0ULL;
+}
+
+static inline u32 qib_get_rcvhdrtail(const struct qib_ctxtdata *rcd)
+{
+	/*
+	 * volatile because it's a DMA target from the chip, routine is
+	 * inlined, and don't want register caching or reordering.
+	 */
+	return (u32) le64_to_cpu(
+		*((volatile __le64 *)rcd->rcvhdrtail_kvaddr)); /* DMA'ed */
+}
+
+static inline u32 qib_get_hdrqtail(const struct qib_ctxtdata *rcd)
+{
+	const struct qib_devdata *dd = rcd->dd;
+	u32 hdrqtail;
+
+	if (dd->flags & QIB_NODMA_RTAIL) {
+		__le32 *rhf_addr;
+		u32 seq;
+
+		rhf_addr = (__le32 *) rcd->rcvhdrq +
+			rcd->head + dd->rhf_offset;
+		seq = qib_hdrget_seq(rhf_addr);
+		hdrqtail = rcd->head;
+		if (seq == rcd->seq_cnt)
+			hdrqtail++;
+	} else
+		hdrqtail = qib_get_rcvhdrtail(rcd);
+
+	return hdrqtail;
+}
+
+/*
+ * sysfs interface.
+ */
+
+extern const char ib_qib_version[];
+
+int qib_device_create(struct qib_devdata *);
+void qib_device_remove(struct qib_devdata *);
+
+int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
+			  struct kobject *kobj);
+int qib_verbs_register_sysfs(struct qib_devdata *);
+void qib_verbs_unregister_sysfs(struct qib_devdata *);
+/* Hook for sysfs read of QSFP */
+extern int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len);
+
+int __init qib_init_qibfs(void);
+int __exit qib_exit_qibfs(void);
+
+int qibfs_add(struct qib_devdata *);
+int qibfs_remove(struct qib_devdata *);
+
+int qib_pcie_init(struct pci_dev *, const struct pci_device_id *);
+int qib_pcie_ddinit(struct qib_devdata *, struct pci_dev *,
+		    const struct pci_device_id *);
+void qib_pcie_ddcleanup(struct qib_devdata *);
+int qib_pcie_params(struct qib_devdata *, u32, u32 *, struct qib_msix_entry *);
+int qib_reinit_intr(struct qib_devdata *);
+void qib_enable_intx(struct pci_dev *);
+void qib_nomsi(struct qib_devdata *);
+void qib_nomsix(struct qib_devdata *);
+void qib_pcie_getcmd(struct qib_devdata *, u16 *, u8 *, u8 *);
+void qib_pcie_reenable(struct qib_devdata *, u16, u8, u8);
+/* interrupts for device */
+u64 qib_int_counter(struct qib_devdata *);
+/* interrupt for all devices */
+u64 qib_sps_ints(void);
+
+/*
+ * dma_addr wrappers - all 0's invalid for hw
+ */
+dma_addr_t qib_map_page(struct pci_dev *, struct page *, unsigned long,
+			  size_t, int);
+const char *qib_get_unit_name(int unit);
+
+/*
+ * Flush write combining store buffers (if present) and perform a write
+ * barrier.
+ */
+#if defined(CONFIG_X86_64)
+#define qib_flush_wc() asm volatile("sfence" : : : "memory")
+#else
+#define qib_flush_wc() wmb() /* no reorder around wc flush */
+#endif
+
+/* global module parameter variables */
+extern unsigned qib_ibmtu;
+extern ushort qib_cfgctxts;
+extern ushort qib_num_cfg_vls;
+extern ushort qib_mini_init; /* If set, do few (ideally 0) writes to chip */
+extern unsigned qib_n_krcv_queues;
+extern unsigned qib_sdma_fetch_arb;
+extern unsigned qib_compat_ddr_negotiate;
+extern int qib_special_trigger;
+extern unsigned qib_numa_aware;
+
+extern struct mutex qib_mutex;
+
+/* Number of seconds before our card status check...  */
+#define STATUS_TIMEOUT 60
+
+#define QIB_DRV_NAME            "ib_qib"
+#define QIB_USER_MINOR_BASE     0
+#define QIB_TRACE_MINOR         127
+#define QIB_DIAGPKT_MINOR       128
+#define QIB_DIAG_MINOR_BASE     129
+#define QIB_NMINORS             255
+
+#define PCI_VENDOR_ID_PATHSCALE 0x1fc1
+#define PCI_VENDOR_ID_QLOGIC 0x1077
+#define PCI_DEVICE_ID_QLOGIC_IB_6120 0x10
+#define PCI_DEVICE_ID_QLOGIC_IB_7220 0x7220
+#define PCI_DEVICE_ID_QLOGIC_IB_7322 0x7322
+
+/*
+ * qib_early_err is used (only!) to print early errors before devdata is
+ * allocated, or when dd->pcidev may not be valid, and at the tail end of
+ * cleanup when devdata may have been freed, etc.  qib_dev_porterr is
+ * the same as qib_dev_err, but is used when the message really needs
+ * the IB port# to be definitive as to what's happening..
+ * All of these go to the trace log, and the trace log entry is done
+ * first to avoid possible serial port delays from printk.
+ */
+#define qib_early_err(dev, fmt, ...) \
+	dev_err(dev, fmt, ##__VA_ARGS__)
+
+#define qib_dev_err(dd, fmt, ...) \
+	dev_err(&(dd)->pcidev->dev, "%s: " fmt, \
+		qib_get_unit_name((dd)->unit), ##__VA_ARGS__)
+
+#define qib_dev_warn(dd, fmt, ...) \
+	dev_warn(&(dd)->pcidev->dev, "%s: " fmt, \
+		qib_get_unit_name((dd)->unit), ##__VA_ARGS__)
+
+#define qib_dev_porterr(dd, port, fmt, ...) \
+	dev_err(&(dd)->pcidev->dev, "%s: IB%u:%u " fmt, \
+		qib_get_unit_name((dd)->unit), (dd)->unit, (port), \
+		##__VA_ARGS__)
+
+#define qib_devinfo(pcidev, fmt, ...) \
+	dev_info(&(pcidev)->dev, fmt, ##__VA_ARGS__)
+
+/*
+ * this is used for formatting hw error messages...
+ */
+struct qib_hwerror_msgs {
+	u64 mask;
+	const char *msg;
+	size_t sz;
+};
+
+#define QLOGIC_IB_HWE_MSG(a, b) { .mask = a, .msg = b }
+
+/* in qib_intr.c... */
+void qib_format_hwerrors(u64 hwerrs,
+			 const struct qib_hwerror_msgs *hwerrmsgs,
+			 size_t nhwerrmsgs, char *msg, size_t lmsg);
+#endif                          /* _QIB_KERNEL_H */
diff --git a/drivers/infiniband/hw/qib/qib_6120_regs.h b/drivers/infiniband/hw/qib/qib_6120_regs.h
new file mode 100644
index 00000000000..e16cb6f7de2
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_6120_regs.h
@@ -0,0 +1,977 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_6120_Revision_OFFS 0x0
+#define QIB_6120_Revision_R_Simulator_LSB 0x3F
+#define QIB_6120_Revision_R_Simulator_RMASK 0x1
+#define QIB_6120_Revision_Reserved_LSB 0x28
+#define QIB_6120_Revision_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_Revision_BoardID_LSB 0x20
+#define QIB_6120_Revision_BoardID_RMASK 0xFF
+#define QIB_6120_Revision_R_SW_LSB 0x18
+#define QIB_6120_Revision_R_SW_RMASK 0xFF
+#define QIB_6120_Revision_R_Arch_LSB 0x10
+#define QIB_6120_Revision_R_Arch_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_6120_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_6120_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_6120_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_6120_Control_OFFS 0x8
+#define QIB_6120_Control_TxLatency_LSB 0x4
+#define QIB_6120_Control_TxLatency_RMASK 0x1
+#define QIB_6120_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_6120_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_6120_Control_LinkEn_LSB 0x2
+#define QIB_6120_Control_LinkEn_RMASK 0x1
+#define QIB_6120_Control_FreezeMode_LSB 0x1
+#define QIB_6120_Control_FreezeMode_RMASK 0x1
+#define QIB_6120_Control_SyncReset_LSB 0x0
+#define QIB_6120_Control_SyncReset_RMASK 0x1
+
+#define QIB_6120_PageAlign_OFFS 0x10
+
+#define QIB_6120_PortCnt_OFFS 0x18
+
+#define QIB_6120_SendRegBase_OFFS 0x30
+
+#define QIB_6120_UserRegBase_OFFS 0x38
+
+#define QIB_6120_CntrRegBase_OFFS 0x40
+
+#define QIB_6120_Scratch_OFFS 0x48
+#define QIB_6120_Scratch_TopHalf_LSB 0x20
+#define QIB_6120_Scratch_TopHalf_RMASK 0xFFFFFFFF
+#define QIB_6120_Scratch_BottomHalf_LSB 0x0
+#define QIB_6120_Scratch_BottomHalf_RMASK 0xFFFFFFFF
+
+#define QIB_6120_IntBlocked_OFFS 0x60
+#define QIB_6120_IntBlocked_ErrorIntBlocked_LSB 0x1F
+#define QIB_6120_IntBlocked_ErrorIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioSetIntBlocked_LSB 0x1E
+#define QIB_6120_IntBlocked_PioSetIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_LSB 0x1D
+#define QIB_6120_IntBlocked_PioBufAvailIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_LSB 0x1C
+#define QIB_6120_IntBlocked_assertGPIOIntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved_LSB 0xF
+#define QIB_6120_IntBlocked_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_LSB 0x10
+#define QIB_6120_IntBlocked_RcvAvail4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_LSB 0xF
+#define QIB_6120_IntBlocked_RcvAvail3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_LSB 0xE
+#define QIB_6120_IntBlocked_RcvAvail2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_LSB 0xD
+#define QIB_6120_IntBlocked_RcvAvail1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_LSB 0xC
+#define QIB_6120_IntBlocked_RcvAvail0IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_Reserved1_LSB 0x5
+#define QIB_6120_IntBlocked_Reserved1_RMASK 0x7F
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_LSB 0x4
+#define QIB_6120_IntBlocked_RcvUrg4IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_LSB 0x3
+#define QIB_6120_IntBlocked_RcvUrg3IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_LSB 0x2
+#define QIB_6120_IntBlocked_RcvUrg2IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_LSB 0x1
+#define QIB_6120_IntBlocked_RcvUrg1IntBlocked_RMASK 0x1
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_LSB 0x0
+#define QIB_6120_IntBlocked_RcvUrg0IntBlocked_RMASK 0x1
+
+#define QIB_6120_IntMask_OFFS 0x68
+#define QIB_6120_IntMask_ErrorIntMask_LSB 0x1F
+#define QIB_6120_IntMask_ErrorIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioSetIntMask_LSB 0x1E
+#define QIB_6120_IntMask_PioSetIntMask_RMASK 0x1
+#define QIB_6120_IntMask_PioBufAvailIntMask_LSB 0x1D
+#define QIB_6120_IntMask_PioBufAvailIntMask_RMASK 0x1
+#define QIB_6120_IntMask_assertGPIOIntMask_LSB 0x1C
+#define QIB_6120_IntMask_assertGPIOIntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved_LSB 0x11
+#define QIB_6120_IntMask_Reserved_RMASK 0x7FF
+#define QIB_6120_IntMask_RcvAvail4IntMask_LSB 0x10
+#define QIB_6120_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail3IntMask_LSB 0xF
+#define QIB_6120_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail2IntMask_LSB 0xE
+#define QIB_6120_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail1IntMask_LSB 0xD
+#define QIB_6120_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvAvail0IntMask_LSB 0xC
+#define QIB_6120_IntMask_RcvAvail0IntMask_RMASK 0x1
+#define QIB_6120_IntMask_Reserved1_LSB 0x5
+#define QIB_6120_IntMask_Reserved1_RMASK 0x7F
+#define QIB_6120_IntMask_RcvUrg4IntMask_LSB 0x4
+#define QIB_6120_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg3IntMask_LSB 0x3
+#define QIB_6120_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg2IntMask_LSB 0x2
+#define QIB_6120_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_LSB 0x1
+#define QIB_6120_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_6120_IntMask_RcvUrg0IntMask_LSB 0x0
+#define QIB_6120_IntMask_RcvUrg0IntMask_RMASK 0x1
+
+#define QIB_6120_IntStatus_OFFS 0x70
+#define QIB_6120_IntStatus_Error_LSB 0x1F
+#define QIB_6120_IntStatus_Error_RMASK 0x1
+#define QIB_6120_IntStatus_PioSent_LSB 0x1E
+#define QIB_6120_IntStatus_PioSent_RMASK 0x1
+#define QIB_6120_IntStatus_PioBufAvail_LSB 0x1D
+#define QIB_6120_IntStatus_PioBufAvail_RMASK 0x1
+#define QIB_6120_IntStatus_assertGPIO_LSB 0x1C
+#define QIB_6120_IntStatus_assertGPIO_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved_LSB 0xF
+#define QIB_6120_IntStatus_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntStatus_RcvAvail4_LSB 0x10
+#define QIB_6120_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail3_LSB 0xF
+#define QIB_6120_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail2_LSB 0xE
+#define QIB_6120_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail1_LSB 0xD
+#define QIB_6120_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvAvail0_LSB 0xC
+#define QIB_6120_IntStatus_RcvAvail0_RMASK 0x1
+#define QIB_6120_IntStatus_Reserved1_LSB 0x5
+#define QIB_6120_IntStatus_Reserved1_RMASK 0x7F
+#define QIB_6120_IntStatus_RcvUrg4_LSB 0x4
+#define QIB_6120_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg3_LSB 0x3
+#define QIB_6120_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg2_LSB 0x2
+#define QIB_6120_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg1_LSB 0x1
+#define QIB_6120_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_6120_IntStatus_RcvUrg0_LSB 0x0
+#define QIB_6120_IntStatus_RcvUrg0_RMASK 0x1
+
+#define QIB_6120_IntClear_OFFS 0x78
+#define QIB_6120_IntClear_ErrorIntClear_LSB 0x1F
+#define QIB_6120_IntClear_ErrorIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioSetIntClear_LSB 0x1E
+#define QIB_6120_IntClear_PioSetIntClear_RMASK 0x1
+#define QIB_6120_IntClear_PioBufAvailIntClear_LSB 0x1D
+#define QIB_6120_IntClear_PioBufAvailIntClear_RMASK 0x1
+#define QIB_6120_IntClear_assertGPIOIntClear_LSB 0x1C
+#define QIB_6120_IntClear_assertGPIOIntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved_LSB 0xF
+#define QIB_6120_IntClear_Reserved_RMASK 0x1FFF
+#define QIB_6120_IntClear_RcvAvail4IntClear_LSB 0x10
+#define QIB_6120_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail3IntClear_LSB 0xF
+#define QIB_6120_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail2IntClear_LSB 0xE
+#define QIB_6120_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail1IntClear_LSB 0xD
+#define QIB_6120_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvAvail0IntClear_LSB 0xC
+#define QIB_6120_IntClear_RcvAvail0IntClear_RMASK 0x1
+#define QIB_6120_IntClear_Reserved1_LSB 0x5
+#define QIB_6120_IntClear_Reserved1_RMASK 0x7F
+#define QIB_6120_IntClear_RcvUrg4IntClear_LSB 0x4
+#define QIB_6120_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg3IntClear_LSB 0x3
+#define QIB_6120_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg2IntClear_LSB 0x2
+#define QIB_6120_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_LSB 0x1
+#define QIB_6120_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_6120_IntClear_RcvUrg0IntClear_LSB 0x0
+#define QIB_6120_IntClear_RcvUrg0IntClear_RMASK 0x1
+
+#define QIB_6120_ErrMask_OFFS 0x80
+#define QIB_6120_ErrMask_Reserved_LSB 0x34
+#define QIB_6120_ErrMask_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrMask_HardwareErrMask_LSB 0x33
+#define QIB_6120_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_ResetNegatedMask_LSB 0x32
+#define QIB_6120_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_6120_ErrMask_InvalidAddrErrMask_LSB 0x31
+#define QIB_6120_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_IBStatusChangedMask_LSB 0x30
+#define QIB_6120_ErrMask_IBStatusChangedMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved1_LSB 0x26
+#define QIB_6120_ErrMask_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_6120_ErrMask_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_6120_ErrMask_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_LSB 0x23
+#define QIB_6120_ErrMask_SendPioArmLaunchErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_6120_ErrMask_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_6120_ErrMask_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendPktLenErrMask_LSB 0x20
+#define QIB_6120_ErrMask_SendPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendUnderRunErrMask_LSB 0x1F
+#define QIB_6120_ErrMask_SendUnderRunErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_6120_ErrMask_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_6120_ErrMask_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_Reserved2_LSB 0x12
+#define QIB_6120_ErrMask_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_6120_ErrMask_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrErrMask_LSB 0x10
+#define QIB_6120_ErrMask_RcvHdrErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_LSB 0xF
+#define QIB_6120_ErrMask_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadTidErrMask_LSB 0xE
+#define QIB_6120_ErrMask_RcvBadTidErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_6120_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_6120_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_LSB 0xB
+#define QIB_6120_ErrMask_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_LSB 0xA
+#define QIB_6120_ErrMask_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvEBPErrMask_LSB 0x9
+#define QIB_6120_ErrMask_RcvEBPErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_6120_ErrMask_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_6120_ErrMask_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_6120_ErrMask_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_6120_ErrMask_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_6120_ErrMask_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_6120_ErrMask_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvICRCErrMask_LSB 0x2
+#define QIB_6120_ErrMask_RcvICRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_LSB 0x1
+#define QIB_6120_ErrMask_RcvVCRCErrMask_RMASK 0x1
+#define QIB_6120_ErrMask_RcvFormatErrMask_LSB 0x0
+#define QIB_6120_ErrMask_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_6120_ErrStatus_OFFS 0x88
+#define QIB_6120_ErrStatus_Reserved_LSB 0x34
+#define QIB_6120_ErrStatus_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrStatus_HardwareErr_LSB 0x33
+#define QIB_6120_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_6120_ErrStatus_ResetNegated_LSB 0x32
+#define QIB_6120_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_6120_ErrStatus_InvalidAddrErr_LSB 0x31
+#define QIB_6120_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_IBStatusChanged_LSB 0x30
+#define QIB_6120_ErrStatus_IBStatusChanged_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved1_LSB 0x26
+#define QIB_6120_ErrStatus_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_LSB 0x25
+#define QIB_6120_ErrStatus_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_6120_ErrStatus_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_LSB 0x23
+#define QIB_6120_ErrStatus_SendPioArmLaunchErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_LSB 0x22
+#define QIB_6120_ErrStatus_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_6120_ErrStatus_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendPktLenErr_LSB 0x20
+#define QIB_6120_ErrStatus_SendPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendUnderRunErr_LSB 0x1F
+#define QIB_6120_ErrStatus_SendUnderRunErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_LSB 0x1E
+#define QIB_6120_ErrStatus_SendMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_SendMinPktLenErr_LSB 0x1D
+#define QIB_6120_ErrStatus_SendMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_Reserved2_LSB 0x12
+#define QIB_6120_ErrStatus_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_LSB 0x11
+#define QIB_6120_ErrStatus_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrErr_LSB 0x10
+#define QIB_6120_ErrStatus_RcvHdrErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrLenErr_LSB 0xF
+#define QIB_6120_ErrStatus_RcvHdrLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadTidErr_LSB 0xE
+#define QIB_6120_ErrStatus_RcvBadTidErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_6120_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_6120_ErrStatus_RcvEgrFullErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvBadVersionErr_LSB 0xB
+#define QIB_6120_ErrStatus_RcvBadVersionErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvIBFlowErr_LSB 0xA
+#define QIB_6120_ErrStatus_RcvIBFlowErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvEBPErr_LSB 0x9
+#define QIB_6120_ErrStatus_RcvEBPErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_6120_ErrStatus_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_6120_ErrStatus_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_LSB 0x6
+#define QIB_6120_ErrStatus_RcvShortPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_LSB 0x5
+#define QIB_6120_ErrStatus_RcvLongPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_LSB 0x4
+#define QIB_6120_ErrStatus_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_LSB 0x3
+#define QIB_6120_ErrStatus_RcvMinPktLenErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvICRCErr_LSB 0x2
+#define QIB_6120_ErrStatus_RcvICRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_LSB 0x1
+#define QIB_6120_ErrStatus_RcvVCRCErr_RMASK 0x1
+#define QIB_6120_ErrStatus_RcvFormatErr_LSB 0x0
+#define QIB_6120_ErrStatus_RcvFormatErr_RMASK 0x1
+
+#define QIB_6120_ErrClear_OFFS 0x90
+#define QIB_6120_ErrClear_Reserved_LSB 0x34
+#define QIB_6120_ErrClear_Reserved_RMASK 0xFFF
+#define QIB_6120_ErrClear_HardwareErrClear_LSB 0x33
+#define QIB_6120_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_ResetNegatedClear_LSB 0x32
+#define QIB_6120_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_6120_ErrClear_InvalidAddrErrClear_LSB 0x31
+#define QIB_6120_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_IBStatusChangedClear_LSB 0x30
+#define QIB_6120_ErrClear_IBStatusChangedClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved1_LSB 0x26
+#define QIB_6120_ErrClear_Reserved1_RMASK 0x3FF
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_6120_ErrClear_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_6120_ErrClear_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_LSB 0x23
+#define QIB_6120_ErrClear_SendPioArmLaunchErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_6120_ErrClear_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_6120_ErrClear_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendPktLenErrClear_LSB 0x20
+#define QIB_6120_ErrClear_SendPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendUnderRunErrClear_LSB 0x1F
+#define QIB_6120_ErrClear_SendUnderRunErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_6120_ErrClear_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_6120_ErrClear_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_Reserved2_LSB 0x12
+#define QIB_6120_ErrClear_Reserved2_RMASK 0x7FF
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_6120_ErrClear_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrErrClear_LSB 0x10
+#define QIB_6120_ErrClear_RcvHdrErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_LSB 0xF
+#define QIB_6120_ErrClear_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadTidErrClear_LSB 0xE
+#define QIB_6120_ErrClear_RcvBadTidErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_6120_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_6120_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_LSB 0xB
+#define QIB_6120_ErrClear_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_LSB 0xA
+#define QIB_6120_ErrClear_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvEBPErrClear_LSB 0x9
+#define QIB_6120_ErrClear_RcvEBPErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_6120_ErrClear_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_6120_ErrClear_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_6120_ErrClear_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_6120_ErrClear_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_6120_ErrClear_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_6120_ErrClear_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvICRCErrClear_LSB 0x2
+#define QIB_6120_ErrClear_RcvICRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_LSB 0x1
+#define QIB_6120_ErrClear_RcvVCRCErrClear_RMASK 0x1
+#define QIB_6120_ErrClear_RcvFormatErrClear_LSB 0x0
+#define QIB_6120_ErrClear_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_6120_HwErrMask_OFFS 0x98
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_LSB 0x3F
+#define QIB_6120_HwErrMask_IBCBusFromSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_LSB 0x3E
+#define QIB_6120_HwErrMask_IBCBusToSPCParityErrMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved_LSB 0x3D
+#define QIB_6120_HwErrMask_Reserved_RMASK 0x1
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_LSB 0x3C
+#define QIB_6120_HwErrMask_IBSerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_LSB 0x3B
+#define QIB_6120_HwErrMask_PCIESerdesQ0PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_LSB 0x3A
+#define QIB_6120_HwErrMask_PCIESerdesQ1PClkNotDetectMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved1_LSB 0x39
+#define QIB_6120_HwErrMask_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_LSB 0x38
+#define QIB_6120_HwErrMask_IBPLLrfSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_LSB 0x37
+#define QIB_6120_HwErrMask_IBPLLfbSlipMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_6120_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved2_LSB 0x33
+#define QIB_6120_HwErrMask_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_LSB 0x2C
+#define QIB_6120_HwErrMask_RXEMemParityErrMask_RMASK 0x7F
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_LSB 0x28
+#define QIB_6120_HwErrMask_TXEMemParityErrMask_RMASK 0xF
+#define QIB_6120_HwErrMask_Reserved3_LSB 0x22
+#define QIB_6120_HwErrMask_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_6120_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_6120_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_6120_HwErrMask_PoisonedTLPMask_LSB 0x1D
+#define QIB_6120_HwErrMask_PoisonedTLPMask_RMASK 0x1
+#define QIB_6120_HwErrMask_Reserved4_LSB 0x6
+#define QIB_6120_HwErrMask_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_LSB 0x0
+#define QIB_6120_HwErrMask_PCIeMemParityErrMask_RMASK 0x3F
+
+#define QIB_6120_HwErrStatus_OFFS 0xA0
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwErrStatus_IBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwErrStatus_IBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved_LSB 0x3D
+#define QIB_6120_HwErrStatus_Reserved_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_LSB 0x3C
+#define QIB_6120_HwErrStatus_IBSerdesPClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_LSB 0x3B
+#define QIB_6120_HwErrStatus_PCIESerdesQ0PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_LSB 0x3A
+#define QIB_6120_HwErrStatus_PCIESerdesQ1PClkNotDetect_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved1_LSB 0x39
+#define QIB_6120_HwErrStatus_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_LSB 0x38
+#define QIB_6120_HwErrStatus_IBPLLrfSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_LSB 0x37
+#define QIB_6120_HwErrStatus_IBPLLfbSlip_RMASK 0x1
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_6120_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved2_LSB 0x33
+#define QIB_6120_HwErrStatus_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrStatus_RXEMemParity_LSB 0x2C
+#define QIB_6120_HwErrStatus_RXEMemParity_RMASK 0x7F
+#define QIB_6120_HwErrStatus_TXEMemParity_LSB 0x28
+#define QIB_6120_HwErrStatus_TXEMemParity_RMASK 0xF
+#define QIB_6120_HwErrStatus_Reserved3_LSB 0x22
+#define QIB_6120_HwErrStatus_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_6120_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_6120_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_6120_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_6120_HwErrStatus_PoisenedTLP_LSB 0x1D
+#define QIB_6120_HwErrStatus_PoisenedTLP_RMASK 0x1
+#define QIB_6120_HwErrStatus_Reserved4_LSB 0x6
+#define QIB_6120_HwErrStatus_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrStatus_PCIeMemParity_LSB 0x0
+#define QIB_6120_HwErrStatus_PCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_HwErrClear_OFFS 0xA8
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_LSB 0x3F
+#define QIB_6120_HwErrClear_IBCBusFromSPCParityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_LSB 0x3E
+#define QIB_6120_HwErrClear_IBCBusToSPCparityErrClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved_LSB 0x3D
+#define QIB_6120_HwErrClear_Reserved_RMASK 0x1
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_LSB 0x3C
+#define QIB_6120_HwErrClear_IBSerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_LSB 0x3B
+#define QIB_6120_HwErrClear_PCIESerdesQ0PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_LSB 0x3A
+#define QIB_6120_HwErrClear_PCIESerdesQ1PClkNotDetectClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved1_LSB 0x39
+#define QIB_6120_HwErrClear_Reserved1_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_LSB 0x38
+#define QIB_6120_HwErrClear_IBPLLrfSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_LSB 0x37
+#define QIB_6120_HwErrClear_IBPLLfbSlipClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_6120_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved2_LSB 0x33
+#define QIB_6120_HwErrClear_Reserved2_RMASK 0x7
+#define QIB_6120_HwErrClear_RXEMemParityClear_LSB 0x2C
+#define QIB_6120_HwErrClear_RXEMemParityClear_RMASK 0x7F
+#define QIB_6120_HwErrClear_TXEMemParityClear_LSB 0x28
+#define QIB_6120_HwErrClear_TXEMemParityClear_RMASK 0xF
+#define QIB_6120_HwErrClear_Reserved3_LSB 0x22
+#define QIB_6120_HwErrClear_Reserved3_RMASK 0x3F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_LSB 0x1F
+#define QIB_6120_HwErrClear_PCIeBusParityClr_RMASK 0x7
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_6120_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_6120_HwErrClear_PoisonedTLPClear_LSB 0x1D
+#define QIB_6120_HwErrClear_PoisonedTLPClear_RMASK 0x1
+#define QIB_6120_HwErrClear_Reserved4_LSB 0x6
+#define QIB_6120_HwErrClear_Reserved4_RMASK 0x7FFFFF
+#define QIB_6120_HwErrClear_PCIeMemParityClr_LSB 0x0
+#define QIB_6120_HwErrClear_PCIeMemParityClr_RMASK 0x3F
+
+#define QIB_6120_HwDiagCtrl_OFFS 0xB0
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_6120_HwDiagCtrl_ForceIBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_LSB 0x3E
+#define QIB_6120_HwDiagCtrl_ForceIBCBusToSPCParityErr_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_6120_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_6120_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_6120_HwDiagCtrl_Reserved_LSB 0x33
+#define QIB_6120_HwDiagCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_LSB 0x2C
+#define QIB_6120_HwDiagCtrl_ForceRxMemParityErr_RMASK 0x7F
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_LSB 0x28
+#define QIB_6120_HwDiagCtrl_ForceTxMemparityErr_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved1_LSB 0x23
+#define QIB_6120_HwDiagCtrl_Reserved1_RMASK 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_6120_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_6120_HwDiagCtrl_Reserved2_LSB 0x6
+#define QIB_6120_HwDiagCtrl_Reserved2_RMASK 0x1FFFFFF
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_LSB 0x0
+#define QIB_6120_HwDiagCtrl_forcePCIeMemParity_RMASK 0x3F
+
+#define QIB_6120_IBCStatus_OFFS 0xC0
+#define QIB_6120_IBCStatus_TxCreditOk_LSB 0x1F
+#define QIB_6120_IBCStatus_TxCreditOk_RMASK 0x1
+#define QIB_6120_IBCStatus_TxReady_LSB 0x1E
+#define QIB_6120_IBCStatus_TxReady_RMASK 0x1
+#define QIB_6120_IBCStatus_Reserved_LSB 0x7
+#define QIB_6120_IBCStatus_Reserved_RMASK 0x7FFFFF
+#define QIB_6120_IBCStatus_LinkState_LSB 0x4
+#define QIB_6120_IBCStatus_LinkState_RMASK 0x7
+#define QIB_6120_IBCStatus_LinkTrainingState_LSB 0x0
+#define QIB_6120_IBCStatus_LinkTrainingState_RMASK 0xF
+
+#define QIB_6120_IBCCtrl_OFFS 0xC8
+#define QIB_6120_IBCCtrl_Loopback_LSB 0x3F
+#define QIB_6120_IBCCtrl_Loopback_RMASK 0x1
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_LSB 0x3E
+#define QIB_6120_IBCCtrl_LinkDownDefaultState_RMASK 0x1
+#define QIB_6120_IBCCtrl_Reserved_LSB 0x2B
+#define QIB_6120_IBCCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_6120_IBCCtrl_CreditScale_LSB 0x28
+#define QIB_6120_IBCCtrl_CreditScale_RMASK 0x7
+#define QIB_6120_IBCCtrl_OverrunThreshold_LSB 0x24
+#define QIB_6120_IBCCtrl_OverrunThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_PhyerrThreshold_LSB 0x20
+#define QIB_6120_IBCCtrl_PhyerrThreshold_RMASK 0xF
+#define QIB_6120_IBCCtrl_Reserved1_LSB 0x1F
+#define QIB_6120_IBCCtrl_Reserved1_RMASK 0x1
+#define QIB_6120_IBCCtrl_MaxPktLen_LSB 0x14
+#define QIB_6120_IBCCtrl_MaxPktLen_RMASK 0x7FF
+#define QIB_6120_IBCCtrl_LinkCmd_LSB 0x12
+#define QIB_6120_IBCCtrl_LinkCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_LinkInitCmd_LSB 0x10
+#define QIB_6120_IBCCtrl_LinkInitCmd_RMASK 0x3
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_LSB 0x8
+#define QIB_6120_IBCCtrl_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_LSB 0x0
+#define QIB_6120_IBCCtrl_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_6120_EXTStatus_OFFS 0xD0
+#define QIB_6120_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_6120_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved_LSB 0x20
+#define QIB_6120_EXTStatus_Reserved_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_Reserved1_LSB 0x10
+#define QIB_6120_EXTStatus_Reserved1_RMASK 0xFFFF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_LSB 0xF
+#define QIB_6120_EXTStatus_MemBISTFoundErr_RMASK 0x1
+#define QIB_6120_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_6120_EXTStatus_MemBISTEndTest_RMASK 0x1
+#define QIB_6120_EXTStatus_Reserved2_LSB 0x0
+#define QIB_6120_EXTStatus_Reserved2_RMASK 0x3FFF
+
+#define QIB_6120_EXTCtrl_OFFS 0xD8
+#define QIB_6120_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_6120_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_6120_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_6120_EXTCtrl_Reserved_LSB 0x4
+#define QIB_6120_EXTCtrl_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_LSB 0x3
+#define QIB_6120_EXTCtrl_LEDPriPortGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_LSB 0x2
+#define QIB_6120_EXTCtrl_LEDPriPortYellowOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_LSB 0x1
+#define QIB_6120_EXTCtrl_LEDGblOkGreenOn_RMASK 0x1
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_LSB 0x0
+#define QIB_6120_EXTCtrl_LEDGblErrRedOff_RMASK 0x1
+
+#define QIB_6120_GPIOOut_OFFS 0xE0
+
+#define QIB_6120_GPIOMask_OFFS 0xE8
+
+#define QIB_6120_GPIOStatus_OFFS 0xF0
+
+#define QIB_6120_GPIOClear_OFFS 0xF8
+
+#define QIB_6120_RcvCtrl_OFFS 0x100
+#define QIB_6120_RcvCtrl_TailUpd_LSB 0x1F
+#define QIB_6120_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_LSB 0x1E
+#define QIB_6120_RcvCtrl_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_6120_RcvCtrl_Reserved_LSB 0x15
+#define QIB_6120_RcvCtrl_Reserved_RMASK 0x1FF
+#define QIB_6120_RcvCtrl_IntrAvail_LSB 0x10
+#define QIB_6120_RcvCtrl_IntrAvail_RMASK 0x1F
+#define QIB_6120_RcvCtrl_Reserved1_LSB 0x9
+#define QIB_6120_RcvCtrl_Reserved1_RMASK 0x7F
+#define QIB_6120_RcvCtrl_Reserved2_LSB 0x5
+#define QIB_6120_RcvCtrl_Reserved2_RMASK 0xF
+#define QIB_6120_RcvCtrl_PortEnable_LSB 0x0
+#define QIB_6120_RcvCtrl_PortEnable_RMASK 0x1F
+
+#define QIB_6120_RcvBTHQP_OFFS 0x108
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_LSB 0x1E
+#define QIB_6120_RcvBTHQP_BTHQP_Mask_RMASK 0x3
+#define QIB_6120_RcvBTHQP_Reserved_LSB 0x18
+#define QIB_6120_RcvBTHQP_Reserved_RMASK 0x3F
+#define QIB_6120_RcvBTHQP_RcvBTHQP_LSB 0x0
+#define QIB_6120_RcvBTHQP_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_6120_RcvHdrSize_OFFS 0x110
+
+#define QIB_6120_RcvHdrCnt_OFFS 0x118
+
+#define QIB_6120_RcvHdrEntSize_OFFS 0x120
+
+#define QIB_6120_RcvTIDBase_OFFS 0x128
+
+#define QIB_6120_RcvTIDCnt_OFFS 0x130
+
+#define QIB_6120_RcvEgrBase_OFFS 0x138
+
+#define QIB_6120_RcvEgrCnt_OFFS 0x140
+
+#define QIB_6120_RcvBufBase_OFFS 0x148
+
+#define QIB_6120_RcvBufSize_OFFS 0x150
+
+#define QIB_6120_RxIntMemBase_OFFS 0x158
+
+#define QIB_6120_RxIntMemSize_OFFS 0x160
+
+#define QIB_6120_RcvPartitionKey_OFFS 0x168
+
+#define QIB_6120_RcvPktLEDCnt_OFFS 0x178
+#define QIB_6120_RcvPktLEDCnt_ONperiod_LSB 0x20
+#define QIB_6120_RcvPktLEDCnt_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_LSB 0x0
+#define QIB_6120_RcvPktLEDCnt_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_6120_SendCtrl_OFFS 0x1C0
+#define QIB_6120_SendCtrl_Disarm_LSB 0x1F
+#define QIB_6120_SendCtrl_Disarm_RMASK 0x1
+#define QIB_6120_SendCtrl_Reserved_LSB 0x17
+#define QIB_6120_SendCtrl_Reserved_RMASK 0xFF
+#define QIB_6120_SendCtrl_DisarmPIOBuf_LSB 0x10
+#define QIB_6120_SendCtrl_DisarmPIOBuf_RMASK 0x7F
+#define QIB_6120_SendCtrl_Reserved1_LSB 0x4
+#define QIB_6120_SendCtrl_Reserved1_RMASK 0xFFF
+#define QIB_6120_SendCtrl_PIOEnable_LSB 0x3
+#define QIB_6120_SendCtrl_PIOEnable_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_LSB 0x2
+#define QIB_6120_SendCtrl_PIOBufAvailUpd_RMASK 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_LSB 0x1
+#define QIB_6120_SendCtrl_PIOIntBufAvail_RMASK 0x1
+#define QIB_6120_SendCtrl_Abort_LSB 0x0
+#define QIB_6120_SendCtrl_Abort_RMASK 0x1
+
+#define QIB_6120_SendPIOBufBase_OFFS 0x1C8
+#define QIB_6120_SendPIOBufBase_Reserved_LSB 0x35
+#define QIB_6120_SendPIOBufBase_Reserved_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_6120_SendPIOBufBase_Reserved1_LSB 0x15
+#define QIB_6120_SendPIOBufBase_Reserved1_RMASK 0x7FF
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_6120_SendPIOSize_OFFS 0x1D0
+#define QIB_6120_SendPIOSize_Reserved_LSB 0x2D
+#define QIB_6120_SendPIOSize_Reserved_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_6120_SendPIOSize_Reserved1_LSB 0xC
+#define QIB_6120_SendPIOSize_Reserved1_RMASK 0xFFFFF
+#define QIB_6120_SendPIOSize_Size_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_6120_SendPIOBufCnt_OFFS 0x1D8
+#define QIB_6120_SendPIOBufCnt_Reserved_LSB 0x24
+#define QIB_6120_SendPIOBufCnt_Reserved_RMASK 0xFFFFFFF
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_LSB 0x20
+#define QIB_6120_SendPIOBufCnt_Num_LargePIO_RMASK 0xF
+#define QIB_6120_SendPIOBufCnt_Reserved1_LSB 0x9
+#define QIB_6120_SendPIOBufCnt_Reserved1_RMASK 0x7FFFFF
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_LSB 0x0
+#define QIB_6120_SendPIOBufCnt_Num_SmallPIO_RMASK 0x1FF
+
+#define QIB_6120_SendPIOAvailAddr_OFFS 0x1E0
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_LSB 0x6
+#define QIB_6120_SendPIOAvailAddr_SendPIOAvailAddr_RMASK 0x3FFFFFFFF
+#define QIB_6120_SendPIOAvailAddr_Reserved_LSB 0x0
+#define QIB_6120_SendPIOAvailAddr_Reserved_RMASK 0x3F
+
+#define QIB_6120_SendBufErr0_OFFS 0x240
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_LSB 0x0
+#define QIB_6120_SendBufErr0_SendBufErrPIO_63_0_RMASK 0x0
+
+#define QIB_6120_RcvHdrAddr0_OFFS 0x280
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_LSB 0x2
+#define QIB_6120_RcvHdrAddr0_RcvHdrAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_RcvHdrTailAddr0_OFFS 0x300
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_LSB 0x2
+#define QIB_6120_RcvHdrTailAddr0_RcvHdrTailAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_6120_RcvHdrTailAddr0_Reserved_LSB 0x0
+#define QIB_6120_RcvHdrTailAddr0_Reserved_RMASK 0x3
+
+#define QIB_6120_SerdesCfg0_OFFS 0x3C0
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_LSB 0x3F
+#define QIB_6120_SerdesCfg0_DisableIBTxIdleDetect_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Reserved_LSB 0x38
+#define QIB_6120_SerdesCfg0_Reserved_RMASK 0x7F
+#define QIB_6120_SerdesCfg0_RxEqCtl_LSB 0x36
+#define QIB_6120_SerdesCfg0_RxEqCtl_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TxTermAdj_LSB 0x34
+#define QIB_6120_SerdesCfg0_TxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_RxTermAdj_LSB 0x32
+#define QIB_6120_SerdesCfg0_RxTermAdj_RMASK 0x3
+#define QIB_6120_SerdesCfg0_TermAdj1_LSB 0x31
+#define QIB_6120_SerdesCfg0_TermAdj1_RMASK 0x1
+#define QIB_6120_SerdesCfg0_TermAdj0_LSB 0x30
+#define QIB_6120_SerdesCfg0_TermAdj0_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKA_LSB 0x2F
+#define QIB_6120_SerdesCfg0_LPBKA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKB_LSB 0x2E
+#define QIB_6120_SerdesCfg0_LPBKB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKC_LSB 0x2D
+#define QIB_6120_SerdesCfg0_LPBKC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_LPBKD_LSB 0x2C
+#define QIB_6120_SerdesCfg0_LPBKD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_PW_LSB 0x2B
+#define QIB_6120_SerdesCfg0_PW_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RefSel_LSB 0x29
+#define QIB_6120_SerdesCfg0_RefSel_RMASK 0x3
+#define QIB_6120_SerdesCfg0_ParReset_LSB 0x28
+#define QIB_6120_SerdesCfg0_ParReset_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ParLPBK_LSB 0x27
+#define QIB_6120_SerdesCfg0_ParLPBK_RMASK 0x1
+#define QIB_6120_SerdesCfg0_OffsetEn_LSB 0x26
+#define QIB_6120_SerdesCfg0_OffsetEn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_Offset_LSB 0x1E
+#define QIB_6120_SerdesCfg0_Offset_RMASK 0xFF
+#define QIB_6120_SerdesCfg0_L2PwrDn_LSB 0x1D
+#define QIB_6120_SerdesCfg0_L2PwrDn_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetPLL_LSB 0x1C
+#define QIB_6120_SerdesCfg0_ResetPLL_RMASK 0x1
+#define QIB_6120_SerdesCfg0_RxTermEnX_LSB 0x18
+#define QIB_6120_SerdesCfg0_RxTermEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_LSB 0x14
+#define QIB_6120_SerdesCfg0_BeaconTxEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxDetEnX_LSB 0x10
+#define QIB_6120_SerdesCfg0_RxDetEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_TxIdeEnX_LSB 0xC
+#define QIB_6120_SerdesCfg0_TxIdeEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_RxIdleEnX_LSB 0x8
+#define QIB_6120_SerdesCfg0_RxIdleEnX_RMASK 0xF
+#define QIB_6120_SerdesCfg0_L1PwrDnA_LSB 0x7
+#define QIB_6120_SerdesCfg0_L1PwrDnA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnB_LSB 0x6
+#define QIB_6120_SerdesCfg0_L1PwrDnB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnC_LSB 0x5
+#define QIB_6120_SerdesCfg0_L1PwrDnC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_L1PwrDnD_LSB 0x4
+#define QIB_6120_SerdesCfg0_L1PwrDnD_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetA_LSB 0x3
+#define QIB_6120_SerdesCfg0_ResetA_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetB_LSB 0x2
+#define QIB_6120_SerdesCfg0_ResetB_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetC_LSB 0x1
+#define QIB_6120_SerdesCfg0_ResetC_RMASK 0x1
+#define QIB_6120_SerdesCfg0_ResetD_LSB 0x0
+#define QIB_6120_SerdesCfg0_ResetD_RMASK 0x1
+
+#define QIB_6120_SerdesStat_OFFS 0x3D0
+#define QIB_6120_SerdesStat_Reserved_LSB 0xC
+#define QIB_6120_SerdesStat_Reserved_RMASK 0xFFFFFFFFFFFFF
+#define QIB_6120_SerdesStat_BeaconDetA_LSB 0xB
+#define QIB_6120_SerdesStat_BeaconDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetB_LSB 0xA
+#define QIB_6120_SerdesStat_BeaconDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetC_LSB 0x9
+#define QIB_6120_SerdesStat_BeaconDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_BeaconDetD_LSB 0x8
+#define QIB_6120_SerdesStat_BeaconDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetA_LSB 0x7
+#define QIB_6120_SerdesStat_RxDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetB_LSB 0x6
+#define QIB_6120_SerdesStat_RxDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetC_LSB 0x5
+#define QIB_6120_SerdesStat_RxDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_RxDetD_LSB 0x4
+#define QIB_6120_SerdesStat_RxDetD_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetA_LSB 0x3
+#define QIB_6120_SerdesStat_TxIdleDetA_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetB_LSB 0x2
+#define QIB_6120_SerdesStat_TxIdleDetB_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_LSB 0x1
+#define QIB_6120_SerdesStat_TxIdleDetC_RMASK 0x1
+#define QIB_6120_SerdesStat_TxIdleDetD_LSB 0x0
+#define QIB_6120_SerdesStat_TxIdleDetD_RMASK 0x1
+
+#define QIB_6120_XGXSCfg_OFFS 0x3D8
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_LSB 0x3F
+#define QIB_6120_XGXSCfg_ArmLaunchErrorDisable_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved_LSB 0x17
+#define QIB_6120_XGXSCfg_Reserved_RMASK 0xFFFFFFFFFF
+#define QIB_6120_XGXSCfg_polarity_inv_LSB 0x13
+#define QIB_6120_XGXSCfg_polarity_inv_RMASK 0xF
+#define QIB_6120_XGXSCfg_link_sync_mask_LSB 0x9
+#define QIB_6120_XGXSCfg_link_sync_mask_RMASK 0x3FF
+#define QIB_6120_XGXSCfg_port_addr_LSB 0x4
+#define QIB_6120_XGXSCfg_port_addr_RMASK 0x1F
+#define QIB_6120_XGXSCfg_mdd_30_LSB 0x3
+#define QIB_6120_XGXSCfg_mdd_30_RMASK 0x1
+#define QIB_6120_XGXSCfg_xcv_resetn_LSB 0x2
+#define QIB_6120_XGXSCfg_xcv_resetn_RMASK 0x1
+#define QIB_6120_XGXSCfg_Reserved1_LSB 0x1
+#define QIB_6120_XGXSCfg_Reserved1_RMASK 0x1
+#define QIB_6120_XGXSCfg_tx_rx_resetn_LSB 0x0
+#define QIB_6120_XGXSCfg_tx_rx_resetn_RMASK 0x1
+
+#define QIB_6120_LBIntCnt_OFFS 0x12000
+
+#define QIB_6120_LBFlowStallCnt_OFFS 0x12008
+
+#define QIB_6120_TxUnsupVLErrCnt_OFFS 0x12018
+
+#define QIB_6120_TxDataPktCnt_OFFS 0x12020
+
+#define QIB_6120_TxFlowPktCnt_OFFS 0x12028
+
+#define QIB_6120_TxDwordCnt_OFFS 0x12030
+
+#define QIB_6120_TxLenErrCnt_OFFS 0x12038
+
+#define QIB_6120_TxMaxMinLenErrCnt_OFFS 0x12040
+
+#define QIB_6120_TxUnderrunCnt_OFFS 0x12048
+
+#define QIB_6120_TxFlowStallCnt_OFFS 0x12050
+
+#define QIB_6120_TxDroppedPktCnt_OFFS 0x12058
+
+#define QIB_6120_RxDroppedPktCnt_OFFS 0x12060
+
+#define QIB_6120_RxDataPktCnt_OFFS 0x12068
+
+#define QIB_6120_RxFlowPktCnt_OFFS 0x12070
+
+#define QIB_6120_RxDwordCnt_OFFS 0x12078
+
+#define QIB_6120_RxLenErrCnt_OFFS 0x12080
+
+#define QIB_6120_RxMaxMinLenErrCnt_OFFS 0x12088
+
+#define QIB_6120_RxICRCErrCnt_OFFS 0x12090
+
+#define QIB_6120_RxVCRCErrCnt_OFFS 0x12098
+
+#define QIB_6120_RxFlowCtrlErrCnt_OFFS 0x120A0
+
+#define QIB_6120_RxBadFormatCnt_OFFS 0x120A8
+
+#define QIB_6120_RxLinkProblemCnt_OFFS 0x120B0
+
+#define QIB_6120_RxEBPCnt_OFFS 0x120B8
+
+#define QIB_6120_RxLPCRCErrCnt_OFFS 0x120C0
+
+#define QIB_6120_RxBufOvflCnt_OFFS 0x120C8
+
+#define QIB_6120_RxTIDFullErrCnt_OFFS 0x120D0
+
+#define QIB_6120_RxTIDValidErrCnt_OFFS 0x120D8
+
+#define QIB_6120_RxPKeyMismatchCnt_OFFS 0x120E0
+
+#define QIB_6120_RxP0HdrEgrOvflCnt_OFFS 0x120E8
+
+#define QIB_6120_IBStatusChangeCnt_OFFS 0x12140
+
+#define QIB_6120_IBLinkErrRecoveryCnt_OFFS 0x12148
+
+#define QIB_6120_IBLinkDownedCnt_OFFS 0x12150
+
+#define QIB_6120_IBSymbolErrCnt_OFFS 0x12158
+
+#define QIB_6120_PcieRetryBufDiagQwordCnt_OFFS 0x12170
+
+#define QIB_6120_RcvEgrArray0_OFFS 0x14000
+
+#define QIB_6120_RcvTIDArray0_OFFS 0x54000
+
+#define QIB_6120_PIOLaunchFIFO_OFFS 0x64000
+
+#define QIB_6120_SendPIOpbcCache_OFFS 0x64800
+
+#define QIB_6120_RcvBuf1_OFFS 0x72000
+
+#define QIB_6120_RcvBuf2_OFFS 0x75000
+
+#define QIB_6120_RcvFlags_OFFS 0x77000
+
+#define QIB_6120_RcvLookupBuf1_OFFS 0x79000
+
+#define QIB_6120_RcvDMABuf_OFFS 0x7B000
+
+#define QIB_6120_MiscRXEIntMem_OFFS 0x7C000
+
+#define QIB_6120_PCIERcvBuf_OFFS 0x80000
+
+#define QIB_6120_PCIERetryBuf_OFFS 0x82000
+
+#define QIB_6120_PCIERcvBufRdToWrAddr_OFFS 0x84000
+
+#define QIB_6120_PIOBuf0_MA_OFFS 0x100000
diff --git a/drivers/infiniband/hw/qib/qib_7220.h b/drivers/infiniband/hw/qib/qib_7220.h
new file mode 100644
index 00000000000..a5356cb4252
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_7220.h
@@ -0,0 +1,149 @@
+#ifndef _QIB_7220_H
+#define _QIB_7220_H
+/*
+ * Copyright (c) 2007, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* grab register-defs auto-generated by HW */
+#include "qib_7220_regs.h"
+
+/* The number of eager receive TIDs for context zero. */
+#define IBA7220_KRCVEGRCNT      2048U
+
+#define IB_7220_LT_STATE_CFGRCVFCFG      0x09
+#define IB_7220_LT_STATE_CFGWAITRMT      0x0a
+#define IB_7220_LT_STATE_TXREVLANES      0x0d
+#define IB_7220_LT_STATE_CFGENH          0x10
+
+struct qib_chip_specific {
+	u64 __iomem *cregbase;
+	u64 *cntrs;
+	u64 *portcntrs;
+	spinlock_t sdepb_lock; /* serdes EPB bus */
+	spinlock_t rcvmod_lock; /* protect rcvctrl shadow changes */
+	spinlock_t gpio_lock; /* RMW of shadows/regs for ExtCtrl and GPIO */
+	u64 hwerrmask;
+	u64 errormask;
+	u64 gpio_out; /* shadow of kr_gpio_out, for rmw ops */
+	u64 gpio_mask; /* shadow the gpio mask register */
+	u64 extctrl; /* shadow the gpio output enable, etc... */
+	u32 ncntrs;
+	u32 nportcntrs;
+	u32 cntrnamelen;
+	u32 portcntrnamelen;
+	u32 numctxts;
+	u32 rcvegrcnt;
+	u32 autoneg_tries;
+	u32 serdes_first_init_done;
+	u32 sdmabufcnt;
+	u32 lastbuf_for_pio;
+	u32 updthresh; /* current AvailUpdThld */
+	u32 updthresh_dflt; /* default AvailUpdThld */
+	int irq;
+	u8 presets_needed;
+	u8 relock_timer_active;
+	char emsgbuf[128];
+	char sdmamsgbuf[192];
+	char bitsmsgbuf[64];
+	struct timer_list relock_timer;
+	unsigned int relock_interval; /* in jiffies */
+};
+
+struct qib_chippport_specific {
+	struct qib_pportdata pportdata;
+	wait_queue_head_t autoneg_wait;
+	struct delayed_work autoneg_work;
+	struct timer_list chase_timer;
+	/*
+	 * these 5 fields are used to establish deltas for IB symbol
+	 * errors and linkrecovery errors.  They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+	u64 ibcctrl; /* kr_ibcctrl shadow */
+	u64 ibcddrctrl; /* kr_ibcddrctrl shadow */
+	unsigned long chase_end;
+	u32 last_delay_mult;
+};
+
+/*
+ * This header file provides the declarations and common definitions
+ * for (mostly) manipulation of the SerDes blocks within the IBA7220.
+ * the functions declared should only be called from within other
+ * 7220-related files such as qib_iba7220.c or qib_sd7220.c.
+ */
+int qib_sd7220_presets(struct qib_devdata *dd);
+int qib_sd7220_init(struct qib_devdata *dd);
+void qib_sd7220_clr_ibpar(struct qib_devdata *);
+/*
+ * Below used for sdnum parameter, selecting one of the two sections
+ * used for PCIe, or the single SerDes used for IB, which is the
+ * only one currently used
+ */
+#define IB_7220_SERDES 2
+
+static inline u32 qib_read_kreg32(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readl((u32 __iomem *)&dd->kregbase[regno]);
+}
+
+static inline u64 qib_read_kreg64(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+
+	return readq(&dd->kregbase[regno]);
+}
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u16 regno, u64 value)
+{
+	if (dd->kregbase)
+		writeq(value, &dd->kregbase[regno]);
+}
+
+void set_7220_relock_poll(struct qib_devdata *, int);
+void shutdown_7220_relock_poll(struct qib_devdata *);
+void toggle_7220_rclkrls(struct qib_devdata *);
+
+
+#endif /* _QIB_7220_H */
diff --git a/drivers/infiniband/hw/qib/qib_7220_regs.h b/drivers/infiniband/hw/qib/qib_7220_regs.h
new file mode 100644
index 00000000000..0da5bb750e5
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_7220_regs.h
@@ -0,0 +1,1496 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_7220_Revision_OFFS 0x0
+#define QIB_7220_Revision_R_Simulator_LSB 0x3F
+#define QIB_7220_Revision_R_Simulator_RMASK 0x1
+#define QIB_7220_Revision_R_Emulation_LSB 0x3E
+#define QIB_7220_Revision_R_Emulation_RMASK 0x1
+#define QIB_7220_Revision_R_Emulation_Revcode_LSB 0x28
+#define QIB_7220_Revision_R_Emulation_Revcode_RMASK 0x3FFFFF
+#define QIB_7220_Revision_BoardID_LSB 0x20
+#define QIB_7220_Revision_BoardID_RMASK 0xFF
+#define QIB_7220_Revision_R_SW_LSB 0x18
+#define QIB_7220_Revision_R_SW_RMASK 0xFF
+#define QIB_7220_Revision_R_Arch_LSB 0x10
+#define QIB_7220_Revision_R_Arch_RMASK 0xFF
+#define QIB_7220_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_7220_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_7220_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_7220_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_7220_Control_OFFS 0x8
+#define QIB_7220_Control_SyncResetExceptPcieIRAMRST_LSB 0x7
+#define QIB_7220_Control_SyncResetExceptPcieIRAMRST_RMASK 0x1
+#define QIB_7220_Control_PCIECplQDiagEn_LSB 0x6
+#define QIB_7220_Control_PCIECplQDiagEn_RMASK 0x1
+#define QIB_7220_Control_Reserved_LSB 0x5
+#define QIB_7220_Control_Reserved_RMASK 0x1
+#define QIB_7220_Control_TxLatency_LSB 0x4
+#define QIB_7220_Control_TxLatency_RMASK 0x1
+#define QIB_7220_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_7220_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_7220_Control_LinkEn_LSB 0x2
+#define QIB_7220_Control_LinkEn_RMASK 0x1
+#define QIB_7220_Control_FreezeMode_LSB 0x1
+#define QIB_7220_Control_FreezeMode_RMASK 0x1
+#define QIB_7220_Control_SyncReset_LSB 0x0
+#define QIB_7220_Control_SyncReset_RMASK 0x1
+
+#define QIB_7220_PageAlign_OFFS 0x10
+
+#define QIB_7220_PortCnt_OFFS 0x18
+
+#define QIB_7220_SendRegBase_OFFS 0x30
+
+#define QIB_7220_UserRegBase_OFFS 0x38
+
+#define QIB_7220_CntrRegBase_OFFS 0x40
+
+#define QIB_7220_Scratch_OFFS 0x48
+
+#define QIB_7220_IntMask_OFFS 0x68
+#define QIB_7220_IntMask_SDmaIntMask_LSB 0x3F
+#define QIB_7220_IntMask_SDmaIntMask_RMASK 0x1
+#define QIB_7220_IntMask_SDmaDisabledMasked_LSB 0x3E
+#define QIB_7220_IntMask_SDmaDisabledMasked_RMASK 0x1
+#define QIB_7220_IntMask_Reserved_LSB 0x31
+#define QIB_7220_IntMask_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntMask_RcvUrg16IntMask_LSB 0x30
+#define QIB_7220_IntMask_RcvUrg16IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg15IntMask_LSB 0x2F
+#define QIB_7220_IntMask_RcvUrg15IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg14IntMask_LSB 0x2E
+#define QIB_7220_IntMask_RcvUrg14IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg13IntMask_LSB 0x2D
+#define QIB_7220_IntMask_RcvUrg13IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg12IntMask_LSB 0x2C
+#define QIB_7220_IntMask_RcvUrg12IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg11IntMask_LSB 0x2B
+#define QIB_7220_IntMask_RcvUrg11IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg10IntMask_LSB 0x2A
+#define QIB_7220_IntMask_RcvUrg10IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg9IntMask_LSB 0x29
+#define QIB_7220_IntMask_RcvUrg9IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg8IntMask_LSB 0x28
+#define QIB_7220_IntMask_RcvUrg8IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg7IntMask_LSB 0x27
+#define QIB_7220_IntMask_RcvUrg7IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg6IntMask_LSB 0x26
+#define QIB_7220_IntMask_RcvUrg6IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg5IntMask_LSB 0x25
+#define QIB_7220_IntMask_RcvUrg5IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg4IntMask_LSB 0x24
+#define QIB_7220_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg3IntMask_LSB 0x23
+#define QIB_7220_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg2IntMask_LSB 0x22
+#define QIB_7220_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg1IntMask_LSB 0x21
+#define QIB_7220_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvUrg0IntMask_LSB 0x20
+#define QIB_7220_IntMask_RcvUrg0IntMask_RMASK 0x1
+#define QIB_7220_IntMask_ErrorIntMask_LSB 0x1F
+#define QIB_7220_IntMask_ErrorIntMask_RMASK 0x1
+#define QIB_7220_IntMask_PioSetIntMask_LSB 0x1E
+#define QIB_7220_IntMask_PioSetIntMask_RMASK 0x1
+#define QIB_7220_IntMask_PioBufAvailIntMask_LSB 0x1D
+#define QIB_7220_IntMask_PioBufAvailIntMask_RMASK 0x1
+#define QIB_7220_IntMask_assertGPIOIntMask_LSB 0x1C
+#define QIB_7220_IntMask_assertGPIOIntMask_RMASK 0x1
+#define QIB_7220_IntMask_IBSerdesTrimDoneIntMask_LSB 0x1B
+#define QIB_7220_IntMask_IBSerdesTrimDoneIntMask_RMASK 0x1
+#define QIB_7220_IntMask_JIntMask_LSB 0x1A
+#define QIB_7220_IntMask_JIntMask_RMASK 0x1
+#define QIB_7220_IntMask_Reserved1_LSB 0x11
+#define QIB_7220_IntMask_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntMask_RcvAvail16IntMask_LSB 0x10
+#define QIB_7220_IntMask_RcvAvail16IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail15IntMask_LSB 0xF
+#define QIB_7220_IntMask_RcvAvail15IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail14IntMask_LSB 0xE
+#define QIB_7220_IntMask_RcvAvail14IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail13IntMask_LSB 0xD
+#define QIB_7220_IntMask_RcvAvail13IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail12IntMask_LSB 0xC
+#define QIB_7220_IntMask_RcvAvail12IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail11IntMask_LSB 0xB
+#define QIB_7220_IntMask_RcvAvail11IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail10IntMask_LSB 0xA
+#define QIB_7220_IntMask_RcvAvail10IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail9IntMask_LSB 0x9
+#define QIB_7220_IntMask_RcvAvail9IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail8IntMask_LSB 0x8
+#define QIB_7220_IntMask_RcvAvail8IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail7IntMask_LSB 0x7
+#define QIB_7220_IntMask_RcvAvail7IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail6IntMask_LSB 0x6
+#define QIB_7220_IntMask_RcvAvail6IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail5IntMask_LSB 0x5
+#define QIB_7220_IntMask_RcvAvail5IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail4IntMask_LSB 0x4
+#define QIB_7220_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail3IntMask_LSB 0x3
+#define QIB_7220_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail2IntMask_LSB 0x2
+#define QIB_7220_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail1IntMask_LSB 0x1
+#define QIB_7220_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_7220_IntMask_RcvAvail0IntMask_LSB 0x0
+#define QIB_7220_IntMask_RcvAvail0IntMask_RMASK 0x1
+
+#define QIB_7220_IntStatus_OFFS 0x70
+#define QIB_7220_IntStatus_SDmaInt_LSB 0x3F
+#define QIB_7220_IntStatus_SDmaInt_RMASK 0x1
+#define QIB_7220_IntStatus_SDmaDisabled_LSB 0x3E
+#define QIB_7220_IntStatus_SDmaDisabled_RMASK 0x1
+#define QIB_7220_IntStatus_Reserved_LSB 0x31
+#define QIB_7220_IntStatus_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntStatus_RcvUrg16_LSB 0x30
+#define QIB_7220_IntStatus_RcvUrg16_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg15_LSB 0x2F
+#define QIB_7220_IntStatus_RcvUrg15_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg14_LSB 0x2E
+#define QIB_7220_IntStatus_RcvUrg14_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg13_LSB 0x2D
+#define QIB_7220_IntStatus_RcvUrg13_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg12_LSB 0x2C
+#define QIB_7220_IntStatus_RcvUrg12_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg11_LSB 0x2B
+#define QIB_7220_IntStatus_RcvUrg11_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg10_LSB 0x2A
+#define QIB_7220_IntStatus_RcvUrg10_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg9_LSB 0x29
+#define QIB_7220_IntStatus_RcvUrg9_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg8_LSB 0x28
+#define QIB_7220_IntStatus_RcvUrg8_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg7_LSB 0x27
+#define QIB_7220_IntStatus_RcvUrg7_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg6_LSB 0x26
+#define QIB_7220_IntStatus_RcvUrg6_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg5_LSB 0x25
+#define QIB_7220_IntStatus_RcvUrg5_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg4_LSB 0x24
+#define QIB_7220_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg3_LSB 0x23
+#define QIB_7220_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg2_LSB 0x22
+#define QIB_7220_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg1_LSB 0x21
+#define QIB_7220_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_7220_IntStatus_RcvUrg0_LSB 0x20
+#define QIB_7220_IntStatus_RcvUrg0_RMASK 0x1
+#define QIB_7220_IntStatus_Error_LSB 0x1F
+#define QIB_7220_IntStatus_Error_RMASK 0x1
+#define QIB_7220_IntStatus_PioSent_LSB 0x1E
+#define QIB_7220_IntStatus_PioSent_RMASK 0x1
+#define QIB_7220_IntStatus_PioBufAvail_LSB 0x1D
+#define QIB_7220_IntStatus_PioBufAvail_RMASK 0x1
+#define QIB_7220_IntStatus_assertGPIO_LSB 0x1C
+#define QIB_7220_IntStatus_assertGPIO_RMASK 0x1
+#define QIB_7220_IntStatus_IBSerdesTrimDone_LSB 0x1B
+#define QIB_7220_IntStatus_IBSerdesTrimDone_RMASK 0x1
+#define QIB_7220_IntStatus_JInt_LSB 0x1A
+#define QIB_7220_IntStatus_JInt_RMASK 0x1
+#define QIB_7220_IntStatus_Reserved1_LSB 0x11
+#define QIB_7220_IntStatus_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntStatus_RcvAvail16_LSB 0x10
+#define QIB_7220_IntStatus_RcvAvail16_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail15_LSB 0xF
+#define QIB_7220_IntStatus_RcvAvail15_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail14_LSB 0xE
+#define QIB_7220_IntStatus_RcvAvail14_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail13_LSB 0xD
+#define QIB_7220_IntStatus_RcvAvail13_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail12_LSB 0xC
+#define QIB_7220_IntStatus_RcvAvail12_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail11_LSB 0xB
+#define QIB_7220_IntStatus_RcvAvail11_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail10_LSB 0xA
+#define QIB_7220_IntStatus_RcvAvail10_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail9_LSB 0x9
+#define QIB_7220_IntStatus_RcvAvail9_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail8_LSB 0x8
+#define QIB_7220_IntStatus_RcvAvail8_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail7_LSB 0x7
+#define QIB_7220_IntStatus_RcvAvail7_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail6_LSB 0x6
+#define QIB_7220_IntStatus_RcvAvail6_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail5_LSB 0x5
+#define QIB_7220_IntStatus_RcvAvail5_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail4_LSB 0x4
+#define QIB_7220_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail3_LSB 0x3
+#define QIB_7220_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail2_LSB 0x2
+#define QIB_7220_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail1_LSB 0x1
+#define QIB_7220_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_7220_IntStatus_RcvAvail0_LSB 0x0
+#define QIB_7220_IntStatus_RcvAvail0_RMASK 0x1
+
+#define QIB_7220_IntClear_OFFS 0x78
+#define QIB_7220_IntClear_SDmaIntClear_LSB 0x3F
+#define QIB_7220_IntClear_SDmaIntClear_RMASK 0x1
+#define QIB_7220_IntClear_SDmaDisabledClear_LSB 0x3E
+#define QIB_7220_IntClear_SDmaDisabledClear_RMASK 0x1
+#define QIB_7220_IntClear_Reserved_LSB 0x31
+#define QIB_7220_IntClear_Reserved_RMASK 0x1FFF
+#define QIB_7220_IntClear_RcvUrg16IntClear_LSB 0x30
+#define QIB_7220_IntClear_RcvUrg16IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg15IntClear_LSB 0x2F
+#define QIB_7220_IntClear_RcvUrg15IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg14IntClear_LSB 0x2E
+#define QIB_7220_IntClear_RcvUrg14IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg13IntClear_LSB 0x2D
+#define QIB_7220_IntClear_RcvUrg13IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg12IntClear_LSB 0x2C
+#define QIB_7220_IntClear_RcvUrg12IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg11IntClear_LSB 0x2B
+#define QIB_7220_IntClear_RcvUrg11IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg10IntClear_LSB 0x2A
+#define QIB_7220_IntClear_RcvUrg10IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg9IntClear_LSB 0x29
+#define QIB_7220_IntClear_RcvUrg9IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg8IntClear_LSB 0x28
+#define QIB_7220_IntClear_RcvUrg8IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg7IntClear_LSB 0x27
+#define QIB_7220_IntClear_RcvUrg7IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg6IntClear_LSB 0x26
+#define QIB_7220_IntClear_RcvUrg6IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg5IntClear_LSB 0x25
+#define QIB_7220_IntClear_RcvUrg5IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg4IntClear_LSB 0x24
+#define QIB_7220_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg3IntClear_LSB 0x23
+#define QIB_7220_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg2IntClear_LSB 0x22
+#define QIB_7220_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg1IntClear_LSB 0x21
+#define QIB_7220_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvUrg0IntClear_LSB 0x20
+#define QIB_7220_IntClear_RcvUrg0IntClear_RMASK 0x1
+#define QIB_7220_IntClear_ErrorIntClear_LSB 0x1F
+#define QIB_7220_IntClear_ErrorIntClear_RMASK 0x1
+#define QIB_7220_IntClear_PioSetIntClear_LSB 0x1E
+#define QIB_7220_IntClear_PioSetIntClear_RMASK 0x1
+#define QIB_7220_IntClear_PioBufAvailIntClear_LSB 0x1D
+#define QIB_7220_IntClear_PioBufAvailIntClear_RMASK 0x1
+#define QIB_7220_IntClear_assertGPIOIntClear_LSB 0x1C
+#define QIB_7220_IntClear_assertGPIOIntClear_RMASK 0x1
+#define QIB_7220_IntClear_IBSerdesTrimDoneClear_LSB 0x1B
+#define QIB_7220_IntClear_IBSerdesTrimDoneClear_RMASK 0x1
+#define QIB_7220_IntClear_JIntClear_LSB 0x1A
+#define QIB_7220_IntClear_JIntClear_RMASK 0x1
+#define QIB_7220_IntClear_Reserved1_LSB 0x11
+#define QIB_7220_IntClear_Reserved1_RMASK 0x1FF
+#define QIB_7220_IntClear_RcvAvail16IntClear_LSB 0x10
+#define QIB_7220_IntClear_RcvAvail16IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail15IntClear_LSB 0xF
+#define QIB_7220_IntClear_RcvAvail15IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail14IntClear_LSB 0xE
+#define QIB_7220_IntClear_RcvAvail14IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail13IntClear_LSB 0xD
+#define QIB_7220_IntClear_RcvAvail13IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail12IntClear_LSB 0xC
+#define QIB_7220_IntClear_RcvAvail12IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail11IntClear_LSB 0xB
+#define QIB_7220_IntClear_RcvAvail11IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail10IntClear_LSB 0xA
+#define QIB_7220_IntClear_RcvAvail10IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail9IntClear_LSB 0x9
+#define QIB_7220_IntClear_RcvAvail9IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail8IntClear_LSB 0x8
+#define QIB_7220_IntClear_RcvAvail8IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail7IntClear_LSB 0x7
+#define QIB_7220_IntClear_RcvAvail7IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail6IntClear_LSB 0x6
+#define QIB_7220_IntClear_RcvAvail6IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail5IntClear_LSB 0x5
+#define QIB_7220_IntClear_RcvAvail5IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail4IntClear_LSB 0x4
+#define QIB_7220_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail3IntClear_LSB 0x3
+#define QIB_7220_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail2IntClear_LSB 0x2
+#define QIB_7220_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail1IntClear_LSB 0x1
+#define QIB_7220_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_7220_IntClear_RcvAvail0IntClear_LSB 0x0
+#define QIB_7220_IntClear_RcvAvail0IntClear_RMASK 0x1
+
+#define QIB_7220_ErrMask_OFFS 0x80
+#define QIB_7220_ErrMask_Reserved_LSB 0x36
+#define QIB_7220_ErrMask_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrMask_InvalidEEPCmdMask_LSB 0x35
+#define QIB_7220_ErrMask_InvalidEEPCmdMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDescAddrMisalignErrMask_LSB 0x34
+#define QIB_7220_ErrMask_SDmaDescAddrMisalignErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_HardwareErrMask_LSB 0x33
+#define QIB_7220_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_ResetNegatedMask_LSB 0x32
+#define QIB_7220_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_7220_ErrMask_InvalidAddrErrMask_LSB 0x31
+#define QIB_7220_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_IBStatusChangedMask_LSB 0x30
+#define QIB_7220_ErrMask_IBStatusChangedMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaUnexpDataErrMask_LSB 0x2F
+#define QIB_7220_ErrMask_SDmaUnexpDataErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaMissingDwErrMask_LSB 0x2E
+#define QIB_7220_ErrMask_SDmaMissingDwErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDwEnErrMask_LSB 0x2D
+#define QIB_7220_ErrMask_SDmaDwEnErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaRpyTagErrMask_LSB 0x2C
+#define QIB_7220_ErrMask_SDmaRpyTagErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDma1stDescErrMask_LSB 0x2B
+#define QIB_7220_ErrMask_SDma1stDescErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaBaseErrMask_LSB 0x2A
+#define QIB_7220_ErrMask_SDmaBaseErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaTailOutOfBoundErrMask_LSB 0x29
+#define QIB_7220_ErrMask_SDmaTailOutOfBoundErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaOutOfBoundErrMask_LSB 0x28
+#define QIB_7220_ErrMask_SDmaOutOfBoundErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaGenMismatchErrMask_LSB 0x27
+#define QIB_7220_ErrMask_SDmaGenMismatchErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendBufMisuseErrMask_LSB 0x26
+#define QIB_7220_ErrMask_SendBufMisuseErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_7220_ErrMask_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_7220_ErrMask_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendPioArmLaunchErrMask_LSB 0x23
+#define QIB_7220_ErrMask_SendPioArmLaunchErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_7220_ErrMask_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_7220_ErrMask_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendPktLenErrMask_LSB 0x20
+#define QIB_7220_ErrMask_SendPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendUnderRunErrMask_LSB 0x1F
+#define QIB_7220_ErrMask_SendUnderRunErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_7220_ErrMask_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_7220_ErrMask_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SDmaDisabledErrMask_LSB 0x1C
+#define QIB_7220_ErrMask_SDmaDisabledErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_SendSpecialTriggerErrMask_LSB 0x1B
+#define QIB_7220_ErrMask_SendSpecialTriggerErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_Reserved1_LSB 0x12
+#define QIB_7220_ErrMask_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrMask_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_7220_ErrMask_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrErrMask_LSB 0x10
+#define QIB_7220_ErrMask_RcvHdrErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrLenErrMask_LSB 0xF
+#define QIB_7220_ErrMask_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvBadTidErrMask_LSB 0xE
+#define QIB_7220_ErrMask_RcvBadTidErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_7220_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_7220_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvBadVersionErrMask_LSB 0xB
+#define QIB_7220_ErrMask_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvIBFlowErrMask_LSB 0xA
+#define QIB_7220_ErrMask_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvEBPErrMask_LSB 0x9
+#define QIB_7220_ErrMask_RcvEBPErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_7220_ErrMask_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_7220_ErrMask_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_7220_ErrMask_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_7220_ErrMask_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_7220_ErrMask_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_7220_ErrMask_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvICRCErrMask_LSB 0x2
+#define QIB_7220_ErrMask_RcvICRCErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvVCRCErrMask_LSB 0x1
+#define QIB_7220_ErrMask_RcvVCRCErrMask_RMASK 0x1
+#define QIB_7220_ErrMask_RcvFormatErrMask_LSB 0x0
+#define QIB_7220_ErrMask_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_7220_ErrStatus_OFFS 0x88
+#define QIB_7220_ErrStatus_Reserved_LSB 0x36
+#define QIB_7220_ErrStatus_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrStatus_InvalidEEPCmdErr_LSB 0x35
+#define QIB_7220_ErrStatus_InvalidEEPCmdErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDescAddrMisalignErr_LSB 0x34
+#define QIB_7220_ErrStatus_SDmaDescAddrMisalignErr_RMASK 0x1
+#define QIB_7220_ErrStatus_HardwareErr_LSB 0x33
+#define QIB_7220_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_7220_ErrStatus_ResetNegated_LSB 0x32
+#define QIB_7220_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_7220_ErrStatus_InvalidAddrErr_LSB 0x31
+#define QIB_7220_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_7220_ErrStatus_IBStatusChanged_LSB 0x30
+#define QIB_7220_ErrStatus_IBStatusChanged_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaUnexpDataErr_LSB 0x2F
+#define QIB_7220_ErrStatus_SDmaUnexpDataErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaMissingDwErr_LSB 0x2E
+#define QIB_7220_ErrStatus_SDmaMissingDwErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDwEnErr_LSB 0x2D
+#define QIB_7220_ErrStatus_SDmaDwEnErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaRpyTagErr_LSB 0x2C
+#define QIB_7220_ErrStatus_SDmaRpyTagErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDma1stDescErr_LSB 0x2B
+#define QIB_7220_ErrStatus_SDma1stDescErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaBaseErr_LSB 0x2A
+#define QIB_7220_ErrStatus_SDmaBaseErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaTailOutOfBoundErr_LSB 0x29
+#define QIB_7220_ErrStatus_SDmaTailOutOfBoundErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaOutOfBoundErr_LSB 0x28
+#define QIB_7220_ErrStatus_SDmaOutOfBoundErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaGenMismatchErr_LSB 0x27
+#define QIB_7220_ErrStatus_SDmaGenMismatchErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendBufMisuseErr_LSB 0x26
+#define QIB_7220_ErrStatus_SendBufMisuseErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnsupportedVLErr_LSB 0x25
+#define QIB_7220_ErrStatus_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_7220_ErrStatus_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendPioArmLaunchErr_LSB 0x23
+#define QIB_7220_ErrStatus_SendPioArmLaunchErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendDroppedDataPktErr_LSB 0x22
+#define QIB_7220_ErrStatus_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_7220_ErrStatus_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendPktLenErr_LSB 0x20
+#define QIB_7220_ErrStatus_SendPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendUnderRunErr_LSB 0x1F
+#define QIB_7220_ErrStatus_SendUnderRunErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendMaxPktLenErr_LSB 0x1E
+#define QIB_7220_ErrStatus_SendMaxPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendMinPktLenErr_LSB 0x1D
+#define QIB_7220_ErrStatus_SendMinPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SDmaDisabledErr_LSB 0x1C
+#define QIB_7220_ErrStatus_SDmaDisabledErr_RMASK 0x1
+#define QIB_7220_ErrStatus_SendSpecialTriggerErr_LSB 0x1B
+#define QIB_7220_ErrStatus_SendSpecialTriggerErr_RMASK 0x1
+#define QIB_7220_ErrStatus_Reserved1_LSB 0x12
+#define QIB_7220_ErrStatus_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrStatus_RcvIBLostLinkErr_LSB 0x11
+#define QIB_7220_ErrStatus_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrErr_LSB 0x10
+#define QIB_7220_ErrStatus_RcvHdrErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrLenErr_LSB 0xF
+#define QIB_7220_ErrStatus_RcvHdrLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvBadTidErr_LSB 0xE
+#define QIB_7220_ErrStatus_RcvBadTidErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_7220_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_7220_ErrStatus_RcvEgrFullErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvBadVersionErr_LSB 0xB
+#define QIB_7220_ErrStatus_RcvBadVersionErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvIBFlowErr_LSB 0xA
+#define QIB_7220_ErrStatus_RcvIBFlowErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvEBPErr_LSB 0x9
+#define QIB_7220_ErrStatus_RcvEBPErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_7220_ErrStatus_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_7220_ErrStatus_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvShortPktLenErr_LSB 0x6
+#define QIB_7220_ErrStatus_RcvShortPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvLongPktLenErr_LSB 0x5
+#define QIB_7220_ErrStatus_RcvLongPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvMaxPktLenErr_LSB 0x4
+#define QIB_7220_ErrStatus_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvMinPktLenErr_LSB 0x3
+#define QIB_7220_ErrStatus_RcvMinPktLenErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvICRCErr_LSB 0x2
+#define QIB_7220_ErrStatus_RcvICRCErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvVCRCErr_LSB 0x1
+#define QIB_7220_ErrStatus_RcvVCRCErr_RMASK 0x1
+#define QIB_7220_ErrStatus_RcvFormatErr_LSB 0x0
+#define QIB_7220_ErrStatus_RcvFormatErr_RMASK 0x1
+
+#define QIB_7220_ErrClear_OFFS 0x90
+#define QIB_7220_ErrClear_Reserved_LSB 0x36
+#define QIB_7220_ErrClear_Reserved_RMASK 0x3FF
+#define QIB_7220_ErrClear_InvalidEEPCmdErrClear_LSB 0x35
+#define QIB_7220_ErrClear_InvalidEEPCmdErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDescAddrMisalignErrClear_LSB 0x34
+#define QIB_7220_ErrClear_SDmaDescAddrMisalignErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_HardwareErrClear_LSB 0x33
+#define QIB_7220_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_ResetNegatedClear_LSB 0x32
+#define QIB_7220_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_7220_ErrClear_InvalidAddrErrClear_LSB 0x31
+#define QIB_7220_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_IBStatusChangedClear_LSB 0x30
+#define QIB_7220_ErrClear_IBStatusChangedClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaUnexpDataErrClear_LSB 0x2F
+#define QIB_7220_ErrClear_SDmaUnexpDataErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaMissingDwErrClear_LSB 0x2E
+#define QIB_7220_ErrClear_SDmaMissingDwErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDwEnErrClear_LSB 0x2D
+#define QIB_7220_ErrClear_SDmaDwEnErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaRpyTagErrClear_LSB 0x2C
+#define QIB_7220_ErrClear_SDmaRpyTagErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDma1stDescErrClear_LSB 0x2B
+#define QIB_7220_ErrClear_SDma1stDescErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaBaseErrClear_LSB 0x2A
+#define QIB_7220_ErrClear_SDmaBaseErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaTailOutOfBoundErrClear_LSB 0x29
+#define QIB_7220_ErrClear_SDmaTailOutOfBoundErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaOutOfBoundErrClear_LSB 0x28
+#define QIB_7220_ErrClear_SDmaOutOfBoundErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaGenMismatchErrClear_LSB 0x27
+#define QIB_7220_ErrClear_SDmaGenMismatchErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendBufMisuseErrClear_LSB 0x26
+#define QIB_7220_ErrClear_SendBufMisuseErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_7220_ErrClear_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_7220_ErrClear_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendPioArmLaunchErrClear_LSB 0x23
+#define QIB_7220_ErrClear_SendPioArmLaunchErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_7220_ErrClear_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_7220_ErrClear_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendPktLenErrClear_LSB 0x20
+#define QIB_7220_ErrClear_SendPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendUnderRunErrClear_LSB 0x1F
+#define QIB_7220_ErrClear_SendUnderRunErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_7220_ErrClear_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_7220_ErrClear_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SDmaDisabledErrClear_LSB 0x1C
+#define QIB_7220_ErrClear_SDmaDisabledErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_SendSpecialTriggerErrClear_LSB 0x1B
+#define QIB_7220_ErrClear_SendSpecialTriggerErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_Reserved1_LSB 0x12
+#define QIB_7220_ErrClear_Reserved1_RMASK 0x1FF
+#define QIB_7220_ErrClear_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_7220_ErrClear_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrErrClear_LSB 0x10
+#define QIB_7220_ErrClear_RcvHdrErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrLenErrClear_LSB 0xF
+#define QIB_7220_ErrClear_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvBadTidErrClear_LSB 0xE
+#define QIB_7220_ErrClear_RcvBadTidErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_7220_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_7220_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvBadVersionErrClear_LSB 0xB
+#define QIB_7220_ErrClear_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvIBFlowErrClear_LSB 0xA
+#define QIB_7220_ErrClear_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvEBPErrClear_LSB 0x9
+#define QIB_7220_ErrClear_RcvEBPErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_7220_ErrClear_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_7220_ErrClear_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_7220_ErrClear_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_7220_ErrClear_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_7220_ErrClear_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_7220_ErrClear_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvICRCErrClear_LSB 0x2
+#define QIB_7220_ErrClear_RcvICRCErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvVCRCErrClear_LSB 0x1
+#define QIB_7220_ErrClear_RcvVCRCErrClear_RMASK 0x1
+#define QIB_7220_ErrClear_RcvFormatErrClear_LSB 0x0
+#define QIB_7220_ErrClear_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_7220_HwErrMask_OFFS 0x98
+#define QIB_7220_HwErrMask_IBCBusFromSPCParityErrMask_LSB 0x3F
+#define QIB_7220_HwErrMask_IBCBusFromSPCParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IBCBusToSPCParityErrMask_LSB 0x3E
+#define QIB_7220_HwErrMask_IBCBusToSPCParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Clk_uC_PLLNotLockedMask_LSB 0x3D
+#define QIB_7220_HwErrMask_Clk_uC_PLLNotLockedMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IBSerdesPClkNotDetectMask_LSB 0x3C
+#define QIB_7220_HwErrMask_IBSerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ3PClkNotDetectMask_LSB 0x3B
+#define QIB_7220_HwErrMask_PCIESerdesQ3PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ2PClkNotDetectMask_LSB 0x3A
+#define QIB_7220_HwErrMask_PCIESerdesQ2PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ1PClkNotDetectMask_LSB 0x39
+#define QIB_7220_HwErrMask_PCIESerdesQ1PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIESerdesQ0PClkNotDetectMask_LSB 0x38
+#define QIB_7220_HwErrMask_PCIESerdesQ0PClkNotDetectMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved_LSB 0x37
+#define QIB_7220_HwErrMask_Reserved_RMASK 0x1
+#define QIB_7220_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_7220_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved1_LSB 0x33
+#define QIB_7220_HwErrMask_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrMask_RXEMemParityErrMask_LSB 0x2C
+#define QIB_7220_HwErrMask_RXEMemParityErrMask_RMASK 0x7F
+#define QIB_7220_HwErrMask_TXEMemParityErrMask_LSB 0x28
+#define QIB_7220_HwErrMask_TXEMemParityErrMask_RMASK 0xF
+#define QIB_7220_HwErrMask_DDSRXEQMemoryParityErrMask_LSB 0x27
+#define QIB_7220_HwErrMask_DDSRXEQMemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_IB_uC_MemoryParityErrMask_LSB 0x26
+#define QIB_7220_HwErrMask_IB_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIEOct1_uC_MemoryParityErrMask_LSB 0x25
+#define QIB_7220_HwErrMask_PCIEOct1_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PCIEOct0_uC_MemoryParityErrMask_LSB 0x24
+#define QIB_7220_HwErrMask_PCIEOct0_uC_MemoryParityErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved2_LSB 0x22
+#define QIB_7220_HwErrMask_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_7220_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_7220_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_7220_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_7220_HwErrMask_PoisonedTLPMask_LSB 0x1D
+#define QIB_7220_HwErrMask_PoisonedTLPMask_RMASK 0x1
+#define QIB_7220_HwErrMask_SDmaMemReadErrMask_LSB 0x1C
+#define QIB_7220_HwErrMask_SDmaMemReadErrMask_RMASK 0x1
+#define QIB_7220_HwErrMask_Reserved3_LSB 0x8
+#define QIB_7220_HwErrMask_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrMask_PCIeMemParityErrMask_LSB 0x0
+#define QIB_7220_HwErrMask_PCIeMemParityErrMask_RMASK 0xFF
+
+#define QIB_7220_HwErrStatus_OFFS 0xA0
+#define QIB_7220_HwErrStatus_IBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_7220_HwErrStatus_IBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_IBCBusToSPCParityErr_LSB 0x3E
+#define QIB_7220_HwErrStatus_IBCBusToSPCParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Clk_uC_PLLNotLocked_LSB 0x3D
+#define QIB_7220_HwErrStatus_Clk_uC_PLLNotLocked_RMASK 0x1
+#define QIB_7220_HwErrStatus_IBSerdesPClkNotDetect_LSB 0x3C
+#define QIB_7220_HwErrStatus_IBSerdesPClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ3PClkNotDetect_LSB 0x3B
+#define QIB_7220_HwErrStatus_PCIESerdesQ3PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ2PClkNotDetect_LSB 0x3A
+#define QIB_7220_HwErrStatus_PCIESerdesQ2PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ1PClkNotDetect_LSB 0x39
+#define QIB_7220_HwErrStatus_PCIESerdesQ1PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIESerdesQ0PClkNotDetect_LSB 0x38
+#define QIB_7220_HwErrStatus_PCIESerdesQ0PClkNotDetect_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved_LSB 0x37
+#define QIB_7220_HwErrStatus_Reserved_RMASK 0x1
+#define QIB_7220_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_7220_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved1_LSB 0x33
+#define QIB_7220_HwErrStatus_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrStatus_RXEMemParity_LSB 0x2C
+#define QIB_7220_HwErrStatus_RXEMemParity_RMASK 0x7F
+#define QIB_7220_HwErrStatus_TXEMemParity_LSB 0x28
+#define QIB_7220_HwErrStatus_TXEMemParity_RMASK 0xF
+#define QIB_7220_HwErrStatus_DDSRXEQMemoryParityErr_LSB 0x27
+#define QIB_7220_HwErrStatus_DDSRXEQMemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_IB_uC_MemoryParityErr_LSB 0x26
+#define QIB_7220_HwErrStatus_IB_uC_MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct1MemoryParityErr_LSB 0x25
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct1MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct0MemoryParityErr_LSB 0x24
+#define QIB_7220_HwErrStatus_PCIE_uC_Oct0MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved2_LSB 0x22
+#define QIB_7220_HwErrStatus_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_7220_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_7220_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_7220_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_7220_HwErrStatus_PoisenedTLP_LSB 0x1D
+#define QIB_7220_HwErrStatus_PoisenedTLP_RMASK 0x1
+#define QIB_7220_HwErrStatus_SDmaMemReadErr_LSB 0x1C
+#define QIB_7220_HwErrStatus_SDmaMemReadErr_RMASK 0x1
+#define QIB_7220_HwErrStatus_Reserved3_LSB 0x8
+#define QIB_7220_HwErrStatus_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrStatus_PCIeMemParity_LSB 0x0
+#define QIB_7220_HwErrStatus_PCIeMemParity_RMASK 0xFF
+
+#define QIB_7220_HwErrClear_OFFS 0xA8
+#define QIB_7220_HwErrClear_IBCBusFromSPCParityErrClear_LSB 0x3F
+#define QIB_7220_HwErrClear_IBCBusFromSPCParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IBCBusToSPCparityErrClear_LSB 0x3E
+#define QIB_7220_HwErrClear_IBCBusToSPCparityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Clk_uC_PLLNotLockedClear_LSB 0x3D
+#define QIB_7220_HwErrClear_Clk_uC_PLLNotLockedClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IBSerdesPClkNotDetectClear_LSB 0x3C
+#define QIB_7220_HwErrClear_IBSerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ3PClkNotDetectClear_LSB 0x3B
+#define QIB_7220_HwErrClear_PCIESerdesQ3PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ2PClkNotDetectClear_LSB 0x3A
+#define QIB_7220_HwErrClear_PCIESerdesQ2PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ1PClkNotDetectClear_LSB 0x39
+#define QIB_7220_HwErrClear_PCIESerdesQ1PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIESerdesQ0PClkNotDetectClear_LSB 0x38
+#define QIB_7220_HwErrClear_PCIESerdesQ0PClkNotDetectClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved_LSB 0x37
+#define QIB_7220_HwErrClear_Reserved_RMASK 0x1
+#define QIB_7220_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_7220_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved1_LSB 0x33
+#define QIB_7220_HwErrClear_Reserved1_RMASK 0x7
+#define QIB_7220_HwErrClear_RXEMemParityClear_LSB 0x2C
+#define QIB_7220_HwErrClear_RXEMemParityClear_RMASK 0x7F
+#define QIB_7220_HwErrClear_TXEMemParityClear_LSB 0x28
+#define QIB_7220_HwErrClear_TXEMemParityClear_RMASK 0xF
+#define QIB_7220_HwErrClear_DDSRXEQMemoryParityErrClear_LSB 0x27
+#define QIB_7220_HwErrClear_DDSRXEQMemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_IB_uC_MemoryParityErrClear_LSB 0x26
+#define QIB_7220_HwErrClear_IB_uC_MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIE_uC_Oct1MemoryParityErrClear_LSB 0x25
+#define QIB_7220_HwErrClear_PCIE_uC_Oct1MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PCIE_uC_Oct0MemoryParityErrClear_LSB 0x24
+#define QIB_7220_HwErrClear_PCIE_uC_Oct0MemoryParityErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved2_LSB 0x22
+#define QIB_7220_HwErrClear_Reserved2_RMASK 0x3
+#define QIB_7220_HwErrClear_PCIeBusParityClr_LSB 0x1F
+#define QIB_7220_HwErrClear_PCIeBusParityClr_RMASK 0x7
+#define QIB_7220_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_7220_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_7220_HwErrClear_PoisonedTLPClear_LSB 0x1D
+#define QIB_7220_HwErrClear_PoisonedTLPClear_RMASK 0x1
+#define QIB_7220_HwErrClear_SDmaMemReadErrClear_LSB 0x1C
+#define QIB_7220_HwErrClear_SDmaMemReadErrClear_RMASK 0x1
+#define QIB_7220_HwErrClear_Reserved3_LSB 0x8
+#define QIB_7220_HwErrClear_Reserved3_RMASK 0xFFFFF
+#define QIB_7220_HwErrClear_PCIeMemParityClr_LSB 0x0
+#define QIB_7220_HwErrClear_PCIeMemParityClr_RMASK 0xFF
+
+#define QIB_7220_HwDiagCtrl_OFFS 0xB0
+#define QIB_7220_HwDiagCtrl_ForceIBCBusFromSPCParityErr_LSB 0x3F
+#define QIB_7220_HwDiagCtrl_ForceIBCBusFromSPCParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForceIBCBusToSPCParityErr_LSB 0x3E
+#define QIB_7220_HwDiagCtrl_ForceIBCBusToSPCParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_7220_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_7220_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_Reserved_LSB 0x33
+#define QIB_7220_HwDiagCtrl_Reserved_RMASK 0x1FF
+#define QIB_7220_HwDiagCtrl_ForceRxMemParityErr_LSB 0x2C
+#define QIB_7220_HwDiagCtrl_ForceRxMemParityErr_RMASK 0x7F
+#define QIB_7220_HwDiagCtrl_ForceTxMemparityErr_LSB 0x28
+#define QIB_7220_HwDiagCtrl_ForceTxMemparityErr_RMASK 0xF
+#define QIB_7220_HwDiagCtrl_ForceDDSRXEQMemoryParityErr_LSB 0x27
+#define QIB_7220_HwDiagCtrl_ForceDDSRXEQMemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForceIB_uC_MemoryParityErr_LSB 0x26
+#define QIB_7220_HwDiagCtrl_ForceIB_uC_MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct1MemoryParityErr_LSB 0x25
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct1MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct0MemoryParityErr_LSB 0x24
+#define QIB_7220_HwDiagCtrl_ForcePCIE_uC_Oct0MemoryParityErr_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_Reserved1_LSB 0x23
+#define QIB_7220_HwDiagCtrl_Reserved1_RMASK 0x1
+#define QIB_7220_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_7220_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_7220_HwDiagCtrl_Reserved2_LSB 0x8
+#define QIB_7220_HwDiagCtrl_Reserved2_RMASK 0x7FFFFF
+#define QIB_7220_HwDiagCtrl_forcePCIeMemParity_LSB 0x0
+#define QIB_7220_HwDiagCtrl_forcePCIeMemParity_RMASK 0xFF
+
+#define QIB_7220_REG_0000B8_OFFS 0xB8
+
+#define QIB_7220_IBCStatus_OFFS 0xC0
+#define QIB_7220_IBCStatus_TxCreditOk_LSB 0x1F
+#define QIB_7220_IBCStatus_TxCreditOk_RMASK 0x1
+#define QIB_7220_IBCStatus_TxReady_LSB 0x1E
+#define QIB_7220_IBCStatus_TxReady_RMASK 0x1
+#define QIB_7220_IBCStatus_Reserved_LSB 0xE
+#define QIB_7220_IBCStatus_Reserved_RMASK 0xFFFF
+#define QIB_7220_IBCStatus_IBTxLaneReversed_LSB 0xD
+#define QIB_7220_IBCStatus_IBTxLaneReversed_RMASK 0x1
+#define QIB_7220_IBCStatus_IBRxLaneReversed_LSB 0xC
+#define QIB_7220_IBCStatus_IBRxLaneReversed_RMASK 0x1
+#define QIB_7220_IBCStatus_IB_SERDES_TRIM_DONE_LSB 0xB
+#define QIB_7220_IBCStatus_IB_SERDES_TRIM_DONE_RMASK 0x1
+#define QIB_7220_IBCStatus_DDS_RXEQ_FAIL_LSB 0xA
+#define QIB_7220_IBCStatus_DDS_RXEQ_FAIL_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkWidthActive_LSB 0x9
+#define QIB_7220_IBCStatus_LinkWidthActive_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkSpeedActive_LSB 0x8
+#define QIB_7220_IBCStatus_LinkSpeedActive_RMASK 0x1
+#define QIB_7220_IBCStatus_LinkState_LSB 0x5
+#define QIB_7220_IBCStatus_LinkState_RMASK 0x7
+#define QIB_7220_IBCStatus_LinkTrainingState_LSB 0x0
+#define QIB_7220_IBCStatus_LinkTrainingState_RMASK 0x1F
+
+#define QIB_7220_IBCCtrl_OFFS 0xC8
+#define QIB_7220_IBCCtrl_Loopback_LSB 0x3F
+#define QIB_7220_IBCCtrl_Loopback_RMASK 0x1
+#define QIB_7220_IBCCtrl_LinkDownDefaultState_LSB 0x3E
+#define QIB_7220_IBCCtrl_LinkDownDefaultState_RMASK 0x1
+#define QIB_7220_IBCCtrl_Reserved_LSB 0x2B
+#define QIB_7220_IBCCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_7220_IBCCtrl_CreditScale_LSB 0x28
+#define QIB_7220_IBCCtrl_CreditScale_RMASK 0x7
+#define QIB_7220_IBCCtrl_OverrunThreshold_LSB 0x24
+#define QIB_7220_IBCCtrl_OverrunThreshold_RMASK 0xF
+#define QIB_7220_IBCCtrl_PhyerrThreshold_LSB 0x20
+#define QIB_7220_IBCCtrl_PhyerrThreshold_RMASK 0xF
+#define QIB_7220_IBCCtrl_MaxPktLen_LSB 0x15
+#define QIB_7220_IBCCtrl_MaxPktLen_RMASK 0x7FF
+#define QIB_7220_IBCCtrl_LinkCmd_LSB 0x13
+#define QIB_7220_IBCCtrl_LinkCmd_RMASK 0x3
+#define QIB_7220_IBCCtrl_LinkInitCmd_LSB 0x10
+#define QIB_7220_IBCCtrl_LinkInitCmd_RMASK 0x7
+#define QIB_7220_IBCCtrl_FlowCtrlWaterMark_LSB 0x8
+#define QIB_7220_IBCCtrl_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_7220_IBCCtrl_FlowCtrlPeriod_LSB 0x0
+#define QIB_7220_IBCCtrl_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_7220_EXTStatus_OFFS 0xD0
+#define QIB_7220_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_7220_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_Reserved_LSB 0x20
+#define QIB_7220_EXTStatus_Reserved_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_Reserved1_LSB 0x10
+#define QIB_7220_EXTStatus_Reserved1_RMASK 0xFFFF
+#define QIB_7220_EXTStatus_MemBISTDisabled_LSB 0xF
+#define QIB_7220_EXTStatus_MemBISTDisabled_RMASK 0x1
+#define QIB_7220_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_7220_EXTStatus_MemBISTEndTest_RMASK 0x1
+#define QIB_7220_EXTStatus_Reserved2_LSB 0x0
+#define QIB_7220_EXTStatus_Reserved2_RMASK 0x3FFF
+
+#define QIB_7220_EXTCtrl_OFFS 0xD8
+#define QIB_7220_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_7220_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_7220_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_7220_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_7220_EXTCtrl_Reserved_LSB 0x4
+#define QIB_7220_EXTCtrl_Reserved_RMASK 0xFFFFFFF
+#define QIB_7220_EXTCtrl_LEDPriPortGreenOn_LSB 0x3
+#define QIB_7220_EXTCtrl_LEDPriPortGreenOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDPriPortYellowOn_LSB 0x2
+#define QIB_7220_EXTCtrl_LEDPriPortYellowOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDGblOkGreenOn_LSB 0x1
+#define QIB_7220_EXTCtrl_LEDGblOkGreenOn_RMASK 0x1
+#define QIB_7220_EXTCtrl_LEDGblErrRedOff_LSB 0x0
+#define QIB_7220_EXTCtrl_LEDGblErrRedOff_RMASK 0x1
+
+#define QIB_7220_GPIOOut_OFFS 0xE0
+
+#define QIB_7220_GPIOMask_OFFS 0xE8
+
+#define QIB_7220_GPIOStatus_OFFS 0xF0
+
+#define QIB_7220_GPIOClear_OFFS 0xF8
+
+#define QIB_7220_RcvCtrl_OFFS 0x100
+#define QIB_7220_RcvCtrl_Reserved_LSB 0x27
+#define QIB_7220_RcvCtrl_Reserved_RMASK 0x1FFFFFF
+#define QIB_7220_RcvCtrl_RcvQPMapEnable_LSB 0x26
+#define QIB_7220_RcvCtrl_RcvQPMapEnable_RMASK 0x1
+#define QIB_7220_RcvCtrl_PortCfg_LSB 0x24
+#define QIB_7220_RcvCtrl_PortCfg_RMASK 0x3
+#define QIB_7220_RcvCtrl_TailUpd_LSB 0x23
+#define QIB_7220_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_7220_RcvCtrl_RcvPartitionKeyDisable_LSB 0x22
+#define QIB_7220_RcvCtrl_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_7220_RcvCtrl_IntrAvail_LSB 0x11
+#define QIB_7220_RcvCtrl_IntrAvail_RMASK 0x1FFFF
+#define QIB_7220_RcvCtrl_PortEnable_LSB 0x0
+#define QIB_7220_RcvCtrl_PortEnable_RMASK 0x1FFFF
+
+#define QIB_7220_RcvBTHQP_OFFS 0x108
+#define QIB_7220_RcvBTHQP_Reserved_LSB 0x18
+#define QIB_7220_RcvBTHQP_Reserved_RMASK 0xFF
+#define QIB_7220_RcvBTHQP_RcvBTHQP_LSB 0x0
+#define QIB_7220_RcvBTHQP_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_7220_RcvHdrSize_OFFS 0x110
+
+#define QIB_7220_RcvHdrCnt_OFFS 0x118
+
+#define QIB_7220_RcvHdrEntSize_OFFS 0x120
+
+#define QIB_7220_RcvTIDBase_OFFS 0x128
+
+#define QIB_7220_RcvTIDCnt_OFFS 0x130
+
+#define QIB_7220_RcvEgrBase_OFFS 0x138
+
+#define QIB_7220_RcvEgrCnt_OFFS 0x140
+
+#define QIB_7220_RcvBufBase_OFFS 0x148
+
+#define QIB_7220_RcvBufSize_OFFS 0x150
+
+#define QIB_7220_RxIntMemBase_OFFS 0x158
+
+#define QIB_7220_RxIntMemSize_OFFS 0x160
+
+#define QIB_7220_RcvPartitionKey_OFFS 0x168
+
+#define QIB_7220_RcvQPMulticastPort_OFFS 0x170
+#define QIB_7220_RcvQPMulticastPort_Reserved_LSB 0x5
+#define QIB_7220_RcvQPMulticastPort_Reserved_RMASK 0x7FFFFFFFFFFFFFF
+#define QIB_7220_RcvQPMulticastPort_RcvQpMcPort_LSB 0x0
+#define QIB_7220_RcvQPMulticastPort_RcvQpMcPort_RMASK 0x1F
+
+#define QIB_7220_RcvPktLEDCnt_OFFS 0x178
+#define QIB_7220_RcvPktLEDCnt_ONperiod_LSB 0x20
+#define QIB_7220_RcvPktLEDCnt_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_7220_RcvPktLEDCnt_OFFperiod_LSB 0x0
+#define QIB_7220_RcvPktLEDCnt_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_7220_IBCDDRCtrl_OFFS 0x180
+#define QIB_7220_IBCDDRCtrl_IB_DLID_MASK_LSB 0x30
+#define QIB_7220_IBCDDRCtrl_IB_DLID_MASK_RMASK 0xFFFF
+#define QIB_7220_IBCDDRCtrl_IB_DLID_LSB 0x20
+#define QIB_7220_IBCDDRCtrl_IB_DLID_RMASK 0xFFFF
+#define QIB_7220_IBCDDRCtrl_Reserved_LSB 0x1B
+#define QIB_7220_IBCDDRCtrl_Reserved_RMASK 0x1F
+#define QIB_7220_IBCDDRCtrl_HRTBT_REQ_LSB 0x1A
+#define QIB_7220_IBCDDRCtrl_HRTBT_REQ_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_HRTBT_PORT_LSB 0x12
+#define QIB_7220_IBCDDRCtrl_HRTBT_PORT_RMASK 0xFF
+#define QIB_7220_IBCDDRCtrl_HRTBT_AUTO_LSB 0x11
+#define QIB_7220_IBCDDRCtrl_HRTBT_AUTO_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_HRTBT_ENB_LSB 0x10
+#define QIB_7220_IBCDDRCtrl_HRTBT_ENB_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_DDS_LSB 0xC
+#define QIB_7220_IBCDDRCtrl_SD_DDS_RMASK 0xF
+#define QIB_7220_IBCDDRCtrl_SD_DDSV_LSB 0xB
+#define QIB_7220_IBCDDRCtrl_SD_DDSV_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_ADD_ENB_LSB 0xA
+#define QIB_7220_IBCDDRCtrl_SD_ADD_ENB_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_RX_EQUAL_ENABLE_LSB 0x9
+#define QIB_7220_IBCDDRCtrl_SD_RX_EQUAL_ENABLE_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_LANE_REV_SUPPORTED_LSB 0x8
+#define QIB_7220_IBCDDRCtrl_IB_LANE_REV_SUPPORTED_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_POLARITY_REV_SUPP_LSB 0x7
+#define QIB_7220_IBCDDRCtrl_IB_POLARITY_REV_SUPP_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_NUM_CHANNELS_LSB 0x5
+#define QIB_7220_IBCDDRCtrl_IB_NUM_CHANNELS_RMASK 0x3
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_QDR_LSB 0x4
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_QDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_DDR_LSB 0x3
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_DDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_SDR_LSB 0x2
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_SDR_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_LSB 0x1
+#define QIB_7220_IBCDDRCtrl_SD_SPEED_RMASK 0x1
+#define QIB_7220_IBCDDRCtrl_IB_ENHANCED_MODE_LSB 0x0
+#define QIB_7220_IBCDDRCtrl_IB_ENHANCED_MODE_RMASK 0x1
+
+#define QIB_7220_HRTBT_GUID_OFFS 0x188
+
+#define QIB_7220_IBCDDRCtrl2_OFFS 0x1A0
+#define QIB_7220_IBCDDRCtrl2_IB_BACK_PORCH_LSB 0x5
+#define QIB_7220_IBCDDRCtrl2_IB_BACK_PORCH_RMASK 0x1F
+#define QIB_7220_IBCDDRCtrl2_IB_FRONT_PORCH_LSB 0x0
+#define QIB_7220_IBCDDRCtrl2_IB_FRONT_PORCH_RMASK 0x1F
+
+#define QIB_7220_IBCDDRStatus_OFFS 0x1A8
+#define QIB_7220_IBCDDRStatus_heartbeat_timed_out_LSB 0x24
+#define QIB_7220_IBCDDRStatus_heartbeat_timed_out_RMASK 0x1
+#define QIB_7220_IBCDDRStatus_heartbeat_crosstalk_LSB 0x20
+#define QIB_7220_IBCDDRStatus_heartbeat_crosstalk_RMASK 0xF
+#define QIB_7220_IBCDDRStatus_RxEqLocalDevice_LSB 0x1E
+#define QIB_7220_IBCDDRStatus_RxEqLocalDevice_RMASK 0x3
+#define QIB_7220_IBCDDRStatus_ReqDDSLocalFromRmt_LSB 0x1A
+#define QIB_7220_IBCDDRStatus_ReqDDSLocalFromRmt_RMASK 0xF
+#define QIB_7220_IBCDDRStatus_LinkRoundTripLatency_LSB 0x0
+#define QIB_7220_IBCDDRStatus_LinkRoundTripLatency_RMASK 0x3FFFFFF
+
+#define QIB_7220_JIntReload_OFFS 0x1B0
+#define QIB_7220_JIntReload_J_limit_reload_LSB 0x10
+#define QIB_7220_JIntReload_J_limit_reload_RMASK 0xFFFF
+#define QIB_7220_JIntReload_J_reload_LSB 0x0
+#define QIB_7220_JIntReload_J_reload_RMASK 0xFFFF
+
+#define QIB_7220_IBNCModeCtrl_OFFS 0x1B8
+#define QIB_7220_IBNCModeCtrl_Reserved_LSB 0x1A
+#define QIB_7220_IBNCModeCtrl_Reserved_RMASK 0x3FFFFFFFFF
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS2_LSB 0x11
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS2_RMASK 0x1FF
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS1_LSB 0x8
+#define QIB_7220_IBNCModeCtrl_TSMCode_TS1_RMASK 0x1FF
+#define QIB_7220_IBNCModeCtrl_Reserved1_LSB 0x3
+#define QIB_7220_IBNCModeCtrl_Reserved1_RMASK 0x1F
+#define QIB_7220_IBNCModeCtrl_TSMEnable_ignore_TSM_on_rx_LSB 0x2
+#define QIB_7220_IBNCModeCtrl_TSMEnable_ignore_TSM_on_rx_RMASK 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS2_LSB 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS2_RMASK 0x1
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS1_LSB 0x0
+#define QIB_7220_IBNCModeCtrl_TSMEnable_send_TS1_RMASK 0x1
+
+#define QIB_7220_SendCtrl_OFFS 0x1C0
+#define QIB_7220_SendCtrl_Disarm_LSB 0x1F
+#define QIB_7220_SendCtrl_Disarm_RMASK 0x1
+#define QIB_7220_SendCtrl_Reserved_LSB 0x1D
+#define QIB_7220_SendCtrl_Reserved_RMASK 0x3
+#define QIB_7220_SendCtrl_AvailUpdThld_LSB 0x18
+#define QIB_7220_SendCtrl_AvailUpdThld_RMASK 0x1F
+#define QIB_7220_SendCtrl_DisarmPIOBuf_LSB 0x10
+#define QIB_7220_SendCtrl_DisarmPIOBuf_RMASK 0xFF
+#define QIB_7220_SendCtrl_Reserved1_LSB 0xD
+#define QIB_7220_SendCtrl_Reserved1_RMASK 0x7
+#define QIB_7220_SendCtrl_SDmaHalt_LSB 0xC
+#define QIB_7220_SendCtrl_SDmaHalt_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaEnable_LSB 0xB
+#define QIB_7220_SendCtrl_SDmaEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaSingleDescriptor_LSB 0xA
+#define QIB_7220_SendCtrl_SDmaSingleDescriptor_RMASK 0x1
+#define QIB_7220_SendCtrl_SDmaIntEnable_LSB 0x9
+#define QIB_7220_SendCtrl_SDmaIntEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_Reserved2_LSB 0x5
+#define QIB_7220_SendCtrl_Reserved2_RMASK 0xF
+#define QIB_7220_SendCtrl_SSpecialTriggerEn_LSB 0x4
+#define QIB_7220_SendCtrl_SSpecialTriggerEn_RMASK 0x1
+#define QIB_7220_SendCtrl_SPioEnable_LSB 0x3
+#define QIB_7220_SendCtrl_SPioEnable_RMASK 0x1
+#define QIB_7220_SendCtrl_SendBufAvailUpd_LSB 0x2
+#define QIB_7220_SendCtrl_SendBufAvailUpd_RMASK 0x1
+#define QIB_7220_SendCtrl_SendIntBufAvail_LSB 0x1
+#define QIB_7220_SendCtrl_SendIntBufAvail_RMASK 0x1
+#define QIB_7220_SendCtrl_Abort_LSB 0x0
+#define QIB_7220_SendCtrl_Abort_RMASK 0x1
+
+#define QIB_7220_SendBufBase_OFFS 0x1C8
+#define QIB_7220_SendBufBase_Reserved_LSB 0x35
+#define QIB_7220_SendBufBase_Reserved_RMASK 0x7FF
+#define QIB_7220_SendBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_7220_SendBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_7220_SendBufBase_Reserved1_LSB 0x15
+#define QIB_7220_SendBufBase_Reserved1_RMASK 0x7FF
+#define QIB_7220_SendBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_7220_SendBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_7220_SendBufSize_OFFS 0x1D0
+#define QIB_7220_SendBufSize_Reserved_LSB 0x2D
+#define QIB_7220_SendBufSize_Reserved_RMASK 0xFFFFF
+#define QIB_7220_SendBufSize_Size_LargePIO_LSB 0x20
+#define QIB_7220_SendBufSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_7220_SendBufSize_Reserved1_LSB 0xC
+#define QIB_7220_SendBufSize_Reserved1_RMASK 0xFFFFF
+#define QIB_7220_SendBufSize_Size_SmallPIO_LSB 0x0
+#define QIB_7220_SendBufSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_7220_SendBufCnt_OFFS 0x1D8
+#define QIB_7220_SendBufCnt_Reserved_LSB 0x24
+#define QIB_7220_SendBufCnt_Reserved_RMASK 0xFFFFFFF
+#define QIB_7220_SendBufCnt_Num_LargeBuffers_LSB 0x20
+#define QIB_7220_SendBufCnt_Num_LargeBuffers_RMASK 0xF
+#define QIB_7220_SendBufCnt_Reserved1_LSB 0x9
+#define QIB_7220_SendBufCnt_Reserved1_RMASK 0x7FFFFF
+#define QIB_7220_SendBufCnt_Num_SmallBuffers_LSB 0x0
+#define QIB_7220_SendBufCnt_Num_SmallBuffers_RMASK 0x1FF
+
+#define QIB_7220_SendBufAvailAddr_OFFS 0x1E0
+#define QIB_7220_SendBufAvailAddr_SendBufAvailAddr_LSB 0x6
+#define QIB_7220_SendBufAvailAddr_SendBufAvailAddr_RMASK 0x3FFFFFFFF
+#define QIB_7220_SendBufAvailAddr_Reserved_LSB 0x0
+#define QIB_7220_SendBufAvailAddr_Reserved_RMASK 0x3F
+
+#define QIB_7220_TxIntMemBase_OFFS 0x1E8
+
+#define QIB_7220_TxIntMemSize_OFFS 0x1F0
+
+#define QIB_7220_SendDmaBase_OFFS 0x1F8
+#define QIB_7220_SendDmaBase_Reserved_LSB 0x30
+#define QIB_7220_SendDmaBase_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaBase_SendDmaBase_LSB 0x0
+#define QIB_7220_SendDmaBase_SendDmaBase_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7220_SendDmaLenGen_OFFS 0x200
+#define QIB_7220_SendDmaLenGen_Reserved_LSB 0x13
+#define QIB_7220_SendDmaLenGen_Reserved_RMASK 0x1FFFFFFFFFFF
+#define QIB_7220_SendDmaLenGen_Generation_LSB 0x10
+#define QIB_7220_SendDmaLenGen_Generation_MSB 0x12
+#define QIB_7220_SendDmaLenGen_Generation_RMASK 0x7
+#define QIB_7220_SendDmaLenGen_Length_LSB 0x0
+#define QIB_7220_SendDmaLenGen_Length_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaTail_OFFS 0x208
+#define QIB_7220_SendDmaTail_Reserved_LSB 0x10
+#define QIB_7220_SendDmaTail_Reserved_RMASK 0xFFFFFFFFFFFF
+#define QIB_7220_SendDmaTail_SendDmaTail_LSB 0x0
+#define QIB_7220_SendDmaTail_SendDmaTail_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaHead_OFFS 0x210
+#define QIB_7220_SendDmaHead_Reserved_LSB 0x30
+#define QIB_7220_SendDmaHead_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_InternalSendDmaHead_LSB 0x20
+#define QIB_7220_SendDmaHead_InternalSendDmaHead_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_Reserved1_LSB 0x10
+#define QIB_7220_SendDmaHead_Reserved1_RMASK 0xFFFF
+#define QIB_7220_SendDmaHead_SendDmaHead_LSB 0x0
+#define QIB_7220_SendDmaHead_SendDmaHead_RMASK 0xFFFF
+
+#define QIB_7220_SendDmaHeadAddr_OFFS 0x218
+#define QIB_7220_SendDmaHeadAddr_Reserved_LSB 0x30
+#define QIB_7220_SendDmaHeadAddr_Reserved_RMASK 0xFFFF
+#define QIB_7220_SendDmaHeadAddr_SendDmaHeadAddr_LSB 0x0
+#define QIB_7220_SendDmaHeadAddr_SendDmaHeadAddr_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7220_SendDmaBufMask0_OFFS 0x220
+#define QIB_7220_SendDmaBufMask0_BufMask_63_0_LSB 0x0
+#define QIB_7220_SendDmaBufMask0_BufMask_63_0_RMASK 0x0
+
+#define QIB_7220_SendDmaStatus_OFFS 0x238
+#define QIB_7220_SendDmaStatus_ScoreBoardDrainInProg_LSB 0x3F
+#define QIB_7220_SendDmaStatus_ScoreBoardDrainInProg_RMASK 0x1
+#define QIB_7220_SendDmaStatus_AbortInProg_LSB 0x3E
+#define QIB_7220_SendDmaStatus_AbortInProg_RMASK 0x1
+#define QIB_7220_SendDmaStatus_InternalSDmaEnable_LSB 0x3D
+#define QIB_7220_SendDmaStatus_InternalSDmaEnable_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbDescIndex_13_0_LSB 0x2F
+#define QIB_7220_SendDmaStatus_ScbDescIndex_13_0_RMASK 0x3FFF
+#define QIB_7220_SendDmaStatus_RpyLowAddr_6_0_LSB 0x28
+#define QIB_7220_SendDmaStatus_RpyLowAddr_6_0_RMASK 0x7F
+#define QIB_7220_SendDmaStatus_RpyTag_7_0_LSB 0x20
+#define QIB_7220_SendDmaStatus_RpyTag_7_0_RMASK 0xFF
+#define QIB_7220_SendDmaStatus_ScbFull_LSB 0x1F
+#define QIB_7220_SendDmaStatus_ScbFull_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbEmpty_LSB 0x1E
+#define QIB_7220_SendDmaStatus_ScbEmpty_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbEntryValid_LSB 0x1D
+#define QIB_7220_SendDmaStatus_ScbEntryValid_RMASK 0x1
+#define QIB_7220_SendDmaStatus_ScbFetchDescFlag_LSB 0x1C
+#define QIB_7220_SendDmaStatus_ScbFetchDescFlag_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoReadyToGo_LSB 0x1B
+#define QIB_7220_SendDmaStatus_SplFifoReadyToGo_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoDisarmed_LSB 0x1A
+#define QIB_7220_SendDmaStatus_SplFifoDisarmed_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoEmpty_LSB 0x19
+#define QIB_7220_SendDmaStatus_SplFifoEmpty_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoFull_LSB 0x18
+#define QIB_7220_SendDmaStatus_SplFifoFull_RMASK 0x1
+#define QIB_7220_SendDmaStatus_SplFifoBufNum_LSB 0x10
+#define QIB_7220_SendDmaStatus_SplFifoBufNum_RMASK 0xFF
+#define QIB_7220_SendDmaStatus_SplFifoDescIndex_LSB 0x0
+#define QIB_7220_SendDmaStatus_SplFifoDescIndex_RMASK 0xFFFF
+
+#define QIB_7220_SendBufErr0_OFFS 0x240
+#define QIB_7220_SendBufErr0_SendBufErr_63_0_LSB 0x0
+#define QIB_7220_SendBufErr0_SendBufErr_63_0_RMASK 0x0
+
+#define QIB_7220_RcvHdrAddr0_OFFS 0x270
+#define QIB_7220_RcvHdrAddr0_RcvHdrAddr0_LSB 0x2
+#define QIB_7220_RcvHdrAddr0_RcvHdrAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_7220_RcvHdrAddr0_Reserved_LSB 0x0
+#define QIB_7220_RcvHdrAddr0_Reserved_RMASK 0x3
+
+#define QIB_7220_RcvHdrTailAddr0_OFFS 0x300
+#define QIB_7220_RcvHdrTailAddr0_RcvHdrTailAddr0_LSB 0x2
+#define QIB_7220_RcvHdrTailAddr0_RcvHdrTailAddr0_RMASK 0x3FFFFFFFFF
+#define QIB_7220_RcvHdrTailAddr0_Reserved_LSB 0x0
+#define QIB_7220_RcvHdrTailAddr0_Reserved_RMASK 0x3
+
+#define QIB_7220_ibsd_epb_access_ctrl_OFFS 0x3C0
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_granted_LSB 0x8
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_granted_RMASK 0x1
+#define QIB_7220_ibsd_epb_access_ctrl_Reserved_LSB 0x1
+#define QIB_7220_ibsd_epb_access_ctrl_Reserved_RMASK 0x7F
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_LSB 0x0
+#define QIB_7220_ibsd_epb_access_ctrl_sw_ib_epb_req_RMASK 0x1
+
+#define QIB_7220_ibsd_epb_transaction_reg_OFFS 0x3C8
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_rdy_LSB 0x1F
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_rdy_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_req_error_LSB 0x1E
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_req_error_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved_LSB 0x1D
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_mem_data_parity_LSB 0x1C
+#define QIB_7220_ibsd_epb_transaction_reg_mem_data_parity_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved1_LSB 0x1B
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved1_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_cs_LSB 0x19
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_cs_RMASK 0x3
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_read_write_LSB 0x18
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_read_write_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved2_LSB 0x17
+#define QIB_7220_ibsd_epb_transaction_reg_Reserved2_RMASK 0x1
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_address_LSB 0x8
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_address_RMASK 0x7FFF
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_data_LSB 0x0
+#define QIB_7220_ibsd_epb_transaction_reg_ib_epb_data_RMASK 0xFF
+
+#define QIB_7220_XGXSCfg_OFFS 0x3D8
+#define QIB_7220_XGXSCfg_sel_link_down_for_fctrl_lane_sync_reset_LSB 0x3F
+#define QIB_7220_XGXSCfg_sel_link_down_for_fctrl_lane_sync_reset_RMASK 0x1
+#define QIB_7220_XGXSCfg_Reserved_LSB 0x13
+#define QIB_7220_XGXSCfg_Reserved_RMASK 0xFFFFFFFFFFF
+#define QIB_7220_XGXSCfg_link_sync_mask_LSB 0x9
+#define QIB_7220_XGXSCfg_link_sync_mask_RMASK 0x3FF
+#define QIB_7220_XGXSCfg_Reserved1_LSB 0x3
+#define QIB_7220_XGXSCfg_Reserved1_RMASK 0x3F
+#define QIB_7220_XGXSCfg_xcv_reset_LSB 0x2
+#define QIB_7220_XGXSCfg_xcv_reset_RMASK 0x1
+#define QIB_7220_XGXSCfg_Reserved2_LSB 0x1
+#define QIB_7220_XGXSCfg_Reserved2_RMASK 0x1
+#define QIB_7220_XGXSCfg_tx_rx_reset_LSB 0x0
+#define QIB_7220_XGXSCfg_tx_rx_reset_RMASK 0x1
+
+#define QIB_7220_IBSerDesCtrl_OFFS 0x3E0
+#define QIB_7220_IBSerDesCtrl_Reserved_LSB 0x2D
+#define QIB_7220_IBSerDesCtrl_Reserved_RMASK 0x7FFFF
+#define QIB_7220_IBSerDesCtrl_INT_uC_LSB 0x2C
+#define QIB_7220_IBSerDesCtrl_INT_uC_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_CKSEL_uC_LSB 0x2A
+#define QIB_7220_IBSerDesCtrl_CKSEL_uC_RMASK 0x3
+#define QIB_7220_IBSerDesCtrl_PLLN_LSB 0x28
+#define QIB_7220_IBSerDesCtrl_PLLN_RMASK 0x3
+#define QIB_7220_IBSerDesCtrl_PLLM_LSB 0x25
+#define QIB_7220_IBSerDesCtrl_PLLM_RMASK 0x7
+#define QIB_7220_IBSerDesCtrl_TXOBPD_LSB 0x24
+#define QIB_7220_IBSerDesCtrl_TXOBPD_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_TWC_LSB 0x23
+#define QIB_7220_IBSerDesCtrl_TWC_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_RXIDLE_LSB 0x22
+#define QIB_7220_IBSerDesCtrl_RXIDLE_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_RXINV_LSB 0x21
+#define QIB_7220_IBSerDesCtrl_RXINV_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_TXINV_LSB 0x20
+#define QIB_7220_IBSerDesCtrl_TXINV_RMASK 0x1
+#define QIB_7220_IBSerDesCtrl_Reserved1_LSB 0x12
+#define QIB_7220_IBSerDesCtrl_Reserved1_RMASK 0x3FFF
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForRXEQ_LSB 0xD
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForRXEQ_RMASK 0x1F
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForDDS_LSB 0x8
+#define QIB_7220_IBSerDesCtrl_NumSerDesRegsToWrForDDS_RMASK 0x1F
+#define QIB_7220_IBSerDesCtrl_Reserved2_LSB 0x1
+#define QIB_7220_IBSerDesCtrl_Reserved2_RMASK 0x7F
+#define QIB_7220_IBSerDesCtrl_ResetIB_uC_Core_LSB 0x0
+#define QIB_7220_IBSerDesCtrl_ResetIB_uC_Core_RMASK 0x1
+
+#define QIB_7220_pciesd_epb_access_ctrl_OFFS 0x400
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_granted_LSB 0x8
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_granted_RMASK 0x1
+#define QIB_7220_pciesd_epb_access_ctrl_Reserved_LSB 0x3
+#define QIB_7220_pciesd_epb_access_ctrl_Reserved_RMASK 0x1F
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcieepb_star_en_LSB 0x1
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcieepb_star_en_RMASK 0x3
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_LSB 0x0
+#define QIB_7220_pciesd_epb_access_ctrl_sw_pcie_epb_req_RMASK 0x1
+
+#define QIB_7220_pciesd_epb_transaction_reg_OFFS 0x408
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_rdy_LSB 0x1F
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_rdy_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_req_error_LSB 0x1E
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_req_error_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved_LSB 0x1D
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_mem_data_parity_LSB 0x1C
+#define QIB_7220_pciesd_epb_transaction_reg_mem_data_parity_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_cs_LSB 0x19
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_cs_RMASK 0x7
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_read_write_LSB 0x18
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_read_write_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved1_LSB 0x17
+#define QIB_7220_pciesd_epb_transaction_reg_Reserved1_RMASK 0x1
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_address_LSB 0x8
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_address_RMASK 0x7FFF
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_data_LSB 0x0
+#define QIB_7220_pciesd_epb_transaction_reg_pcie_epb_data_RMASK 0xFF
+
+#define QIB_7220_SerDes_DDSRXEQ0_OFFS 0x500
+#define QIB_7220_SerDes_DDSRXEQ0_reg_addr_LSB 0x4
+#define QIB_7220_SerDes_DDSRXEQ0_reg_addr_RMASK 0x3F
+#define QIB_7220_SerDes_DDSRXEQ0_element_num_LSB 0x0
+#define QIB_7220_SerDes_DDSRXEQ0_element_num_RMASK 0xF
+
+#define QIB_7220_LBIntCnt_OFFS 0x13000
+
+#define QIB_7220_LBFlowStallCnt_OFFS 0x13008
+
+#define QIB_7220_TxSDmaDescCnt_OFFS 0x13010
+
+#define QIB_7220_TxUnsupVLErrCnt_OFFS 0x13018
+
+#define QIB_7220_TxDataPktCnt_OFFS 0x13020
+
+#define QIB_7220_TxFlowPktCnt_OFFS 0x13028
+
+#define QIB_7220_TxDwordCnt_OFFS 0x13030
+
+#define QIB_7220_TxLenErrCnt_OFFS 0x13038
+
+#define QIB_7220_TxMaxMinLenErrCnt_OFFS 0x13040
+
+#define QIB_7220_TxUnderrunCnt_OFFS 0x13048
+
+#define QIB_7220_TxFlowStallCnt_OFFS 0x13050
+
+#define QIB_7220_TxDroppedPktCnt_OFFS 0x13058
+
+#define QIB_7220_RxDroppedPktCnt_OFFS 0x13060
+
+#define QIB_7220_RxDataPktCnt_OFFS 0x13068
+
+#define QIB_7220_RxFlowPktCnt_OFFS 0x13070
+
+#define QIB_7220_RxDwordCnt_OFFS 0x13078
+
+#define QIB_7220_RxLenErrCnt_OFFS 0x13080
+
+#define QIB_7220_RxMaxMinLenErrCnt_OFFS 0x13088
+
+#define QIB_7220_RxICRCErrCnt_OFFS 0x13090
+
+#define QIB_7220_RxVCRCErrCnt_OFFS 0x13098
+
+#define QIB_7220_RxFlowCtrlViolCnt_OFFS 0x130A0
+
+#define QIB_7220_RxVersionErrCnt_OFFS 0x130A8
+
+#define QIB_7220_RxLinkMalformCnt_OFFS 0x130B0
+
+#define QIB_7220_RxEBPCnt_OFFS 0x130B8
+
+#define QIB_7220_RxLPCRCErrCnt_OFFS 0x130C0
+
+#define QIB_7220_RxBufOvflCnt_OFFS 0x130C8
+
+#define QIB_7220_RxTIDFullErrCnt_OFFS 0x130D0
+
+#define QIB_7220_RxTIDValidErrCnt_OFFS 0x130D8
+
+#define QIB_7220_RxPKeyMismatchCnt_OFFS 0x130E0
+
+#define QIB_7220_RxP0HdrEgrOvflCnt_OFFS 0x130E8
+
+#define QIB_7220_IBStatusChangeCnt_OFFS 0x13170
+
+#define QIB_7220_IBLinkErrRecoveryCnt_OFFS 0x13178
+
+#define QIB_7220_IBLinkDownedCnt_OFFS 0x13180
+
+#define QIB_7220_IBSymbolErrCnt_OFFS 0x13188
+
+#define QIB_7220_RxVL15DroppedPktCnt_OFFS 0x13190
+
+#define QIB_7220_RxOtherLocalPhyErrCnt_OFFS 0x13198
+
+#define QIB_7220_PcieRetryBufDiagQwordCnt_OFFS 0x131A0
+
+#define QIB_7220_ExcessBufferOvflCnt_OFFS 0x131A8
+
+#define QIB_7220_LocalLinkIntegrityErrCnt_OFFS 0x131B0
+
+#define QIB_7220_RxVlErrCnt_OFFS 0x131B8
+
+#define QIB_7220_RxDlidFltrCnt_OFFS 0x131C0
+
+#define QIB_7220_CNT_0131C8_OFFS 0x131C8
+
+#define QIB_7220_PSStat_OFFS 0x13200
+
+#define QIB_7220_PSStart_OFFS 0x13208
+
+#define QIB_7220_PSInterval_OFFS 0x13210
+
+#define QIB_7220_PSRcvDataCount_OFFS 0x13218
+
+#define QIB_7220_PSRcvPktsCount_OFFS 0x13220
+
+#define QIB_7220_PSXmitDataCount_OFFS 0x13228
+
+#define QIB_7220_PSXmitPktsCount_OFFS 0x13230
+
+#define QIB_7220_PSXmitWaitCount_OFFS 0x13238
+
+#define QIB_7220_CNT_013240_OFFS 0x13240
+
+#define QIB_7220_RcvEgrArray_OFFS 0x14000
+
+#define QIB_7220_MEM_038000_OFFS 0x38000
+
+#define QIB_7220_RcvTIDArray0_OFFS 0x53000
+
+#define QIB_7220_PIOLaunchFIFO_OFFS 0x64000
+
+#define QIB_7220_MEM_064480_OFFS 0x64480
+
+#define QIB_7220_SendPIOpbcCache_OFFS 0x64800
+
+#define QIB_7220_MEM_064C80_OFFS 0x64C80
+
+#define QIB_7220_PreLaunchFIFO_OFFS 0x65000
+
+#define QIB_7220_MEM_065080_OFFS 0x65080
+
+#define QIB_7220_ScoreBoard_OFFS 0x65400
+
+#define QIB_7220_MEM_065440_OFFS 0x65440
+
+#define QIB_7220_DescriptorFIFO_OFFS 0x65800
+
+#define QIB_7220_MEM_065880_OFFS 0x65880
+
+#define QIB_7220_RcvBuf1_OFFS 0x72000
+
+#define QIB_7220_MEM_074800_OFFS 0x74800
+
+#define QIB_7220_RcvBuf2_OFFS 0x75000
+
+#define QIB_7220_MEM_076400_OFFS 0x76400
+
+#define QIB_7220_RcvFlags_OFFS 0x77000
+
+#define QIB_7220_MEM_078400_OFFS 0x78400
+
+#define QIB_7220_RcvLookupBuf1_OFFS 0x79000
+
+#define QIB_7220_MEM_07A400_OFFS 0x7A400
+
+#define QIB_7220_RcvDMADatBuf_OFFS 0x7B000
+
+#define QIB_7220_RcvDMAHdrBuf_OFFS 0x7B800
+
+#define QIB_7220_MiscRXEIntMem_OFFS 0x7C000
+
+#define QIB_7220_MEM_07D400_OFFS 0x7D400
+
+#define QIB_7220_PCIERcvBuf_OFFS 0x80000
+
+#define QIB_7220_PCIERetryBuf_OFFS 0x84000
+
+#define QIB_7220_PCIERcvBufRdToWrAddr_OFFS 0x88000
+
+#define QIB_7220_PCIECplBuf_OFFS 0x90000
+
+#define QIB_7220_IBSerDesMappTable_OFFS 0x94000
+
+#define QIB_7220_MEM_095000_OFFS 0x95000
+
+#define QIB_7220_SendBuf0_MA_OFFS 0x100000
+
+#define QIB_7220_MEM_1A0000_OFFS 0x1A0000
diff --git a/drivers/infiniband/hw/qib/qib_7322_regs.h b/drivers/infiniband/hw/qib/qib_7322_regs.h
new file mode 100644
index 00000000000..32dc81ff8d4
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_7322_regs.h
@@ -0,0 +1,3163 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/* This file is mechanically generated from RTL. Any hand-edits will be lost! */
+
+#define QIB_7322_Revision_OFFS 0x0
+#define QIB_7322_Revision_DEF 0x0000000002010601
+#define QIB_7322_Revision_R_Simulator_LSB 0x3F
+#define QIB_7322_Revision_R_Simulator_MSB 0x3F
+#define QIB_7322_Revision_R_Simulator_RMASK 0x1
+#define QIB_7322_Revision_R_Emulation_LSB 0x3E
+#define QIB_7322_Revision_R_Emulation_MSB 0x3E
+#define QIB_7322_Revision_R_Emulation_RMASK 0x1
+#define QIB_7322_Revision_R_Emulation_Revcode_LSB 0x28
+#define QIB_7322_Revision_R_Emulation_Revcode_MSB 0x3D
+#define QIB_7322_Revision_R_Emulation_Revcode_RMASK 0x3FFFFF
+#define QIB_7322_Revision_BoardID_LSB 0x20
+#define QIB_7322_Revision_BoardID_MSB 0x27
+#define QIB_7322_Revision_BoardID_RMASK 0xFF
+#define QIB_7322_Revision_R_SW_LSB 0x18
+#define QIB_7322_Revision_R_SW_MSB 0x1F
+#define QIB_7322_Revision_R_SW_RMASK 0xFF
+#define QIB_7322_Revision_R_Arch_LSB 0x10
+#define QIB_7322_Revision_R_Arch_MSB 0x17
+#define QIB_7322_Revision_R_Arch_RMASK 0xFF
+#define QIB_7322_Revision_R_ChipRevMajor_LSB 0x8
+#define QIB_7322_Revision_R_ChipRevMajor_MSB 0xF
+#define QIB_7322_Revision_R_ChipRevMajor_RMASK 0xFF
+#define QIB_7322_Revision_R_ChipRevMinor_LSB 0x0
+#define QIB_7322_Revision_R_ChipRevMinor_MSB 0x7
+#define QIB_7322_Revision_R_ChipRevMinor_RMASK 0xFF
+
+#define QIB_7322_Control_OFFS 0x8
+#define QIB_7322_Control_DEF 0x0000000000000000
+#define QIB_7322_Control_PCIECplQDiagEn_LSB 0x6
+#define QIB_7322_Control_PCIECplQDiagEn_MSB 0x6
+#define QIB_7322_Control_PCIECplQDiagEn_RMASK 0x1
+#define QIB_7322_Control_PCIEPostQDiagEn_LSB 0x5
+#define QIB_7322_Control_PCIEPostQDiagEn_MSB 0x5
+#define QIB_7322_Control_PCIEPostQDiagEn_RMASK 0x1
+#define QIB_7322_Control_SDmaDescFetchPriorityEn_LSB 0x4
+#define QIB_7322_Control_SDmaDescFetchPriorityEn_MSB 0x4
+#define QIB_7322_Control_SDmaDescFetchPriorityEn_RMASK 0x1
+#define QIB_7322_Control_PCIERetryBufDiagEn_LSB 0x3
+#define QIB_7322_Control_PCIERetryBufDiagEn_MSB 0x3
+#define QIB_7322_Control_PCIERetryBufDiagEn_RMASK 0x1
+#define QIB_7322_Control_FreezeMode_LSB 0x1
+#define QIB_7322_Control_FreezeMode_MSB 0x1
+#define QIB_7322_Control_FreezeMode_RMASK 0x1
+#define QIB_7322_Control_SyncReset_LSB 0x0
+#define QIB_7322_Control_SyncReset_MSB 0x0
+#define QIB_7322_Control_SyncReset_RMASK 0x1
+
+#define QIB_7322_PageAlign_OFFS 0x10
+#define QIB_7322_PageAlign_DEF 0x0000000000001000
+
+#define QIB_7322_ContextCnt_OFFS 0x18
+#define QIB_7322_ContextCnt_DEF 0x0000000000000012
+
+#define QIB_7322_Scratch_OFFS 0x20
+#define QIB_7322_Scratch_DEF 0x0000000000000000
+
+#define QIB_7322_CntrRegBase_OFFS 0x28
+#define QIB_7322_CntrRegBase_DEF 0x0000000000011000
+
+#define QIB_7322_SendRegBase_OFFS 0x30
+#define QIB_7322_SendRegBase_DEF 0x0000000000003000
+
+#define QIB_7322_UserRegBase_OFFS 0x38
+#define QIB_7322_UserRegBase_DEF 0x0000000000200000
+
+#define QIB_7322_IntMask_OFFS 0x68
+#define QIB_7322_IntMask_DEF 0x0000000000000000
+#define QIB_7322_IntMask_SDmaIntMask_1_LSB 0x3F
+#define QIB_7322_IntMask_SDmaIntMask_1_MSB 0x3F
+#define QIB_7322_IntMask_SDmaIntMask_1_RMASK 0x1
+#define QIB_7322_IntMask_SDmaIntMask_0_LSB 0x3E
+#define QIB_7322_IntMask_SDmaIntMask_0_MSB 0x3E
+#define QIB_7322_IntMask_SDmaIntMask_0_RMASK 0x1
+#define QIB_7322_IntMask_SDmaProgressIntMask_1_LSB 0x3D
+#define QIB_7322_IntMask_SDmaProgressIntMask_1_MSB 0x3D
+#define QIB_7322_IntMask_SDmaProgressIntMask_1_RMASK 0x1
+#define QIB_7322_IntMask_SDmaProgressIntMask_0_LSB 0x3C
+#define QIB_7322_IntMask_SDmaProgressIntMask_0_MSB 0x3C
+#define QIB_7322_IntMask_SDmaProgressIntMask_0_RMASK 0x1
+#define QIB_7322_IntMask_SDmaIdleIntMask_1_LSB 0x3B
+#define QIB_7322_IntMask_SDmaIdleIntMask_1_MSB 0x3B
+#define QIB_7322_IntMask_SDmaIdleIntMask_1_RMASK 0x1
+#define QIB_7322_IntMask_SDmaIdleIntMask_0_LSB 0x3A
+#define QIB_7322_IntMask_SDmaIdleIntMask_0_MSB 0x3A
+#define QIB_7322_IntMask_SDmaIdleIntMask_0_RMASK 0x1
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_1_LSB 0x39
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_1_MSB 0x39
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_1_RMASK 0x1
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_0_LSB 0x38
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_0_MSB 0x38
+#define QIB_7322_IntMask_SDmaCleanupDoneMask_0_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg17IntMask_LSB 0x31
+#define QIB_7322_IntMask_RcvUrg17IntMask_MSB 0x31
+#define QIB_7322_IntMask_RcvUrg17IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg16IntMask_LSB 0x30
+#define QIB_7322_IntMask_RcvUrg16IntMask_MSB 0x30
+#define QIB_7322_IntMask_RcvUrg16IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg15IntMask_LSB 0x2F
+#define QIB_7322_IntMask_RcvUrg15IntMask_MSB 0x2F
+#define QIB_7322_IntMask_RcvUrg15IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg14IntMask_LSB 0x2E
+#define QIB_7322_IntMask_RcvUrg14IntMask_MSB 0x2E
+#define QIB_7322_IntMask_RcvUrg14IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg13IntMask_LSB 0x2D
+#define QIB_7322_IntMask_RcvUrg13IntMask_MSB 0x2D
+#define QIB_7322_IntMask_RcvUrg13IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg12IntMask_LSB 0x2C
+#define QIB_7322_IntMask_RcvUrg12IntMask_MSB 0x2C
+#define QIB_7322_IntMask_RcvUrg12IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg11IntMask_LSB 0x2B
+#define QIB_7322_IntMask_RcvUrg11IntMask_MSB 0x2B
+#define QIB_7322_IntMask_RcvUrg11IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg10IntMask_LSB 0x2A
+#define QIB_7322_IntMask_RcvUrg10IntMask_MSB 0x2A
+#define QIB_7322_IntMask_RcvUrg10IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg9IntMask_LSB 0x29
+#define QIB_7322_IntMask_RcvUrg9IntMask_MSB 0x29
+#define QIB_7322_IntMask_RcvUrg9IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg8IntMask_LSB 0x28
+#define QIB_7322_IntMask_RcvUrg8IntMask_MSB 0x28
+#define QIB_7322_IntMask_RcvUrg8IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg7IntMask_LSB 0x27
+#define QIB_7322_IntMask_RcvUrg7IntMask_MSB 0x27
+#define QIB_7322_IntMask_RcvUrg7IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg6IntMask_LSB 0x26
+#define QIB_7322_IntMask_RcvUrg6IntMask_MSB 0x26
+#define QIB_7322_IntMask_RcvUrg6IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg5IntMask_LSB 0x25
+#define QIB_7322_IntMask_RcvUrg5IntMask_MSB 0x25
+#define QIB_7322_IntMask_RcvUrg5IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg4IntMask_LSB 0x24
+#define QIB_7322_IntMask_RcvUrg4IntMask_MSB 0x24
+#define QIB_7322_IntMask_RcvUrg4IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg3IntMask_LSB 0x23
+#define QIB_7322_IntMask_RcvUrg3IntMask_MSB 0x23
+#define QIB_7322_IntMask_RcvUrg3IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg2IntMask_LSB 0x22
+#define QIB_7322_IntMask_RcvUrg2IntMask_MSB 0x22
+#define QIB_7322_IntMask_RcvUrg2IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg1IntMask_LSB 0x21
+#define QIB_7322_IntMask_RcvUrg1IntMask_MSB 0x21
+#define QIB_7322_IntMask_RcvUrg1IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvUrg0IntMask_LSB 0x20
+#define QIB_7322_IntMask_RcvUrg0IntMask_MSB 0x20
+#define QIB_7322_IntMask_RcvUrg0IntMask_RMASK 0x1
+#define QIB_7322_IntMask_ErrIntMask_1_LSB 0x1F
+#define QIB_7322_IntMask_ErrIntMask_1_MSB 0x1F
+#define QIB_7322_IntMask_ErrIntMask_1_RMASK 0x1
+#define QIB_7322_IntMask_ErrIntMask_0_LSB 0x1E
+#define QIB_7322_IntMask_ErrIntMask_0_MSB 0x1E
+#define QIB_7322_IntMask_ErrIntMask_0_RMASK 0x1
+#define QIB_7322_IntMask_ErrIntMask_LSB 0x1D
+#define QIB_7322_IntMask_ErrIntMask_MSB 0x1D
+#define QIB_7322_IntMask_ErrIntMask_RMASK 0x1
+#define QIB_7322_IntMask_AssertGPIOIntMask_LSB 0x1C
+#define QIB_7322_IntMask_AssertGPIOIntMask_MSB 0x1C
+#define QIB_7322_IntMask_AssertGPIOIntMask_RMASK 0x1
+#define QIB_7322_IntMask_SendDoneIntMask_1_LSB 0x19
+#define QIB_7322_IntMask_SendDoneIntMask_1_MSB 0x19
+#define QIB_7322_IntMask_SendDoneIntMask_1_RMASK 0x1
+#define QIB_7322_IntMask_SendDoneIntMask_0_LSB 0x18
+#define QIB_7322_IntMask_SendDoneIntMask_0_MSB 0x18
+#define QIB_7322_IntMask_SendDoneIntMask_0_RMASK 0x1
+#define QIB_7322_IntMask_SendBufAvailIntMask_LSB 0x17
+#define QIB_7322_IntMask_SendBufAvailIntMask_MSB 0x17
+#define QIB_7322_IntMask_SendBufAvailIntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail17IntMask_LSB 0x11
+#define QIB_7322_IntMask_RcvAvail17IntMask_MSB 0x11
+#define QIB_7322_IntMask_RcvAvail17IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail16IntMask_LSB 0x10
+#define QIB_7322_IntMask_RcvAvail16IntMask_MSB 0x10
+#define QIB_7322_IntMask_RcvAvail16IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail15IntMask_LSB 0xF
+#define QIB_7322_IntMask_RcvAvail15IntMask_MSB 0xF
+#define QIB_7322_IntMask_RcvAvail15IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail14IntMask_LSB 0xE
+#define QIB_7322_IntMask_RcvAvail14IntMask_MSB 0xE
+#define QIB_7322_IntMask_RcvAvail14IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail13IntMask_LSB 0xD
+#define QIB_7322_IntMask_RcvAvail13IntMask_MSB 0xD
+#define QIB_7322_IntMask_RcvAvail13IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail12IntMask_LSB 0xC
+#define QIB_7322_IntMask_RcvAvail12IntMask_MSB 0xC
+#define QIB_7322_IntMask_RcvAvail12IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail11IntMask_LSB 0xB
+#define QIB_7322_IntMask_RcvAvail11IntMask_MSB 0xB
+#define QIB_7322_IntMask_RcvAvail11IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail10IntMask_LSB 0xA
+#define QIB_7322_IntMask_RcvAvail10IntMask_MSB 0xA
+#define QIB_7322_IntMask_RcvAvail10IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail9IntMask_LSB 0x9
+#define QIB_7322_IntMask_RcvAvail9IntMask_MSB 0x9
+#define QIB_7322_IntMask_RcvAvail9IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail8IntMask_LSB 0x8
+#define QIB_7322_IntMask_RcvAvail8IntMask_MSB 0x8
+#define QIB_7322_IntMask_RcvAvail8IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail7IntMask_LSB 0x7
+#define QIB_7322_IntMask_RcvAvail7IntMask_MSB 0x7
+#define QIB_7322_IntMask_RcvAvail7IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail6IntMask_LSB 0x6
+#define QIB_7322_IntMask_RcvAvail6IntMask_MSB 0x6
+#define QIB_7322_IntMask_RcvAvail6IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail5IntMask_LSB 0x5
+#define QIB_7322_IntMask_RcvAvail5IntMask_MSB 0x5
+#define QIB_7322_IntMask_RcvAvail5IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail4IntMask_LSB 0x4
+#define QIB_7322_IntMask_RcvAvail4IntMask_MSB 0x4
+#define QIB_7322_IntMask_RcvAvail4IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail3IntMask_LSB 0x3
+#define QIB_7322_IntMask_RcvAvail3IntMask_MSB 0x3
+#define QIB_7322_IntMask_RcvAvail3IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail2IntMask_LSB 0x2
+#define QIB_7322_IntMask_RcvAvail2IntMask_MSB 0x2
+#define QIB_7322_IntMask_RcvAvail2IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail1IntMask_LSB 0x1
+#define QIB_7322_IntMask_RcvAvail1IntMask_MSB 0x1
+#define QIB_7322_IntMask_RcvAvail1IntMask_RMASK 0x1
+#define QIB_7322_IntMask_RcvAvail0IntMask_LSB 0x0
+#define QIB_7322_IntMask_RcvAvail0IntMask_MSB 0x0
+#define QIB_7322_IntMask_RcvAvail0IntMask_RMASK 0x1
+
+#define QIB_7322_IntStatus_OFFS 0x70
+#define QIB_7322_IntStatus_DEF 0x0000000000000000
+#define QIB_7322_IntStatus_SDmaInt_1_LSB 0x3F
+#define QIB_7322_IntStatus_SDmaInt_1_MSB 0x3F
+#define QIB_7322_IntStatus_SDmaInt_1_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaInt_0_LSB 0x3E
+#define QIB_7322_IntStatus_SDmaInt_0_MSB 0x3E
+#define QIB_7322_IntStatus_SDmaInt_0_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaProgressInt_1_LSB 0x3D
+#define QIB_7322_IntStatus_SDmaProgressInt_1_MSB 0x3D
+#define QIB_7322_IntStatus_SDmaProgressInt_1_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaProgressInt_0_LSB 0x3C
+#define QIB_7322_IntStatus_SDmaProgressInt_0_MSB 0x3C
+#define QIB_7322_IntStatus_SDmaProgressInt_0_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaIdleInt_1_LSB 0x3B
+#define QIB_7322_IntStatus_SDmaIdleInt_1_MSB 0x3B
+#define QIB_7322_IntStatus_SDmaIdleInt_1_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaIdleInt_0_LSB 0x3A
+#define QIB_7322_IntStatus_SDmaIdleInt_0_MSB 0x3A
+#define QIB_7322_IntStatus_SDmaIdleInt_0_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaCleanupDone_1_LSB 0x39
+#define QIB_7322_IntStatus_SDmaCleanupDone_1_MSB 0x39
+#define QIB_7322_IntStatus_SDmaCleanupDone_1_RMASK 0x1
+#define QIB_7322_IntStatus_SDmaCleanupDone_0_LSB 0x38
+#define QIB_7322_IntStatus_SDmaCleanupDone_0_MSB 0x38
+#define QIB_7322_IntStatus_SDmaCleanupDone_0_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg17_LSB 0x31
+#define QIB_7322_IntStatus_RcvUrg17_MSB 0x31
+#define QIB_7322_IntStatus_RcvUrg17_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg16_LSB 0x30
+#define QIB_7322_IntStatus_RcvUrg16_MSB 0x30
+#define QIB_7322_IntStatus_RcvUrg16_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg15_LSB 0x2F
+#define QIB_7322_IntStatus_RcvUrg15_MSB 0x2F
+#define QIB_7322_IntStatus_RcvUrg15_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg14_LSB 0x2E
+#define QIB_7322_IntStatus_RcvUrg14_MSB 0x2E
+#define QIB_7322_IntStatus_RcvUrg14_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg13_LSB 0x2D
+#define QIB_7322_IntStatus_RcvUrg13_MSB 0x2D
+#define QIB_7322_IntStatus_RcvUrg13_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg12_LSB 0x2C
+#define QIB_7322_IntStatus_RcvUrg12_MSB 0x2C
+#define QIB_7322_IntStatus_RcvUrg12_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg11_LSB 0x2B
+#define QIB_7322_IntStatus_RcvUrg11_MSB 0x2B
+#define QIB_7322_IntStatus_RcvUrg11_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg10_LSB 0x2A
+#define QIB_7322_IntStatus_RcvUrg10_MSB 0x2A
+#define QIB_7322_IntStatus_RcvUrg10_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg9_LSB 0x29
+#define QIB_7322_IntStatus_RcvUrg9_MSB 0x29
+#define QIB_7322_IntStatus_RcvUrg9_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg8_LSB 0x28
+#define QIB_7322_IntStatus_RcvUrg8_MSB 0x28
+#define QIB_7322_IntStatus_RcvUrg8_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg7_LSB 0x27
+#define QIB_7322_IntStatus_RcvUrg7_MSB 0x27
+#define QIB_7322_IntStatus_RcvUrg7_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg6_LSB 0x26
+#define QIB_7322_IntStatus_RcvUrg6_MSB 0x26
+#define QIB_7322_IntStatus_RcvUrg6_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg5_LSB 0x25
+#define QIB_7322_IntStatus_RcvUrg5_MSB 0x25
+#define QIB_7322_IntStatus_RcvUrg5_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg4_LSB 0x24
+#define QIB_7322_IntStatus_RcvUrg4_MSB 0x24
+#define QIB_7322_IntStatus_RcvUrg4_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg3_LSB 0x23
+#define QIB_7322_IntStatus_RcvUrg3_MSB 0x23
+#define QIB_7322_IntStatus_RcvUrg3_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg2_LSB 0x22
+#define QIB_7322_IntStatus_RcvUrg2_MSB 0x22
+#define QIB_7322_IntStatus_RcvUrg2_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg1_LSB 0x21
+#define QIB_7322_IntStatus_RcvUrg1_MSB 0x21
+#define QIB_7322_IntStatus_RcvUrg1_RMASK 0x1
+#define QIB_7322_IntStatus_RcvUrg0_LSB 0x20
+#define QIB_7322_IntStatus_RcvUrg0_MSB 0x20
+#define QIB_7322_IntStatus_RcvUrg0_RMASK 0x1
+#define QIB_7322_IntStatus_Err_1_LSB 0x1F
+#define QIB_7322_IntStatus_Err_1_MSB 0x1F
+#define QIB_7322_IntStatus_Err_1_RMASK 0x1
+#define QIB_7322_IntStatus_Err_0_LSB 0x1E
+#define QIB_7322_IntStatus_Err_0_MSB 0x1E
+#define QIB_7322_IntStatus_Err_0_RMASK 0x1
+#define QIB_7322_IntStatus_Err_LSB 0x1D
+#define QIB_7322_IntStatus_Err_MSB 0x1D
+#define QIB_7322_IntStatus_Err_RMASK 0x1
+#define QIB_7322_IntStatus_AssertGPIO_LSB 0x1C
+#define QIB_7322_IntStatus_AssertGPIO_MSB 0x1C
+#define QIB_7322_IntStatus_AssertGPIO_RMASK 0x1
+#define QIB_7322_IntStatus_SendDone_1_LSB 0x19
+#define QIB_7322_IntStatus_SendDone_1_MSB 0x19
+#define QIB_7322_IntStatus_SendDone_1_RMASK 0x1
+#define QIB_7322_IntStatus_SendDone_0_LSB 0x18
+#define QIB_7322_IntStatus_SendDone_0_MSB 0x18
+#define QIB_7322_IntStatus_SendDone_0_RMASK 0x1
+#define QIB_7322_IntStatus_SendBufAvail_LSB 0x17
+#define QIB_7322_IntStatus_SendBufAvail_MSB 0x17
+#define QIB_7322_IntStatus_SendBufAvail_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail17_LSB 0x11
+#define QIB_7322_IntStatus_RcvAvail17_MSB 0x11
+#define QIB_7322_IntStatus_RcvAvail17_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail16_LSB 0x10
+#define QIB_7322_IntStatus_RcvAvail16_MSB 0x10
+#define QIB_7322_IntStatus_RcvAvail16_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail15_LSB 0xF
+#define QIB_7322_IntStatus_RcvAvail15_MSB 0xF
+#define QIB_7322_IntStatus_RcvAvail15_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail14_LSB 0xE
+#define QIB_7322_IntStatus_RcvAvail14_MSB 0xE
+#define QIB_7322_IntStatus_RcvAvail14_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail13_LSB 0xD
+#define QIB_7322_IntStatus_RcvAvail13_MSB 0xD
+#define QIB_7322_IntStatus_RcvAvail13_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail12_LSB 0xC
+#define QIB_7322_IntStatus_RcvAvail12_MSB 0xC
+#define QIB_7322_IntStatus_RcvAvail12_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail11_LSB 0xB
+#define QIB_7322_IntStatus_RcvAvail11_MSB 0xB
+#define QIB_7322_IntStatus_RcvAvail11_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail10_LSB 0xA
+#define QIB_7322_IntStatus_RcvAvail10_MSB 0xA
+#define QIB_7322_IntStatus_RcvAvail10_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail9_LSB 0x9
+#define QIB_7322_IntStatus_RcvAvail9_MSB 0x9
+#define QIB_7322_IntStatus_RcvAvail9_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail8_LSB 0x8
+#define QIB_7322_IntStatus_RcvAvail8_MSB 0x8
+#define QIB_7322_IntStatus_RcvAvail8_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail7_LSB 0x7
+#define QIB_7322_IntStatus_RcvAvail7_MSB 0x7
+#define QIB_7322_IntStatus_RcvAvail7_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail6_LSB 0x6
+#define QIB_7322_IntStatus_RcvAvail6_MSB 0x6
+#define QIB_7322_IntStatus_RcvAvail6_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail5_LSB 0x5
+#define QIB_7322_IntStatus_RcvAvail5_MSB 0x5
+#define QIB_7322_IntStatus_RcvAvail5_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail4_LSB 0x4
+#define QIB_7322_IntStatus_RcvAvail4_MSB 0x4
+#define QIB_7322_IntStatus_RcvAvail4_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail3_LSB 0x3
+#define QIB_7322_IntStatus_RcvAvail3_MSB 0x3
+#define QIB_7322_IntStatus_RcvAvail3_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail2_LSB 0x2
+#define QIB_7322_IntStatus_RcvAvail2_MSB 0x2
+#define QIB_7322_IntStatus_RcvAvail2_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail1_LSB 0x1
+#define QIB_7322_IntStatus_RcvAvail1_MSB 0x1
+#define QIB_7322_IntStatus_RcvAvail1_RMASK 0x1
+#define QIB_7322_IntStatus_RcvAvail0_LSB 0x0
+#define QIB_7322_IntStatus_RcvAvail0_MSB 0x0
+#define QIB_7322_IntStatus_RcvAvail0_RMASK 0x1
+
+#define QIB_7322_IntClear_OFFS 0x78
+#define QIB_7322_IntClear_DEF 0x0000000000000000
+#define QIB_7322_IntClear_SDmaIntClear_1_LSB 0x3F
+#define QIB_7322_IntClear_SDmaIntClear_1_MSB 0x3F
+#define QIB_7322_IntClear_SDmaIntClear_1_RMASK 0x1
+#define QIB_7322_IntClear_SDmaIntClear_0_LSB 0x3E
+#define QIB_7322_IntClear_SDmaIntClear_0_MSB 0x3E
+#define QIB_7322_IntClear_SDmaIntClear_0_RMASK 0x1
+#define QIB_7322_IntClear_SDmaProgressIntClear_1_LSB 0x3D
+#define QIB_7322_IntClear_SDmaProgressIntClear_1_MSB 0x3D
+#define QIB_7322_IntClear_SDmaProgressIntClear_1_RMASK 0x1
+#define QIB_7322_IntClear_SDmaProgressIntClear_0_LSB 0x3C
+#define QIB_7322_IntClear_SDmaProgressIntClear_0_MSB 0x3C
+#define QIB_7322_IntClear_SDmaProgressIntClear_0_RMASK 0x1
+#define QIB_7322_IntClear_SDmaIdleIntClear_1_LSB 0x3B
+#define QIB_7322_IntClear_SDmaIdleIntClear_1_MSB 0x3B
+#define QIB_7322_IntClear_SDmaIdleIntClear_1_RMASK 0x1
+#define QIB_7322_IntClear_SDmaIdleIntClear_0_LSB 0x3A
+#define QIB_7322_IntClear_SDmaIdleIntClear_0_MSB 0x3A
+#define QIB_7322_IntClear_SDmaIdleIntClear_0_RMASK 0x1
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_1_LSB 0x39
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_1_MSB 0x39
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_1_RMASK 0x1
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_0_LSB 0x38
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_0_MSB 0x38
+#define QIB_7322_IntClear_SDmaCleanupDoneClear_0_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg17IntClear_LSB 0x31
+#define QIB_7322_IntClear_RcvUrg17IntClear_MSB 0x31
+#define QIB_7322_IntClear_RcvUrg17IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg16IntClear_LSB 0x30
+#define QIB_7322_IntClear_RcvUrg16IntClear_MSB 0x30
+#define QIB_7322_IntClear_RcvUrg16IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg15IntClear_LSB 0x2F
+#define QIB_7322_IntClear_RcvUrg15IntClear_MSB 0x2F
+#define QIB_7322_IntClear_RcvUrg15IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg14IntClear_LSB 0x2E
+#define QIB_7322_IntClear_RcvUrg14IntClear_MSB 0x2E
+#define QIB_7322_IntClear_RcvUrg14IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg13IntClear_LSB 0x2D
+#define QIB_7322_IntClear_RcvUrg13IntClear_MSB 0x2D
+#define QIB_7322_IntClear_RcvUrg13IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg12IntClear_LSB 0x2C
+#define QIB_7322_IntClear_RcvUrg12IntClear_MSB 0x2C
+#define QIB_7322_IntClear_RcvUrg12IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg11IntClear_LSB 0x2B
+#define QIB_7322_IntClear_RcvUrg11IntClear_MSB 0x2B
+#define QIB_7322_IntClear_RcvUrg11IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg10IntClear_LSB 0x2A
+#define QIB_7322_IntClear_RcvUrg10IntClear_MSB 0x2A
+#define QIB_7322_IntClear_RcvUrg10IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg9IntClear_LSB 0x29
+#define QIB_7322_IntClear_RcvUrg9IntClear_MSB 0x29
+#define QIB_7322_IntClear_RcvUrg9IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg8IntClear_LSB 0x28
+#define QIB_7322_IntClear_RcvUrg8IntClear_MSB 0x28
+#define QIB_7322_IntClear_RcvUrg8IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg7IntClear_LSB 0x27
+#define QIB_7322_IntClear_RcvUrg7IntClear_MSB 0x27
+#define QIB_7322_IntClear_RcvUrg7IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg6IntClear_LSB 0x26
+#define QIB_7322_IntClear_RcvUrg6IntClear_MSB 0x26
+#define QIB_7322_IntClear_RcvUrg6IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg5IntClear_LSB 0x25
+#define QIB_7322_IntClear_RcvUrg5IntClear_MSB 0x25
+#define QIB_7322_IntClear_RcvUrg5IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg4IntClear_LSB 0x24
+#define QIB_7322_IntClear_RcvUrg4IntClear_MSB 0x24
+#define QIB_7322_IntClear_RcvUrg4IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg3IntClear_LSB 0x23
+#define QIB_7322_IntClear_RcvUrg3IntClear_MSB 0x23
+#define QIB_7322_IntClear_RcvUrg3IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg2IntClear_LSB 0x22
+#define QIB_7322_IntClear_RcvUrg2IntClear_MSB 0x22
+#define QIB_7322_IntClear_RcvUrg2IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg1IntClear_LSB 0x21
+#define QIB_7322_IntClear_RcvUrg1IntClear_MSB 0x21
+#define QIB_7322_IntClear_RcvUrg1IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvUrg0IntClear_LSB 0x20
+#define QIB_7322_IntClear_RcvUrg0IntClear_MSB 0x20
+#define QIB_7322_IntClear_RcvUrg0IntClear_RMASK 0x1
+#define QIB_7322_IntClear_ErrIntClear_1_LSB 0x1F
+#define QIB_7322_IntClear_ErrIntClear_1_MSB 0x1F
+#define QIB_7322_IntClear_ErrIntClear_1_RMASK 0x1
+#define QIB_7322_IntClear_ErrIntClear_0_LSB 0x1E
+#define QIB_7322_IntClear_ErrIntClear_0_MSB 0x1E
+#define QIB_7322_IntClear_ErrIntClear_0_RMASK 0x1
+#define QIB_7322_IntClear_ErrIntClear_LSB 0x1D
+#define QIB_7322_IntClear_ErrIntClear_MSB 0x1D
+#define QIB_7322_IntClear_ErrIntClear_RMASK 0x1
+#define QIB_7322_IntClear_AssertGPIOIntClear_LSB 0x1C
+#define QIB_7322_IntClear_AssertGPIOIntClear_MSB 0x1C
+#define QIB_7322_IntClear_AssertGPIOIntClear_RMASK 0x1
+#define QIB_7322_IntClear_SendDoneIntClear_1_LSB 0x19
+#define QIB_7322_IntClear_SendDoneIntClear_1_MSB 0x19
+#define QIB_7322_IntClear_SendDoneIntClear_1_RMASK 0x1
+#define QIB_7322_IntClear_SendDoneIntClear_0_LSB 0x18
+#define QIB_7322_IntClear_SendDoneIntClear_0_MSB 0x18
+#define QIB_7322_IntClear_SendDoneIntClear_0_RMASK 0x1
+#define QIB_7322_IntClear_SendBufAvailIntClear_LSB 0x17
+#define QIB_7322_IntClear_SendBufAvailIntClear_MSB 0x17
+#define QIB_7322_IntClear_SendBufAvailIntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail17IntClear_LSB 0x11
+#define QIB_7322_IntClear_RcvAvail17IntClear_MSB 0x11
+#define QIB_7322_IntClear_RcvAvail17IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail16IntClear_LSB 0x10
+#define QIB_7322_IntClear_RcvAvail16IntClear_MSB 0x10
+#define QIB_7322_IntClear_RcvAvail16IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail15IntClear_LSB 0xF
+#define QIB_7322_IntClear_RcvAvail15IntClear_MSB 0xF
+#define QIB_7322_IntClear_RcvAvail15IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail14IntClear_LSB 0xE
+#define QIB_7322_IntClear_RcvAvail14IntClear_MSB 0xE
+#define QIB_7322_IntClear_RcvAvail14IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail13IntClear_LSB 0xD
+#define QIB_7322_IntClear_RcvAvail13IntClear_MSB 0xD
+#define QIB_7322_IntClear_RcvAvail13IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail12IntClear_LSB 0xC
+#define QIB_7322_IntClear_RcvAvail12IntClear_MSB 0xC
+#define QIB_7322_IntClear_RcvAvail12IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail11IntClear_LSB 0xB
+#define QIB_7322_IntClear_RcvAvail11IntClear_MSB 0xB
+#define QIB_7322_IntClear_RcvAvail11IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail10IntClear_LSB 0xA
+#define QIB_7322_IntClear_RcvAvail10IntClear_MSB 0xA
+#define QIB_7322_IntClear_RcvAvail10IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail9IntClear_LSB 0x9
+#define QIB_7322_IntClear_RcvAvail9IntClear_MSB 0x9
+#define QIB_7322_IntClear_RcvAvail9IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail8IntClear_LSB 0x8
+#define QIB_7322_IntClear_RcvAvail8IntClear_MSB 0x8
+#define QIB_7322_IntClear_RcvAvail8IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail7IntClear_LSB 0x7
+#define QIB_7322_IntClear_RcvAvail7IntClear_MSB 0x7
+#define QIB_7322_IntClear_RcvAvail7IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail6IntClear_LSB 0x6
+#define QIB_7322_IntClear_RcvAvail6IntClear_MSB 0x6
+#define QIB_7322_IntClear_RcvAvail6IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail5IntClear_LSB 0x5
+#define QIB_7322_IntClear_RcvAvail5IntClear_MSB 0x5
+#define QIB_7322_IntClear_RcvAvail5IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail4IntClear_LSB 0x4
+#define QIB_7322_IntClear_RcvAvail4IntClear_MSB 0x4
+#define QIB_7322_IntClear_RcvAvail4IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail3IntClear_LSB 0x3
+#define QIB_7322_IntClear_RcvAvail3IntClear_MSB 0x3
+#define QIB_7322_IntClear_RcvAvail3IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail2IntClear_LSB 0x2
+#define QIB_7322_IntClear_RcvAvail2IntClear_MSB 0x2
+#define QIB_7322_IntClear_RcvAvail2IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail1IntClear_LSB 0x1
+#define QIB_7322_IntClear_RcvAvail1IntClear_MSB 0x1
+#define QIB_7322_IntClear_RcvAvail1IntClear_RMASK 0x1
+#define QIB_7322_IntClear_RcvAvail0IntClear_LSB 0x0
+#define QIB_7322_IntClear_RcvAvail0IntClear_MSB 0x0
+#define QIB_7322_IntClear_RcvAvail0IntClear_RMASK 0x1
+
+#define QIB_7322_ErrMask_OFFS 0x80
+#define QIB_7322_ErrMask_DEF 0x0000000000000000
+#define QIB_7322_ErrMask_ResetNegatedMask_LSB 0x3F
+#define QIB_7322_ErrMask_ResetNegatedMask_MSB 0x3F
+#define QIB_7322_ErrMask_ResetNegatedMask_RMASK 0x1
+#define QIB_7322_ErrMask_HardwareErrMask_LSB 0x3E
+#define QIB_7322_ErrMask_HardwareErrMask_MSB 0x3E
+#define QIB_7322_ErrMask_HardwareErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_InvalidAddrErrMask_LSB 0x3D
+#define QIB_7322_ErrMask_InvalidAddrErrMask_MSB 0x3D
+#define QIB_7322_ErrMask_InvalidAddrErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SDmaVL15ErrMask_LSB 0x38
+#define QIB_7322_ErrMask_SDmaVL15ErrMask_MSB 0x38
+#define QIB_7322_ErrMask_SDmaVL15ErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SBufVL15MisUseErrMask_LSB 0x37
+#define QIB_7322_ErrMask_SBufVL15MisUseErrMask_MSB 0x37
+#define QIB_7322_ErrMask_SBufVL15MisUseErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_InvalidEEPCmdMask_LSB 0x35
+#define QIB_7322_ErrMask_InvalidEEPCmdMask_MSB 0x35
+#define QIB_7322_ErrMask_InvalidEEPCmdMask_RMASK 0x1
+#define QIB_7322_ErrMask_RcvContextShareErrMask_LSB 0x34
+#define QIB_7322_ErrMask_RcvContextShareErrMask_MSB 0x34
+#define QIB_7322_ErrMask_RcvContextShareErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SendVLMismatchErrMask_LSB 0x24
+#define QIB_7322_ErrMask_SendVLMismatchErrMask_MSB 0x24
+#define QIB_7322_ErrMask_SendVLMismatchErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SendArmLaunchErrMask_LSB 0x23
+#define QIB_7322_ErrMask_SendArmLaunchErrMask_MSB 0x23
+#define QIB_7322_ErrMask_SendArmLaunchErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SendSpecialTriggerErrMask_LSB 0x1B
+#define QIB_7322_ErrMask_SendSpecialTriggerErrMask_MSB 0x1B
+#define QIB_7322_ErrMask_SendSpecialTriggerErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SDmaWrongPortErrMask_LSB 0x1A
+#define QIB_7322_ErrMask_SDmaWrongPortErrMask_MSB 0x1A
+#define QIB_7322_ErrMask_SDmaWrongPortErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_SDmaBufMaskDuplicateErrMask_LSB 0x19
+#define QIB_7322_ErrMask_SDmaBufMaskDuplicateErrMask_MSB 0x19
+#define QIB_7322_ErrMask_SDmaBufMaskDuplicateErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_RcvHdrFullErrMask_LSB 0xD
+#define QIB_7322_ErrMask_RcvHdrFullErrMask_MSB 0xD
+#define QIB_7322_ErrMask_RcvHdrFullErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_RcvEgrFullErrMask_LSB 0xC
+#define QIB_7322_ErrMask_RcvEgrFullErrMask_MSB 0xC
+#define QIB_7322_ErrMask_RcvEgrFullErrMask_RMASK 0x1
+
+#define QIB_7322_ErrStatus_OFFS 0x88
+#define QIB_7322_ErrStatus_DEF 0x0000000000000000
+#define QIB_7322_ErrStatus_ResetNegated_LSB 0x3F
+#define QIB_7322_ErrStatus_ResetNegated_MSB 0x3F
+#define QIB_7322_ErrStatus_ResetNegated_RMASK 0x1
+#define QIB_7322_ErrStatus_HardwareErr_LSB 0x3E
+#define QIB_7322_ErrStatus_HardwareErr_MSB 0x3E
+#define QIB_7322_ErrStatus_HardwareErr_RMASK 0x1
+#define QIB_7322_ErrStatus_InvalidAddrErr_LSB 0x3D
+#define QIB_7322_ErrStatus_InvalidAddrErr_MSB 0x3D
+#define QIB_7322_ErrStatus_InvalidAddrErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SDmaVL15Err_LSB 0x38
+#define QIB_7322_ErrStatus_SDmaVL15Err_MSB 0x38
+#define QIB_7322_ErrStatus_SDmaVL15Err_RMASK 0x1
+#define QIB_7322_ErrStatus_SBufVL15MisUseErr_LSB 0x37
+#define QIB_7322_ErrStatus_SBufVL15MisUseErr_MSB 0x37
+#define QIB_7322_ErrStatus_SBufVL15MisUseErr_RMASK 0x1
+#define QIB_7322_ErrStatus_InvalidEEPCmdErr_LSB 0x35
+#define QIB_7322_ErrStatus_InvalidEEPCmdErr_MSB 0x35
+#define QIB_7322_ErrStatus_InvalidEEPCmdErr_RMASK 0x1
+#define QIB_7322_ErrStatus_RcvContextShareErr_LSB 0x34
+#define QIB_7322_ErrStatus_RcvContextShareErr_MSB 0x34
+#define QIB_7322_ErrStatus_RcvContextShareErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SendVLMismatchErr_LSB 0x24
+#define QIB_7322_ErrStatus_SendVLMismatchErr_MSB 0x24
+#define QIB_7322_ErrStatus_SendVLMismatchErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SendArmLaunchErr_LSB 0x23
+#define QIB_7322_ErrStatus_SendArmLaunchErr_MSB 0x23
+#define QIB_7322_ErrStatus_SendArmLaunchErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SendSpecialTriggerErr_LSB 0x1B
+#define QIB_7322_ErrStatus_SendSpecialTriggerErr_MSB 0x1B
+#define QIB_7322_ErrStatus_SendSpecialTriggerErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SDmaWrongPortErr_LSB 0x1A
+#define QIB_7322_ErrStatus_SDmaWrongPortErr_MSB 0x1A
+#define QIB_7322_ErrStatus_SDmaWrongPortErr_RMASK 0x1
+#define QIB_7322_ErrStatus_SDmaBufMaskDuplicateErr_LSB 0x19
+#define QIB_7322_ErrStatus_SDmaBufMaskDuplicateErr_MSB 0x19
+#define QIB_7322_ErrStatus_SDmaBufMaskDuplicateErr_RMASK 0x1
+#define QIB_7322_ErrStatus_RcvHdrFullErr_LSB 0xD
+#define QIB_7322_ErrStatus_RcvHdrFullErr_MSB 0xD
+#define QIB_7322_ErrStatus_RcvHdrFullErr_RMASK 0x1
+#define QIB_7322_ErrStatus_RcvEgrFullErr_LSB 0xC
+#define QIB_7322_ErrStatus_RcvEgrFullErr_MSB 0xC
+#define QIB_7322_ErrStatus_RcvEgrFullErr_RMASK 0x1
+
+#define QIB_7322_ErrClear_OFFS 0x90
+#define QIB_7322_ErrClear_DEF 0x0000000000000000
+#define QIB_7322_ErrClear_ResetNegatedClear_LSB 0x3F
+#define QIB_7322_ErrClear_ResetNegatedClear_MSB 0x3F
+#define QIB_7322_ErrClear_ResetNegatedClear_RMASK 0x1
+#define QIB_7322_ErrClear_HardwareErrClear_LSB 0x3E
+#define QIB_7322_ErrClear_HardwareErrClear_MSB 0x3E
+#define QIB_7322_ErrClear_HardwareErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_InvalidAddrErrClear_LSB 0x3D
+#define QIB_7322_ErrClear_InvalidAddrErrClear_MSB 0x3D
+#define QIB_7322_ErrClear_InvalidAddrErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SDmaVL15ErrClear_LSB 0x38
+#define QIB_7322_ErrClear_SDmaVL15ErrClear_MSB 0x38
+#define QIB_7322_ErrClear_SDmaVL15ErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SBufVL15MisUseErrClear_LSB 0x37
+#define QIB_7322_ErrClear_SBufVL15MisUseErrClear_MSB 0x37
+#define QIB_7322_ErrClear_SBufVL15MisUseErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_InvalidEEPCmdErrClear_LSB 0x35
+#define QIB_7322_ErrClear_InvalidEEPCmdErrClear_MSB 0x35
+#define QIB_7322_ErrClear_InvalidEEPCmdErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_RcvContextShareErrClear_LSB 0x34
+#define QIB_7322_ErrClear_RcvContextShareErrClear_MSB 0x34
+#define QIB_7322_ErrClear_RcvContextShareErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SendVLMismatchErrMask_LSB 0x24
+#define QIB_7322_ErrClear_SendVLMismatchErrMask_MSB 0x24
+#define QIB_7322_ErrClear_SendVLMismatchErrMask_RMASK 0x1
+#define QIB_7322_ErrClear_SendArmLaunchErrClear_LSB 0x23
+#define QIB_7322_ErrClear_SendArmLaunchErrClear_MSB 0x23
+#define QIB_7322_ErrClear_SendArmLaunchErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SendSpecialTriggerErrClear_LSB 0x1B
+#define QIB_7322_ErrClear_SendSpecialTriggerErrClear_MSB 0x1B
+#define QIB_7322_ErrClear_SendSpecialTriggerErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SDmaWrongPortErrClear_LSB 0x1A
+#define QIB_7322_ErrClear_SDmaWrongPortErrClear_MSB 0x1A
+#define QIB_7322_ErrClear_SDmaWrongPortErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_SDmaBufMaskDuplicateErrClear_LSB 0x19
+#define QIB_7322_ErrClear_SDmaBufMaskDuplicateErrClear_MSB 0x19
+#define QIB_7322_ErrClear_SDmaBufMaskDuplicateErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_RcvHdrFullErrClear_LSB 0xD
+#define QIB_7322_ErrClear_RcvHdrFullErrClear_MSB 0xD
+#define QIB_7322_ErrClear_RcvHdrFullErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_RcvEgrFullErrClear_LSB 0xC
+#define QIB_7322_ErrClear_RcvEgrFullErrClear_MSB 0xC
+#define QIB_7322_ErrClear_RcvEgrFullErrClear_RMASK 0x1
+
+#define QIB_7322_HwErrMask_OFFS 0x98
+#define QIB_7322_HwErrMask_DEF 0x0000000000000000
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_1_LSB 0x3F
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_1_MSB 0x3F
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_1_RMASK 0x1
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_0_LSB 0x3E
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_0_MSB 0x3E
+#define QIB_7322_HwErrMask_IBSerdesPClkNotDetectMask_0_RMASK 0x1
+#define QIB_7322_HwErrMask_PCIESerdesPClkNotDetectMask_LSB 0x37
+#define QIB_7322_HwErrMask_PCIESerdesPClkNotDetectMask_MSB 0x37
+#define QIB_7322_HwErrMask_PCIESerdesPClkNotDetectMask_RMASK 0x1
+#define QIB_7322_HwErrMask_PowerOnBISTFailedMask_LSB 0x36
+#define QIB_7322_HwErrMask_PowerOnBISTFailedMask_MSB 0x36
+#define QIB_7322_HwErrMask_PowerOnBISTFailedMask_RMASK 0x1
+#define QIB_7322_HwErrMask_TempsenseTholdReachedMask_LSB 0x35
+#define QIB_7322_HwErrMask_TempsenseTholdReachedMask_MSB 0x35
+#define QIB_7322_HwErrMask_TempsenseTholdReachedMask_RMASK 0x1
+#define QIB_7322_HwErrMask_MemoryErrMask_LSB 0x30
+#define QIB_7322_HwErrMask_MemoryErrMask_MSB 0x30
+#define QIB_7322_HwErrMask_MemoryErrMask_RMASK 0x1
+#define QIB_7322_HwErrMask_pcie_phy_txParityErr_LSB 0x22
+#define QIB_7322_HwErrMask_pcie_phy_txParityErr_MSB 0x22
+#define QIB_7322_HwErrMask_pcie_phy_txParityErr_RMASK 0x1
+#define QIB_7322_HwErrMask_PCIeBusParityErrMask_LSB 0x1F
+#define QIB_7322_HwErrMask_PCIeBusParityErrMask_MSB 0x21
+#define QIB_7322_HwErrMask_PCIeBusParityErrMask_RMASK 0x7
+#define QIB_7322_HwErrMask_PcieCplTimeoutMask_LSB 0x1E
+#define QIB_7322_HwErrMask_PcieCplTimeoutMask_MSB 0x1E
+#define QIB_7322_HwErrMask_PcieCplTimeoutMask_RMASK 0x1
+#define QIB_7322_HwErrMask_PciePoisonedTLPMask_LSB 0x1D
+#define QIB_7322_HwErrMask_PciePoisonedTLPMask_MSB 0x1D
+#define QIB_7322_HwErrMask_PciePoisonedTLPMask_RMASK 0x1
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_1_LSB 0x1C
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_1_MSB 0x1C
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_1_RMASK 0x1
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_0_LSB 0x1B
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_0_MSB 0x1B
+#define QIB_7322_HwErrMask_SDmaMemReadErrMask_0_RMASK 0x1
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_1_LSB 0xF
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_1_MSB 0xF
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_1_RMASK 0x1
+#define QIB_7322_HwErrMask_IBCBusToSPCParityErrMask_1_LSB 0xE
+#define QIB_7322_HwErrMask_IBCBusToSPCParityErrMask_1_MSB 0xE
+#define QIB_7322_HwErrMask_IBCBusToSPCParityErrMask_1_RMASK 0x1
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_0_LSB 0xD
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_0_MSB 0xD
+#define QIB_7322_HwErrMask_IBCBusFromSPCParityErrMask_0_RMASK 0x1
+#define QIB_7322_HwErrMask_statusValidNoEopMask_LSB 0xC
+#define QIB_7322_HwErrMask_statusValidNoEopMask_MSB 0xC
+#define QIB_7322_HwErrMask_statusValidNoEopMask_RMASK 0x1
+#define QIB_7322_HwErrMask_LATriggeredMask_LSB 0xB
+#define QIB_7322_HwErrMask_LATriggeredMask_MSB 0xB
+#define QIB_7322_HwErrMask_LATriggeredMask_RMASK 0x1
+
+#define QIB_7322_HwErrStatus_OFFS 0xA0
+#define QIB_7322_HwErrStatus_DEF 0x0000000000000000
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_1_LSB 0x3F
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_1_MSB 0x3F
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_1_RMASK 0x1
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_0_LSB 0x3E
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_0_MSB 0x3E
+#define QIB_7322_HwErrStatus_IBSerdesPClkNotDetect_0_RMASK 0x1
+#define QIB_7322_HwErrStatus_PCIESerdesPClkNotDetect_LSB 0x37
+#define QIB_7322_HwErrStatus_PCIESerdesPClkNotDetect_MSB 0x37
+#define QIB_7322_HwErrStatus_PCIESerdesPClkNotDetect_RMASK 0x1
+#define QIB_7322_HwErrStatus_PowerOnBISTFailed_LSB 0x36
+#define QIB_7322_HwErrStatus_PowerOnBISTFailed_MSB 0x36
+#define QIB_7322_HwErrStatus_PowerOnBISTFailed_RMASK 0x1
+#define QIB_7322_HwErrStatus_TempsenseTholdReached_LSB 0x35
+#define QIB_7322_HwErrStatus_TempsenseTholdReached_MSB 0x35
+#define QIB_7322_HwErrStatus_TempsenseTholdReached_RMASK 0x1
+#define QIB_7322_HwErrStatus_MemoryErr_LSB 0x30
+#define QIB_7322_HwErrStatus_MemoryErr_MSB 0x30
+#define QIB_7322_HwErrStatus_MemoryErr_RMASK 0x1
+#define QIB_7322_HwErrStatus_pcie_phy_txParityErr_LSB 0x22
+#define QIB_7322_HwErrStatus_pcie_phy_txParityErr_MSB 0x22
+#define QIB_7322_HwErrStatus_pcie_phy_txParityErr_RMASK 0x1
+#define QIB_7322_HwErrStatus_PCIeBusParity_LSB 0x1F
+#define QIB_7322_HwErrStatus_PCIeBusParity_MSB 0x21
+#define QIB_7322_HwErrStatus_PCIeBusParity_RMASK 0x7
+#define QIB_7322_HwErrStatus_PcieCplTimeout_LSB 0x1E
+#define QIB_7322_HwErrStatus_PcieCplTimeout_MSB 0x1E
+#define QIB_7322_HwErrStatus_PcieCplTimeout_RMASK 0x1
+#define QIB_7322_HwErrStatus_PciePoisonedTLP_LSB 0x1D
+#define QIB_7322_HwErrStatus_PciePoisonedTLP_MSB 0x1D
+#define QIB_7322_HwErrStatus_PciePoisonedTLP_RMASK 0x1
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_1_LSB 0x1C
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_1_MSB 0x1C
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_1_RMASK 0x1
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_0_LSB 0x1B
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_0_MSB 0x1B
+#define QIB_7322_HwErrStatus_SDmaMemReadErr_0_RMASK 0x1
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_1_LSB 0xF
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_1_MSB 0xF
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_1_RMASK 0x1
+#define QIB_7322_HwErrStatus_IBCBusToSPCParityErr_1_LSB 0xE
+#define QIB_7322_HwErrStatus_IBCBusToSPCParityErr_1_MSB 0xE
+#define QIB_7322_HwErrStatus_IBCBusToSPCParityErr_1_RMASK 0x1
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_0_LSB 0xD
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_0_MSB 0xD
+#define QIB_7322_HwErrStatus_IBCBusFromSPCParityErr_0_RMASK 0x1
+#define QIB_7322_HwErrStatus_statusValidNoEop_LSB 0xC
+#define QIB_7322_HwErrStatus_statusValidNoEop_MSB 0xC
+#define QIB_7322_HwErrStatus_statusValidNoEop_RMASK 0x1
+#define QIB_7322_HwErrStatus_LATriggered_LSB 0xB
+#define QIB_7322_HwErrStatus_LATriggered_MSB 0xB
+#define QIB_7322_HwErrStatus_LATriggered_RMASK 0x1
+
+#define QIB_7322_HwErrClear_OFFS 0xA8
+#define QIB_7322_HwErrClear_DEF 0x0000000000000000
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_1_LSB 0x3F
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_1_MSB 0x3F
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_1_RMASK 0x1
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_0_LSB 0x3E
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_0_MSB 0x3E
+#define QIB_7322_HwErrClear_IBSerdesPClkNotDetectClear_0_RMASK 0x1
+#define QIB_7322_HwErrClear_PCIESerdesPClkNotDetectClear_LSB 0x37
+#define QIB_7322_HwErrClear_PCIESerdesPClkNotDetectClear_MSB 0x37
+#define QIB_7322_HwErrClear_PCIESerdesPClkNotDetectClear_RMASK 0x1
+#define QIB_7322_HwErrClear_PowerOnBISTFailedClear_LSB 0x36
+#define QIB_7322_HwErrClear_PowerOnBISTFailedClear_MSB 0x36
+#define QIB_7322_HwErrClear_PowerOnBISTFailedClear_RMASK 0x1
+#define QIB_7322_HwErrClear_TempsenseTholdReachedClear_LSB 0x35
+#define QIB_7322_HwErrClear_TempsenseTholdReachedClear_MSB 0x35
+#define QIB_7322_HwErrClear_TempsenseTholdReachedClear_RMASK 0x1
+#define QIB_7322_HwErrClear_MemoryErrClear_LSB 0x30
+#define QIB_7322_HwErrClear_MemoryErrClear_MSB 0x30
+#define QIB_7322_HwErrClear_MemoryErrClear_RMASK 0x1
+#define QIB_7322_HwErrClear_pcie_phy_txParityErr_LSB 0x22
+#define QIB_7322_HwErrClear_pcie_phy_txParityErr_MSB 0x22
+#define QIB_7322_HwErrClear_pcie_phy_txParityErr_RMASK 0x1
+#define QIB_7322_HwErrClear_PCIeBusParityClear_LSB 0x1F
+#define QIB_7322_HwErrClear_PCIeBusParityClear_MSB 0x21
+#define QIB_7322_HwErrClear_PCIeBusParityClear_RMASK 0x7
+#define QIB_7322_HwErrClear_PcieCplTimeoutClear_LSB 0x1E
+#define QIB_7322_HwErrClear_PcieCplTimeoutClear_MSB 0x1E
+#define QIB_7322_HwErrClear_PcieCplTimeoutClear_RMASK 0x1
+#define QIB_7322_HwErrClear_PciePoisonedTLPClear_LSB 0x1D
+#define QIB_7322_HwErrClear_PciePoisonedTLPClear_MSB 0x1D
+#define QIB_7322_HwErrClear_PciePoisonedTLPClear_RMASK 0x1
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_1_LSB 0x1C
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_1_MSB 0x1C
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_1_RMASK 0x1
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_0_LSB 0x1B
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_0_MSB 0x1B
+#define QIB_7322_HwErrClear_SDmaMemReadErrClear_0_RMASK 0x1
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_1_LSB 0xF
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_1_MSB 0xF
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_1_RMASK 0x1
+#define QIB_7322_HwErrClear_IBCBusToSPCParityErrClear_1_LSB 0xE
+#define QIB_7322_HwErrClear_IBCBusToSPCParityErrClear_1_MSB 0xE
+#define QIB_7322_HwErrClear_IBCBusToSPCParityErrClear_1_RMASK 0x1
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_0_LSB 0xD
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_0_MSB 0xD
+#define QIB_7322_HwErrClear_IBCBusFromSPCParityErrClear_0_RMASK 0x1
+#define QIB_7322_HwErrClear_statusValidNoEopClear_LSB 0xC
+#define QIB_7322_HwErrClear_statusValidNoEopClear_MSB 0xC
+#define QIB_7322_HwErrClear_statusValidNoEopClear_RMASK 0x1
+#define QIB_7322_HwErrClear_LATriggeredClear_LSB 0xB
+#define QIB_7322_HwErrClear_LATriggeredClear_MSB 0xB
+#define QIB_7322_HwErrClear_LATriggeredClear_RMASK 0x1
+
+#define QIB_7322_HwDiagCtrl_OFFS 0xB0
+#define QIB_7322_HwDiagCtrl_DEF 0x0000000000000000
+#define QIB_7322_HwDiagCtrl_Diagnostic_LSB 0x3F
+#define QIB_7322_HwDiagCtrl_Diagnostic_MSB 0x3F
+#define QIB_7322_HwDiagCtrl_Diagnostic_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_CounterWrEnable_LSB 0x3D
+#define QIB_7322_HwDiagCtrl_CounterWrEnable_MSB 0x3D
+#define QIB_7322_HwDiagCtrl_CounterWrEnable_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_CounterDisable_LSB 0x3C
+#define QIB_7322_HwDiagCtrl_CounterDisable_MSB 0x3C
+#define QIB_7322_HwDiagCtrl_CounterDisable_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_forcePCIeBusParity_LSB 0x1F
+#define QIB_7322_HwDiagCtrl_forcePCIeBusParity_MSB 0x22
+#define QIB_7322_HwDiagCtrl_forcePCIeBusParity_RMASK 0xF
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_1_LSB 0xF
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_1_MSB 0xF
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_1_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_1_LSB 0xE
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_1_MSB 0xE
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_1_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_0_LSB 0xD
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_0_MSB 0xD
+#define QIB_7322_HwDiagCtrl_ForceIBCBusFromSPCParityErr_0_RMASK 0x1
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_0_LSB 0xC
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_0_MSB 0xC
+#define QIB_7322_HwDiagCtrl_ForceIBCBusToSPCParityErr_0_RMASK 0x1
+
+#define QIB_7322_EXTStatus_OFFS 0xC0
+#define QIB_7322_EXTStatus_DEF 0x000000000000X000
+#define QIB_7322_EXTStatus_GPIOIn_LSB 0x30
+#define QIB_7322_EXTStatus_GPIOIn_MSB 0x3F
+#define QIB_7322_EXTStatus_GPIOIn_RMASK 0xFFFF
+#define QIB_7322_EXTStatus_MemBISTDisabled_LSB 0xF
+#define QIB_7322_EXTStatus_MemBISTDisabled_MSB 0xF
+#define QIB_7322_EXTStatus_MemBISTDisabled_RMASK 0x1
+#define QIB_7322_EXTStatus_MemBISTEndTest_LSB 0xE
+#define QIB_7322_EXTStatus_MemBISTEndTest_MSB 0xE
+#define QIB_7322_EXTStatus_MemBISTEndTest_RMASK 0x1
+
+#define QIB_7322_EXTCtrl_OFFS 0xC8
+#define QIB_7322_EXTCtrl_DEF 0x0000000000000000
+#define QIB_7322_EXTCtrl_GPIOOe_LSB 0x30
+#define QIB_7322_EXTCtrl_GPIOOe_MSB 0x3F
+#define QIB_7322_EXTCtrl_GPIOOe_RMASK 0xFFFF
+#define QIB_7322_EXTCtrl_GPIOInvert_LSB 0x20
+#define QIB_7322_EXTCtrl_GPIOInvert_MSB 0x2F
+#define QIB_7322_EXTCtrl_GPIOInvert_RMASK 0xFFFF
+#define QIB_7322_EXTCtrl_LEDPort1GreenOn_LSB 0x3
+#define QIB_7322_EXTCtrl_LEDPort1GreenOn_MSB 0x3
+#define QIB_7322_EXTCtrl_LEDPort1GreenOn_RMASK 0x1
+#define QIB_7322_EXTCtrl_LEDPort1YellowOn_LSB 0x2
+#define QIB_7322_EXTCtrl_LEDPort1YellowOn_MSB 0x2
+#define QIB_7322_EXTCtrl_LEDPort1YellowOn_RMASK 0x1
+#define QIB_7322_EXTCtrl_LEDPort0GreenOn_LSB 0x1
+#define QIB_7322_EXTCtrl_LEDPort0GreenOn_MSB 0x1
+#define QIB_7322_EXTCtrl_LEDPort0GreenOn_RMASK 0x1
+#define QIB_7322_EXTCtrl_LEDPort0YellowOn_LSB 0x0
+#define QIB_7322_EXTCtrl_LEDPort0YellowOn_MSB 0x0
+#define QIB_7322_EXTCtrl_LEDPort0YellowOn_RMASK 0x1
+
+#define QIB_7322_GPIOOut_OFFS 0xE0
+#define QIB_7322_GPIOOut_DEF 0x0000000000000000
+
+#define QIB_7322_GPIOMask_OFFS 0xE8
+#define QIB_7322_GPIOMask_DEF 0x0000000000000000
+
+#define QIB_7322_GPIOStatus_OFFS 0xF0
+#define QIB_7322_GPIOStatus_DEF 0x0000000000000000
+
+#define QIB_7322_GPIOClear_OFFS 0xF8
+#define QIB_7322_GPIOClear_DEF 0x0000000000000000
+
+#define QIB_7322_RcvCtrl_OFFS 0x100
+#define QIB_7322_RcvCtrl_DEF 0x0000000000000000
+#define QIB_7322_RcvCtrl_TidReDirect_LSB 0x30
+#define QIB_7322_RcvCtrl_TidReDirect_MSB 0x3F
+#define QIB_7322_RcvCtrl_TidReDirect_RMASK 0xFFFF
+#define QIB_7322_RcvCtrl_TailUpd_LSB 0x2F
+#define QIB_7322_RcvCtrl_TailUpd_MSB 0x2F
+#define QIB_7322_RcvCtrl_TailUpd_RMASK 0x1
+#define QIB_7322_RcvCtrl_XrcTypeCode_LSB 0x2C
+#define QIB_7322_RcvCtrl_XrcTypeCode_MSB 0x2E
+#define QIB_7322_RcvCtrl_XrcTypeCode_RMASK 0x7
+#define QIB_7322_RcvCtrl_TidFlowEnable_LSB 0x2B
+#define QIB_7322_RcvCtrl_TidFlowEnable_MSB 0x2B
+#define QIB_7322_RcvCtrl_TidFlowEnable_RMASK 0x1
+#define QIB_7322_RcvCtrl_ContextCfg_LSB 0x29
+#define QIB_7322_RcvCtrl_ContextCfg_MSB 0x2A
+#define QIB_7322_RcvCtrl_ContextCfg_RMASK 0x3
+#define QIB_7322_RcvCtrl_IntrAvail_LSB 0x14
+#define QIB_7322_RcvCtrl_IntrAvail_MSB 0x25
+#define QIB_7322_RcvCtrl_IntrAvail_RMASK 0x3FFFF
+#define QIB_7322_RcvCtrl_dontDropRHQFull_LSB 0x0
+#define QIB_7322_RcvCtrl_dontDropRHQFull_MSB 0x11
+#define QIB_7322_RcvCtrl_dontDropRHQFull_RMASK 0x3FFFF
+
+#define QIB_7322_RcvHdrSize_OFFS 0x110
+#define QIB_7322_RcvHdrSize_DEF 0x0000000000000000
+
+#define QIB_7322_RcvHdrCnt_OFFS 0x118
+#define QIB_7322_RcvHdrCnt_DEF 0x0000000000000000
+
+#define QIB_7322_RcvHdrEntSize_OFFS 0x120
+#define QIB_7322_RcvHdrEntSize_DEF 0x0000000000000000
+
+#define QIB_7322_RcvTIDBase_OFFS 0x128
+#define QIB_7322_RcvTIDBase_DEF 0x0000000000050000
+
+#define QIB_7322_RcvTIDCnt_OFFS 0x130
+#define QIB_7322_RcvTIDCnt_DEF 0x0000000000000200
+
+#define QIB_7322_RcvEgrBase_OFFS 0x138
+#define QIB_7322_RcvEgrBase_DEF 0x0000000000014000
+
+#define QIB_7322_RcvEgrCnt_OFFS 0x140
+#define QIB_7322_RcvEgrCnt_DEF 0x0000000000001000
+
+#define QIB_7322_RcvBufBase_OFFS 0x148
+#define QIB_7322_RcvBufBase_DEF 0x0000000000080000
+
+#define QIB_7322_RcvBufSize_OFFS 0x150
+#define QIB_7322_RcvBufSize_DEF 0x0000000000005000
+
+#define QIB_7322_RxIntMemBase_OFFS 0x158
+#define QIB_7322_RxIntMemBase_DEF 0x0000000000077000
+
+#define QIB_7322_RxIntMemSize_OFFS 0x160
+#define QIB_7322_RxIntMemSize_DEF 0x0000000000007000
+
+#define QIB_7322_feature_mask_OFFS 0x190
+#define QIB_7322_feature_mask_DEF 0x00000000000000XX
+
+#define QIB_7322_active_feature_mask_OFFS 0x198
+#define QIB_7322_active_feature_mask_DEF 0x00000000000000XX
+#define QIB_7322_active_feature_mask_Port1_QDR_Enabled_LSB 0x5
+#define QIB_7322_active_feature_mask_Port1_QDR_Enabled_MSB 0x5
+#define QIB_7322_active_feature_mask_Port1_QDR_Enabled_RMASK 0x1
+#define QIB_7322_active_feature_mask_Port1_DDR_Enabled_LSB 0x4
+#define QIB_7322_active_feature_mask_Port1_DDR_Enabled_MSB 0x4
+#define QIB_7322_active_feature_mask_Port1_DDR_Enabled_RMASK 0x1
+#define QIB_7322_active_feature_mask_Port1_SDR_Enabled_LSB 0x3
+#define QIB_7322_active_feature_mask_Port1_SDR_Enabled_MSB 0x3
+#define QIB_7322_active_feature_mask_Port1_SDR_Enabled_RMASK 0x1
+#define QIB_7322_active_feature_mask_Port0_QDR_Enabled_LSB 0x2
+#define QIB_7322_active_feature_mask_Port0_QDR_Enabled_MSB 0x2
+#define QIB_7322_active_feature_mask_Port0_QDR_Enabled_RMASK 0x1
+#define QIB_7322_active_feature_mask_Port0_DDR_Enabled_LSB 0x1
+#define QIB_7322_active_feature_mask_Port0_DDR_Enabled_MSB 0x1
+#define QIB_7322_active_feature_mask_Port0_DDR_Enabled_RMASK 0x1
+#define QIB_7322_active_feature_mask_Port0_SDR_Enabled_LSB 0x0
+#define QIB_7322_active_feature_mask_Port0_SDR_Enabled_MSB 0x0
+#define QIB_7322_active_feature_mask_Port0_SDR_Enabled_RMASK 0x1
+
+#define QIB_7322_SendCtrl_OFFS 0x1C0
+#define QIB_7322_SendCtrl_DEF 0x0000000000000000
+#define QIB_7322_SendCtrl_Disarm_LSB 0x1F
+#define QIB_7322_SendCtrl_Disarm_MSB 0x1F
+#define QIB_7322_SendCtrl_Disarm_RMASK 0x1
+#define QIB_7322_SendCtrl_SendBufAvailPad64Byte_LSB 0x1D
+#define QIB_7322_SendCtrl_SendBufAvailPad64Byte_MSB 0x1D
+#define QIB_7322_SendCtrl_SendBufAvailPad64Byte_RMASK 0x1
+#define QIB_7322_SendCtrl_AvailUpdThld_LSB 0x18
+#define QIB_7322_SendCtrl_AvailUpdThld_MSB 0x1C
+#define QIB_7322_SendCtrl_AvailUpdThld_RMASK 0x1F
+#define QIB_7322_SendCtrl_DisarmSendBuf_LSB 0x10
+#define QIB_7322_SendCtrl_DisarmSendBuf_MSB 0x17
+#define QIB_7322_SendCtrl_DisarmSendBuf_RMASK 0xFF
+#define QIB_7322_SendCtrl_SpecialTriggerEn_LSB 0x4
+#define QIB_7322_SendCtrl_SpecialTriggerEn_MSB 0x4
+#define QIB_7322_SendCtrl_SpecialTriggerEn_RMASK 0x1
+#define QIB_7322_SendCtrl_SendBufAvailUpd_LSB 0x2
+#define QIB_7322_SendCtrl_SendBufAvailUpd_MSB 0x2
+#define QIB_7322_SendCtrl_SendBufAvailUpd_RMASK 0x1
+#define QIB_7322_SendCtrl_SendIntBufAvail_LSB 0x1
+#define QIB_7322_SendCtrl_SendIntBufAvail_MSB 0x1
+#define QIB_7322_SendCtrl_SendIntBufAvail_RMASK 0x1
+
+#define QIB_7322_SendBufBase_OFFS 0x1C8
+#define QIB_7322_SendBufBase_DEF 0x0018000000100000
+#define QIB_7322_SendBufBase_BaseAddr_LargePIO_LSB 0x20
+#define QIB_7322_SendBufBase_BaseAddr_LargePIO_MSB 0x34
+#define QIB_7322_SendBufBase_BaseAddr_LargePIO_RMASK 0x1FFFFF
+#define QIB_7322_SendBufBase_BaseAddr_SmallPIO_LSB 0x0
+#define QIB_7322_SendBufBase_BaseAddr_SmallPIO_MSB 0x14
+#define QIB_7322_SendBufBase_BaseAddr_SmallPIO_RMASK 0x1FFFFF
+
+#define QIB_7322_SendBufSize_OFFS 0x1D0
+#define QIB_7322_SendBufSize_DEF 0x0000108000000880
+#define QIB_7322_SendBufSize_Size_LargePIO_LSB 0x20
+#define QIB_7322_SendBufSize_Size_LargePIO_MSB 0x2C
+#define QIB_7322_SendBufSize_Size_LargePIO_RMASK 0x1FFF
+#define QIB_7322_SendBufSize_Size_SmallPIO_LSB 0x0
+#define QIB_7322_SendBufSize_Size_SmallPIO_MSB 0xB
+#define QIB_7322_SendBufSize_Size_SmallPIO_RMASK 0xFFF
+
+#define QIB_7322_SendBufCnt_OFFS 0x1D8
+#define QIB_7322_SendBufCnt_DEF 0x0000002000000080
+#define QIB_7322_SendBufCnt_Num_LargeBuffers_LSB 0x20
+#define QIB_7322_SendBufCnt_Num_LargeBuffers_MSB 0x25
+#define QIB_7322_SendBufCnt_Num_LargeBuffers_RMASK 0x3F
+#define QIB_7322_SendBufCnt_Num_SmallBuffers_LSB 0x0
+#define QIB_7322_SendBufCnt_Num_SmallBuffers_MSB 0x8
+#define QIB_7322_SendBufCnt_Num_SmallBuffers_RMASK 0x1FF
+
+#define QIB_7322_SendBufAvailAddr_OFFS 0x1E0
+#define QIB_7322_SendBufAvailAddr_DEF 0x0000000000000000
+#define QIB_7322_SendBufAvailAddr_SendBufAvailAddr_LSB 0x6
+#define QIB_7322_SendBufAvailAddr_SendBufAvailAddr_MSB 0x27
+#define QIB_7322_SendBufAvailAddr_SendBufAvailAddr_RMASK 0x3FFFFFFFF
+
+#define QIB_7322_SendBufErr0_OFFS 0x240
+#define QIB_7322_SendBufErr0_DEF 0x0000000000000000
+#define QIB_7322_SendBufErr0_SendBufErr_63_0_LSB 0x0
+#define QIB_7322_SendBufErr0_SendBufErr_63_0_MSB 0x3F
+#define QIB_7322_SendBufErr0_SendBufErr_63_0_RMASK 0x0
+
+#define QIB_7322_AvailUpdCount_OFFS 0x268
+#define QIB_7322_AvailUpdCount_DEF 0x0000000000000000
+#define QIB_7322_AvailUpdCount_AvailUpdCount_LSB 0x0
+#define QIB_7322_AvailUpdCount_AvailUpdCount_MSB 0x4
+#define QIB_7322_AvailUpdCount_AvailUpdCount_RMASK 0x1F
+
+#define QIB_7322_RcvHdrAddr0_OFFS 0x280
+#define QIB_7322_RcvHdrAddr0_DEF 0x0000000000000000
+#define QIB_7322_RcvHdrAddr0_RcvHdrAddr_LSB 0x2
+#define QIB_7322_RcvHdrAddr0_RcvHdrAddr_MSB 0x27
+#define QIB_7322_RcvHdrAddr0_RcvHdrAddr_RMASK 0x3FFFFFFFFF
+
+#define QIB_7322_RcvHdrTailAddr0_OFFS 0x340
+#define QIB_7322_RcvHdrTailAddr0_DEF 0x0000000000000000
+#define QIB_7322_RcvHdrTailAddr0_RcvHdrTailAddr_LSB 0x2
+#define QIB_7322_RcvHdrTailAddr0_RcvHdrTailAddr_MSB 0x27
+#define QIB_7322_RcvHdrTailAddr0_RcvHdrTailAddr_RMASK 0x3FFFFFFFFF
+
+#define QIB_7322_ahb_access_ctrl_OFFS 0x460
+#define QIB_7322_ahb_access_ctrl_DEF 0x0000000000000000
+#define QIB_7322_ahb_access_ctrl_sw_sel_ahb_trgt_LSB 0x1
+#define QIB_7322_ahb_access_ctrl_sw_sel_ahb_trgt_MSB 0x2
+#define QIB_7322_ahb_access_ctrl_sw_sel_ahb_trgt_RMASK 0x3
+#define QIB_7322_ahb_access_ctrl_sw_ahb_sel_LSB 0x0
+#define QIB_7322_ahb_access_ctrl_sw_ahb_sel_MSB 0x0
+#define QIB_7322_ahb_access_ctrl_sw_ahb_sel_RMASK 0x1
+
+#define QIB_7322_ahb_transaction_reg_OFFS 0x468
+#define QIB_7322_ahb_transaction_reg_DEF 0x0000000080000000
+#define QIB_7322_ahb_transaction_reg_ahb_data_LSB 0x20
+#define QIB_7322_ahb_transaction_reg_ahb_data_MSB 0x3F
+#define QIB_7322_ahb_transaction_reg_ahb_data_RMASK 0xFFFFFFFF
+#define QIB_7322_ahb_transaction_reg_ahb_rdy_LSB 0x1F
+#define QIB_7322_ahb_transaction_reg_ahb_rdy_MSB 0x1F
+#define QIB_7322_ahb_transaction_reg_ahb_rdy_RMASK 0x1
+#define QIB_7322_ahb_transaction_reg_ahb_req_err_LSB 0x1E
+#define QIB_7322_ahb_transaction_reg_ahb_req_err_MSB 0x1E
+#define QIB_7322_ahb_transaction_reg_ahb_req_err_RMASK 0x1
+#define QIB_7322_ahb_transaction_reg_write_not_read_LSB 0x1B
+#define QIB_7322_ahb_transaction_reg_write_not_read_MSB 0x1B
+#define QIB_7322_ahb_transaction_reg_write_not_read_RMASK 0x1
+#define QIB_7322_ahb_transaction_reg_ahb_address_LSB 0x10
+#define QIB_7322_ahb_transaction_reg_ahb_address_MSB 0x1A
+#define QIB_7322_ahb_transaction_reg_ahb_address_RMASK 0x7FF
+
+#define QIB_7322_SPC_JTAG_ACCESS_REG_OFFS 0x470
+#define QIB_7322_SPC_JTAG_ACCESS_REG_DEF 0x0000000000000001
+#define QIB_7322_SPC_JTAG_ACCESS_REG_SPC_JTAG_ACCESS_EN_LSB 0xA
+#define QIB_7322_SPC_JTAG_ACCESS_REG_SPC_JTAG_ACCESS_EN_MSB 0xA
+#define QIB_7322_SPC_JTAG_ACCESS_REG_SPC_JTAG_ACCESS_EN_RMASK 0x1
+#define QIB_7322_SPC_JTAG_ACCESS_REG_bist_en_LSB 0x5
+#define QIB_7322_SPC_JTAG_ACCESS_REG_bist_en_MSB 0x9
+#define QIB_7322_SPC_JTAG_ACCESS_REG_bist_en_RMASK 0x1F
+#define QIB_7322_SPC_JTAG_ACCESS_REG_opcode_LSB 0x3
+#define QIB_7322_SPC_JTAG_ACCESS_REG_opcode_MSB 0x4
+#define QIB_7322_SPC_JTAG_ACCESS_REG_opcode_RMASK 0x3
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdi_LSB 0x2
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdi_MSB 0x2
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdi_RMASK 0x1
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdo_LSB 0x1
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdo_MSB 0x1
+#define QIB_7322_SPC_JTAG_ACCESS_REG_tdo_RMASK 0x1
+#define QIB_7322_SPC_JTAG_ACCESS_REG_rdy_LSB 0x0
+#define QIB_7322_SPC_JTAG_ACCESS_REG_rdy_MSB 0x0
+#define QIB_7322_SPC_JTAG_ACCESS_REG_rdy_RMASK 0x1
+
+#define QIB_7322_SendCheckMask0_OFFS 0x4C0
+#define QIB_7322_SendCheckMask0_DEF 0x0000000000000000
+#define QIB_7322_SendCheckMask0_SendCheckMask_63_32_LSB 0x0
+#define QIB_7322_SendCheckMask0_SendCheckMask_63_32_MSB 0x3F
+#define QIB_7322_SendCheckMask0_SendCheckMask_63_32_RMASK 0x0
+
+#define QIB_7322_SendGRHCheckMask0_OFFS 0x4E0
+#define QIB_7322_SendGRHCheckMask0_DEF 0x0000000000000000
+#define QIB_7322_SendGRHCheckMask0_SendGRHCheckMask_63_32_LSB 0x0
+#define QIB_7322_SendGRHCheckMask0_SendGRHCheckMask_63_32_MSB 0x3F
+#define QIB_7322_SendGRHCheckMask0_SendGRHCheckMask_63_32_RMASK 0x0
+
+#define QIB_7322_SendIBPacketMask0_OFFS 0x500
+#define QIB_7322_SendIBPacketMask0_DEF 0x0000000000000000
+#define QIB_7322_SendIBPacketMask0_SendIBPacketMask_63_32_LSB 0x0
+#define QIB_7322_SendIBPacketMask0_SendIBPacketMask_63_32_MSB 0x3F
+#define QIB_7322_SendIBPacketMask0_SendIBPacketMask_63_32_RMASK 0x0
+
+#define QIB_7322_IntRedirect0_OFFS 0x540
+#define QIB_7322_IntRedirect0_DEF 0x0000000000000000
+#define QIB_7322_IntRedirect0_vec11_LSB 0x37
+#define QIB_7322_IntRedirect0_vec11_MSB 0x3B
+#define QIB_7322_IntRedirect0_vec11_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec10_LSB 0x32
+#define QIB_7322_IntRedirect0_vec10_MSB 0x36
+#define QIB_7322_IntRedirect0_vec10_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec9_LSB 0x2D
+#define QIB_7322_IntRedirect0_vec9_MSB 0x31
+#define QIB_7322_IntRedirect0_vec9_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec8_LSB 0x28
+#define QIB_7322_IntRedirect0_vec8_MSB 0x2C
+#define QIB_7322_IntRedirect0_vec8_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec7_LSB 0x23
+#define QIB_7322_IntRedirect0_vec7_MSB 0x27
+#define QIB_7322_IntRedirect0_vec7_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec6_LSB 0x1E
+#define QIB_7322_IntRedirect0_vec6_MSB 0x22
+#define QIB_7322_IntRedirect0_vec6_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec5_LSB 0x19
+#define QIB_7322_IntRedirect0_vec5_MSB 0x1D
+#define QIB_7322_IntRedirect0_vec5_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec4_LSB 0x14
+#define QIB_7322_IntRedirect0_vec4_MSB 0x18
+#define QIB_7322_IntRedirect0_vec4_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec3_LSB 0xF
+#define QIB_7322_IntRedirect0_vec3_MSB 0x13
+#define QIB_7322_IntRedirect0_vec3_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec2_LSB 0xA
+#define QIB_7322_IntRedirect0_vec2_MSB 0xE
+#define QIB_7322_IntRedirect0_vec2_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec1_LSB 0x5
+#define QIB_7322_IntRedirect0_vec1_MSB 0x9
+#define QIB_7322_IntRedirect0_vec1_RMASK 0x1F
+#define QIB_7322_IntRedirect0_vec0_LSB 0x0
+#define QIB_7322_IntRedirect0_vec0_MSB 0x4
+#define QIB_7322_IntRedirect0_vec0_RMASK 0x1F
+
+#define QIB_7322_Int_Granted_OFFS 0x570
+#define QIB_7322_Int_Granted_DEF 0x0000000000000000
+
+#define QIB_7322_vec_clr_without_int_OFFS 0x578
+#define QIB_7322_vec_clr_without_int_DEF 0x0000000000000000
+
+#define QIB_7322_DCACtrlA_OFFS 0x580
+#define QIB_7322_DCACtrlA_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlA_SendDMAHead1DCAEnable_LSB 0x4
+#define QIB_7322_DCACtrlA_SendDMAHead1DCAEnable_MSB 0x4
+#define QIB_7322_DCACtrlA_SendDMAHead1DCAEnable_RMASK 0x1
+#define QIB_7322_DCACtrlA_SendDMAHead0DCAEnable_LSB 0x3
+#define QIB_7322_DCACtrlA_SendDMAHead0DCAEnable_MSB 0x3
+#define QIB_7322_DCACtrlA_SendDMAHead0DCAEnable_RMASK 0x1
+#define QIB_7322_DCACtrlA_RcvTailUpdDCAEnable_LSB 0x2
+#define QIB_7322_DCACtrlA_RcvTailUpdDCAEnable_MSB 0x2
+#define QIB_7322_DCACtrlA_RcvTailUpdDCAEnable_RMASK 0x1
+#define QIB_7322_DCACtrlA_EagerDCAEnable_LSB 0x1
+#define QIB_7322_DCACtrlA_EagerDCAEnable_MSB 0x1
+#define QIB_7322_DCACtrlA_EagerDCAEnable_RMASK 0x1
+#define QIB_7322_DCACtrlA_RcvHdrqDCAEnable_LSB 0x0
+#define QIB_7322_DCACtrlA_RcvHdrqDCAEnable_MSB 0x0
+#define QIB_7322_DCACtrlA_RcvHdrqDCAEnable_RMASK 0x1
+
+#define QIB_7322_DCACtrlB_OFFS 0x588
+#define QIB_7322_DCACtrlB_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAXfrCnt_LSB 0x36
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAXfrCnt_MSB 0x3B
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAOPH_LSB 0x2E
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAOPH_MSB 0x35
+#define QIB_7322_DCACtrlB_RcvHdrq3DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAXfrCnt_LSB 0x28
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAXfrCnt_MSB 0x2D
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAOPH_LSB 0x20
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAOPH_MSB 0x27
+#define QIB_7322_DCACtrlB_RcvHdrq2DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAXfrCnt_LSB 0x16
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAXfrCnt_MSB 0x1B
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAOPH_LSB 0xE
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAOPH_MSB 0x15
+#define QIB_7322_DCACtrlB_RcvHdrq1DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAXfrCnt_LSB 0x8
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAXfrCnt_MSB 0xD
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAOPH_LSB 0x0
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAOPH_MSB 0x7
+#define QIB_7322_DCACtrlB_RcvHdrq0DCAOPH_RMASK 0xFF
+
+#define QIB_7322_DCACtrlC_OFFS 0x590
+#define QIB_7322_DCACtrlC_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAXfrCnt_LSB 0x36
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAXfrCnt_MSB 0x3B
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAOPH_LSB 0x2E
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAOPH_MSB 0x35
+#define QIB_7322_DCACtrlC_RcvHdrq7DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAXfrCnt_LSB 0x28
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAXfrCnt_MSB 0x2D
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAOPH_LSB 0x20
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAOPH_MSB 0x27
+#define QIB_7322_DCACtrlC_RcvHdrq6DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAXfrCnt_LSB 0x16
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAXfrCnt_MSB 0x1B
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAOPH_LSB 0xE
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAOPH_MSB 0x15
+#define QIB_7322_DCACtrlC_RcvHdrq5DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAXfrCnt_LSB 0x8
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAXfrCnt_MSB 0xD
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAOPH_LSB 0x0
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAOPH_MSB 0x7
+#define QIB_7322_DCACtrlC_RcvHdrq4DCAOPH_RMASK 0xFF
+
+#define QIB_7322_DCACtrlD_OFFS 0x598
+#define QIB_7322_DCACtrlD_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAXfrCnt_LSB 0x36
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAXfrCnt_MSB 0x3B
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAOPH_LSB 0x2E
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAOPH_MSB 0x35
+#define QIB_7322_DCACtrlD_RcvHdrq11DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAXfrCnt_LSB 0x28
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAXfrCnt_MSB 0x2D
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAOPH_LSB 0x20
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAOPH_MSB 0x27
+#define QIB_7322_DCACtrlD_RcvHdrq10DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAXfrCnt_LSB 0x16
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAXfrCnt_MSB 0x1B
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAOPH_LSB 0xE
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAOPH_MSB 0x15
+#define QIB_7322_DCACtrlD_RcvHdrq9DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAXfrCnt_LSB 0x8
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAXfrCnt_MSB 0xD
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAOPH_LSB 0x0
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAOPH_MSB 0x7
+#define QIB_7322_DCACtrlD_RcvHdrq8DCAOPH_RMASK 0xFF
+
+#define QIB_7322_DCACtrlE_OFFS 0x5A0
+#define QIB_7322_DCACtrlE_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAXfrCnt_LSB 0x36
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAXfrCnt_MSB 0x3B
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAOPH_LSB 0x2E
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAOPH_MSB 0x35
+#define QIB_7322_DCACtrlE_RcvHdrq15DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAXfrCnt_LSB 0x28
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAXfrCnt_MSB 0x2D
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAOPH_LSB 0x20
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAOPH_MSB 0x27
+#define QIB_7322_DCACtrlE_RcvHdrq14DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAXfrCnt_LSB 0x16
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAXfrCnt_MSB 0x1B
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAOPH_LSB 0xE
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAOPH_MSB 0x15
+#define QIB_7322_DCACtrlE_RcvHdrq13DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAXfrCnt_LSB 0x8
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAXfrCnt_MSB 0xD
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAOPH_LSB 0x0
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAOPH_MSB 0x7
+#define QIB_7322_DCACtrlE_RcvHdrq12DCAOPH_RMASK 0xFF
+
+#define QIB_7322_DCACtrlF_OFFS 0x5A8
+#define QIB_7322_DCACtrlF_DEF 0x0000000000000000
+#define QIB_7322_DCACtrlF_SendDma1DCAOPH_LSB 0x28
+#define QIB_7322_DCACtrlF_SendDma1DCAOPH_MSB 0x2F
+#define QIB_7322_DCACtrlF_SendDma1DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlF_SendDma0DCAOPH_LSB 0x20
+#define QIB_7322_DCACtrlF_SendDma0DCAOPH_MSB 0x27
+#define QIB_7322_DCACtrlF_SendDma0DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAXfrCnt_LSB 0x16
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAXfrCnt_MSB 0x1B
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAOPH_LSB 0xE
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAOPH_MSB 0x15
+#define QIB_7322_DCACtrlF_RcvHdrq17DCAOPH_RMASK 0xFF
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAXfrCnt_LSB 0x8
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAXfrCnt_MSB 0xD
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAXfrCnt_RMASK 0x3F
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAOPH_LSB 0x0
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAOPH_MSB 0x7
+#define QIB_7322_DCACtrlF_RcvHdrq16DCAOPH_RMASK 0xFF
+
+#define QIB_7322_RcvAvailTimeOut0_OFFS 0xC00
+#define QIB_7322_RcvAvailTimeOut0_DEF 0x0000000000000000
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOCount_LSB 0x10
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOCount_MSB 0x1F
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOCount_RMASK 0xFFFF
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOReload_LSB 0x0
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOReload_MSB 0xF
+#define QIB_7322_RcvAvailTimeOut0_RcvAvailTOReload_RMASK 0xFFFF
+
+#define QIB_7322_CntrRegBase_0_OFFS 0x1028
+#define QIB_7322_CntrRegBase_0_DEF 0x0000000000012000
+
+#define QIB_7322_ErrMask_0_OFFS 0x1080
+#define QIB_7322_ErrMask_0_DEF 0x0000000000000000
+#define QIB_7322_ErrMask_0_IBStatusChangedMask_LSB 0x3A
+#define QIB_7322_ErrMask_0_IBStatusChangedMask_MSB 0x3A
+#define QIB_7322_ErrMask_0_IBStatusChangedMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SHeadersErrMask_LSB 0x39
+#define QIB_7322_ErrMask_0_SHeadersErrMask_MSB 0x39
+#define QIB_7322_ErrMask_0_SHeadersErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_VL15BufMisuseErrMask_LSB 0x36
+#define QIB_7322_ErrMask_0_VL15BufMisuseErrMask_MSB 0x36
+#define QIB_7322_ErrMask_0_VL15BufMisuseErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaHaltErrMask_LSB 0x31
+#define QIB_7322_ErrMask_0_SDmaHaltErrMask_MSB 0x31
+#define QIB_7322_ErrMask_0_SDmaHaltErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaDescAddrMisalignErrMask_LSB 0x30
+#define QIB_7322_ErrMask_0_SDmaDescAddrMisalignErrMask_MSB 0x30
+#define QIB_7322_ErrMask_0_SDmaDescAddrMisalignErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaUnexpDataErrMask_LSB 0x2F
+#define QIB_7322_ErrMask_0_SDmaUnexpDataErrMask_MSB 0x2F
+#define QIB_7322_ErrMask_0_SDmaUnexpDataErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaMissingDwErrMask_LSB 0x2E
+#define QIB_7322_ErrMask_0_SDmaMissingDwErrMask_MSB 0x2E
+#define QIB_7322_ErrMask_0_SDmaMissingDwErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaDwEnErrMask_LSB 0x2D
+#define QIB_7322_ErrMask_0_SDmaDwEnErrMask_MSB 0x2D
+#define QIB_7322_ErrMask_0_SDmaDwEnErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaRpyTagErrMask_LSB 0x2C
+#define QIB_7322_ErrMask_0_SDmaRpyTagErrMask_MSB 0x2C
+#define QIB_7322_ErrMask_0_SDmaRpyTagErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDma1stDescErrMask_LSB 0x2B
+#define QIB_7322_ErrMask_0_SDma1stDescErrMask_MSB 0x2B
+#define QIB_7322_ErrMask_0_SDma1stDescErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaBaseErrMask_LSB 0x2A
+#define QIB_7322_ErrMask_0_SDmaBaseErrMask_MSB 0x2A
+#define QIB_7322_ErrMask_0_SDmaBaseErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaTailOutOfBoundErrMask_LSB 0x29
+#define QIB_7322_ErrMask_0_SDmaTailOutOfBoundErrMask_MSB 0x29
+#define QIB_7322_ErrMask_0_SDmaTailOutOfBoundErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaOutOfBoundErrMask_LSB 0x28
+#define QIB_7322_ErrMask_0_SDmaOutOfBoundErrMask_MSB 0x28
+#define QIB_7322_ErrMask_0_SDmaOutOfBoundErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SDmaGenMismatchErrMask_LSB 0x27
+#define QIB_7322_ErrMask_0_SDmaGenMismatchErrMask_MSB 0x27
+#define QIB_7322_ErrMask_0_SDmaGenMismatchErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendBufMisuseErrMask_LSB 0x26
+#define QIB_7322_ErrMask_0_SendBufMisuseErrMask_MSB 0x26
+#define QIB_7322_ErrMask_0_SendBufMisuseErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendUnsupportedVLErrMask_LSB 0x25
+#define QIB_7322_ErrMask_0_SendUnsupportedVLErrMask_MSB 0x25
+#define QIB_7322_ErrMask_0_SendUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendUnexpectedPktNumErrMask_LSB 0x24
+#define QIB_7322_ErrMask_0_SendUnexpectedPktNumErrMask_MSB 0x24
+#define QIB_7322_ErrMask_0_SendUnexpectedPktNumErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendDroppedDataPktErrMask_LSB 0x22
+#define QIB_7322_ErrMask_0_SendDroppedDataPktErrMask_MSB 0x22
+#define QIB_7322_ErrMask_0_SendDroppedDataPktErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendDroppedSmpPktErrMask_LSB 0x21
+#define QIB_7322_ErrMask_0_SendDroppedSmpPktErrMask_MSB 0x21
+#define QIB_7322_ErrMask_0_SendDroppedSmpPktErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendPktLenErrMask_LSB 0x20
+#define QIB_7322_ErrMask_0_SendPktLenErrMask_MSB 0x20
+#define QIB_7322_ErrMask_0_SendPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendUnderRunErrMask_LSB 0x1F
+#define QIB_7322_ErrMask_0_SendUnderRunErrMask_MSB 0x1F
+#define QIB_7322_ErrMask_0_SendUnderRunErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendMaxPktLenErrMask_LSB 0x1E
+#define QIB_7322_ErrMask_0_SendMaxPktLenErrMask_MSB 0x1E
+#define QIB_7322_ErrMask_0_SendMaxPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_SendMinPktLenErrMask_LSB 0x1D
+#define QIB_7322_ErrMask_0_SendMinPktLenErrMask_MSB 0x1D
+#define QIB_7322_ErrMask_0_SendMinPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvIBLostLinkErrMask_LSB 0x11
+#define QIB_7322_ErrMask_0_RcvIBLostLinkErrMask_MSB 0x11
+#define QIB_7322_ErrMask_0_RcvIBLostLinkErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvHdrErrMask_LSB 0x10
+#define QIB_7322_ErrMask_0_RcvHdrErrMask_MSB 0x10
+#define QIB_7322_ErrMask_0_RcvHdrErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvHdrLenErrMask_LSB 0xF
+#define QIB_7322_ErrMask_0_RcvHdrLenErrMask_MSB 0xF
+#define QIB_7322_ErrMask_0_RcvHdrLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvBadTidErrMask_LSB 0xE
+#define QIB_7322_ErrMask_0_RcvBadTidErrMask_MSB 0xE
+#define QIB_7322_ErrMask_0_RcvBadTidErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvBadVersionErrMask_LSB 0xB
+#define QIB_7322_ErrMask_0_RcvBadVersionErrMask_MSB 0xB
+#define QIB_7322_ErrMask_0_RcvBadVersionErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvIBFlowErrMask_LSB 0xA
+#define QIB_7322_ErrMask_0_RcvIBFlowErrMask_MSB 0xA
+#define QIB_7322_ErrMask_0_RcvIBFlowErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvEBPErrMask_LSB 0x9
+#define QIB_7322_ErrMask_0_RcvEBPErrMask_MSB 0x9
+#define QIB_7322_ErrMask_0_RcvEBPErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvUnsupportedVLErrMask_LSB 0x8
+#define QIB_7322_ErrMask_0_RcvUnsupportedVLErrMask_MSB 0x8
+#define QIB_7322_ErrMask_0_RcvUnsupportedVLErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvUnexpectedCharErrMask_LSB 0x7
+#define QIB_7322_ErrMask_0_RcvUnexpectedCharErrMask_MSB 0x7
+#define QIB_7322_ErrMask_0_RcvUnexpectedCharErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvShortPktLenErrMask_LSB 0x6
+#define QIB_7322_ErrMask_0_RcvShortPktLenErrMask_MSB 0x6
+#define QIB_7322_ErrMask_0_RcvShortPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvLongPktLenErrMask_LSB 0x5
+#define QIB_7322_ErrMask_0_RcvLongPktLenErrMask_MSB 0x5
+#define QIB_7322_ErrMask_0_RcvLongPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvMaxPktLenErrMask_LSB 0x4
+#define QIB_7322_ErrMask_0_RcvMaxPktLenErrMask_MSB 0x4
+#define QIB_7322_ErrMask_0_RcvMaxPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvMinPktLenErrMask_LSB 0x3
+#define QIB_7322_ErrMask_0_RcvMinPktLenErrMask_MSB 0x3
+#define QIB_7322_ErrMask_0_RcvMinPktLenErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvICRCErrMask_LSB 0x2
+#define QIB_7322_ErrMask_0_RcvICRCErrMask_MSB 0x2
+#define QIB_7322_ErrMask_0_RcvICRCErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvVCRCErrMask_LSB 0x1
+#define QIB_7322_ErrMask_0_RcvVCRCErrMask_MSB 0x1
+#define QIB_7322_ErrMask_0_RcvVCRCErrMask_RMASK 0x1
+#define QIB_7322_ErrMask_0_RcvFormatErrMask_LSB 0x0
+#define QIB_7322_ErrMask_0_RcvFormatErrMask_MSB 0x0
+#define QIB_7322_ErrMask_0_RcvFormatErrMask_RMASK 0x1
+
+#define QIB_7322_ErrStatus_0_OFFS 0x1088
+#define QIB_7322_ErrStatus_0_DEF 0x0000000000000000
+#define QIB_7322_ErrStatus_0_IBStatusChanged_LSB 0x3A
+#define QIB_7322_ErrStatus_0_IBStatusChanged_MSB 0x3A
+#define QIB_7322_ErrStatus_0_IBStatusChanged_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SHeadersErr_LSB 0x39
+#define QIB_7322_ErrStatus_0_SHeadersErr_MSB 0x39
+#define QIB_7322_ErrStatus_0_SHeadersErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_VL15BufMisuseErr_LSB 0x36
+#define QIB_7322_ErrStatus_0_VL15BufMisuseErr_MSB 0x36
+#define QIB_7322_ErrStatus_0_VL15BufMisuseErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaHaltErr_LSB 0x31
+#define QIB_7322_ErrStatus_0_SDmaHaltErr_MSB 0x31
+#define QIB_7322_ErrStatus_0_SDmaHaltErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaDescAddrMisalignErr_LSB 0x30
+#define QIB_7322_ErrStatus_0_SDmaDescAddrMisalignErr_MSB 0x30
+#define QIB_7322_ErrStatus_0_SDmaDescAddrMisalignErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaUnexpDataErr_LSB 0x2F
+#define QIB_7322_ErrStatus_0_SDmaUnexpDataErr_MSB 0x2F
+#define QIB_7322_ErrStatus_0_SDmaUnexpDataErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaMissingDwErr_LSB 0x2E
+#define QIB_7322_ErrStatus_0_SDmaMissingDwErr_MSB 0x2E
+#define QIB_7322_ErrStatus_0_SDmaMissingDwErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaDwEnErr_LSB 0x2D
+#define QIB_7322_ErrStatus_0_SDmaDwEnErr_MSB 0x2D
+#define QIB_7322_ErrStatus_0_SDmaDwEnErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaRpyTagErr_LSB 0x2C
+#define QIB_7322_ErrStatus_0_SDmaRpyTagErr_MSB 0x2C
+#define QIB_7322_ErrStatus_0_SDmaRpyTagErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDma1stDescErr_LSB 0x2B
+#define QIB_7322_ErrStatus_0_SDma1stDescErr_MSB 0x2B
+#define QIB_7322_ErrStatus_0_SDma1stDescErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaBaseErr_LSB 0x2A
+#define QIB_7322_ErrStatus_0_SDmaBaseErr_MSB 0x2A
+#define QIB_7322_ErrStatus_0_SDmaBaseErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaTailOutOfBoundErr_LSB 0x29
+#define QIB_7322_ErrStatus_0_SDmaTailOutOfBoundErr_MSB 0x29
+#define QIB_7322_ErrStatus_0_SDmaTailOutOfBoundErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaOutOfBoundErr_LSB 0x28
+#define QIB_7322_ErrStatus_0_SDmaOutOfBoundErr_MSB 0x28
+#define QIB_7322_ErrStatus_0_SDmaOutOfBoundErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SDmaGenMismatchErr_LSB 0x27
+#define QIB_7322_ErrStatus_0_SDmaGenMismatchErr_MSB 0x27
+#define QIB_7322_ErrStatus_0_SDmaGenMismatchErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendBufMisuseErr_LSB 0x26
+#define QIB_7322_ErrStatus_0_SendBufMisuseErr_MSB 0x26
+#define QIB_7322_ErrStatus_0_SendBufMisuseErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendUnsupportedVLErr_LSB 0x25
+#define QIB_7322_ErrStatus_0_SendUnsupportedVLErr_MSB 0x25
+#define QIB_7322_ErrStatus_0_SendUnsupportedVLErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendUnexpectedPktNumErr_LSB 0x24
+#define QIB_7322_ErrStatus_0_SendUnexpectedPktNumErr_MSB 0x24
+#define QIB_7322_ErrStatus_0_SendUnexpectedPktNumErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendDroppedDataPktErr_LSB 0x22
+#define QIB_7322_ErrStatus_0_SendDroppedDataPktErr_MSB 0x22
+#define QIB_7322_ErrStatus_0_SendDroppedDataPktErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendDroppedSmpPktErr_LSB 0x21
+#define QIB_7322_ErrStatus_0_SendDroppedSmpPktErr_MSB 0x21
+#define QIB_7322_ErrStatus_0_SendDroppedSmpPktErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendPktLenErr_LSB 0x20
+#define QIB_7322_ErrStatus_0_SendPktLenErr_MSB 0x20
+#define QIB_7322_ErrStatus_0_SendPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendUnderRunErr_LSB 0x1F
+#define QIB_7322_ErrStatus_0_SendUnderRunErr_MSB 0x1F
+#define QIB_7322_ErrStatus_0_SendUnderRunErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendMaxPktLenErr_LSB 0x1E
+#define QIB_7322_ErrStatus_0_SendMaxPktLenErr_MSB 0x1E
+#define QIB_7322_ErrStatus_0_SendMaxPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_SendMinPktLenErr_LSB 0x1D
+#define QIB_7322_ErrStatus_0_SendMinPktLenErr_MSB 0x1D
+#define QIB_7322_ErrStatus_0_SendMinPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvIBLostLinkErr_LSB 0x11
+#define QIB_7322_ErrStatus_0_RcvIBLostLinkErr_MSB 0x11
+#define QIB_7322_ErrStatus_0_RcvIBLostLinkErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvHdrErr_LSB 0x10
+#define QIB_7322_ErrStatus_0_RcvHdrErr_MSB 0x10
+#define QIB_7322_ErrStatus_0_RcvHdrErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvHdrLenErr_LSB 0xF
+#define QIB_7322_ErrStatus_0_RcvHdrLenErr_MSB 0xF
+#define QIB_7322_ErrStatus_0_RcvHdrLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvBadTidErr_LSB 0xE
+#define QIB_7322_ErrStatus_0_RcvBadTidErr_MSB 0xE
+#define QIB_7322_ErrStatus_0_RcvBadTidErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvBadVersionErr_LSB 0xB
+#define QIB_7322_ErrStatus_0_RcvBadVersionErr_MSB 0xB
+#define QIB_7322_ErrStatus_0_RcvBadVersionErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvIBFlowErr_LSB 0xA
+#define QIB_7322_ErrStatus_0_RcvIBFlowErr_MSB 0xA
+#define QIB_7322_ErrStatus_0_RcvIBFlowErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvEBPErr_LSB 0x9
+#define QIB_7322_ErrStatus_0_RcvEBPErr_MSB 0x9
+#define QIB_7322_ErrStatus_0_RcvEBPErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvUnsupportedVLErr_LSB 0x8
+#define QIB_7322_ErrStatus_0_RcvUnsupportedVLErr_MSB 0x8
+#define QIB_7322_ErrStatus_0_RcvUnsupportedVLErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvUnexpectedCharErr_LSB 0x7
+#define QIB_7322_ErrStatus_0_RcvUnexpectedCharErr_MSB 0x7
+#define QIB_7322_ErrStatus_0_RcvUnexpectedCharErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvShortPktLenErr_LSB 0x6
+#define QIB_7322_ErrStatus_0_RcvShortPktLenErr_MSB 0x6
+#define QIB_7322_ErrStatus_0_RcvShortPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvLongPktLenErr_LSB 0x5
+#define QIB_7322_ErrStatus_0_RcvLongPktLenErr_MSB 0x5
+#define QIB_7322_ErrStatus_0_RcvLongPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvMaxPktLenErr_LSB 0x4
+#define QIB_7322_ErrStatus_0_RcvMaxPktLenErr_MSB 0x4
+#define QIB_7322_ErrStatus_0_RcvMaxPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvMinPktLenErr_LSB 0x3
+#define QIB_7322_ErrStatus_0_RcvMinPktLenErr_MSB 0x3
+#define QIB_7322_ErrStatus_0_RcvMinPktLenErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvICRCErr_LSB 0x2
+#define QIB_7322_ErrStatus_0_RcvICRCErr_MSB 0x2
+#define QIB_7322_ErrStatus_0_RcvICRCErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvVCRCErr_LSB 0x1
+#define QIB_7322_ErrStatus_0_RcvVCRCErr_MSB 0x1
+#define QIB_7322_ErrStatus_0_RcvVCRCErr_RMASK 0x1
+#define QIB_7322_ErrStatus_0_RcvFormatErr_LSB 0x0
+#define QIB_7322_ErrStatus_0_RcvFormatErr_MSB 0x0
+#define QIB_7322_ErrStatus_0_RcvFormatErr_RMASK 0x1
+
+#define QIB_7322_ErrClear_0_OFFS 0x1090
+#define QIB_7322_ErrClear_0_DEF 0x0000000000000000
+#define QIB_7322_ErrClear_0_IBStatusChangedClear_LSB 0x3A
+#define QIB_7322_ErrClear_0_IBStatusChangedClear_MSB 0x3A
+#define QIB_7322_ErrClear_0_IBStatusChangedClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SHeadersErrClear_LSB 0x39
+#define QIB_7322_ErrClear_0_SHeadersErrClear_MSB 0x39
+#define QIB_7322_ErrClear_0_SHeadersErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_VL15BufMisuseErrClear_LSB 0x36
+#define QIB_7322_ErrClear_0_VL15BufMisuseErrClear_MSB 0x36
+#define QIB_7322_ErrClear_0_VL15BufMisuseErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaHaltErrClear_LSB 0x31
+#define QIB_7322_ErrClear_0_SDmaHaltErrClear_MSB 0x31
+#define QIB_7322_ErrClear_0_SDmaHaltErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaDescAddrMisalignErrClear_LSB 0x30
+#define QIB_7322_ErrClear_0_SDmaDescAddrMisalignErrClear_MSB 0x30
+#define QIB_7322_ErrClear_0_SDmaDescAddrMisalignErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaUnexpDataErrClear_LSB 0x2F
+#define QIB_7322_ErrClear_0_SDmaUnexpDataErrClear_MSB 0x2F
+#define QIB_7322_ErrClear_0_SDmaUnexpDataErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaMissingDwErrClear_LSB 0x2E
+#define QIB_7322_ErrClear_0_SDmaMissingDwErrClear_MSB 0x2E
+#define QIB_7322_ErrClear_0_SDmaMissingDwErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaDwEnErrClear_LSB 0x2D
+#define QIB_7322_ErrClear_0_SDmaDwEnErrClear_MSB 0x2D
+#define QIB_7322_ErrClear_0_SDmaDwEnErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaRpyTagErrClear_LSB 0x2C
+#define QIB_7322_ErrClear_0_SDmaRpyTagErrClear_MSB 0x2C
+#define QIB_7322_ErrClear_0_SDmaRpyTagErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDma1stDescErrClear_LSB 0x2B
+#define QIB_7322_ErrClear_0_SDma1stDescErrClear_MSB 0x2B
+#define QIB_7322_ErrClear_0_SDma1stDescErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaBaseErrClear_LSB 0x2A
+#define QIB_7322_ErrClear_0_SDmaBaseErrClear_MSB 0x2A
+#define QIB_7322_ErrClear_0_SDmaBaseErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaTailOutOfBoundErrClear_LSB 0x29
+#define QIB_7322_ErrClear_0_SDmaTailOutOfBoundErrClear_MSB 0x29
+#define QIB_7322_ErrClear_0_SDmaTailOutOfBoundErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaOutOfBoundErrClear_LSB 0x28
+#define QIB_7322_ErrClear_0_SDmaOutOfBoundErrClear_MSB 0x28
+#define QIB_7322_ErrClear_0_SDmaOutOfBoundErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SDmaGenMismatchErrClear_LSB 0x27
+#define QIB_7322_ErrClear_0_SDmaGenMismatchErrClear_MSB 0x27
+#define QIB_7322_ErrClear_0_SDmaGenMismatchErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendBufMisuseErrClear_LSB 0x26
+#define QIB_7322_ErrClear_0_SendBufMisuseErrClear_MSB 0x26
+#define QIB_7322_ErrClear_0_SendBufMisuseErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendUnsupportedVLErrClear_LSB 0x25
+#define QIB_7322_ErrClear_0_SendUnsupportedVLErrClear_MSB 0x25
+#define QIB_7322_ErrClear_0_SendUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendUnexpectedPktNumErrClear_LSB 0x24
+#define QIB_7322_ErrClear_0_SendUnexpectedPktNumErrClear_MSB 0x24
+#define QIB_7322_ErrClear_0_SendUnexpectedPktNumErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendDroppedDataPktErrClear_LSB 0x22
+#define QIB_7322_ErrClear_0_SendDroppedDataPktErrClear_MSB 0x22
+#define QIB_7322_ErrClear_0_SendDroppedDataPktErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendDroppedSmpPktErrClear_LSB 0x21
+#define QIB_7322_ErrClear_0_SendDroppedSmpPktErrClear_MSB 0x21
+#define QIB_7322_ErrClear_0_SendDroppedSmpPktErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendPktLenErrClear_LSB 0x20
+#define QIB_7322_ErrClear_0_SendPktLenErrClear_MSB 0x20
+#define QIB_7322_ErrClear_0_SendPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendUnderRunErrClear_LSB 0x1F
+#define QIB_7322_ErrClear_0_SendUnderRunErrClear_MSB 0x1F
+#define QIB_7322_ErrClear_0_SendUnderRunErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendMaxPktLenErrClear_LSB 0x1E
+#define QIB_7322_ErrClear_0_SendMaxPktLenErrClear_MSB 0x1E
+#define QIB_7322_ErrClear_0_SendMaxPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_SendMinPktLenErrClear_LSB 0x1D
+#define QIB_7322_ErrClear_0_SendMinPktLenErrClear_MSB 0x1D
+#define QIB_7322_ErrClear_0_SendMinPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvIBLostLinkErrClear_LSB 0x11
+#define QIB_7322_ErrClear_0_RcvIBLostLinkErrClear_MSB 0x11
+#define QIB_7322_ErrClear_0_RcvIBLostLinkErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvHdrErrClear_LSB 0x10
+#define QIB_7322_ErrClear_0_RcvHdrErrClear_MSB 0x10
+#define QIB_7322_ErrClear_0_RcvHdrErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvHdrLenErrClear_LSB 0xF
+#define QIB_7322_ErrClear_0_RcvHdrLenErrClear_MSB 0xF
+#define QIB_7322_ErrClear_0_RcvHdrLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvBadTidErrClear_LSB 0xE
+#define QIB_7322_ErrClear_0_RcvBadTidErrClear_MSB 0xE
+#define QIB_7322_ErrClear_0_RcvBadTidErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvBadVersionErrClear_LSB 0xB
+#define QIB_7322_ErrClear_0_RcvBadVersionErrClear_MSB 0xB
+#define QIB_7322_ErrClear_0_RcvBadVersionErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvIBFlowErrClear_LSB 0xA
+#define QIB_7322_ErrClear_0_RcvIBFlowErrClear_MSB 0xA
+#define QIB_7322_ErrClear_0_RcvIBFlowErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvEBPErrClear_LSB 0x9
+#define QIB_7322_ErrClear_0_RcvEBPErrClear_MSB 0x9
+#define QIB_7322_ErrClear_0_RcvEBPErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvUnsupportedVLErrClear_LSB 0x8
+#define QIB_7322_ErrClear_0_RcvUnsupportedVLErrClear_MSB 0x8
+#define QIB_7322_ErrClear_0_RcvUnsupportedVLErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvUnexpectedCharErrClear_LSB 0x7
+#define QIB_7322_ErrClear_0_RcvUnexpectedCharErrClear_MSB 0x7
+#define QIB_7322_ErrClear_0_RcvUnexpectedCharErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvShortPktLenErrClear_LSB 0x6
+#define QIB_7322_ErrClear_0_RcvShortPktLenErrClear_MSB 0x6
+#define QIB_7322_ErrClear_0_RcvShortPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvLongPktLenErrClear_LSB 0x5
+#define QIB_7322_ErrClear_0_RcvLongPktLenErrClear_MSB 0x5
+#define QIB_7322_ErrClear_0_RcvLongPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvMaxPktLenErrClear_LSB 0x4
+#define QIB_7322_ErrClear_0_RcvMaxPktLenErrClear_MSB 0x4
+#define QIB_7322_ErrClear_0_RcvMaxPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvMinPktLenErrClear_LSB 0x3
+#define QIB_7322_ErrClear_0_RcvMinPktLenErrClear_MSB 0x3
+#define QIB_7322_ErrClear_0_RcvMinPktLenErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvICRCErrClear_LSB 0x2
+#define QIB_7322_ErrClear_0_RcvICRCErrClear_MSB 0x2
+#define QIB_7322_ErrClear_0_RcvICRCErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvVCRCErrClear_LSB 0x1
+#define QIB_7322_ErrClear_0_RcvVCRCErrClear_MSB 0x1
+#define QIB_7322_ErrClear_0_RcvVCRCErrClear_RMASK 0x1
+#define QIB_7322_ErrClear_0_RcvFormatErrClear_LSB 0x0
+#define QIB_7322_ErrClear_0_RcvFormatErrClear_MSB 0x0
+#define QIB_7322_ErrClear_0_RcvFormatErrClear_RMASK 0x1
+
+#define QIB_7322_TXEStatus_0_OFFS 0x10B8
+#define QIB_7322_TXEStatus_0_DEF 0x0000000XC00080FF
+#define QIB_7322_TXEStatus_0_TXE_IBC_Idle_LSB 0x1F
+#define QIB_7322_TXEStatus_0_TXE_IBC_Idle_MSB 0x1F
+#define QIB_7322_TXEStatus_0_TXE_IBC_Idle_RMASK 0x1
+#define QIB_7322_TXEStatus_0_RmFifoEmpty_LSB 0x1E
+#define QIB_7322_TXEStatus_0_RmFifoEmpty_MSB 0x1E
+#define QIB_7322_TXEStatus_0_RmFifoEmpty_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL15_LSB 0xF
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL15_MSB 0xF
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL15_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL7_LSB 0x7
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL7_MSB 0x7
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL7_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL6_LSB 0x6
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL6_MSB 0x6
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL6_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL5_LSB 0x5
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL5_MSB 0x5
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL5_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL4_LSB 0x4
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL4_MSB 0x4
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL4_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL3_LSB 0x3
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL3_MSB 0x3
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL3_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL2_LSB 0x2
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL2_MSB 0x2
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL2_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL1_LSB 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL1_MSB 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL1_RMASK 0x1
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL0_LSB 0x0
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL0_MSB 0x0
+#define QIB_7322_TXEStatus_0_LaFifoEmpty_VL0_RMASK 0x1
+
+#define QIB_7322_RcvCtrl_0_OFFS 0x1100
+#define QIB_7322_RcvCtrl_0_DEF 0x0000000000000000
+#define QIB_7322_RcvCtrl_0_RcvResetCredit_LSB 0x2A
+#define QIB_7322_RcvCtrl_0_RcvResetCredit_MSB 0x2A
+#define QIB_7322_RcvCtrl_0_RcvResetCredit_RMASK 0x1
+#define QIB_7322_RcvCtrl_0_RcvPartitionKeyDisable_LSB 0x29
+#define QIB_7322_RcvCtrl_0_RcvPartitionKeyDisable_MSB 0x29
+#define QIB_7322_RcvCtrl_0_RcvPartitionKeyDisable_RMASK 0x1
+#define QIB_7322_RcvCtrl_0_RcvQPMapEnable_LSB 0x28
+#define QIB_7322_RcvCtrl_0_RcvQPMapEnable_MSB 0x28
+#define QIB_7322_RcvCtrl_0_RcvQPMapEnable_RMASK 0x1
+#define QIB_7322_RcvCtrl_0_RcvIBPortEnable_LSB 0x27
+#define QIB_7322_RcvCtrl_0_RcvIBPortEnable_MSB 0x27
+#define QIB_7322_RcvCtrl_0_RcvIBPortEnable_RMASK 0x1
+#define QIB_7322_RcvCtrl_0_ContextEnableUser_LSB 0x2
+#define QIB_7322_RcvCtrl_0_ContextEnableUser_MSB 0x11
+#define QIB_7322_RcvCtrl_0_ContextEnableUser_RMASK 0xFFFF
+#define QIB_7322_RcvCtrl_0_ContextEnableKernel_LSB 0x0
+#define QIB_7322_RcvCtrl_0_ContextEnableKernel_MSB 0x0
+#define QIB_7322_RcvCtrl_0_ContextEnableKernel_RMASK 0x1
+
+#define QIB_7322_RcvBTHQP_0_OFFS 0x1108
+#define QIB_7322_RcvBTHQP_0_DEF 0x0000000000000000
+#define QIB_7322_RcvBTHQP_0_RcvBTHQP_LSB 0x0
+#define QIB_7322_RcvBTHQP_0_RcvBTHQP_MSB 0x17
+#define QIB_7322_RcvBTHQP_0_RcvBTHQP_RMASK 0xFFFFFF
+
+#define QIB_7322_RcvQPMapTableA_0_OFFS 0x1110
+#define QIB_7322_RcvQPMapTableA_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext5_LSB 0x19
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext5_MSB 0x1D
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext5_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext4_LSB 0x14
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext4_MSB 0x18
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext4_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext3_LSB 0xF
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext3_MSB 0x13
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext3_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext2_LSB 0xA
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext2_MSB 0xE
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext2_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext1_LSB 0x5
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext1_MSB 0x9
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext1_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext0_LSB 0x0
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext0_MSB 0x4
+#define QIB_7322_RcvQPMapTableA_0_RcvQPMapContext0_RMASK 0x1F
+
+#define QIB_7322_RcvQPMapTableB_0_OFFS 0x1118
+#define QIB_7322_RcvQPMapTableB_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext11_LSB 0x19
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext11_MSB 0x1D
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext11_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext10_LSB 0x14
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext10_MSB 0x18
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext10_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext9_LSB 0xF
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext9_MSB 0x13
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext9_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext8_LSB 0xA
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext8_MSB 0xE
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext8_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext7_LSB 0x5
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext7_MSB 0x9
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext7_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext6_LSB 0x0
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext6_MSB 0x4
+#define QIB_7322_RcvQPMapTableB_0_RcvQPMapContext6_RMASK 0x1F
+
+#define QIB_7322_RcvQPMapTableC_0_OFFS 0x1120
+#define QIB_7322_RcvQPMapTableC_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext17_LSB 0x19
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext17_MSB 0x1D
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext17_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext16_LSB 0x14
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext16_MSB 0x18
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext16_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext15_LSB 0xF
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext15_MSB 0x13
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext15_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext14_LSB 0xA
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext14_MSB 0xE
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext14_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext13_LSB 0x5
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext13_MSB 0x9
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext13_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext12_LSB 0x0
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext12_MSB 0x4
+#define QIB_7322_RcvQPMapTableC_0_RcvQPMapContext12_RMASK 0x1F
+
+#define QIB_7322_RcvQPMapTableD_0_OFFS 0x1128
+#define QIB_7322_RcvQPMapTableD_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext23_LSB 0x19
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext23_MSB 0x1D
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext23_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext22_LSB 0x14
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext22_MSB 0x18
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext22_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext21_LSB 0xF
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext21_MSB 0x13
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext21_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext20_LSB 0xA
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext20_MSB 0xE
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext20_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext19_LSB 0x5
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext19_MSB 0x9
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext19_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext18_LSB 0x0
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext18_MSB 0x4
+#define QIB_7322_RcvQPMapTableD_0_RcvQPMapContext18_RMASK 0x1F
+
+#define QIB_7322_RcvQPMapTableE_0_OFFS 0x1130
+#define QIB_7322_RcvQPMapTableE_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext29_LSB 0x19
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext29_MSB 0x1D
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext29_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext28_LSB 0x14
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext28_MSB 0x18
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext28_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext27_LSB 0xF
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext27_MSB 0x13
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext27_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext26_LSB 0xA
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext26_MSB 0xE
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext26_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext25_LSB 0x5
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext25_MSB 0x9
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext25_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext24_LSB 0x0
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext24_MSB 0x4
+#define QIB_7322_RcvQPMapTableE_0_RcvQPMapContext24_RMASK 0x1F
+
+#define QIB_7322_RcvQPMapTableF_0_OFFS 0x1138
+#define QIB_7322_RcvQPMapTableF_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext31_LSB 0x5
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext31_MSB 0x9
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext31_RMASK 0x1F
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext30_LSB 0x0
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext30_MSB 0x4
+#define QIB_7322_RcvQPMapTableF_0_RcvQPMapContext30_RMASK 0x1F
+
+#define QIB_7322_PSStat_0_OFFS 0x1140
+#define QIB_7322_PSStat_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSStart_0_OFFS 0x1148
+#define QIB_7322_PSStart_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSInterval_0_OFFS 0x1150
+#define QIB_7322_PSInterval_0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvStatus_0_OFFS 0x1160
+#define QIB_7322_RcvStatus_0_DEF 0x0000000000000000
+#define QIB_7322_RcvStatus_0_DmaeqBlockingContext_LSB 0x1
+#define QIB_7322_RcvStatus_0_DmaeqBlockingContext_MSB 0x5
+#define QIB_7322_RcvStatus_0_DmaeqBlockingContext_RMASK 0x1F
+#define QIB_7322_RcvStatus_0_RxPktInProgress_LSB 0x0
+#define QIB_7322_RcvStatus_0_RxPktInProgress_MSB 0x0
+#define QIB_7322_RcvStatus_0_RxPktInProgress_RMASK 0x1
+
+#define QIB_7322_RcvPartitionKey_0_OFFS 0x1168
+#define QIB_7322_RcvPartitionKey_0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvQPMulticastContext_0_OFFS 0x1170
+#define QIB_7322_RcvQPMulticastContext_0_DEF 0x0000000000000000
+#define QIB_7322_RcvQPMulticastContext_0_RcvQpMcContext_LSB 0x0
+#define QIB_7322_RcvQPMulticastContext_0_RcvQpMcContext_MSB 0x4
+#define QIB_7322_RcvQPMulticastContext_0_RcvQpMcContext_RMASK 0x1F
+
+#define QIB_7322_RcvPktLEDCnt_0_OFFS 0x1178
+#define QIB_7322_RcvPktLEDCnt_0_DEF 0x0000000000000000
+#define QIB_7322_RcvPktLEDCnt_0_ONperiod_LSB 0x20
+#define QIB_7322_RcvPktLEDCnt_0_ONperiod_MSB 0x3F
+#define QIB_7322_RcvPktLEDCnt_0_ONperiod_RMASK 0xFFFFFFFF
+#define QIB_7322_RcvPktLEDCnt_0_OFFperiod_LSB 0x0
+#define QIB_7322_RcvPktLEDCnt_0_OFFperiod_MSB 0x1F
+#define QIB_7322_RcvPktLEDCnt_0_OFFperiod_RMASK 0xFFFFFFFF
+
+#define QIB_7322_SendDmaIdleCnt_0_OFFS 0x1180
+#define QIB_7322_SendDmaIdleCnt_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaIdleCnt_0_SendDmaIdleCnt_LSB 0x0
+#define QIB_7322_SendDmaIdleCnt_0_SendDmaIdleCnt_MSB 0xF
+#define QIB_7322_SendDmaIdleCnt_0_SendDmaIdleCnt_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaReloadCnt_0_OFFS 0x1188
+#define QIB_7322_SendDmaReloadCnt_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaReloadCnt_0_SendDmaReloadCnt_LSB 0x0
+#define QIB_7322_SendDmaReloadCnt_0_SendDmaReloadCnt_MSB 0xF
+#define QIB_7322_SendDmaReloadCnt_0_SendDmaReloadCnt_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaDescCnt_0_OFFS 0x1190
+#define QIB_7322_SendDmaDescCnt_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaDescCnt_0_SendDmaDescCnt_LSB 0x0
+#define QIB_7322_SendDmaDescCnt_0_SendDmaDescCnt_MSB 0xF
+#define QIB_7322_SendDmaDescCnt_0_SendDmaDescCnt_RMASK 0xFFFF
+
+#define QIB_7322_SendCtrl_0_OFFS 0x11C0
+#define QIB_7322_SendCtrl_0_DEF 0x0000000000000000
+#define QIB_7322_SendCtrl_0_IBVLArbiterEn_LSB 0xF
+#define QIB_7322_SendCtrl_0_IBVLArbiterEn_MSB 0xF
+#define QIB_7322_SendCtrl_0_IBVLArbiterEn_RMASK 0x1
+#define QIB_7322_SendCtrl_0_TxeDrainRmFifo_LSB 0xE
+#define QIB_7322_SendCtrl_0_TxeDrainRmFifo_MSB 0xE
+#define QIB_7322_SendCtrl_0_TxeDrainRmFifo_RMASK 0x1
+#define QIB_7322_SendCtrl_0_TxeDrainLaFifo_LSB 0xD
+#define QIB_7322_SendCtrl_0_TxeDrainLaFifo_MSB 0xD
+#define QIB_7322_SendCtrl_0_TxeDrainLaFifo_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SDmaHalt_LSB 0xC
+#define QIB_7322_SendCtrl_0_SDmaHalt_MSB 0xC
+#define QIB_7322_SendCtrl_0_SDmaHalt_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SDmaEnable_LSB 0xB
+#define QIB_7322_SendCtrl_0_SDmaEnable_MSB 0xB
+#define QIB_7322_SendCtrl_0_SDmaEnable_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SDmaSingleDescriptor_LSB 0xA
+#define QIB_7322_SendCtrl_0_SDmaSingleDescriptor_MSB 0xA
+#define QIB_7322_SendCtrl_0_SDmaSingleDescriptor_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SDmaIntEnable_LSB 0x9
+#define QIB_7322_SendCtrl_0_SDmaIntEnable_MSB 0x9
+#define QIB_7322_SendCtrl_0_SDmaIntEnable_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SDmaCleanup_LSB 0x8
+#define QIB_7322_SendCtrl_0_SDmaCleanup_MSB 0x8
+#define QIB_7322_SendCtrl_0_SDmaCleanup_RMASK 0x1
+#define QIB_7322_SendCtrl_0_ForceCreditUpToDate_LSB 0x7
+#define QIB_7322_SendCtrl_0_ForceCreditUpToDate_MSB 0x7
+#define QIB_7322_SendCtrl_0_ForceCreditUpToDate_RMASK 0x1
+#define QIB_7322_SendCtrl_0_SendEnable_LSB 0x3
+#define QIB_7322_SendCtrl_0_SendEnable_MSB 0x3
+#define QIB_7322_SendCtrl_0_SendEnable_RMASK 0x1
+#define QIB_7322_SendCtrl_0_TxeBypassIbc_LSB 0x1
+#define QIB_7322_SendCtrl_0_TxeBypassIbc_MSB 0x1
+#define QIB_7322_SendCtrl_0_TxeBypassIbc_RMASK 0x1
+#define QIB_7322_SendCtrl_0_TxeAbortIbc_LSB 0x0
+#define QIB_7322_SendCtrl_0_TxeAbortIbc_MSB 0x0
+#define QIB_7322_SendCtrl_0_TxeAbortIbc_RMASK 0x1
+
+#define QIB_7322_SendDmaBase_0_OFFS 0x11F8
+#define QIB_7322_SendDmaBase_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaBase_0_SendDmaBase_LSB 0x0
+#define QIB_7322_SendDmaBase_0_SendDmaBase_MSB 0x2F
+#define QIB_7322_SendDmaBase_0_SendDmaBase_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7322_SendDmaLenGen_0_OFFS 0x1200
+#define QIB_7322_SendDmaLenGen_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaLenGen_0_Generation_LSB 0x10
+#define QIB_7322_SendDmaLenGen_0_Generation_MSB 0x12
+#define QIB_7322_SendDmaLenGen_0_Generation_RMASK 0x7
+#define QIB_7322_SendDmaLenGen_0_Length_LSB 0x0
+#define QIB_7322_SendDmaLenGen_0_Length_MSB 0xF
+#define QIB_7322_SendDmaLenGen_0_Length_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaTail_0_OFFS 0x1208
+#define QIB_7322_SendDmaTail_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaTail_0_SendDmaTail_LSB 0x0
+#define QIB_7322_SendDmaTail_0_SendDmaTail_MSB 0xF
+#define QIB_7322_SendDmaTail_0_SendDmaTail_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaHead_0_OFFS 0x1210
+#define QIB_7322_SendDmaHead_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaHead_0_InternalSendDmaHead_LSB 0x20
+#define QIB_7322_SendDmaHead_0_InternalSendDmaHead_MSB 0x2F
+#define QIB_7322_SendDmaHead_0_InternalSendDmaHead_RMASK 0xFFFF
+#define QIB_7322_SendDmaHead_0_SendDmaHead_LSB 0x0
+#define QIB_7322_SendDmaHead_0_SendDmaHead_MSB 0xF
+#define QIB_7322_SendDmaHead_0_SendDmaHead_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaHeadAddr_0_OFFS 0x1218
+#define QIB_7322_SendDmaHeadAddr_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaHeadAddr_0_SendDmaHeadAddr_LSB 0x0
+#define QIB_7322_SendDmaHeadAddr_0_SendDmaHeadAddr_MSB 0x2F
+#define QIB_7322_SendDmaHeadAddr_0_SendDmaHeadAddr_RMASK 0xFFFFFFFFFFFF
+
+#define QIB_7322_SendDmaBufMask0_0_OFFS 0x1220
+#define QIB_7322_SendDmaBufMask0_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaBufMask0_0_BufMask_63_0_LSB 0x0
+#define QIB_7322_SendDmaBufMask0_0_BufMask_63_0_MSB 0x3F
+#define QIB_7322_SendDmaBufMask0_0_BufMask_63_0_RMASK 0x0
+
+#define QIB_7322_SendDmaStatus_0_OFFS 0x1238
+#define QIB_7322_SendDmaStatus_0_DEF 0x0000000042000000
+#define QIB_7322_SendDmaStatus_0_ScoreBoardDrainInProg_LSB 0x3F
+#define QIB_7322_SendDmaStatus_0_ScoreBoardDrainInProg_MSB 0x3F
+#define QIB_7322_SendDmaStatus_0_ScoreBoardDrainInProg_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_HaltInProg_LSB 0x3E
+#define QIB_7322_SendDmaStatus_0_HaltInProg_MSB 0x3E
+#define QIB_7322_SendDmaStatus_0_HaltInProg_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_InternalSDmaHalt_LSB 0x3D
+#define QIB_7322_SendDmaStatus_0_InternalSDmaHalt_MSB 0x3D
+#define QIB_7322_SendDmaStatus_0_InternalSDmaHalt_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_ScbDescIndex_13_0_LSB 0x2F
+#define QIB_7322_SendDmaStatus_0_ScbDescIndex_13_0_MSB 0x3C
+#define QIB_7322_SendDmaStatus_0_ScbDescIndex_13_0_RMASK 0x3FFF
+#define QIB_7322_SendDmaStatus_0_RpyLowAddr_6_0_LSB 0x28
+#define QIB_7322_SendDmaStatus_0_RpyLowAddr_6_0_MSB 0x2E
+#define QIB_7322_SendDmaStatus_0_RpyLowAddr_6_0_RMASK 0x7F
+#define QIB_7322_SendDmaStatus_0_RpyTag_7_0_LSB 0x20
+#define QIB_7322_SendDmaStatus_0_RpyTag_7_0_MSB 0x27
+#define QIB_7322_SendDmaStatus_0_RpyTag_7_0_RMASK 0xFF
+#define QIB_7322_SendDmaStatus_0_ScbFull_LSB 0x1F
+#define QIB_7322_SendDmaStatus_0_ScbFull_MSB 0x1F
+#define QIB_7322_SendDmaStatus_0_ScbFull_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_ScbEmpty_LSB 0x1E
+#define QIB_7322_SendDmaStatus_0_ScbEmpty_MSB 0x1E
+#define QIB_7322_SendDmaStatus_0_ScbEmpty_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_ScbEntryValid_LSB 0x1D
+#define QIB_7322_SendDmaStatus_0_ScbEntryValid_MSB 0x1D
+#define QIB_7322_SendDmaStatus_0_ScbEntryValid_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_ScbFetchDescFlag_LSB 0x1C
+#define QIB_7322_SendDmaStatus_0_ScbFetchDescFlag_MSB 0x1C
+#define QIB_7322_SendDmaStatus_0_ScbFetchDescFlag_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_SplFifoReadyToGo_LSB 0x1B
+#define QIB_7322_SendDmaStatus_0_SplFifoReadyToGo_MSB 0x1B
+#define QIB_7322_SendDmaStatus_0_SplFifoReadyToGo_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_SplFifoDisarmed_LSB 0x1A
+#define QIB_7322_SendDmaStatus_0_SplFifoDisarmed_MSB 0x1A
+#define QIB_7322_SendDmaStatus_0_SplFifoDisarmed_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_SplFifoEmpty_LSB 0x19
+#define QIB_7322_SendDmaStatus_0_SplFifoEmpty_MSB 0x19
+#define QIB_7322_SendDmaStatus_0_SplFifoEmpty_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_SplFifoFull_LSB 0x18
+#define QIB_7322_SendDmaStatus_0_SplFifoFull_MSB 0x18
+#define QIB_7322_SendDmaStatus_0_SplFifoFull_RMASK 0x1
+#define QIB_7322_SendDmaStatus_0_SplFifoBufNum_LSB 0x10
+#define QIB_7322_SendDmaStatus_0_SplFifoBufNum_MSB 0x17
+#define QIB_7322_SendDmaStatus_0_SplFifoBufNum_RMASK 0xFF
+#define QIB_7322_SendDmaStatus_0_SplFifoDescIndex_LSB 0x0
+#define QIB_7322_SendDmaStatus_0_SplFifoDescIndex_MSB 0xF
+#define QIB_7322_SendDmaStatus_0_SplFifoDescIndex_RMASK 0xFFFF
+
+#define QIB_7322_SendDmaPriorityThld_0_OFFS 0x1258
+#define QIB_7322_SendDmaPriorityThld_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaPriorityThld_0_PriorityThreshold_LSB 0x0
+#define QIB_7322_SendDmaPriorityThld_0_PriorityThreshold_MSB 0x3
+#define QIB_7322_SendDmaPriorityThld_0_PriorityThreshold_RMASK 0xF
+
+#define QIB_7322_SendHdrErrSymptom_0_OFFS 0x1260
+#define QIB_7322_SendHdrErrSymptom_0_DEF 0x0000000000000000
+#define QIB_7322_SendHdrErrSymptom_0_NonKeyPacket_LSB 0x6
+#define QIB_7322_SendHdrErrSymptom_0_NonKeyPacket_MSB 0x6
+#define QIB_7322_SendHdrErrSymptom_0_NonKeyPacket_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_GRHFail_LSB 0x5
+#define QIB_7322_SendHdrErrSymptom_0_GRHFail_MSB 0x5
+#define QIB_7322_SendHdrErrSymptom_0_GRHFail_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_PkeyFail_LSB 0x4
+#define QIB_7322_SendHdrErrSymptom_0_PkeyFail_MSB 0x4
+#define QIB_7322_SendHdrErrSymptom_0_PkeyFail_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_QPFail_LSB 0x3
+#define QIB_7322_SendHdrErrSymptom_0_QPFail_MSB 0x3
+#define QIB_7322_SendHdrErrSymptom_0_QPFail_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_SLIDFail_LSB 0x2
+#define QIB_7322_SendHdrErrSymptom_0_SLIDFail_MSB 0x2
+#define QIB_7322_SendHdrErrSymptom_0_SLIDFail_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_RawIPV6_LSB 0x1
+#define QIB_7322_SendHdrErrSymptom_0_RawIPV6_MSB 0x1
+#define QIB_7322_SendHdrErrSymptom_0_RawIPV6_RMASK 0x1
+#define QIB_7322_SendHdrErrSymptom_0_PacketTooSmall_LSB 0x0
+#define QIB_7322_SendHdrErrSymptom_0_PacketTooSmall_MSB 0x0
+#define QIB_7322_SendHdrErrSymptom_0_PacketTooSmall_RMASK 0x1
+
+#define QIB_7322_RxCreditVL0_0_OFFS 0x1280
+#define QIB_7322_RxCreditVL0_0_DEF 0x0000000000000000
+#define QIB_7322_RxCreditVL0_0_RxBufrConsumedVL_LSB 0x10
+#define QIB_7322_RxCreditVL0_0_RxBufrConsumedVL_MSB 0x1B
+#define QIB_7322_RxCreditVL0_0_RxBufrConsumedVL_RMASK 0xFFF
+#define QIB_7322_RxCreditVL0_0_RxMaxCreditVL_LSB 0x0
+#define QIB_7322_RxCreditVL0_0_RxMaxCreditVL_MSB 0xB
+#define QIB_7322_RxCreditVL0_0_RxMaxCreditVL_RMASK 0xFFF
+
+#define QIB_7322_SendDmaBufUsed0_0_OFFS 0x1480
+#define QIB_7322_SendDmaBufUsed0_0_DEF 0x0000000000000000
+#define QIB_7322_SendDmaBufUsed0_0_BufUsed_63_0_LSB 0x0
+#define QIB_7322_SendDmaBufUsed0_0_BufUsed_63_0_MSB 0x3F
+#define QIB_7322_SendDmaBufUsed0_0_BufUsed_63_0_RMASK 0x0
+
+#define QIB_7322_SendCheckControl_0_OFFS 0x14A8
+#define QIB_7322_SendCheckControl_0_DEF 0x0000000000000000
+#define QIB_7322_SendCheckControl_0_PKey_En_LSB 0x4
+#define QIB_7322_SendCheckControl_0_PKey_En_MSB 0x4
+#define QIB_7322_SendCheckControl_0_PKey_En_RMASK 0x1
+#define QIB_7322_SendCheckControl_0_BTHQP_En_LSB 0x3
+#define QIB_7322_SendCheckControl_0_BTHQP_En_MSB 0x3
+#define QIB_7322_SendCheckControl_0_BTHQP_En_RMASK 0x1
+#define QIB_7322_SendCheckControl_0_SLID_En_LSB 0x2
+#define QIB_7322_SendCheckControl_0_SLID_En_MSB 0x2
+#define QIB_7322_SendCheckControl_0_SLID_En_RMASK 0x1
+#define QIB_7322_SendCheckControl_0_RawIPV6_En_LSB 0x1
+#define QIB_7322_SendCheckControl_0_RawIPV6_En_MSB 0x1
+#define QIB_7322_SendCheckControl_0_RawIPV6_En_RMASK 0x1
+#define QIB_7322_SendCheckControl_0_PacketTooSmall_En_LSB 0x0
+#define QIB_7322_SendCheckControl_0_PacketTooSmall_En_MSB 0x0
+#define QIB_7322_SendCheckControl_0_PacketTooSmall_En_RMASK 0x1
+
+#define QIB_7322_SendIBSLIDMask_0_OFFS 0x14B0
+#define QIB_7322_SendIBSLIDMask_0_DEF 0x0000000000000000
+#define QIB_7322_SendIBSLIDMask_0_SendIBSLIDMask_15_0_LSB 0x0
+#define QIB_7322_SendIBSLIDMask_0_SendIBSLIDMask_15_0_MSB 0xF
+#define QIB_7322_SendIBSLIDMask_0_SendIBSLIDMask_15_0_RMASK 0xFFFF
+
+#define QIB_7322_SendIBSLIDAssign_0_OFFS 0x14B8
+#define QIB_7322_SendIBSLIDAssign_0_DEF 0x0000000000000000
+#define QIB_7322_SendIBSLIDAssign_0_SendIBSLIDAssign_15_0_LSB 0x0
+#define QIB_7322_SendIBSLIDAssign_0_SendIBSLIDAssign_15_0_MSB 0xF
+#define QIB_7322_SendIBSLIDAssign_0_SendIBSLIDAssign_15_0_RMASK 0xFFFF
+
+#define QIB_7322_IBCStatusA_0_OFFS 0x1540
+#define QIB_7322_IBCStatusA_0_DEF 0x0000000000000X02
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL7_LSB 0x27
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL7_MSB 0x27
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL7_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL6_LSB 0x26
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL6_MSB 0x26
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL6_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL5_LSB 0x25
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL5_MSB 0x25
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL5_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL4_LSB 0x24
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL4_MSB 0x24
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL4_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL3_LSB 0x23
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL3_MSB 0x23
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL3_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL2_LSB 0x22
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL2_MSB 0x22
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL2_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL1_LSB 0x21
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL1_MSB 0x21
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL1_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL0_LSB 0x20
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL0_MSB 0x20
+#define QIB_7322_IBCStatusA_0_TxCreditOk_VL0_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_TxReady_LSB 0x1E
+#define QIB_7322_IBCStatusA_0_TxReady_MSB 0x1E
+#define QIB_7322_IBCStatusA_0_TxReady_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_LinkSpeedQDR_LSB 0x1D
+#define QIB_7322_IBCStatusA_0_LinkSpeedQDR_MSB 0x1D
+#define QIB_7322_IBCStatusA_0_LinkSpeedQDR_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_ScrambleCapRemote_LSB 0xF
+#define QIB_7322_IBCStatusA_0_ScrambleCapRemote_MSB 0xF
+#define QIB_7322_IBCStatusA_0_ScrambleCapRemote_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_ScrambleEn_LSB 0xE
+#define QIB_7322_IBCStatusA_0_ScrambleEn_MSB 0xE
+#define QIB_7322_IBCStatusA_0_ScrambleEn_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_IBTxLaneReversed_LSB 0xD
+#define QIB_7322_IBCStatusA_0_IBTxLaneReversed_MSB 0xD
+#define QIB_7322_IBCStatusA_0_IBTxLaneReversed_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_IBRxLaneReversed_LSB 0xC
+#define QIB_7322_IBCStatusA_0_IBRxLaneReversed_MSB 0xC
+#define QIB_7322_IBCStatusA_0_IBRxLaneReversed_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_DDS_RXEQ_FAIL_LSB 0xA
+#define QIB_7322_IBCStatusA_0_DDS_RXEQ_FAIL_MSB 0xA
+#define QIB_7322_IBCStatusA_0_DDS_RXEQ_FAIL_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_LinkWidthActive_LSB 0x9
+#define QIB_7322_IBCStatusA_0_LinkWidthActive_MSB 0x9
+#define QIB_7322_IBCStatusA_0_LinkWidthActive_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_LinkSpeedActive_LSB 0x8
+#define QIB_7322_IBCStatusA_0_LinkSpeedActive_MSB 0x8
+#define QIB_7322_IBCStatusA_0_LinkSpeedActive_RMASK 0x1
+#define QIB_7322_IBCStatusA_0_LinkState_LSB 0x5
+#define QIB_7322_IBCStatusA_0_LinkState_MSB 0x7
+#define QIB_7322_IBCStatusA_0_LinkState_RMASK 0x7
+#define QIB_7322_IBCStatusA_0_LinkTrainingState_LSB 0x0
+#define QIB_7322_IBCStatusA_0_LinkTrainingState_MSB 0x4
+#define QIB_7322_IBCStatusA_0_LinkTrainingState_RMASK 0x1F
+
+#define QIB_7322_IBCStatusB_0_OFFS 0x1548
+#define QIB_7322_IBCStatusB_0_DEF 0x00000000XXXXXXXX
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_debug_LSB 0x27
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_debug_MSB 0x27
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_debug_RMASK 0x1
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_reached_threshold_LSB 0x26
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_reached_threshold_MSB 0x26
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_reached_threshold_RMASK 0x1
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_started_LSB 0x25
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_started_MSB 0x25
+#define QIB_7322_IBCStatusB_0_ibsd_adaptation_timer_started_RMASK 0x1
+#define QIB_7322_IBCStatusB_0_heartbeat_timed_out_LSB 0x24
+#define QIB_7322_IBCStatusB_0_heartbeat_timed_out_MSB 0x24
+#define QIB_7322_IBCStatusB_0_heartbeat_timed_out_RMASK 0x1
+#define QIB_7322_IBCStatusB_0_heartbeat_crosstalk_LSB 0x20
+#define QIB_7322_IBCStatusB_0_heartbeat_crosstalk_MSB 0x23
+#define QIB_7322_IBCStatusB_0_heartbeat_crosstalk_RMASK 0xF
+#define QIB_7322_IBCStatusB_0_RxEqLocalDevice_LSB 0x1E
+#define QIB_7322_IBCStatusB_0_RxEqLocalDevice_MSB 0x1F
+#define QIB_7322_IBCStatusB_0_RxEqLocalDevice_RMASK 0x3
+#define QIB_7322_IBCStatusB_0_ReqDDSLocalFromRmt_LSB 0x1A
+#define QIB_7322_IBCStatusB_0_ReqDDSLocalFromRmt_MSB 0x1D
+#define QIB_7322_IBCStatusB_0_ReqDDSLocalFromRmt_RMASK 0xF
+#define QIB_7322_IBCStatusB_0_LinkRoundTripLatency_LSB 0x0
+#define QIB_7322_IBCStatusB_0_LinkRoundTripLatency_MSB 0x19
+#define QIB_7322_IBCStatusB_0_LinkRoundTripLatency_RMASK 0x3FFFFFF
+
+#define QIB_7322_IBCCtrlA_0_OFFS 0x1560
+#define QIB_7322_IBCCtrlA_0_DEF 0x0000000000000000
+#define QIB_7322_IBCCtrlA_0_Loopback_LSB 0x3F
+#define QIB_7322_IBCCtrlA_0_Loopback_MSB 0x3F
+#define QIB_7322_IBCCtrlA_0_Loopback_RMASK 0x1
+#define QIB_7322_IBCCtrlA_0_LinkDownDefaultState_LSB 0x3E
+#define QIB_7322_IBCCtrlA_0_LinkDownDefaultState_MSB 0x3E
+#define QIB_7322_IBCCtrlA_0_LinkDownDefaultState_RMASK 0x1
+#define QIB_7322_IBCCtrlA_0_IBLinkEn_LSB 0x3D
+#define QIB_7322_IBCCtrlA_0_IBLinkEn_MSB 0x3D
+#define QIB_7322_IBCCtrlA_0_IBLinkEn_RMASK 0x1
+#define QIB_7322_IBCCtrlA_0_IBStatIntReductionEn_LSB 0x3C
+#define QIB_7322_IBCCtrlA_0_IBStatIntReductionEn_MSB 0x3C
+#define QIB_7322_IBCCtrlA_0_IBStatIntReductionEn_RMASK 0x1
+#define QIB_7322_IBCCtrlA_0_NumVLane_LSB 0x30
+#define QIB_7322_IBCCtrlA_0_NumVLane_MSB 0x32
+#define QIB_7322_IBCCtrlA_0_NumVLane_RMASK 0x7
+#define QIB_7322_IBCCtrlA_0_OverrunThreshold_LSB 0x24
+#define QIB_7322_IBCCtrlA_0_OverrunThreshold_MSB 0x27
+#define QIB_7322_IBCCtrlA_0_OverrunThreshold_RMASK 0xF
+#define QIB_7322_IBCCtrlA_0_PhyerrThreshold_LSB 0x20
+#define QIB_7322_IBCCtrlA_0_PhyerrThreshold_MSB 0x23
+#define QIB_7322_IBCCtrlA_0_PhyerrThreshold_RMASK 0xF
+#define QIB_7322_IBCCtrlA_0_MaxPktLen_LSB 0x15
+#define QIB_7322_IBCCtrlA_0_MaxPktLen_MSB 0x1F
+#define QIB_7322_IBCCtrlA_0_MaxPktLen_RMASK 0x7FF
+#define QIB_7322_IBCCtrlA_0_LinkCmd_LSB 0x13
+#define QIB_7322_IBCCtrlA_0_LinkCmd_MSB 0x14
+#define QIB_7322_IBCCtrlA_0_LinkCmd_RMASK 0x3
+#define QIB_7322_IBCCtrlA_0_LinkInitCmd_LSB 0x10
+#define QIB_7322_IBCCtrlA_0_LinkInitCmd_MSB 0x12
+#define QIB_7322_IBCCtrlA_0_LinkInitCmd_RMASK 0x7
+#define QIB_7322_IBCCtrlA_0_FlowCtrlWaterMark_LSB 0x8
+#define QIB_7322_IBCCtrlA_0_FlowCtrlWaterMark_MSB 0xF
+#define QIB_7322_IBCCtrlA_0_FlowCtrlWaterMark_RMASK 0xFF
+#define QIB_7322_IBCCtrlA_0_FlowCtrlPeriod_LSB 0x0
+#define QIB_7322_IBCCtrlA_0_FlowCtrlPeriod_MSB 0x7
+#define QIB_7322_IBCCtrlA_0_FlowCtrlPeriod_RMASK 0xFF
+
+#define QIB_7322_IBCCtrlB_0_OFFS 0x1568
+#define QIB_7322_IBCCtrlB_0_DEF 0x00000000000305FF
+#define QIB_7322_IBCCtrlB_0_IB_DLID_MASK_LSB 0x30
+#define QIB_7322_IBCCtrlB_0_IB_DLID_MASK_MSB 0x3F
+#define QIB_7322_IBCCtrlB_0_IB_DLID_MASK_RMASK 0xFFFF
+#define QIB_7322_IBCCtrlB_0_IB_DLID_LSB 0x20
+#define QIB_7322_IBCCtrlB_0_IB_DLID_MSB 0x2F
+#define QIB_7322_IBCCtrlB_0_IB_DLID_RMASK 0xFFFF
+#define QIB_7322_IBCCtrlB_0_IB_ENABLE_FILT_DPKT_LSB 0x1B
+#define QIB_7322_IBCCtrlB_0_IB_ENABLE_FILT_DPKT_MSB 0x1B
+#define QIB_7322_IBCCtrlB_0_IB_ENABLE_FILT_DPKT_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_HRTBT_REQ_LSB 0x1A
+#define QIB_7322_IBCCtrlB_0_HRTBT_REQ_MSB 0x1A
+#define QIB_7322_IBCCtrlB_0_HRTBT_REQ_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_HRTBT_PORT_LSB 0x12
+#define QIB_7322_IBCCtrlB_0_HRTBT_PORT_MSB 0x19
+#define QIB_7322_IBCCtrlB_0_HRTBT_PORT_RMASK 0xFF
+#define QIB_7322_IBCCtrlB_0_HRTBT_AUTO_LSB 0x11
+#define QIB_7322_IBCCtrlB_0_HRTBT_AUTO_MSB 0x11
+#define QIB_7322_IBCCtrlB_0_HRTBT_AUTO_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_HRTBT_ENB_LSB 0x10
+#define QIB_7322_IBCCtrlB_0_HRTBT_ENB_MSB 0x10
+#define QIB_7322_IBCCtrlB_0_HRTBT_ENB_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_DDS_LSB 0xC
+#define QIB_7322_IBCCtrlB_0_SD_DDS_MSB 0xF
+#define QIB_7322_IBCCtrlB_0_SD_DDS_RMASK 0xF
+#define QIB_7322_IBCCtrlB_0_SD_DDSV_LSB 0xB
+#define QIB_7322_IBCCtrlB_0_SD_DDSV_MSB 0xB
+#define QIB_7322_IBCCtrlB_0_SD_DDSV_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_ADD_ENB_LSB 0xA
+#define QIB_7322_IBCCtrlB_0_SD_ADD_ENB_MSB 0xA
+#define QIB_7322_IBCCtrlB_0_SD_ADD_ENB_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_RX_EQUAL_ENABLE_LSB 0x9
+#define QIB_7322_IBCCtrlB_0_SD_RX_EQUAL_ENABLE_MSB 0x9
+#define QIB_7322_IBCCtrlB_0_SD_RX_EQUAL_ENABLE_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_IB_LANE_REV_SUPPORTED_LSB 0x8
+#define QIB_7322_IBCCtrlB_0_IB_LANE_REV_SUPPORTED_MSB 0x8
+#define QIB_7322_IBCCtrlB_0_IB_LANE_REV_SUPPORTED_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_IB_POLARITY_REV_SUPP_LSB 0x7
+#define QIB_7322_IBCCtrlB_0_IB_POLARITY_REV_SUPP_MSB 0x7
+#define QIB_7322_IBCCtrlB_0_IB_POLARITY_REV_SUPP_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_IB_NUM_CHANNELS_LSB 0x5
+#define QIB_7322_IBCCtrlB_0_IB_NUM_CHANNELS_MSB 0x6
+#define QIB_7322_IBCCtrlB_0_IB_NUM_CHANNELS_RMASK 0x3
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_QDR_LSB 0x4
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_QDR_MSB 0x4
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_QDR_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_DDR_LSB 0x3
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_DDR_MSB 0x3
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_DDR_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_SDR_LSB 0x2
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_SDR_MSB 0x2
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_SDR_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_LSB 0x1
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_MSB 0x1
+#define QIB_7322_IBCCtrlB_0_SD_SPEED_RMASK 0x1
+#define QIB_7322_IBCCtrlB_0_IB_ENHANCED_MODE_LSB 0x0
+#define QIB_7322_IBCCtrlB_0_IB_ENHANCED_MODE_MSB 0x0
+#define QIB_7322_IBCCtrlB_0_IB_ENHANCED_MODE_RMASK 0x1
+
+#define QIB_7322_IBCCtrlC_0_OFFS 0x1570
+#define QIB_7322_IBCCtrlC_0_DEF 0x0000000000000301
+#define QIB_7322_IBCCtrlC_0_IB_BACK_PORCH_LSB 0x5
+#define QIB_7322_IBCCtrlC_0_IB_BACK_PORCH_MSB 0x9
+#define QIB_7322_IBCCtrlC_0_IB_BACK_PORCH_RMASK 0x1F
+#define QIB_7322_IBCCtrlC_0_IB_FRONT_PORCH_LSB 0x0
+#define QIB_7322_IBCCtrlC_0_IB_FRONT_PORCH_MSB 0x4
+#define QIB_7322_IBCCtrlC_0_IB_FRONT_PORCH_RMASK 0x1F
+
+#define QIB_7322_HRTBT_GUID_0_OFFS 0x1588
+#define QIB_7322_HRTBT_GUID_0_DEF 0x0000000000000000
+
+#define QIB_7322_IB_SDTEST_IF_TX_0_OFFS 0x1590
+#define QIB_7322_IB_SDTEST_IF_TX_0_DEF 0x0000000000000000
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_RX_CFG_LSB 0x30
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_RX_CFG_MSB 0x3F
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_RX_CFG_RMASK 0xFFFF
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_TX_CFG_LSB 0x20
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_TX_CFG_MSB 0x2F
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_TX_CFG_RMASK 0xFFFF
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_SPEED_LSB 0xD
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_SPEED_MSB 0xF
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_SPEED_RMASK 0x7
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_OPCODE_LSB 0xB
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_OPCODE_MSB 0xC
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_TX_OPCODE_RMASK 0x3
+#define QIB_7322_IB_SDTEST_IF_TX_0_CREDIT_CHANGE_LSB 0x4
+#define QIB_7322_IB_SDTEST_IF_TX_0_CREDIT_CHANGE_MSB 0x4
+#define QIB_7322_IB_SDTEST_IF_TX_0_CREDIT_CHANGE_RMASK 0x1
+#define QIB_7322_IB_SDTEST_IF_TX_0_VL_CAP_LSB 0x2
+#define QIB_7322_IB_SDTEST_IF_TX_0_VL_CAP_MSB 0x3
+#define QIB_7322_IB_SDTEST_IF_TX_0_VL_CAP_RMASK 0x3
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_3_TX_VALID_LSB 0x1
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_3_TX_VALID_MSB 0x1
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_3_TX_VALID_RMASK 0x1
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_T_TX_VALID_LSB 0x0
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_T_TX_VALID_MSB 0x0
+#define QIB_7322_IB_SDTEST_IF_TX_0_TS_T_TX_VALID_RMASK 0x1
+
+#define QIB_7322_IB_SDTEST_IF_RX_0_OFFS 0x1598
+#define QIB_7322_IB_SDTEST_IF_RX_0_DEF 0x0000000000000000
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_RX_CFG_LSB 0x30
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_RX_CFG_MSB 0x3F
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_RX_CFG_RMASK 0xFFFF
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_TX_CFG_LSB 0x20
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_TX_CFG_MSB 0x2F
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_TX_CFG_RMASK 0xFFFF
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_B_LSB 0x18
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_B_MSB 0x1F
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_B_RMASK 0xFF
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_A_LSB 0x10
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_A_MSB 0x17
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_RX_A_RMASK 0xFF
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_3_RX_VALID_LSB 0x1
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_3_RX_VALID_MSB 0x1
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_3_RX_VALID_RMASK 0x1
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_T_RX_VALID_LSB 0x0
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_T_RX_VALID_MSB 0x0
+#define QIB_7322_IB_SDTEST_IF_RX_0_TS_T_RX_VALID_RMASK 0x1
+
+#define QIB_7322_IBNCModeCtrl_0_OFFS 0x15B8
+#define QIB_7322_IBNCModeCtrl_0_DEF 0x0000000000000000
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteForce_LSB 0x22
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteForce_MSB 0x22
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteForce_RMASK 0x1
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteMask_LSB 0x21
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteMask_MSB 0x21
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapRemoteMask_RMASK 0x1
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapLocal_LSB 0x20
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapLocal_MSB 0x20
+#define QIB_7322_IBNCModeCtrl_0_ScrambleCapLocal_RMASK 0x1
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS2_LSB 0x11
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS2_MSB 0x19
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS2_RMASK 0x1FF
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS1_LSB 0x8
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS1_MSB 0x10
+#define QIB_7322_IBNCModeCtrl_0_TSMCode_TS1_RMASK 0x1FF
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_ignore_TSM_on_rx_LSB 0x2
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_ignore_TSM_on_rx_MSB 0x2
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_ignore_TSM_on_rx_RMASK 0x1
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS2_LSB 0x1
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS2_MSB 0x1
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS2_RMASK 0x1
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS1_LSB 0x0
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS1_MSB 0x0
+#define QIB_7322_IBNCModeCtrl_0_TSMEnable_send_TS1_RMASK 0x1
+
+#define QIB_7322_IBSerdesStatus_0_OFFS 0x15D0
+#define QIB_7322_IBSerdesStatus_0_DEF 0x0000000000000000
+
+#define QIB_7322_IBPCSConfig_0_OFFS 0x15D8
+#define QIB_7322_IBPCSConfig_0_DEF 0x0000000000000007
+#define QIB_7322_IBPCSConfig_0_link_sync_mask_LSB 0x9
+#define QIB_7322_IBPCSConfig_0_link_sync_mask_MSB 0x12
+#define QIB_7322_IBPCSConfig_0_link_sync_mask_RMASK 0x3FF
+#define QIB_7322_IBPCSConfig_0_xcv_rreset_LSB 0x2
+#define QIB_7322_IBPCSConfig_0_xcv_rreset_MSB 0x2
+#define QIB_7322_IBPCSConfig_0_xcv_rreset_RMASK 0x1
+#define QIB_7322_IBPCSConfig_0_xcv_treset_LSB 0x1
+#define QIB_7322_IBPCSConfig_0_xcv_treset_MSB 0x1
+#define QIB_7322_IBPCSConfig_0_xcv_treset_RMASK 0x1
+#define QIB_7322_IBPCSConfig_0_tx_rx_reset_LSB 0x0
+#define QIB_7322_IBPCSConfig_0_tx_rx_reset_MSB 0x0
+#define QIB_7322_IBPCSConfig_0_tx_rx_reset_RMASK 0x1
+
+#define QIB_7322_IBSerdesCtrl_0_OFFS 0x15E0
+#define QIB_7322_IBSerdesCtrl_0_DEF 0x0000000000FFA00F
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_QDR_LSB 0x1A
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_QDR_MSB 0x1A
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_QDR_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_DDR_LSB 0x19
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_DDR_MSB 0x19
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_DDR_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_SDR_LSB 0x18
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_SDR_MSB 0x18
+#define QIB_7322_IBSerdesCtrl_0_DISABLE_RXLATOFF_SDR_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_CHANNEL_RESET_N_LSB 0x14
+#define QIB_7322_IBSerdesCtrl_0_CHANNEL_RESET_N_MSB 0x17
+#define QIB_7322_IBSerdesCtrl_0_CHANNEL_RESET_N_RMASK 0xF
+#define QIB_7322_IBSerdesCtrl_0_CGMODE_LSB 0x10
+#define QIB_7322_IBSerdesCtrl_0_CGMODE_MSB 0x13
+#define QIB_7322_IBSerdesCtrl_0_CGMODE_RMASK 0xF
+#define QIB_7322_IBSerdesCtrl_0_IB_LAT_MODE_LSB 0xF
+#define QIB_7322_IBSerdesCtrl_0_IB_LAT_MODE_MSB 0xF
+#define QIB_7322_IBSerdesCtrl_0_IB_LAT_MODE_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_RXLOSEN_LSB 0xD
+#define QIB_7322_IBSerdesCtrl_0_RXLOSEN_MSB 0xD
+#define QIB_7322_IBSerdesCtrl_0_RXLOSEN_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_LPEN_LSB 0xC
+#define QIB_7322_IBSerdesCtrl_0_LPEN_MSB 0xC
+#define QIB_7322_IBSerdesCtrl_0_LPEN_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_PLLPD_LSB 0xB
+#define QIB_7322_IBSerdesCtrl_0_PLLPD_MSB 0xB
+#define QIB_7322_IBSerdesCtrl_0_PLLPD_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_TXPD_LSB 0xA
+#define QIB_7322_IBSerdesCtrl_0_TXPD_MSB 0xA
+#define QIB_7322_IBSerdesCtrl_0_TXPD_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_RXPD_LSB 0x9
+#define QIB_7322_IBSerdesCtrl_0_RXPD_MSB 0x9
+#define QIB_7322_IBSerdesCtrl_0_RXPD_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_TXIDLE_LSB 0x8
+#define QIB_7322_IBSerdesCtrl_0_TXIDLE_MSB 0x8
+#define QIB_7322_IBSerdesCtrl_0_TXIDLE_RMASK 0x1
+#define QIB_7322_IBSerdesCtrl_0_CMODE_LSB 0x0
+#define QIB_7322_IBSerdesCtrl_0_CMODE_MSB 0x6
+#define QIB_7322_IBSerdesCtrl_0_CMODE_RMASK 0x7F
+
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_OFFS 0x1600
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_DEF 0x0000000000000000
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_tx_override_deemphasis_select_LSB 0x1F
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_tx_override_deemphasis_select_MSB 0x1F
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_tx_override_deemphasis_select_RMASK 0x1
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_reset_tx_deemphasis_override_LSB 0x1E
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_reset_tx_deemphasis_override_MSB 0x1E
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_reset_tx_deemphasis_override_RMASK 0x1
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txampcntl_d2a_LSB 0xE
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txampcntl_d2a_MSB 0x11
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txampcntl_d2a_RMASK 0xF
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txc0_ena_LSB 0x9
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txc0_ena_MSB 0xD
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txc0_ena_RMASK 0x1F
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcp1_ena_LSB 0x5
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcp1_ena_MSB 0x8
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcp1_ena_RMASK 0xF
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_xtra_emph0_LSB 0x3
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_xtra_emph0_MSB 0x4
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_xtra_emph0_RMASK 0x3
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_ena_LSB 0x0
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_ena_MSB 0x2
+#define QIB_7322_IBSD_TX_DEEMPHASIS_OVERRIDE_0_txcn1_ena_RMASK 0x7
+
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_OFFS 0x1640
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenagain_sdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenale_sdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_STATIC_SDR_0_static_disable_rxenadfe_sdr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_OFFS 0x1648
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenagain_sdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenale_sdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_SDR_0_dyn_disable_rxenadfe_sdr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_OFFS 0x1650
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenagain_ddr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenale_ddr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_STATIC_DDR_0_static_disable_rxenadfe_ddr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_OFFS 0x1658
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenagain_ddr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenale_ddr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_DDR_0_dyn_disable_rxenadfe_ddr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_OFFS 0x1660
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenagain_qdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenale_qdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_STATIC_QDR_0_static_disable_rxenadfe_qdr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_OFFS 0x1668
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_DEF 0x0000000000000000
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch3_LSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch3_MSB 0x27
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch2_LSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch2_MSB 0x26
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch1_LSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch1_MSB 0x25
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch0_LSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch0_MSB 0x24
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenagain_qdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch3_LSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch3_MSB 0x23
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch3_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch2_LSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch2_MSB 0x22
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch2_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch1_LSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch1_MSB 0x21
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch1_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch0_LSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch0_MSB 0x20
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenale_qdr_ch0_RMASK 0x1
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch3_LSB 0x18
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch3_MSB 0x1F
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch3_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch2_LSB 0x10
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch2_MSB 0x17
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch2_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch1_LSB 0x8
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch1_MSB 0xF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch1_RMASK 0xFF
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch0_LSB 0x0
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch0_MSB 0x7
+#define QIB_7322_ADAPT_DISABLE_DYNAMIC_QDR_0_dyn_disable_rxenadfe_qdr_ch0_RMASK 0xFF
+
+#define QIB_7322_ADAPT_DISABLE_TIMER_THRESHOLD_0_OFFS 0x1670
+#define QIB_7322_ADAPT_DISABLE_TIMER_THRESHOLD_0_DEF 0x0000000000000000
+
+#define QIB_7322_HighPriorityLimit_0_OFFS 0x1BC0
+#define QIB_7322_HighPriorityLimit_0_DEF 0x0000000000000000
+#define QIB_7322_HighPriorityLimit_0_Limit_LSB 0x0
+#define QIB_7322_HighPriorityLimit_0_Limit_MSB 0x7
+#define QIB_7322_HighPriorityLimit_0_Limit_RMASK 0xFF
+
+#define QIB_7322_LowPriority0_0_OFFS 0x1C00
+#define QIB_7322_LowPriority0_0_DEF 0x0000000000000000
+#define QIB_7322_LowPriority0_0_VirtualLane_LSB 0x10
+#define QIB_7322_LowPriority0_0_VirtualLane_MSB 0x12
+#define QIB_7322_LowPriority0_0_VirtualLane_RMASK 0x7
+#define QIB_7322_LowPriority0_0_Weight_LSB 0x0
+#define QIB_7322_LowPriority0_0_Weight_MSB 0x7
+#define QIB_7322_LowPriority0_0_Weight_RMASK 0xFF
+
+#define QIB_7322_HighPriority0_0_OFFS 0x1E00
+#define QIB_7322_HighPriority0_0_DEF 0x0000000000000000
+#define QIB_7322_HighPriority0_0_VirtualLane_LSB 0x10
+#define QIB_7322_HighPriority0_0_VirtualLane_MSB 0x12
+#define QIB_7322_HighPriority0_0_VirtualLane_RMASK 0x7
+#define QIB_7322_HighPriority0_0_Weight_LSB 0x0
+#define QIB_7322_HighPriority0_0_Weight_MSB 0x7
+#define QIB_7322_HighPriority0_0_Weight_RMASK 0xFF
+
+#define QIB_7322_CntrRegBase_1_OFFS 0x2028
+#define QIB_7322_CntrRegBase_1_DEF 0x0000000000013000
+
+#define QIB_7322_RcvQPMulticastContext_1_OFFS 0x2170
+
+#define QIB_7322_SendCtrl_1_OFFS 0x21C0
+
+#define QIB_7322_SendBufAvail0_OFFS 0x3000
+#define QIB_7322_SendBufAvail0_DEF 0x0000000000000000
+#define QIB_7322_SendBufAvail0_SendBuf_31_0_LSB 0x0
+#define QIB_7322_SendBufAvail0_SendBuf_31_0_MSB 0x3F
+#define QIB_7322_SendBufAvail0_SendBuf_31_0_RMASK 0x0
+
+#define QIB_7322_MsixTable_OFFS 0x8000
+#define QIB_7322_MsixTable_DEF 0x0000000000000000
+
+#define QIB_7322_MsixPba_OFFS 0x9000
+#define QIB_7322_MsixPba_DEF 0x0000000000000000
+
+#define QIB_7322_LAMemory_OFFS 0xA000
+#define QIB_7322_LAMemory_DEF 0x0000000000000000
+
+#define QIB_7322_LBIntCnt_OFFS 0x11000
+#define QIB_7322_LBIntCnt_DEF 0x0000000000000000
+
+#define QIB_7322_LBFlowStallCnt_OFFS 0x11008
+#define QIB_7322_LBFlowStallCnt_DEF 0x0000000000000000
+
+#define QIB_7322_RxTIDFullErrCnt_OFFS 0x110D0
+#define QIB_7322_RxTIDFullErrCnt_DEF 0x0000000000000000
+
+#define QIB_7322_RxTIDValidErrCnt_OFFS 0x110D8
+#define QIB_7322_RxTIDValidErrCnt_DEF 0x0000000000000000
+
+#define QIB_7322_RxP0HdrEgrOvflCnt_OFFS 0x110E8
+#define QIB_7322_RxP0HdrEgrOvflCnt_DEF 0x0000000000000000
+
+#define QIB_7322_PcieRetryBufDiagQwordCnt_OFFS 0x111A0
+#define QIB_7322_PcieRetryBufDiagQwordCnt_DEF 0x0000000000000000
+
+#define QIB_7322_RxTidFlowDropCnt_OFFS 0x111E0
+#define QIB_7322_RxTidFlowDropCnt_DEF 0x0000000000000000
+
+#define QIB_7322_LBIntCnt_0_OFFS 0x12000
+#define QIB_7322_LBIntCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxCreditUpToDateTimeOut_0_OFFS 0x12008
+#define QIB_7322_TxCreditUpToDateTimeOut_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxSDmaDescCnt_0_OFFS 0x12010
+#define QIB_7322_TxSDmaDescCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxUnsupVLErrCnt_0_OFFS 0x12018
+#define QIB_7322_TxUnsupVLErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxDataPktCnt_0_OFFS 0x12020
+#define QIB_7322_TxDataPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxFlowPktCnt_0_OFFS 0x12028
+#define QIB_7322_TxFlowPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxDwordCnt_0_OFFS 0x12030
+#define QIB_7322_TxDwordCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxLenErrCnt_0_OFFS 0x12038
+#define QIB_7322_TxLenErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxMaxMinLenErrCnt_0_OFFS 0x12040
+#define QIB_7322_TxMaxMinLenErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxUnderrunCnt_0_OFFS 0x12048
+#define QIB_7322_TxUnderrunCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxFlowStallCnt_0_OFFS 0x12050
+#define QIB_7322_TxFlowStallCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxDroppedPktCnt_0_OFFS 0x12058
+#define QIB_7322_TxDroppedPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxDroppedPktCnt_0_OFFS 0x12060
+#define QIB_7322_RxDroppedPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxDataPktCnt_0_OFFS 0x12068
+#define QIB_7322_RxDataPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxFlowPktCnt_0_OFFS 0x12070
+#define QIB_7322_RxFlowPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxDwordCnt_0_OFFS 0x12078
+#define QIB_7322_RxDwordCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxLenErrCnt_0_OFFS 0x12080
+#define QIB_7322_RxLenErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxMaxMinLenErrCnt_0_OFFS 0x12088
+#define QIB_7322_RxMaxMinLenErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxICRCErrCnt_0_OFFS 0x12090
+#define QIB_7322_RxICRCErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxVCRCErrCnt_0_OFFS 0x12098
+#define QIB_7322_RxVCRCErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxFlowCtrlViolCnt_0_OFFS 0x120A0
+#define QIB_7322_RxFlowCtrlViolCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxVersionErrCnt_0_OFFS 0x120A8
+#define QIB_7322_RxVersionErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxLinkMalformCnt_0_OFFS 0x120B0
+#define QIB_7322_RxLinkMalformCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxEBPCnt_0_OFFS 0x120B8
+#define QIB_7322_RxEBPCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxLPCRCErrCnt_0_OFFS 0x120C0
+#define QIB_7322_RxLPCRCErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxBufOvflCnt_0_OFFS 0x120C8
+#define QIB_7322_RxBufOvflCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxLenTruncateCnt_0_OFFS 0x120D0
+#define QIB_7322_RxLenTruncateCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxPKeyMismatchCnt_0_OFFS 0x120E0
+#define QIB_7322_RxPKeyMismatchCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_IBLinkDownedCnt_0_OFFS 0x12180
+#define QIB_7322_IBLinkDownedCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_IBSymbolErrCnt_0_OFFS 0x12188
+#define QIB_7322_IBSymbolErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_IBStatusChangeCnt_0_OFFS 0x12190
+#define QIB_7322_IBStatusChangeCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_IBLinkErrRecoveryCnt_0_OFFS 0x12198
+#define QIB_7322_IBLinkErrRecoveryCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_ExcessBufferOvflCnt_0_OFFS 0x121A8
+#define QIB_7322_ExcessBufferOvflCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_LocalLinkIntegrityErrCnt_0_OFFS 0x121B0
+#define QIB_7322_LocalLinkIntegrityErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxVlErrCnt_0_OFFS 0x121B8
+#define QIB_7322_RxVlErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxDlidFltrCnt_0_OFFS 0x121C0
+#define QIB_7322_RxDlidFltrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxVL15DroppedPktCnt_0_OFFS 0x121C8
+#define QIB_7322_RxVL15DroppedPktCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxOtherLocalPhyErrCnt_0_OFFS 0x121D0
+#define QIB_7322_RxOtherLocalPhyErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_RxQPInvalidContextCnt_0_OFFS 0x121D8
+#define QIB_7322_RxQPInvalidContextCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_TxHeadersErrCnt_0_OFFS 0x121F8
+#define QIB_7322_TxHeadersErrCnt_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSRcvDataCount_0_OFFS 0x12218
+#define QIB_7322_PSRcvDataCount_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSRcvPktsCount_0_OFFS 0x12220
+#define QIB_7322_PSRcvPktsCount_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitDataCount_0_OFFS 0x12228
+#define QIB_7322_PSXmitDataCount_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitPktsCount_0_OFFS 0x12230
+#define QIB_7322_PSXmitPktsCount_0_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitWaitCount_0_OFFS 0x12238
+#define QIB_7322_PSXmitWaitCount_0_DEF 0x0000000000000000
+
+#define QIB_7322_LBIntCnt_1_OFFS 0x13000
+#define QIB_7322_LBIntCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxCreditUpToDateTimeOut_1_OFFS 0x13008
+#define QIB_7322_TxCreditUpToDateTimeOut_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxSDmaDescCnt_1_OFFS 0x13010
+#define QIB_7322_TxSDmaDescCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxUnsupVLErrCnt_1_OFFS 0x13018
+#define QIB_7322_TxUnsupVLErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxDataPktCnt_1_OFFS 0x13020
+#define QIB_7322_TxDataPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxFlowPktCnt_1_OFFS 0x13028
+#define QIB_7322_TxFlowPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxDwordCnt_1_OFFS 0x13030
+#define QIB_7322_TxDwordCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxLenErrCnt_1_OFFS 0x13038
+#define QIB_7322_TxLenErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxMaxMinLenErrCnt_1_OFFS 0x13040
+#define QIB_7322_TxMaxMinLenErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxUnderrunCnt_1_OFFS 0x13048
+#define QIB_7322_TxUnderrunCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxFlowStallCnt_1_OFFS 0x13050
+#define QIB_7322_TxFlowStallCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxDroppedPktCnt_1_OFFS 0x13058
+#define QIB_7322_TxDroppedPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxDroppedPktCnt_1_OFFS 0x13060
+#define QIB_7322_RxDroppedPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxDataPktCnt_1_OFFS 0x13068
+#define QIB_7322_RxDataPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxFlowPktCnt_1_OFFS 0x13070
+#define QIB_7322_RxFlowPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxDwordCnt_1_OFFS 0x13078
+#define QIB_7322_RxDwordCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxLenErrCnt_1_OFFS 0x13080
+#define QIB_7322_RxLenErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxMaxMinLenErrCnt_1_OFFS 0x13088
+#define QIB_7322_RxMaxMinLenErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxICRCErrCnt_1_OFFS 0x13090
+#define QIB_7322_RxICRCErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxVCRCErrCnt_1_OFFS 0x13098
+#define QIB_7322_RxVCRCErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxFlowCtrlViolCnt_1_OFFS 0x130A0
+#define QIB_7322_RxFlowCtrlViolCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxVersionErrCnt_1_OFFS 0x130A8
+#define QIB_7322_RxVersionErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxLinkMalformCnt_1_OFFS 0x130B0
+#define QIB_7322_RxLinkMalformCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxEBPCnt_1_OFFS 0x130B8
+#define QIB_7322_RxEBPCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxLPCRCErrCnt_1_OFFS 0x130C0
+#define QIB_7322_RxLPCRCErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxBufOvflCnt_1_OFFS 0x130C8
+#define QIB_7322_RxBufOvflCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxLenTruncateCnt_1_OFFS 0x130D0
+#define QIB_7322_RxLenTruncateCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxPKeyMismatchCnt_1_OFFS 0x130E0
+#define QIB_7322_RxPKeyMismatchCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_IBLinkDownedCnt_1_OFFS 0x13180
+#define QIB_7322_IBLinkDownedCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_IBSymbolErrCnt_1_OFFS 0x13188
+#define QIB_7322_IBSymbolErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_IBStatusChangeCnt_1_OFFS 0x13190
+#define QIB_7322_IBStatusChangeCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_IBLinkErrRecoveryCnt_1_OFFS 0x13198
+#define QIB_7322_IBLinkErrRecoveryCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_ExcessBufferOvflCnt_1_OFFS 0x131A8
+#define QIB_7322_ExcessBufferOvflCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_LocalLinkIntegrityErrCnt_1_OFFS 0x131B0
+#define QIB_7322_LocalLinkIntegrityErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxVlErrCnt_1_OFFS 0x131B8
+#define QIB_7322_RxVlErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxDlidFltrCnt_1_OFFS 0x131C0
+#define QIB_7322_RxDlidFltrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxVL15DroppedPktCnt_1_OFFS 0x131C8
+#define QIB_7322_RxVL15DroppedPktCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxOtherLocalPhyErrCnt_1_OFFS 0x131D0
+#define QIB_7322_RxOtherLocalPhyErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_RxQPInvalidContextCnt_1_OFFS 0x131D8
+#define QIB_7322_RxQPInvalidContextCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_TxHeadersErrCnt_1_OFFS 0x131F8
+#define QIB_7322_TxHeadersErrCnt_1_DEF 0x0000000000000000
+
+#define QIB_7322_PSRcvDataCount_1_OFFS 0x13218
+#define QIB_7322_PSRcvDataCount_1_DEF 0x0000000000000000
+
+#define QIB_7322_PSRcvPktsCount_1_OFFS 0x13220
+#define QIB_7322_PSRcvPktsCount_1_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitDataCount_1_OFFS 0x13228
+#define QIB_7322_PSXmitDataCount_1_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitPktsCount_1_OFFS 0x13230
+#define QIB_7322_PSXmitPktsCount_1_DEF 0x0000000000000000
+
+#define QIB_7322_PSXmitWaitCount_1_OFFS 0x13238
+#define QIB_7322_PSXmitWaitCount_1_DEF 0x0000000000000000
+
+#define QIB_7322_RcvEgrArray_OFFS 0x14000
+#define QIB_7322_RcvEgrArray_DEF 0x0000000000000000
+#define QIB_7322_RcvEgrArray_RT_BufSize_LSB 0x25
+#define QIB_7322_RcvEgrArray_RT_BufSize_MSB 0x27
+#define QIB_7322_RcvEgrArray_RT_BufSize_RMASK 0x7
+#define QIB_7322_RcvEgrArray_RT_Addr_LSB 0x0
+#define QIB_7322_RcvEgrArray_RT_Addr_MSB 0x24
+#define QIB_7322_RcvEgrArray_RT_Addr_RMASK 0x1FFFFFFFFF
+
+#define QIB_7322_RcvTIDArray0_OFFS 0x50000
+#define QIB_7322_RcvTIDArray0_DEF 0x0000000000000000
+#define QIB_7322_RcvTIDArray0_RT_BufSize_LSB 0x25
+#define QIB_7322_RcvTIDArray0_RT_BufSize_MSB 0x27
+#define QIB_7322_RcvTIDArray0_RT_BufSize_RMASK 0x7
+#define QIB_7322_RcvTIDArray0_RT_Addr_LSB 0x0
+#define QIB_7322_RcvTIDArray0_RT_Addr_MSB 0x24
+#define QIB_7322_RcvTIDArray0_RT_Addr_RMASK 0x1FFFFFFFFF
+
+#define QIB_7322_IBSD_DDS_MAP_TABLE_0_OFFS 0xD0000
+#define QIB_7322_IBSD_DDS_MAP_TABLE_0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvHdrTail0_OFFS 0x200000
+#define QIB_7322_RcvHdrTail0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvHdrHead0_OFFS 0x200008
+#define QIB_7322_RcvHdrHead0_DEF 0x0000000000000000
+#define QIB_7322_RcvHdrHead0_counter_LSB 0x20
+#define QIB_7322_RcvHdrHead0_counter_MSB 0x2F
+#define QIB_7322_RcvHdrHead0_counter_RMASK 0xFFFF
+#define QIB_7322_RcvHdrHead0_RcvHeadPointer_LSB 0x0
+#define QIB_7322_RcvHdrHead0_RcvHeadPointer_MSB 0x1F
+#define QIB_7322_RcvHdrHead0_RcvHeadPointer_RMASK 0xFFFFFFFF
+
+#define QIB_7322_RcvEgrIndexTail0_OFFS 0x200010
+#define QIB_7322_RcvEgrIndexTail0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvEgrIndexHead0_OFFS 0x200018
+#define QIB_7322_RcvEgrIndexHead0_DEF 0x0000000000000000
+
+#define QIB_7322_RcvTIDFlowTable0_OFFS 0x201000
+#define QIB_7322_RcvTIDFlowTable0_DEF 0x0000000000000000
+#define QIB_7322_RcvTIDFlowTable0_GenMismatch_LSB 0x1C
+#define QIB_7322_RcvTIDFlowTable0_GenMismatch_MSB 0x1C
+#define QIB_7322_RcvTIDFlowTable0_GenMismatch_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_SeqMismatch_LSB 0x1B
+#define QIB_7322_RcvTIDFlowTable0_SeqMismatch_MSB 0x1B
+#define QIB_7322_RcvTIDFlowTable0_SeqMismatch_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_KeepOnGenErr_LSB 0x16
+#define QIB_7322_RcvTIDFlowTable0_KeepOnGenErr_MSB 0x16
+#define QIB_7322_RcvTIDFlowTable0_KeepOnGenErr_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_KeepAfterSeqErr_LSB 0x15
+#define QIB_7322_RcvTIDFlowTable0_KeepAfterSeqErr_MSB 0x15
+#define QIB_7322_RcvTIDFlowTable0_KeepAfterSeqErr_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_HdrSuppEnabled_LSB 0x14
+#define QIB_7322_RcvTIDFlowTable0_HdrSuppEnabled_MSB 0x14
+#define QIB_7322_RcvTIDFlowTable0_HdrSuppEnabled_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_FlowValid_LSB 0x13
+#define QIB_7322_RcvTIDFlowTable0_FlowValid_MSB 0x13
+#define QIB_7322_RcvTIDFlowTable0_FlowValid_RMASK 0x1
+#define QIB_7322_RcvTIDFlowTable0_GenVal_LSB 0xB
+#define QIB_7322_RcvTIDFlowTable0_GenVal_MSB 0x12
+#define QIB_7322_RcvTIDFlowTable0_GenVal_RMASK 0xFF
+#define QIB_7322_RcvTIDFlowTable0_SeqNum_LSB 0x0
+#define QIB_7322_RcvTIDFlowTable0_SeqNum_MSB 0xA
+#define QIB_7322_RcvTIDFlowTable0_SeqNum_RMASK 0x7FF
diff --git a/drivers/infiniband/hw/qib/qib_common.h b/drivers/infiniband/hw/qib/qib_common.h
new file mode 100644
index 00000000000..5670ace27c6
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_common.h
@@ -0,0 +1,812 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef _QIB_COMMON_H
+#define _QIB_COMMON_H
+
+/*
+ * This file contains defines, structures, etc. that are used
+ * to communicate between kernel and user code.
+ */
+
+/* This is the IEEE-assigned OUI for QLogic Inc. QLogic_IB */
+#define QIB_SRC_OUI_1 0x00
+#define QIB_SRC_OUI_2 0x11
+#define QIB_SRC_OUI_3 0x75
+
+/* version of protocol header (known to chip also). In the long run,
+ * we should be able to generate and accept a range of version numbers;
+ * for now we only accept one, and it's compiled in.
+ */
+#define IPS_PROTO_VERSION 2
+
+/*
+ * These are compile time constants that you may want to enable or disable
+ * if you are trying to debug problems with code or performance.
+ * QIB_VERBOSE_TRACING define as 1 if you want additional tracing in
+ * fastpath code
+ * QIB_TRACE_REGWRITES define as 1 if you want register writes to be
+ * traced in faspath code
+ * _QIB_TRACING define as 0 if you want to remove all tracing in a
+ * compilation unit
+ */
+
+/*
+ * The value in the BTH QP field that QLogic_IB uses to differentiate
+ * an qlogic_ib protocol IB packet vs standard IB transport
+ * This it needs to be even (0x656b78), because the LSB is sometimes
+ * used for the MSB of context. The change may cause a problem
+ * interoperating with older software.
+ */
+#define QIB_KD_QP 0x656b78
+
+/*
+ * These are the status bits readable (in ascii form, 64bit value)
+ * from the "status" sysfs file.  For binary compatibility, values
+ * must remain as is; removed states can be reused for different
+ * purposes.
+ */
+#define QIB_STATUS_INITTED       0x1    /* basic initialization done */
+/* Chip has been found and initted */
+#define QIB_STATUS_CHIP_PRESENT 0x20
+/* IB link is at ACTIVE, usable for data traffic */
+#define QIB_STATUS_IB_READY     0x40
+/* link is configured, LID, MTU, etc. have been set */
+#define QIB_STATUS_IB_CONF      0x80
+/* A Fatal hardware error has occurred. */
+#define QIB_STATUS_HWERROR     0x200
+
+/*
+ * The list of usermode accessible registers.  Also see Reg_* later in file.
+ */
+enum qib_ureg {
+	/* (RO)  DMA RcvHdr to be used next. */
+	ur_rcvhdrtail = 0,
+	/* (RW)  RcvHdr entry to be processed next by host. */
+	ur_rcvhdrhead = 1,
+	/* (RO)  Index of next Eager index to use. */
+	ur_rcvegrindextail = 2,
+	/* (RW)  Eager TID to be processed next */
+	ur_rcvegrindexhead = 3,
+	/* For internal use only; max register number. */
+	_QIB_UregMax
+};
+
+/* bit values for spi_runtime_flags */
+#define QIB_RUNTIME_PCIE                0x0002
+#define QIB_RUNTIME_FORCE_WC_ORDER      0x0004
+#define QIB_RUNTIME_RCVHDR_COPY         0x0008
+#define QIB_RUNTIME_MASTER              0x0010
+#define QIB_RUNTIME_RCHK                0x0020
+#define QIB_RUNTIME_NODMA_RTAIL         0x0080
+#define QIB_RUNTIME_SPECIAL_TRIGGER     0x0100
+#define QIB_RUNTIME_SDMA                0x0200
+#define QIB_RUNTIME_FORCE_PIOAVAIL      0x0400
+#define QIB_RUNTIME_PIO_REGSWAPPED      0x0800
+#define QIB_RUNTIME_CTXT_MSB_IN_QP      0x1000
+#define QIB_RUNTIME_CTXT_REDIRECT       0x2000
+#define QIB_RUNTIME_HDRSUPP             0x4000
+
+/*
+ * This structure is returned by qib_userinit() immediately after
+ * open to get implementation-specific info, and info specific to this
+ * instance.
+ *
+ * This struct must have explict pad fields where type sizes
+ * may result in different alignments between 32 and 64 bit
+ * programs, since the 64 bit * bit kernel requires the user code
+ * to have matching offsets
+ */
+struct qib_base_info {
+	/* version of hardware, for feature checking. */
+	__u32 spi_hw_version;
+	/* version of software, for feature checking. */
+	__u32 spi_sw_version;
+	/* QLogic_IB context assigned, goes into sent packets */
+	__u16 spi_ctxt;
+	__u16 spi_subctxt;
+	/*
+	 * IB MTU, packets IB data must be less than this.
+	 * The MTU is in bytes, and will be a multiple of 4 bytes.
+	 */
+	__u32 spi_mtu;
+	/*
+	 * Size of a PIO buffer.  Any given packet's total size must be less
+	 * than this (in words).  Included is the starting control word, so
+	 * if 513 is returned, then total pkt size is 512 words or less.
+	 */
+	__u32 spi_piosize;
+	/* size of the TID cache in qlogic_ib, in entries */
+	__u32 spi_tidcnt;
+	/* size of the TID Eager list in qlogic_ib, in entries */
+	__u32 spi_tidegrcnt;
+	/* size of a single receive header queue entry in words. */
+	__u32 spi_rcvhdrent_size;
+	/*
+	 * Count of receive header queue entries allocated.
+	 * This may be less than the spu_rcvhdrcnt passed in!.
+	 */
+	__u32 spi_rcvhdr_cnt;
+
+	/* per-chip and other runtime features bitmap (QIB_RUNTIME_*) */
+	__u32 spi_runtime_flags;
+
+	/* address where hardware receive header queue is mapped */
+	__u64 spi_rcvhdr_base;
+
+	/* user program. */
+
+	/* base address of eager TID receive buffers used by hardware. */
+	__u64 spi_rcv_egrbufs;
+
+	/* Allocated by initialization code, not by protocol. */
+
+	/*
+	 * Size of each TID buffer in host memory, starting at
+	 * spi_rcv_egrbufs.  The buffers are virtually contiguous.
+	 */
+	__u32 spi_rcv_egrbufsize;
+	/*
+	 * The special QP (queue pair) value that identifies an qlogic_ib
+	 * protocol packet from standard IB packets.  More, probably much
+	 * more, to be added.
+	 */
+	__u32 spi_qpair;
+
+	/*
+	 * User register base for init code, not to be used directly by
+	 * protocol or applications.  Always points to chip registers,
+	 * for normal or shared context.
+	 */
+	__u64 spi_uregbase;
+	/*
+	 * Maximum buffer size in bytes that can be used in a single TID
+	 * entry (assuming the buffer is aligned to this boundary).  This is
+	 * the minimum of what the hardware and software support Guaranteed
+	 * to be a power of 2.
+	 */
+	__u32 spi_tid_maxsize;
+	/*
+	 * alignment of each pio send buffer (byte count
+	 * to add to spi_piobufbase to get to second buffer)
+	 */
+	__u32 spi_pioalign;
+	/*
+	 * The index of the first pio buffer available to this process;
+	 * needed to do lookup in spi_pioavailaddr; not added to
+	 * spi_piobufbase.
+	 */
+	__u32 spi_pioindex;
+	 /* number of buffers mapped for this process */
+	__u32 spi_piocnt;
+
+	/*
+	 * Base address of writeonly pio buffers for this process.
+	 * Each buffer has spi_piosize words, and is aligned on spi_pioalign
+	 * boundaries.  spi_piocnt buffers are mapped from this address
+	 */
+	__u64 spi_piobufbase;
+
+	/*
+	 * Base address of readonly memory copy of the pioavail registers.
+	 * There are 2 bits for each buffer.
+	 */
+	__u64 spi_pioavailaddr;
+
+	/*
+	 * Address where driver updates a copy of the interface and driver
+	 * status (QIB_STATUS_*) as a 64 bit value.  It's followed by a
+	 * link status qword (formerly combined with driver status), then a
+	 * string indicating hardware error, if there was one.
+	 */
+	__u64 spi_status;
+
+	/* number of chip ctxts available to user processes */
+	__u32 spi_nctxts;
+	__u16 spi_unit; /* unit number of chip we are using */
+	__u16 spi_port; /* IB port number we are using */
+	/* num bufs in each contiguous set */
+	__u32 spi_rcv_egrperchunk;
+	/* size in bytes of each contiguous set */
+	__u32 spi_rcv_egrchunksize;
+	/* total size of mmap to cover full rcvegrbuffers */
+	__u32 spi_rcv_egrbuftotlen;
+	__u32 spi_rhf_offset; /* dword offset in hdrqent for rcvhdr flags */
+	/* address of readonly memory copy of the rcvhdrq tail register. */
+	__u64 spi_rcvhdr_tailaddr;
+
+	/*
+	 * shared memory pages for subctxts if ctxt is shared; these cover
+	 * all the processes in the group sharing a single context.
+	 * all have enough space for the num_subcontexts value on this job.
+	 */
+	__u64 spi_subctxt_uregbase;
+	__u64 spi_subctxt_rcvegrbuf;
+	__u64 spi_subctxt_rcvhdr_base;
+
+	/* shared memory page for send buffer disarm status */
+	__u64 spi_sendbuf_status;
+} __attribute__ ((aligned(8)));
+
+/*
+ * This version number is given to the driver by the user code during
+ * initialization in the spu_userversion field of qib_user_info, so
+ * the driver can check for compatibility with user code.
+ *
+ * The major version changes when data structures
+ * change in an incompatible way.  The driver must be the same or higher
+ * for initialization to succeed.  In some cases, a higher version
+ * driver will not interoperate with older software, and initialization
+ * will return an error.
+ */
+#define QIB_USER_SWMAJOR 1
+
+/*
+ * Minor version differences are always compatible
+ * a within a major version, however if user software is larger
+ * than driver software, some new features and/or structure fields
+ * may not be implemented; the user code must deal with this if it
+ * cares, or it must abort after initialization reports the difference.
+ */
+#define QIB_USER_SWMINOR 13
+
+#define QIB_USER_SWVERSION ((QIB_USER_SWMAJOR << 16) | QIB_USER_SWMINOR)
+
+#ifndef QIB_KERN_TYPE
+#define QIB_KERN_TYPE 0
+#endif
+
+/*
+ * Similarly, this is the kernel version going back to the user.  It's
+ * slightly different, in that we want to tell if the driver was built as
+ * part of a QLogic release, or from the driver from openfabrics.org,
+ * kernel.org, or a standard distribution, for support reasons.
+ * The high bit is 0 for non-QLogic and 1 for QLogic-built/supplied.
+ *
+ * It's returned by the driver to the user code during initialization in the
+ * spi_sw_version field of qib_base_info, so the user code can in turn
+ * check for compatibility with the kernel.
+*/
+#define QIB_KERN_SWVERSION ((QIB_KERN_TYPE << 31) | QIB_USER_SWVERSION)
+
+/*
+ * Define the driver version number.  This is something that refers only
+ * to the driver itself, not the software interfaces it supports.
+ */
+#define QIB_DRIVER_VERSION_BASE "1.11"
+
+/* create the final driver version string */
+#ifdef QIB_IDSTR
+#define QIB_DRIVER_VERSION QIB_DRIVER_VERSION_BASE " " QIB_IDSTR
+#else
+#define QIB_DRIVER_VERSION QIB_DRIVER_VERSION_BASE
+#endif
+
+/*
+ * If the unit is specified via open, HCA choice is fixed.  If port is
+ * specified, it's also fixed.  Otherwise we try to spread contexts
+ * across ports and HCAs, using different algorithims.  WITHIN is
+ * the old default, prior to this mechanism.
+ */
+#define QIB_PORT_ALG_ACROSS 0 /* round robin contexts across HCAs, then
+			       * ports; this is the default */
+#define QIB_PORT_ALG_WITHIN 1 /* use all contexts on an HCA (round robin
+			       * active ports within), then next HCA */
+#define QIB_PORT_ALG_COUNT 2 /* number of algorithm choices */
+
+/*
+ * This structure is passed to qib_userinit() to tell the driver where
+ * user code buffers are, sizes, etc.   The offsets and sizes of the
+ * fields must remain unchanged, for binary compatibility.  It can
+ * be extended, if userversion is changed so user code can tell, if needed
+ */
+struct qib_user_info {
+	/*
+	 * version of user software, to detect compatibility issues.
+	 * Should be set to QIB_USER_SWVERSION.
+	 */
+	__u32 spu_userversion;
+
+	__u32 _spu_unused2;
+
+	/* size of struct base_info to write to */
+	__u32 spu_base_info_size;
+
+	__u32 spu_port_alg; /* which QIB_PORT_ALG_*; unused user minor < 11 */
+
+	/*
+	 * If two or more processes wish to share a context, each process
+	 * must set the spu_subctxt_cnt and spu_subctxt_id to the same
+	 * values.  The only restriction on the spu_subctxt_id is that
+	 * it be unique for a given node.
+	 */
+	__u16 spu_subctxt_cnt;
+	__u16 spu_subctxt_id;
+
+	__u32 spu_port; /* IB port requested by user if > 0 */
+
+	/*
+	 * address of struct base_info to write to
+	 */
+	__u64 spu_base_info;
+
+} __attribute__ ((aligned(8)));
+
+/* User commands. */
+
+/* 16 available, was: old set up userspace (for old user code) */
+#define QIB_CMD_CTXT_INFO       17      /* find out what resources we got */
+#define QIB_CMD_RECV_CTRL       18      /* control receipt of packets */
+#define QIB_CMD_TID_UPDATE      19      /* update expected TID entries */
+#define QIB_CMD_TID_FREE        20      /* free expected TID entries */
+#define QIB_CMD_SET_PART_KEY    21      /* add partition key */
+/* 22 available, was: return info on slave processes (for old user code) */
+#define QIB_CMD_ASSIGN_CTXT     23      /* allocate HCA and ctxt */
+#define QIB_CMD_USER_INIT       24      /* set up userspace */
+#define QIB_CMD_UNUSED_1        25
+#define QIB_CMD_UNUSED_2        26
+#define QIB_CMD_PIOAVAILUPD     27      /* force an update of PIOAvail reg */
+#define QIB_CMD_POLL_TYPE       28      /* set the kind of polling we want */
+#define QIB_CMD_ARMLAUNCH_CTRL  29      /* armlaunch detection control */
+/* 30 is unused */
+#define QIB_CMD_SDMA_INFLIGHT   31      /* sdma inflight counter request */
+#define QIB_CMD_SDMA_COMPLETE   32      /* sdma completion counter request */
+/* 33 available, was a testing feature  */
+#define QIB_CMD_DISARM_BUFS     34      /* disarm send buffers w/ errors */
+#define QIB_CMD_ACK_EVENT       35      /* ack & clear bits */
+#define QIB_CMD_CPUS_LIST       36      /* list of cpus allocated, for pinned
+					 * processes: qib_cpus_list */
+
+/*
+ * QIB_CMD_ACK_EVENT obsoletes QIB_CMD_DISARM_BUFS, but we keep it for
+ * compatibility with libraries from previous release.   The ACK_EVENT
+ * will take appropriate driver action (if any, just DISARM for now),
+ * then clear the bits passed in as part of the mask.  These bits are
+ * in the first 64bit word at spi_sendbuf_status, and are passed to
+ * the driver in the event_mask union as well.
+ */
+#define _QIB_EVENT_DISARM_BUFS_BIT	0
+#define _QIB_EVENT_LINKDOWN_BIT		1
+#define _QIB_EVENT_LID_CHANGE_BIT	2
+#define _QIB_EVENT_LMC_CHANGE_BIT	3
+#define _QIB_EVENT_SL2VL_CHANGE_BIT	4
+#define _QIB_MAX_EVENT_BIT _QIB_EVENT_SL2VL_CHANGE_BIT
+
+#define QIB_EVENT_DISARM_BUFS_BIT	(1UL << _QIB_EVENT_DISARM_BUFS_BIT)
+#define QIB_EVENT_LINKDOWN_BIT		(1UL << _QIB_EVENT_LINKDOWN_BIT)
+#define QIB_EVENT_LID_CHANGE_BIT	(1UL << _QIB_EVENT_LID_CHANGE_BIT)
+#define QIB_EVENT_LMC_CHANGE_BIT	(1UL << _QIB_EVENT_LMC_CHANGE_BIT)
+#define QIB_EVENT_SL2VL_CHANGE_BIT	(1UL << _QIB_EVENT_SL2VL_CHANGE_BIT)
+
+
+/*
+ * Poll types
+ */
+#define QIB_POLL_TYPE_ANYRCV     0x0
+#define QIB_POLL_TYPE_URGENT     0x1
+
+struct qib_ctxt_info {
+	__u16 num_active;       /* number of active units */
+	__u16 unit;             /* unit (chip) assigned to caller */
+	__u16 port;             /* IB port assigned to caller (1-based) */
+	__u16 ctxt;             /* ctxt on unit assigned to caller */
+	__u16 subctxt;          /* subctxt on unit assigned to caller */
+	__u16 num_ctxts;        /* number of ctxts available on unit */
+	__u16 num_subctxts;     /* number of subctxts opened on ctxt */
+	__u16 rec_cpu;          /* cpu # for affinity (ffff if none) */
+};
+
+struct qib_tid_info {
+	__u32 tidcnt;
+	/* make structure same size in 32 and 64 bit */
+	__u32 tid__unused;
+	/* virtual address of first page in transfer */
+	__u64 tidvaddr;
+	/* pointer (same size 32/64 bit) to __u16 tid array */
+	__u64 tidlist;
+
+	/*
+	 * pointer (same size 32/64 bit) to bitmap of TIDs used
+	 * for this call; checked for being large enough at open
+	 */
+	__u64 tidmap;
+};
+
+struct qib_cmd {
+	__u32 type;                     /* command type */
+	union {
+		struct qib_tid_info tid_info;
+		struct qib_user_info user_info;
+
+		/*
+		 * address in userspace where we should put the sdma
+		 * inflight counter
+		 */
+		__u64 sdma_inflight;
+		/*
+		 * address in userspace where we should put the sdma
+		 * completion counter
+		 */
+		__u64 sdma_complete;
+		/* address in userspace of struct qib_ctxt_info to
+		   write result to */
+		__u64 ctxt_info;
+		/* enable/disable receipt of packets */
+		__u32 recv_ctrl;
+		/* enable/disable armlaunch errors (non-zero to enable) */
+		__u32 armlaunch_ctrl;
+		/* partition key to set */
+		__u16 part_key;
+		/* user address of __u32 bitmask of active slaves */
+		__u64 slave_mask_addr;
+		/* type of polling we want */
+		__u16 poll_type;
+		/* back pressure enable bit for one particular context */
+		__u8 ctxt_bp;
+		/* qib_user_event_ack(), IPATH_EVENT_* bits */
+		__u64 event_mask;
+	} cmd;
+};
+
+struct qib_iovec {
+	/* Pointer to data, but same size 32 and 64 bit */
+	__u64 iov_base;
+
+	/*
+	 * Length of data; don't need 64 bits, but want
+	 * qib_sendpkt to remain same size as before 32 bit changes, so...
+	 */
+	__u64 iov_len;
+};
+
+/*
+ * Describes a single packet for send.  Each packet can have one or more
+ * buffers, but the total length (exclusive of IB headers) must be less
+ * than the MTU, and if using the PIO method, entire packet length,
+ * including IB headers, must be less than the qib_piosize value (words).
+ * Use of this necessitates including sys/uio.h
+ */
+struct __qib_sendpkt {
+	__u32 sps_flags;        /* flags for packet (TBD) */
+	__u32 sps_cnt;          /* number of entries to use in sps_iov */
+	/* array of iov's describing packet. TEMPORARY */
+	struct qib_iovec sps_iov[4];
+};
+
+/*
+ * Diagnostics can send a packet by "writing" the following
+ * structs to the diag data special file.
+ * This allows a custom
+ * pbc (+ static rate) qword, so that special modes and deliberate
+ * changes to CRCs can be used. The elements were also re-ordered
+ * for better alignment and to avoid padding issues.
+ */
+#define _DIAG_XPKT_VERS 3
+struct qib_diag_xpkt {
+	__u16 version;
+	__u16 unit;
+	__u16 port;
+	__u16 len;
+	__u64 data;
+	__u64 pbc_wd;
+};
+
+/*
+ * Data layout in I2C flash (for GUID, etc.)
+ * All fields are little-endian binary unless otherwise stated
+ */
+#define QIB_FLASH_VERSION 2
+struct qib_flash {
+	/* flash layout version (QIB_FLASH_VERSION) */
+	__u8 if_fversion;
+	/* checksum protecting if_length bytes */
+	__u8 if_csum;
+	/*
+	 * valid length (in use, protected by if_csum), including
+	 * if_fversion and if_csum themselves)
+	 */
+	__u8 if_length;
+	/* the GUID, in network order */
+	__u8 if_guid[8];
+	/* number of GUIDs to use, starting from if_guid */
+	__u8 if_numguid;
+	/* the (last 10 characters of) board serial number, in ASCII */
+	char if_serial[12];
+	/* board mfg date (YYYYMMDD ASCII) */
+	char if_mfgdate[8];
+	/* last board rework/test date (YYYYMMDD ASCII) */
+	char if_testdate[8];
+	/* logging of error counts, TBD */
+	__u8 if_errcntp[4];
+	/* powered on hours, updated at driver unload */
+	__u8 if_powerhour[2];
+	/* ASCII free-form comment field */
+	char if_comment[32];
+	/* Backwards compatible prefix for longer QLogic Serial Numbers */
+	char if_sprefix[4];
+	/* 82 bytes used, min flash size is 128 bytes */
+	__u8 if_future[46];
+};
+
+/*
+ * These are the counters implemented in the chip, and are listed in order.
+ * The InterCaps naming is taken straight from the chip spec.
+ */
+struct qlogic_ib_counters {
+	__u64 LBIntCnt;
+	__u64 LBFlowStallCnt;
+	__u64 TxSDmaDescCnt;    /* was Reserved1 */
+	__u64 TxUnsupVLErrCnt;
+	__u64 TxDataPktCnt;
+	__u64 TxFlowPktCnt;
+	__u64 TxDwordCnt;
+	__u64 TxLenErrCnt;
+	__u64 TxMaxMinLenErrCnt;
+	__u64 TxUnderrunCnt;
+	__u64 TxFlowStallCnt;
+	__u64 TxDroppedPktCnt;
+	__u64 RxDroppedPktCnt;
+	__u64 RxDataPktCnt;
+	__u64 RxFlowPktCnt;
+	__u64 RxDwordCnt;
+	__u64 RxLenErrCnt;
+	__u64 RxMaxMinLenErrCnt;
+	__u64 RxICRCErrCnt;
+	__u64 RxVCRCErrCnt;
+	__u64 RxFlowCtrlErrCnt;
+	__u64 RxBadFormatCnt;
+	__u64 RxLinkProblemCnt;
+	__u64 RxEBPCnt;
+	__u64 RxLPCRCErrCnt;
+	__u64 RxBufOvflCnt;
+	__u64 RxTIDFullErrCnt;
+	__u64 RxTIDValidErrCnt;
+	__u64 RxPKeyMismatchCnt;
+	__u64 RxP0HdrEgrOvflCnt;
+	__u64 RxP1HdrEgrOvflCnt;
+	__u64 RxP2HdrEgrOvflCnt;
+	__u64 RxP3HdrEgrOvflCnt;
+	__u64 RxP4HdrEgrOvflCnt;
+	__u64 RxP5HdrEgrOvflCnt;
+	__u64 RxP6HdrEgrOvflCnt;
+	__u64 RxP7HdrEgrOvflCnt;
+	__u64 RxP8HdrEgrOvflCnt;
+	__u64 RxP9HdrEgrOvflCnt;
+	__u64 RxP10HdrEgrOvflCnt;
+	__u64 RxP11HdrEgrOvflCnt;
+	__u64 RxP12HdrEgrOvflCnt;
+	__u64 RxP13HdrEgrOvflCnt;
+	__u64 RxP14HdrEgrOvflCnt;
+	__u64 RxP15HdrEgrOvflCnt;
+	__u64 RxP16HdrEgrOvflCnt;
+	__u64 IBStatusChangeCnt;
+	__u64 IBLinkErrRecoveryCnt;
+	__u64 IBLinkDownedCnt;
+	__u64 IBSymbolErrCnt;
+	__u64 RxVL15DroppedPktCnt;
+	__u64 RxOtherLocalPhyErrCnt;
+	__u64 PcieRetryBufDiagQwordCnt;
+	__u64 ExcessBufferOvflCnt;
+	__u64 LocalLinkIntegrityErrCnt;
+	__u64 RxVlErrCnt;
+	__u64 RxDlidFltrCnt;
+};
+
+/*
+ * The next set of defines are for packet headers, and chip register
+ * and memory bits that are visible to and/or used by user-mode software.
+ */
+
+/* RcvHdrFlags bits */
+#define QLOGIC_IB_RHF_LENGTH_MASK 0x7FF
+#define QLOGIC_IB_RHF_LENGTH_SHIFT 0
+#define QLOGIC_IB_RHF_RCVTYPE_MASK 0x7
+#define QLOGIC_IB_RHF_RCVTYPE_SHIFT 11
+#define QLOGIC_IB_RHF_EGRINDEX_MASK 0xFFF
+#define QLOGIC_IB_RHF_EGRINDEX_SHIFT 16
+#define QLOGIC_IB_RHF_SEQ_MASK 0xF
+#define QLOGIC_IB_RHF_SEQ_SHIFT 0
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_MASK 0x7FF
+#define QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT 4
+#define QLOGIC_IB_RHF_H_ICRCERR   0x80000000
+#define QLOGIC_IB_RHF_H_VCRCERR   0x40000000
+#define QLOGIC_IB_RHF_H_PARITYERR 0x20000000
+#define QLOGIC_IB_RHF_H_LENERR    0x10000000
+#define QLOGIC_IB_RHF_H_MTUERR    0x08000000
+#define QLOGIC_IB_RHF_H_IHDRERR   0x04000000
+#define QLOGIC_IB_RHF_H_TIDERR    0x02000000
+#define QLOGIC_IB_RHF_H_MKERR     0x01000000
+#define QLOGIC_IB_RHF_H_IBERR     0x00800000
+#define QLOGIC_IB_RHF_H_ERR_MASK  0xFF800000
+#define QLOGIC_IB_RHF_L_USE_EGR   0x80000000
+#define QLOGIC_IB_RHF_L_SWA       0x00008000
+#define QLOGIC_IB_RHF_L_SWB       0x00004000
+
+/* qlogic_ib header fields */
+#define QLOGIC_IB_I_VERS_MASK 0xF
+#define QLOGIC_IB_I_VERS_SHIFT 28
+#define QLOGIC_IB_I_CTXT_MASK 0xF
+#define QLOGIC_IB_I_CTXT_SHIFT 24
+#define QLOGIC_IB_I_TID_MASK 0x7FF
+#define QLOGIC_IB_I_TID_SHIFT 13
+#define QLOGIC_IB_I_OFFSET_MASK 0x1FFF
+#define QLOGIC_IB_I_OFFSET_SHIFT 0
+
+/* K_PktFlags bits */
+#define QLOGIC_IB_KPF_INTR 0x1
+#define QLOGIC_IB_KPF_SUBCTXT_MASK 0x3
+#define QLOGIC_IB_KPF_SUBCTXT_SHIFT 1
+
+#define QLOGIC_IB_MAX_SUBCTXT   4
+
+/* SendPIO per-buffer control */
+#define QLOGIC_IB_SP_TEST    0x40
+#define QLOGIC_IB_SP_TESTEBP 0x20
+#define QLOGIC_IB_SP_TRIGGER_SHIFT  15
+
+/* SendPIOAvail bits */
+#define QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT 1
+#define QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT 0
+
+/* qlogic_ib header format */
+struct qib_header {
+	/*
+	 * Version - 4 bits, Context - 4 bits, TID - 10 bits and Offset -
+	 * 14 bits before ECO change ~28 Dec 03.  After that, Vers 4,
+	 * Context 4, TID 11, offset 13.
+	 */
+	__le32 ver_ctxt_tid_offset;
+	__le16 chksum;
+	__le16 pkt_flags;
+};
+
+/*
+ * qlogic_ib user message header format.
+ * This structure contains the first 4 fields common to all protocols
+ * that employ qlogic_ib.
+ */
+struct qib_message_header {
+	__be16 lrh[4];
+	__be32 bth[3];
+	/* fields below this point are in host byte order */
+	struct qib_header iph;
+	/* fields below are simplified, but should match PSM */
+	/* some are accessed by driver when packet spliting is needed */
+	__u8 sub_opcode;
+	__u8 flags;
+	__u16 commidx;
+	__u32 ack_seq_num;
+	__u8 flowid;
+	__u8 hdr_dlen;
+	__u16 mqhdr;
+	__u32 uwords[4];
+};
+
+/* sequence number bits for message */
+union qib_seqnum {
+	struct {
+		__u32 seq:11;
+		__u32 gen:8;
+		__u32 flow:5;
+	};
+	struct {
+		__u32 pkt:16;
+		__u32 msg:8;
+	};
+	__u32 val;
+};
+
+/* qib receiving-dma tid-session-member */
+struct qib_tid_session_member {
+	__u16 tid;
+	__u16 offset;
+	__u16 length;
+};
+
+/* IB - LRH header consts */
+#define QIB_LRH_GRH 0x0003      /* 1. word of IB LRH - next header: GRH */
+#define QIB_LRH_BTH 0x0002      /* 1. word of IB LRH - next header: BTH */
+
+/* misc. */
+#define SIZE_OF_CRC 1
+
+#define QIB_DEFAULT_P_KEY 0xFFFF
+#define QIB_PERMISSIVE_LID 0xFFFF
+#define QIB_AETH_CREDIT_SHIFT 24
+#define QIB_AETH_CREDIT_MASK 0x1F
+#define QIB_AETH_CREDIT_INVAL 0x1F
+#define QIB_PSN_MASK 0xFFFFFF
+#define QIB_MSN_MASK 0xFFFFFF
+#define QIB_QPN_MASK 0xFFFFFF
+#define QIB_MULTICAST_LID_BASE 0xC000
+#define QIB_EAGER_TID_ID QLOGIC_IB_I_TID_MASK
+#define QIB_MULTICAST_QPN 0xFFFFFF
+
+/* Receive Header Queue: receive type (from qlogic_ib) */
+#define RCVHQ_RCV_TYPE_EXPECTED  0
+#define RCVHQ_RCV_TYPE_EAGER     1
+#define RCVHQ_RCV_TYPE_NON_KD    2
+#define RCVHQ_RCV_TYPE_ERROR     3
+
+#define QIB_HEADER_QUEUE_WORDS 9
+
+/* functions for extracting fields from rcvhdrq entries for the driver.
+ */
+static inline __u32 qib_hdrget_err_flags(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[1]) & QLOGIC_IB_RHF_H_ERR_MASK;
+}
+
+static inline __u32 qib_hdrget_rcv_type(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_RCVTYPE_SHIFT) &
+		QLOGIC_IB_RHF_RCVTYPE_MASK;
+}
+
+static inline __u32 qib_hdrget_length_in_bytes(const __le32 *rbuf)
+{
+	return ((__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_LENGTH_SHIFT) &
+		QLOGIC_IB_RHF_LENGTH_MASK) << 2;
+}
+
+static inline __u32 qib_hdrget_index(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[0]) >> QLOGIC_IB_RHF_EGRINDEX_SHIFT) &
+		QLOGIC_IB_RHF_EGRINDEX_MASK;
+}
+
+static inline __u32 qib_hdrget_seq(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_SEQ_SHIFT) &
+		QLOGIC_IB_RHF_SEQ_MASK;
+}
+
+static inline __u32 qib_hdrget_offset(const __le32 *rbuf)
+{
+	return (__le32_to_cpu(rbuf[1]) >> QLOGIC_IB_RHF_HDRQ_OFFSET_SHIFT) &
+		QLOGIC_IB_RHF_HDRQ_OFFSET_MASK;
+}
+
+static inline __u32 qib_hdrget_use_egr_buf(const __le32 *rbuf)
+{
+	return __le32_to_cpu(rbuf[0]) & QLOGIC_IB_RHF_L_USE_EGR;
+}
+
+static inline __u32 qib_hdrget_qib_ver(__le32 hdrword)
+{
+	return (__le32_to_cpu(hdrword) >> QLOGIC_IB_I_VERS_SHIFT) &
+		QLOGIC_IB_I_VERS_MASK;
+}
+
+#endif                          /* _QIB_COMMON_H */
diff --git a/drivers/infiniband/hw/qib/qib_cq.c b/drivers/infiniband/hw/qib/qib_cq.c
new file mode 100644
index 00000000000..ab4e11cfab1
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_cq.c
@@ -0,0 +1,540 @@
+/*
+ * Copyright (c) 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006, 2007, 2008, 2010 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/kthread.h>
+
+#include "qib_verbs.h"
+#include "qib.h"
+
+/**
+ * qib_cq_enter - add a new entry to the completion queue
+ * @cq: completion queue
+ * @entry: work completion entry to add
+ * @sig: true if @entry is a solicitated entry
+ *
+ * This may be called with qp->s_lock held.
+ */
+void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int solicited)
+{
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	u32 head;
+	u32 next;
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	/*
+	 * Note that the head pointer might be writable by user processes.
+	 * Take care to verify it is a sane value.
+	 */
+	wc = cq->queue;
+	head = wc->head;
+	if (head >= (unsigned) cq->ibcq.cqe) {
+		head = cq->ibcq.cqe;
+		next = 0;
+	} else
+		next = head + 1;
+	if (unlikely(next == wc->tail)) {
+		spin_unlock_irqrestore(&cq->lock, flags);
+		if (cq->ibcq.event_handler) {
+			struct ib_event ev;
+
+			ev.device = cq->ibcq.device;
+			ev.element.cq = &cq->ibcq;
+			ev.event = IB_EVENT_CQ_ERR;
+			cq->ibcq.event_handler(&ev, cq->ibcq.cq_context);
+		}
+		return;
+	}
+	if (cq->ip) {
+		wc->uqueue[head].wr_id = entry->wr_id;
+		wc->uqueue[head].status = entry->status;
+		wc->uqueue[head].opcode = entry->opcode;
+		wc->uqueue[head].vendor_err = entry->vendor_err;
+		wc->uqueue[head].byte_len = entry->byte_len;
+		wc->uqueue[head].ex.imm_data =
+			(__u32 __force)entry->ex.imm_data;
+		wc->uqueue[head].qp_num = entry->qp->qp_num;
+		wc->uqueue[head].src_qp = entry->src_qp;
+		wc->uqueue[head].wc_flags = entry->wc_flags;
+		wc->uqueue[head].pkey_index = entry->pkey_index;
+		wc->uqueue[head].slid = entry->slid;
+		wc->uqueue[head].sl = entry->sl;
+		wc->uqueue[head].dlid_path_bits = entry->dlid_path_bits;
+		wc->uqueue[head].port_num = entry->port_num;
+		/* Make sure entry is written before the head index. */
+		smp_wmb();
+	} else
+		wc->kqueue[head] = *entry;
+	wc->head = next;
+
+	if (cq->notify == IB_CQ_NEXT_COMP ||
+	    (cq->notify == IB_CQ_SOLICITED &&
+	     (solicited || entry->status != IB_WC_SUCCESS))) {
+		struct kthread_worker *worker;
+		/*
+		 * This will cause send_complete() to be called in
+		 * another thread.
+		 */
+		smp_rmb();
+		worker = cq->dd->worker;
+		if (likely(worker)) {
+			cq->notify = IB_CQ_NONE;
+			cq->triggered++;
+			queue_kthread_work(worker, &cq->comptask);
+		}
+	}
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+}
+
+/**
+ * qib_poll_cq - poll for work completion entries
+ * @ibcq: the completion queue to poll
+ * @num_entries: the maximum number of entries to return
+ * @entry: pointer to array where work completions are placed
+ *
+ * Returns the number of completion entries polled.
+ *
+ * This may be called from interrupt context.  Also called by ib_poll_cq()
+ * in the generic verbs code.
+ */
+int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *wc;
+	unsigned long flags;
+	int npolled;
+	u32 tail;
+
+	/* The kernel can only poll a kernel completion queue */
+	if (cq->ip) {
+		npolled = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&cq->lock, flags);
+
+	wc = cq->queue;
+	tail = wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	for (npolled = 0; npolled < num_entries; ++npolled, ++entry) {
+		if (tail == wc->head)
+			break;
+		/* The kernel doesn't need a RMB since it has the lock. */
+		*entry = wc->kqueue[tail];
+		if (tail >= cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	wc->tail = tail;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+bail:
+	return npolled;
+}
+
+static void send_complete(struct kthread_work *work)
+{
+	struct qib_cq *cq = container_of(work, struct qib_cq, comptask);
+
+	/*
+	 * The completion handler will most likely rearm the notification
+	 * and poll for all pending entries.  If a new completion entry
+	 * is added while we are in this routine, queue_work()
+	 * won't call us again until we return so we check triggered to
+	 * see if we need to call the handler again.
+	 */
+	for (;;) {
+		u8 triggered = cq->triggered;
+
+		/*
+		 * IPoIB connected mode assumes the callback is from a
+		 * soft IRQ. We simulate this by blocking "bottom halves".
+		 * See the implementation for ipoib_cm_handle_tx_wc(),
+		 * netif_tx_lock_bh() and netif_tx_lock().
+		 */
+		local_bh_disable();
+		cq->ibcq.comp_handler(&cq->ibcq, cq->ibcq.cq_context);
+		local_bh_enable();
+
+		if (cq->triggered == triggered)
+			return;
+	}
+}
+
+/**
+ * qib_create_cq - create a completion queue
+ * @ibdev: the device this completion queue is attached to
+ * @entries: the minimum size of the completion queue
+ * @context: unused by the QLogic_IB driver
+ * @udata: user data for libibverbs.so
+ *
+ * Returns a pointer to the completion queue or negative errno values
+ * for failure.
+ *
+ * Called by ib_create_cq() in the generic verbs code.
+ */
+struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
+			    int comp_vector, struct ib_ucontext *context,
+			    struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_cq *cq;
+	struct qib_cq_wc *wc;
+	struct ib_cq *ret;
+	u32 sz;
+
+	if (entries < 1 || entries > ib_qib_max_cqes) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	/* Allocate the completion queue structure. */
+	cq = kmalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Allocate the completion queue entries and head/tail pointers.
+	 * This is allocated separately so that it can be resized and
+	 * also mapped into user space.
+	 * We need to use vmalloc() in order to support mmap and large
+	 * numbers of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (entries + 1);
+	else
+		sz += sizeof(struct ib_wc) * (entries + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_cq;
+	}
+
+	/*
+	 * Return the address of the WC as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+
+		cq->ip = qib_create_mmap_info(dev, sz, context, wc);
+		if (!cq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wc;
+		}
+
+		err = ib_copy_to_udata(udata, &cq->ip->offset,
+				       sizeof(cq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		cq->ip = NULL;
+
+	spin_lock(&dev->n_cqs_lock);
+	if (dev->n_cqs_allocated == ib_qib_max_cqs) {
+		spin_unlock(&dev->n_cqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_cqs_allocated++;
+	spin_unlock(&dev->n_cqs_lock);
+
+	if (cq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&cq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	/*
+	 * ib_create_cq() will initialize cq->ibcq except for cq->ibcq.cqe.
+	 * The number of entries should be >= the number requested or return
+	 * an error.
+	 */
+	cq->dd = dd_from_dev(dev);
+	cq->ibcq.cqe = entries;
+	cq->notify = IB_CQ_NONE;
+	cq->triggered = 0;
+	spin_lock_init(&cq->lock);
+	init_kthread_work(&cq->comptask, send_complete);
+	wc->head = 0;
+	wc->tail = 0;
+	cq->queue = wc;
+
+	ret = &cq->ibcq;
+
+	goto done;
+
+bail_ip:
+	kfree(cq->ip);
+bail_wc:
+	vfree(wc);
+bail_cq:
+	kfree(cq);
+done:
+	return ret;
+}
+
+/**
+ * qib_destroy_cq - destroy a completion queue
+ * @ibcq: the completion queue to destroy.
+ *
+ * Returns 0 for success.
+ *
+ * Called by ib_destroy_cq() in the generic verbs code.
+ */
+int qib_destroy_cq(struct ib_cq *ibcq)
+{
+	struct qib_ibdev *dev = to_idev(ibcq->device);
+	struct qib_cq *cq = to_icq(ibcq);
+
+	flush_kthread_work(&cq->comptask);
+	spin_lock(&dev->n_cqs_lock);
+	dev->n_cqs_allocated--;
+	spin_unlock(&dev->n_cqs_lock);
+	if (cq->ip)
+		kref_put(&cq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(cq->queue);
+	kfree(cq);
+
+	return 0;
+}
+
+/**
+ * qib_req_notify_cq - change the notification type for a completion queue
+ * @ibcq: the completion queue
+ * @notify_flags: the type of notification to request
+ *
+ * Returns 0 for success.
+ *
+ * This may be called from interrupt context.  Also called by
+ * ib_req_notify_cq() in the generic verbs code.
+ */
+int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&cq->lock, flags);
+	/*
+	 * Don't change IB_CQ_NEXT_COMP to IB_CQ_SOLICITED but allow
+	 * any other transitions (see C11-31 and C11-32 in ch. 11.4.2.2).
+	 */
+	if (cq->notify != IB_CQ_NEXT_COMP)
+		cq->notify = notify_flags & IB_CQ_SOLICITED_MASK;
+
+	if ((notify_flags & IB_CQ_REPORT_MISSED_EVENTS) &&
+	    cq->queue->head != cq->queue->tail)
+		ret = 1;
+
+	spin_unlock_irqrestore(&cq->lock, flags);
+
+	return ret;
+}
+
+/**
+ * qib_resize_cq - change the size of the CQ
+ * @ibcq: the completion queue
+ *
+ * Returns 0 for success.
+ */
+int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata)
+{
+	struct qib_cq *cq = to_icq(ibcq);
+	struct qib_cq_wc *old_wc;
+	struct qib_cq_wc *wc;
+	u32 head, tail, n;
+	int ret;
+	u32 sz;
+
+	if (cqe < 1 || cqe > ib_qib_max_cqes) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	sz = sizeof(*wc);
+	if (udata && udata->outlen >= sizeof(__u64))
+		sz += sizeof(struct ib_uverbs_wc) * (cqe + 1);
+	else
+		sz += sizeof(struct ib_wc) * (cqe + 1);
+	wc = vmalloc_user(sz);
+	if (!wc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	/* Check that we can write the offset to mmap. */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		__u64 offset = 0;
+
+		ret = ib_copy_to_udata(udata, &offset, sizeof(offset));
+		if (ret)
+			goto bail_free;
+	}
+
+	spin_lock_irq(&cq->lock);
+	/*
+	 * Make sure head and tail are sane since they
+	 * might be user writable.
+	 */
+	old_wc = cq->queue;
+	head = old_wc->head;
+	if (head > (u32) cq->ibcq.cqe)
+		head = (u32) cq->ibcq.cqe;
+	tail = old_wc->tail;
+	if (tail > (u32) cq->ibcq.cqe)
+		tail = (u32) cq->ibcq.cqe;
+	if (head < tail)
+		n = cq->ibcq.cqe + 1 + head - tail;
+	else
+		n = head - tail;
+	if (unlikely((u32)cqe < n)) {
+		ret = -EINVAL;
+		goto bail_unlock;
+	}
+	for (n = 0; tail != head; n++) {
+		if (cq->ip)
+			wc->uqueue[n] = old_wc->uqueue[tail];
+		else
+			wc->kqueue[n] = old_wc->kqueue[tail];
+		if (tail == (u32) cq->ibcq.cqe)
+			tail = 0;
+		else
+			tail++;
+	}
+	cq->ibcq.cqe = cqe;
+	wc->head = n;
+	wc->tail = 0;
+	cq->queue = wc;
+	spin_unlock_irq(&cq->lock);
+
+	vfree(old_wc);
+
+	if (cq->ip) {
+		struct qib_ibdev *dev = to_idev(ibcq->device);
+		struct qib_mmap_info *ip = cq->ip;
+
+		qib_update_mmap_info(dev, ip, sz, wc);
+
+		/*
+		 * Return the offset to mmap.
+		 * See qib_mmap() for details.
+		 */
+		if (udata && udata->outlen >= sizeof(__u64)) {
+			ret = ib_copy_to_udata(udata, &ip->offset,
+					       sizeof(ip->offset));
+			if (ret)
+				goto bail;
+		}
+
+		spin_lock_irq(&dev->pending_lock);
+		if (list_empty(&ip->pending_mmaps))
+			list_add(&ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = 0;
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&cq->lock);
+bail_free:
+	vfree(wc);
+bail:
+	return ret;
+}
+
+int qib_cq_init(struct qib_devdata *dd)
+{
+	int ret = 0;
+	int cpu;
+	struct task_struct *task;
+
+	if (dd->worker)
+		return 0;
+	dd->worker = kzalloc(sizeof(*dd->worker), GFP_KERNEL);
+	if (!dd->worker)
+		return -ENOMEM;
+	init_kthread_worker(dd->worker);
+	task = kthread_create_on_node(
+		kthread_worker_fn,
+		dd->worker,
+		dd->assigned_node_id,
+		"qib_cq%d", dd->unit);
+	if (IS_ERR(task))
+		goto task_fail;
+	cpu = cpumask_first(cpumask_of_node(dd->assigned_node_id));
+	kthread_bind(task, cpu);
+	wake_up_process(task);
+out:
+	return ret;
+task_fail:
+	ret = PTR_ERR(task);
+	kfree(dd->worker);
+	dd->worker = NULL;
+	goto out;
+}
+
+void qib_cq_exit(struct qib_devdata *dd)
+{
+	struct kthread_worker *worker;
+
+	worker = dd->worker;
+	if (!worker)
+		return;
+	/* blocks future queuing from send_complete() */
+	dd->worker = NULL;
+	smp_wmb();
+	flush_kthread_worker(worker);
+	kthread_stop(worker->task);
+	kfree(worker);
+}
diff --git a/drivers/infiniband/hw/qib/qib_debugfs.c b/drivers/infiniband/hw/qib/qib_debugfs.c
new file mode 100644
index 00000000000..799a0c3bffc
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_debugfs.c
@@ -0,0 +1,283 @@
+#ifdef CONFIG_DEBUG_FS
+/*
+ * Copyright (c) 2013 Intel Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/debugfs.h>
+#include <linux/seq_file.h>
+#include <linux/kernel.h>
+#include <linux/export.h>
+
+#include "qib.h"
+#include "qib_verbs.h"
+#include "qib_debugfs.h"
+
+static struct dentry *qib_dbg_root;
+
+#define DEBUGFS_FILE(name) \
+static const struct seq_operations _##name##_seq_ops = { \
+	.start = _##name##_seq_start, \
+	.next  = _##name##_seq_next, \
+	.stop  = _##name##_seq_stop, \
+	.show  = _##name##_seq_show \
+}; \
+static int _##name##_open(struct inode *inode, struct file *s) \
+{ \
+	struct seq_file *seq; \
+	int ret; \
+	ret =  seq_open(s, &_##name##_seq_ops); \
+	if (ret) \
+		return ret; \
+	seq = s->private_data; \
+	seq->private = inode->i_private; \
+	return 0; \
+} \
+static const struct file_operations _##name##_file_ops = { \
+	.owner   = THIS_MODULE, \
+	.open    = _##name##_open, \
+	.read    = seq_read, \
+	.llseek  = seq_lseek, \
+	.release = seq_release \
+};
+
+#define DEBUGFS_FILE_CREATE(name) \
+do { \
+	struct dentry *ent; \
+	ent = debugfs_create_file(#name , 0400, ibd->qib_ibdev_dbg, \
+		ibd, &_##name##_file_ops); \
+	if (!ent) \
+		pr_warn("create of " #name " failed\n"); \
+} while (0)
+
+static void *_opcode_stats_seq_start(struct seq_file *s, loff_t *pos)
+{
+	struct qib_opcode_stats_perctx *opstats;
+
+	if (*pos >= ARRAY_SIZE(opstats->stats))
+		return NULL;
+	return pos;
+}
+
+static void *_opcode_stats_seq_next(struct seq_file *s, void *v, loff_t *pos)
+{
+	struct qib_opcode_stats_perctx *opstats;
+
+	++*pos;
+	if (*pos >= ARRAY_SIZE(opstats->stats))
+		return NULL;
+	return pos;
+}
+
+
+static void _opcode_stats_seq_stop(struct seq_file *s, void *v)
+{
+	/* nothing allocated */
+}
+
+static int _opcode_stats_seq_show(struct seq_file *s, void *v)
+{
+	loff_t *spos = v;
+	loff_t i = *spos, j;
+	u64 n_packets = 0, n_bytes = 0;
+	struct qib_ibdev *ibd = (struct qib_ibdev *)s->private;
+	struct qib_devdata *dd = dd_from_dev(ibd);
+
+	for (j = 0; j < dd->first_user_ctxt; j++) {
+		if (!dd->rcd[j])
+			continue;
+		n_packets += dd->rcd[j]->opstats->stats[i].n_packets;
+		n_bytes += dd->rcd[j]->opstats->stats[i].n_bytes;
+	}
+	if (!n_packets && !n_bytes)
+		return SEQ_SKIP;
+	seq_printf(s, "%02llx %llu/%llu\n", i,
+		(unsigned long long) n_packets,
+		(unsigned long long) n_bytes);
+
+	return 0;
+}
+
+DEBUGFS_FILE(opcode_stats)
+
+static void *_ctx_stats_seq_start(struct seq_file *s, loff_t *pos)
+{
+	struct qib_ibdev *ibd = (struct qib_ibdev *)s->private;
+	struct qib_devdata *dd = dd_from_dev(ibd);
+
+	if (!*pos)
+		return SEQ_START_TOKEN;
+	if (*pos >= dd->first_user_ctxt)
+		return NULL;
+	return pos;
+}
+
+static void *_ctx_stats_seq_next(struct seq_file *s, void *v, loff_t *pos)
+{
+	struct qib_ibdev *ibd = (struct qib_ibdev *)s->private;
+	struct qib_devdata *dd = dd_from_dev(ibd);
+
+	if (v == SEQ_START_TOKEN)
+		return pos;
+
+	++*pos;
+	if (*pos >= dd->first_user_ctxt)
+		return NULL;
+	return pos;
+}
+
+static void _ctx_stats_seq_stop(struct seq_file *s, void *v)
+{
+	/* nothing allocated */
+}
+
+static int _ctx_stats_seq_show(struct seq_file *s, void *v)
+{
+	loff_t *spos;
+	loff_t i, j;
+	u64 n_packets = 0;
+	struct qib_ibdev *ibd = (struct qib_ibdev *)s->private;
+	struct qib_devdata *dd = dd_from_dev(ibd);
+
+	if (v == SEQ_START_TOKEN) {
+		seq_puts(s, "Ctx:npkts\n");
+		return 0;
+	}
+
+	spos = v;
+	i = *spos;
+
+	if (!dd->rcd[i])
+		return SEQ_SKIP;
+
+	for (j = 0; j < ARRAY_SIZE(dd->rcd[i]->opstats->stats); j++)
+		n_packets += dd->rcd[i]->opstats->stats[j].n_packets;
+
+	if (!n_packets)
+		return SEQ_SKIP;
+
+	seq_printf(s, "  %llu:%llu\n", i, n_packets);
+	return 0;
+}
+
+DEBUGFS_FILE(ctx_stats)
+
+static void *_qp_stats_seq_start(struct seq_file *s, loff_t *pos)
+{
+	struct qib_qp_iter *iter;
+	loff_t n = *pos;
+
+	iter = qib_qp_iter_init(s->private);
+	if (!iter)
+		return NULL;
+
+	while (n--) {
+		if (qib_qp_iter_next(iter)) {
+			kfree(iter);
+			return NULL;
+		}
+	}
+
+	return iter;
+}
+
+static void *_qp_stats_seq_next(struct seq_file *s, void *iter_ptr,
+				   loff_t *pos)
+{
+	struct qib_qp_iter *iter = iter_ptr;
+
+	(*pos)++;
+
+	if (qib_qp_iter_next(iter)) {
+		kfree(iter);
+		return NULL;
+	}
+
+	return iter;
+}
+
+static void _qp_stats_seq_stop(struct seq_file *s, void *iter_ptr)
+{
+	/* nothing for now */
+}
+
+static int _qp_stats_seq_show(struct seq_file *s, void *iter_ptr)
+{
+	struct qib_qp_iter *iter = iter_ptr;
+
+	if (!iter)
+		return 0;
+
+	qib_qp_iter_print(s, iter);
+
+	return 0;
+}
+
+DEBUGFS_FILE(qp_stats)
+
+void qib_dbg_ibdev_init(struct qib_ibdev *ibd)
+{
+	char name[10];
+
+	snprintf(name, sizeof(name), "qib%d", dd_from_dev(ibd)->unit);
+	ibd->qib_ibdev_dbg = debugfs_create_dir(name, qib_dbg_root);
+	if (!ibd->qib_ibdev_dbg) {
+		pr_warn("create of %s failed\n", name);
+		return;
+	}
+	DEBUGFS_FILE_CREATE(opcode_stats);
+	DEBUGFS_FILE_CREATE(ctx_stats);
+	DEBUGFS_FILE_CREATE(qp_stats);
+	return;
+}
+
+void qib_dbg_ibdev_exit(struct qib_ibdev *ibd)
+{
+	if (!qib_dbg_root)
+		goto out;
+	debugfs_remove_recursive(ibd->qib_ibdev_dbg);
+out:
+	ibd->qib_ibdev_dbg = NULL;
+}
+
+void qib_dbg_init(void)
+{
+	qib_dbg_root = debugfs_create_dir(QIB_DRV_NAME, NULL);
+	if (!qib_dbg_root)
+		pr_warn("init of debugfs failed\n");
+}
+
+void qib_dbg_exit(void)
+{
+	debugfs_remove_recursive(qib_dbg_root);
+	qib_dbg_root = NULL;
+}
+
+#endif
+
diff --git a/drivers/infiniband/hw/qib/qib_debugfs.h b/drivers/infiniband/hw/qib/qib_debugfs.h
new file mode 100644
index 00000000000..7ae983a91b8
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_debugfs.h
@@ -0,0 +1,45 @@
+#ifndef _QIB_DEBUGFS_H
+#define _QIB_DEBUGFS_H
+
+#ifdef CONFIG_DEBUG_FS
+/*
+ * Copyright (c) 2013 Intel Corporation.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+struct qib_ibdev;
+void qib_dbg_ibdev_init(struct qib_ibdev *ibd);
+void qib_dbg_ibdev_exit(struct qib_ibdev *ibd);
+void qib_dbg_init(void);
+void qib_dbg_exit(void);
+
+#endif
+
+#endif                          /* _QIB_DEBUGFS_H */
diff --git a/drivers/infiniband/hw/qib/qib_diag.c b/drivers/infiniband/hw/qib/qib_diag.c
new file mode 100644
index 00000000000..5dfda4c5cc9
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_diag.c
@@ -0,0 +1,911 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file contains support for diagnostic functions.  It is accessed by
+ * opening the qib_diag device, normally minor number 129.  Diagnostic use
+ * of the QLogic_IB chip may render the chip or board unusable until the
+ * driver is unloaded, or in some cases, until the system is rebooted.
+ *
+ * Accesses to the chip through this interface are not similar to going
+ * through the /sys/bus/pci resource mmap interface.
+ */
+
+#include <linux/io.h>
+#include <linux/pci.h>
+#include <linux/poll.h>
+#include <linux/vmalloc.h>
+#include <linux/export.h>
+#include <linux/fs.h>
+#include <linux/uaccess.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+#undef pr_fmt
+#define pr_fmt(fmt) QIB_DRV_NAME ": " fmt
+
+/*
+ * Each client that opens the diag device must read then write
+ * offset 0, to prevent lossage from random cat or od. diag_state
+ * sequences this "handshake".
+ */
+enum diag_state { UNUSED = 0, OPENED, INIT, READY };
+
+/* State for an individual client. PID so children cannot abuse handshake */
+static struct qib_diag_client {
+	struct qib_diag_client *next;
+	struct qib_devdata *dd;
+	pid_t pid;
+	enum diag_state state;
+} *client_pool;
+
+/*
+ * Get a client struct. Recycled if possible, else kmalloc.
+ * Must be called with qib_mutex held
+ */
+static struct qib_diag_client *get_client(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	dc = client_pool;
+	if (dc)
+		/* got from pool remove it and use */
+		client_pool = dc->next;
+	else
+		/* None in pool, alloc and init */
+		dc = kmalloc(sizeof *dc, GFP_KERNEL);
+
+	if (dc) {
+		dc->next = NULL;
+		dc->dd = dd;
+		dc->pid = current->pid;
+		dc->state = OPENED;
+	}
+	return dc;
+}
+
+/*
+ * Return to pool. Must be called with qib_mutex held
+ */
+static void return_client(struct qib_diag_client *dc)
+{
+	struct qib_devdata *dd = dc->dd;
+	struct qib_diag_client *tdc, *rdc;
+
+	rdc = NULL;
+	if (dc == dd->diag_client) {
+		dd->diag_client = dc->next;
+		rdc = dc;
+	} else {
+		tdc = dc->dd->diag_client;
+		while (tdc) {
+			if (dc == tdc->next) {
+				tdc->next = dc->next;
+				rdc = dc;
+				break;
+			}
+			tdc = tdc->next;
+		}
+	}
+	if (rdc) {
+		rdc->state = UNUSED;
+		rdc->dd = NULL;
+		rdc->pid = 0;
+		rdc->next = client_pool;
+		client_pool = rdc;
+	}
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp);
+static int qib_diag_release(struct inode *in, struct file *fp);
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off);
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off);
+
+static const struct file_operations diag_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diag_write,
+	.read = qib_diag_read,
+	.open = qib_diag_open,
+	.release = qib_diag_release,
+	.llseek = default_llseek,
+};
+
+static atomic_t diagpkt_count = ATOMIC_INIT(0);
+static struct cdev *diagpkt_cdev;
+static struct device *diagpkt_device;
+
+static ssize_t qib_diagpkt_write(struct file *fp, const char __user *data,
+				 size_t count, loff_t *off);
+
+static const struct file_operations diagpkt_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_diagpkt_write,
+	.llseek = noop_llseek,
+};
+
+int qib_diag_add(struct qib_devdata *dd)
+{
+	char name[16];
+	int ret = 0;
+
+	if (atomic_inc_return(&diagpkt_count) == 1) {
+		ret = qib_cdev_init(QIB_DIAGPKT_MINOR, "ipath_diagpkt",
+				    &diagpkt_file_ops, &diagpkt_cdev,
+				    &diagpkt_device);
+		if (ret)
+			goto done;
+	}
+
+	snprintf(name, sizeof(name), "ipath_diag%d", dd->unit);
+	ret = qib_cdev_init(QIB_DIAG_MINOR_BASE + dd->unit, name,
+			    &diag_file_ops, &dd->diag_cdev,
+			    &dd->diag_device);
+done:
+	return ret;
+}
+
+static void qib_unregister_observers(struct qib_devdata *dd);
+
+void qib_diag_remove(struct qib_devdata *dd)
+{
+	struct qib_diag_client *dc;
+
+	if (atomic_dec_and_test(&diagpkt_count))
+		qib_cdev_cleanup(&diagpkt_cdev, &diagpkt_device);
+
+	qib_cdev_cleanup(&dd->diag_cdev, &dd->diag_device);
+
+	/*
+	 * Return all diag_clients of this device. There should be none,
+	 * as we are "guaranteed" that no clients are still open
+	 */
+	while (dd->diag_client)
+		return_client(dd->diag_client);
+
+	/* Now clean up all unused client structs */
+	while (client_pool) {
+		dc = client_pool;
+		client_pool = dc->next;
+		kfree(dc);
+	}
+	/* Clean up observer list */
+	qib_unregister_observers(dd);
+}
+
+/* qib_remap_ioaddr32 - remap an offset into chip address space to __iomem *
+ *
+ * @dd: the qlogic_ib device
+ * @offs: the offset in chip-space
+ * @cntp: Pointer to max (byte) count for transfer starting at offset
+ * This returns a u32 __iomem * so it can be used for both 64 and 32-bit
+ * mapping. It is needed because with the use of PAT for control of
+ * write-combining, the logically contiguous address-space of the chip
+ * may be split into virtually non-contiguous spaces, with different
+ * attributes, which are them mapped to contiguous physical space
+ * based from the first BAR.
+ *
+ * The code below makes the same assumptions as were made in
+ * init_chip_wc_pat() (qib_init.c), copied here:
+ * Assumes chip address space looks like:
+ *		- kregs + sregs + cregs + uregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *	or:
+ *		- kregs + sregs + cregs (in any order)
+ *		- piobufs (2K and 4K bufs in either order)
+ *		- uregs
+ *
+ * If cntp is non-NULL, returns how many bytes from offset can be accessed
+ * Returns 0 if the offset is not mapped.
+ */
+static u32 __iomem *qib_remap_ioaddr32(struct qib_devdata *dd, u32 offset,
+				       u32 *cntp)
+{
+	u32 kreglen;
+	u32 snd_bottom, snd_lim = 0;
+	u32 __iomem *krb32 = (u32 __iomem *)dd->kregbase;
+	u32 __iomem *map = NULL;
+	u32 cnt = 0;
+	u32 tot4k, offs4k;
+
+	/* First, simplest case, offset is within the first map. */
+	kreglen = (dd->kregend - dd->kregbase) * sizeof(u64);
+	if (offset < kreglen) {
+		map = krb32 + (offset / sizeof(u32));
+		cnt = kreglen - offset;
+		goto mapped;
+	}
+
+	/*
+	 * Next check for user regs, the next most common case,
+	 * and a cheap check because if they are not in the first map
+	 * they are last in chip.
+	 */
+	if (dd->userbase) {
+		/* If user regs mapped, they are after send, so set limit. */
+		u32 ulim = (dd->cfgctxts * dd->ureg_align) + dd->uregbase;
+		if (!dd->piovl15base)
+			snd_lim = dd->uregbase;
+		krb32 = (u32 __iomem *)dd->userbase;
+		if (offset >= dd->uregbase && offset < ulim) {
+			map = krb32 + (offset - dd->uregbase) / sizeof(u32);
+			cnt = ulim - offset;
+			goto mapped;
+		}
+	}
+
+	/*
+	 * Lastly, check for offset within Send Buffers.
+	 * This is gnarly because struct devdata is deliberately vague
+	 * about things like 7322 VL15 buffers, and we are not in
+	 * chip-specific code here, so should not make many assumptions.
+	 * The one we _do_ make is that the only chip that has more sndbufs
+	 * than we admit is the 7322, and it has userregs above that, so
+	 * we know the snd_lim.
+	 */
+	/* Assume 2K buffers are first. */
+	snd_bottom = dd->pio2k_bufbase;
+	if (snd_lim == 0) {
+		u32 tot2k = dd->piobcnt2k * ALIGN(dd->piosize2k, dd->palign);
+		snd_lim = snd_bottom + tot2k;
+	}
+	/* If 4k buffers exist, account for them by bumping
+	 * appropriate limit.
+	 */
+	tot4k = dd->piobcnt4k * dd->align4k;
+	offs4k = dd->piobufbase >> 32;
+	if (dd->piobcnt4k) {
+		if (snd_bottom > offs4k)
+			snd_bottom = offs4k;
+		else {
+			/* 4k above 2k. Bump snd_lim, if needed*/
+			if (!dd->userbase || dd->piovl15base)
+				snd_lim = offs4k + tot4k;
+		}
+	}
+	/*
+	 * Judgement call: can we ignore the space between SendBuffs and
+	 * UserRegs, where we would like to see vl15 buffs, but not more?
+	 */
+	if (offset >= snd_bottom && offset < snd_lim) {
+		offset -= snd_bottom;
+		map = (u32 __iomem *)dd->piobase + (offset / sizeof(u32));
+		cnt = snd_lim - offset;
+	}
+
+	if (!map && offs4k && dd->piovl15base) {
+		snd_lim = offs4k + tot4k + 2 * dd->align4k;
+		if (offset >= (offs4k + tot4k) && offset < snd_lim) {
+			map = (u32 __iomem *)dd->piovl15base +
+				((offset - (offs4k + tot4k)) / sizeof(u32));
+			cnt = snd_lim - offset;
+		}
+	}
+
+mapped:
+	if (cntp)
+		*cntp = cnt;
+	return map;
+}
+
+/*
+ * qib_read_umem64 - read a 64-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy (multiple of 32 bits)
+ *
+ * This function also localizes all chip memory accesses.
+ * The copy should be written such that we read full cacheline packets
+ * from the chip.  This is usually used for a single qword
+ *
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+static int qib_read_umem64(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (const u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data = readq(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(u64))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem64 - write a 64-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: the number of bytes to copy (multiple of 32 bits)
+ *
+ * This is usually used for a single qword
+ * NOTE:  This assumes the chip address is 64-bit aligned.
+ */
+
+static int qib_write_umem64(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u64 __iomem *reg_addr;
+	const u64 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = (u64 __iomem *)qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u64));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u64 data;
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writeq(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u64);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_read_umem32 - read a 32-bit quantity from the chip into user space
+ * @dd: the qlogic_ib device
+ * @uaddr: the location to store the data in user memory
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @count: number of bytes to copy
+ *
+ * read 32 bit values, not 64 bit; for memories that only
+ * support 32 bit reads; usually a single dword.
+ */
+static int qib_read_umem32(struct qib_devdata *dd, void __user *uaddr,
+			   u32 regoffs, size_t count)
+{
+	const u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	/* not very efficient, but it works for now */
+	while (reg_addr < reg_end) {
+		u32 data = readl(reg_addr);
+
+		if (copy_to_user(uaddr, &data, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+/*
+ * qib_write_umem32 - write a 32-bit quantity to the chip from user space
+ * @dd: the qlogic_ib device
+ * @regoffs: the offset from BAR0 (_NOT_ full pointer, anymore)
+ * @uaddr: the source of the data in user memory
+ * @count: number of bytes to copy
+ *
+ * write 32 bit values, not 64 bit; for memories that only
+ * support 32 bit write; usually a single dword.
+ */
+
+static int qib_write_umem32(struct qib_devdata *dd, u32 regoffs,
+			    const void __user *uaddr, size_t count)
+{
+	u32 __iomem *reg_addr;
+	const u32 __iomem *reg_end;
+	u32 limit;
+	int ret;
+
+	reg_addr = qib_remap_ioaddr32(dd, regoffs, &limit);
+	if (reg_addr == NULL || limit == 0 || !(dd->flags & QIB_PRESENT)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (count >= limit)
+		count = limit;
+	reg_end = reg_addr + (count / sizeof(u32));
+
+	while (reg_addr < reg_end) {
+		u32 data;
+
+		if (copy_from_user(&data, uaddr, sizeof(data))) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		writel(data, reg_addr);
+
+		reg_addr++;
+		uaddr += sizeof(u32);
+	}
+	ret = 0;
+bail:
+	return ret;
+}
+
+static int qib_diag_open(struct inode *in, struct file *fp)
+{
+	int unit = iminor(in) - QIB_DIAG_MINOR_BASE;
+	struct qib_devdata *dd;
+	struct qib_diag_client *dc;
+	int ret;
+
+	mutex_lock(&qib_mutex);
+
+	dd = qib_lookup(unit);
+
+	if (dd == NULL || !(dd->flags & QIB_PRESENT) ||
+	    !dd->kregbase) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	dc = get_client(dd);
+	if (!dc) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	dc->next = dd->diag_client;
+	dd->diag_client = dc;
+	fp->private_data = dc;
+	ret = 0;
+bail:
+	mutex_unlock(&qib_mutex);
+
+	return ret;
+}
+
+/**
+ * qib_diagpkt_write - write an IB packet
+ * @fp: the diag data device file pointer
+ * @data: qib_diag_pkt structure saying where to get the packet
+ * @count: size of data to write
+ * @off: unused by this code
+ */
+static ssize_t qib_diagpkt_write(struct file *fp,
+				 const char __user *data,
+				 size_t count, loff_t *off)
+{
+	u32 __iomem *piobuf;
+	u32 plen, pbufn, maxlen_reserve;
+	struct qib_diag_xpkt dp;
+	u32 *tmpbuf = NULL;
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	ssize_t ret = 0;
+
+	if (count != sizeof(dp)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (copy_from_user(&dp, data, sizeof(dp))) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	dd = qib_lookup(dp.unit);
+	if (!dd || !(dd->flags & QIB_PRESENT) || !dd->kregbase) {
+		ret = -ENODEV;
+		goto bail;
+	}
+	if (!(dd->flags & QIB_INITTED)) {
+		/* no hardware, freeze, etc. */
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	if (dp.version != _DIAG_XPKT_VERS) {
+		qib_dev_err(dd, "Invalid version %u for diagpkt_write\n",
+			    dp.version);
+		ret = -EINVAL;
+		goto bail;
+	}
+	/* send count must be an exact number of dwords */
+	if (dp.len & 3) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	if (!dp.port || dp.port > dd->num_pports) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	ppd = &dd->pport[dp.port - 1];
+
+	/*
+	 * need total length before first word written, plus 2 Dwords. One Dword
+	 * is for padding so we get the full user data when not aligned on
+	 * a word boundary. The other Dword is to make sure we have room for the
+	 * ICRC which gets tacked on later.
+	 */
+	maxlen_reserve = 2 * sizeof(u32);
+	if (dp.len > ppd->ibmaxlen - maxlen_reserve) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	plen = sizeof(u32) + dp.len;
+
+	tmpbuf = vmalloc(plen);
+	if (!tmpbuf) {
+		qib_devinfo(dd->pcidev,
+			"Unable to allocate tmp buffer, failing\n");
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmpbuf,
+			   (const void __user *) (unsigned long) dp.data,
+			   dp.len)) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	plen >>= 2;             /* in dwords */
+
+	if (dp.pbc_wd == 0)
+		dp.pbc_wd = plen;
+
+	piobuf = dd->f_getsendbuf(ppd, dp.pbc_wd, &pbufn);
+	if (!piobuf) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	/* disarm it just to be extra sure */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbufn));
+
+	/* disable header check on pbufn for this packet */
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_DIS1, NULL);
+
+	writeq(dp.pbc_wd, piobuf);
+	/*
+	 * Copy all but the trigger word, then flush, so it's written
+	 * to chip before trigger word, then write trigger word, then
+	 * flush again, so packet is sent.
+	 */
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, tmpbuf, plen - 1);
+		qib_flush_wc();
+		__raw_writel(tmpbuf[plen - 1], piobuf + plen + 1);
+	} else
+		qib_pio_copy(piobuf + 2, tmpbuf, plen);
+
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+
+	/*
+	 * Ensure buffer is written to the chip, then re-enable
+	 * header checks (if supported by chip).  The txchk
+	 * code will ensure seen by chip before returning.
+	 */
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pbufn);
+	dd->f_txchk_change(dd, pbufn, 1, TXCHK_CHG_TYPE_ENAB1, NULL);
+
+	ret = sizeof(dp);
+
+bail:
+	vfree(tmpbuf);
+	return ret;
+}
+
+static int qib_diag_release(struct inode *in, struct file *fp)
+{
+	mutex_lock(&qib_mutex);
+	return_client(fp->private_data);
+	fp->private_data = NULL;
+	mutex_unlock(&qib_mutex);
+	return 0;
+}
+
+/*
+ * Chip-specific code calls to register its interest in
+ * a specific range.
+ */
+struct diag_observer_list_elt {
+	struct diag_observer_list_elt *next;
+	const struct diag_observer *op;
+};
+
+int qib_register_observer(struct qib_devdata *dd,
+			  const struct diag_observer *op)
+{
+	struct diag_observer_list_elt *olp;
+	unsigned long flags;
+
+	if (!dd || !op)
+		return -EINVAL;
+	olp = vmalloc(sizeof *olp);
+	if (!olp) {
+		pr_err("vmalloc for observer failed\n");
+		return -ENOMEM;
+	}
+
+	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+	olp->op = op;
+	olp->next = dd->diag_observer_list;
+	dd->diag_observer_list = olp;
+	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+
+	return 0;
+}
+
+/* Remove all registered observers when device is closed */
+static void qib_unregister_observers(struct qib_devdata *dd)
+{
+	struct diag_observer_list_elt *olp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+	olp = dd->diag_observer_list;
+	while (olp) {
+		/* Pop one observer, let go of lock */
+		dd->diag_observer_list = olp->next;
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		vfree(olp);
+		/* try again. */
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		olp = dd->diag_observer_list;
+	}
+	spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+}
+
+/*
+ * Find the observer, if any, for the specified address. Initial implementation
+ * is simple stack of observers. This must be called with diag transaction
+ * lock held.
+ */
+static const struct diag_observer *diag_get_observer(struct qib_devdata *dd,
+						     u32 addr)
+{
+	struct diag_observer_list_elt *olp;
+	const struct diag_observer *op = NULL;
+
+	olp = dd->diag_observer_list;
+	while (olp) {
+		op = olp->op;
+		if (addr >= op->bottom && addr <= op->top)
+			break;
+		olp = olp->next;
+	}
+	if (!olp)
+		op = NULL;
+
+	return op;
+}
+
+static ssize_t qib_diag_read(struct file *fp, char __user *data,
+			     size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY && (*off || count != 8))
+		ret = -EINVAL;  /* prevent cat /dev/qib_diag* */
+	else {
+		unsigned long flags;
+		u64 data64 = 0;
+		int use_32;
+		const struct diag_observer *op;
+
+		use_32 = (count % 8) || (*off % 8);
+		ret = -1;
+		spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+		/*
+		 * Check for observer on this address range.
+		 * we only support a single 32 or 64-bit read
+		 * via observer, currently.
+		 */
+		op = diag_get_observer(dd, *off);
+		if (op) {
+			u32 offset = *off;
+			ret = op->hook(dd, op, offset, &data64, 0, use_32);
+		}
+		/*
+		 * We need to release lock before any copy_to_user(),
+		 * whether implicit in qib_read_umem* or explicit below.
+		 */
+		spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit rd
+				 */
+				ret = qib_read_umem32(dd, data, (u32) *off,
+						      count);
+			else
+				ret = qib_read_umem64(dd, data, (u32) *off,
+						      count);
+		} else if (ret == count) {
+			/* Below finishes case where observer existed */
+			ret = copy_to_user(data, &data64, use_32 ?
+					   sizeof(u32) : sizeof(u64));
+			if (ret)
+				ret = -EFAULT;
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == OPENED)
+			dc->state = INIT;
+	}
+bail:
+	return ret;
+}
+
+static ssize_t qib_diag_write(struct file *fp, const char __user *data,
+			      size_t count, loff_t *off)
+{
+	struct qib_diag_client *dc = fp->private_data;
+	struct qib_devdata *dd = dc->dd;
+	void __iomem *kreg_base;
+	ssize_t ret;
+
+	if (dc->pid != current->pid) {
+		ret = -EPERM;
+		goto bail;
+	}
+
+	kreg_base = dd->kregbase;
+
+	if (count == 0)
+		ret = 0;
+	else if ((count % 4) || (*off % 4))
+		/* address or length is not 32-bit aligned, hence invalid */
+		ret = -EINVAL;
+	else if (dc->state < READY &&
+		((*off || count != 8) || dc->state != INIT))
+		/* No writes except second-step of init seq */
+		ret = -EINVAL;  /* before any other write allowed */
+	else {
+		unsigned long flags;
+		const struct diag_observer *op = NULL;
+		int use_32 =  (count % 8) || (*off % 8);
+
+		/*
+		 * Check for observer on this address range.
+		 * We only support a single 32 or 64-bit write
+		 * via observer, currently. This helps, because
+		 * we would otherwise have to jump through hoops
+		 * to make "diag transaction" meaningful when we
+		 * cannot do a copy_from_user while holding the lock.
+		 */
+		if (count == 4 || count == 8) {
+			u64 data64;
+			u32 offset = *off;
+			ret = copy_from_user(&data64, data, count);
+			if (ret) {
+				ret = -EFAULT;
+				goto bail;
+			}
+			spin_lock_irqsave(&dd->qib_diag_trans_lock, flags);
+			op = diag_get_observer(dd, *off);
+			if (op)
+				ret = op->hook(dd, op, offset, &data64, ~0Ull,
+					       use_32);
+			spin_unlock_irqrestore(&dd->qib_diag_trans_lock, flags);
+		}
+
+		if (!op) {
+			if (use_32)
+				/*
+				 * Address or length is not 64-bit aligned;
+				 * do 32-bit write
+				 */
+				ret = qib_write_umem32(dd, (u32) *off, data,
+						       count);
+			else
+				ret = qib_write_umem64(dd, (u32) *off, data,
+						       count);
+		}
+	}
+
+	if (ret >= 0) {
+		*off += count;
+		ret = count;
+		if (dc->state == INIT)
+			dc->state = READY; /* all read/write OK now */
+	}
+bail:
+	return ret;
+}
diff --git a/drivers/infiniband/hw/qib/qib_dma.c b/drivers/infiniband/hw/qib/qib_dma.c
new file mode 100644
index 00000000000..59fe092b4b0
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_dma.c
@@ -0,0 +1,169 @@
+/*
+ * Copyright (c) 2006, 2009, 2010 QLogic, Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/types.h>
+#include <linux/scatterlist.h>
+
+#include "qib_verbs.h"
+
+#define BAD_DMA_ADDRESS ((u64) 0)
+
+/*
+ * The following functions implement driver specific replacements
+ * for the ib_dma_*() functions.
+ *
+ * These functions return kernel virtual addresses instead of
+ * device bus addresses since the driver uses the CPU to copy
+ * data instead of using hardware DMA.
+ */
+
+static int qib_mapping_error(struct ib_device *dev, u64 dma_addr)
+{
+	return dma_addr == BAD_DMA_ADDRESS;
+}
+
+static u64 qib_dma_map_single(struct ib_device *dev, void *cpu_addr,
+			      size_t size, enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+	return (u64) cpu_addr;
+}
+
+static void qib_dma_unmap_single(struct ib_device *dev, u64 addr, size_t size,
+				 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static u64 qib_dma_map_page(struct ib_device *dev, struct page *page,
+			    unsigned long offset, size_t size,
+			    enum dma_data_direction direction)
+{
+	u64 addr;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	if (offset + size > PAGE_SIZE) {
+		addr = BAD_DMA_ADDRESS;
+		goto done;
+	}
+
+	addr = (u64) page_address(page);
+	if (addr)
+		addr += offset;
+	/* TODO: handle highmem pages */
+
+done:
+	return addr;
+}
+
+static void qib_dma_unmap_page(struct ib_device *dev, u64 addr, size_t size,
+			       enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static int qib_map_sg(struct ib_device *dev, struct scatterlist *sgl,
+		      int nents, enum dma_data_direction direction)
+{
+	struct scatterlist *sg;
+	u64 addr;
+	int i;
+	int ret = nents;
+
+	BUG_ON(!valid_dma_direction(direction));
+
+	for_each_sg(sgl, sg, nents, i) {
+		addr = (u64) page_address(sg_page(sg));
+		/* TODO: handle highmem pages */
+		if (!addr) {
+			ret = 0;
+			break;
+		}
+		sg->dma_address = addr + sg->offset;
+#ifdef CONFIG_NEED_SG_DMA_LENGTH
+		sg->dma_length = sg->length;
+#endif
+	}
+	return ret;
+}
+
+static void qib_unmap_sg(struct ib_device *dev,
+			 struct scatterlist *sg, int nents,
+			 enum dma_data_direction direction)
+{
+	BUG_ON(!valid_dma_direction(direction));
+}
+
+static void qib_sync_single_for_cpu(struct ib_device *dev, u64 addr,
+				    size_t size, enum dma_data_direction dir)
+{
+}
+
+static void qib_sync_single_for_device(struct ib_device *dev, u64 addr,
+				       size_t size,
+				       enum dma_data_direction dir)
+{
+}
+
+static void *qib_dma_alloc_coherent(struct ib_device *dev, size_t size,
+				    u64 *dma_handle, gfp_t flag)
+{
+	struct page *p;
+	void *addr = NULL;
+
+	p = alloc_pages(flag, get_order(size));
+	if (p)
+		addr = page_address(p);
+	if (dma_handle)
+		*dma_handle = (u64) addr;
+	return addr;
+}
+
+static void qib_dma_free_coherent(struct ib_device *dev, size_t size,
+				  void *cpu_addr, u64 dma_handle)
+{
+	free_pages((unsigned long) cpu_addr, get_order(size));
+}
+
+struct ib_dma_mapping_ops qib_dma_mapping_ops = {
+	.mapping_error = qib_mapping_error,
+	.map_single = qib_dma_map_single,
+	.unmap_single = qib_dma_unmap_single,
+	.map_page = qib_dma_map_page,
+	.unmap_page = qib_dma_unmap_page,
+	.map_sg = qib_map_sg,
+	.unmap_sg = qib_unmap_sg,
+	.sync_single_for_cpu = qib_sync_single_for_cpu,
+	.sync_single_for_device = qib_sync_single_for_device,
+	.alloc_coherent = qib_dma_alloc_coherent,
+	.free_coherent = qib_dma_free_coherent
+};
diff --git a/drivers/infiniband/hw/qib/qib_driver.c b/drivers/infiniband/hw/qib/qib_driver.c
new file mode 100644
index 00000000000..5bee08f16d7
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_driver.c
@@ -0,0 +1,817 @@
+/*
+ * Copyright (c) 2013 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+#include <linux/module.h>
+#include <linux/prefetch.h>
+
+#include "qib.h"
+
+/*
+ * The size has to be longer than this string, so we can append
+ * board/chip information to it in the init code.
+ */
+const char ib_qib_version[] = QIB_DRIVER_VERSION "\n";
+
+DEFINE_SPINLOCK(qib_devs_lock);
+LIST_HEAD(qib_dev_list);
+DEFINE_MUTEX(qib_mutex);	/* general driver use */
+
+unsigned qib_ibmtu;
+module_param_named(ibmtu, qib_ibmtu, uint, S_IRUGO);
+MODULE_PARM_DESC(ibmtu, "Set max IB MTU (0=2KB, 1=256, 2=512, ... 5=4096");
+
+unsigned qib_compat_ddr_negotiate = 1;
+module_param_named(compat_ddr_negotiate, qib_compat_ddr_negotiate, uint,
+		   S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(compat_ddr_negotiate,
+		 "Attempt pre-IBTA 1.2 DDR speed negotiation");
+
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_AUTHOR("Intel <ibsupport@intel.com>");
+MODULE_DESCRIPTION("Intel IB driver");
+MODULE_VERSION(QIB_DRIVER_VERSION);
+
+/*
+ * QIB_PIO_MAXIBHDR is the max IB header size allowed for in our
+ * PIO send buffers.  This is well beyond anything currently
+ * defined in the InfiniBand spec.
+ */
+#define QIB_PIO_MAXIBHDR 128
+
+/*
+ * QIB_MAX_PKT_RCV is the max # if packets processed per receive interrupt.
+ */
+#define QIB_MAX_PKT_RECV 64
+
+struct qlogic_ib_stats qib_stats;
+
+const char *qib_get_unit_name(int unit)
+{
+	static char iname[16];
+
+	snprintf(iname, sizeof iname, "infinipath%u", unit);
+	return iname;
+}
+
+/*
+ * Return count of units with at least one port ACTIVE.
+ */
+int qib_count_active_units(void)
+{
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx, nunits_active = 0;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		if (!(dd->flags & QIB_PRESENT) || !dd->kregbase)
+			continue;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE))) {
+				nunits_active++;
+				break;
+			}
+		}
+	}
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+	return nunits_active;
+}
+
+/*
+ * Return count of all units, optionally return in arguments
+ * the number of usable (present) units, and the number of
+ * ports that are up.
+ */
+int qib_count_units(int *npresentp, int *nupp)
+{
+	int nunits = 0, npresent = 0, nup = 0;
+	struct qib_devdata *dd;
+	unsigned long flags;
+	int pidx;
+	struct qib_pportdata *ppd;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		nunits++;
+		if ((dd->flags & QIB_PRESENT) && dd->kregbase)
+			npresent++;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			if (ppd->lid && (ppd->lflags & (QIBL_LINKINIT |
+					 QIBL_LINKARMED | QIBL_LINKACTIVE)))
+				nup++;
+		}
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	if (npresentp)
+		*npresentp = npresent;
+	if (nupp)
+		*nupp = nup;
+
+	return nunits;
+}
+
+/**
+ * qib_wait_linkstate - wait for an IB link state change to occur
+ * @dd: the qlogic_ib device
+ * @state: the state to wait for
+ * @msecs: the number of milliseconds to wait
+ *
+ * wait up to msecs milliseconds for IB link state change to occur for
+ * now, take the easy polling route.  Currently used only by
+ * qib_set_linkstate.  Returns 0 if state reached, otherwise
+ * -ETIMEDOUT state can have multiple states set, for any of several
+ * transitions.
+ */
+int qib_wait_linkstate(struct qib_pportdata *ppd, u32 state, int msecs)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	if (ppd->state_wanted) {
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		ret = -EBUSY;
+		goto bail;
+	}
+	ppd->state_wanted = state;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	wait_event_interruptible_timeout(ppd->state_wait,
+					 (ppd->lflags & state),
+					 msecs_to_jiffies(msecs));
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->state_wanted = 0;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	if (!(ppd->lflags & state))
+		ret = -ETIMEDOUT;
+	else
+		ret = 0;
+bail:
+	return ret;
+}
+
+int qib_set_linkstate(struct qib_pportdata *ppd, u8 newstate)
+{
+	u32 lstate;
+	int ret;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	switch (newstate) {
+	case QIB_IB_LINKDOWN_ONLY:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_NOP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_POLL);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_SLEEP:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_SLEEP);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKDOWN_DISABLE:
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_DOWN | IB_LINKINITCMD_DISABLE);
+		/* don't wait */
+		ret = 0;
+		goto bail;
+
+	case QIB_IB_LINKARM:
+		if (ppd->lflags & QIBL_LINKARMED) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & (QIBL_LINKINIT | QIBL_LINKACTIVE))) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		/*
+		 * Since the port can be ACTIVE when we ask for ARMED,
+		 * clear QIBL_LINKV so we can wait for a transition.
+		 * If the link isn't ARMED, then something else happened
+		 * and there is no point waiting for ARMED.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ARMED | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKV;
+		break;
+
+	case QIB_IB_LINKACTIVE:
+		if (ppd->lflags & QIBL_LINKACTIVE) {
+			ret = 0;
+			goto bail;
+		}
+		if (!(ppd->lflags & QIBL_LINKARMED)) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LSTATE,
+				 IB_LINKCMD_ACTIVE | IB_LINKINITCMD_NOP);
+		lstate = QIBL_LINKACTIVE;
+		break;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+	ret = qib_wait_linkstate(ppd, lstate, 10);
+
+bail:
+	return ret;
+}
+
+/*
+ * Get address of eager buffer from it's index (allocated in chunks, not
+ * contiguous).
+ */
+static inline void *qib_get_egrbuf(const struct qib_ctxtdata *rcd, u32 etail)
+{
+	const u32 chunk = etail >> rcd->rcvegrbufs_perchunk_shift;
+	const u32 idx =  etail & ((u32)rcd->rcvegrbufs_perchunk - 1);
+
+	return rcd->rcvegrbuf[chunk] + (idx << rcd->dd->rcvegrbufsize_shift);
+}
+
+/*
+ * Returns 1 if error was a CRC, else 0.
+ * Needed for some chip's synthesized error counters.
+ */
+static u32 qib_rcv_hdrerr(struct qib_ctxtdata *rcd, struct qib_pportdata *ppd,
+			  u32 ctxt, u32 eflags, u32 l, u32 etail,
+			  __le32 *rhf_addr, struct qib_message_header *rhdr)
+{
+	u32 ret = 0;
+
+	if (eflags & (QLOGIC_IB_RHF_H_ICRCERR | QLOGIC_IB_RHF_H_VCRCERR))
+		ret = 1;
+	else if (eflags == QLOGIC_IB_RHF_H_TIDERR) {
+		/* For TIDERR and RC QPs premptively schedule a NAK */
+		struct qib_ib_header *hdr = (struct qib_ib_header *) rhdr;
+		struct qib_other_headers *ohdr = NULL;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		struct qib_qp *qp = NULL;
+		u32 tlen = qib_hdrget_length_in_bytes(rhf_addr);
+		u16 lid  = be16_to_cpu(hdr->lrh[1]);
+		int lnh = be16_to_cpu(hdr->lrh[0]) & 3;
+		u32 qp_num;
+		u32 opcode;
+		u32 psn;
+		int diff;
+
+		/* Sanity check packet */
+		if (tlen < 24)
+			goto drop;
+
+		if (lid < QIB_MULTICAST_LID_BASE) {
+			lid &= ~((1 << ppd->lmc) - 1);
+			if (unlikely(lid != ppd->lid))
+				goto drop;
+		}
+
+		/* Check for GRH */
+		if (lnh == QIB_LRH_BTH)
+			ohdr = &hdr->u.oth;
+		else if (lnh == QIB_LRH_GRH) {
+			u32 vtf;
+
+			ohdr = &hdr->u.l.oth;
+			if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
+				goto drop;
+			vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
+			if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
+				goto drop;
+		} else
+			goto drop;
+
+		/* Get opcode and PSN from packet */
+		opcode = be32_to_cpu(ohdr->bth[0]);
+		opcode >>= 24;
+		psn = be32_to_cpu(ohdr->bth[2]);
+
+		/* Get the destination QP number. */
+		qp_num = be32_to_cpu(ohdr->bth[1]) & QIB_QPN_MASK;
+		if (qp_num != QIB_MULTICAST_QPN) {
+			int ruc_res;
+			qp = qib_lookup_qpn(ibp, qp_num);
+			if (!qp)
+				goto drop;
+
+			/*
+			 * Handle only RC QPs - for other QP types drop error
+			 * packet.
+			 */
+			spin_lock(&qp->r_lock);
+
+			/* Check for valid receive state. */
+			if (!(ib_qib_state_ops[qp->state] &
+			      QIB_PROCESS_RECV_OK)) {
+				ibp->n_pkt_drops++;
+				goto unlock;
+			}
+
+			switch (qp->ibqp.qp_type) {
+			case IB_QPT_RC:
+				ruc_res =
+					qib_ruc_check_hdr(
+						ibp, hdr,
+						lnh == QIB_LRH_GRH,
+						qp,
+						be32_to_cpu(ohdr->bth[0]));
+				if (ruc_res)
+					goto unlock;
+
+				/* Only deal with RDMA Writes for now */
+				if (opcode <
+				    IB_OPCODE_RC_RDMA_READ_RESPONSE_FIRST) {
+					diff = qib_cmp24(psn, qp->r_psn);
+					if (!qp->r_nak_state && diff >= 0) {
+						ibp->n_rc_seqnak++;
+						qp->r_nak_state =
+							IB_NAK_PSN_ERROR;
+						/* Use the expected PSN. */
+						qp->r_ack_psn = qp->r_psn;
+						/*
+						 * Wait to send the sequence
+						 * NAK until all packets
+						 * in the receive queue have
+						 * been processed.
+						 * Otherwise, we end up
+						 * propagating congestion.
+						 */
+						if (list_empty(&qp->rspwait)) {
+							qp->r_flags |=
+								QIB_R_RSP_NAK;
+							atomic_inc(
+								&qp->refcount);
+							list_add_tail(
+							 &qp->rspwait,
+							 &rcd->qp_wait_list);
+						}
+					} /* Out of sequence NAK */
+				} /* QP Request NAKs */
+				break;
+			case IB_QPT_SMI:
+			case IB_QPT_GSI:
+			case IB_QPT_UD:
+			case IB_QPT_UC:
+			default:
+				/* For now don't handle any other QP types */
+				break;
+			}
+
+unlock:
+			spin_unlock(&qp->r_lock);
+			/*
+			 * Notify qib_destroy_qp() if it is waiting
+			 * for us to finish.
+			 */
+			if (atomic_dec_and_test(&qp->refcount))
+				wake_up(&qp->wait);
+		} /* Unicast QP */
+	} /* Valid packet with TIDErr */
+
+drop:
+	return ret;
+}
+
+/*
+ * qib_kreceive - receive a packet
+ * @rcd: the qlogic_ib context
+ * @llic: gets count of good packets needed to clear lli,
+ *          (used with chips that need need to track crcs for lli)
+ *
+ * called from interrupt handler for errors or receive interrupt
+ * Returns number of CRC error packets, needed by some chips for
+ * local link integrity tracking.   crcs are adjusted down by following
+ * good packets, if any, and count of good packets is also tracked.
+ */
+u32 qib_kreceive(struct qib_ctxtdata *rcd, u32 *llic, u32 *npkts)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	__le32 *rhf_addr;
+	void *ebuf;
+	const u32 rsize = dd->rcvhdrentsize;        /* words */
+	const u32 maxcnt = dd->rcvhdrcnt * rsize;   /* words */
+	u32 etail = -1, l, hdrqtail;
+	struct qib_message_header *hdr;
+	u32 eflags, etype, tlen, i = 0, updegr = 0, crcs = 0;
+	int last;
+	u64 lval;
+	struct qib_qp *qp, *nqp;
+
+	l = rcd->head;
+	rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+	if (dd->flags & QIB_NODMA_RTAIL) {
+		u32 seq = qib_hdrget_seq(rhf_addr);
+		if (seq != rcd->seq_cnt)
+			goto bail;
+		hdrqtail = 0;
+	} else {
+		hdrqtail = qib_get_rcvhdrtail(rcd);
+		if (l == hdrqtail)
+			goto bail;
+		smp_rmb();  /* prevent speculative reads of dma'ed hdrq */
+	}
+
+	for (last = 0, i = 1; !last; i += !last) {
+		hdr = dd->f_get_msgheader(dd, rhf_addr);
+		eflags = qib_hdrget_err_flags(rhf_addr);
+		etype = qib_hdrget_rcv_type(rhf_addr);
+		/* total length */
+		tlen = qib_hdrget_length_in_bytes(rhf_addr);
+		ebuf = NULL;
+		if ((dd->flags & QIB_NODMA_RTAIL) ?
+		    qib_hdrget_use_egr_buf(rhf_addr) :
+		    (etype != RCVHQ_RCV_TYPE_EXPECTED)) {
+			etail = qib_hdrget_index(rhf_addr);
+			updegr = 1;
+			if (tlen > sizeof(*hdr) ||
+			    etype >= RCVHQ_RCV_TYPE_NON_KD) {
+				ebuf = qib_get_egrbuf(rcd, etail);
+				prefetch_range(ebuf, tlen - sizeof(*hdr));
+			}
+		}
+		if (!eflags) {
+			u16 lrh_len = be16_to_cpu(hdr->lrh[2]) << 2;
+
+			if (lrh_len != tlen) {
+				qib_stats.sps_lenerrs++;
+				goto move_along;
+			}
+		}
+		if (etype == RCVHQ_RCV_TYPE_NON_KD && !eflags &&
+		    ebuf == NULL &&
+		    tlen > (dd->rcvhdrentsize - 2 + 1 -
+				qib_hdrget_offset(rhf_addr)) << 2) {
+			goto move_along;
+		}
+
+		/*
+		 * Both tiderr and qibhdrerr are set for all plain IB
+		 * packets; only qibhdrerr should be set.
+		 */
+		if (unlikely(eflags))
+			crcs += qib_rcv_hdrerr(rcd, ppd, rcd->ctxt, eflags, l,
+					       etail, rhf_addr, hdr);
+		else if (etype == RCVHQ_RCV_TYPE_NON_KD) {
+			qib_ib_rcv(rcd, hdr, ebuf, tlen);
+			if (crcs)
+				crcs--;
+			else if (llic && *llic)
+				--*llic;
+		}
+move_along:
+		l += rsize;
+		if (l >= maxcnt)
+			l = 0;
+		if (i == QIB_MAX_PKT_RECV)
+			last = 1;
+
+		rhf_addr = (__le32 *) rcd->rcvhdrq + l + dd->rhf_offset;
+		if (dd->flags & QIB_NODMA_RTAIL) {
+			u32 seq = qib_hdrget_seq(rhf_addr);
+
+			if (++rcd->seq_cnt > 13)
+				rcd->seq_cnt = 1;
+			if (seq != rcd->seq_cnt)
+				last = 1;
+		} else if (l == hdrqtail)
+			last = 1;
+		/*
+		 * Update head regs etc., every 16 packets, if not last pkt,
+		 * to help prevent rcvhdrq overflows, when many packets
+		 * are processed and queue is nearly full.
+		 * Don't request an interrupt for intermediate updates.
+		 */
+		lval = l;
+		if (!last && !(i & 0xf)) {
+			dd->f_update_usrhead(rcd, lval, updegr, etail, i);
+			updegr = 0;
+		}
+	}
+	/*
+	 * Notify qib_destroy_qp() if it is waiting
+	 * for lookaside_qp to finish.
+	 */
+	if (rcd->lookaside_qp) {
+		if (atomic_dec_and_test(&rcd->lookaside_qp->refcount))
+			wake_up(&rcd->lookaside_qp->wait);
+		rcd->lookaside_qp = NULL;
+	}
+
+	rcd->head = l;
+
+	/*
+	 * Iterate over all QPs waiting to respond.
+	 * The list won't change since the IRQ is only run on one CPU.
+	 */
+	list_for_each_entry_safe(qp, nqp, &rcd->qp_wait_list, rspwait) {
+		list_del_init(&qp->rspwait);
+		if (qp->r_flags & QIB_R_RSP_NAK) {
+			qp->r_flags &= ~QIB_R_RSP_NAK;
+			qib_send_rc_ack(qp);
+		}
+		if (qp->r_flags & QIB_R_RSP_SEND) {
+			unsigned long flags;
+
+			qp->r_flags &= ~QIB_R_RSP_SEND;
+			spin_lock_irqsave(&qp->s_lock, flags);
+			if (ib_qib_state_ops[qp->state] &
+					QIB_PROCESS_OR_FLUSH_SEND)
+				qib_schedule_send(qp);
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+		}
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+
+bail:
+	/* Report number of packets consumed */
+	if (npkts)
+		*npkts = i;
+
+	/*
+	 * Always write head at end, and setup rcv interrupt, even
+	 * if no packets were processed.
+	 */
+	lval = (u64)rcd->head | dd->rhdrhead_intr_off;
+	dd->f_update_usrhead(rcd, lval, updegr, etail, i);
+	return crcs;
+}
+
+/**
+ * qib_set_mtu - set the MTU
+ * @ppd: the perport data
+ * @arg: the new MTU
+ *
+ * We can handle "any" incoming size, the issue here is whether we
+ * need to restrict our outgoing size.   For now, we don't do any
+ * sanity checking on this, and we don't deal with what happens to
+ * programs that are already running when the size changes.
+ * NOTE: changing the MTU will usually cause the IBC to go back to
+ * link INIT state...
+ */
+int qib_set_mtu(struct qib_pportdata *ppd, u16 arg)
+{
+	u32 piosize;
+	int ret, chk;
+
+	if (arg != 256 && arg != 512 && arg != 1024 && arg != 2048 &&
+	    arg != 4096) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	chk = ib_mtu_enum_to_int(qib_ibmtu);
+	if (chk > 0 && arg > chk) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	piosize = ppd->ibmaxlen;
+	ppd->ibmtu = arg;
+
+	if (arg >= (piosize - QIB_PIO_MAXIBHDR)) {
+		/* Only if it's not the initial value (or reset to it) */
+		if (piosize != ppd->init_ibmaxlen) {
+			if (arg > piosize && arg <= ppd->init_ibmaxlen)
+				piosize = ppd->init_ibmaxlen - 2 * sizeof(u32);
+			ppd->ibmaxlen = piosize;
+		}
+	} else if ((arg + QIB_PIO_MAXIBHDR) != ppd->ibmaxlen) {
+		piosize = arg + QIB_PIO_MAXIBHDR - 2 * sizeof(u32);
+		ppd->ibmaxlen = piosize;
+	}
+
+	ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_MTU, 0);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_set_lid(struct qib_pportdata *ppd, u32 lid, u8 lmc)
+{
+	struct qib_devdata *dd = ppd->dd;
+	ppd->lid = lid;
+	ppd->lmc = lmc;
+
+	dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LIDLMC,
+			 lid | (~((1U << lmc) - 1)) << 16);
+
+	qib_devinfo(dd->pcidev, "IB%u:%u got a lid: 0x%x\n",
+		    dd->unit, ppd->port, lid);
+
+	return 0;
+}
+
+/*
+ * Following deal with the "obviously simple" task of overriding the state
+ * of the LEDS, which normally indicate link physical and logical status.
+ * The complications arise in dealing with different hardware mappings
+ * and the board-dependent routine being called from interrupts.
+ * and then there's the requirement to _flash_ them.
+ */
+#define LED_OVER_FREQ_SHIFT 8
+#define LED_OVER_FREQ_MASK (0xFF<<LED_OVER_FREQ_SHIFT)
+/* Below is "non-zero" to force override, but both actual LEDs are off */
+#define LED_OVER_BOTH_OFF (8)
+
+static void qib_run_led_override(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff;
+	int ph_idx;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	ph_idx = ppd->led_override_phase++ & 1;
+	ppd->led_override = ppd->led_override_vals[ph_idx];
+	timeoff = ppd->led_override_timeoff;
+
+	dd->f_setextled(ppd, 1);
+	/*
+	 * don't re-fire the timer if user asked for it to be off; we let
+	 * it fire one more time after they turn it off to simplify
+	 */
+	if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+		mod_timer(&ppd->led_override_timer, jiffies + timeoff);
+}
+
+void qib_set_led_override(struct qib_pportdata *ppd, unsigned int val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int timeoff, freq;
+
+	if (!(dd->flags & QIB_INITTED))
+		return;
+
+	/* First check if we are blinking. If not, use 1HZ polling */
+	timeoff = HZ;
+	freq = (val & LED_OVER_FREQ_MASK) >> LED_OVER_FREQ_SHIFT;
+
+	if (freq) {
+		/* For blink, set each phase from one nybble of val */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = (val >> 4) & 0xF;
+		timeoff = (HZ << 4)/freq;
+	} else {
+		/* Non-blink set both phases the same. */
+		ppd->led_override_vals[0] = val & 0xF;
+		ppd->led_override_vals[1] = val & 0xF;
+	}
+	ppd->led_override_timeoff = timeoff;
+
+	/*
+	 * If the timer has not already been started, do so. Use a "quick"
+	 * timeout so the function will be called soon, to look at our request.
+	 */
+	if (atomic_inc_return(&ppd->led_override_timer_active) == 1) {
+		/* Need to start timer */
+		init_timer(&ppd->led_override_timer);
+		ppd->led_override_timer.function = qib_run_led_override;
+		ppd->led_override_timer.data = (unsigned long) ppd;
+		ppd->led_override_timer.expires = jiffies + 1;
+		add_timer(&ppd->led_override_timer);
+	} else {
+		if (ppd->led_override_vals[0] || ppd->led_override_vals[1])
+			mod_timer(&ppd->led_override_timer, jiffies + 1);
+		atomic_dec(&ppd->led_override_timer_active);
+	}
+}
+
+/**
+ * qib_reset_device - reset the chip if possible
+ * @unit: the device to reset
+ *
+ * Whether or not reset is successful, we attempt to re-initialize the chip
+ * (that is, much like a driver unload/reload).  We clear the INITTED flag
+ * so that the various entry points will fail until we reinitialize.  For
+ * now, we only allow this if no user contexts are open that use chip resources
+ */
+int qib_reset_device(int unit)
+{
+	int ret, i;
+	struct qib_devdata *dd = qib_lookup(unit);
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	int pidx;
+
+	if (!dd) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	qib_devinfo(dd->pcidev, "Reset on unit %u requested\n", unit);
+
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT)) {
+		qib_devinfo(dd->pcidev,
+			"Invalid unit number %u or not initialized or not present\n",
+			unit);
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	if (dd->rcd)
+		for (i = dd->first_user_ctxt; i < dd->cfgctxts; i++) {
+			if (!dd->rcd[i] || !dd->rcd[i]->cnt)
+				continue;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			ret = -EBUSY;
+			goto bail;
+		}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (atomic_read(&ppd->led_override_timer_active)) {
+			/* Need to stop LED timer, _then_ shut off LEDs */
+			del_timer_sync(&ppd->led_override_timer);
+			atomic_set(&ppd->led_override_timer_active, 0);
+		}
+
+		/* Shut off LEDs after we are sure timer is not running */
+		ppd->led_override = LED_OVER_BOTH_OFF;
+		dd->f_setextled(ppd, 0);
+		if (dd->flags & QIB_HAS_SEND_DMA)
+			qib_teardown_sdma(ppd);
+	}
+
+	ret = dd->f_reset(dd);
+	if (ret == 1)
+		ret = qib_init(dd, 1);
+	else
+		ret = -EAGAIN;
+	if (ret)
+		qib_dev_err(dd,
+			"Reinitialize unit %u after reset failed with %d\n",
+			unit, ret);
+	else
+		qib_devinfo(dd->pcidev,
+			"Reinitialized unit %u after resetting\n",
+			unit);
+
+bail:
+	return ret;
+}
diff --git a/drivers/infiniband/hw/qib/qib_eeprom.c b/drivers/infiniband/hw/qib/qib_eeprom.c
new file mode 100644
index 00000000000..4d5d71aaa2b
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_eeprom.c
@@ -0,0 +1,456 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * Functions specific to the serial EEPROM on cards handled by ib_qib.
+ * The actual serail interface code is in qib_twsi.c. This file is a client
+ */
+
+/**
+ * qib_eeprom_read - receives bytes from the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: address to read from
+ * @buffer: where to store result
+ * @len: number of bytes to receive
+ */
+int qib_eeprom_read(struct qib_devdata *dd, u8 eeprom_offset,
+		    void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = qib_twsi_reset(dd);
+		if (ret)
+			qib_dev_err(dd, "EEPROM Reset for read failed\n");
+		else
+			ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev,
+					      eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+/*
+ * Actually update the eeprom, first doing write enable if
+ * needed, then restoring write enable state.
+ * Must be called with eep_lock held
+ */
+static int eeprom_write_with_enable(struct qib_devdata *dd, u8 offset,
+		     const void *buf, int len)
+{
+	int ret, pwen;
+
+	pwen = dd->f_eeprom_wen(dd, 1);
+	ret = qib_twsi_reset(dd);
+	if (ret)
+		qib_dev_err(dd, "EEPROM Reset for write failed\n");
+	else
+		ret = qib_twsi_blk_wr(dd, dd->twsi_eeprom_dev,
+				      offset, buf, len);
+	dd->f_eeprom_wen(dd, pwen);
+	return ret;
+}
+
+/**
+ * qib_eeprom_write - writes data to the eeprom via I2C
+ * @dd: the qlogic_ib device
+ * @eeprom_offset: where to place data
+ * @buffer: data to write
+ * @len: number of bytes to write
+ */
+int qib_eeprom_write(struct qib_devdata *dd, u8 eeprom_offset,
+		     const void *buff, int len)
+{
+	int ret;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (!ret) {
+		ret = eeprom_write_with_enable(dd, eeprom_offset, buff, len);
+		mutex_unlock(&dd->eep_lock);
+	}
+
+	return ret;
+}
+
+static u8 flash_csum(struct qib_flash *ifp, int adjust)
+{
+	u8 *ip = (u8 *) ifp;
+	u8 csum = 0, len;
+
+	/*
+	 * Limit length checksummed to max length of actual data.
+	 * Checksum of erased eeprom will still be bad, but we avoid
+	 * reading past the end of the buffer we were passed.
+	 */
+	len = ifp->if_length;
+	if (len > sizeof(struct qib_flash))
+		len = sizeof(struct qib_flash);
+	while (len--)
+		csum += *ip++;
+	csum -= ifp->if_csum;
+	csum = ~csum;
+	if (adjust)
+		ifp->if_csum = csum;
+
+	return csum;
+}
+
+/**
+ * qib_get_eeprom_info- get the GUID et al. from the TSWI EEPROM device
+ * @dd: the qlogic_ib device
+ *
+ * We have the capability to use the nguid field, and get
+ * the guid from the first chip's flash, to use for all of them.
+ */
+void qib_get_eeprom_info(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	__be64 guid;
+	int len, eep_stat;
+	u8 csum, *bguid;
+	int t = dd->unit;
+	struct qib_devdata *dd0 = qib_lookup(0);
+
+	if (t && dd0->nguid > 1 && t <= dd0->nguid) {
+		u8 oguid;
+		dd->base_guid = dd0->base_guid;
+		bguid = (u8 *) &dd->base_guid;
+
+		oguid = bguid[7];
+		bguid[7] += t;
+		if (oguid > bguid[7]) {
+			if (bguid[6] == 0xff) {
+				if (bguid[5] == 0xff) {
+					qib_dev_err(dd,
+						"Can't set %s GUID from base, wraps to OUI!\n",
+						qib_get_unit_name(t));
+					dd->base_guid = 0;
+					goto bail;
+				}
+				bguid[5]++;
+			}
+			bguid[6]++;
+		}
+		dd->nguid = 1;
+		goto bail;
+	}
+
+	/*
+	 * Read full flash, not just currently used part, since it may have
+	 * been written with a newer definition.
+	 * */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	if (!buf) {
+		qib_dev_err(dd,
+			"Couldn't allocate memory to read %u bytes from eeprom for GUID\n",
+			len);
+		goto bail;
+	}
+
+	/*
+	 * Use "public" eeprom read function, which does locking and
+	 * figures out device. This will migrate to chip-specific.
+	 */
+	eep_stat = qib_eeprom_read(dd, 0, buf, len);
+
+	if (eep_stat) {
+		qib_dev_err(dd, "Failed reading GUID from eeprom\n");
+		goto done;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		qib_devinfo(dd->pcidev,
+			"Bad I2C flash checksum: 0x%x, not 0x%x\n",
+			csum, ifp->if_csum);
+		goto done;
+	}
+	if (*(__be64 *) ifp->if_guid == cpu_to_be64(0) ||
+	    *(__be64 *) ifp->if_guid == ~cpu_to_be64(0)) {
+		qib_dev_err(dd,
+			"Invalid GUID %llx from flash; ignoring\n",
+			*(unsigned long long *) ifp->if_guid);
+		/* don't allow GUID if all 0 or all 1's */
+		goto done;
+	}
+
+	/* complain, but allow it */
+	if (*(u64 *) ifp->if_guid == 0x100007511000000ULL)
+		qib_devinfo(dd->pcidev,
+			"Warning, GUID %llx is default, probably not correct!\n",
+			*(unsigned long long *) ifp->if_guid);
+
+	bguid = ifp->if_guid;
+	if (!bguid[0] && !bguid[1] && !bguid[2]) {
+		/*
+		 * Original incorrect GUID format in flash; fix in
+		 * core copy, by shifting up 2 octets; don't need to
+		 * change top octet, since both it and shifted are 0.
+		 */
+		bguid[1] = bguid[3];
+		bguid[2] = bguid[4];
+		bguid[3] = 0;
+		bguid[4] = 0;
+		guid = *(__be64 *) ifp->if_guid;
+	} else
+		guid = *(__be64 *) ifp->if_guid;
+	dd->base_guid = guid;
+	dd->nguid = ifp->if_numguid;
+	/*
+	 * Things are slightly complicated by the desire to transparently
+	 * support both the Pathscale 10-digit serial number and the QLogic
+	 * 13-character version.
+	 */
+	if ((ifp->if_fversion > 1) && ifp->if_sprefix[0] &&
+	    ((u8 *) ifp->if_sprefix)[0] != 0xFF) {
+		char *snp = dd->serial;
+
+		/*
+		 * This board has a Serial-prefix, which is stored
+		 * elsewhere for backward-compatibility.
+		 */
+		memcpy(snp, ifp->if_sprefix, sizeof ifp->if_sprefix);
+		snp[sizeof ifp->if_sprefix] = '\0';
+		len = strlen(snp);
+		snp += len;
+		len = (sizeof dd->serial) - len;
+		if (len > sizeof ifp->if_serial)
+			len = sizeof ifp->if_serial;
+		memcpy(snp, ifp->if_serial, len);
+	} else
+		memcpy(dd->serial, ifp->if_serial,
+		       sizeof ifp->if_serial);
+	if (!strstr(ifp->if_comment, "Tested successfully"))
+		qib_dev_err(dd,
+			"Board SN %s did not pass functional test: %s\n",
+			dd->serial, ifp->if_comment);
+
+	memcpy(&dd->eep_st_errs, &ifp->if_errcntp, QIB_EEP_LOG_CNT);
+	/*
+	 * Power-on (actually "active") hours are kept as little-endian value
+	 * in EEPROM, but as seconds in a (possibly as small as 24-bit)
+	 * atomic_t while running.
+	 */
+	atomic_set(&dd->active_time, 0);
+	dd->eep_hrs = ifp->if_powerhour[0] | (ifp->if_powerhour[1] << 8);
+
+done:
+	vfree(buf);
+
+bail:;
+}
+
+/**
+ * qib_update_eeprom_log - copy active-time and error counters to eeprom
+ * @dd: the qlogic_ib device
+ *
+ * Although the time is kept as seconds in the qib_devdata struct, it is
+ * rounded to hours for re-write, as we have only 16 bits in EEPROM.
+ * First-cut code reads whole (expected) struct qib_flash, modifies,
+ * re-writes. Future direction: read/write only what we need, assuming
+ * that the EEPROM had to have been "good enough" for driver init, and
+ * if not, we aren't making it worse.
+ *
+ */
+int qib_update_eeprom_log(struct qib_devdata *dd)
+{
+	void *buf;
+	struct qib_flash *ifp;
+	int len, hi_water;
+	uint32_t new_time, new_hrs;
+	u8 csum;
+	int ret, idx;
+	unsigned long flags;
+
+	/* first, check if we actually need to do anything. */
+	ret = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		if (dd->eep_st_new_errs[idx]) {
+			ret = 1;
+			break;
+		}
+	}
+	new_time = atomic_read(&dd->active_time);
+
+	if (ret == 0 && new_time < 3600)
+		goto bail;
+
+	/*
+	 * The quick-check above determined that there is something worthy
+	 * of logging, so get current contents and do a more detailed idea.
+	 * read full flash, not just currently used part, since it may have
+	 * been written with a newer definition
+	 */
+	len = sizeof(struct qib_flash);
+	buf = vmalloc(len);
+	ret = 1;
+	if (!buf) {
+		qib_dev_err(dd,
+			"Couldn't allocate memory to read %u bytes from eeprom for logging\n",
+			len);
+		goto bail;
+	}
+
+	/* Grab semaphore and read current EEPROM. If we get an
+	 * error, let go, but if not, keep it until we finish write.
+	 */
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret) {
+		qib_dev_err(dd, "Unable to acquire EEPROM for logging\n");
+		goto free_bail;
+	}
+	ret = qib_twsi_blk_rd(dd, dd->twsi_eeprom_dev, 0, buf, len);
+	if (ret) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "Unable read EEPROM for logging\n");
+		goto free_bail;
+	}
+	ifp = (struct qib_flash *)buf;
+
+	csum = flash_csum(ifp, 0);
+	if (csum != ifp->if_csum) {
+		mutex_unlock(&dd->eep_lock);
+		qib_dev_err(dd, "EEPROM cks err (0x%02X, S/B 0x%02X)\n",
+			    csum, ifp->if_csum);
+		ret = 1;
+		goto free_bail;
+	}
+	hi_water = 0;
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		int new_val = dd->eep_st_new_errs[idx];
+		if (new_val) {
+			/*
+			 * If we have seen any errors, add to EEPROM values
+			 * We need to saturate at 0xFF (255) and we also
+			 * would need to adjust the checksum if we were
+			 * trying to minimize EEPROM traffic
+			 * Note that we add to actual current count in EEPROM,
+			 * in case it was altered while we were running.
+			 */
+			new_val += ifp->if_errcntp[idx];
+			if (new_val > 0xFF)
+				new_val = 0xFF;
+			if (ifp->if_errcntp[idx] != new_val) {
+				ifp->if_errcntp[idx] = new_val;
+				hi_water = offsetof(struct qib_flash,
+						    if_errcntp) + idx;
+			}
+			/*
+			 * update our shadow (used to minimize EEPROM
+			 * traffic), to match what we are about to write.
+			 */
+			dd->eep_st_errs[idx] = new_val;
+			dd->eep_st_new_errs[idx] = 0;
+		}
+	}
+	/*
+	 * Now update active-time. We would like to round to the nearest hour
+	 * but unless atomic_t are sure to be proper signed ints we cannot,
+	 * because we need to account for what we "transfer" to EEPROM and
+	 * if we log an hour at 31 minutes, then we would need to set
+	 * active_time to -29 to accurately count the _next_ hour.
+	 */
+	if (new_time >= 3600) {
+		new_hrs = new_time / 3600;
+		atomic_sub((new_hrs * 3600), &dd->active_time);
+		new_hrs += dd->eep_hrs;
+		if (new_hrs > 0xFFFF)
+			new_hrs = 0xFFFF;
+		dd->eep_hrs = new_hrs;
+		if ((new_hrs & 0xFF) != ifp->if_powerhour[0]) {
+			ifp->if_powerhour[0] = new_hrs & 0xFF;
+			hi_water = offsetof(struct qib_flash, if_powerhour);
+		}
+		if ((new_hrs >> 8) != ifp->if_powerhour[1]) {
+			ifp->if_powerhour[1] = new_hrs >> 8;
+			hi_water = offsetof(struct qib_flash, if_powerhour) + 1;
+		}
+	}
+	/*
+	 * There is a tiny possibility that we could somehow fail to write
+	 * the EEPROM after updating our shadows, but problems from holding
+	 * the spinlock too long are a much bigger issue.
+	 */
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+	if (hi_water) {
+		/* we made some change to the data, uopdate cksum and write */
+		csum = flash_csum(ifp, 1);
+		ret = eeprom_write_with_enable(dd, 0, buf, hi_water + 1);
+	}
+	mutex_unlock(&dd->eep_lock);
+	if (ret)
+		qib_dev_err(dd, "Failed updating EEPROM\n");
+
+free_bail:
+	vfree(buf);
+bail:
+	return ret;
+}
+
+/**
+ * qib_inc_eeprom_err - increment one of the four error counters
+ * that are logged to EEPROM.
+ * @dd: the qlogic_ib device
+ * @eidx: 0..3, the counter to increment
+ * @incr: how much to add
+ *
+ * Each counter is 8-bits, and saturates at 255 (0xFF). They
+ * are copied to the EEPROM (aka flash) whenever qib_update_eeprom_log()
+ * is called, but it can only be called in a context that allows sleep.
+ * This function can be called even at interrupt level.
+ */
+void qib_inc_eeprom_err(struct qib_devdata *dd, u32 eidx, u32 incr)
+{
+	uint new_val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	new_val = dd->eep_st_new_errs[eidx] + incr;
+	if (new_val > 255)
+		new_val = 255;
+	dd->eep_st_new_errs[eidx] = new_val;
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+}
diff --git a/drivers/infiniband/hw/qib/qib_file_ops.c b/drivers/infiniband/hw/qib/qib_file_ops.c
new file mode 100644
index 00000000000..b15e34eeef6
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_file_ops.c
@@ -0,0 +1,2410 @@
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/poll.h>
+#include <linux/cdev.h>
+#include <linux/swap.h>
+#include <linux/vmalloc.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/aio.h>
+#include <linux/jiffies.h>
+#include <asm/pgtable.h>
+#include <linux/delay.h>
+#include <linux/export.h>
+
+#include "qib.h"
+#include "qib_common.h"
+#include "qib_user_sdma.h"
+
+#undef pr_fmt
+#define pr_fmt(fmt) QIB_DRV_NAME ": " fmt
+
+static int qib_open(struct inode *, struct file *);
+static int qib_close(struct inode *, struct file *);
+static ssize_t qib_write(struct file *, const char __user *, size_t, loff_t *);
+static ssize_t qib_aio_write(struct kiocb *, const struct iovec *,
+			     unsigned long, loff_t);
+static unsigned int qib_poll(struct file *, struct poll_table_struct *);
+static int qib_mmapf(struct file *, struct vm_area_struct *);
+
+static const struct file_operations qib_file_ops = {
+	.owner = THIS_MODULE,
+	.write = qib_write,
+	.aio_write = qib_aio_write,
+	.open = qib_open,
+	.release = qib_close,
+	.poll = qib_poll,
+	.mmap = qib_mmapf,
+	.llseek = noop_llseek,
+};
+
+/*
+ * Convert kernel virtual addresses to physical addresses so they don't
+ * potentially conflict with the chip addresses used as mmap offsets.
+ * It doesn't really matter what mmap offset we use as long as we can
+ * interpret it correctly.
+ */
+static u64 cvt_kvaddr(void *p)
+{
+	struct page *page;
+	u64 paddr = 0;
+
+	page = vmalloc_to_page(p);
+	if (page)
+		paddr = page_to_pfn(page) << PAGE_SHIFT;
+
+	return paddr;
+}
+
+static int qib_get_base_info(struct file *fp, void __user *ubase,
+			     size_t ubase_size)
+{
+	struct qib_ctxtdata *rcd = ctxt_fp(fp);
+	int ret = 0;
+	struct qib_base_info *kinfo = NULL;
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	unsigned subctxt_cnt;
+	int shared, master;
+	size_t sz;
+
+	subctxt_cnt = rcd->subctxt_cnt;
+	if (!subctxt_cnt) {
+		shared = 0;
+		master = 0;
+		subctxt_cnt = 1;
+	} else {
+		shared = 1;
+		master = !subctxt_fp(fp);
+	}
+
+	sz = sizeof(*kinfo);
+	/* If context sharing is not requested, allow the old size structure */
+	if (!shared)
+		sz -= 7 * sizeof(u64);
+	if (ubase_size < sz) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	kinfo = kzalloc(sizeof(*kinfo), GFP_KERNEL);
+	if (kinfo == NULL) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	ret = dd->f_get_base_info(rcd, kinfo);
+	if (ret < 0)
+		goto bail;
+
+	kinfo->spi_rcvhdr_cnt = dd->rcvhdrcnt;
+	kinfo->spi_rcvhdrent_size = dd->rcvhdrentsize;
+	kinfo->spi_tidegrcnt = rcd->rcvegrcnt;
+	kinfo->spi_rcv_egrbufsize = dd->rcvegrbufsize;
+	/*
+	 * have to mmap whole thing
+	 */
+	kinfo->spi_rcv_egrbuftotlen =
+		rcd->rcvegrbuf_chunks * rcd->rcvegrbuf_size;
+	kinfo->spi_rcv_egrperchunk = rcd->rcvegrbufs_perchunk;
+	kinfo->spi_rcv_egrchunksize = kinfo->spi_rcv_egrbuftotlen /
+		rcd->rcvegrbuf_chunks;
+	kinfo->spi_tidcnt = dd->rcvtidcnt / subctxt_cnt;
+	if (master)
+		kinfo->spi_tidcnt += dd->rcvtidcnt % subctxt_cnt;
+	/*
+	 * for this use, may be cfgctxts summed over all chips that
+	 * are are configured and present
+	 */
+	kinfo->spi_nctxts = dd->cfgctxts;
+	/* unit (chip/board) our context is on */
+	kinfo->spi_unit = dd->unit;
+	kinfo->spi_port = ppd->port;
+	/* for now, only a single page */
+	kinfo->spi_tid_maxsize = PAGE_SIZE;
+
+	/*
+	 * Doing this per context, and based on the skip value, etc.  This has
+	 * to be the actual buffer size, since the protocol code treats it
+	 * as an array.
+	 *
+	 * These have to be set to user addresses in the user code via mmap.
+	 * These values are used on return to user code for the mmap target
+	 * addresses only.  For 32 bit, same 44 bit address problem, so use
+	 * the physical address, not virtual.  Before 2.6.11, using the
+	 * page_address() macro worked, but in 2.6.11, even that returns the
+	 * full 64 bit address (upper bits all 1's).  So far, using the
+	 * physical addresses (or chip offsets, for chip mapping) works, but
+	 * no doubt some future kernel release will change that, and we'll be
+	 * on to yet another method of dealing with this.
+	 * Normally only one of rcvhdr_tailaddr or rhf_offset is useful
+	 * since the chips with non-zero rhf_offset don't normally
+	 * enable tail register updates to host memory, but for testing,
+	 * both can be enabled and used.
+	 */
+	kinfo->spi_rcvhdr_base = (u64) rcd->rcvhdrq_phys;
+	kinfo->spi_rcvhdr_tailaddr = (u64) rcd->rcvhdrqtailaddr_phys;
+	kinfo->spi_rhf_offset = dd->rhf_offset;
+	kinfo->spi_rcv_egrbufs = (u64) rcd->rcvegr_phys;
+	kinfo->spi_pioavailaddr = (u64) dd->pioavailregs_phys;
+	/* setup per-unit (not port) status area for user programs */
+	kinfo->spi_status = (u64) kinfo->spi_pioavailaddr +
+		(char *) ppd->statusp -
+		(char *) dd->pioavailregs_dma;
+	kinfo->spi_uregbase = (u64) dd->uregbase + dd->ureg_align * rcd->ctxt;
+	if (!shared) {
+		kinfo->spi_piocnt = rcd->piocnt;
+		kinfo->spi_piobufbase = (u64) rcd->piobufs;
+		kinfo->spi_sendbuf_status = cvt_kvaddr(rcd->user_event_mask);
+	} else if (master) {
+		kinfo->spi_piocnt = (rcd->piocnt / subctxt_cnt) +
+				    (rcd->piocnt % subctxt_cnt);
+		/* Master's PIO buffers are after all the slave's */
+		kinfo->spi_piobufbase = (u64) rcd->piobufs +
+			dd->palign *
+			(rcd->piocnt - kinfo->spi_piocnt);
+	} else {
+		unsigned slave = subctxt_fp(fp) - 1;
+
+		kinfo->spi_piocnt = rcd->piocnt / subctxt_cnt;
+		kinfo->spi_piobufbase = (u64) rcd->piobufs +
+			dd->palign * kinfo->spi_piocnt * slave;
+	}
+
+	if (shared) {
+		kinfo->spi_sendbuf_status =
+			cvt_kvaddr(&rcd->user_event_mask[subctxt_fp(fp)]);
+		/* only spi_subctxt_* fields should be set in this block! */
+		kinfo->spi_subctxt_uregbase = cvt_kvaddr(rcd->subctxt_uregbase);
+
+		kinfo->spi_subctxt_rcvegrbuf =
+			cvt_kvaddr(rcd->subctxt_rcvegrbuf);
+		kinfo->spi_subctxt_rcvhdr_base =
+			cvt_kvaddr(rcd->subctxt_rcvhdr_base);
+	}
+
+	/*
+	 * All user buffers are 2KB buffers.  If we ever support
+	 * giving 4KB buffers to user processes, this will need some
+	 * work.  Can't use piobufbase directly, because it has
+	 * both 2K and 4K buffer base values.
+	 */
+	kinfo->spi_pioindex = (kinfo->spi_piobufbase - dd->pio2k_bufbase) /
+		dd->palign;
+	kinfo->spi_pioalign = dd->palign;
+	kinfo->spi_qpair = QIB_KD_QP;
+	/*
+	 * user mode PIO buffers are always 2KB, even when 4KB can
+	 * be received, and sent via the kernel; this is ibmaxlen
+	 * for 2K MTU.
+	 */
+	kinfo->spi_piosize = dd->piosize2k - 2 * sizeof(u32);
+	kinfo->spi_mtu = ppd->ibmaxlen; /* maxlen, not ibmtu */
+	kinfo->spi_ctxt = rcd->ctxt;
+	kinfo->spi_subctxt = subctxt_fp(fp);
+	kinfo->spi_sw_version = QIB_KERN_SWVERSION;
+	kinfo->spi_sw_version |= 1U << 31; /* QLogic-built, not kernel.org */
+	kinfo->spi_hw_version = dd->revision;
+
+	if (master)
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_MASTER;
+
+	sz = (ubase_size < sizeof(*kinfo)) ? ubase_size : sizeof(*kinfo);
+	if (copy_to_user(ubase, kinfo, sz))
+		ret = -EFAULT;
+bail:
+	kfree(kinfo);
+	return ret;
+}
+
+/**
+ * qib_tid_update - update a context TID
+ * @rcd: the context
+ * @fp: the qib device file
+ * @ti: the TID information
+ *
+ * The new implementation as of Oct 2004 is that the driver assigns
+ * the tid and returns it to the caller.   To reduce search time, we
+ * keep a cursor for each context, walking the shadow tid array to find
+ * one that's not in use.
+ *
+ * For now, if we can't allocate the full list, we fail, although
+ * in the long run, we'll allocate as many as we can, and the
+ * caller will deal with that by trying the remaining pages later.
+ * That means that when we fail, we have to mark the tids as not in
+ * use again, in our shadow copy.
+ *
+ * It's up to the caller to free the tids when they are done.
+ * We'll unlock the pages as they free them.
+ *
+ * Also, right now we are locking one page at a time, but since
+ * the intended use of this routine is for a single group of
+ * virtually contiguous pages, that should change to improve
+ * performance.
+ */
+static int qib_tid_update(struct qib_ctxtdata *rcd, struct file *fp,
+			  const struct qib_tid_info *ti)
+{
+	int ret = 0, ntids;
+	u32 tid, ctxttid, cnt, i, tidcnt, tidoff;
+	u16 *tidlist;
+	struct qib_devdata *dd = rcd->dd;
+	u64 physaddr;
+	unsigned long vaddr;
+	u64 __iomem *tidbase;
+	unsigned long tidmap[8];
+	struct page **pagep = NULL;
+	unsigned subctxt = subctxt_fp(fp);
+
+	if (!dd->pageshadow) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	cnt = ti->tidcnt;
+	if (!cnt) {
+		ret = -EFAULT;
+		goto done;
+	}
+	ctxttid = rcd->ctxt * dd->rcvtidcnt;
+	if (!rcd->subctxt_cnt) {
+		tidcnt = dd->rcvtidcnt;
+		tid = rcd->tidcursor;
+		tidoff = 0;
+	} else if (!subctxt) {
+		tidcnt = (dd->rcvtidcnt / rcd->subctxt_cnt) +
+			 (dd->rcvtidcnt % rcd->subctxt_cnt);
+		tidoff = dd->rcvtidcnt - tidcnt;
+		ctxttid += tidoff;
+		tid = tidcursor_fp(fp);
+	} else {
+		tidcnt = dd->rcvtidcnt / rcd->subctxt_cnt;
+		tidoff = tidcnt * (subctxt - 1);
+		ctxttid += tidoff;
+		tid = tidcursor_fp(fp);
+	}
+	if (cnt > tidcnt) {
+		/* make sure it all fits in tid_pg_list */
+		qib_devinfo(dd->pcidev,
+			"Process tried to allocate %u TIDs, only trying max (%u)\n",
+			cnt, tidcnt);
+		cnt = tidcnt;
+	}
+	pagep = (struct page **) rcd->tid_pg_list;
+	tidlist = (u16 *) &pagep[dd->rcvtidcnt];
+	pagep += tidoff;
+	tidlist += tidoff;
+
+	memset(tidmap, 0, sizeof(tidmap));
+	/* before decrement; chip actual # */
+	ntids = tidcnt;
+	tidbase = (u64 __iomem *) (((char __iomem *) dd->kregbase) +
+				   dd->rcvtidbase +
+				   ctxttid * sizeof(*tidbase));
+
+	/* virtual address of first page in transfer */
+	vaddr = ti->tidvaddr;
+	if (!access_ok(VERIFY_WRITE, (void __user *) vaddr,
+		       cnt * PAGE_SIZE)) {
+		ret = -EFAULT;
+		goto done;
+	}
+	ret = qib_get_user_pages(vaddr, cnt, pagep);
+	if (ret) {
+		/*
+		 * if (ret == -EBUSY)
+		 * We can't continue because the pagep array won't be
+		 * initialized. This should never happen,
+		 * unless perhaps the user has mpin'ed the pages
+		 * themselves.
+		 */
+		qib_devinfo(dd->pcidev,
+			 "Failed to lock addr %p, %u pages: "
+			 "errno %d\n", (void *) vaddr, cnt, -ret);
+		goto done;
+	}
+	for (i = 0; i < cnt; i++, vaddr += PAGE_SIZE) {
+		for (; ntids--; tid++) {
+			if (tid == tidcnt)
+				tid = 0;
+			if (!dd->pageshadow[ctxttid + tid])
+				break;
+		}
+		if (ntids < 0) {
+			/*
+			 * Oops, wrapped all the way through their TIDs,
+			 * and didn't have enough free; see comments at
+			 * start of routine
+			 */
+			i--;    /* last tidlist[i] not filled in */
+			ret = -ENOMEM;
+			break;
+		}
+		tidlist[i] = tid + tidoff;
+		/* we "know" system pages and TID pages are same size */
+		dd->pageshadow[ctxttid + tid] = pagep[i];
+		dd->physshadow[ctxttid + tid] =
+			qib_map_page(dd->pcidev, pagep[i], 0, PAGE_SIZE,
+				     PCI_DMA_FROMDEVICE);
+		/*
+		 * don't need atomic or it's overhead
+		 */
+		__set_bit(tid, tidmap);
+		physaddr = dd->physshadow[ctxttid + tid];
+		/* PERFORMANCE: below should almost certainly be cached */
+		dd->f_put_tid(dd, &tidbase[tid],
+				  RCVHQ_RCV_TYPE_EXPECTED, physaddr);
+		/*
+		 * don't check this tid in qib_ctxtshadow, since we
+		 * just filled it in; start with the next one.
+		 */
+		tid++;
+	}
+
+	if (ret) {
+		u32 limit;
+cleanup:
+		/* jump here if copy out of updated info failed... */
+		/* same code that's in qib_free_tid() */
+		limit = sizeof(tidmap) * BITS_PER_BYTE;
+		if (limit > tidcnt)
+			/* just in case size changes in future */
+			limit = tidcnt;
+		tid = find_first_bit((const unsigned long *)tidmap, limit);
+		for (; tid < limit; tid++) {
+			if (!test_bit(tid, tidmap))
+				continue;
+			if (dd->pageshadow[ctxttid + tid]) {
+				dma_addr_t phys;
+
+				phys = dd->physshadow[ctxttid + tid];
+				dd->physshadow[ctxttid + tid] = dd->tidinvalid;
+				/* PERFORMANCE: below should almost certainly
+				 * be cached
+				 */
+				dd->f_put_tid(dd, &tidbase[tid],
+					      RCVHQ_RCV_TYPE_EXPECTED,
+					      dd->tidinvalid);
+				pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+					       PCI_DMA_FROMDEVICE);
+				dd->pageshadow[ctxttid + tid] = NULL;
+			}
+		}
+		qib_release_user_pages(pagep, cnt);
+	} else {
+		/*
+		 * Copy the updated array, with qib_tid's filled in, back
+		 * to user.  Since we did the copy in already, this "should
+		 * never fail" If it does, we have to clean up...
+		 */
+		if (copy_to_user((void __user *)
+				 (unsigned long) ti->tidlist,
+				 tidlist, cnt * sizeof(*tidlist))) {
+			ret = -EFAULT;
+			goto cleanup;
+		}
+		if (copy_to_user((void __user *) (unsigned long) ti->tidmap,
+				 tidmap, sizeof tidmap)) {
+			ret = -EFAULT;
+			goto cleanup;
+		}
+		if (tid == tidcnt)
+			tid = 0;
+		if (!rcd->subctxt_cnt)
+			rcd->tidcursor = tid;
+		else
+			tidcursor_fp(fp) = tid;
+	}
+
+done:
+	return ret;
+}
+
+/**
+ * qib_tid_free - free a context TID
+ * @rcd: the context
+ * @subctxt: the subcontext
+ * @ti: the TID info
+ *
+ * right now we are unlocking one page at a time, but since
+ * the intended use of this routine is for a single group of
+ * virtually contiguous pages, that should change to improve
+ * performance.  We check that the TID is in range for this context
+ * but otherwise don't check validity; if user has an error and
+ * frees the wrong tid, it's only their own data that can thereby
+ * be corrupted.  We do check that the TID was in use, for sanity
+ * We always use our idea of the saved address, not the address that
+ * they pass in to us.
+ */
+static int qib_tid_free(struct qib_ctxtdata *rcd, unsigned subctxt,
+			const struct qib_tid_info *ti)
+{
+	int ret = 0;
+	u32 tid, ctxttid, cnt, limit, tidcnt;
+	struct qib_devdata *dd = rcd->dd;
+	u64 __iomem *tidbase;
+	unsigned long tidmap[8];
+
+	if (!dd->pageshadow) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	if (copy_from_user(tidmap, (void __user *)(unsigned long)ti->tidmap,
+			   sizeof tidmap)) {
+		ret = -EFAULT;
+		goto done;
+	}
+
+	ctxttid = rcd->ctxt * dd->rcvtidcnt;
+	if (!rcd->subctxt_cnt)
+		tidcnt = dd->rcvtidcnt;
+	else if (!subctxt) {
+		tidcnt = (dd->rcvtidcnt / rcd->subctxt_cnt) +
+			 (dd->rcvtidcnt % rcd->subctxt_cnt);
+		ctxttid += dd->rcvtidcnt - tidcnt;
+	} else {
+		tidcnt = dd->rcvtidcnt / rcd->subctxt_cnt;
+		ctxttid += tidcnt * (subctxt - 1);
+	}
+	tidbase = (u64 __iomem *) ((char __iomem *)(dd->kregbase) +
+				   dd->rcvtidbase +
+				   ctxttid * sizeof(*tidbase));
+
+	limit = sizeof(tidmap) * BITS_PER_BYTE;
+	if (limit > tidcnt)
+		/* just in case size changes in future */
+		limit = tidcnt;
+	tid = find_first_bit(tidmap, limit);
+	for (cnt = 0; tid < limit; tid++) {
+		/*
+		 * small optimization; if we detect a run of 3 or so without
+		 * any set, use find_first_bit again.  That's mainly to
+		 * accelerate the case where we wrapped, so we have some at
+		 * the beginning, and some at the end, and a big gap
+		 * in the middle.
+		 */
+		if (!test_bit(tid, tidmap))
+			continue;
+		cnt++;
+		if (dd->pageshadow[ctxttid + tid]) {
+			struct page *p;
+			dma_addr_t phys;
+
+			p = dd->pageshadow[ctxttid + tid];
+			dd->pageshadow[ctxttid + tid] = NULL;
+			phys = dd->physshadow[ctxttid + tid];
+			dd->physshadow[ctxttid + tid] = dd->tidinvalid;
+			/* PERFORMANCE: below should almost certainly be
+			 * cached
+			 */
+			dd->f_put_tid(dd, &tidbase[tid],
+				      RCVHQ_RCV_TYPE_EXPECTED, dd->tidinvalid);
+			pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+				       PCI_DMA_FROMDEVICE);
+			qib_release_user_pages(&p, 1);
+		}
+	}
+done:
+	return ret;
+}
+
+/**
+ * qib_set_part_key - set a partition key
+ * @rcd: the context
+ * @key: the key
+ *
+ * We can have up to 4 active at a time (other than the default, which is
+ * always allowed).  This is somewhat tricky, since multiple contexts may set
+ * the same key, so we reference count them, and clean up at exit.  All 4
+ * partition keys are packed into a single qlogic_ib register.  It's an
+ * error for a process to set the same pkey multiple times.  We provide no
+ * mechanism to de-allocate a pkey at this time, we may eventually need to
+ * do that.  I've used the atomic operations, and no locking, and only make
+ * a single pass through what's available.  This should be more than
+ * adequate for some time. I'll think about spinlocks or the like if and as
+ * it's necessary.
+ */
+static int qib_set_part_key(struct qib_ctxtdata *rcd, u16 key)
+{
+	struct qib_pportdata *ppd = rcd->ppd;
+	int i, any = 0, pidx = -1;
+	u16 lkey = key & 0x7FFF;
+	int ret;
+
+	if (lkey == (QIB_DEFAULT_P_KEY & 0x7FFF)) {
+		/* nothing to do; this key always valid */
+		ret = 0;
+		goto bail;
+	}
+
+	if (!lkey) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Set the full membership bit, because it has to be
+	 * set in the register or the packet, and it seems
+	 * cleaner to set in the register than to force all
+	 * callers to set it.
+	 */
+	key |= 0x8000;
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		if (!rcd->pkeys[i] && pidx == -1)
+			pidx = i;
+		if (rcd->pkeys[i] == key) {
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (pidx == -1) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (any = i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i]) {
+			any++;
+			continue;
+		}
+		if (ppd->pkeys[i] == key) {
+			atomic_t *pkrefs = &ppd->pkeyrefs[i];
+
+			if (atomic_inc_return(pkrefs) > 1) {
+				rcd->pkeys[pidx] = key;
+				ret = 0;
+				goto bail;
+			} else {
+				/*
+				 * lost race, decrement count, catch below
+				 */
+				atomic_dec(pkrefs);
+				any++;
+			}
+		}
+		if ((ppd->pkeys[i] & 0x7FFF) == lkey) {
+			/*
+			 * It makes no sense to have both the limited and
+			 * full membership PKEY set at the same time since
+			 * the unlimited one will disable the limited one.
+			 */
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (!any) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (any = i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i] &&
+		    atomic_inc_return(&ppd->pkeyrefs[i]) == 1) {
+			rcd->pkeys[pidx] = key;
+			ppd->pkeys[i] = key;
+			(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+			ret = 0;
+			goto bail;
+		}
+	}
+	ret = -EBUSY;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_manage_rcvq - manage a context's receive queue
+ * @rcd: the context
+ * @subctxt: the subcontext
+ * @start_stop: action to carry out
+ *
+ * start_stop == 0 disables receive on the context, for use in queue
+ * overflow conditions.  start_stop==1 re-enables, to be used to
+ * re-init the software copy of the head register
+ */
+static int qib_manage_rcvq(struct qib_ctxtdata *rcd, unsigned subctxt,
+			   int start_stop)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned int rcvctrl_op;
+
+	if (subctxt)
+		goto bail;
+	/* atomically clear receive enable ctxt. */
+	if (start_stop) {
+		/*
+		 * On enable, force in-memory copy of the tail register to
+		 * 0, so that protocol code doesn't have to worry about
+		 * whether or not the chip has yet updated the in-memory
+		 * copy or not on return from the system call. The chip
+		 * always resets it's tail register back to 0 on a
+		 * transition from disabled to enabled.
+		 */
+		if (rcd->rcvhdrtail_kvaddr)
+			qib_clear_rcvhdrtail(rcd);
+		rcvctrl_op = QIB_RCVCTRL_CTXT_ENB;
+	} else
+		rcvctrl_op = QIB_RCVCTRL_CTXT_DIS;
+	dd->f_rcvctrl(rcd->ppd, rcvctrl_op, rcd->ctxt);
+	/* always; new head should be equal to new tail; see above */
+bail:
+	return 0;
+}
+
+static void qib_clean_part_key(struct qib_ctxtdata *rcd,
+			       struct qib_devdata *dd)
+{
+	int i, j, pchanged = 0;
+	u64 oldpkey;
+	struct qib_pportdata *ppd = rcd->ppd;
+
+	/* for debugging only */
+	oldpkey = (u64) ppd->pkeys[0] |
+		((u64) ppd->pkeys[1] << 16) |
+		((u64) ppd->pkeys[2] << 32) |
+		((u64) ppd->pkeys[3] << 48);
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		if (!rcd->pkeys[i])
+			continue;
+		for (j = 0; j < ARRAY_SIZE(ppd->pkeys); j++) {
+			/* check for match independent of the global bit */
+			if ((ppd->pkeys[j] & 0x7fff) !=
+			    (rcd->pkeys[i] & 0x7fff))
+				continue;
+			if (atomic_dec_and_test(&ppd->pkeyrefs[j])) {
+				ppd->pkeys[j] = 0;
+				pchanged++;
+			}
+			break;
+		}
+		rcd->pkeys[i] = 0;
+	}
+	if (pchanged)
+		(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+}
+
+/* common code for the mappings on dma_alloc_coherent mem */
+static int qib_mmap_mem(struct vm_area_struct *vma, struct qib_ctxtdata *rcd,
+			unsigned len, void *kvaddr, u32 write_ok, char *what)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned long pfn;
+	int ret;
+
+	if ((vma->vm_end - vma->vm_start) > len) {
+		qib_devinfo(dd->pcidev,
+			 "FAIL on %s: len %lx > %x\n", what,
+			 vma->vm_end - vma->vm_start, len);
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	/*
+	 * shared context user code requires rcvhdrq mapped r/w, others
+	 * only allowed readonly mapping.
+	 */
+	if (!write_ok) {
+		if (vma->vm_flags & VM_WRITE) {
+			qib_devinfo(dd->pcidev,
+				 "%s must be mapped readonly\n", what);
+			ret = -EPERM;
+			goto bail;
+		}
+
+		/* don't allow them to later change with mprotect */
+		vma->vm_flags &= ~VM_MAYWRITE;
+	}
+
+	pfn = virt_to_phys(kvaddr) >> PAGE_SHIFT;
+	ret = remap_pfn_range(vma, vma->vm_start, pfn,
+			      len, vma->vm_page_prot);
+	if (ret)
+		qib_devinfo(dd->pcidev,
+			"%s ctxt%u mmap of %lx, %x bytes failed: %d\n",
+			what, rcd->ctxt, pfn, len, ret);
+bail:
+	return ret;
+}
+
+static int mmap_ureg(struct vm_area_struct *vma, struct qib_devdata *dd,
+		     u64 ureg)
+{
+	unsigned long phys;
+	unsigned long sz;
+	int ret;
+
+	/*
+	 * This is real hardware, so use io_remap.  This is the mechanism
+	 * for the user process to update the head registers for their ctxt
+	 * in the chip.
+	 */
+	sz = dd->flags & QIB_HAS_HDRSUPP ? 2 * PAGE_SIZE : PAGE_SIZE;
+	if ((vma->vm_end - vma->vm_start) > sz) {
+		qib_devinfo(dd->pcidev,
+			"FAIL mmap userreg: reqlen %lx > PAGE\n",
+			vma->vm_end - vma->vm_start);
+		ret = -EFAULT;
+	} else {
+		phys = dd->physaddr + ureg;
+		vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+
+		vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+		ret = io_remap_pfn_range(vma, vma->vm_start,
+					 phys >> PAGE_SHIFT,
+					 vma->vm_end - vma->vm_start,
+					 vma->vm_page_prot);
+	}
+	return ret;
+}
+
+static int mmap_piobufs(struct vm_area_struct *vma,
+			struct qib_devdata *dd,
+			struct qib_ctxtdata *rcd,
+			unsigned piobufs, unsigned piocnt)
+{
+	unsigned long phys;
+	int ret;
+
+	/*
+	 * When we map the PIO buffers in the chip, we want to map them as
+	 * writeonly, no read possible; unfortunately, x86 doesn't allow
+	 * for this in hardware, but we still prevent users from asking
+	 * for it.
+	 */
+	if ((vma->vm_end - vma->vm_start) > (piocnt * dd->palign)) {
+		qib_devinfo(dd->pcidev,
+			"FAIL mmap piobufs: reqlen %lx > PAGE\n",
+			 vma->vm_end - vma->vm_start);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	phys = dd->physaddr + piobufs;
+
+#if defined(__powerpc__)
+	/* There isn't a generic way to specify writethrough mappings */
+	pgprot_val(vma->vm_page_prot) |= _PAGE_NO_CACHE;
+	pgprot_val(vma->vm_page_prot) |= _PAGE_WRITETHRU;
+	pgprot_val(vma->vm_page_prot) &= ~_PAGE_GUARDED;
+#endif
+
+	/*
+	 * don't allow them to later change to readable with mprotect (for when
+	 * not initially mapped readable, as is normally the case)
+	 */
+	vma->vm_flags &= ~VM_MAYREAD;
+	vma->vm_flags |= VM_DONTCOPY | VM_DONTEXPAND;
+
+	if (qib_wc_pat)
+		vma->vm_page_prot = pgprot_writecombine(vma->vm_page_prot);
+
+	ret = io_remap_pfn_range(vma, vma->vm_start, phys >> PAGE_SHIFT,
+				 vma->vm_end - vma->vm_start,
+				 vma->vm_page_prot);
+bail:
+	return ret;
+}
+
+static int mmap_rcvegrbufs(struct vm_area_struct *vma,
+			   struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned long start, size;
+	size_t total_size, i;
+	unsigned long pfn;
+	int ret;
+
+	size = rcd->rcvegrbuf_size;
+	total_size = rcd->rcvegrbuf_chunks * size;
+	if ((vma->vm_end - vma->vm_start) > total_size) {
+		qib_devinfo(dd->pcidev,
+			"FAIL on egr bufs: reqlen %lx > actual %lx\n",
+			 vma->vm_end - vma->vm_start,
+			 (unsigned long) total_size);
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (vma->vm_flags & VM_WRITE) {
+		qib_devinfo(dd->pcidev,
+			"Can't map eager buffers as writable (flags=%lx)\n",
+			vma->vm_flags);
+		ret = -EPERM;
+		goto bail;
+	}
+	/* don't allow them to later change to writeable with mprotect */
+	vma->vm_flags &= ~VM_MAYWRITE;
+
+	start = vma->vm_start;
+
+	for (i = 0; i < rcd->rcvegrbuf_chunks; i++, start += size) {
+		pfn = virt_to_phys(rcd->rcvegrbuf[i]) >> PAGE_SHIFT;
+		ret = remap_pfn_range(vma, start, pfn, size,
+				      vma->vm_page_prot);
+		if (ret < 0)
+			goto bail;
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/*
+ * qib_file_vma_fault - handle a VMA page fault.
+ */
+static int qib_file_vma_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
+{
+	struct page *page;
+
+	page = vmalloc_to_page((void *)(vmf->pgoff << PAGE_SHIFT));
+	if (!page)
+		return VM_FAULT_SIGBUS;
+
+	get_page(page);
+	vmf->page = page;
+
+	return 0;
+}
+
+static struct vm_operations_struct qib_file_vm_ops = {
+	.fault = qib_file_vma_fault,
+};
+
+static int mmap_kvaddr(struct vm_area_struct *vma, u64 pgaddr,
+		       struct qib_ctxtdata *rcd, unsigned subctxt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned subctxt_cnt;
+	unsigned long len;
+	void *addr;
+	size_t size;
+	int ret = 0;
+
+	subctxt_cnt = rcd->subctxt_cnt;
+	size = rcd->rcvegrbuf_chunks * rcd->rcvegrbuf_size;
+
+	/*
+	 * Each process has all the subctxt uregbase, rcvhdrq, and
+	 * rcvegrbufs mmapped - as an array for all the processes,
+	 * and also separately for this process.
+	 */
+	if (pgaddr == cvt_kvaddr(rcd->subctxt_uregbase)) {
+		addr = rcd->subctxt_uregbase;
+		size = PAGE_SIZE * subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvhdr_base)) {
+		addr = rcd->subctxt_rcvhdr_base;
+		size = rcd->rcvhdrq_size * subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvegrbuf)) {
+		addr = rcd->subctxt_rcvegrbuf;
+		size *= subctxt_cnt;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_uregbase +
+					PAGE_SIZE * subctxt)) {
+		addr = rcd->subctxt_uregbase + PAGE_SIZE * subctxt;
+		size = PAGE_SIZE;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvhdr_base +
+					rcd->rcvhdrq_size * subctxt)) {
+		addr = rcd->subctxt_rcvhdr_base +
+			rcd->rcvhdrq_size * subctxt;
+		size = rcd->rcvhdrq_size;
+	} else if (pgaddr == cvt_kvaddr(&rcd->user_event_mask[subctxt])) {
+		addr = rcd->user_event_mask;
+		size = PAGE_SIZE;
+	} else if (pgaddr == cvt_kvaddr(rcd->subctxt_rcvegrbuf +
+					size * subctxt)) {
+		addr = rcd->subctxt_rcvegrbuf + size * subctxt;
+		/* rcvegrbufs are read-only on the slave */
+		if (vma->vm_flags & VM_WRITE) {
+			qib_devinfo(dd->pcidev,
+				 "Can't map eager buffers as "
+				 "writable (flags=%lx)\n", vma->vm_flags);
+			ret = -EPERM;
+			goto bail;
+		}
+		/*
+		 * Don't allow permission to later change to writeable
+		 * with mprotect.
+		 */
+		vma->vm_flags &= ~VM_MAYWRITE;
+	} else
+		goto bail;
+	len = vma->vm_end - vma->vm_start;
+	if (len > size) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	vma->vm_pgoff = (unsigned long) addr >> PAGE_SHIFT;
+	vma->vm_ops = &qib_file_vm_ops;
+	vma->vm_flags |= VM_DONTEXPAND | VM_DONTDUMP;
+	ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_mmapf - mmap various structures into user space
+ * @fp: the file pointer
+ * @vma: the VM area
+ *
+ * We use this to have a shared buffer between the kernel and the user code
+ * for the rcvhdr queue, egr buffers, and the per-context user regs and pio
+ * buffers in the chip.  We have the open and close entries so we can bump
+ * the ref count and keep the driver from being unloaded while still mapped.
+ */
+static int qib_mmapf(struct file *fp, struct vm_area_struct *vma)
+{
+	struct qib_ctxtdata *rcd;
+	struct qib_devdata *dd;
+	u64 pgaddr, ureg;
+	unsigned piobufs, piocnt;
+	int ret, match = 1;
+
+	rcd = ctxt_fp(fp);
+	if (!rcd || !(vma->vm_flags & VM_SHARED)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	dd = rcd->dd;
+
+	/*
+	 * This is the qib_do_user_init() code, mapping the shared buffers
+	 * and per-context user registers into the user process. The address
+	 * referred to by vm_pgoff is the file offset passed via mmap().
+	 * For shared contexts, this is the kernel vmalloc() address of the
+	 * pages to share with the master.
+	 * For non-shared or master ctxts, this is a physical address.
+	 * We only do one mmap for each space mapped.
+	 */
+	pgaddr = vma->vm_pgoff << PAGE_SHIFT;
+
+	/*
+	 * Check for 0 in case one of the allocations failed, but user
+	 * called mmap anyway.
+	 */
+	if (!pgaddr)  {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Physical addresses must fit in 40 bits for our hardware.
+	 * Check for kernel virtual addresses first, anything else must
+	 * match a HW or memory address.
+	 */
+	ret = mmap_kvaddr(vma, pgaddr, rcd, subctxt_fp(fp));
+	if (ret) {
+		if (ret > 0)
+			ret = 0;
+		goto bail;
+	}
+
+	ureg = dd->uregbase + dd->ureg_align * rcd->ctxt;
+	if (!rcd->subctxt_cnt) {
+		/* ctxt is not shared */
+		piocnt = rcd->piocnt;
+		piobufs = rcd->piobufs;
+	} else if (!subctxt_fp(fp)) {
+		/* caller is the master */
+		piocnt = (rcd->piocnt / rcd->subctxt_cnt) +
+			 (rcd->piocnt % rcd->subctxt_cnt);
+		piobufs = rcd->piobufs +
+			dd->palign * (rcd->piocnt - piocnt);
+	} else {
+		unsigned slave = subctxt_fp(fp) - 1;
+
+		/* caller is a slave */
+		piocnt = rcd->piocnt / rcd->subctxt_cnt;
+		piobufs = rcd->piobufs + dd->palign * piocnt * slave;
+	}
+
+	if (pgaddr == ureg)
+		ret = mmap_ureg(vma, dd, ureg);
+	else if (pgaddr == piobufs)
+		ret = mmap_piobufs(vma, dd, rcd, piobufs, piocnt);
+	else if (pgaddr == dd->pioavailregs_phys)
+		/* in-memory copy of pioavail registers */
+		ret = qib_mmap_mem(vma, rcd, PAGE_SIZE,
+				   (void *) dd->pioavailregs_dma, 0,
+				   "pioavail registers");
+	else if (pgaddr == rcd->rcvegr_phys)
+		ret = mmap_rcvegrbufs(vma, rcd);
+	else if (pgaddr == (u64) rcd->rcvhdrq_phys)
+		/*
+		 * The rcvhdrq itself; multiple pages, contiguous
+		 * from an i/o perspective.  Shared contexts need
+		 * to map r/w, so we allow writing.
+		 */
+		ret = qib_mmap_mem(vma, rcd, rcd->rcvhdrq_size,
+				   rcd->rcvhdrq, 1, "rcvhdrq");
+	else if (pgaddr == (u64) rcd->rcvhdrqtailaddr_phys)
+		/* in-memory copy of rcvhdrq tail register */
+		ret = qib_mmap_mem(vma, rcd, PAGE_SIZE,
+				   rcd->rcvhdrtail_kvaddr, 0,
+				   "rcvhdrq tail");
+	else
+		match = 0;
+	if (!match)
+		ret = -EINVAL;
+
+	vma->vm_private_data = NULL;
+
+	if (ret < 0)
+		qib_devinfo(dd->pcidev,
+			 "mmap Failure %d: off %llx len %lx\n",
+			 -ret, (unsigned long long)pgaddr,
+			 vma->vm_end - vma->vm_start);
+bail:
+	return ret;
+}
+
+static unsigned int qib_poll_urgent(struct qib_ctxtdata *rcd,
+				    struct file *fp,
+				    struct poll_table_struct *pt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned pollflag;
+
+	poll_wait(fp, &rcd->wait, pt);
+
+	spin_lock_irq(&dd->uctxt_lock);
+	if (rcd->urgent != rcd->urgent_poll) {
+		pollflag = POLLIN | POLLRDNORM;
+		rcd->urgent_poll = rcd->urgent;
+	} else {
+		pollflag = 0;
+		set_bit(QIB_CTXT_WAITING_URG, &rcd->flag);
+	}
+	spin_unlock_irq(&dd->uctxt_lock);
+
+	return pollflag;
+}
+
+static unsigned int qib_poll_next(struct qib_ctxtdata *rcd,
+				  struct file *fp,
+				  struct poll_table_struct *pt)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned pollflag;
+
+	poll_wait(fp, &rcd->wait, pt);
+
+	spin_lock_irq(&dd->uctxt_lock);
+	if (dd->f_hdrqempty(rcd)) {
+		set_bit(QIB_CTXT_WAITING_RCV, &rcd->flag);
+		dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_INTRAVAIL_ENB, rcd->ctxt);
+		pollflag = 0;
+	} else
+		pollflag = POLLIN | POLLRDNORM;
+	spin_unlock_irq(&dd->uctxt_lock);
+
+	return pollflag;
+}
+
+static unsigned int qib_poll(struct file *fp, struct poll_table_struct *pt)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned pollflag;
+
+	rcd = ctxt_fp(fp);
+	if (!rcd)
+		pollflag = POLLERR;
+	else if (rcd->poll_type == QIB_POLL_TYPE_URGENT)
+		pollflag = qib_poll_urgent(rcd, fp, pt);
+	else  if (rcd->poll_type == QIB_POLL_TYPE_ANYRCV)
+		pollflag = qib_poll_next(rcd, fp, pt);
+	else /* invalid */
+		pollflag = POLLERR;
+
+	return pollflag;
+}
+
+static void assign_ctxt_affinity(struct file *fp, struct qib_devdata *dd)
+{
+	struct qib_filedata *fd = fp->private_data;
+	const unsigned int weight = cpumask_weight(&current->cpus_allowed);
+	const struct cpumask *local_mask = cpumask_of_pcibus(dd->pcidev->bus);
+	int local_cpu;
+
+	/*
+	 * If process has NOT already set it's affinity, select and
+	 * reserve a processor for it on the local NUMA node.
+	 */
+	if ((weight >= qib_cpulist_count) &&
+		(cpumask_weight(local_mask) <= qib_cpulist_count)) {
+		for_each_cpu(local_cpu, local_mask)
+			if (!test_and_set_bit(local_cpu, qib_cpulist)) {
+				fd->rec_cpu_num = local_cpu;
+				return;
+			}
+	}
+
+	/*
+	 * If process has NOT already set it's affinity, select and
+	 * reserve a processor for it, as a rendevous for all
+	 * users of the driver.  If they don't actually later
+	 * set affinity to this cpu, or set it to some other cpu,
+	 * it just means that sooner or later we don't recommend
+	 * a cpu, and let the scheduler do it's best.
+	 */
+	if (weight >= qib_cpulist_count) {
+		int cpu;
+		cpu = find_first_zero_bit(qib_cpulist,
+					  qib_cpulist_count);
+		if (cpu == qib_cpulist_count)
+			qib_dev_err(dd,
+			"no cpus avail for affinity PID %u\n",
+			current->pid);
+		else {
+			__set_bit(cpu, qib_cpulist);
+			fd->rec_cpu_num = cpu;
+		}
+	}
+}
+
+/*
+ * Check that userland and driver are compatible for subcontexts.
+ */
+static int qib_compatible_subctxts(int user_swmajor, int user_swminor)
+{
+	/* this code is written long-hand for clarity */
+	if (QIB_USER_SWMAJOR != user_swmajor) {
+		/* no promise of compatibility if major mismatch */
+		return 0;
+	}
+	if (QIB_USER_SWMAJOR == 1) {
+		switch (QIB_USER_SWMINOR) {
+		case 0:
+		case 1:
+		case 2:
+			/* no subctxt implementation so cannot be compatible */
+			return 0;
+		case 3:
+			/* 3 is only compatible with itself */
+			return user_swminor == 3;
+		default:
+			/* >= 4 are compatible (or are expected to be) */
+			return user_swminor <= QIB_USER_SWMINOR;
+		}
+	}
+	/* make no promises yet for future major versions */
+	return 0;
+}
+
+static int init_subctxts(struct qib_devdata *dd,
+			 struct qib_ctxtdata *rcd,
+			 const struct qib_user_info *uinfo)
+{
+	int ret = 0;
+	unsigned num_subctxts;
+	size_t size;
+
+	/*
+	 * If the user is requesting zero subctxts,
+	 * skip the subctxt allocation.
+	 */
+	if (uinfo->spu_subctxt_cnt <= 0)
+		goto bail;
+	num_subctxts = uinfo->spu_subctxt_cnt;
+
+	/* Check for subctxt compatibility */
+	if (!qib_compatible_subctxts(uinfo->spu_userversion >> 16,
+		uinfo->spu_userversion & 0xffff)) {
+		qib_devinfo(dd->pcidev,
+			 "Mismatched user version (%d.%d) and driver "
+			 "version (%d.%d) while context sharing. Ensure "
+			 "that driver and library are from the same "
+			 "release.\n",
+			 (int) (uinfo->spu_userversion >> 16),
+			 (int) (uinfo->spu_userversion & 0xffff),
+			 QIB_USER_SWMAJOR, QIB_USER_SWMINOR);
+		goto bail;
+	}
+	if (num_subctxts > QLOGIC_IB_MAX_SUBCTXT) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	rcd->subctxt_uregbase = vmalloc_user(PAGE_SIZE * num_subctxts);
+	if (!rcd->subctxt_uregbase) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	/* Note: rcd->rcvhdrq_size isn't initialized yet. */
+	size = ALIGN(dd->rcvhdrcnt * dd->rcvhdrentsize *
+		     sizeof(u32), PAGE_SIZE) * num_subctxts;
+	rcd->subctxt_rcvhdr_base = vmalloc_user(size);
+	if (!rcd->subctxt_rcvhdr_base) {
+		ret = -ENOMEM;
+		goto bail_ureg;
+	}
+
+	rcd->subctxt_rcvegrbuf = vmalloc_user(rcd->rcvegrbuf_chunks *
+					      rcd->rcvegrbuf_size *
+					      num_subctxts);
+	if (!rcd->subctxt_rcvegrbuf) {
+		ret = -ENOMEM;
+		goto bail_rhdr;
+	}
+
+	rcd->subctxt_cnt = uinfo->spu_subctxt_cnt;
+	rcd->subctxt_id = uinfo->spu_subctxt_id;
+	rcd->active_slaves = 1;
+	rcd->redirect_seq_cnt = 1;
+	set_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag);
+	goto bail;
+
+bail_rhdr:
+	vfree(rcd->subctxt_rcvhdr_base);
+bail_ureg:
+	vfree(rcd->subctxt_uregbase);
+	rcd->subctxt_uregbase = NULL;
+bail:
+	return ret;
+}
+
+static int setup_ctxt(struct qib_pportdata *ppd, int ctxt,
+		      struct file *fp, const struct qib_user_info *uinfo)
+{
+	struct qib_filedata *fd = fp->private_data;
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	void *ptmp = NULL;
+	int ret;
+	int numa_id;
+
+	assign_ctxt_affinity(fp, dd);
+
+	numa_id = qib_numa_aware ? ((fd->rec_cpu_num != -1) ?
+		cpu_to_node(fd->rec_cpu_num) :
+		numa_node_id()) : dd->assigned_node_id;
+
+	rcd = qib_create_ctxtdata(ppd, ctxt, numa_id);
+
+	/*
+	 * Allocate memory for use in qib_tid_update() at open to
+	 * reduce cost of expected send setup per message segment
+	 */
+	if (rcd)
+		ptmp = kmalloc(dd->rcvtidcnt * sizeof(u16) +
+			       dd->rcvtidcnt * sizeof(struct page **),
+			       GFP_KERNEL);
+
+	if (!rcd || !ptmp) {
+		qib_dev_err(dd,
+			"Unable to allocate ctxtdata memory, failing open\n");
+		ret = -ENOMEM;
+		goto bailerr;
+	}
+	rcd->userversion = uinfo->spu_userversion;
+	ret = init_subctxts(dd, rcd, uinfo);
+	if (ret)
+		goto bailerr;
+	rcd->tid_pg_list = ptmp;
+	rcd->pid = current->pid;
+	init_waitqueue_head(&dd->rcd[ctxt]->wait);
+	strlcpy(rcd->comm, current->comm, sizeof(rcd->comm));
+	ctxt_fp(fp) = rcd;
+	qib_stats.sps_ctxts++;
+	dd->freectxts--;
+	ret = 0;
+	goto bail;
+
+bailerr:
+	if (fd->rec_cpu_num != -1)
+		__clear_bit(fd->rec_cpu_num, qib_cpulist);
+
+	dd->rcd[ctxt] = NULL;
+	kfree(rcd);
+	kfree(ptmp);
+bail:
+	return ret;
+}
+
+static inline int usable(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+
+	return dd && (dd->flags & QIB_PRESENT) && dd->kregbase && ppd->lid &&
+		(ppd->lflags & QIBL_LINKACTIVE);
+}
+
+/*
+ * Select a context on the given device, either using a requested port
+ * or the port based on the context number.
+ */
+static int choose_port_ctxt(struct file *fp, struct qib_devdata *dd, u32 port,
+			    const struct qib_user_info *uinfo)
+{
+	struct qib_pportdata *ppd = NULL;
+	int ret, ctxt;
+
+	if (port) {
+		if (!usable(dd->pport + port - 1)) {
+			ret = -ENETDOWN;
+			goto done;
+		} else
+			ppd = dd->pport + port - 1;
+	}
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts && dd->rcd[ctxt];
+	     ctxt++)
+		;
+	if (ctxt == dd->cfgctxts) {
+		ret = -EBUSY;
+		goto done;
+	}
+	if (!ppd) {
+		u32 pidx = ctxt % dd->num_pports;
+		if (usable(dd->pport + pidx))
+			ppd = dd->pport + pidx;
+		else {
+			for (pidx = 0; pidx < dd->num_pports && !ppd;
+			     pidx++)
+				if (usable(dd->pport + pidx))
+					ppd = dd->pport + pidx;
+		}
+	}
+	ret = ppd ? setup_ctxt(ppd, ctxt, fp, uinfo) : -ENETDOWN;
+done:
+	return ret;
+}
+
+static int find_free_ctxt(int unit, struct file *fp,
+			  const struct qib_user_info *uinfo)
+{
+	struct qib_devdata *dd = qib_lookup(unit);
+	int ret;
+
+	if (!dd || (uinfo->spu_port && uinfo->spu_port > dd->num_pports))
+		ret = -ENODEV;
+	else
+		ret = choose_port_ctxt(fp, dd, uinfo->spu_port, uinfo);
+
+	return ret;
+}
+
+static int get_a_ctxt(struct file *fp, const struct qib_user_info *uinfo,
+		      unsigned alg)
+{
+	struct qib_devdata *udd = NULL;
+	int ret = 0, devmax, npresent, nup, ndev, dusable = 0, i;
+	u32 port = uinfo->spu_port, ctxt;
+
+	devmax = qib_count_units(&npresent, &nup);
+	if (!npresent) {
+		ret = -ENXIO;
+		goto done;
+	}
+	if (nup == 0) {
+		ret = -ENETDOWN;
+		goto done;
+	}
+
+	if (alg == QIB_PORT_ALG_ACROSS) {
+		unsigned inuse = ~0U;
+		/* find device (with ACTIVE ports) with fewest ctxts in use */
+		for (ndev = 0; ndev < devmax; ndev++) {
+			struct qib_devdata *dd = qib_lookup(ndev);
+			unsigned cused = 0, cfree = 0, pusable = 0;
+			if (!dd)
+				continue;
+			if (port && port <= dd->num_pports &&
+			    usable(dd->pport + port - 1))
+				pusable = 1;
+			else
+				for (i = 0; i < dd->num_pports; i++)
+					if (usable(dd->pport + i))
+						pusable++;
+			if (!pusable)
+				continue;
+			for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts;
+			     ctxt++)
+				if (dd->rcd[ctxt])
+					cused++;
+				else
+					cfree++;
+			if (cfree && cused < inuse) {
+				udd = dd;
+				inuse = cused;
+			}
+		}
+		if (udd) {
+			ret = choose_port_ctxt(fp, udd, port, uinfo);
+			goto done;
+		}
+	} else {
+		for (ndev = 0; ndev < devmax; ndev++) {
+			struct qib_devdata *dd = qib_lookup(ndev);
+			if (dd) {
+				ret = choose_port_ctxt(fp, dd, port, uinfo);
+				if (!ret)
+					goto done;
+				if (ret == -EBUSY)
+					dusable++;
+			}
+		}
+	}
+	ret = dusable ? -EBUSY : -ENETDOWN;
+
+done:
+	return ret;
+}
+
+static int find_shared_ctxt(struct file *fp,
+			    const struct qib_user_info *uinfo)
+{
+	int devmax, ndev, i;
+	int ret = 0;
+
+	devmax = qib_count_units(NULL, NULL);
+
+	for (ndev = 0; ndev < devmax; ndev++) {
+		struct qib_devdata *dd = qib_lookup(ndev);
+
+		/* device portion of usable() */
+		if (!(dd && (dd->flags & QIB_PRESENT) && dd->kregbase))
+			continue;
+		for (i = dd->first_user_ctxt; i < dd->cfgctxts; i++) {
+			struct qib_ctxtdata *rcd = dd->rcd[i];
+
+			/* Skip ctxts which are not yet open */
+			if (!rcd || !rcd->cnt)
+				continue;
+			/* Skip ctxt if it doesn't match the requested one */
+			if (rcd->subctxt_id != uinfo->spu_subctxt_id)
+				continue;
+			/* Verify the sharing process matches the master */
+			if (rcd->subctxt_cnt != uinfo->spu_subctxt_cnt ||
+			    rcd->userversion != uinfo->spu_userversion ||
+			    rcd->cnt >= rcd->subctxt_cnt) {
+				ret = -EINVAL;
+				goto done;
+			}
+			ctxt_fp(fp) = rcd;
+			subctxt_fp(fp) = rcd->cnt++;
+			rcd->subpid[subctxt_fp(fp)] = current->pid;
+			tidcursor_fp(fp) = 0;
+			rcd->active_slaves |= 1 << subctxt_fp(fp);
+			ret = 1;
+			goto done;
+		}
+	}
+
+done:
+	return ret;
+}
+
+static int qib_open(struct inode *in, struct file *fp)
+{
+	/* The real work is performed later in qib_assign_ctxt() */
+	fp->private_data = kzalloc(sizeof(struct qib_filedata), GFP_KERNEL);
+	if (fp->private_data) /* no cpu affinity by default */
+		((struct qib_filedata *)fp->private_data)->rec_cpu_num = -1;
+	return fp->private_data ? 0 : -ENOMEM;
+}
+
+static int find_hca(unsigned int cpu, int *unit)
+{
+	int ret = 0, devmax, npresent, nup, ndev;
+
+	*unit = -1;
+
+	devmax = qib_count_units(&npresent, &nup);
+	if (!npresent) {
+		ret = -ENXIO;
+		goto done;
+	}
+	if (!nup) {
+		ret = -ENETDOWN;
+		goto done;
+	}
+	for (ndev = 0; ndev < devmax; ndev++) {
+		struct qib_devdata *dd = qib_lookup(ndev);
+		if (dd) {
+			if (pcibus_to_node(dd->pcidev->bus) < 0) {
+				ret = -EINVAL;
+				goto done;
+			}
+			if (cpu_to_node(cpu) ==
+				pcibus_to_node(dd->pcidev->bus)) {
+				*unit = ndev;
+				goto done;
+			}
+		}
+	}
+done:
+	return ret;
+}
+
+static int do_qib_user_sdma_queue_create(struct file *fp)
+{
+	struct qib_filedata *fd = fp->private_data;
+	struct qib_ctxtdata *rcd = fd->rcd;
+	struct qib_devdata *dd = rcd->dd;
+
+	if (dd->flags & QIB_HAS_SEND_DMA) {
+
+		fd->pq = qib_user_sdma_queue_create(&dd->pcidev->dev,
+						    dd->unit,
+						    rcd->ctxt,
+						    fd->subctxt);
+		if (!fd->pq)
+			return -ENOMEM;
+	}
+
+	return 0;
+}
+
+/*
+ * Get ctxt early, so can set affinity prior to memory allocation.
+ */
+static int qib_assign_ctxt(struct file *fp, const struct qib_user_info *uinfo)
+{
+	int ret;
+	int i_minor;
+	unsigned swmajor, swminor, alg = QIB_PORT_ALG_ACROSS;
+
+	/* Check to be sure we haven't already initialized this file */
+	if (ctxt_fp(fp)) {
+		ret = -EINVAL;
+		goto done;
+	}
+
+	/* for now, if major version is different, bail */
+	swmajor = uinfo->spu_userversion >> 16;
+	if (swmajor != QIB_USER_SWMAJOR) {
+		ret = -ENODEV;
+		goto done;
+	}
+
+	swminor = uinfo->spu_userversion & 0xffff;
+
+	if (swminor >= 11 && uinfo->spu_port_alg < QIB_PORT_ALG_COUNT)
+		alg = uinfo->spu_port_alg;
+
+	mutex_lock(&qib_mutex);
+
+	if (qib_compatible_subctxts(swmajor, swminor) &&
+	    uinfo->spu_subctxt_cnt) {
+		ret = find_shared_ctxt(fp, uinfo);
+		if (ret > 0) {
+			ret = do_qib_user_sdma_queue_create(fp);
+			if (!ret)
+				assign_ctxt_affinity(fp, (ctxt_fp(fp))->dd);
+			goto done_ok;
+		}
+	}
+
+	i_minor = iminor(file_inode(fp)) - QIB_USER_MINOR_BASE;
+	if (i_minor)
+		ret = find_free_ctxt(i_minor - 1, fp, uinfo);
+	else {
+		int unit;
+		const unsigned int cpu = cpumask_first(&current->cpus_allowed);
+		const unsigned int weight =
+			cpumask_weight(&current->cpus_allowed);
+
+		if (weight == 1 && !test_bit(cpu, qib_cpulist))
+			if (!find_hca(cpu, &unit) && unit >= 0)
+				if (!find_free_ctxt(unit, fp, uinfo)) {
+					ret = 0;
+					goto done_chk_sdma;
+				}
+		ret = get_a_ctxt(fp, uinfo, alg);
+	}
+
+done_chk_sdma:
+	if (!ret)
+		ret = do_qib_user_sdma_queue_create(fp);
+done_ok:
+	mutex_unlock(&qib_mutex);
+
+done:
+	return ret;
+}
+
+
+static int qib_do_user_init(struct file *fp,
+			    const struct qib_user_info *uinfo)
+{
+	int ret;
+	struct qib_ctxtdata *rcd = ctxt_fp(fp);
+	struct qib_devdata *dd;
+	unsigned uctxt;
+
+	/* Subctxts don't need to initialize anything since master did it. */
+	if (subctxt_fp(fp)) {
+		ret = wait_event_interruptible(rcd->wait,
+			!test_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag));
+		goto bail;
+	}
+
+	dd = rcd->dd;
+
+	/* some ctxts may get extra buffers, calculate that here */
+	uctxt = rcd->ctxt - dd->first_user_ctxt;
+	if (uctxt < dd->ctxts_extrabuf) {
+		rcd->piocnt = dd->pbufsctxt + 1;
+		rcd->pio_base = rcd->piocnt * uctxt;
+	} else {
+		rcd->piocnt = dd->pbufsctxt;
+		rcd->pio_base = rcd->piocnt * uctxt +
+			dd->ctxts_extrabuf;
+	}
+
+	/*
+	 * All user buffers are 2KB buffers.  If we ever support
+	 * giving 4KB buffers to user processes, this will need some
+	 * work.  Can't use piobufbase directly, because it has
+	 * both 2K and 4K buffer base values.  So check and handle.
+	 */
+	if ((rcd->pio_base + rcd->piocnt) > dd->piobcnt2k) {
+		if (rcd->pio_base >= dd->piobcnt2k) {
+			qib_dev_err(dd,
+				    "%u:ctxt%u: no 2KB buffers available\n",
+				    dd->unit, rcd->ctxt);
+			ret = -ENOBUFS;
+			goto bail;
+		}
+		rcd->piocnt = dd->piobcnt2k - rcd->pio_base;
+		qib_dev_err(dd, "Ctxt%u: would use 4KB bufs, using %u\n",
+			    rcd->ctxt, rcd->piocnt);
+	}
+
+	rcd->piobufs = dd->pio2k_bufbase + rcd->pio_base * dd->palign;
+	qib_chg_pioavailkernel(dd, rcd->pio_base, rcd->piocnt,
+			       TXCHK_CHG_TYPE_USER, rcd);
+	/*
+	 * try to ensure that processes start up with consistent avail update
+	 * for their own range, at least.   If system very quiet, it might
+	 * have the in-memory copy out of date at startup for this range of
+	 * buffers, when a context gets re-used.  Do after the chg_pioavail
+	 * and before the rest of setup, so it's "almost certain" the dma
+	 * will have occurred (can't 100% guarantee, but should be many
+	 * decimals of 9s, with this ordering), given how much else happens
+	 * after this.
+	 */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+
+	/*
+	 * Now allocate the rcvhdr Q and eager TIDs; skip the TID
+	 * array for time being.  If rcd->ctxt > chip-supported,
+	 * we need to do extra stuff here to handle by handling overflow
+	 * through ctxt 0, someday
+	 */
+	ret = qib_create_rcvhdrq(dd, rcd);
+	if (!ret)
+		ret = qib_setup_eagerbufs(rcd);
+	if (ret)
+		goto bail_pio;
+
+	rcd->tidcursor = 0; /* start at beginning after open */
+
+	/* initialize poll variables... */
+	rcd->urgent = 0;
+	rcd->urgent_poll = 0;
+
+	/*
+	 * Now enable the ctxt for receive.
+	 * For chips that are set to DMA the tail register to memory
+	 * when they change (and when the update bit transitions from
+	 * 0 to 1.  So for those chips, we turn it off and then back on.
+	 * This will (very briefly) affect any other open ctxts, but the
+	 * duration is very short, and therefore isn't an issue.  We
+	 * explicitly set the in-memory tail copy to 0 beforehand, so we
+	 * don't have to wait to be sure the DMA update has happened
+	 * (chip resets head/tail to 0 on transition to enable).
+	 */
+	if (rcd->rcvhdrtail_kvaddr)
+		qib_clear_rcvhdrtail(rcd);
+
+	dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_CTXT_ENB | QIB_RCVCTRL_TIDFLOW_ENB,
+		      rcd->ctxt);
+
+	/* Notify any waiting slaves */
+	if (rcd->subctxt_cnt) {
+		clear_bit(QIB_CTXT_MASTER_UNINIT, &rcd->flag);
+		wake_up(&rcd->wait);
+	}
+	return 0;
+
+bail_pio:
+	qib_chg_pioavailkernel(dd, rcd->pio_base, rcd->piocnt,
+			       TXCHK_CHG_TYPE_KERN, rcd);
+bail:
+	return ret;
+}
+
+/**
+ * unlock_exptid - unlock any expected TID entries context still had in use
+ * @rcd: ctxt
+ *
+ * We don't actually update the chip here, because we do a bulk update
+ * below, using f_clear_tids.
+ */
+static void unlock_expected_tids(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	int ctxt_tidbase = rcd->ctxt * dd->rcvtidcnt;
+	int i, cnt = 0, maxtid = ctxt_tidbase + dd->rcvtidcnt;
+
+	for (i = ctxt_tidbase; i < maxtid; i++) {
+		struct page *p = dd->pageshadow[i];
+		dma_addr_t phys;
+
+		if (!p)
+			continue;
+
+		phys = dd->physshadow[i];
+		dd->physshadow[i] = dd->tidinvalid;
+		dd->pageshadow[i] = NULL;
+		pci_unmap_page(dd->pcidev, phys, PAGE_SIZE,
+			       PCI_DMA_FROMDEVICE);
+		qib_release_user_pages(&p, 1);
+		cnt++;
+	}
+}
+
+static int qib_close(struct inode *in, struct file *fp)
+{
+	int ret = 0;
+	struct qib_filedata *fd;
+	struct qib_ctxtdata *rcd;
+	struct qib_devdata *dd;
+	unsigned long flags;
+	unsigned ctxt;
+	pid_t pid;
+
+	mutex_lock(&qib_mutex);
+
+	fd = fp->private_data;
+	fp->private_data = NULL;
+	rcd = fd->rcd;
+	if (!rcd) {
+		mutex_unlock(&qib_mutex);
+		goto bail;
+	}
+
+	dd = rcd->dd;
+
+	/* ensure all pio buffer writes in progress are flushed */
+	qib_flush_wc();
+
+	/* drain user sdma queue */
+	if (fd->pq) {
+		qib_user_sdma_queue_drain(rcd->ppd, fd->pq);
+		qib_user_sdma_queue_destroy(fd->pq);
+	}
+
+	if (fd->rec_cpu_num != -1)
+		__clear_bit(fd->rec_cpu_num, qib_cpulist);
+
+	if (--rcd->cnt) {
+		/*
+		 * XXX If the master closes the context before the slave(s),
+		 * revoke the mmap for the eager receive queue so
+		 * the slave(s) don't wait for receive data forever.
+		 */
+		rcd->active_slaves &= ~(1 << fd->subctxt);
+		rcd->subpid[fd->subctxt] = 0;
+		mutex_unlock(&qib_mutex);
+		goto bail;
+	}
+
+	/* early; no interrupt users after this */
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	ctxt = rcd->ctxt;
+	dd->rcd[ctxt] = NULL;
+	pid = rcd->pid;
+	rcd->pid = 0;
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+
+	if (rcd->rcvwait_to || rcd->piowait_to ||
+	    rcd->rcvnowait || rcd->pionowait) {
+		rcd->rcvwait_to = 0;
+		rcd->piowait_to = 0;
+		rcd->rcvnowait = 0;
+		rcd->pionowait = 0;
+	}
+	if (rcd->flag)
+		rcd->flag = 0;
+
+	if (dd->kregbase) {
+		/* atomically clear receive enable ctxt and intr avail. */
+		dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_CTXT_DIS |
+				  QIB_RCVCTRL_INTRAVAIL_DIS, ctxt);
+
+		/* clean up the pkeys for this ctxt user */
+		qib_clean_part_key(rcd, dd);
+		qib_disarm_piobufs(dd, rcd->pio_base, rcd->piocnt);
+		qib_chg_pioavailkernel(dd, rcd->pio_base,
+				       rcd->piocnt, TXCHK_CHG_TYPE_KERN, NULL);
+
+		dd->f_clear_tids(dd, rcd);
+
+		if (dd->pageshadow)
+			unlock_expected_tids(rcd);
+		qib_stats.sps_ctxts--;
+		dd->freectxts++;
+	}
+
+	mutex_unlock(&qib_mutex);
+	qib_free_ctxtdata(dd, rcd); /* after releasing the mutex */
+
+bail:
+	kfree(fd);
+	return ret;
+}
+
+static int qib_ctxt_info(struct file *fp, struct qib_ctxt_info __user *uinfo)
+{
+	struct qib_ctxt_info info;
+	int ret;
+	size_t sz;
+	struct qib_ctxtdata *rcd = ctxt_fp(fp);
+	struct qib_filedata *fd;
+
+	fd = fp->private_data;
+
+	info.num_active = qib_count_active_units();
+	info.unit = rcd->dd->unit;
+	info.port = rcd->ppd->port;
+	info.ctxt = rcd->ctxt;
+	info.subctxt =  subctxt_fp(fp);
+	/* Number of user ctxts available for this device. */
+	info.num_ctxts = rcd->dd->cfgctxts - rcd->dd->first_user_ctxt;
+	info.num_subctxts = rcd->subctxt_cnt;
+	info.rec_cpu = fd->rec_cpu_num;
+	sz = sizeof(info);
+
+	if (copy_to_user(uinfo, &info, sz)) {
+		ret = -EFAULT;
+		goto bail;
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+static int qib_sdma_get_inflight(struct qib_user_sdma_queue *pq,
+				 u32 __user *inflightp)
+{
+	const u32 val = qib_user_sdma_inflight_counter(pq);
+
+	if (put_user(val, inflightp))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int qib_sdma_get_complete(struct qib_pportdata *ppd,
+				 struct qib_user_sdma_queue *pq,
+				 u32 __user *completep)
+{
+	u32 val;
+	int err;
+
+	if (!pq)
+		return -EINVAL;
+
+	err = qib_user_sdma_make_progress(ppd, pq);
+	if (err < 0)
+		return err;
+
+	val = qib_user_sdma_complete_counter(pq);
+	if (put_user(val, completep))
+		return -EFAULT;
+
+	return 0;
+}
+
+static int disarm_req_delay(struct qib_ctxtdata *rcd)
+{
+	int ret = 0;
+
+	if (!usable(rcd->ppd)) {
+		int i;
+		/*
+		 * if link is down, or otherwise not usable, delay
+		 * the caller up to 30 seconds, so we don't thrash
+		 * in trying to get the chip back to ACTIVE, and
+		 * set flag so they make the call again.
+		 */
+		if (rcd->user_event_mask) {
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				&rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[i]);
+		}
+		for (i = 0; !usable(rcd->ppd) && i < 300; i++)
+			msleep(100);
+		ret = -ENETDOWN;
+	}
+	return ret;
+}
+
+/*
+ * Find all user contexts in use, and set the specified bit in their
+ * event mask.
+ * See also find_ctxt() for a similar use, that is specific to send buffers.
+ */
+int qib_set_uevent_bits(struct qib_pportdata *ppd, const int evtbit)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned ctxt;
+	int ret = 0;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->dd->uctxt_lock, flags);
+	for (ctxt = ppd->dd->first_user_ctxt; ctxt < ppd->dd->cfgctxts;
+	     ctxt++) {
+		rcd = ppd->dd->rcd[ctxt];
+		if (!rcd)
+			continue;
+		if (rcd->user_event_mask) {
+			int i;
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(evtbit, &rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(evtbit, &rcd->user_event_mask[i]);
+		}
+		ret = 1;
+		break;
+	}
+	spin_unlock_irqrestore(&ppd->dd->uctxt_lock, flags);
+
+	return ret;
+}
+
+/*
+ * clear the event notifier events for this context.
+ * For the DISARM_BUFS case, we also take action (this obsoletes
+ * the older QIB_CMD_DISARM_BUFS, but we keep it for backwards
+ * compatibility.
+ * Other bits don't currently require actions, just atomically clear.
+ * User process then performs actions appropriate to bit having been
+ * set, if desired, and checks again in future.
+ */
+static int qib_user_event_ack(struct qib_ctxtdata *rcd, int subctxt,
+			      unsigned long events)
+{
+	int ret = 0, i;
+
+	for (i = 0; i <= _QIB_MAX_EVENT_BIT; i++) {
+		if (!test_bit(i, &events))
+			continue;
+		if (i == _QIB_EVENT_DISARM_BUFS_BIT) {
+			(void)qib_disarm_piobufs_ifneeded(rcd);
+			ret = disarm_req_delay(rcd);
+		} else
+			clear_bit(i, &rcd->user_event_mask[subctxt]);
+	}
+	return ret;
+}
+
+static ssize_t qib_write(struct file *fp, const char __user *data,
+			 size_t count, loff_t *off)
+{
+	const struct qib_cmd __user *ucmd;
+	struct qib_ctxtdata *rcd;
+	const void __user *src;
+	size_t consumed, copy = 0;
+	struct qib_cmd cmd;
+	ssize_t ret = 0;
+	void *dest;
+
+	if (count < sizeof(cmd.type)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	ucmd = (const struct qib_cmd __user *) data;
+
+	if (copy_from_user(&cmd.type, &ucmd->type, sizeof(cmd.type))) {
+		ret = -EFAULT;
+		goto bail;
+	}
+
+	consumed = sizeof(cmd.type);
+
+	switch (cmd.type) {
+	case QIB_CMD_ASSIGN_CTXT:
+	case QIB_CMD_USER_INIT:
+		copy = sizeof(cmd.cmd.user_info);
+		dest = &cmd.cmd.user_info;
+		src = &ucmd->cmd.user_info;
+		break;
+
+	case QIB_CMD_RECV_CTRL:
+		copy = sizeof(cmd.cmd.recv_ctrl);
+		dest = &cmd.cmd.recv_ctrl;
+		src = &ucmd->cmd.recv_ctrl;
+		break;
+
+	case QIB_CMD_CTXT_INFO:
+		copy = sizeof(cmd.cmd.ctxt_info);
+		dest = &cmd.cmd.ctxt_info;
+		src = &ucmd->cmd.ctxt_info;
+		break;
+
+	case QIB_CMD_TID_UPDATE:
+	case QIB_CMD_TID_FREE:
+		copy = sizeof(cmd.cmd.tid_info);
+		dest = &cmd.cmd.tid_info;
+		src = &ucmd->cmd.tid_info;
+		break;
+
+	case QIB_CMD_SET_PART_KEY:
+		copy = sizeof(cmd.cmd.part_key);
+		dest = &cmd.cmd.part_key;
+		src = &ucmd->cmd.part_key;
+		break;
+
+	case QIB_CMD_DISARM_BUFS:
+	case QIB_CMD_PIOAVAILUPD: /* force an update of PIOAvail reg */
+		copy = 0;
+		src = NULL;
+		dest = NULL;
+		break;
+
+	case QIB_CMD_POLL_TYPE:
+		copy = sizeof(cmd.cmd.poll_type);
+		dest = &cmd.cmd.poll_type;
+		src = &ucmd->cmd.poll_type;
+		break;
+
+	case QIB_CMD_ARMLAUNCH_CTRL:
+		copy = sizeof(cmd.cmd.armlaunch_ctrl);
+		dest = &cmd.cmd.armlaunch_ctrl;
+		src = &ucmd->cmd.armlaunch_ctrl;
+		break;
+
+	case QIB_CMD_SDMA_INFLIGHT:
+		copy = sizeof(cmd.cmd.sdma_inflight);
+		dest = &cmd.cmd.sdma_inflight;
+		src = &ucmd->cmd.sdma_inflight;
+		break;
+
+	case QIB_CMD_SDMA_COMPLETE:
+		copy = sizeof(cmd.cmd.sdma_complete);
+		dest = &cmd.cmd.sdma_complete;
+		src = &ucmd->cmd.sdma_complete;
+		break;
+
+	case QIB_CMD_ACK_EVENT:
+		copy = sizeof(cmd.cmd.event_mask);
+		dest = &cmd.cmd.event_mask;
+		src = &ucmd->cmd.event_mask;
+		break;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (copy) {
+		if ((count - consumed) < copy) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		if (copy_from_user(dest, src, copy)) {
+			ret = -EFAULT;
+			goto bail;
+		}
+		consumed += copy;
+	}
+
+	rcd = ctxt_fp(fp);
+	if (!rcd && cmd.type != QIB_CMD_ASSIGN_CTXT) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	switch (cmd.type) {
+	case QIB_CMD_ASSIGN_CTXT:
+		ret = qib_assign_ctxt(fp, &cmd.cmd.user_info);
+		if (ret)
+			goto bail;
+		break;
+
+	case QIB_CMD_USER_INIT:
+		ret = qib_do_user_init(fp, &cmd.cmd.user_info);
+		if (ret)
+			goto bail;
+		ret = qib_get_base_info(fp, (void __user *) (unsigned long)
+					cmd.cmd.user_info.spu_base_info,
+					cmd.cmd.user_info.spu_base_info_size);
+		break;
+
+	case QIB_CMD_RECV_CTRL:
+		ret = qib_manage_rcvq(rcd, subctxt_fp(fp), cmd.cmd.recv_ctrl);
+		break;
+
+	case QIB_CMD_CTXT_INFO:
+		ret = qib_ctxt_info(fp, (struct qib_ctxt_info __user *)
+				    (unsigned long) cmd.cmd.ctxt_info);
+		break;
+
+	case QIB_CMD_TID_UPDATE:
+		ret = qib_tid_update(rcd, fp, &cmd.cmd.tid_info);
+		break;
+
+	case QIB_CMD_TID_FREE:
+		ret = qib_tid_free(rcd, subctxt_fp(fp), &cmd.cmd.tid_info);
+		break;
+
+	case QIB_CMD_SET_PART_KEY:
+		ret = qib_set_part_key(rcd, cmd.cmd.part_key);
+		break;
+
+	case QIB_CMD_DISARM_BUFS:
+		(void)qib_disarm_piobufs_ifneeded(rcd);
+		ret = disarm_req_delay(rcd);
+		break;
+
+	case QIB_CMD_PIOAVAILUPD:
+		qib_force_pio_avail_update(rcd->dd);
+		break;
+
+	case QIB_CMD_POLL_TYPE:
+		rcd->poll_type = cmd.cmd.poll_type;
+		break;
+
+	case QIB_CMD_ARMLAUNCH_CTRL:
+		rcd->dd->f_set_armlaunch(rcd->dd, cmd.cmd.armlaunch_ctrl);
+		break;
+
+	case QIB_CMD_SDMA_INFLIGHT:
+		ret = qib_sdma_get_inflight(user_sdma_queue_fp(fp),
+					    (u32 __user *) (unsigned long)
+					    cmd.cmd.sdma_inflight);
+		break;
+
+	case QIB_CMD_SDMA_COMPLETE:
+		ret = qib_sdma_get_complete(rcd->ppd,
+					    user_sdma_queue_fp(fp),
+					    (u32 __user *) (unsigned long)
+					    cmd.cmd.sdma_complete);
+		break;
+
+	case QIB_CMD_ACK_EVENT:
+		ret = qib_user_event_ack(rcd, subctxt_fp(fp),
+					 cmd.cmd.event_mask);
+		break;
+	}
+
+	if (ret >= 0)
+		ret = consumed;
+
+bail:
+	return ret;
+}
+
+static ssize_t qib_aio_write(struct kiocb *iocb, const struct iovec *iov,
+			     unsigned long dim, loff_t off)
+{
+	struct qib_filedata *fp = iocb->ki_filp->private_data;
+	struct qib_ctxtdata *rcd = ctxt_fp(iocb->ki_filp);
+	struct qib_user_sdma_queue *pq = fp->pq;
+
+	if (!dim || !pq)
+		return -EINVAL;
+
+	return qib_user_sdma_writev(rcd, pq, iov, dim);
+}
+
+static struct class *qib_class;
+static dev_t qib_dev;
+
+int qib_cdev_init(int minor, const char *name,
+		  const struct file_operations *fops,
+		  struct cdev **cdevp, struct device **devp)
+{
+	const dev_t dev = MKDEV(MAJOR(qib_dev), minor);
+	struct cdev *cdev;
+	struct device *device = NULL;
+	int ret;
+
+	cdev = cdev_alloc();
+	if (!cdev) {
+		pr_err("Could not allocate cdev for minor %d, %s\n",
+		       minor, name);
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	cdev->owner = THIS_MODULE;
+	cdev->ops = fops;
+	kobject_set_name(&cdev->kobj, name);
+
+	ret = cdev_add(cdev, dev, 1);
+	if (ret < 0) {
+		pr_err("Could not add cdev for minor %d, %s (err %d)\n",
+		       minor, name, -ret);
+		goto err_cdev;
+	}
+
+	device = device_create(qib_class, NULL, dev, NULL, "%s", name);
+	if (!IS_ERR(device))
+		goto done;
+	ret = PTR_ERR(device);
+	device = NULL;
+	pr_err("Could not create device for minor %d, %s (err %d)\n",
+	       minor, name, -ret);
+err_cdev:
+	cdev_del(cdev);
+	cdev = NULL;
+done:
+	*cdevp = cdev;
+	*devp = device;
+	return ret;
+}
+
+void qib_cdev_cleanup(struct cdev **cdevp, struct device **devp)
+{
+	struct device *device = *devp;
+
+	if (device) {
+		device_unregister(device);
+		*devp = NULL;
+	}
+
+	if (*cdevp) {
+		cdev_del(*cdevp);
+		*cdevp = NULL;
+	}
+}
+
+static struct cdev *wildcard_cdev;
+static struct device *wildcard_device;
+
+int __init qib_dev_init(void)
+{
+	int ret;
+
+	ret = alloc_chrdev_region(&qib_dev, 0, QIB_NMINORS, QIB_DRV_NAME);
+	if (ret < 0) {
+		pr_err("Could not allocate chrdev region (err %d)\n", -ret);
+		goto done;
+	}
+
+	qib_class = class_create(THIS_MODULE, "ipath");
+	if (IS_ERR(qib_class)) {
+		ret = PTR_ERR(qib_class);
+		pr_err("Could not create device class (err %d)\n", -ret);
+		unregister_chrdev_region(qib_dev, QIB_NMINORS);
+	}
+
+done:
+	return ret;
+}
+
+void qib_dev_cleanup(void)
+{
+	if (qib_class) {
+		class_destroy(qib_class);
+		qib_class = NULL;
+	}
+
+	unregister_chrdev_region(qib_dev, QIB_NMINORS);
+}
+
+static atomic_t user_count = ATOMIC_INIT(0);
+
+static void qib_user_remove(struct qib_devdata *dd)
+{
+	if (atomic_dec_return(&user_count) == 0)
+		qib_cdev_cleanup(&wildcard_cdev, &wildcard_device);
+
+	qib_cdev_cleanup(&dd->user_cdev, &dd->user_device);
+}
+
+static int qib_user_add(struct qib_devdata *dd)
+{
+	char name[10];
+	int ret;
+
+	if (atomic_inc_return(&user_count) == 1) {
+		ret = qib_cdev_init(0, "ipath", &qib_file_ops,
+				    &wildcard_cdev, &wildcard_device);
+		if (ret)
+			goto done;
+	}
+
+	snprintf(name, sizeof(name), "ipath%d", dd->unit);
+	ret = qib_cdev_init(dd->unit + 1, name, &qib_file_ops,
+			    &dd->user_cdev, &dd->user_device);
+	if (ret)
+		qib_user_remove(dd);
+done:
+	return ret;
+}
+
+/*
+ * Create per-unit files in /dev
+ */
+int qib_device_create(struct qib_devdata *dd)
+{
+	int r, ret;
+
+	r = qib_user_add(dd);
+	ret = qib_diag_add(dd);
+	if (r && !ret)
+		ret = r;
+	return ret;
+}
+
+/*
+ * Remove per-unit files in /dev
+ * void, core kernel returns no errors for this stuff
+ */
+void qib_device_remove(struct qib_devdata *dd)
+{
+	qib_user_remove(dd);
+	qib_diag_remove(dd);
+}
diff --git a/drivers/infiniband/hw/qib/qib_fs.c b/drivers/infiniband/hw/qib/qib_fs.c
new file mode 100644
index 00000000000..cab610ccd50
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_fs.c
@@ -0,0 +1,621 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/fs.h>
+#include <linux/mount.h>
+#include <linux/pagemap.h>
+#include <linux/init.h>
+#include <linux/namei.h>
+
+#include "qib.h"
+
+#define QIBFS_MAGIC 0x726a77
+
+static struct super_block *qib_super;
+
+#define private2dd(file) (file_inode(file)->i_private)
+
+static int qibfs_mknod(struct inode *dir, struct dentry *dentry,
+		       umode_t mode, const struct file_operations *fops,
+		       void *data)
+{
+	int error;
+	struct inode *inode = new_inode(dir->i_sb);
+
+	if (!inode) {
+		error = -EPERM;
+		goto bail;
+	}
+
+	inode->i_ino = get_next_ino();
+	inode->i_mode = mode;
+	inode->i_uid = GLOBAL_ROOT_UID;
+	inode->i_gid = GLOBAL_ROOT_GID;
+	inode->i_blocks = 0;
+	inode->i_atime = CURRENT_TIME;
+	inode->i_mtime = inode->i_atime;
+	inode->i_ctime = inode->i_atime;
+	inode->i_private = data;
+	if (S_ISDIR(mode)) {
+		inode->i_op = &simple_dir_inode_operations;
+		inc_nlink(inode);
+		inc_nlink(dir);
+	}
+
+	inode->i_fop = fops;
+
+	d_instantiate(dentry, inode);
+	error = 0;
+
+bail:
+	return error;
+}
+
+static int create_file(const char *name, umode_t mode,
+		       struct dentry *parent, struct dentry **dentry,
+		       const struct file_operations *fops, void *data)
+{
+	int error;
+
+	*dentry = NULL;
+	mutex_lock(&parent->d_inode->i_mutex);
+	*dentry = lookup_one_len(name, parent, strlen(name));
+	if (!IS_ERR(*dentry))
+		error = qibfs_mknod(parent->d_inode, *dentry,
+				    mode, fops, data);
+	else
+		error = PTR_ERR(*dentry);
+	mutex_unlock(&parent->d_inode->i_mutex);
+
+	return error;
+}
+
+static ssize_t driver_stats_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	qib_stats.sps_ints = qib_sps_ints();
+	return simple_read_from_buffer(buf, count, ppos, &qib_stats,
+				       sizeof qib_stats);
+}
+
+/*
+ * driver stats field names, one line per stat, single string.  Used by
+ * programs like ipathstats to print the stats in a way which works for
+ * different versions of drivers, without changing program source.
+ * if qlogic_ib_stats changes, this needs to change.  Names need to be
+ * 12 chars or less (w/o newline), for proper display by ipathstats utility.
+ */
+static const char qib_statnames[] =
+	"KernIntr\n"
+	"ErrorIntr\n"
+	"Tx_Errs\n"
+	"Rcv_Errs\n"
+	"H/W_Errs\n"
+	"NoPIOBufs\n"
+	"CtxtsOpen\n"
+	"RcvLen_Errs\n"
+	"EgrBufFull\n"
+	"EgrHdrFull\n"
+	;
+
+static ssize_t driver_names_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	return simple_read_from_buffer(buf, count, ppos, qib_statnames,
+		sizeof qib_statnames - 1); /* no null */
+}
+
+static const struct file_operations driver_ops[] = {
+	{ .read = driver_stats_read, .llseek = generic_file_llseek, },
+	{ .read = driver_names_read, .llseek = generic_file_llseek, },
+};
+
+/* read the per-device counters */
+static ssize_t dev_counters_read(struct file *file, char __user *buf,
+				 size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	size_t avail;
+	struct qib_devdata *dd = private2dd(file);
+
+	avail = dd->f_read_cntrs(dd, *ppos, NULL, &counters);
+	return simple_read_from_buffer(buf, count, ppos, counters, avail);
+}
+
+/* read the per-device counters */
+static ssize_t dev_names_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	size_t avail;
+	struct qib_devdata *dd = private2dd(file);
+
+	avail = dd->f_read_cntrs(dd, *ppos, &names, NULL);
+	return simple_read_from_buffer(buf, count, ppos, names, avail);
+}
+
+static const struct file_operations cntr_ops[] = {
+	{ .read = dev_counters_read, .llseek = generic_file_llseek, },
+	{ .read = dev_names_read, .llseek = generic_file_llseek, },
+};
+
+/*
+ * Could use file_inode(file)->i_ino to figure out which file,
+ * instead of separate routine for each, but for now, this works...
+ */
+
+/* read the per-port names (same for each port) */
+static ssize_t portnames_read(struct file *file, char __user *buf,
+			      size_t count, loff_t *ppos)
+{
+	char *names;
+	size_t avail;
+	struct qib_devdata *dd = private2dd(file);
+
+	avail = dd->f_read_portcntrs(dd, *ppos, 0, &names, NULL);
+	return simple_read_from_buffer(buf, count, ppos, names, avail);
+}
+
+/* read the per-port counters for port 1 (pidx 0) */
+static ssize_t portcntrs_1_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	size_t avail;
+	struct qib_devdata *dd = private2dd(file);
+
+	avail = dd->f_read_portcntrs(dd, *ppos, 0, NULL, &counters);
+	return simple_read_from_buffer(buf, count, ppos, counters, avail);
+}
+
+/* read the per-port counters for port 2 (pidx 1) */
+static ssize_t portcntrs_2_read(struct file *file, char __user *buf,
+				size_t count, loff_t *ppos)
+{
+	u64 *counters;
+	size_t avail;
+	struct qib_devdata *dd = private2dd(file);
+
+	avail = dd->f_read_portcntrs(dd, *ppos, 1, NULL, &counters);
+	return simple_read_from_buffer(buf, count, ppos, counters, avail);
+}
+
+static const struct file_operations portcntr_ops[] = {
+	{ .read = portnames_read, .llseek = generic_file_llseek, },
+	{ .read = portcntrs_1_read, .llseek = generic_file_llseek, },
+	{ .read = portcntrs_2_read, .llseek = generic_file_llseek, },
+};
+
+/*
+ * read the per-port QSFP data for port 1 (pidx 0)
+ */
+static ssize_t qsfp_1_read(struct file *file, char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd = private2dd(file);
+	char *tmp;
+	int ret;
+
+	tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	ret = qib_qsfp_dump(dd->pport, tmp, PAGE_SIZE);
+	if (ret > 0)
+		ret = simple_read_from_buffer(buf, count, ppos, tmp, ret);
+	kfree(tmp);
+	return ret;
+}
+
+/*
+ * read the per-port QSFP data for port 2 (pidx 1)
+ */
+static ssize_t qsfp_2_read(struct file *file, char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd = private2dd(file);
+	char *tmp;
+	int ret;
+
+	if (dd->num_pports < 2)
+		return -ENODEV;
+
+	tmp = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!tmp)
+		return -ENOMEM;
+
+	ret = qib_qsfp_dump(dd->pport + 1, tmp, PAGE_SIZE);
+	if (ret > 0)
+		ret = simple_read_from_buffer(buf, count, ppos, tmp, ret);
+	kfree(tmp);
+	return ret;
+}
+
+static const struct file_operations qsfp_ops[] = {
+	{ .read = qsfp_1_read, .llseek = generic_file_llseek, },
+	{ .read = qsfp_2_read, .llseek = generic_file_llseek, },
+};
+
+static ssize_t flash_read(struct file *file, char __user *buf,
+			  size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos < 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (pos >= sizeof(struct qib_flash)) {
+		ret = 0;
+		goto bail;
+	}
+
+	if (count > sizeof(struct qib_flash) - pos)
+		count = sizeof(struct qib_flash) - pos;
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_read(dd, pos, tmp, count)) {
+		qib_dev_err(dd, "failed to read from flash\n");
+		ret = -ENXIO;
+		goto bail_tmp;
+	}
+
+	if (copy_to_user(buf, tmp, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static ssize_t flash_write(struct file *file, const char __user *buf,
+			   size_t count, loff_t *ppos)
+{
+	struct qib_devdata *dd;
+	ssize_t ret;
+	loff_t pos;
+	char *tmp;
+
+	pos = *ppos;
+
+	if (pos != 0) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	if (count != sizeof(struct qib_flash)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	tmp = kmalloc(count, GFP_KERNEL);
+	if (!tmp) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	if (copy_from_user(tmp, buf, count)) {
+		ret = -EFAULT;
+		goto bail_tmp;
+	}
+
+	dd = private2dd(file);
+	if (qib_eeprom_write(dd, pos, tmp, count)) {
+		ret = -ENXIO;
+		qib_dev_err(dd, "failed to write to flash\n");
+		goto bail_tmp;
+	}
+
+	*ppos = pos + count;
+	ret = count;
+
+bail_tmp:
+	kfree(tmp);
+
+bail:
+	return ret;
+}
+
+static const struct file_operations flash_ops = {
+	.read = flash_read,
+	.write = flash_write,
+	.llseek = default_llseek,
+};
+
+static int add_cntr_files(struct super_block *sb, struct qib_devdata *dd)
+{
+	struct dentry *dir, *tmp;
+	char unit[10];
+	int ret, i;
+
+	/* create the per-unit directory */
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	ret = create_file(unit, S_IFDIR|S_IRUGO|S_IXUGO, sb->s_root, &dir,
+			  &simple_dir_operations, dd);
+	if (ret) {
+		pr_err("create_file(%s) failed: %d\n", unit, ret);
+		goto bail;
+	}
+
+	/* create the files in the new directory */
+	ret = create_file("counters", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[0], dd);
+	if (ret) {
+		pr_err("create_file(%s/counters) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("counter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &cntr_ops[1], dd);
+	if (ret) {
+		pr_err("create_file(%s/counter_names) failed: %d\n",
+		       unit, ret);
+		goto bail;
+	}
+	ret = create_file("portcounter_names", S_IFREG|S_IRUGO, dir, &tmp,
+			  &portcntr_ops[0], dd);
+	if (ret) {
+		pr_err("create_file(%s/%s) failed: %d\n",
+		       unit, "portcounter_names", ret);
+		goto bail;
+	}
+	for (i = 1; i <= dd->num_pports; i++) {
+		char fname[24];
+
+		sprintf(fname, "port%dcounters", i);
+		/* create the files in the new directory */
+		ret = create_file(fname, S_IFREG|S_IRUGO, dir, &tmp,
+				  &portcntr_ops[i], dd);
+		if (ret) {
+			pr_err("create_file(%s/%s) failed: %d\n",
+				unit, fname, ret);
+			goto bail;
+		}
+		if (!(dd->flags & QIB_HAS_QSFP))
+			continue;
+		sprintf(fname, "qsfp%d", i);
+		ret = create_file(fname, S_IFREG|S_IRUGO, dir, &tmp,
+				  &qsfp_ops[i - 1], dd);
+		if (ret) {
+			pr_err("create_file(%s/%s) failed: %d\n",
+				unit, fname, ret);
+			goto bail;
+		}
+	}
+
+	ret = create_file("flash", S_IFREG|S_IWUSR|S_IRUGO, dir, &tmp,
+			  &flash_ops, dd);
+	if (ret)
+		pr_err("create_file(%s/flash) failed: %d\n",
+			unit, ret);
+bail:
+	return ret;
+}
+
+static int remove_file(struct dentry *parent, char *name)
+{
+	struct dentry *tmp;
+	int ret;
+
+	tmp = lookup_one_len(name, parent, strlen(name));
+
+	if (IS_ERR(tmp)) {
+		ret = PTR_ERR(tmp);
+		goto bail;
+	}
+
+	spin_lock(&tmp->d_lock);
+	if (!(d_unhashed(tmp) && tmp->d_inode)) {
+		__d_drop(tmp);
+		spin_unlock(&tmp->d_lock);
+		simple_unlink(parent->d_inode, tmp);
+	} else {
+		spin_unlock(&tmp->d_lock);
+	}
+	dput(tmp);
+
+	ret = 0;
+bail:
+	/*
+	 * We don't expect clients to care about the return value, but
+	 * it's there if they need it.
+	 */
+	return ret;
+}
+
+static int remove_device_files(struct super_block *sb,
+			       struct qib_devdata *dd)
+{
+	struct dentry *dir, *root;
+	char unit[10];
+	int ret, i;
+
+	root = dget(sb->s_root);
+	mutex_lock(&root->d_inode->i_mutex);
+	snprintf(unit, sizeof unit, "%u", dd->unit);
+	dir = lookup_one_len(unit, root, strlen(unit));
+
+	if (IS_ERR(dir)) {
+		ret = PTR_ERR(dir);
+		pr_err("Lookup of %s failed\n", unit);
+		goto bail;
+	}
+
+	mutex_lock(&dir->d_inode->i_mutex);
+	remove_file(dir, "counters");
+	remove_file(dir, "counter_names");
+	remove_file(dir, "portcounter_names");
+	for (i = 0; i < dd->num_pports; i++) {
+		char fname[24];
+
+		sprintf(fname, "port%dcounters", i + 1);
+		remove_file(dir, fname);
+		if (dd->flags & QIB_HAS_QSFP) {
+			sprintf(fname, "qsfp%d", i + 1);
+			remove_file(dir, fname);
+		}
+	}
+	remove_file(dir, "flash");
+	mutex_unlock(&dir->d_inode->i_mutex);
+	ret = simple_rmdir(root->d_inode, dir);
+	d_delete(dir);
+	dput(dir);
+
+bail:
+	mutex_unlock(&root->d_inode->i_mutex);
+	dput(root);
+	return ret;
+}
+
+/*
+ * This fills everything in when the fs is mounted, to handle umount/mount
+ * after device init.  The direct add_cntr_files() call handles adding
+ * them from the init code, when the fs is already mounted.
+ */
+static int qibfs_fill_super(struct super_block *sb, void *data, int silent)
+{
+	struct qib_devdata *dd, *tmp;
+	unsigned long flags;
+	int ret;
+
+	static struct tree_descr files[] = {
+		[2] = {"driver_stats", &driver_ops[0], S_IRUGO},
+		[3] = {"driver_stats_names", &driver_ops[1], S_IRUGO},
+		{""},
+	};
+
+	ret = simple_fill_super(sb, QIBFS_MAGIC, files);
+	if (ret) {
+		pr_err("simple_fill_super failed: %d\n", ret);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	list_for_each_entry_safe(dd, tmp, &qib_dev_list, list) {
+		spin_unlock_irqrestore(&qib_devs_lock, flags);
+		ret = add_cntr_files(sb, dd);
+		if (ret)
+			goto bail;
+		spin_lock_irqsave(&qib_devs_lock, flags);
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+bail:
+	return ret;
+}
+
+static struct dentry *qibfs_mount(struct file_system_type *fs_type, int flags,
+			const char *dev_name, void *data)
+{
+	struct dentry *ret;
+	ret = mount_single(fs_type, flags, data, qibfs_fill_super);
+	if (!IS_ERR(ret))
+		qib_super = ret->d_sb;
+	return ret;
+}
+
+static void qibfs_kill_super(struct super_block *s)
+{
+	kill_litter_super(s);
+	qib_super = NULL;
+}
+
+int qibfs_add(struct qib_devdata *dd)
+{
+	int ret;
+
+	/*
+	 * On first unit initialized, qib_super will not yet exist
+	 * because nobody has yet tried to mount the filesystem, so
+	 * we can't consider that to be an error; if an error occurs
+	 * during the mount, that will get a complaint, so this is OK.
+	 * add_cntr_files() for all units is done at mount from
+	 * qibfs_fill_super(), so one way or another, everything works.
+	 */
+	if (qib_super == NULL)
+		ret = 0;
+	else
+		ret = add_cntr_files(qib_super, dd);
+	return ret;
+}
+
+int qibfs_remove(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	if (qib_super)
+		ret = remove_device_files(qib_super, dd);
+
+	return ret;
+}
+
+static struct file_system_type qibfs_fs_type = {
+	.owner =        THIS_MODULE,
+	.name =         "ipathfs",
+	.mount =        qibfs_mount,
+	.kill_sb =      qibfs_kill_super,
+};
+MODULE_ALIAS_FS("ipathfs");
+
+int __init qib_init_qibfs(void)
+{
+	return register_filesystem(&qibfs_fs_type);
+}
+
+int __exit qib_exit_qibfs(void)
+{
+	return unregister_filesystem(&qibfs_fs_type);
+}
diff --git a/drivers/infiniband/hw/qib/qib_iba6120.c b/drivers/infiniband/hw/qib/qib_iba6120.c
new file mode 100644
index 00000000000..d68266ac761
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_iba6120.c
@@ -0,0 +1,3599 @@
+/*
+ * Copyright (c) 2013 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/*
+ * This file contains all of the code that is specific to the
+ * QLogic_IB 6120 PCIe chip.
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <rdma/ib_verbs.h>
+
+#include "qib.h"
+#include "qib_6120_regs.h"
+
+static void qib_6120_setup_setextled(struct qib_pportdata *, u32);
+static void sendctrl_6120_mod(struct qib_pportdata *ppd, u32 op);
+static u8 qib_6120_phys_portstate(u64);
+static u32 qib_6120_iblink_state(u64);
+
+/*
+ * This file contains all the chip-specific register information and
+ * access functions for the Intel Intel_IB PCI-Express chip.
+ *
+ */
+
+/* KREG_IDX uses machine-generated #defines */
+#define KREG_IDX(regname) (QIB_6120_##regname##_OFFS / sizeof(u64))
+
+/* Use defines to tie machine-generated names to lower-case names */
+#define kr_extctrl KREG_IDX(EXTCtrl)
+#define kr_extstatus KREG_IDX(EXTStatus)
+#define kr_gpio_clear KREG_IDX(GPIOClear)
+#define kr_gpio_mask KREG_IDX(GPIOMask)
+#define kr_gpio_out KREG_IDX(GPIOOut)
+#define kr_gpio_status KREG_IDX(GPIOStatus)
+#define kr_rcvctrl KREG_IDX(RcvCtrl)
+#define kr_sendctrl KREG_IDX(SendCtrl)
+#define kr_partitionkey KREG_IDX(RcvPartitionKey)
+#define kr_hwdiagctrl KREG_IDX(HwDiagCtrl)
+#define kr_ibcstatus KREG_IDX(IBCStatus)
+#define kr_ibcctrl KREG_IDX(IBCCtrl)
+#define kr_sendbuffererror KREG_IDX(SendBufErr0)
+#define kr_rcvbthqp KREG_IDX(RcvBTHQP)
+#define kr_counterregbase KREG_IDX(CntrRegBase)
+#define kr_palign KREG_IDX(PageAlign)
+#define kr_rcvegrbase KREG_IDX(RcvEgrBase)
+#define kr_rcvegrcnt KREG_IDX(RcvEgrCnt)
+#define kr_rcvhdrcnt KREG_IDX(RcvHdrCnt)
+#define kr_rcvhdrentsize KREG_IDX(RcvHdrEntSize)
+#define kr_rcvhdrsize KREG_IDX(RcvHdrSize)
+#define kr_rcvtidbase KREG_IDX(RcvTIDBase)
+#define kr_rcvtidcnt KREG_IDX(RcvTIDCnt)
+#define kr_scratch KREG_IDX(Scratch)
+#define kr_sendctrl KREG_IDX(SendCtrl)
+#define kr_sendpioavailaddr KREG_IDX(SendPIOAvailAddr)
+#define kr_sendpiobufbase KREG_IDX(SendPIOBufBase)
+#define kr_sendpiobufcnt KREG_IDX(SendPIOBufCnt)
+#define kr_sendpiosize KREG_IDX(SendPIOSize)
+#define kr_sendregbase KREG_IDX(SendRegBase)
+#define kr_userregbase KREG_IDX(UserRegBase)
+#define kr_control KREG_IDX(Control)
+#define kr_intclear KREG_IDX(IntClear)
+#define kr_intmask KREG_IDX(IntMask)
+#define kr_intstatus KREG_IDX(IntStatus)
+#define kr_errclear KREG_IDX(ErrClear)
+#define kr_errmask KREG_IDX(ErrMask)
+#define kr_errstatus KREG_IDX(ErrStatus)
+#define kr_hwerrclear KREG_IDX(HwErrClear)
+#define kr_hwerrmask KREG_IDX(HwErrMask)
+#define kr_hwerrstatus KREG_IDX(HwErrStatus)
+#define kr_revision KREG_IDX(Revision)
+#define kr_portcnt KREG_IDX(PortCnt)
+#define kr_serdes_cfg0 KREG_IDX(SerdesCfg0)
+#define kr_serdes_cfg1 (kr_serdes_cfg0 + 1)
+#define kr_serdes_stat KREG_IDX(SerdesStat)
+#define kr_xgxs_cfg KREG_IDX(XGXSCfg)
+
+/* These must only be written via qib_write_kreg_ctxt() */
+#define kr_rcvhdraddr KREG_IDX(RcvHdrAddr0)
+#define kr_rcvhdrtailaddr KREG_IDX(RcvHdrTailAddr0)
+
+#define CREG_IDX(regname) ((QIB_6120_##regname##_OFFS - \
+			QIB_6120_LBIntCnt_OFFS) / sizeof(u64))
+
+#define cr_badformat CREG_IDX(RxBadFormatCnt)
+#define cr_erricrc CREG_IDX(RxICRCErrCnt)
+#define cr_errlink CREG_IDX(RxLinkProblemCnt)
+#define cr_errlpcrc CREG_IDX(RxLPCRCErrCnt)
+#define cr_errpkey CREG_IDX(RxPKeyMismatchCnt)
+#define cr_rcvflowctrl_err CREG_IDX(RxFlowCtrlErrCnt)
+#define cr_err_rlen CREG_IDX(RxLenErrCnt)
+#define cr_errslen CREG_IDX(TxLenErrCnt)
+#define cr_errtidfull CREG_IDX(RxTIDFullErrCnt)
+#define cr_errtidvalid CREG_IDX(RxTIDValidErrCnt)
+#define cr_errvcrc CREG_IDX(RxVCRCErrCnt)
+#define cr_ibstatuschange CREG_IDX(IBStatusChangeCnt)
+#define cr_lbint CREG_IDX(LBIntCnt)
+#define cr_invalidrlen CREG_IDX(RxMaxMinLenErrCnt)
+#define cr_invalidslen CREG_IDX(TxMaxMinLenErrCnt)
+#define cr_lbflowstall CREG_IDX(LBFlowStallCnt)
+#define cr_pktrcv CREG_IDX(RxDataPktCnt)
+#define cr_pktrcvflowctrl CREG_IDX(RxFlowPktCnt)
+#define cr_pktsend CREG_IDX(TxDataPktCnt)
+#define cr_pktsendflow CREG_IDX(TxFlowPktCnt)
+#define cr_portovfl CREG_IDX(RxP0HdrEgrOvflCnt)
+#define cr_rcvebp CREG_IDX(RxEBPCnt)
+#define cr_rcvovfl CREG_IDX(RxBufOvflCnt)
+#define cr_senddropped CREG_IDX(TxDroppedPktCnt)
+#define cr_sendstall CREG_IDX(TxFlowStallCnt)
+#define cr_sendunderrun CREG_IDX(TxUnderrunCnt)
+#define cr_wordrcv CREG_IDX(RxDwordCnt)
+#define cr_wordsend CREG_IDX(TxDwordCnt)
+#define cr_txunsupvl CREG_IDX(TxUnsupVLErrCnt)
+#define cr_rxdroppkt CREG_IDX(RxDroppedPktCnt)
+#define cr_iblinkerrrecov CREG_IDX(IBLinkErrRecoveryCnt)
+#define cr_iblinkdown CREG_IDX(IBLinkDownedCnt)
+#define cr_ibsymbolerr CREG_IDX(IBSymbolErrCnt)
+
+#define SYM_RMASK(regname, fldname) ((u64)              \
+	QIB_6120_##regname##_##fldname##_RMASK)
+#define SYM_MASK(regname, fldname) ((u64)               \
+	QIB_6120_##regname##_##fldname##_RMASK <<       \
+	 QIB_6120_##regname##_##fldname##_LSB)
+#define SYM_LSB(regname, fldname) (QIB_6120_##regname##_##fldname##_LSB)
+
+#define SYM_FIELD(value, regname, fldname) ((u64) \
+	(((value) >> SYM_LSB(regname, fldname)) & \
+	 SYM_RMASK(regname, fldname)))
+#define ERR_MASK(fldname) SYM_MASK(ErrMask, fldname##Mask)
+#define HWE_MASK(fldname) SYM_MASK(HwErrMask, fldname##Mask)
+
+/* link training states, from IBC */
+#define IB_6120_LT_STATE_DISABLED        0x00
+#define IB_6120_LT_STATE_LINKUP          0x01
+#define IB_6120_LT_STATE_POLLACTIVE      0x02
+#define IB_6120_LT_STATE_POLLQUIET       0x03
+#define IB_6120_LT_STATE_SLEEPDELAY      0x04
+#define IB_6120_LT_STATE_SLEEPQUIET      0x05
+#define IB_6120_LT_STATE_CFGDEBOUNCE     0x08
+#define IB_6120_LT_STATE_CFGRCVFCFG      0x09
+#define IB_6120_LT_STATE_CFGWAITRMT      0x0a
+#define IB_6120_LT_STATE_CFGIDLE 0x0b
+#define IB_6120_LT_STATE_RECOVERRETRAIN  0x0c
+#define IB_6120_LT_STATE_RECOVERWAITRMT  0x0e
+#define IB_6120_LT_STATE_RECOVERIDLE     0x0f
+
+/* link state machine states from IBC */
+#define IB_6120_L_STATE_DOWN             0x0
+#define IB_6120_L_STATE_INIT             0x1
+#define IB_6120_L_STATE_ARM              0x2
+#define IB_6120_L_STATE_ACTIVE           0x3
+#define IB_6120_L_STATE_ACT_DEFER        0x4
+
+static const u8 qib_6120_physportstate[0x20] = {
+	[IB_6120_LT_STATE_DISABLED] = IB_PHYSPORTSTATE_DISABLED,
+	[IB_6120_LT_STATE_LINKUP] = IB_PHYSPORTSTATE_LINKUP,
+	[IB_6120_LT_STATE_POLLACTIVE] = IB_PHYSPORTSTATE_POLL,
+	[IB_6120_LT_STATE_POLLQUIET] = IB_PHYSPORTSTATE_POLL,
+	[IB_6120_LT_STATE_SLEEPDELAY] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_6120_LT_STATE_SLEEPQUIET] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_6120_LT_STATE_CFGDEBOUNCE] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_6120_LT_STATE_CFGRCVFCFG] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_6120_LT_STATE_CFGWAITRMT] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_6120_LT_STATE_CFGIDLE] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_6120_LT_STATE_RECOVERRETRAIN] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_6120_LT_STATE_RECOVERWAITRMT] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_6120_LT_STATE_RECOVERIDLE] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[0x10] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x11] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x12] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x13] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x14] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x15] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x16] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x17] = IB_PHYSPORTSTATE_CFG_TRAIN
+};
+
+
+struct qib_chip_specific {
+	u64 __iomem *cregbase;
+	u64 *cntrs;
+	u64 *portcntrs;
+	void *dummy_hdrq;   /* used after ctxt close */
+	dma_addr_t dummy_hdrq_phys;
+	spinlock_t kernel_tid_lock; /* no back to back kernel TID writes */
+	spinlock_t user_tid_lock; /* no back to back user TID writes */
+	spinlock_t rcvmod_lock; /* protect rcvctrl shadow changes */
+	spinlock_t gpio_lock; /* RMW of shadows/regs for ExtCtrl and GPIO */
+	u64 hwerrmask;
+	u64 errormask;
+	u64 gpio_out; /* shadow of kr_gpio_out, for rmw ops */
+	u64 gpio_mask; /* shadow the gpio mask register */
+	u64 extctrl; /* shadow the gpio output enable, etc... */
+	/*
+	 * these 5 fields are used to establish deltas for IB symbol
+	 * errors and linkrecovery errors.  They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+	u64 ibcctrl; /* shadow for kr_ibcctrl */
+	u32 lastlinkrecov; /* link recovery issue */
+	int irq;
+	u32 cntrnamelen;
+	u32 portcntrnamelen;
+	u32 ncntrs;
+	u32 nportcntrs;
+	/* used with gpio interrupts to implement IB counters */
+	u32 rxfc_unsupvl_errs;
+	u32 overrun_thresh_errs;
+	/*
+	 * these count only cases where _successive_ LocalLinkIntegrity
+	 * errors were seen in the receive headers of IB standard packets
+	 */
+	u32 lli_errs;
+	u32 lli_counter;
+	u64 lli_thresh;
+	u64 sword; /* total dwords sent (sample result) */
+	u64 rword; /* total dwords received (sample result) */
+	u64 spkts; /* total packets sent (sample result) */
+	u64 rpkts; /* total packets received (sample result) */
+	u64 xmit_wait; /* # of ticks no data sent (sample result) */
+	struct timer_list pma_timer;
+	char emsgbuf[128];
+	char bitsmsgbuf[64];
+	u8 pma_sample_status;
+};
+
+/* ibcctrl bits */
+#define QLOGIC_IB_IBCC_LINKINITCMD_DISABLE 1
+/* cycle through TS1/TS2 till OK */
+#define QLOGIC_IB_IBCC_LINKINITCMD_POLL 2
+/* wait for TS1, then go on */
+#define QLOGIC_IB_IBCC_LINKINITCMD_SLEEP 3
+#define QLOGIC_IB_IBCC_LINKINITCMD_SHIFT 16
+
+#define QLOGIC_IB_IBCC_LINKCMD_DOWN 1           /* move to 0x11 */
+#define QLOGIC_IB_IBCC_LINKCMD_ARMED 2          /* move to 0x21 */
+#define QLOGIC_IB_IBCC_LINKCMD_ACTIVE 3 /* move to 0x31 */
+#define QLOGIC_IB_IBCC_LINKCMD_SHIFT 18
+
+/*
+ * We could have a single register get/put routine, that takes a group type,
+ * but this is somewhat clearer and cleaner.  It also gives us some error
+ * checking.  64 bit register reads should always work, but are inefficient
+ * on opteron (the northbridge always generates 2 separate HT 32 bit reads),
+ * so we use kreg32 wherever possible.  User register and counter register
+ * reads are always 32 bit reads, so only one form of those routines.
+ */
+
+/**
+ * qib_read_ureg32 - read 32-bit virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @ctxt: context number
+ *
+ * Return the contents of a register that is virtualized to be per context.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
+ */
+static inline u32 qib_read_ureg32(const struct qib_devdata *dd,
+				  enum qib_ureg regno, int ctxt)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+
+	if (dd->userbase)
+		return readl(regno + (u64 __iomem *)
+			     ((char __iomem *)dd->userbase +
+			      dd->ureg_align * ctxt));
+	else
+		return readl(regno + (u64 __iomem *)
+			     (dd->uregbase +
+			      (char __iomem *)dd->kregbase +
+			      dd->ureg_align * ctxt));
+}
+
+/**
+ * qib_write_ureg - write 32-bit virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @value: value
+ * @ctxt: context
+ *
+ * Write the contents of a register that is virtualized to be per context.
+ */
+static inline void qib_write_ureg(const struct qib_devdata *dd,
+				  enum qib_ureg regno, u64 value, int ctxt)
+{
+	u64 __iomem *ubase;
+	if (dd->userbase)
+		ubase = (u64 __iomem *)
+			((char __iomem *) dd->userbase +
+			 dd->ureg_align * ctxt);
+	else
+		ubase = (u64 __iomem *)
+			(dd->uregbase +
+			 (char __iomem *) dd->kregbase +
+			 dd->ureg_align * ctxt);
+
+	if (dd->kregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &ubase[regno]);
+}
+
+static inline u32 qib_read_kreg32(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readl((u32 __iomem *)&dd->kregbase[regno]);
+}
+
+static inline u64 qib_read_kreg64(const struct qib_devdata *dd,
+				  const u16 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+
+	return readq(&dd->kregbase[regno]);
+}
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u16 regno, u64 value)
+{
+	if (dd->kregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &dd->kregbase[regno]);
+}
+
+/**
+ * qib_write_kreg_ctxt - write a device's per-ctxt 64-bit kernel register
+ * @dd: the qlogic_ib device
+ * @regno: the register number to write
+ * @ctxt: the context containing the register
+ * @value: the value to write
+ */
+static inline void qib_write_kreg_ctxt(const struct qib_devdata *dd,
+				       const u16 regno, unsigned ctxt,
+				       u64 value)
+{
+	qib_write_kreg(dd, regno + ctxt, value);
+}
+
+static inline void write_6120_creg(const struct qib_devdata *dd,
+				   u16 regno, u64 value)
+{
+	if (dd->cspec->cregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &dd->cspec->cregbase[regno]);
+}
+
+static inline u64 read_6120_creg(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readq(&dd->cspec->cregbase[regno]);
+}
+
+static inline u32 read_6120_creg32(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readl(&dd->cspec->cregbase[regno]);
+}
+
+/* kr_control bits */
+#define QLOGIC_IB_C_RESET 1U
+
+/* kr_intstatus, kr_intclear, kr_intmask bits */
+#define QLOGIC_IB_I_RCVURG_MASK ((1U << 5) - 1)
+#define QLOGIC_IB_I_RCVURG_SHIFT 0
+#define QLOGIC_IB_I_RCVAVAIL_MASK ((1U << 5) - 1)
+#define QLOGIC_IB_I_RCVAVAIL_SHIFT 12
+
+#define QLOGIC_IB_C_FREEZEMODE 0x00000002
+#define QLOGIC_IB_C_LINKENABLE 0x00000004
+#define QLOGIC_IB_I_ERROR               0x0000000080000000ULL
+#define QLOGIC_IB_I_SPIOSENT            0x0000000040000000ULL
+#define QLOGIC_IB_I_SPIOBUFAVAIL        0x0000000020000000ULL
+#define QLOGIC_IB_I_GPIO                0x0000000010000000ULL
+#define QLOGIC_IB_I_BITSEXTANT \
+		((QLOGIC_IB_I_RCVURG_MASK << QLOGIC_IB_I_RCVURG_SHIFT) | \
+		(QLOGIC_IB_I_RCVAVAIL_MASK << \
+		 QLOGIC_IB_I_RCVAVAIL_SHIFT) | \
+		QLOGIC_IB_I_ERROR | QLOGIC_IB_I_SPIOSENT | \
+		QLOGIC_IB_I_SPIOBUFAVAIL | QLOGIC_IB_I_GPIO)
+
+/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
+#define QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK  0x000000000000003fULL
+#define QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT 0
+#define QLOGIC_IB_HWE_PCIEPOISONEDTLP      0x0000000010000000ULL
+#define QLOGIC_IB_HWE_PCIECPLTIMEOUT       0x0000000020000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYXTLH    0x0000000040000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYXADM    0x0000000080000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYRADM    0x0000000100000000ULL
+#define QLOGIC_IB_HWE_COREPLL_FBSLIP       0x0080000000000000ULL
+#define QLOGIC_IB_HWE_COREPLL_RFSLIP       0x0100000000000000ULL
+#define QLOGIC_IB_HWE_PCIE1PLLFAILED       0x0400000000000000ULL
+#define QLOGIC_IB_HWE_PCIE0PLLFAILED       0x0800000000000000ULL
+#define QLOGIC_IB_HWE_SERDESPLLFAILED      0x1000000000000000ULL
+
+
+/* kr_extstatus bits */
+#define QLOGIC_IB_EXTS_FREQSEL 0x2
+#define QLOGIC_IB_EXTS_SERDESSEL 0x4
+#define QLOGIC_IB_EXTS_MEMBIST_ENDTEST     0x0000000000004000
+#define QLOGIC_IB_EXTS_MEMBIST_FOUND       0x0000000000008000
+
+/* kr_xgxsconfig bits */
+#define QLOGIC_IB_XGXS_RESET          0x5ULL
+
+#define _QIB_GPIO_SDA_NUM 1
+#define _QIB_GPIO_SCL_NUM 0
+
+/* Bits in GPIO for the added IB link interrupts */
+#define GPIO_RXUVL_BIT 3
+#define GPIO_OVRUN_BIT 4
+#define GPIO_LLI_BIT 5
+#define GPIO_ERRINTR_MASK 0x38
+
+
+#define QLOGIC_IB_RT_BUFSIZE_MASK 0xe0000000ULL
+#define QLOGIC_IB_RT_BUFSIZE_SHIFTVAL(tid) \
+	((((tid) & QLOGIC_IB_RT_BUFSIZE_MASK) >> 29) + 11 - 1)
+#define QLOGIC_IB_RT_BUFSIZE(tid) (1 << QLOGIC_IB_RT_BUFSIZE_SHIFTVAL(tid))
+#define QLOGIC_IB_RT_IS_VALID(tid) \
+	(((tid) & QLOGIC_IB_RT_BUFSIZE_MASK) && \
+	 ((((tid) & QLOGIC_IB_RT_BUFSIZE_MASK) != QLOGIC_IB_RT_BUFSIZE_MASK)))
+#define QLOGIC_IB_RT_ADDR_MASK 0x1FFFFFFFULL /* 29 bits valid */
+#define QLOGIC_IB_RT_ADDR_SHIFT 10
+
+#define QLOGIC_IB_R_INTRAVAIL_SHIFT 16
+#define QLOGIC_IB_R_TAILUPD_SHIFT 31
+#define IBA6120_R_PKEY_DIS_SHIFT 30
+
+#define PBC_6120_VL15_SEND_CTRL (1ULL << 31) /* pbc; VL15; link_buf only */
+
+#define IBCBUSFRSPCPARITYERR HWE_MASK(IBCBusFromSPCParityErr)
+#define IBCBUSTOSPCPARITYERR HWE_MASK(IBCBusToSPCParityErr)
+
+#define SYM_MASK_BIT(regname, fldname, bit) ((u64) \
+	((1ULL << (SYM_LSB(regname, fldname) + (bit)))))
+
+#define TXEMEMPARITYERR_PIOBUF \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 0)
+#define TXEMEMPARITYERR_PIOPBC \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 1)
+#define TXEMEMPARITYERR_PIOLAUNCHFIFO \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 2)
+
+#define RXEMEMPARITYERR_RCVBUF \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 0)
+#define RXEMEMPARITYERR_LOOKUPQ \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 1)
+#define RXEMEMPARITYERR_EXPTID \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 2)
+#define RXEMEMPARITYERR_EAGERTID \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 3)
+#define RXEMEMPARITYERR_FLAGBUF \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 4)
+#define RXEMEMPARITYERR_DATAINFO \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 5)
+#define RXEMEMPARITYERR_HDRINFO \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 6)
+
+/* 6120 specific hardware errors... */
+static const struct qib_hwerror_msgs qib_6120_hwerror_msgs[] = {
+	/* generic hardware errors */
+	QLOGIC_IB_HWE_MSG(IBCBUSFRSPCPARITYERR, "QIB2IB Parity"),
+	QLOGIC_IB_HWE_MSG(IBCBUSTOSPCPARITYERR, "IB2QIB Parity"),
+
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOBUF,
+			  "TXE PIOBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOPBC,
+			  "TXE PIOPBC Memory Parity"),
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOLAUNCHFIFO,
+			  "TXE PIOLAUNCHFIFO Memory Parity"),
+
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_RCVBUF,
+			  "RXE RCVBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_LOOKUPQ,
+			  "RXE LOOKUPQ Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_EAGERTID,
+			  "RXE EAGERTID Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_EXPTID,
+			  "RXE EXPTID Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_FLAGBUF,
+			  "RXE FLAGBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_DATAINFO,
+			  "RXE DATAINFO Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_HDRINFO,
+			  "RXE HDRINFO Memory Parity"),
+
+	/* chip-specific hardware errors */
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEPOISONEDTLP,
+			  "PCIe Poisoned TLP"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIECPLTIMEOUT,
+			  "PCIe completion timeout"),
+	/*
+	 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
+	 * parity or memory parity error failures, because most likely we
+	 * won't be able to talk to the core of the chip.  Nonetheless, we
+	 * might see them, if they are in parts of the PCIe core that aren't
+	 * essential.
+	 */
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE1PLLFAILED,
+			  "PCIePLL1"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE0PLLFAILED,
+			  "PCIePLL0"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYXTLH,
+			  "PCIe XTLH core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYXADM,
+			  "PCIe ADM TX core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYRADM,
+			  "PCIe ADM RX core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_SERDESPLLFAILED,
+			  "SerDes PLL"),
+};
+
+#define TXE_PIO_PARITY (TXEMEMPARITYERR_PIOBUF | TXEMEMPARITYERR_PIOPBC)
+#define _QIB_PLL_FAIL (QLOGIC_IB_HWE_COREPLL_FBSLIP |   \
+		QLOGIC_IB_HWE_COREPLL_RFSLIP)
+
+	/* variables for sanity checking interrupt and errors */
+#define IB_HWE_BITSEXTANT \
+	(HWE_MASK(RXEMemParityErr) |					\
+	 HWE_MASK(TXEMemParityErr) |					\
+	 (QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK <<			\
+	  QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT) |			\
+	 QLOGIC_IB_HWE_PCIE1PLLFAILED |					\
+	 QLOGIC_IB_HWE_PCIE0PLLFAILED |					\
+	 QLOGIC_IB_HWE_PCIEPOISONEDTLP |				\
+	 QLOGIC_IB_HWE_PCIECPLTIMEOUT |					\
+	 QLOGIC_IB_HWE_PCIEBUSPARITYXTLH |				\
+	 QLOGIC_IB_HWE_PCIEBUSPARITYXADM |				\
+	 QLOGIC_IB_HWE_PCIEBUSPARITYRADM |				\
+	 HWE_MASK(PowerOnBISTFailed) |					\
+	 QLOGIC_IB_HWE_COREPLL_FBSLIP |					\
+	 QLOGIC_IB_HWE_COREPLL_RFSLIP |					\
+	 QLOGIC_IB_HWE_SERDESPLLFAILED |				\
+	 HWE_MASK(IBCBusToSPCParityErr) |				\
+	 HWE_MASK(IBCBusFromSPCParityErr))
+
+#define IB_E_BITSEXTANT \
+	(ERR_MASK(RcvFormatErr) | ERR_MASK(RcvVCRCErr) |		\
+	 ERR_MASK(RcvICRCErr) | ERR_MASK(RcvMinPktLenErr) |		\
+	 ERR_MASK(RcvMaxPktLenErr) | ERR_MASK(RcvLongPktLenErr) |	\
+	 ERR_MASK(RcvShortPktLenErr) | ERR_MASK(RcvUnexpectedCharErr) | \
+	 ERR_MASK(RcvUnsupportedVLErr) | ERR_MASK(RcvEBPErr) |		\
+	 ERR_MASK(RcvIBFlowErr) | ERR_MASK(RcvBadVersionErr) |		\
+	 ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr) |		\
+	 ERR_MASK(RcvBadTidErr) | ERR_MASK(RcvHdrLenErr) |		\
+	 ERR_MASK(RcvHdrErr) | ERR_MASK(RcvIBLostLinkErr) |		\
+	 ERR_MASK(SendMinPktLenErr) | ERR_MASK(SendMaxPktLenErr) |	\
+	 ERR_MASK(SendUnderRunErr) | ERR_MASK(SendPktLenErr) |		\
+	 ERR_MASK(SendDroppedSmpPktErr) |				\
+	 ERR_MASK(SendDroppedDataPktErr) |				\
+	 ERR_MASK(SendPioArmLaunchErr) |				\
+	 ERR_MASK(SendUnexpectedPktNumErr) |				\
+	 ERR_MASK(SendUnsupportedVLErr) | ERR_MASK(IBStatusChanged) |	\
+	 ERR_MASK(InvalidAddrErr) | ERR_MASK(ResetNegated) |		\
+	 ERR_MASK(HardwareErr))
+
+#define QLOGIC_IB_E_PKTERRS ( \
+		ERR_MASK(SendPktLenErr) |				\
+		ERR_MASK(SendDroppedDataPktErr) |			\
+		ERR_MASK(RcvVCRCErr) |					\
+		ERR_MASK(RcvICRCErr) |					\
+		ERR_MASK(RcvShortPktLenErr) |				\
+		ERR_MASK(RcvEBPErr))
+
+/* These are all rcv-related errors which we want to count for stats */
+#define E_SUM_PKTERRS						\
+	(ERR_MASK(RcvHdrLenErr) | ERR_MASK(RcvBadTidErr) |		\
+	 ERR_MASK(RcvBadVersionErr) | ERR_MASK(RcvHdrErr) |		\
+	 ERR_MASK(RcvLongPktLenErr) | ERR_MASK(RcvShortPktLenErr) |	\
+	 ERR_MASK(RcvMaxPktLenErr) | ERR_MASK(RcvMinPktLenErr) |	\
+	 ERR_MASK(RcvFormatErr) | ERR_MASK(RcvUnsupportedVLErr) |	\
+	 ERR_MASK(RcvUnexpectedCharErr) | ERR_MASK(RcvEBPErr))
+
+/* These are all send-related errors which we want to count for stats */
+#define E_SUM_ERRS							\
+	(ERR_MASK(SendPioArmLaunchErr) |				\
+	 ERR_MASK(SendUnexpectedPktNumErr) |				\
+	 ERR_MASK(SendDroppedDataPktErr) |				\
+	 ERR_MASK(SendDroppedSmpPktErr) |				\
+	 ERR_MASK(SendMaxPktLenErr) | ERR_MASK(SendUnsupportedVLErr) |	\
+	 ERR_MASK(SendMinPktLenErr) | ERR_MASK(SendPktLenErr) |		\
+	 ERR_MASK(InvalidAddrErr))
+
+/*
+ * this is similar to E_SUM_ERRS, but can't ignore armlaunch, don't ignore
+ * errors not related to freeze and cancelling buffers.  Can't ignore
+ * armlaunch because could get more while still cleaning up, and need
+ * to cancel those as they happen.
+ */
+#define E_SPKT_ERRS_IGNORE \
+	(ERR_MASK(SendDroppedDataPktErr) |				\
+	 ERR_MASK(SendDroppedSmpPktErr) |				\
+	 ERR_MASK(SendMaxPktLenErr) | ERR_MASK(SendMinPktLenErr) |	\
+	 ERR_MASK(SendPktLenErr))
+
+/*
+ * these are errors that can occur when the link changes state while
+ * a packet is being sent or received.  This doesn't cover things
+ * like EBP or VCRC that can be the result of a sending having the
+ * link change state, so we receive a "known bad" packet.
+ */
+#define E_SUM_LINK_PKTERRS		\
+	(ERR_MASK(SendDroppedDataPktErr) |				\
+	 ERR_MASK(SendDroppedSmpPktErr) |				\
+	 ERR_MASK(SendMinPktLenErr) | ERR_MASK(SendPktLenErr) |		\
+	 ERR_MASK(RcvShortPktLenErr) | ERR_MASK(RcvMinPktLenErr) |	\
+	 ERR_MASK(RcvUnexpectedCharErr))
+
+static void qib_6120_put_tid_2(struct qib_devdata *, u64 __iomem *,
+			       u32, unsigned long);
+
+/*
+ * On platforms using this chip, and not having ordered WC stores, we
+ * can get TXE parity errors due to speculative reads to the PIO buffers,
+ * and this, due to a chip issue can result in (many) false parity error
+ * reports.  So it's a debug print on those, and an info print on systems
+ * where the speculative reads don't occur.
+ */
+static void qib_6120_txe_recover(struct qib_devdata *dd)
+{
+	if (!qib_unordered_wc())
+		qib_devinfo(dd->pcidev,
+			    "Recovering from TXE PIO parity error\n");
+}
+
+/* enable/disable chip from delivering interrupts */
+static void qib_6120_set_intr_state(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		if (dd->flags & QIB_BADINTR)
+			return;
+		qib_write_kreg(dd, kr_intmask, ~0ULL);
+		/* force re-interrupt of any pending interrupts. */
+		qib_write_kreg(dd, kr_intclear, 0ULL);
+	} else
+		qib_write_kreg(dd, kr_intmask, 0ULL);
+}
+
+/*
+ * Try to cleanup as much as possible for anything that might have gone
+ * wrong while in freeze mode, such as pio buffers being written by user
+ * processes (causing armlaunch), send errors due to going into freeze mode,
+ * etc., and try to avoid causing extra interrupts while doing so.
+ * Forcibly update the in-memory pioavail register copies after cleanup
+ * because the chip won't do it while in freeze mode (the register values
+ * themselves are kept correct).
+ * Make sure that we don't lose any important interrupts by using the chip
+ * feature that says that writing 0 to a bit in *clear that is set in
+ * *status will cause an interrupt to be generated again (if allowed by
+ * the *mask value).
+ * This is in chip-specific code because of all of the register accesses,
+ * even though the details are similar on most chips
+ */
+static void qib_6120_clear_freeze(struct qib_devdata *dd)
+{
+	/* disable error interrupts, to avoid confusion */
+	qib_write_kreg(dd, kr_errmask, 0ULL);
+
+	/* also disable interrupts; errormask is sometimes overwriten */
+	qib_6120_set_intr_state(dd, 0);
+
+	qib_cancel_sends(dd->pport);
+
+	/* clear the freeze, and be sure chip saw it */
+	qib_write_kreg(dd, kr_control, dd->control);
+	qib_read_kreg32(dd, kr_scratch);
+
+	/* force in-memory update now we are out of freeze */
+	qib_force_pio_avail_update(dd);
+
+	/*
+	 * force new interrupt if any hwerr, error or interrupt bits are
+	 * still set, and clear "safe" send packet errors related to freeze
+	 * and cancelling sends.  Re-enable error interrupts before possible
+	 * force of re-interrupt on pending interrupts.
+	 */
+	qib_write_kreg(dd, kr_hwerrclear, 0ULL);
+	qib_write_kreg(dd, kr_errclear, E_SPKT_ERRS_IGNORE);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+	qib_6120_set_intr_state(dd, 1);
+}
+
+/**
+ * qib_handle_6120_hwerrors - display hardware errors.
+ * @dd: the qlogic_ib device
+ * @msg: the output buffer
+ * @msgl: the size of the output buffer
+ *
+ * Use same msg buffer as regular errors to avoid excessive stack
+ * use.  Most hardware errors are catastrophic, but for right now,
+ * we'll print them and continue.  Reuse the same message buffer as
+ * handle_6120_errors() to avoid excessive stack usage.
+ */
+static void qib_handle_6120_hwerrors(struct qib_devdata *dd, char *msg,
+				     size_t msgl)
+{
+	u64 hwerrs;
+	u32 bits, ctrl;
+	int isfatal = 0;
+	char *bitsmsg;
+	int log_idx;
+
+	hwerrs = qib_read_kreg64(dd, kr_hwerrstatus);
+	if (!hwerrs)
+		return;
+	if (hwerrs == ~0ULL) {
+		qib_dev_err(dd,
+			"Read of hardware error status failed (all bits set); ignoring\n");
+		return;
+	}
+	qib_stats.sps_hwerrs++;
+
+	/* Always clear the error status register, except MEMBISTFAIL,
+	 * regardless of whether we continue or stop using the chip.
+	 * We want that set so we know it failed, even across driver reload.
+	 * We'll still ignore it in the hwerrmask.  We do this partly for
+	 * diagnostics, but also for support */
+	qib_write_kreg(dd, kr_hwerrclear,
+		       hwerrs & ~HWE_MASK(PowerOnBISTFailed));
+
+	hwerrs &= dd->cspec->hwerrmask;
+
+	/* We log some errors to EEPROM, check if we have any of those. */
+	for (log_idx = 0; log_idx < QIB_EEP_LOG_CNT; ++log_idx)
+		if (hwerrs & dd->eep_st_masks[log_idx].hwerrs_to_log)
+			qib_inc_eeprom_err(dd, log_idx, 1);
+
+	/*
+	 * Make sure we get this much out, unless told to be quiet,
+	 * or it's occurred within the last 5 seconds.
+	 */
+	if (hwerrs & ~(TXE_PIO_PARITY | RXEMEMPARITYERR_EAGERTID))
+		qib_devinfo(dd->pcidev,
+			"Hardware error: hwerr=0x%llx (cleared)\n",
+			(unsigned long long) hwerrs);
+
+	if (hwerrs & ~IB_HWE_BITSEXTANT)
+		qib_dev_err(dd,
+			"hwerror interrupt with unknown errors %llx set\n",
+			(unsigned long long)(hwerrs & ~IB_HWE_BITSEXTANT));
+
+	ctrl = qib_read_kreg32(dd, kr_control);
+	if ((ctrl & QLOGIC_IB_C_FREEZEMODE) && !dd->diag_client) {
+		/*
+		 * Parity errors in send memory are recoverable,
+		 * just cancel the send (if indicated in * sendbuffererror),
+		 * count the occurrence, unfreeze (if no other handled
+		 * hardware error bits are set), and continue. They can
+		 * occur if a processor speculative read is done to the PIO
+		 * buffer while we are sending a packet, for example.
+		 */
+		if (hwerrs & TXE_PIO_PARITY) {
+			qib_6120_txe_recover(dd);
+			hwerrs &= ~TXE_PIO_PARITY;
+		}
+
+		if (!hwerrs) {
+			static u32 freeze_cnt;
+
+			freeze_cnt++;
+			qib_6120_clear_freeze(dd);
+		} else
+			isfatal = 1;
+	}
+
+	*msg = '\0';
+
+	if (hwerrs & HWE_MASK(PowerOnBISTFailed)) {
+		isfatal = 1;
+		strlcat(msg,
+			"[Memory BIST test failed, InfiniPath hardware unusable]",
+			msgl);
+		/* ignore from now on, so disable until driver reloaded */
+		dd->cspec->hwerrmask &= ~HWE_MASK(PowerOnBISTFailed);
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	qib_format_hwerrors(hwerrs, qib_6120_hwerror_msgs,
+			    ARRAY_SIZE(qib_6120_hwerror_msgs), msg, msgl);
+
+	bitsmsg = dd->cspec->bitsmsgbuf;
+	if (hwerrs & (QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK <<
+		      QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT)) {
+		bits = (u32) ((hwerrs >>
+			       QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT) &
+			      QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK);
+		snprintf(bitsmsg, sizeof dd->cspec->bitsmsgbuf,
+			 "[PCIe Mem Parity Errs %x] ", bits);
+		strlcat(msg, bitsmsg, msgl);
+	}
+
+	if (hwerrs & _QIB_PLL_FAIL) {
+		isfatal = 1;
+		snprintf(bitsmsg, sizeof dd->cspec->bitsmsgbuf,
+			 "[PLL failed (%llx), InfiniPath hardware unusable]",
+			 (unsigned long long) hwerrs & _QIB_PLL_FAIL);
+		strlcat(msg, bitsmsg, msgl);
+		/* ignore from now on, so disable until driver reloaded */
+		dd->cspec->hwerrmask &= ~(hwerrs & _QIB_PLL_FAIL);
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	if (hwerrs & QLOGIC_IB_HWE_SERDESPLLFAILED) {
+		/*
+		 * If it occurs, it is left masked since the external
+		 * interface is unused
+		 */
+		dd->cspec->hwerrmask &= ~QLOGIC_IB_HWE_SERDESPLLFAILED;
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	if (hwerrs)
+		/*
+		 * if any set that we aren't ignoring; only
+		 * make the complaint once, in case it's stuck
+		 * or recurring, and we get here multiple
+		 * times.
+		 */
+		qib_dev_err(dd, "%s hardware error\n", msg);
+	else
+		*msg = 0; /* recovered from all of them */
+
+	if (isfatal && !dd->diag_client) {
+		qib_dev_err(dd,
+			"Fatal Hardware Error, no longer usable, SN %.16s\n",
+			dd->serial);
+		/*
+		 * for /sys status file and user programs to print; if no
+		 * trailing brace is copied, we'll know it was truncated.
+		 */
+		if (dd->freezemsg)
+			snprintf(dd->freezemsg, dd->freezelen,
+				 "{%s}", msg);
+		qib_disable_after_error(dd);
+	}
+}
+
+/*
+ * Decode the error status into strings, deciding whether to always
+ * print * it or not depending on "normal packet errors" vs everything
+ * else.   Return 1 if "real" errors, otherwise 0 if only packet
+ * errors, so caller can decide what to print with the string.
+ */
+static int qib_decode_6120_err(struct qib_devdata *dd, char *buf, size_t blen,
+			       u64 err)
+{
+	int iserr = 1;
+
+	*buf = '\0';
+	if (err & QLOGIC_IB_E_PKTERRS) {
+		if (!(err & ~QLOGIC_IB_E_PKTERRS))
+			iserr = 0;
+		if ((err & ERR_MASK(RcvICRCErr)) &&
+		    !(err&(ERR_MASK(RcvVCRCErr)|ERR_MASK(RcvEBPErr))))
+			strlcat(buf, "CRC ", blen);
+		if (!iserr)
+			goto done;
+	}
+	if (err & ERR_MASK(RcvHdrLenErr))
+		strlcat(buf, "rhdrlen ", blen);
+	if (err & ERR_MASK(RcvBadTidErr))
+		strlcat(buf, "rbadtid ", blen);
+	if (err & ERR_MASK(RcvBadVersionErr))
+		strlcat(buf, "rbadversion ", blen);
+	if (err & ERR_MASK(RcvHdrErr))
+		strlcat(buf, "rhdr ", blen);
+	if (err & ERR_MASK(RcvLongPktLenErr))
+		strlcat(buf, "rlongpktlen ", blen);
+	if (err & ERR_MASK(RcvMaxPktLenErr))
+		strlcat(buf, "rmaxpktlen ", blen);
+	if (err & ERR_MASK(RcvMinPktLenErr))
+		strlcat(buf, "rminpktlen ", blen);
+	if (err & ERR_MASK(SendMinPktLenErr))
+		strlcat(buf, "sminpktlen ", blen);
+	if (err & ERR_MASK(RcvFormatErr))
+		strlcat(buf, "rformaterr ", blen);
+	if (err & ERR_MASK(RcvUnsupportedVLErr))
+		strlcat(buf, "runsupvl ", blen);
+	if (err & ERR_MASK(RcvUnexpectedCharErr))
+		strlcat(buf, "runexpchar ", blen);
+	if (err & ERR_MASK(RcvIBFlowErr))
+		strlcat(buf, "ribflow ", blen);
+	if (err & ERR_MASK(SendUnderRunErr))
+		strlcat(buf, "sunderrun ", blen);
+	if (err & ERR_MASK(SendPioArmLaunchErr))
+		strlcat(buf, "spioarmlaunch ", blen);
+	if (err & ERR_MASK(SendUnexpectedPktNumErr))
+		strlcat(buf, "sunexperrpktnum ", blen);
+	if (err & ERR_MASK(SendDroppedSmpPktErr))
+		strlcat(buf, "sdroppedsmppkt ", blen);
+	if (err & ERR_MASK(SendMaxPktLenErr))
+		strlcat(buf, "smaxpktlen ", blen);
+	if (err & ERR_MASK(SendUnsupportedVLErr))
+		strlcat(buf, "sunsupVL ", blen);
+	if (err & ERR_MASK(InvalidAddrErr))
+		strlcat(buf, "invalidaddr ", blen);
+	if (err & ERR_MASK(RcvEgrFullErr))
+		strlcat(buf, "rcvegrfull ", blen);
+	if (err & ERR_MASK(RcvHdrFullErr))
+		strlcat(buf, "rcvhdrfull ", blen);
+	if (err & ERR_MASK(IBStatusChanged))
+		strlcat(buf, "ibcstatuschg ", blen);
+	if (err & ERR_MASK(RcvIBLostLinkErr))
+		strlcat(buf, "riblostlink ", blen);
+	if (err & ERR_MASK(HardwareErr))
+		strlcat(buf, "hardware ", blen);
+	if (err & ERR_MASK(ResetNegated))
+		strlcat(buf, "reset ", blen);
+done:
+	return iserr;
+}
+
+/*
+ * Called when we might have an error that is specific to a particular
+ * PIO buffer, and may need to cancel that buffer, so it can be re-used.
+ */
+static void qib_disarm_6120_senderrbufs(struct qib_pportdata *ppd)
+{
+	unsigned long sbuf[2];
+	struct qib_devdata *dd = ppd->dd;
+
+	/*
+	 * It's possible that sendbuffererror could have bits set; might
+	 * have already done this as a result of hardware error handling.
+	 */
+	sbuf[0] = qib_read_kreg64(dd, kr_sendbuffererror);
+	sbuf[1] = qib_read_kreg64(dd, kr_sendbuffererror + 1);
+
+	if (sbuf[0] || sbuf[1])
+		qib_disarm_piobufs_set(dd, sbuf,
+				       dd->piobcnt2k + dd->piobcnt4k);
+}
+
+static int chk_6120_linkrecovery(struct qib_devdata *dd, u64 ibcs)
+{
+	int ret = 1;
+	u32 ibstate = qib_6120_iblink_state(ibcs);
+	u32 linkrecov = read_6120_creg32(dd, cr_iblinkerrrecov);
+
+	if (linkrecov != dd->cspec->lastlinkrecov) {
+		/* and no more until active again */
+		dd->cspec->lastlinkrecov = 0;
+		qib_set_linkstate(dd->pport, QIB_IB_LINKDOWN);
+		ret = 0;
+	}
+	if (ibstate == IB_PORT_ACTIVE)
+		dd->cspec->lastlinkrecov =
+			read_6120_creg32(dd, cr_iblinkerrrecov);
+	return ret;
+}
+
+static void handle_6120_errors(struct qib_devdata *dd, u64 errs)
+{
+	char *msg;
+	u64 ignore_this_time = 0;
+	u64 iserr = 0;
+	int log_idx;
+	struct qib_pportdata *ppd = dd->pport;
+	u64 mask;
+
+	/* don't report errors that are masked */
+	errs &= dd->cspec->errormask;
+	msg = dd->cspec->emsgbuf;
+
+	/* do these first, they are most important */
+	if (errs & ERR_MASK(HardwareErr))
+		qib_handle_6120_hwerrors(dd, msg, sizeof dd->cspec->emsgbuf);
+	else
+		for (log_idx = 0; log_idx < QIB_EEP_LOG_CNT; ++log_idx)
+			if (errs & dd->eep_st_masks[log_idx].errs_to_log)
+				qib_inc_eeprom_err(dd, log_idx, 1);
+
+	if (errs & ~IB_E_BITSEXTANT)
+		qib_dev_err(dd,
+			"error interrupt with unknown errors %llx set\n",
+			(unsigned long long) (errs & ~IB_E_BITSEXTANT));
+
+	if (errs & E_SUM_ERRS) {
+		qib_disarm_6120_senderrbufs(ppd);
+		if ((errs & E_SUM_LINK_PKTERRS) &&
+		    !(ppd->lflags & QIBL_LINKACTIVE)) {
+			/*
+			 * This can happen when trying to bring the link
+			 * up, but the IB link changes state at the "wrong"
+			 * time. The IB logic then complains that the packet
+			 * isn't valid.  We don't want to confuse people, so
+			 * we just don't print them, except at debug
+			 */
+			ignore_this_time = errs & E_SUM_LINK_PKTERRS;
+		}
+	} else if ((errs & E_SUM_LINK_PKTERRS) &&
+		   !(ppd->lflags & QIBL_LINKACTIVE)) {
+		/*
+		 * This can happen when SMA is trying to bring the link
+		 * up, but the IB link changes state at the "wrong" time.
+		 * The IB logic then complains that the packet isn't
+		 * valid.  We don't want to confuse people, so we just
+		 * don't print them, except at debug
+		 */
+		ignore_this_time = errs & E_SUM_LINK_PKTERRS;
+	}
+
+	qib_write_kreg(dd, kr_errclear, errs);
+
+	errs &= ~ignore_this_time;
+	if (!errs)
+		goto done;
+
+	/*
+	 * The ones we mask off are handled specially below
+	 * or above.
+	 */
+	mask = ERR_MASK(IBStatusChanged) | ERR_MASK(RcvEgrFullErr) |
+		ERR_MASK(RcvHdrFullErr) | ERR_MASK(HardwareErr);
+	qib_decode_6120_err(dd, msg, sizeof dd->cspec->emsgbuf, errs & ~mask);
+
+	if (errs & E_SUM_PKTERRS)
+		qib_stats.sps_rcverrs++;
+	if (errs & E_SUM_ERRS)
+		qib_stats.sps_txerrs++;
+
+	iserr = errs & ~(E_SUM_PKTERRS | QLOGIC_IB_E_PKTERRS);
+
+	if (errs & ERR_MASK(IBStatusChanged)) {
+		u64 ibcs = qib_read_kreg64(dd, kr_ibcstatus);
+		u32 ibstate = qib_6120_iblink_state(ibcs);
+		int handle = 1;
+
+		if (ibstate != IB_PORT_INIT && dd->cspec->lastlinkrecov)
+			handle = chk_6120_linkrecovery(dd, ibcs);
+		/*
+		 * Since going into a recovery state causes the link state
+		 * to go down and since recovery is transitory, it is better
+		 * if we "miss" ever seeing the link training state go into
+		 * recovery (i.e., ignore this transition for link state
+		 * special handling purposes) without updating lastibcstat.
+		 */
+		if (handle && qib_6120_phys_portstate(ibcs) ==
+					    IB_PHYSPORTSTATE_LINK_ERR_RECOVER)
+			handle = 0;
+		if (handle)
+			qib_handle_e_ibstatuschanged(ppd, ibcs);
+	}
+
+	if (errs & ERR_MASK(ResetNegated)) {
+		qib_dev_err(dd,
+			"Got reset, requires re-init (unload and reload driver)\n");
+		dd->flags &= ~QIB_INITTED;  /* needs re-init */
+		/* mark as having had error */
+		*dd->devstatusp |= QIB_STATUS_HWERROR;
+		*dd->pport->statusp &= ~QIB_STATUS_IB_CONF;
+	}
+
+	if (*msg && iserr)
+		qib_dev_porterr(dd, ppd->port, "%s error\n", msg);
+
+	if (ppd->state_wanted & ppd->lflags)
+		wake_up_interruptible(&ppd->state_wait);
+
+	/*
+	 * If there were hdrq or egrfull errors, wake up any processes
+	 * waiting in poll.  We used to try to check which contexts had
+	 * the overflow, but given the cost of that and the chip reads
+	 * to support it, it's better to just wake everybody up if we
+	 * get an overflow; waiters can poll again if it's not them.
+	 */
+	if (errs & (ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr))) {
+		qib_handle_urcv(dd, ~0U);
+		if (errs & ERR_MASK(RcvEgrFullErr))
+			qib_stats.sps_buffull++;
+		else
+			qib_stats.sps_hdrfull++;
+	}
+done:
+	return;
+}
+
+/**
+ * qib_6120_init_hwerrors - enable hardware errors
+ * @dd: the qlogic_ib device
+ *
+ * now that we have finished initializing everything that might reasonably
+ * cause a hardware error, and cleared those errors bits as they occur,
+ * we can enable hardware errors in the mask (potentially enabling
+ * freeze mode), and enable hardware errors as errors (along with
+ * everything else) in errormask
+ */
+static void qib_6120_init_hwerrors(struct qib_devdata *dd)
+{
+	u64 val;
+	u64 extsval;
+
+	extsval = qib_read_kreg64(dd, kr_extstatus);
+
+	if (!(extsval & QLOGIC_IB_EXTS_MEMBIST_ENDTEST))
+		qib_dev_err(dd, "MemBIST did not complete!\n");
+
+	/* init so all hwerrors interrupt, and enter freeze, ajdust below */
+	val = ~0ULL;
+	if (dd->minrev < 2) {
+		/*
+		 * Avoid problem with internal interface bus parity
+		 * checking. Fixed in Rev2.
+		 */
+		val &= ~QLOGIC_IB_HWE_PCIEBUSPARITYRADM;
+	}
+	/* avoid some intel cpu's speculative read freeze mode issue */
+	val &= ~TXEMEMPARITYERR_PIOBUF;
+
+	dd->cspec->hwerrmask = val;
+
+	qib_write_kreg(dd, kr_hwerrclear, ~HWE_MASK(PowerOnBISTFailed));
+	qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+
+	/* clear all */
+	qib_write_kreg(dd, kr_errclear, ~0ULL);
+	/* enable errors that are masked, at least this first time. */
+	qib_write_kreg(dd, kr_errmask, ~0ULL);
+	dd->cspec->errormask = qib_read_kreg64(dd, kr_errmask);
+	/* clear any interrupts up to this point (ints still not enabled) */
+	qib_write_kreg(dd, kr_intclear, ~0ULL);
+
+	qib_write_kreg(dd, kr_rcvbthqp,
+		       dd->qpn_mask << (QIB_6120_RcvBTHQP_BTHQP_Mask_LSB - 1) |
+		       QIB_KD_QP);
+}
+
+/*
+ * Disable and enable the armlaunch error.  Used for PIO bandwidth testing
+ * on chips that are count-based, rather than trigger-based.  There is no
+ * reference counting, but that's also fine, given the intended use.
+ * Only chip-specific because it's all register accesses
+ */
+static void qib_set_6120_armlaunch(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		qib_write_kreg(dd, kr_errclear,
+			       ERR_MASK(SendPioArmLaunchErr));
+		dd->cspec->errormask |= ERR_MASK(SendPioArmLaunchErr);
+	} else
+		dd->cspec->errormask &= ~ERR_MASK(SendPioArmLaunchErr);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+}
+
+/*
+ * Formerly took parameter <which> in pre-shifted,
+ * pre-merged form with LinkCmd and LinkInitCmd
+ * together, and assuming the zero was NOP.
+ */
+static void qib_set_ib_6120_lstate(struct qib_pportdata *ppd, u16 linkcmd,
+				   u16 linitcmd)
+{
+	u64 mod_wd;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	if (linitcmd == QLOGIC_IB_IBCC_LINKINITCMD_DISABLE) {
+		/*
+		 * If we are told to disable, note that so link-recovery
+		 * code does not attempt to bring us back up.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	} else if (linitcmd || linkcmd == QLOGIC_IB_IBCC_LINKCMD_DOWN) {
+		/*
+		 * Any other linkinitcmd will lead to LINKDOWN and then
+		 * to INIT (if all is well), so clear flag to let
+		 * link-recovery code attempt to bring us back up.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+
+	mod_wd = (linkcmd << QLOGIC_IB_IBCC_LINKCMD_SHIFT) |
+		(linitcmd << QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+
+	qib_write_kreg(dd, kr_ibcctrl, dd->cspec->ibcctrl | mod_wd);
+	/* write to chip to prevent back-to-back writes of control reg */
+	qib_write_kreg(dd, kr_scratch, 0);
+}
+
+/**
+ * qib_6120_bringup_serdes - bring up the serdes
+ * @dd: the qlogic_ib device
+ */
+static int qib_6120_bringup_serdes(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 val, config1, prev_val, hwstat, ibc;
+
+	/* Put IBC in reset, sends disabled */
+	dd->control &= ~QLOGIC_IB_C_LINKENABLE;
+	qib_write_kreg(dd, kr_control, 0ULL);
+
+	dd->cspec->ibdeltainprog = 1;
+	dd->cspec->ibsymsnap = read_6120_creg32(dd, cr_ibsymbolerr);
+	dd->cspec->iblnkerrsnap = read_6120_creg32(dd, cr_iblinkerrrecov);
+
+	/* flowcontrolwatermark is in units of KBytes */
+	ibc = 0x5ULL << SYM_LSB(IBCCtrl, FlowCtrlWaterMark);
+	/*
+	 * How often flowctrl sent.  More or less in usecs; balance against
+	 * watermark value, so that in theory senders always get a flow
+	 * control update in time to not let the IB link go idle.
+	 */
+	ibc |= 0x3ULL << SYM_LSB(IBCCtrl, FlowCtrlPeriod);
+	/* max error tolerance */
+	dd->cspec->lli_thresh = 0xf;
+	ibc |= (u64) dd->cspec->lli_thresh << SYM_LSB(IBCCtrl, PhyerrThreshold);
+	/* use "real" buffer space for */
+	ibc |= 4ULL << SYM_LSB(IBCCtrl, CreditScale);
+	/* IB credit flow control. */
+	ibc |= 0xfULL << SYM_LSB(IBCCtrl, OverrunThreshold);
+	/*
+	 * set initial max size pkt IBC will send, including ICRC; it's the
+	 * PIO buffer size in dwords, less 1; also see qib_set_mtu()
+	 */
+	ibc |= ((u64)(ppd->ibmaxlen >> 2) + 1) << SYM_LSB(IBCCtrl, MaxPktLen);
+	dd->cspec->ibcctrl = ibc; /* without linkcmd or linkinitcmd! */
+
+	/* initially come up waiting for TS1, without sending anything. */
+	val = dd->cspec->ibcctrl | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE <<
+		QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+	qib_write_kreg(dd, kr_ibcctrl, val);
+
+	val = qib_read_kreg64(dd, kr_serdes_cfg0);
+	config1 = qib_read_kreg64(dd, kr_serdes_cfg1);
+
+	/*
+	 * Force reset on, also set rxdetect enable.  Must do before reading
+	 * serdesstatus at least for simulation, or some of the bits in
+	 * serdes status will come back as undefined and cause simulation
+	 * failures
+	 */
+	val |= SYM_MASK(SerdesCfg0, ResetPLL) |
+		SYM_MASK(SerdesCfg0, RxDetEnX) |
+		(SYM_MASK(SerdesCfg0, L1PwrDnA) |
+		 SYM_MASK(SerdesCfg0, L1PwrDnB) |
+		 SYM_MASK(SerdesCfg0, L1PwrDnC) |
+		 SYM_MASK(SerdesCfg0, L1PwrDnD));
+	qib_write_kreg(dd, kr_serdes_cfg0, val);
+	/* be sure chip saw it */
+	qib_read_kreg64(dd, kr_scratch);
+	udelay(5);              /* need pll reset set at least for a bit */
+	/*
+	 * after PLL is reset, set the per-lane Resets and TxIdle and
+	 * clear the PLL reset and rxdetect (to get falling edge).
+	 * Leave L1PWR bits set (permanently)
+	 */
+	val &= ~(SYM_MASK(SerdesCfg0, RxDetEnX) |
+		 SYM_MASK(SerdesCfg0, ResetPLL) |
+		 (SYM_MASK(SerdesCfg0, L1PwrDnA) |
+		  SYM_MASK(SerdesCfg0, L1PwrDnB) |
+		  SYM_MASK(SerdesCfg0, L1PwrDnC) |
+		  SYM_MASK(SerdesCfg0, L1PwrDnD)));
+	val |= (SYM_MASK(SerdesCfg0, ResetA) |
+		SYM_MASK(SerdesCfg0, ResetB) |
+		SYM_MASK(SerdesCfg0, ResetC) |
+		SYM_MASK(SerdesCfg0, ResetD)) |
+		SYM_MASK(SerdesCfg0, TxIdeEnX);
+	qib_write_kreg(dd, kr_serdes_cfg0, val);
+	/* be sure chip saw it */
+	(void) qib_read_kreg64(dd, kr_scratch);
+	/* need PLL reset clear for at least 11 usec before lane
+	 * resets cleared; give it a few more to be sure */
+	udelay(15);
+	val &= ~((SYM_MASK(SerdesCfg0, ResetA) |
+		  SYM_MASK(SerdesCfg0, ResetB) |
+		  SYM_MASK(SerdesCfg0, ResetC) |
+		  SYM_MASK(SerdesCfg0, ResetD)) |
+		 SYM_MASK(SerdesCfg0, TxIdeEnX));
+
+	qib_write_kreg(dd, kr_serdes_cfg0, val);
+	/* be sure chip saw it */
+	(void) qib_read_kreg64(dd, kr_scratch);
+
+	val = qib_read_kreg64(dd, kr_xgxs_cfg);
+	prev_val = val;
+	if (val & QLOGIC_IB_XGXS_RESET)
+		val &= ~QLOGIC_IB_XGXS_RESET;
+	if (SYM_FIELD(val, XGXSCfg, polarity_inv) != ppd->rx_pol_inv) {
+		/* need to compensate for Tx inversion in partner */
+		val &= ~SYM_MASK(XGXSCfg, polarity_inv);
+		val |= (u64)ppd->rx_pol_inv << SYM_LSB(XGXSCfg, polarity_inv);
+	}
+	if (val != prev_val)
+		qib_write_kreg(dd, kr_xgxs_cfg, val);
+
+	val = qib_read_kreg64(dd, kr_serdes_cfg0);
+
+	/* clear current and de-emphasis bits */
+	config1 &= ~0x0ffffffff00ULL;
+	/* set current to 20ma */
+	config1 |= 0x00000000000ULL;
+	/* set de-emphasis to -5.68dB */
+	config1 |= 0x0cccc000000ULL;
+	qib_write_kreg(dd, kr_serdes_cfg1, config1);
+
+	/* base and port guid same for single port */
+	ppd->guid = dd->base_guid;
+
+	/*
+	 * the process of setting and un-resetting the serdes normally
+	 * causes a serdes PLL error, so check for that and clear it
+	 * here.  Also clearr hwerr bit in errstatus, but not others.
+	 */
+	hwstat = qib_read_kreg64(dd, kr_hwerrstatus);
+	if (hwstat) {
+		/* should just have PLL, clear all set, in an case */
+		qib_write_kreg(dd, kr_hwerrclear, hwstat);
+		qib_write_kreg(dd, kr_errclear, ERR_MASK(HardwareErr));
+	}
+
+	dd->control |= QLOGIC_IB_C_LINKENABLE;
+	dd->control &= ~QLOGIC_IB_C_FREEZEMODE;
+	qib_write_kreg(dd, kr_control, dd->control);
+
+	return 0;
+}
+
+/**
+ * qib_6120_quiet_serdes - set serdes to txidle
+ * @ppd: physical port of the qlogic_ib device
+ * Called when driver is being unloaded
+ */
+static void qib_6120_quiet_serdes(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 val;
+
+	qib_set_ib_6120_lstate(ppd, 0, QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+
+	/* disable IBC */
+	dd->control &= ~QLOGIC_IB_C_LINKENABLE;
+	qib_write_kreg(dd, kr_control,
+		       dd->control | QLOGIC_IB_C_FREEZEMODE);
+
+	if (dd->cspec->ibsymdelta || dd->cspec->iblnkerrdelta ||
+	    dd->cspec->ibdeltainprog) {
+		u64 diagc;
+
+		/* enable counter writes */
+		diagc = qib_read_kreg64(dd, kr_hwdiagctrl);
+		qib_write_kreg(dd, kr_hwdiagctrl,
+			       diagc | SYM_MASK(HwDiagCtrl, CounterWrEnable));
+
+		if (dd->cspec->ibsymdelta || dd->cspec->ibdeltainprog) {
+			val = read_6120_creg32(dd, cr_ibsymbolerr);
+			if (dd->cspec->ibdeltainprog)
+				val -= val - dd->cspec->ibsymsnap;
+			val -= dd->cspec->ibsymdelta;
+			write_6120_creg(dd, cr_ibsymbolerr, val);
+		}
+		if (dd->cspec->iblnkerrdelta || dd->cspec->ibdeltainprog) {
+			val = read_6120_creg32(dd, cr_iblinkerrrecov);
+			if (dd->cspec->ibdeltainprog)
+				val -= val - dd->cspec->iblnkerrsnap;
+			val -= dd->cspec->iblnkerrdelta;
+			write_6120_creg(dd, cr_iblinkerrrecov, val);
+		}
+
+		/* and disable counter writes */
+		qib_write_kreg(dd, kr_hwdiagctrl, diagc);
+	}
+
+	val = qib_read_kreg64(dd, kr_serdes_cfg0);
+	val |= SYM_MASK(SerdesCfg0, TxIdeEnX);
+	qib_write_kreg(dd, kr_serdes_cfg0, val);
+}
+
+/**
+ * qib_6120_setup_setextled - set the state of the two external LEDs
+ * @dd: the qlogic_ib device
+ * @on: whether the link is up or not
+ *
+ * The exact combo of LEDs if on is true is determined by looking
+ * at the ibcstatus.
+
+ * These LEDs indicate the physical and logical state of IB link.
+ * For this chip (at least with recommended board pinouts), LED1
+ * is Yellow (logical state) and LED2 is Green (physical state),
+ *
+ * Note:  We try to match the Mellanox HCA LED behavior as best
+ * we can.  Green indicates physical link state is OK (something is
+ * plugged in, and we can train).
+ * Amber indicates the link is logically up (ACTIVE).
+ * Mellanox further blinks the amber LED to indicate data packet
+ * activity, but we have no hardware support for that, so it would
+ * require waking up every 10-20 msecs and checking the counters
+ * on the chip, and then turning the LED off if appropriate.  That's
+ * visible overhead, so not something we will do.
+ *
+ */
+static void qib_6120_setup_setextled(struct qib_pportdata *ppd, u32 on)
+{
+	u64 extctl, val, lst, ltst;
+	unsigned long flags;
+	struct qib_devdata *dd = ppd->dd;
+
+	/*
+	 * The diags use the LED to indicate diag info, so we leave
+	 * the external LED alone when the diags are running.
+	 */
+	if (dd->diag_client)
+		return;
+
+	/* Allow override of LED display for, e.g. Locating system in rack */
+	if (ppd->led_override) {
+		ltst = (ppd->led_override & QIB_LED_PHYS) ?
+			IB_PHYSPORTSTATE_LINKUP : IB_PHYSPORTSTATE_DISABLED,
+		lst = (ppd->led_override & QIB_LED_LOG) ?
+			IB_PORT_ACTIVE : IB_PORT_DOWN;
+	} else if (on) {
+		val = qib_read_kreg64(dd, kr_ibcstatus);
+		ltst = qib_6120_phys_portstate(val);
+		lst = qib_6120_iblink_state(val);
+	} else {
+		ltst = 0;
+		lst = 0;
+	}
+
+	spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+	extctl = dd->cspec->extctrl & ~(SYM_MASK(EXTCtrl, LEDPriPortGreenOn) |
+				 SYM_MASK(EXTCtrl, LEDPriPortYellowOn));
+
+	if (ltst == IB_PHYSPORTSTATE_LINKUP)
+		extctl |= SYM_MASK(EXTCtrl, LEDPriPortYellowOn);
+	if (lst == IB_PORT_ACTIVE)
+		extctl |= SYM_MASK(EXTCtrl, LEDPriPortGreenOn);
+	dd->cspec->extctrl = extctl;
+	qib_write_kreg(dd, kr_extctrl, extctl);
+	spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+}
+
+static void qib_6120_free_irq(struct qib_devdata *dd)
+{
+	if (dd->cspec->irq) {
+		free_irq(dd->cspec->irq, dd);
+		dd->cspec->irq = 0;
+	}
+	qib_nomsi(dd);
+}
+
+/**
+ * qib_6120_setup_cleanup - clean up any per-chip chip-specific stuff
+ * @dd: the qlogic_ib device
+ *
+ * This is called during driver unload.
+*/
+static void qib_6120_setup_cleanup(struct qib_devdata *dd)
+{
+	qib_6120_free_irq(dd);
+	kfree(dd->cspec->cntrs);
+	kfree(dd->cspec->portcntrs);
+	if (dd->cspec->dummy_hdrq) {
+		dma_free_coherent(&dd->pcidev->dev,
+				  ALIGN(dd->rcvhdrcnt *
+					dd->rcvhdrentsize *
+					sizeof(u32), PAGE_SIZE),
+				  dd->cspec->dummy_hdrq,
+				  dd->cspec->dummy_hdrq_phys);
+		dd->cspec->dummy_hdrq = NULL;
+	}
+}
+
+static void qib_wantpiobuf_6120_intr(struct qib_devdata *dd, u32 needint)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+	if (needint)
+		dd->sendctrl |= SYM_MASK(SendCtrl, PIOIntBufAvail);
+	else
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, PIOIntBufAvail);
+	qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+}
+
+/*
+ * handle errors and unusual events first, separate function
+ * to improve cache hits for fast path interrupt handling
+ */
+static noinline void unlikely_6120_intr(struct qib_devdata *dd, u64 istat)
+{
+	if (unlikely(istat & ~QLOGIC_IB_I_BITSEXTANT))
+		qib_dev_err(dd, "interrupt with unknown interrupts %Lx set\n",
+			    istat & ~QLOGIC_IB_I_BITSEXTANT);
+
+	if (istat & QLOGIC_IB_I_ERROR) {
+		u64 estat = 0;
+
+		qib_stats.sps_errints++;
+		estat = qib_read_kreg64(dd, kr_errstatus);
+		if (!estat)
+			qib_devinfo(dd->pcidev,
+				"error interrupt (%Lx), but no error bits set!\n",
+				istat);
+		handle_6120_errors(dd, estat);
+	}
+
+	if (istat & QLOGIC_IB_I_GPIO) {
+		u32 gpiostatus;
+		u32 to_clear = 0;
+
+		/*
+		 * GPIO_3..5 on IBA6120 Rev2 chips indicate
+		 * errors that we need to count.
+		 */
+		gpiostatus = qib_read_kreg32(dd, kr_gpio_status);
+		/* First the error-counter case. */
+		if (gpiostatus & GPIO_ERRINTR_MASK) {
+			/* want to clear the bits we see asserted. */
+			to_clear |= (gpiostatus & GPIO_ERRINTR_MASK);
+
+			/*
+			 * Count appropriately, clear bits out of our copy,
+			 * as they have been "handled".
+			 */
+			if (gpiostatus & (1 << GPIO_RXUVL_BIT))
+				dd->cspec->rxfc_unsupvl_errs++;
+			if (gpiostatus & (1 << GPIO_OVRUN_BIT))
+				dd->cspec->overrun_thresh_errs++;
+			if (gpiostatus & (1 << GPIO_LLI_BIT))
+				dd->cspec->lli_errs++;
+			gpiostatus &= ~GPIO_ERRINTR_MASK;
+		}
+		if (gpiostatus) {
+			/*
+			 * Some unexpected bits remain. If they could have
+			 * caused the interrupt, complain and clear.
+			 * To avoid repetition of this condition, also clear
+			 * the mask. It is almost certainly due to error.
+			 */
+			const u32 mask = qib_read_kreg32(dd, kr_gpio_mask);
+
+			/*
+			 * Also check that the chip reflects our shadow,
+			 * and report issues, If they caused the interrupt.
+			 * we will suppress by refreshing from the shadow.
+			 */
+			if (mask & gpiostatus) {
+				to_clear |= (gpiostatus & mask);
+				dd->cspec->gpio_mask &= ~(gpiostatus & mask);
+				qib_write_kreg(dd, kr_gpio_mask,
+					       dd->cspec->gpio_mask);
+			}
+		}
+		if (to_clear)
+			qib_write_kreg(dd, kr_gpio_clear, (u64) to_clear);
+	}
+}
+
+static irqreturn_t qib_6120intr(int irq, void *data)
+{
+	struct qib_devdata *dd = data;
+	irqreturn_t ret;
+	u32 istat, ctxtrbits, rmask, crcs = 0;
+	unsigned i;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT) {
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		ret = IRQ_HANDLED;
+		goto bail;
+	}
+
+	istat = qib_read_kreg32(dd, kr_intstatus);
+
+	if (unlikely(!istat)) {
+		ret = IRQ_NONE; /* not our interrupt, or already handled */
+		goto bail;
+	}
+	if (unlikely(istat == -1)) {
+		qib_bad_intrstatus(dd);
+		/* don't know if it was our interrupt or not */
+		ret = IRQ_NONE;
+		goto bail;
+	}
+
+	this_cpu_inc(*dd->int_counter);
+
+	if (unlikely(istat & (~QLOGIC_IB_I_BITSEXTANT |
+			      QLOGIC_IB_I_GPIO | QLOGIC_IB_I_ERROR)))
+		unlikely_6120_intr(dd, istat);
+
+	/*
+	 * Clear the interrupt bits we found set, relatively early, so we
+	 * "know" know the chip will have seen this by the time we process
+	 * the queue, and will re-interrupt if necessary.  The processor
+	 * itself won't take the interrupt again until we return.
+	 */
+	qib_write_kreg(dd, kr_intclear, istat);
+
+	/*
+	 * Handle kernel receive queues before checking for pio buffers
+	 * available since receives can overflow; piobuf waiters can afford
+	 * a few extra cycles, since they were waiting anyway.
+	 */
+	ctxtrbits = istat &
+		((QLOGIC_IB_I_RCVAVAIL_MASK << QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+		 (QLOGIC_IB_I_RCVURG_MASK << QLOGIC_IB_I_RCVURG_SHIFT));
+	if (ctxtrbits) {
+		rmask = (1U << QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+			(1U << QLOGIC_IB_I_RCVURG_SHIFT);
+		for (i = 0; i < dd->first_user_ctxt; i++) {
+			if (ctxtrbits & rmask) {
+				ctxtrbits &= ~rmask;
+				crcs += qib_kreceive(dd->rcd[i],
+						     &dd->cspec->lli_counter,
+						     NULL);
+			}
+			rmask <<= 1;
+		}
+		if (crcs) {
+			u32 cntr = dd->cspec->lli_counter;
+			cntr += crcs;
+			if (cntr) {
+				if (cntr > dd->cspec->lli_thresh) {
+					dd->cspec->lli_counter = 0;
+					dd->cspec->lli_errs++;
+				} else
+					dd->cspec->lli_counter += cntr;
+			}
+		}
+
+
+		if (ctxtrbits) {
+			ctxtrbits =
+				(ctxtrbits >> QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+				(ctxtrbits >> QLOGIC_IB_I_RCVURG_SHIFT);
+			qib_handle_urcv(dd, ctxtrbits);
+		}
+	}
+
+	if ((istat & QLOGIC_IB_I_SPIOBUFAVAIL) && (dd->flags & QIB_INITTED))
+		qib_ib_piobufavail(dd);
+
+	ret = IRQ_HANDLED;
+bail:
+	return ret;
+}
+
+/*
+ * Set up our chip-specific interrupt handler
+ * The interrupt type has already been setup, so
+ * we just need to do the registration and error checking.
+ */
+static void qib_setup_6120_interrupt(struct qib_devdata *dd)
+{
+	/*
+	 * If the chip supports added error indication via GPIO pins,
+	 * enable interrupts on those bits so the interrupt routine
+	 * can count the events. Also set flag so interrupt routine
+	 * can know they are expected.
+	 */
+	if (SYM_FIELD(dd->revision, Revision_R,
+		      ChipRevMinor) > 1) {
+		/* Rev2+ reports extra errors via internal GPIO pins */
+		dd->cspec->gpio_mask |= GPIO_ERRINTR_MASK;
+		qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+	}
+
+	if (!dd->cspec->irq)
+		qib_dev_err(dd,
+			"irq is 0, BIOS error?  Interrupts won't work\n");
+	else {
+		int ret;
+		ret = request_irq(dd->cspec->irq, qib_6120intr, 0,
+				  QIB_DRV_NAME, dd);
+		if (ret)
+			qib_dev_err(dd,
+				"Couldn't setup interrupt (irq=%d): %d\n",
+				dd->cspec->irq, ret);
+	}
+}
+
+/**
+ * pe_boardname - fill in the board name
+ * @dd: the qlogic_ib device
+ *
+ * info is based on the board revision register
+ */
+static void pe_boardname(struct qib_devdata *dd)
+{
+	char *n;
+	u32 boardid, namelen;
+
+	boardid = SYM_FIELD(dd->revision, Revision,
+			    BoardID);
+
+	switch (boardid) {
+	case 2:
+		n = "InfiniPath_QLE7140";
+		break;
+	default:
+		qib_dev_err(dd, "Unknown 6120 board with ID %u\n", boardid);
+		n = "Unknown_InfiniPath_6120";
+		break;
+	}
+	namelen = strlen(n) + 1;
+	dd->boardname = kmalloc(namelen, GFP_KERNEL);
+	if (!dd->boardname)
+		qib_dev_err(dd, "Failed allocation for board name: %s\n", n);
+	else
+		snprintf(dd->boardname, namelen, "%s", n);
+
+	if (dd->majrev != 4 || !dd->minrev || dd->minrev > 2)
+		qib_dev_err(dd,
+			"Unsupported InfiniPath hardware revision %u.%u!\n",
+			dd->majrev, dd->minrev);
+
+	snprintf(dd->boardversion, sizeof(dd->boardversion),
+		 "ChipABI %u.%u, %s, InfiniPath%u %u.%u, SW Compat %u\n",
+		 QIB_CHIP_VERS_MAJ, QIB_CHIP_VERS_MIN, dd->boardname,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, Arch),
+		 dd->majrev, dd->minrev,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, SW));
+
+}
+
+/*
+ * This routine sleeps, so it can only be called from user context, not
+ * from interrupt context.  If we need interrupt context, we can split
+ * it into two routines.
+ */
+static int qib_6120_setup_reset(struct qib_devdata *dd)
+{
+	u64 val;
+	int i;
+	int ret;
+	u16 cmdval;
+	u8 int_line, clinesz;
+
+	qib_pcie_getcmd(dd, &cmdval, &int_line, &clinesz);
+
+	/* Use ERROR so it shows up in logs, etc. */
+	qib_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->unit);
+
+	/* no interrupts till re-initted */
+	qib_6120_set_intr_state(dd, 0);
+
+	dd->cspec->ibdeltainprog = 0;
+	dd->cspec->ibsymdelta = 0;
+	dd->cspec->iblnkerrdelta = 0;
+
+	/*
+	 * Keep chip from being accessed until we are ready.  Use
+	 * writeq() directly, to allow the write even though QIB_PRESENT
+	 * isn't set.
+	 */
+	dd->flags &= ~(QIB_INITTED | QIB_PRESENT);
+	/* so we check interrupts work again */
+	dd->z_int_counter = qib_int_counter(dd);
+	val = dd->control | QLOGIC_IB_C_RESET;
+	writeq(val, &dd->kregbase[kr_control]);
+	mb(); /* prevent compiler re-ordering around actual reset */
+
+	for (i = 1; i <= 5; i++) {
+		/*
+		 * Allow MBIST, etc. to complete; longer on each retry.
+		 * We sometimes get machine checks from bus timeout if no
+		 * response, so for now, make it *really* long.
+		 */
+		msleep(1000 + (1 + i) * 2000);
+
+		qib_pcie_reenable(dd, cmdval, int_line, clinesz);
+
+		/*
+		 * Use readq directly, so we don't need to mark it as PRESENT
+		 * until we get a successful indication that all is well.
+		 */
+		val = readq(&dd->kregbase[kr_revision]);
+		if (val == dd->revision) {
+			dd->flags |= QIB_PRESENT; /* it's back */
+			ret = qib_reinit_intr(dd);
+			goto bail;
+		}
+	}
+	ret = 0; /* failed */
+
+bail:
+	if (ret) {
+		if (qib_pcie_params(dd, dd->lbus_width, NULL, NULL))
+			qib_dev_err(dd,
+				"Reset failed to setup PCIe or interrupts; continuing anyway\n");
+		/* clear the reset error, init error/hwerror mask */
+		qib_6120_init_hwerrors(dd);
+		/* for Rev2 error interrupts; nop for rev 1 */
+		qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+		/* clear the reset error, init error/hwerror mask */
+		qib_6120_init_hwerrors(dd);
+	}
+	return ret;
+}
+
+/**
+ * qib_6120_put_tid - write a TID in chip
+ * @dd: the qlogic_ib device
+ * @tidptr: pointer to the expected TID (in chip) to update
+ * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0)
+ * for expected
+ * @pa: physical address of in memory buffer; tidinvalid if freeing
+ *
+ * This exists as a separate routine to allow for special locking etc.
+ * It's used for both the full cleanup on exit, as well as the normal
+ * setup and teardown.
+ */
+static void qib_6120_put_tid(struct qib_devdata *dd, u64 __iomem *tidptr,
+			     u32 type, unsigned long pa)
+{
+	u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
+	unsigned long flags;
+	int tidx;
+	spinlock_t *tidlockp; /* select appropriate spinlock */
+
+	if (!dd->kregbase)
+		return;
+
+	if (pa != dd->tidinvalid) {
+		if (pa & ((1U << 11) - 1)) {
+			qib_dev_err(dd, "Physaddr %lx not 2KB aligned!\n",
+				    pa);
+			return;
+		}
+		pa >>= 11;
+		if (pa & ~QLOGIC_IB_RT_ADDR_MASK) {
+			qib_dev_err(dd,
+				"Physical page address 0x%lx larger than supported\n",
+				pa);
+			return;
+		}
+
+		if (type == RCVHQ_RCV_TYPE_EAGER)
+			pa |= dd->tidtemplate;
+		else /* for now, always full 4KB page */
+			pa |= 2 << 29;
+	}
+
+	/*
+	 * Avoid chip issue by writing the scratch register
+	 * before and after the TID, and with an io write barrier.
+	 * We use a spinlock around the writes, so they can't intermix
+	 * with other TID (eager or expected) writes (the chip problem
+	 * is triggered by back to back TID writes). Unfortunately, this
+	 * call can be done from interrupt level for the ctxt 0 eager TIDs,
+	 * so we have to use irqsave locks.
+	 */
+	/*
+	 * Assumes tidptr always > egrtidbase
+	 * if type == RCVHQ_RCV_TYPE_EAGER.
+	 */
+	tidx = tidptr - dd->egrtidbase;
+
+	tidlockp = (type == RCVHQ_RCV_TYPE_EAGER && tidx < dd->rcvhdrcnt)
+		? &dd->cspec->kernel_tid_lock : &dd->cspec->user_tid_lock;
+	spin_lock_irqsave(tidlockp, flags);
+	qib_write_kreg(dd, kr_scratch, 0xfeeddeaf);
+	writel(pa, tidp32);
+	qib_write_kreg(dd, kr_scratch, 0xdeadbeef);
+	mmiowb();
+	spin_unlock_irqrestore(tidlockp, flags);
+}
+
+/**
+ * qib_6120_put_tid_2 - write a TID in chip, Revision 2 or higher
+ * @dd: the qlogic_ib device
+ * @tidptr: pointer to the expected TID (in chip) to update
+ * @tidtype: RCVHQ_RCV_TYPE_EAGER (1) for eager, RCVHQ_RCV_TYPE_EXPECTED (0)
+ * for expected
+ * @pa: physical address of in memory buffer; tidinvalid if freeing
+ *
+ * This exists as a separate routine to allow for selection of the
+ * appropriate "flavor". The static calls in cleanup just use the
+ * revision-agnostic form, as they are not performance critical.
+ */
+static void qib_6120_put_tid_2(struct qib_devdata *dd, u64 __iomem *tidptr,
+			       u32 type, unsigned long pa)
+{
+	u32 __iomem *tidp32 = (u32 __iomem *)tidptr;
+	u32 tidx;
+
+	if (!dd->kregbase)
+		return;
+
+	if (pa != dd->tidinvalid) {
+		if (pa & ((1U << 11) - 1)) {
+			qib_dev_err(dd, "Physaddr %lx not 2KB aligned!\n",
+				    pa);
+			return;
+		}
+		pa >>= 11;
+		if (pa & ~QLOGIC_IB_RT_ADDR_MASK) {
+			qib_dev_err(dd,
+				"Physical page address 0x%lx larger than supported\n",
+				pa);
+			return;
+		}
+
+		if (type == RCVHQ_RCV_TYPE_EAGER)
+			pa |= dd->tidtemplate;
+		else /* for now, always full 4KB page */
+			pa |= 2 << 29;
+	}
+	tidx = tidptr - dd->egrtidbase;
+	writel(pa, tidp32);
+	mmiowb();
+}
+
+
+/**
+ * qib_6120_clear_tids - clear all TID entries for a context, expected and eager
+ * @dd: the qlogic_ib device
+ * @ctxt: the context
+ *
+ * clear all TID entries for a context, expected and eager.
+ * Used from qib_close().  On this chip, TIDs are only 32 bits,
+ * not 64, but they are still on 64 bit boundaries, so tidbase
+ * is declared as u64 * for the pointer math, even though we write 32 bits
+ */
+static void qib_6120_clear_tids(struct qib_devdata *dd,
+				struct qib_ctxtdata *rcd)
+{
+	u64 __iomem *tidbase;
+	unsigned long tidinv;
+	u32 ctxt;
+	int i;
+
+	if (!dd->kregbase || !rcd)
+		return;
+
+	ctxt = rcd->ctxt;
+
+	tidinv = dd->tidinvalid;
+	tidbase = (u64 __iomem *)
+		((char __iomem *)(dd->kregbase) +
+		 dd->rcvtidbase +
+		 ctxt * dd->rcvtidcnt * sizeof(*tidbase));
+
+	for (i = 0; i < dd->rcvtidcnt; i++)
+		/* use func pointer because could be one of two funcs */
+		dd->f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
+				  tidinv);
+
+	tidbase = (u64 __iomem *)
+		((char __iomem *)(dd->kregbase) +
+		 dd->rcvegrbase +
+		 rcd->rcvegr_tid_base * sizeof(*tidbase));
+
+	for (i = 0; i < rcd->rcvegrcnt; i++)
+		/* use func pointer because could be one of two funcs */
+		dd->f_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
+				  tidinv);
+}
+
+/**
+ * qib_6120_tidtemplate - setup constants for TID updates
+ * @dd: the qlogic_ib device
+ *
+ * We setup stuff that we use a lot, to avoid calculating each time
+ */
+static void qib_6120_tidtemplate(struct qib_devdata *dd)
+{
+	u32 egrsize = dd->rcvegrbufsize;
+
+	/*
+	 * For now, we always allocate 4KB buffers (at init) so we can
+	 * receive max size packets.  We may want a module parameter to
+	 * specify 2KB or 4KB and/or make be per ctxt instead of per device
+	 * for those who want to reduce memory footprint.  Note that the
+	 * rcvhdrentsize size must be large enough to hold the largest
+	 * IB header (currently 96 bytes) that we expect to handle (plus of
+	 * course the 2 dwords of RHF).
+	 */
+	if (egrsize == 2048)
+		dd->tidtemplate = 1U << 29;
+	else if (egrsize == 4096)
+		dd->tidtemplate = 2U << 29;
+	dd->tidinvalid = 0;
+}
+
+int __attribute__((weak)) qib_unordered_wc(void)
+{
+	return 0;
+}
+
+/**
+ * qib_6120_get_base_info - set chip-specific flags for user code
+ * @rcd: the qlogic_ib ctxt
+ * @kbase: qib_base_info pointer
+ *
+ * We set the PCIE flag because the lower bandwidth on PCIe vs
+ * HyperTransport can affect some user packet algorithms.
+ */
+static int qib_6120_get_base_info(struct qib_ctxtdata *rcd,
+				  struct qib_base_info *kinfo)
+{
+	if (qib_unordered_wc())
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_FORCE_WC_ORDER;
+
+	kinfo->spi_runtime_flags |= QIB_RUNTIME_PCIE |
+		QIB_RUNTIME_FORCE_PIOAVAIL | QIB_RUNTIME_PIO_REGSWAPPED;
+	return 0;
+}
+
+
+static struct qib_message_header *
+qib_6120_get_msgheader(struct qib_devdata *dd, __le32 *rhf_addr)
+{
+	return (struct qib_message_header *)
+		&rhf_addr[sizeof(u64) / sizeof(u32)];
+}
+
+static void qib_6120_config_ctxts(struct qib_devdata *dd)
+{
+	dd->ctxtcnt = qib_read_kreg32(dd, kr_portcnt);
+	if (qib_n_krcv_queues > 1) {
+		dd->first_user_ctxt = qib_n_krcv_queues * dd->num_pports;
+		if (dd->first_user_ctxt > dd->ctxtcnt)
+			dd->first_user_ctxt = dd->ctxtcnt;
+		dd->qpn_mask = dd->first_user_ctxt <= 2 ? 2 : 6;
+	} else
+		dd->first_user_ctxt = dd->num_pports;
+	dd->n_krcv_queues = dd->first_user_ctxt;
+}
+
+static void qib_update_6120_usrhead(struct qib_ctxtdata *rcd, u64 hd,
+				    u32 updegr, u32 egrhd, u32 npkts)
+{
+	if (updegr)
+		qib_write_ureg(rcd->dd, ur_rcvegrindexhead, egrhd, rcd->ctxt);
+	mmiowb();
+	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
+	mmiowb();
+}
+
+static u32 qib_6120_hdrqempty(struct qib_ctxtdata *rcd)
+{
+	u32 head, tail;
+
+	head = qib_read_ureg32(rcd->dd, ur_rcvhdrhead, rcd->ctxt);
+	if (rcd->rcvhdrtail_kvaddr)
+		tail = qib_get_rcvhdrtail(rcd);
+	else
+		tail = qib_read_ureg32(rcd->dd, ur_rcvhdrtail, rcd->ctxt);
+	return head == tail;
+}
+
+/*
+ * Used when we close any ctxt, for DMA already in flight
+ * at close.  Can't be done until we know hdrq size, so not
+ * early in chip init.
+ */
+static void alloc_dummy_hdrq(struct qib_devdata *dd)
+{
+	dd->cspec->dummy_hdrq = dma_alloc_coherent(&dd->pcidev->dev,
+					dd->rcd[0]->rcvhdrq_size,
+					&dd->cspec->dummy_hdrq_phys,
+					GFP_ATOMIC | __GFP_COMP);
+	if (!dd->cspec->dummy_hdrq) {
+		qib_devinfo(dd->pcidev, "Couldn't allocate dummy hdrq\n");
+		/* fallback to just 0'ing */
+		dd->cspec->dummy_hdrq_phys = 0UL;
+	}
+}
+
+/*
+ * Modify the RCVCTRL register in chip-specific way. This
+ * is a function because bit positions and (future) register
+ * location is chip-specific, but the needed operations are
+ * generic. <op> is a bit-mask because we often want to
+ * do multiple modifications.
+ */
+static void rcvctrl_6120_mod(struct qib_pportdata *ppd, unsigned int op,
+			     int ctxt)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 mask, val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+
+	if (op & QIB_RCVCTRL_TAILUPD_ENB)
+		dd->rcvctrl |= (1ULL << QLOGIC_IB_R_TAILUPD_SHIFT);
+	if (op & QIB_RCVCTRL_TAILUPD_DIS)
+		dd->rcvctrl &= ~(1ULL << QLOGIC_IB_R_TAILUPD_SHIFT);
+	if (op & QIB_RCVCTRL_PKEY_ENB)
+		dd->rcvctrl &= ~(1ULL << IBA6120_R_PKEY_DIS_SHIFT);
+	if (op & QIB_RCVCTRL_PKEY_DIS)
+		dd->rcvctrl |= (1ULL << IBA6120_R_PKEY_DIS_SHIFT);
+	if (ctxt < 0)
+		mask = (1ULL << dd->ctxtcnt) - 1;
+	else
+		mask = (1ULL << ctxt);
+	if (op & QIB_RCVCTRL_CTXT_ENB) {
+		/* always done for specific ctxt */
+		dd->rcvctrl |= (mask << SYM_LSB(RcvCtrl, PortEnable));
+		if (!(dd->flags & QIB_NODMA_RTAIL))
+			dd->rcvctrl |= 1ULL << QLOGIC_IB_R_TAILUPD_SHIFT;
+		/* Write these registers before the context is enabled. */
+		qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr, ctxt,
+			dd->rcd[ctxt]->rcvhdrqtailaddr_phys);
+		qib_write_kreg_ctxt(dd, kr_rcvhdraddr, ctxt,
+			dd->rcd[ctxt]->rcvhdrq_phys);
+
+		if (ctxt == 0 && !dd->cspec->dummy_hdrq)
+			alloc_dummy_hdrq(dd);
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS)
+		dd->rcvctrl &= ~(mask << SYM_LSB(RcvCtrl, PortEnable));
+	if (op & QIB_RCVCTRL_INTRAVAIL_ENB)
+		dd->rcvctrl |= (mask << QLOGIC_IB_R_INTRAVAIL_SHIFT);
+	if (op & QIB_RCVCTRL_INTRAVAIL_DIS)
+		dd->rcvctrl &= ~(mask << QLOGIC_IB_R_INTRAVAIL_SHIFT);
+	qib_write_kreg(dd, kr_rcvctrl, dd->rcvctrl);
+	if ((op & QIB_RCVCTRL_INTRAVAIL_ENB) && dd->rhdrhead_intr_off) {
+		/* arm rcv interrupt */
+		val = qib_read_ureg32(dd, ur_rcvhdrhead, ctxt) |
+			dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	}
+	if (op & QIB_RCVCTRL_CTXT_ENB) {
+		/*
+		 * Init the context registers also; if we were
+		 * disabled, tail and head should both be zero
+		 * already from the enable, but since we don't
+		 * know, we have to do it explicitly.
+		 */
+		val = qib_read_ureg32(dd, ur_rcvegrindextail, ctxt);
+		qib_write_ureg(dd, ur_rcvegrindexhead, val, ctxt);
+
+		val = qib_read_ureg32(dd, ur_rcvhdrtail, ctxt);
+		dd->rcd[ctxt]->head = val;
+		/* If kctxt, interrupt on next receive. */
+		if (ctxt < dd->first_user_ctxt)
+			val |= dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS) {
+		/*
+		 * Be paranoid, and never write 0's to these, just use an
+		 * unused page.  Of course,
+		 * rcvhdraddr points to a large chunk of memory, so this
+		 * could still trash things, but at least it won't trash
+		 * page 0, and by disabling the ctxt, it should stop "soon",
+		 * even if a packet or two is in already in flight after we
+		 * disabled the ctxt.  Only 6120 has this issue.
+		 */
+		if (ctxt >= 0) {
+			qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr, ctxt,
+					    dd->cspec->dummy_hdrq_phys);
+			qib_write_kreg_ctxt(dd, kr_rcvhdraddr, ctxt,
+					    dd->cspec->dummy_hdrq_phys);
+		} else {
+			unsigned i;
+
+			for (i = 0; i < dd->cfgctxts; i++) {
+				qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr,
+					    i, dd->cspec->dummy_hdrq_phys);
+				qib_write_kreg_ctxt(dd, kr_rcvhdraddr,
+					    i, dd->cspec->dummy_hdrq_phys);
+			}
+		}
+	}
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+}
+
+/*
+ * Modify the SENDCTRL register in chip-specific way. This
+ * is a function there may be multiple such registers with
+ * slightly different layouts. Only operations actually used
+ * are implemented yet.
+ * Chip requires no back-back sendctrl writes, so write
+ * scratch register after writing sendctrl
+ */
+static void sendctrl_6120_mod(struct qib_pportdata *ppd, u32 op)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 tmp_dd_sendctrl;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+
+	/* First the ones that are "sticky", saved in shadow */
+	if (op & QIB_SENDCTRL_CLEAR)
+		dd->sendctrl = 0;
+	if (op & QIB_SENDCTRL_SEND_DIS)
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, PIOEnable);
+	else if (op & QIB_SENDCTRL_SEND_ENB)
+		dd->sendctrl |= SYM_MASK(SendCtrl, PIOEnable);
+	if (op & QIB_SENDCTRL_AVAIL_DIS)
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, PIOBufAvailUpd);
+	else if (op & QIB_SENDCTRL_AVAIL_ENB)
+		dd->sendctrl |= SYM_MASK(SendCtrl, PIOBufAvailUpd);
+
+	if (op & QIB_SENDCTRL_DISARM_ALL) {
+		u32 i, last;
+
+		tmp_dd_sendctrl = dd->sendctrl;
+		/*
+		 * disarm any that are not yet launched, disabling sends
+		 * and updates until done.
+		 */
+		last = dd->piobcnt2k + dd->piobcnt4k;
+		tmp_dd_sendctrl &=
+			~(SYM_MASK(SendCtrl, PIOEnable) |
+			  SYM_MASK(SendCtrl, PIOBufAvailUpd));
+		for (i = 0; i < last; i++) {
+			qib_write_kreg(dd, kr_sendctrl, tmp_dd_sendctrl |
+				       SYM_MASK(SendCtrl, Disarm) | i);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+	}
+
+	tmp_dd_sendctrl = dd->sendctrl;
+
+	if (op & QIB_SENDCTRL_FLUSH)
+		tmp_dd_sendctrl |= SYM_MASK(SendCtrl, Abort);
+	if (op & QIB_SENDCTRL_DISARM)
+		tmp_dd_sendctrl |= SYM_MASK(SendCtrl, Disarm) |
+			((op & QIB_6120_SendCtrl_DisarmPIOBuf_RMASK) <<
+			 SYM_LSB(SendCtrl, DisarmPIOBuf));
+	if (op & QIB_SENDCTRL_AVAIL_BLIP)
+		tmp_dd_sendctrl &= ~SYM_MASK(SendCtrl, PIOBufAvailUpd);
+
+	qib_write_kreg(dd, kr_sendctrl, tmp_dd_sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	if (op & QIB_SENDCTRL_AVAIL_BLIP) {
+		qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+
+	if (op & QIB_SENDCTRL_FLUSH) {
+		u32 v;
+		/*
+		 * ensure writes have hit chip, then do a few
+		 * more reads, to allow DMA of pioavail registers
+		 * to occur, so in-memory copy is in sync with
+		 * the chip.  Not always safe to sleep.
+		 */
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		qib_read_kreg32(dd, kr_scratch);
+	}
+}
+
+/**
+ * qib_portcntr_6120 - read a per-port counter
+ * @dd: the qlogic_ib device
+ * @creg: the counter to snapshot
+ */
+static u64 qib_portcntr_6120(struct qib_pportdata *ppd, u32 reg)
+{
+	u64 ret = 0ULL;
+	struct qib_devdata *dd = ppd->dd;
+	u16 creg;
+	/* 0xffff for unimplemented or synthesized counters */
+	static const u16 xlator[] = {
+		[QIBPORTCNTR_PKTSEND] = cr_pktsend,
+		[QIBPORTCNTR_WORDSEND] = cr_wordsend,
+		[QIBPORTCNTR_PSXMITDATA] = 0xffff,
+		[QIBPORTCNTR_PSXMITPKTS] = 0xffff,
+		[QIBPORTCNTR_PSXMITWAIT] = 0xffff,
+		[QIBPORTCNTR_SENDSTALL] = cr_sendstall,
+		[QIBPORTCNTR_PKTRCV] = cr_pktrcv,
+		[QIBPORTCNTR_PSRCVDATA] = 0xffff,
+		[QIBPORTCNTR_PSRCVPKTS] = 0xffff,
+		[QIBPORTCNTR_RCVEBP] = cr_rcvebp,
+		[QIBPORTCNTR_RCVOVFL] = cr_rcvovfl,
+		[QIBPORTCNTR_WORDRCV] = cr_wordrcv,
+		[QIBPORTCNTR_RXDROPPKT] = cr_rxdroppkt,
+		[QIBPORTCNTR_RXLOCALPHYERR] = 0xffff,
+		[QIBPORTCNTR_RXVLERR] = 0xffff,
+		[QIBPORTCNTR_ERRICRC] = cr_erricrc,
+		[QIBPORTCNTR_ERRVCRC] = cr_errvcrc,
+		[QIBPORTCNTR_ERRLPCRC] = cr_errlpcrc,
+		[QIBPORTCNTR_BADFORMAT] = cr_badformat,
+		[QIBPORTCNTR_ERR_RLEN] = cr_err_rlen,
+		[QIBPORTCNTR_IBSYMBOLERR] = cr_ibsymbolerr,
+		[QIBPORTCNTR_INVALIDRLEN] = cr_invalidrlen,
+		[QIBPORTCNTR_UNSUPVL] = cr_txunsupvl,
+		[QIBPORTCNTR_EXCESSBUFOVFL] = 0xffff,
+		[QIBPORTCNTR_ERRLINK] = cr_errlink,
+		[QIBPORTCNTR_IBLINKDOWN] = cr_iblinkdown,
+		[QIBPORTCNTR_IBLINKERRRECOV] = cr_iblinkerrrecov,
+		[QIBPORTCNTR_LLI] = 0xffff,
+		[QIBPORTCNTR_PSINTERVAL] = 0xffff,
+		[QIBPORTCNTR_PSSTART] = 0xffff,
+		[QIBPORTCNTR_PSSTAT] = 0xffff,
+		[QIBPORTCNTR_VL15PKTDROP] = 0xffff,
+		[QIBPORTCNTR_ERRPKEY] = cr_errpkey,
+		[QIBPORTCNTR_KHDROVFL] = 0xffff,
+	};
+
+	if (reg >= ARRAY_SIZE(xlator)) {
+		qib_devinfo(ppd->dd->pcidev,
+			 "Unimplemented portcounter %u\n", reg);
+		goto done;
+	}
+	creg = xlator[reg];
+
+	/* handle counters requests not implemented as chip counters */
+	if (reg == QIBPORTCNTR_LLI)
+		ret = dd->cspec->lli_errs;
+	else if (reg == QIBPORTCNTR_EXCESSBUFOVFL)
+		ret = dd->cspec->overrun_thresh_errs;
+	else if (reg == QIBPORTCNTR_KHDROVFL) {
+		int i;
+
+		/* sum over all kernel contexts */
+		for (i = 0; i < dd->first_user_ctxt; i++)
+			ret += read_6120_creg32(dd, cr_portovfl + i);
+	} else if (reg == QIBPORTCNTR_PSSTAT)
+		ret = dd->cspec->pma_sample_status;
+	if (creg == 0xffff)
+		goto done;
+
+	/*
+	 * only fast incrementing counters are 64bit; use 32 bit reads to
+	 * avoid two independent reads when on opteron
+	 */
+	if (creg == cr_wordsend || creg == cr_wordrcv ||
+	    creg == cr_pktsend || creg == cr_pktrcv)
+		ret = read_6120_creg(dd, creg);
+	else
+		ret = read_6120_creg32(dd, creg);
+	if (creg == cr_ibsymbolerr) {
+		if (dd->cspec->ibdeltainprog)
+			ret -= ret - dd->cspec->ibsymsnap;
+		ret -= dd->cspec->ibsymdelta;
+	} else if (creg == cr_iblinkerrrecov) {
+		if (dd->cspec->ibdeltainprog)
+			ret -= ret - dd->cspec->iblnkerrsnap;
+		ret -= dd->cspec->iblnkerrdelta;
+	}
+	if (reg == QIBPORTCNTR_RXDROPPKT) /* add special cased count */
+		ret += dd->cspec->rxfc_unsupvl_errs;
+
+done:
+	return ret;
+}
+
+/*
+ * Device counter names (not port-specific), one line per stat,
+ * single string.  Used by utilities like ipathstats to print the stats
+ * in a way which works for different versions of drivers, without changing
+ * the utility.  Names need to be 12 chars or less (w/o newline), for proper
+ * display by utility.
+ * Non-error counters are first.
+ * Start of "error" conters is indicated by a leading "E " on the first
+ * "error" counter, and doesn't count in label length.
+ * The EgrOvfl list needs to be last so we truncate them at the configured
+ * context count for the device.
+ * cntr6120indices contains the corresponding register indices.
+ */
+static const char cntr6120names[] =
+	"Interrupts\n"
+	"HostBusStall\n"
+	"E RxTIDFull\n"
+	"RxTIDInvalid\n"
+	"Ctxt0EgrOvfl\n"
+	"Ctxt1EgrOvfl\n"
+	"Ctxt2EgrOvfl\n"
+	"Ctxt3EgrOvfl\n"
+	"Ctxt4EgrOvfl\n";
+
+static const size_t cntr6120indices[] = {
+	cr_lbint,
+	cr_lbflowstall,
+	cr_errtidfull,
+	cr_errtidvalid,
+	cr_portovfl + 0,
+	cr_portovfl + 1,
+	cr_portovfl + 2,
+	cr_portovfl + 3,
+	cr_portovfl + 4,
+};
+
+/*
+ * same as cntr6120names and cntr6120indices, but for port-specific counters.
+ * portcntr6120indices is somewhat complicated by some registers needing
+ * adjustments of various kinds, and those are ORed with _PORT_VIRT_FLAG
+ */
+static const char portcntr6120names[] =
+	"TxPkt\n"
+	"TxFlowPkt\n"
+	"TxWords\n"
+	"RxPkt\n"
+	"RxFlowPkt\n"
+	"RxWords\n"
+	"TxFlowStall\n"
+	"E IBStatusChng\n"
+	"IBLinkDown\n"
+	"IBLnkRecov\n"
+	"IBRxLinkErr\n"
+	"IBSymbolErr\n"
+	"RxLLIErr\n"
+	"RxBadFormat\n"
+	"RxBadLen\n"
+	"RxBufOvrfl\n"
+	"RxEBP\n"
+	"RxFlowCtlErr\n"
+	"RxICRCerr\n"
+	"RxLPCRCerr\n"
+	"RxVCRCerr\n"
+	"RxInvalLen\n"
+	"RxInvalPKey\n"
+	"RxPktDropped\n"
+	"TxBadLength\n"
+	"TxDropped\n"
+	"TxInvalLen\n"
+	"TxUnderrun\n"
+	"TxUnsupVL\n"
+	;
+
+#define _PORT_VIRT_FLAG 0x8000 /* "virtual", need adjustments */
+static const size_t portcntr6120indices[] = {
+	QIBPORTCNTR_PKTSEND | _PORT_VIRT_FLAG,
+	cr_pktsendflow,
+	QIBPORTCNTR_WORDSEND | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_PKTRCV | _PORT_VIRT_FLAG,
+	cr_pktrcvflowctrl,
+	QIBPORTCNTR_WORDRCV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_SENDSTALL | _PORT_VIRT_FLAG,
+	cr_ibstatuschange,
+	QIBPORTCNTR_IBLINKDOWN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBLINKERRRECOV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLINK | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBSYMBOLERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_LLI | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_BADFORMAT | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERR_RLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVOVFL | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVEBP | _PORT_VIRT_FLAG,
+	cr_rcvflowctrl_err,
+	QIBPORTCNTR_ERRICRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLPCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRVCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_INVALIDRLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRPKEY | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RXDROPPKT | _PORT_VIRT_FLAG,
+	cr_invalidslen,
+	cr_senddropped,
+	cr_errslen,
+	cr_sendunderrun,
+	cr_txunsupvl,
+};
+
+/* do all the setup to make the counter reads efficient later */
+static void init_6120_cntrnames(struct qib_devdata *dd)
+{
+	int i, j = 0;
+	char *s;
+
+	for (i = 0, s = (char *)cntr6120names; s && j <= dd->cfgctxts;
+	     i++) {
+		/* we always have at least one counter before the egrovfl */
+		if (!j && !strncmp("Ctxt0EgrOvfl", s + 1, 12))
+			j = 1;
+		s = strchr(s + 1, '\n');
+		if (s && j)
+			j++;
+	}
+	dd->cspec->ncntrs = i;
+	if (!s)
+		/* full list; size is without terminating null */
+		dd->cspec->cntrnamelen = sizeof(cntr6120names) - 1;
+	else
+		dd->cspec->cntrnamelen = 1 + s - cntr6120names;
+	dd->cspec->cntrs = kmalloc(dd->cspec->ncntrs
+		* sizeof(u64), GFP_KERNEL);
+	if (!dd->cspec->cntrs)
+		qib_dev_err(dd, "Failed allocation for counters\n");
+
+	for (i = 0, s = (char *)portcntr6120names; s; i++)
+		s = strchr(s + 1, '\n');
+	dd->cspec->nportcntrs = i - 1;
+	dd->cspec->portcntrnamelen = sizeof(portcntr6120names) - 1;
+	dd->cspec->portcntrs = kmalloc(dd->cspec->nportcntrs
+		* sizeof(u64), GFP_KERNEL);
+	if (!dd->cspec->portcntrs)
+		qib_dev_err(dd, "Failed allocation for portcounters\n");
+}
+
+static u32 qib_read_6120cntrs(struct qib_devdata *dd, loff_t pos, char **namep,
+			      u64 **cntrp)
+{
+	u32 ret;
+
+	if (namep) {
+		ret = dd->cspec->cntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+		else
+			*namep = (char *)cntr6120names;
+	} else {
+		u64 *cntr = dd->cspec->cntrs;
+		int i;
+
+		ret = dd->cspec->ncntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+		if (pos >= ret) {
+			ret = 0; /* final read after getting everything */
+			goto done;
+		}
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->ncntrs; i++)
+			*cntr++ = read_6120_creg32(dd, cntr6120indices[i]);
+	}
+done:
+	return ret;
+}
+
+static u32 qib_read_6120portcntrs(struct qib_devdata *dd, loff_t pos, u32 port,
+				  char **namep, u64 **cntrp)
+{
+	u32 ret;
+
+	if (namep) {
+		ret = dd->cspec->portcntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+		else
+			*namep = (char *)portcntr6120names;
+	} else {
+		u64 *cntr = dd->cspec->portcntrs;
+		struct qib_pportdata *ppd = &dd->pport[port];
+		int i;
+
+		ret = dd->cspec->nportcntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->nportcntrs; i++) {
+			if (portcntr6120indices[i] & _PORT_VIRT_FLAG)
+				*cntr++ = qib_portcntr_6120(ppd,
+					portcntr6120indices[i] &
+					~_PORT_VIRT_FLAG);
+			else
+				*cntr++ = read_6120_creg32(dd,
+					   portcntr6120indices[i]);
+		}
+	}
+done:
+	return ret;
+}
+
+static void qib_chk_6120_errormask(struct qib_devdata *dd)
+{
+	static u32 fixed;
+	u32 ctrl;
+	unsigned long errormask;
+	unsigned long hwerrs;
+
+	if (!dd->cspec->errormask || !(dd->flags & QIB_INITTED))
+		return;
+
+	errormask = qib_read_kreg64(dd, kr_errmask);
+
+	if (errormask == dd->cspec->errormask)
+		return;
+	fixed++;
+
+	hwerrs = qib_read_kreg64(dd, kr_hwerrstatus);
+	ctrl = qib_read_kreg32(dd, kr_control);
+
+	qib_write_kreg(dd, kr_errmask,
+		dd->cspec->errormask);
+
+	if ((hwerrs & dd->cspec->hwerrmask) ||
+	    (ctrl & QLOGIC_IB_C_FREEZEMODE)) {
+		qib_write_kreg(dd, kr_hwerrclear, 0ULL);
+		qib_write_kreg(dd, kr_errclear, 0ULL);
+		/* force re-interrupt of pending events, just in case */
+		qib_write_kreg(dd, kr_intclear, 0ULL);
+		qib_devinfo(dd->pcidev,
+			 "errormask fixed(%u) %lx->%lx, ctrl %x hwerr %lx\n",
+			 fixed, errormask, (unsigned long)dd->cspec->errormask,
+			 ctrl, hwerrs);
+	}
+}
+
+/**
+ * qib_get_faststats - get word counters from chip before they overflow
+ * @opaque - contains a pointer to the qlogic_ib device qib_devdata
+ *
+ * This needs more work; in particular, decision on whether we really
+ * need traffic_wds done the way it is
+ * called from add_timer
+ */
+static void qib_get_6120_faststats(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *) opaque;
+	struct qib_pportdata *ppd = dd->pport;
+	unsigned long flags;
+	u64 traffic_wds;
+
+	/*
+	 * don't access the chip while running diags, or memory diags can
+	 * fail
+	 */
+	if (!(dd->flags & QIB_INITTED) || dd->diag_client)
+		/* but re-arm the timer, for diags case; won't hurt other */
+		goto done;
+
+	/*
+	 * We now try to maintain an activity timer, based on traffic
+	 * exceeding a threshold, so we need to check the word-counts
+	 * even if they are 64-bit.
+	 */
+	traffic_wds = qib_portcntr_6120(ppd, cr_wordsend) +
+		qib_portcntr_6120(ppd, cr_wordrcv);
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	traffic_wds -= dd->traffic_wds;
+	dd->traffic_wds += traffic_wds;
+	if (traffic_wds  >= QIB_TRAFFIC_ACTIVE_THRESHOLD)
+		atomic_add(5, &dd->active_time); /* S/B #define */
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+
+	qib_chk_6120_errormask(dd);
+done:
+	mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
+}
+
+/* no interrupt fallback for these chips */
+static int qib_6120_nointr_fallback(struct qib_devdata *dd)
+{
+	return 0;
+}
+
+/*
+ * reset the XGXS (between serdes and IBC).  Slightly less intrusive
+ * than resetting the IBC or external link state, and useful in some
+ * cases to cause some retraining.  To do this right, we reset IBC
+ * as well.
+ */
+static void qib_6120_xgxs_reset(struct qib_pportdata *ppd)
+{
+	u64 val, prev_val;
+	struct qib_devdata *dd = ppd->dd;
+
+	prev_val = qib_read_kreg64(dd, kr_xgxs_cfg);
+	val = prev_val | QLOGIC_IB_XGXS_RESET;
+	prev_val &= ~QLOGIC_IB_XGXS_RESET; /* be sure */
+	qib_write_kreg(dd, kr_control,
+		       dd->control & ~QLOGIC_IB_C_LINKENABLE);
+	qib_write_kreg(dd, kr_xgxs_cfg, val);
+	qib_read_kreg32(dd, kr_scratch);
+	qib_write_kreg(dd, kr_xgxs_cfg, prev_val);
+	qib_write_kreg(dd, kr_control, dd->control);
+}
+
+static int qib_6120_get_ib_cfg(struct qib_pportdata *ppd, int which)
+{
+	int ret;
+
+	switch (which) {
+	case QIB_IB_CFG_LWID:
+		ret = ppd->link_width_active;
+		break;
+
+	case QIB_IB_CFG_SPD:
+		ret = ppd->link_speed_active;
+		break;
+
+	case QIB_IB_CFG_LWID_ENB:
+		ret = ppd->link_width_enabled;
+		break;
+
+	case QIB_IB_CFG_SPD_ENB:
+		ret = ppd->link_speed_enabled;
+		break;
+
+	case QIB_IB_CFG_OP_VLS:
+		ret = ppd->vls_operational;
+		break;
+
+	case QIB_IB_CFG_VL_HIGH_CAP:
+		ret = 0;
+		break;
+
+	case QIB_IB_CFG_VL_LOW_CAP:
+		ret = 0;
+		break;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		ret = SYM_FIELD(ppd->dd->cspec->ibcctrl, IBCCtrl,
+				OverrunThreshold);
+		break;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		ret = SYM_FIELD(ppd->dd->cspec->ibcctrl, IBCCtrl,
+				PhyerrThreshold);
+		break;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		ret = (ppd->dd->cspec->ibcctrl &
+		       SYM_MASK(IBCCtrl, LinkDownDefaultState)) ?
+			IB_LINKINITCMD_SLEEP : IB_LINKINITCMD_POLL;
+		break;
+
+	case QIB_IB_CFG_HRTBT: /* Get Heartbeat off/enable/auto */
+		ret = 0; /* no heartbeat on this chip */
+		break;
+
+	case QIB_IB_CFG_PMA_TICKS:
+		ret = 250; /* 1 usec. */
+		break;
+
+	default:
+		ret =  -EINVAL;
+		break;
+	}
+	return ret;
+}
+
+/*
+ * We assume range checking is already done, if needed.
+ */
+static int qib_6120_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = 0;
+	u64 val64;
+	u16 lcmd, licmd;
+
+	switch (which) {
+	case QIB_IB_CFG_LWID_ENB:
+		ppd->link_width_enabled = val;
+		break;
+
+	case QIB_IB_CFG_SPD_ENB:
+		ppd->link_speed_enabled = val;
+		break;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		val64 = SYM_FIELD(dd->cspec->ibcctrl, IBCCtrl,
+				  OverrunThreshold);
+		if (val64 != val) {
+			dd->cspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, OverrunThreshold);
+			dd->cspec->ibcctrl |= (u64) val <<
+				SYM_LSB(IBCCtrl, OverrunThreshold);
+			qib_write_kreg(dd, kr_ibcctrl, dd->cspec->ibcctrl);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+		break;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		val64 = SYM_FIELD(dd->cspec->ibcctrl, IBCCtrl,
+				  PhyerrThreshold);
+		if (val64 != val) {
+			dd->cspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, PhyerrThreshold);
+			dd->cspec->ibcctrl |= (u64) val <<
+				SYM_LSB(IBCCtrl, PhyerrThreshold);
+			qib_write_kreg(dd, kr_ibcctrl, dd->cspec->ibcctrl);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+		break;
+
+	case QIB_IB_CFG_PKEYS: /* update pkeys */
+		val64 = (u64) ppd->pkeys[0] | ((u64) ppd->pkeys[1] << 16) |
+			((u64) ppd->pkeys[2] << 32) |
+			((u64) ppd->pkeys[3] << 48);
+		qib_write_kreg(dd, kr_partitionkey, val64);
+		break;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		if (val == IB_LINKINITCMD_POLL)
+			dd->cspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, LinkDownDefaultState);
+		else /* SLEEP */
+			dd->cspec->ibcctrl |=
+				SYM_MASK(IBCCtrl, LinkDownDefaultState);
+		qib_write_kreg(dd, kr_ibcctrl, dd->cspec->ibcctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+		break;
+
+	case QIB_IB_CFG_MTU: /* update the MTU in IBC */
+		/*
+		 * Update our housekeeping variables, and set IBC max
+		 * size, same as init code; max IBC is max we allow in
+		 * buffer, less the qword pbc, plus 1 for ICRC, in dwords
+		 * Set even if it's unchanged, print debug message only
+		 * on changes.
+		 */
+		val = (ppd->ibmaxlen >> 2) + 1;
+		dd->cspec->ibcctrl &= ~SYM_MASK(IBCCtrl, MaxPktLen);
+		dd->cspec->ibcctrl |= (u64)val <<
+			SYM_LSB(IBCCtrl, MaxPktLen);
+		qib_write_kreg(dd, kr_ibcctrl, dd->cspec->ibcctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+		break;
+
+	case QIB_IB_CFG_LSTATE: /* set the IB link state */
+		switch (val & 0xffff0000) {
+		case IB_LINKCMD_DOWN:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_DOWN;
+			if (!dd->cspec->ibdeltainprog) {
+				dd->cspec->ibdeltainprog = 1;
+				dd->cspec->ibsymsnap =
+					read_6120_creg32(dd, cr_ibsymbolerr);
+				dd->cspec->iblnkerrsnap =
+					read_6120_creg32(dd, cr_iblinkerrrecov);
+			}
+			break;
+
+		case IB_LINKCMD_ARMED:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ARMED;
+			break;
+
+		case IB_LINKCMD_ACTIVE:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ACTIVE;
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkcmd req 0x%x\n", val >> 16);
+			goto bail;
+		}
+		switch (val & 0xffff) {
+		case IB_LINKINITCMD_NOP:
+			licmd = 0;
+			break;
+
+		case IB_LINKINITCMD_POLL:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_POLL;
+			break;
+
+		case IB_LINKINITCMD_SLEEP:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_SLEEP;
+			break;
+
+		case IB_LINKINITCMD_DISABLE:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_DISABLE;
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkinitcmd req 0x%x\n",
+				    val & 0xffff);
+			goto bail;
+		}
+		qib_set_ib_6120_lstate(ppd, lcmd, licmd);
+		goto bail;
+
+	case QIB_IB_CFG_HRTBT:
+		ret = -EINVAL;
+		break;
+
+	default:
+		ret = -EINVAL;
+	}
+bail:
+	return ret;
+}
+
+static int qib_6120_set_loopback(struct qib_pportdata *ppd, const char *what)
+{
+	int ret = 0;
+	if (!strncmp(what, "ibc", 3)) {
+		ppd->dd->cspec->ibcctrl |= SYM_MASK(IBCCtrl, Loopback);
+		qib_devinfo(ppd->dd->pcidev, "Enabling IB%u:%u IBC loopback\n",
+			 ppd->dd->unit, ppd->port);
+	} else if (!strncmp(what, "off", 3)) {
+		ppd->dd->cspec->ibcctrl &= ~SYM_MASK(IBCCtrl, Loopback);
+		qib_devinfo(ppd->dd->pcidev,
+			"Disabling IB%u:%u IBC loopback (normal)\n",
+			ppd->dd->unit, ppd->port);
+	} else
+		ret = -EINVAL;
+	if (!ret) {
+		qib_write_kreg(ppd->dd, kr_ibcctrl, ppd->dd->cspec->ibcctrl);
+		qib_write_kreg(ppd->dd, kr_scratch, 0);
+	}
+	return ret;
+}
+
+static void pma_6120_timer(unsigned long data)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)data;
+	struct qib_chip_specific *cs = ppd->dd->cspec;
+	struct qib_ibport *ibp = &ppd->ibport_data;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	if (cs->pma_sample_status == IB_PMA_SAMPLE_STATUS_STARTED) {
+		cs->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
+		qib_snapshot_counters(ppd, &cs->sword, &cs->rword,
+				      &cs->spkts, &cs->rpkts, &cs->xmit_wait);
+		mod_timer(&cs->pma_timer,
+			  jiffies + usecs_to_jiffies(ibp->pma_sample_interval));
+	} else if (cs->pma_sample_status == IB_PMA_SAMPLE_STATUS_RUNNING) {
+		u64 ta, tb, tc, td, te;
+
+		cs->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
+		qib_snapshot_counters(ppd, &ta, &tb, &tc, &td, &te);
+
+		cs->sword = ta - cs->sword;
+		cs->rword = tb - cs->rword;
+		cs->spkts = tc - cs->spkts;
+		cs->rpkts = td - cs->rpkts;
+		cs->xmit_wait = te - cs->xmit_wait;
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+}
+
+/*
+ * Note that the caller has the ibp->lock held.
+ */
+static void qib_set_cntr_6120_sample(struct qib_pportdata *ppd, u32 intv,
+				     u32 start)
+{
+	struct qib_chip_specific *cs = ppd->dd->cspec;
+
+	if (start && intv) {
+		cs->pma_sample_status = IB_PMA_SAMPLE_STATUS_STARTED;
+		mod_timer(&cs->pma_timer, jiffies + usecs_to_jiffies(start));
+	} else if (intv) {
+		cs->pma_sample_status = IB_PMA_SAMPLE_STATUS_RUNNING;
+		qib_snapshot_counters(ppd, &cs->sword, &cs->rword,
+				      &cs->spkts, &cs->rpkts, &cs->xmit_wait);
+		mod_timer(&cs->pma_timer, jiffies + usecs_to_jiffies(intv));
+	} else {
+		cs->pma_sample_status = IB_PMA_SAMPLE_STATUS_DONE;
+		cs->sword = 0;
+		cs->rword = 0;
+		cs->spkts = 0;
+		cs->rpkts = 0;
+		cs->xmit_wait = 0;
+	}
+}
+
+static u32 qib_6120_iblink_state(u64 ibcs)
+{
+	u32 state = (u32)SYM_FIELD(ibcs, IBCStatus, LinkState);
+
+	switch (state) {
+	case IB_6120_L_STATE_INIT:
+		state = IB_PORT_INIT;
+		break;
+	case IB_6120_L_STATE_ARM:
+		state = IB_PORT_ARMED;
+		break;
+	case IB_6120_L_STATE_ACTIVE:
+		/* fall through */
+	case IB_6120_L_STATE_ACT_DEFER:
+		state = IB_PORT_ACTIVE;
+		break;
+	default: /* fall through */
+	case IB_6120_L_STATE_DOWN:
+		state = IB_PORT_DOWN;
+		break;
+	}
+	return state;
+}
+
+/* returns the IBTA port state, rather than the IBC link training state */
+static u8 qib_6120_phys_portstate(u64 ibcs)
+{
+	u8 state = (u8)SYM_FIELD(ibcs, IBCStatus, LinkTrainingState);
+	return qib_6120_physportstate[state];
+}
+
+static int qib_6120_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags &= ~QIBL_IB_FORCE_NOTIFY;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	if (ibup) {
+		if (ppd->dd->cspec->ibdeltainprog) {
+			ppd->dd->cspec->ibdeltainprog = 0;
+			ppd->dd->cspec->ibsymdelta +=
+				read_6120_creg32(ppd->dd, cr_ibsymbolerr) -
+					ppd->dd->cspec->ibsymsnap;
+			ppd->dd->cspec->iblnkerrdelta +=
+				read_6120_creg32(ppd->dd, cr_iblinkerrrecov) -
+					ppd->dd->cspec->iblnkerrsnap;
+		}
+		qib_hol_init(ppd);
+	} else {
+		ppd->dd->cspec->lli_counter = 0;
+		if (!ppd->dd->cspec->ibdeltainprog) {
+			ppd->dd->cspec->ibdeltainprog = 1;
+			ppd->dd->cspec->ibsymsnap =
+				read_6120_creg32(ppd->dd, cr_ibsymbolerr);
+			ppd->dd->cspec->iblnkerrsnap =
+				read_6120_creg32(ppd->dd, cr_iblinkerrrecov);
+		}
+		qib_hol_down(ppd);
+	}
+
+	qib_6120_setup_setextled(ppd, ibup);
+
+	return 0;
+}
+
+/* Does read/modify/write to appropriate registers to
+ * set output and direction bits selected by mask.
+ * these are in their canonical postions (e.g. lsb of
+ * dir will end up in D48 of extctrl on existing chips).
+ * returns contents of GP Inputs.
+ */
+static int gpio_6120_mod(struct qib_devdata *dd, u32 out, u32 dir, u32 mask)
+{
+	u64 read_val, new_out;
+	unsigned long flags;
+
+	if (mask) {
+		/* some bits being written, lock access to GPIO */
+		dir &= mask;
+		out &= mask;
+		spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+		dd->cspec->extctrl &= ~((u64)mask << SYM_LSB(EXTCtrl, GPIOOe));
+		dd->cspec->extctrl |= ((u64) dir << SYM_LSB(EXTCtrl, GPIOOe));
+		new_out = (dd->cspec->gpio_out & ~mask) | out;
+
+		qib_write_kreg(dd, kr_extctrl, dd->cspec->extctrl);
+		qib_write_kreg(dd, kr_gpio_out, new_out);
+		dd->cspec->gpio_out = new_out;
+		spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+	}
+	/*
+	 * It is unlikely that a read at this time would get valid
+	 * data on a pin whose direction line was set in the same
+	 * call to this function. We include the read here because
+	 * that allows us to potentially combine a change on one pin with
+	 * a read on another, and because the old code did something like
+	 * this.
+	 */
+	read_val = qib_read_kreg64(dd, kr_extstatus);
+	return SYM_FIELD(read_val, EXTStatus, GPIOIn);
+}
+
+/*
+ * Read fundamental info we need to use the chip.  These are
+ * the registers that describe chip capabilities, and are
+ * saved in shadow registers.
+ */
+static void get_6120_chip_params(struct qib_devdata *dd)
+{
+	u64 val;
+	u32 piobufs;
+	int mtu;
+
+	dd->uregbase = qib_read_kreg32(dd, kr_userregbase);
+
+	dd->rcvtidcnt = qib_read_kreg32(dd, kr_rcvtidcnt);
+	dd->rcvtidbase = qib_read_kreg32(dd, kr_rcvtidbase);
+	dd->rcvegrbase = qib_read_kreg32(dd, kr_rcvegrbase);
+	dd->palign = qib_read_kreg32(dd, kr_palign);
+	dd->piobufbase = qib_read_kreg64(dd, kr_sendpiobufbase);
+	dd->pio2k_bufbase = dd->piobufbase & 0xffffffff;
+
+	dd->rcvhdrcnt = qib_read_kreg32(dd, kr_rcvegrcnt);
+
+	val = qib_read_kreg64(dd, kr_sendpiosize);
+	dd->piosize2k = val & ~0U;
+	dd->piosize4k = val >> 32;
+
+	mtu = ib_mtu_enum_to_int(qib_ibmtu);
+	if (mtu == -1)
+		mtu = QIB_DEFAULT_MTU;
+	dd->pport->ibmtu = (u32)mtu;
+
+	val = qib_read_kreg64(dd, kr_sendpiobufcnt);
+	dd->piobcnt2k = val & ~0U;
+	dd->piobcnt4k = val >> 32;
+	dd->last_pio = dd->piobcnt4k + dd->piobcnt2k - 1;
+	/* these may be adjusted in init_chip_wc_pat() */
+	dd->pio2kbase = (u32 __iomem *)
+		(((char __iomem *)dd->kregbase) + dd->pio2k_bufbase);
+	if (dd->piobcnt4k) {
+		dd->pio4kbase = (u32 __iomem *)
+			(((char __iomem *) dd->kregbase) +
+			 (dd->piobufbase >> 32));
+		/*
+		 * 4K buffers take 2 pages; we use roundup just to be
+		 * paranoid; we calculate it once here, rather than on
+		 * ever buf allocate
+		 */
+		dd->align4k = ALIGN(dd->piosize4k, dd->palign);
+	}
+
+	piobufs = dd->piobcnt4k + dd->piobcnt2k;
+
+	dd->pioavregs = ALIGN(piobufs, sizeof(u64) * BITS_PER_BYTE / 2) /
+		(sizeof(u64) * BITS_PER_BYTE / 2);
+}
+
+/*
+ * The chip base addresses in cspec and cpspec have to be set
+ * after possible init_chip_wc_pat(), rather than in
+ * get_6120_chip_params(), so split out as separate function
+ */
+static void set_6120_baseaddrs(struct qib_devdata *dd)
+{
+	u32 cregbase;
+	cregbase = qib_read_kreg32(dd, kr_counterregbase);
+	dd->cspec->cregbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase + cregbase);
+
+	dd->egrtidbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase + dd->rcvegrbase);
+}
+
+/*
+ * Write the final few registers that depend on some of the
+ * init setup.  Done late in init, just before bringing up
+ * the serdes.
+ */
+static int qib_late_6120_initreg(struct qib_devdata *dd)
+{
+	int ret = 0;
+	u64 val;
+
+	qib_write_kreg(dd, kr_rcvhdrentsize, dd->rcvhdrentsize);
+	qib_write_kreg(dd, kr_rcvhdrsize, dd->rcvhdrsize);
+	qib_write_kreg(dd, kr_rcvhdrcnt, dd->rcvhdrcnt);
+	qib_write_kreg(dd, kr_sendpioavailaddr, dd->pioavailregs_phys);
+	val = qib_read_kreg64(dd, kr_sendpioavailaddr);
+	if (val != dd->pioavailregs_phys) {
+		qib_dev_err(dd,
+			"Catastrophic software error, SendPIOAvailAddr written as %lx, read back as %llx\n",
+			(unsigned long) dd->pioavailregs_phys,
+			(unsigned long long) val);
+		ret = -EINVAL;
+	}
+	return ret;
+}
+
+static int init_6120_variables(struct qib_devdata *dd)
+{
+	int ret = 0;
+	struct qib_pportdata *ppd;
+	u32 sbufs;
+
+	ppd = (struct qib_pportdata *)(dd + 1);
+	dd->pport = ppd;
+	dd->num_pports = 1;
+
+	dd->cspec = (struct qib_chip_specific *)(ppd + dd->num_pports);
+	ppd->cpspec = NULL; /* not used in this chip */
+
+	spin_lock_init(&dd->cspec->kernel_tid_lock);
+	spin_lock_init(&dd->cspec->user_tid_lock);
+	spin_lock_init(&dd->cspec->rcvmod_lock);
+	spin_lock_init(&dd->cspec->gpio_lock);
+
+	/* we haven't yet set QIB_PRESENT, so use read directly */
+	dd->revision = readq(&dd->kregbase[kr_revision]);
+
+	if ((dd->revision & 0xffffffffU) == 0xffffffffU) {
+		qib_dev_err(dd,
+			"Revision register read failure, giving up initialization\n");
+		ret = -ENODEV;
+		goto bail;
+	}
+	dd->flags |= QIB_PRESENT;  /* now register routines work */
+
+	dd->majrev = (u8) SYM_FIELD(dd->revision, Revision_R,
+				    ChipRevMajor);
+	dd->minrev = (u8) SYM_FIELD(dd->revision, Revision_R,
+				    ChipRevMinor);
+
+	get_6120_chip_params(dd);
+	pe_boardname(dd); /* fill in boardname */
+
+	/*
+	 * GPIO bits for TWSI data and clock,
+	 * used for serial EEPROM.
+	 */
+	dd->gpio_sda_num = _QIB_GPIO_SDA_NUM;
+	dd->gpio_scl_num = _QIB_GPIO_SCL_NUM;
+	dd->twsi_eeprom_dev = QIB_TWSI_NO_DEV;
+
+	if (qib_unordered_wc())
+		dd->flags |= QIB_PIO_FLUSH_WC;
+
+	/*
+	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
+	 * 2 is Some Misc, 3 is reserved for future.
+	 */
+	dd->eep_st_masks[0].hwerrs_to_log = HWE_MASK(TXEMemParityErr);
+
+	/* Ignore errors in PIO/PBC on systems with unordered write-combining */
+	if (qib_unordered_wc())
+		dd->eep_st_masks[0].hwerrs_to_log &= ~TXE_PIO_PARITY;
+
+	dd->eep_st_masks[1].hwerrs_to_log = HWE_MASK(RXEMemParityErr);
+
+	dd->eep_st_masks[2].errs_to_log = ERR_MASK(ResetNegated);
+
+	ret = qib_init_pportdata(ppd, dd, 0, 1);
+	if (ret)
+		goto bail;
+	ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
+	ppd->link_speed_supported = QIB_IB_SDR;
+	ppd->link_width_enabled = IB_WIDTH_4X;
+	ppd->link_speed_enabled = ppd->link_speed_supported;
+	/* these can't change for this chip, so set once */
+	ppd->link_width_active = ppd->link_width_enabled;
+	ppd->link_speed_active = ppd->link_speed_enabled;
+	ppd->vls_supported = IB_VL_VL0;
+	ppd->vls_operational = ppd->vls_supported;
+
+	dd->rcvhdrentsize = QIB_RCVHDR_ENTSIZE;
+	dd->rcvhdrsize = QIB_DFLT_RCVHDRSIZE;
+	dd->rhf_offset = 0;
+
+	/* we always allocate at least 2048 bytes for eager buffers */
+	ret = ib_mtu_enum_to_int(qib_ibmtu);
+	dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
+
+	qib_6120_tidtemplate(dd);
+
+	/*
+	 * We can request a receive interrupt for 1 or
+	 * more packets from current offset.  For now, we set this
+	 * up for a single packet.
+	 */
+	dd->rhdrhead_intr_off = 1ULL << 32;
+
+	/* setup the stats timer; the add_timer is done at end of init */
+	init_timer(&dd->stats_timer);
+	dd->stats_timer.function = qib_get_6120_faststats;
+	dd->stats_timer.data = (unsigned long) dd;
+
+	init_timer(&dd->cspec->pma_timer);
+	dd->cspec->pma_timer.function = pma_6120_timer;
+	dd->cspec->pma_timer.data = (unsigned long) ppd;
+
+	dd->ureg_align = qib_read_kreg32(dd, kr_palign);
+
+	dd->piosize2kmax_dwords = dd->piosize2k >> 2;
+	qib_6120_config_ctxts(dd);
+	qib_set_ctxtcnt(dd);
+
+	if (qib_wc_pat) {
+		ret = init_chip_wc_pat(dd, 0);
+		if (ret)
+			goto bail;
+	}
+	set_6120_baseaddrs(dd); /* set chip access pointers now */
+
+	ret = 0;
+	if (qib_mini_init)
+		goto bail;
+
+	qib_num_cfg_vls = 1; /* if any 6120's, only one VL */
+
+	ret = qib_create_ctxts(dd);
+	init_6120_cntrnames(dd);
+
+	/* use all of 4KB buffers for the kernel, otherwise 16 */
+	sbufs = dd->piobcnt4k ?  dd->piobcnt4k : 16;
+
+	dd->lastctxt_piobuf = dd->piobcnt2k + dd->piobcnt4k - sbufs;
+	dd->pbufsctxt = dd->lastctxt_piobuf /
+		(dd->cfgctxts - dd->first_user_ctxt);
+
+	if (ret)
+		goto bail;
+bail:
+	return ret;
+}
+
+/*
+ * For this chip, we want to use the same buffer every time
+ * when we are trying to bring the link up (they are always VL15
+ * packets).  At that link state the packet should always go out immediately
+ * (or at least be discarded at the tx interface if the link is down).
+ * If it doesn't, and the buffer isn't available, that means some other
+ * sender has gotten ahead of us, and is preventing our packet from going
+ * out.  In that case, we flush all packets, and try again.  If that still
+ * fails, we fail the request, and hope things work the next time around.
+ *
+ * We don't need very complicated heuristics on whether the packet had
+ * time to go out or not, since even at SDR 1X, it goes out in very short
+ * time periods, covered by the chip reads done here and as part of the
+ * flush.
+ */
+static u32 __iomem *get_6120_link_buf(struct qib_pportdata *ppd, u32 *bnum)
+{
+	u32 __iomem *buf;
+	u32 lbuf = ppd->dd->piobcnt2k + ppd->dd->piobcnt4k - 1;
+
+	/*
+	 * always blip to get avail list updated, since it's almost
+	 * always needed, and is fairly cheap.
+	 */
+	sendctrl_6120_mod(ppd->dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+	qib_read_kreg64(ppd->dd, kr_scratch); /* extra chip flush */
+	buf = qib_getsendbuf_range(ppd->dd, bnum, lbuf, lbuf);
+	if (buf)
+		goto done;
+
+	sendctrl_6120_mod(ppd, QIB_SENDCTRL_DISARM_ALL | QIB_SENDCTRL_FLUSH |
+			  QIB_SENDCTRL_AVAIL_BLIP);
+	ppd->dd->upd_pio_shadow  = 1; /* update our idea of what's busy */
+	qib_read_kreg64(ppd->dd, kr_scratch); /* extra chip flush */
+	buf = qib_getsendbuf_range(ppd->dd, bnum, lbuf, lbuf);
+done:
+	return buf;
+}
+
+static u32 __iomem *qib_6120_getsendbuf(struct qib_pportdata *ppd, u64 pbc,
+					u32 *pbufnum)
+{
+	u32 first, last, plen = pbc & QIB_PBC_LENGTH_MASK;
+	struct qib_devdata *dd = ppd->dd;
+	u32 __iomem *buf;
+
+	if (((pbc >> 32) & PBC_6120_VL15_SEND_CTRL) &&
+		!(ppd->lflags & (QIBL_IB_AUTONEG_INPROG | QIBL_LINKACTIVE)))
+		buf = get_6120_link_buf(ppd, pbufnum);
+	else {
+
+		if ((plen + 1) > dd->piosize2kmax_dwords)
+			first = dd->piobcnt2k;
+		else
+			first = 0;
+		/* try 4k if all 2k busy, so same last for both sizes */
+		last = dd->piobcnt2k + dd->piobcnt4k - 1;
+		buf = qib_getsendbuf_range(dd, pbufnum, first, last);
+	}
+	return buf;
+}
+
+static int init_sdma_6120_regs(struct qib_pportdata *ppd)
+{
+	return -ENODEV;
+}
+
+static u16 qib_sdma_6120_gethead(struct qib_pportdata *ppd)
+{
+	return 0;
+}
+
+static int qib_sdma_6120_busy(struct qib_pportdata *ppd)
+{
+	return 0;
+}
+
+static void qib_sdma_update_6120_tail(struct qib_pportdata *ppd, u16 tail)
+{
+}
+
+static void qib_6120_sdma_sendctrl(struct qib_pportdata *ppd, unsigned op)
+{
+}
+
+static void qib_sdma_set_6120_desc_cnt(struct qib_pportdata *ppd, unsigned cnt)
+{
+}
+
+/*
+ * the pbc doesn't need a VL15 indicator, but we need it for link_buf.
+ * The chip ignores the bit if set.
+ */
+static u32 qib_6120_setpbc_control(struct qib_pportdata *ppd, u32 plen,
+				   u8 srate, u8 vl)
+{
+	return vl == 15 ? PBC_6120_VL15_SEND_CTRL : 0;
+}
+
+static void qib_6120_initvl15_bufs(struct qib_devdata *dd)
+{
+}
+
+static void qib_6120_init_ctxt(struct qib_ctxtdata *rcd)
+{
+	rcd->rcvegrcnt = rcd->dd->rcvhdrcnt;
+	rcd->rcvegr_tid_base = rcd->ctxt * rcd->rcvegrcnt;
+}
+
+static void qib_6120_txchk_change(struct qib_devdata *dd, u32 start,
+	u32 len, u32 avail, struct qib_ctxtdata *rcd)
+{
+}
+
+static void writescratch(struct qib_devdata *dd, u32 val)
+{
+	(void) qib_write_kreg(dd, kr_scratch, val);
+}
+
+static int qib_6120_tempsense_rd(struct qib_devdata *dd, int regnum)
+{
+	return -ENXIO;
+}
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+static int qib_6120_notify_dca(struct qib_devdata *dd, unsigned long event)
+{
+	return 0;
+}
+#endif
+
+/* Dummy function, as 6120 boards never disable EEPROM Write */
+static int qib_6120_eeprom_wen(struct qib_devdata *dd, int wen)
+{
+	return 1;
+}
+
+/**
+ * qib_init_iba6120_funcs - set up the chip-specific function pointers
+ * @pdev: pci_dev of the qlogic_ib device
+ * @ent: pci_device_id matching this chip
+ *
+ * This is global, and is called directly at init to set up the
+ * chip-specific function pointers for later use.
+ *
+ * It also allocates/partially-inits the qib_devdata struct for
+ * this device.
+ */
+struct qib_devdata *qib_init_iba6120_funcs(struct pci_dev *pdev,
+					   const struct pci_device_id *ent)
+{
+	struct qib_devdata *dd;
+	int ret;
+
+	dd = qib_alloc_devdata(pdev, sizeof(struct qib_pportdata) +
+			       sizeof(struct qib_chip_specific));
+	if (IS_ERR(dd))
+		goto bail;
+
+	dd->f_bringup_serdes    = qib_6120_bringup_serdes;
+	dd->f_cleanup           = qib_6120_setup_cleanup;
+	dd->f_clear_tids        = qib_6120_clear_tids;
+	dd->f_free_irq          = qib_6120_free_irq;
+	dd->f_get_base_info     = qib_6120_get_base_info;
+	dd->f_get_msgheader     = qib_6120_get_msgheader;
+	dd->f_getsendbuf        = qib_6120_getsendbuf;
+	dd->f_gpio_mod          = gpio_6120_mod;
+	dd->f_eeprom_wen	= qib_6120_eeprom_wen;
+	dd->f_hdrqempty         = qib_6120_hdrqempty;
+	dd->f_ib_updown         = qib_6120_ib_updown;
+	dd->f_init_ctxt         = qib_6120_init_ctxt;
+	dd->f_initvl15_bufs     = qib_6120_initvl15_bufs;
+	dd->f_intr_fallback     = qib_6120_nointr_fallback;
+	dd->f_late_initreg      = qib_late_6120_initreg;
+	dd->f_setpbc_control    = qib_6120_setpbc_control;
+	dd->f_portcntr          = qib_portcntr_6120;
+	dd->f_put_tid           = (dd->minrev >= 2) ?
+				      qib_6120_put_tid_2 :
+				      qib_6120_put_tid;
+	dd->f_quiet_serdes      = qib_6120_quiet_serdes;
+	dd->f_rcvctrl           = rcvctrl_6120_mod;
+	dd->f_read_cntrs        = qib_read_6120cntrs;
+	dd->f_read_portcntrs    = qib_read_6120portcntrs;
+	dd->f_reset             = qib_6120_setup_reset;
+	dd->f_init_sdma_regs    = init_sdma_6120_regs;
+	dd->f_sdma_busy         = qib_sdma_6120_busy;
+	dd->f_sdma_gethead      = qib_sdma_6120_gethead;
+	dd->f_sdma_sendctrl     = qib_6120_sdma_sendctrl;
+	dd->f_sdma_set_desc_cnt = qib_sdma_set_6120_desc_cnt;
+	dd->f_sdma_update_tail  = qib_sdma_update_6120_tail;
+	dd->f_sendctrl          = sendctrl_6120_mod;
+	dd->f_set_armlaunch     = qib_set_6120_armlaunch;
+	dd->f_set_cntr_sample   = qib_set_cntr_6120_sample;
+	dd->f_iblink_state      = qib_6120_iblink_state;
+	dd->f_ibphys_portstate  = qib_6120_phys_portstate;
+	dd->f_get_ib_cfg        = qib_6120_get_ib_cfg;
+	dd->f_set_ib_cfg        = qib_6120_set_ib_cfg;
+	dd->f_set_ib_loopback   = qib_6120_set_loopback;
+	dd->f_set_intr_state    = qib_6120_set_intr_state;
+	dd->f_setextled         = qib_6120_setup_setextled;
+	dd->f_txchk_change      = qib_6120_txchk_change;
+	dd->f_update_usrhead    = qib_update_6120_usrhead;
+	dd->f_wantpiobuf_intr   = qib_wantpiobuf_6120_intr;
+	dd->f_xgxs_reset        = qib_6120_xgxs_reset;
+	dd->f_writescratch      = writescratch;
+	dd->f_tempsense_rd	= qib_6120_tempsense_rd;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dd->f_notify_dca = qib_6120_notify_dca;
+#endif
+	/*
+	 * Do remaining pcie setup and save pcie values in dd.
+	 * Any error printing is already done by the init code.
+	 * On return, we have the chip mapped and accessible,
+	 * but chip registers are not set up until start of
+	 * init_6120_variables.
+	 */
+	ret = qib_pcie_ddinit(dd, pdev, ent);
+	if (ret < 0)
+		goto bail_free;
+
+	/* initialize chip-specific variables */
+	ret = init_6120_variables(dd);
+	if (ret)
+		goto bail_cleanup;
+
+	if (qib_mini_init)
+		goto bail;
+
+	if (qib_pcie_params(dd, 8, NULL, NULL))
+		qib_dev_err(dd,
+			"Failed to setup PCIe or interrupts; continuing anyway\n");
+	dd->cspec->irq = pdev->irq; /* save IRQ */
+
+	/* clear diagctrl register, in case diags were running and crashed */
+	qib_write_kreg(dd, kr_hwdiagctrl, 0);
+
+	if (qib_read_kreg64(dd, kr_hwerrstatus) &
+	    QLOGIC_IB_HWE_SERDESPLLFAILED)
+		qib_write_kreg(dd, kr_hwerrclear,
+			       QLOGIC_IB_HWE_SERDESPLLFAILED);
+
+	/* setup interrupt handler (interrupt type handled above) */
+	qib_setup_6120_interrupt(dd);
+	/* Note that qpn_mask is set by qib_6120_config_ctxts() first */
+	qib_6120_init_hwerrors(dd);
+
+	goto bail;
+
+bail_cleanup:
+	qib_pcie_ddcleanup(dd);
+bail_free:
+	qib_free_devdata(dd);
+	dd = ERR_PTR(ret);
+bail:
+	return dd;
+}
diff --git a/drivers/infiniband/hw/qib/qib_iba7220.c b/drivers/infiniband/hw/qib/qib_iba7220.c
new file mode 100644
index 00000000000..7dec89fdc12
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_iba7220.c
@@ -0,0 +1,4659 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/*
+ * This file contains all of the code that is specific to the
+ * QLogic_IB 7220 chip (except that specific to the SerDes)
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/module.h>
+#include <linux/io.h>
+#include <rdma/ib_verbs.h>
+
+#include "qib.h"
+#include "qib_7220.h"
+
+static void qib_setup_7220_setextled(struct qib_pportdata *, u32);
+static void qib_7220_handle_hwerrors(struct qib_devdata *, char *, size_t);
+static void sendctrl_7220_mod(struct qib_pportdata *ppd, u32 op);
+static u32 qib_7220_iblink_state(u64);
+static u8 qib_7220_phys_portstate(u64);
+static void qib_sdma_update_7220_tail(struct qib_pportdata *, u16);
+static void qib_set_ib_7220_lstate(struct qib_pportdata *, u16, u16);
+
+/*
+ * This file contains almost all the chip-specific register information and
+ * access functions for the QLogic QLogic_IB 7220 PCI-Express chip, with the
+ * exception of SerDes support, which in in qib_sd7220.c.
+ */
+
+/* Below uses machine-generated qib_chipnum_regs.h file */
+#define KREG_IDX(regname) (QIB_7220_##regname##_OFFS / sizeof(u64))
+
+/* Use defines to tie machine-generated names to lower-case names */
+#define kr_control KREG_IDX(Control)
+#define kr_counterregbase KREG_IDX(CntrRegBase)
+#define kr_errclear KREG_IDX(ErrClear)
+#define kr_errmask KREG_IDX(ErrMask)
+#define kr_errstatus KREG_IDX(ErrStatus)
+#define kr_extctrl KREG_IDX(EXTCtrl)
+#define kr_extstatus KREG_IDX(EXTStatus)
+#define kr_gpio_clear KREG_IDX(GPIOClear)
+#define kr_gpio_mask KREG_IDX(GPIOMask)
+#define kr_gpio_out KREG_IDX(GPIOOut)
+#define kr_gpio_status KREG_IDX(GPIOStatus)
+#define kr_hrtbt_guid KREG_IDX(HRTBT_GUID)
+#define kr_hwdiagctrl KREG_IDX(HwDiagCtrl)
+#define kr_hwerrclear KREG_IDX(HwErrClear)
+#define kr_hwerrmask KREG_IDX(HwErrMask)
+#define kr_hwerrstatus KREG_IDX(HwErrStatus)
+#define kr_ibcctrl KREG_IDX(IBCCtrl)
+#define kr_ibcddrctrl KREG_IDX(IBCDDRCtrl)
+#define kr_ibcddrstatus KREG_IDX(IBCDDRStatus)
+#define kr_ibcstatus KREG_IDX(IBCStatus)
+#define kr_ibserdesctrl KREG_IDX(IBSerDesCtrl)
+#define kr_intclear KREG_IDX(IntClear)
+#define kr_intmask KREG_IDX(IntMask)
+#define kr_intstatus KREG_IDX(IntStatus)
+#define kr_ncmodectrl KREG_IDX(IBNCModeCtrl)
+#define kr_palign KREG_IDX(PageAlign)
+#define kr_partitionkey KREG_IDX(RcvPartitionKey)
+#define kr_portcnt KREG_IDX(PortCnt)
+#define kr_rcvbthqp KREG_IDX(RcvBTHQP)
+#define kr_rcvctrl KREG_IDX(RcvCtrl)
+#define kr_rcvegrbase KREG_IDX(RcvEgrBase)
+#define kr_rcvegrcnt KREG_IDX(RcvEgrCnt)
+#define kr_rcvhdrcnt KREG_IDX(RcvHdrCnt)
+#define kr_rcvhdrentsize KREG_IDX(RcvHdrEntSize)
+#define kr_rcvhdrsize KREG_IDX(RcvHdrSize)
+#define kr_rcvpktledcnt KREG_IDX(RcvPktLEDCnt)
+#define kr_rcvtidbase KREG_IDX(RcvTIDBase)
+#define kr_rcvtidcnt KREG_IDX(RcvTIDCnt)
+#define kr_revision KREG_IDX(Revision)
+#define kr_scratch KREG_IDX(Scratch)
+#define kr_sendbuffererror KREG_IDX(SendBufErr0)
+#define kr_sendctrl KREG_IDX(SendCtrl)
+#define kr_senddmabase KREG_IDX(SendDmaBase)
+#define kr_senddmabufmask0 KREG_IDX(SendDmaBufMask0)
+#define kr_senddmabufmask1 (KREG_IDX(SendDmaBufMask0) + 1)
+#define kr_senddmabufmask2 (KREG_IDX(SendDmaBufMask0) + 2)
+#define kr_senddmahead KREG_IDX(SendDmaHead)
+#define kr_senddmaheadaddr KREG_IDX(SendDmaHeadAddr)
+#define kr_senddmalengen KREG_IDX(SendDmaLenGen)
+#define kr_senddmastatus KREG_IDX(SendDmaStatus)
+#define kr_senddmatail KREG_IDX(SendDmaTail)
+#define kr_sendpioavailaddr KREG_IDX(SendBufAvailAddr)
+#define kr_sendpiobufbase KREG_IDX(SendBufBase)
+#define kr_sendpiobufcnt KREG_IDX(SendBufCnt)
+#define kr_sendpiosize KREG_IDX(SendBufSize)
+#define kr_sendregbase KREG_IDX(SendRegBase)
+#define kr_userregbase KREG_IDX(UserRegBase)
+#define kr_xgxs_cfg KREG_IDX(XGXSCfg)
+
+/* These must only be written via qib_write_kreg_ctxt() */
+#define kr_rcvhdraddr KREG_IDX(RcvHdrAddr0)
+#define kr_rcvhdrtailaddr KREG_IDX(RcvHdrTailAddr0)
+
+
+#define CREG_IDX(regname) ((QIB_7220_##regname##_OFFS - \
+			QIB_7220_LBIntCnt_OFFS) / sizeof(u64))
+
+#define cr_badformat CREG_IDX(RxVersionErrCnt)
+#define cr_erricrc CREG_IDX(RxICRCErrCnt)
+#define cr_errlink CREG_IDX(RxLinkMalformCnt)
+#define cr_errlpcrc CREG_IDX(RxLPCRCErrCnt)
+#define cr_errpkey CREG_IDX(RxPKeyMismatchCnt)
+#define cr_rcvflowctrl_err CREG_IDX(RxFlowCtrlViolCnt)
+#define cr_err_rlen CREG_IDX(RxLenErrCnt)
+#define cr_errslen CREG_IDX(TxLenErrCnt)
+#define cr_errtidfull CREG_IDX(RxTIDFullErrCnt)
+#define cr_errtidvalid CREG_IDX(RxTIDValidErrCnt)
+#define cr_errvcrc CREG_IDX(RxVCRCErrCnt)
+#define cr_ibstatuschange CREG_IDX(IBStatusChangeCnt)
+#define cr_lbint CREG_IDX(LBIntCnt)
+#define cr_invalidrlen CREG_IDX(RxMaxMinLenErrCnt)
+#define cr_invalidslen CREG_IDX(TxMaxMinLenErrCnt)
+#define cr_lbflowstall CREG_IDX(LBFlowStallCnt)
+#define cr_pktrcv CREG_IDX(RxDataPktCnt)
+#define cr_pktrcvflowctrl CREG_IDX(RxFlowPktCnt)
+#define cr_pktsend CREG_IDX(TxDataPktCnt)
+#define cr_pktsendflow CREG_IDX(TxFlowPktCnt)
+#define cr_portovfl CREG_IDX(RxP0HdrEgrOvflCnt)
+#define cr_rcvebp CREG_IDX(RxEBPCnt)
+#define cr_rcvovfl CREG_IDX(RxBufOvflCnt)
+#define cr_senddropped CREG_IDX(TxDroppedPktCnt)
+#define cr_sendstall CREG_IDX(TxFlowStallCnt)
+#define cr_sendunderrun CREG_IDX(TxUnderrunCnt)
+#define cr_wordrcv CREG_IDX(RxDwordCnt)
+#define cr_wordsend CREG_IDX(TxDwordCnt)
+#define cr_txunsupvl CREG_IDX(TxUnsupVLErrCnt)
+#define cr_rxdroppkt CREG_IDX(RxDroppedPktCnt)
+#define cr_iblinkerrrecov CREG_IDX(IBLinkErrRecoveryCnt)
+#define cr_iblinkdown CREG_IDX(IBLinkDownedCnt)
+#define cr_ibsymbolerr CREG_IDX(IBSymbolErrCnt)
+#define cr_vl15droppedpkt CREG_IDX(RxVL15DroppedPktCnt)
+#define cr_rxotherlocalphyerr CREG_IDX(RxOtherLocalPhyErrCnt)
+#define cr_excessbufferovfl CREG_IDX(ExcessBufferOvflCnt)
+#define cr_locallinkintegrityerr CREG_IDX(LocalLinkIntegrityErrCnt)
+#define cr_rxvlerr CREG_IDX(RxVlErrCnt)
+#define cr_rxdlidfltr CREG_IDX(RxDlidFltrCnt)
+#define cr_psstat CREG_IDX(PSStat)
+#define cr_psstart CREG_IDX(PSStart)
+#define cr_psinterval CREG_IDX(PSInterval)
+#define cr_psrcvdatacount CREG_IDX(PSRcvDataCount)
+#define cr_psrcvpktscount CREG_IDX(PSRcvPktsCount)
+#define cr_psxmitdatacount CREG_IDX(PSXmitDataCount)
+#define cr_psxmitpktscount CREG_IDX(PSXmitPktsCount)
+#define cr_psxmitwaitcount CREG_IDX(PSXmitWaitCount)
+#define cr_txsdmadesc CREG_IDX(TxSDmaDescCnt)
+#define cr_pcieretrydiag CREG_IDX(PcieRetryBufDiagQwordCnt)
+
+#define SYM_RMASK(regname, fldname) ((u64)              \
+	QIB_7220_##regname##_##fldname##_RMASK)
+#define SYM_MASK(regname, fldname) ((u64)               \
+	QIB_7220_##regname##_##fldname##_RMASK <<       \
+	 QIB_7220_##regname##_##fldname##_LSB)
+#define SYM_LSB(regname, fldname) (QIB_7220_##regname##_##fldname##_LSB)
+#define SYM_FIELD(value, regname, fldname) ((u64) \
+	(((value) >> SYM_LSB(regname, fldname)) & \
+	 SYM_RMASK(regname, fldname)))
+#define ERR_MASK(fldname) SYM_MASK(ErrMask, fldname##Mask)
+#define HWE_MASK(fldname) SYM_MASK(HwErrMask, fldname##Mask)
+
+/* ibcctrl bits */
+#define QLOGIC_IB_IBCC_LINKINITCMD_DISABLE 1
+/* cycle through TS1/TS2 till OK */
+#define QLOGIC_IB_IBCC_LINKINITCMD_POLL 2
+/* wait for TS1, then go on */
+#define QLOGIC_IB_IBCC_LINKINITCMD_SLEEP 3
+#define QLOGIC_IB_IBCC_LINKINITCMD_SHIFT 16
+
+#define QLOGIC_IB_IBCC_LINKCMD_DOWN 1           /* move to 0x11 */
+#define QLOGIC_IB_IBCC_LINKCMD_ARMED 2          /* move to 0x21 */
+#define QLOGIC_IB_IBCC_LINKCMD_ACTIVE 3 /* move to 0x31 */
+
+#define BLOB_7220_IBCHG 0x81
+
+/*
+ * We could have a single register get/put routine, that takes a group type,
+ * but this is somewhat clearer and cleaner.  It also gives us some error
+ * checking.  64 bit register reads should always work, but are inefficient
+ * on opteron (the northbridge always generates 2 separate HT 32 bit reads),
+ * so we use kreg32 wherever possible.  User register and counter register
+ * reads are always 32 bit reads, so only one form of those routines.
+ */
+
+/**
+ * qib_read_ureg32 - read 32-bit virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @ctxt: context number
+ *
+ * Return the contents of a register that is virtualized to be per context.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
+ */
+static inline u32 qib_read_ureg32(const struct qib_devdata *dd,
+				  enum qib_ureg regno, int ctxt)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+
+	if (dd->userbase)
+		return readl(regno + (u64 __iomem *)
+			     ((char __iomem *)dd->userbase +
+			      dd->ureg_align * ctxt));
+	else
+		return readl(regno + (u64 __iomem *)
+			     (dd->uregbase +
+			      (char __iomem *)dd->kregbase +
+			      dd->ureg_align * ctxt));
+}
+
+/**
+ * qib_write_ureg - write 32-bit virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @value: value
+ * @ctxt: context
+ *
+ * Write the contents of a register that is virtualized to be per context.
+ */
+static inline void qib_write_ureg(const struct qib_devdata *dd,
+				  enum qib_ureg regno, u64 value, int ctxt)
+{
+	u64 __iomem *ubase;
+
+	if (dd->userbase)
+		ubase = (u64 __iomem *)
+			((char __iomem *) dd->userbase +
+			 dd->ureg_align * ctxt);
+	else
+		ubase = (u64 __iomem *)
+			(dd->uregbase +
+			 (char __iomem *) dd->kregbase +
+			 dd->ureg_align * ctxt);
+
+	if (dd->kregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &ubase[regno]);
+}
+
+/**
+ * qib_write_kreg_ctxt - write a device's per-ctxt 64-bit kernel register
+ * @dd: the qlogic_ib device
+ * @regno: the register number to write
+ * @ctxt: the context containing the register
+ * @value: the value to write
+ */
+static inline void qib_write_kreg_ctxt(const struct qib_devdata *dd,
+				       const u16 regno, unsigned ctxt,
+				       u64 value)
+{
+	qib_write_kreg(dd, regno + ctxt, value);
+}
+
+static inline void write_7220_creg(const struct qib_devdata *dd,
+				   u16 regno, u64 value)
+{
+	if (dd->cspec->cregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &dd->cspec->cregbase[regno]);
+}
+
+static inline u64 read_7220_creg(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readq(&dd->cspec->cregbase[regno]);
+}
+
+static inline u32 read_7220_creg32(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readl(&dd->cspec->cregbase[regno]);
+}
+
+/* kr_revision bits */
+#define QLOGIC_IB_R_EMULATORREV_MASK ((1ULL << 22) - 1)
+#define QLOGIC_IB_R_EMULATORREV_SHIFT 40
+
+/* kr_control bits */
+#define QLOGIC_IB_C_RESET (1U << 7)
+
+/* kr_intstatus, kr_intclear, kr_intmask bits */
+#define QLOGIC_IB_I_RCVURG_MASK ((1ULL << 17) - 1)
+#define QLOGIC_IB_I_RCVURG_SHIFT 32
+#define QLOGIC_IB_I_RCVAVAIL_MASK ((1ULL << 17) - 1)
+#define QLOGIC_IB_I_RCVAVAIL_SHIFT 0
+#define QLOGIC_IB_I_SERDESTRIMDONE (1ULL << 27)
+
+#define QLOGIC_IB_C_FREEZEMODE 0x00000002
+#define QLOGIC_IB_C_LINKENABLE 0x00000004
+
+#define QLOGIC_IB_I_SDMAINT             0x8000000000000000ULL
+#define QLOGIC_IB_I_SDMADISABLED        0x4000000000000000ULL
+#define QLOGIC_IB_I_ERROR               0x0000000080000000ULL
+#define QLOGIC_IB_I_SPIOSENT            0x0000000040000000ULL
+#define QLOGIC_IB_I_SPIOBUFAVAIL        0x0000000020000000ULL
+#define QLOGIC_IB_I_GPIO                0x0000000010000000ULL
+
+/* variables for sanity checking interrupt and errors */
+#define QLOGIC_IB_I_BITSEXTANT \
+		(QLOGIC_IB_I_SDMAINT | QLOGIC_IB_I_SDMADISABLED | \
+		(QLOGIC_IB_I_RCVURG_MASK << QLOGIC_IB_I_RCVURG_SHIFT) | \
+		(QLOGIC_IB_I_RCVAVAIL_MASK << \
+		 QLOGIC_IB_I_RCVAVAIL_SHIFT) | \
+		QLOGIC_IB_I_ERROR | QLOGIC_IB_I_SPIOSENT | \
+		QLOGIC_IB_I_SPIOBUFAVAIL | QLOGIC_IB_I_GPIO | \
+		QLOGIC_IB_I_SERDESTRIMDONE)
+
+#define IB_HWE_BITSEXTANT \
+	       (HWE_MASK(RXEMemParityErr) | \
+		HWE_MASK(TXEMemParityErr) | \
+		(QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK <<	 \
+		 QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT) | \
+		QLOGIC_IB_HWE_PCIE1PLLFAILED | \
+		QLOGIC_IB_HWE_PCIE0PLLFAILED | \
+		QLOGIC_IB_HWE_PCIEPOISONEDTLP | \
+		QLOGIC_IB_HWE_PCIECPLTIMEOUT | \
+		QLOGIC_IB_HWE_PCIEBUSPARITYXTLH | \
+		QLOGIC_IB_HWE_PCIEBUSPARITYXADM | \
+		QLOGIC_IB_HWE_PCIEBUSPARITYRADM | \
+		HWE_MASK(PowerOnBISTFailed) |	  \
+		QLOGIC_IB_HWE_COREPLL_FBSLIP | \
+		QLOGIC_IB_HWE_COREPLL_RFSLIP | \
+		QLOGIC_IB_HWE_SERDESPLLFAILED | \
+		HWE_MASK(IBCBusToSPCParityErr) | \
+		HWE_MASK(IBCBusFromSPCParityErr) | \
+		QLOGIC_IB_HWE_PCIECPLDATAQUEUEERR | \
+		QLOGIC_IB_HWE_PCIECPLHDRQUEUEERR | \
+		QLOGIC_IB_HWE_SDMAMEMREADERR | \
+		QLOGIC_IB_HWE_CLK_UC_PLLNOTLOCKED | \
+		QLOGIC_IB_HWE_PCIESERDESQ0PCLKNOTDETECT | \
+		QLOGIC_IB_HWE_PCIESERDESQ1PCLKNOTDETECT | \
+		QLOGIC_IB_HWE_PCIESERDESQ2PCLKNOTDETECT | \
+		QLOGIC_IB_HWE_PCIESERDESQ3PCLKNOTDETECT | \
+		QLOGIC_IB_HWE_DDSRXEQMEMORYPARITYERR | \
+		QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR | \
+		QLOGIC_IB_HWE_PCIE_UC_OCT0MEMORYPARITYERR | \
+		QLOGIC_IB_HWE_PCIE_UC_OCT1MEMORYPARITYERR)
+
+#define IB_E_BITSEXTANT							\
+	(ERR_MASK(RcvFormatErr) | ERR_MASK(RcvVCRCErr) |		\
+	 ERR_MASK(RcvICRCErr) | ERR_MASK(RcvMinPktLenErr) |		\
+	 ERR_MASK(RcvMaxPktLenErr) | ERR_MASK(RcvLongPktLenErr) |	\
+	 ERR_MASK(RcvShortPktLenErr) | ERR_MASK(RcvUnexpectedCharErr) | \
+	 ERR_MASK(RcvUnsupportedVLErr) | ERR_MASK(RcvEBPErr) |		\
+	 ERR_MASK(RcvIBFlowErr) | ERR_MASK(RcvBadVersionErr) |		\
+	 ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr) |		\
+	 ERR_MASK(RcvBadTidErr) | ERR_MASK(RcvHdrLenErr) |		\
+	 ERR_MASK(RcvHdrErr) | ERR_MASK(RcvIBLostLinkErr) |		\
+	 ERR_MASK(SendSpecialTriggerErr) |				\
+	 ERR_MASK(SDmaDisabledErr) | ERR_MASK(SendMinPktLenErr) |	\
+	 ERR_MASK(SendMaxPktLenErr) | ERR_MASK(SendUnderRunErr) |	\
+	 ERR_MASK(SendPktLenErr) | ERR_MASK(SendDroppedSmpPktErr) |	\
+	 ERR_MASK(SendDroppedDataPktErr) |				\
+	 ERR_MASK(SendPioArmLaunchErr) |				\
+	 ERR_MASK(SendUnexpectedPktNumErr) |				\
+	 ERR_MASK(SendUnsupportedVLErr) | ERR_MASK(SendBufMisuseErr) |	\
+	 ERR_MASK(SDmaGenMismatchErr) | ERR_MASK(SDmaOutOfBoundErr) |	\
+	 ERR_MASK(SDmaTailOutOfBoundErr) | ERR_MASK(SDmaBaseErr) |	\
+	 ERR_MASK(SDma1stDescErr) | ERR_MASK(SDmaRpyTagErr) |		\
+	 ERR_MASK(SDmaDwEnErr) | ERR_MASK(SDmaMissingDwErr) |		\
+	 ERR_MASK(SDmaUnexpDataErr) |					\
+	 ERR_MASK(IBStatusChanged) | ERR_MASK(InvalidAddrErr) |		\
+	 ERR_MASK(ResetNegated) | ERR_MASK(HardwareErr) |		\
+	 ERR_MASK(SDmaDescAddrMisalignErr) |				\
+	 ERR_MASK(InvalidEEPCmd))
+
+/* kr_hwerrclear, kr_hwerrmask, kr_hwerrstatus, bits */
+#define QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK  0x00000000000000ffULL
+#define QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT 0
+#define QLOGIC_IB_HWE_PCIEPOISONEDTLP      0x0000000010000000ULL
+#define QLOGIC_IB_HWE_PCIECPLTIMEOUT       0x0000000020000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYXTLH    0x0000000040000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYXADM    0x0000000080000000ULL
+#define QLOGIC_IB_HWE_PCIEBUSPARITYRADM    0x0000000100000000ULL
+#define QLOGIC_IB_HWE_COREPLL_FBSLIP       0x0080000000000000ULL
+#define QLOGIC_IB_HWE_COREPLL_RFSLIP       0x0100000000000000ULL
+#define QLOGIC_IB_HWE_PCIE1PLLFAILED       0x0400000000000000ULL
+#define QLOGIC_IB_HWE_PCIE0PLLFAILED       0x0800000000000000ULL
+#define QLOGIC_IB_HWE_SERDESPLLFAILED      0x1000000000000000ULL
+/* specific to this chip */
+#define QLOGIC_IB_HWE_PCIECPLDATAQUEUEERR         0x0000000000000040ULL
+#define QLOGIC_IB_HWE_PCIECPLHDRQUEUEERR          0x0000000000000080ULL
+#define QLOGIC_IB_HWE_SDMAMEMREADERR              0x0000000010000000ULL
+#define QLOGIC_IB_HWE_CLK_UC_PLLNOTLOCKED          0x2000000000000000ULL
+#define QLOGIC_IB_HWE_PCIESERDESQ0PCLKNOTDETECT   0x0100000000000000ULL
+#define QLOGIC_IB_HWE_PCIESERDESQ1PCLKNOTDETECT   0x0200000000000000ULL
+#define QLOGIC_IB_HWE_PCIESERDESQ2PCLKNOTDETECT   0x0400000000000000ULL
+#define QLOGIC_IB_HWE_PCIESERDESQ3PCLKNOTDETECT   0x0800000000000000ULL
+#define QLOGIC_IB_HWE_DDSRXEQMEMORYPARITYERR       0x0000008000000000ULL
+#define QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR        0x0000004000000000ULL
+#define QLOGIC_IB_HWE_PCIE_UC_OCT0MEMORYPARITYERR 0x0000001000000000ULL
+#define QLOGIC_IB_HWE_PCIE_UC_OCT1MEMORYPARITYERR 0x0000002000000000ULL
+
+#define IBA7220_IBCC_LINKCMD_SHIFT 19
+
+/* kr_ibcddrctrl bits */
+#define IBA7220_IBC_DLIDLMC_MASK        0xFFFFFFFFUL
+#define IBA7220_IBC_DLIDLMC_SHIFT       32
+
+#define IBA7220_IBC_HRTBT_MASK  (SYM_RMASK(IBCDDRCtrl, HRTBT_AUTO) | \
+				 SYM_RMASK(IBCDDRCtrl, HRTBT_ENB))
+#define IBA7220_IBC_HRTBT_SHIFT SYM_LSB(IBCDDRCtrl, HRTBT_ENB)
+
+#define IBA7220_IBC_LANE_REV_SUPPORTED (1<<8)
+#define IBA7220_IBC_LREV_MASK   1
+#define IBA7220_IBC_LREV_SHIFT  8
+#define IBA7220_IBC_RXPOL_MASK  1
+#define IBA7220_IBC_RXPOL_SHIFT 7
+#define IBA7220_IBC_WIDTH_SHIFT 5
+#define IBA7220_IBC_WIDTH_MASK  0x3
+#define IBA7220_IBC_WIDTH_1X_ONLY       (0 << IBA7220_IBC_WIDTH_SHIFT)
+#define IBA7220_IBC_WIDTH_4X_ONLY       (1 << IBA7220_IBC_WIDTH_SHIFT)
+#define IBA7220_IBC_WIDTH_AUTONEG       (2 << IBA7220_IBC_WIDTH_SHIFT)
+#define IBA7220_IBC_SPEED_AUTONEG       (1 << 1)
+#define IBA7220_IBC_SPEED_SDR           (1 << 2)
+#define IBA7220_IBC_SPEED_DDR           (1 << 3)
+#define IBA7220_IBC_SPEED_AUTONEG_MASK  (0x7 << 1)
+#define IBA7220_IBC_IBTA_1_2_MASK       (1)
+
+/* kr_ibcddrstatus */
+/* link latency shift is 0, don't bother defining */
+#define IBA7220_DDRSTAT_LINKLAT_MASK    0x3ffffff
+
+/* kr_extstatus bits */
+#define QLOGIC_IB_EXTS_FREQSEL 0x2
+#define QLOGIC_IB_EXTS_SERDESSEL 0x4
+#define QLOGIC_IB_EXTS_MEMBIST_ENDTEST     0x0000000000004000
+#define QLOGIC_IB_EXTS_MEMBIST_DISABLED    0x0000000000008000
+
+/* kr_xgxsconfig bits */
+#define QLOGIC_IB_XGXS_RESET          0x5ULL
+#define QLOGIC_IB_XGXS_FC_SAFE        (1ULL << 63)
+
+/* kr_rcvpktledcnt */
+#define IBA7220_LEDBLINK_ON_SHIFT 32 /* 4ns period on after packet */
+#define IBA7220_LEDBLINK_OFF_SHIFT 0 /* 4ns period off before next on */
+
+#define _QIB_GPIO_SDA_NUM 1
+#define _QIB_GPIO_SCL_NUM 0
+#define QIB_TWSI_EEPROM_DEV 0xA2 /* All Production 7220 cards. */
+#define QIB_TWSI_TEMP_DEV 0x98
+
+/* HW counter clock is at 4nsec */
+#define QIB_7220_PSXMITWAIT_CHECK_RATE 4000
+
+#define IBA7220_R_INTRAVAIL_SHIFT 17
+#define IBA7220_R_PKEY_DIS_SHIFT 34
+#define IBA7220_R_TAILUPD_SHIFT 35
+#define IBA7220_R_CTXTCFG_SHIFT 36
+
+#define IBA7220_HDRHEAD_PKTINT_SHIFT 32 /* interrupt cnt in upper 32 bits */
+
+/*
+ * the size bits give us 2^N, in KB units.  0 marks as invalid,
+ * and 7 is reserved.  We currently use only 2KB and 4KB
+ */
+#define IBA7220_TID_SZ_SHIFT 37 /* shift to 3bit size selector */
+#define IBA7220_TID_SZ_2K (1UL << IBA7220_TID_SZ_SHIFT) /* 2KB */
+#define IBA7220_TID_SZ_4K (2UL << IBA7220_TID_SZ_SHIFT) /* 4KB */
+#define IBA7220_TID_PA_SHIFT 11U /* TID addr in chip stored w/o low bits */
+#define PBC_7220_VL15_SEND (1ULL << 63) /* pbc; VL15, no credit check */
+#define PBC_7220_VL15_SEND_CTRL (1ULL << 31) /* control version of same */
+
+#define AUTONEG_TRIES 5 /* sequential retries to negotiate DDR */
+
+/* packet rate matching delay multiplier */
+static u8 rate_to_delay[2][2] = {
+	/* 1x, 4x */
+	{   8, 2 }, /* SDR */
+	{   4, 1 }  /* DDR */
+};
+
+static u8 ib_rate_to_delay[IB_RATE_120_GBPS + 1] = {
+	[IB_RATE_2_5_GBPS] = 8,
+	[IB_RATE_5_GBPS] = 4,
+	[IB_RATE_10_GBPS] = 2,
+	[IB_RATE_20_GBPS] = 1
+};
+
+#define IBA7220_LINKSPEED_SHIFT SYM_LSB(IBCStatus, LinkSpeedActive)
+#define IBA7220_LINKWIDTH_SHIFT SYM_LSB(IBCStatus, LinkWidthActive)
+
+/* link training states, from IBC */
+#define IB_7220_LT_STATE_DISABLED        0x00
+#define IB_7220_LT_STATE_LINKUP          0x01
+#define IB_7220_LT_STATE_POLLACTIVE      0x02
+#define IB_7220_LT_STATE_POLLQUIET       0x03
+#define IB_7220_LT_STATE_SLEEPDELAY      0x04
+#define IB_7220_LT_STATE_SLEEPQUIET      0x05
+#define IB_7220_LT_STATE_CFGDEBOUNCE     0x08
+#define IB_7220_LT_STATE_CFGRCVFCFG      0x09
+#define IB_7220_LT_STATE_CFGWAITRMT      0x0a
+#define IB_7220_LT_STATE_CFGIDLE 0x0b
+#define IB_7220_LT_STATE_RECOVERRETRAIN  0x0c
+#define IB_7220_LT_STATE_RECOVERWAITRMT  0x0e
+#define IB_7220_LT_STATE_RECOVERIDLE     0x0f
+
+/* link state machine states from IBC */
+#define IB_7220_L_STATE_DOWN             0x0
+#define IB_7220_L_STATE_INIT             0x1
+#define IB_7220_L_STATE_ARM              0x2
+#define IB_7220_L_STATE_ACTIVE           0x3
+#define IB_7220_L_STATE_ACT_DEFER        0x4
+
+static const u8 qib_7220_physportstate[0x20] = {
+	[IB_7220_LT_STATE_DISABLED] = IB_PHYSPORTSTATE_DISABLED,
+	[IB_7220_LT_STATE_LINKUP] = IB_PHYSPORTSTATE_LINKUP,
+	[IB_7220_LT_STATE_POLLACTIVE] = IB_PHYSPORTSTATE_POLL,
+	[IB_7220_LT_STATE_POLLQUIET] = IB_PHYSPORTSTATE_POLL,
+	[IB_7220_LT_STATE_SLEEPDELAY] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_7220_LT_STATE_SLEEPQUIET] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_7220_LT_STATE_CFGDEBOUNCE] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7220_LT_STATE_CFGRCVFCFG] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7220_LT_STATE_CFGWAITRMT] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7220_LT_STATE_CFGIDLE] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7220_LT_STATE_RECOVERRETRAIN] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_7220_LT_STATE_RECOVERWAITRMT] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_7220_LT_STATE_RECOVERIDLE] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[0x10] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x11] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x12] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x13] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x14] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x15] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x16] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x17] = IB_PHYSPORTSTATE_CFG_TRAIN
+};
+
+int qib_special_trigger;
+module_param_named(special_trigger, qib_special_trigger, int, S_IRUGO);
+MODULE_PARM_DESC(special_trigger, "Enable SpecialTrigger arm/launch");
+
+#define IBCBUSFRSPCPARITYERR HWE_MASK(IBCBusFromSPCParityErr)
+#define IBCBUSTOSPCPARITYERR HWE_MASK(IBCBusToSPCParityErr)
+
+#define SYM_MASK_BIT(regname, fldname, bit) ((u64) \
+	(1ULL << (SYM_LSB(regname, fldname) + (bit))))
+
+#define TXEMEMPARITYERR_PIOBUF \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 0)
+#define TXEMEMPARITYERR_PIOPBC \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 1)
+#define TXEMEMPARITYERR_PIOLAUNCHFIFO \
+	SYM_MASK_BIT(HwErrMask, TXEMemParityErrMask, 2)
+
+#define RXEMEMPARITYERR_RCVBUF \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 0)
+#define RXEMEMPARITYERR_LOOKUPQ \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 1)
+#define RXEMEMPARITYERR_EXPTID \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 2)
+#define RXEMEMPARITYERR_EAGERTID \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 3)
+#define RXEMEMPARITYERR_FLAGBUF \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 4)
+#define RXEMEMPARITYERR_DATAINFO \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 5)
+#define RXEMEMPARITYERR_HDRINFO \
+	SYM_MASK_BIT(HwErrMask, RXEMemParityErrMask, 6)
+
+/* 7220 specific hardware errors... */
+static const struct qib_hwerror_msgs qib_7220_hwerror_msgs[] = {
+	/* generic hardware errors */
+	QLOGIC_IB_HWE_MSG(IBCBUSFRSPCPARITYERR, "QIB2IB Parity"),
+	QLOGIC_IB_HWE_MSG(IBCBUSTOSPCPARITYERR, "IB2QIB Parity"),
+
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOBUF,
+			  "TXE PIOBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOPBC,
+			  "TXE PIOPBC Memory Parity"),
+	QLOGIC_IB_HWE_MSG(TXEMEMPARITYERR_PIOLAUNCHFIFO,
+			  "TXE PIOLAUNCHFIFO Memory Parity"),
+
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_RCVBUF,
+			  "RXE RCVBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_LOOKUPQ,
+			  "RXE LOOKUPQ Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_EAGERTID,
+			  "RXE EAGERTID Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_EXPTID,
+			  "RXE EXPTID Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_FLAGBUF,
+			  "RXE FLAGBUF Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_DATAINFO,
+			  "RXE DATAINFO Memory Parity"),
+	QLOGIC_IB_HWE_MSG(RXEMEMPARITYERR_HDRINFO,
+			  "RXE HDRINFO Memory Parity"),
+
+	/* chip-specific hardware errors */
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEPOISONEDTLP,
+			  "PCIe Poisoned TLP"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIECPLTIMEOUT,
+			  "PCIe completion timeout"),
+	/*
+	 * In practice, it's unlikely wthat we'll see PCIe PLL, or bus
+	 * parity or memory parity error failures, because most likely we
+	 * won't be able to talk to the core of the chip.  Nonetheless, we
+	 * might see them, if they are in parts of the PCIe core that aren't
+	 * essential.
+	 */
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE1PLLFAILED,
+			  "PCIePLL1"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE0PLLFAILED,
+			  "PCIePLL0"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYXTLH,
+			  "PCIe XTLH core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYXADM,
+			  "PCIe ADM TX core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIEBUSPARITYRADM,
+			  "PCIe ADM RX core parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_SERDESPLLFAILED,
+			  "SerDes PLL"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIECPLDATAQUEUEERR,
+			  "PCIe cpl header queue"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIECPLHDRQUEUEERR,
+			  "PCIe cpl data queue"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_SDMAMEMREADERR,
+			  "Send DMA memory read"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_CLK_UC_PLLNOTLOCKED,
+			  "uC PLL clock not locked"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIESERDESQ0PCLKNOTDETECT,
+			  "PCIe serdes Q0 no clock"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIESERDESQ1PCLKNOTDETECT,
+			  "PCIe serdes Q1 no clock"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIESERDESQ2PCLKNOTDETECT,
+			  "PCIe serdes Q2 no clock"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIESERDESQ3PCLKNOTDETECT,
+			  "PCIe serdes Q3 no clock"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_DDSRXEQMEMORYPARITYERR,
+			  "DDS RXEQ memory parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR,
+			  "IB uC memory parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE_UC_OCT0MEMORYPARITYERR,
+			  "PCIe uC oct0 memory parity"),
+	QLOGIC_IB_HWE_MSG(QLOGIC_IB_HWE_PCIE_UC_OCT1MEMORYPARITYERR,
+			  "PCIe uC oct1 memory parity"),
+};
+
+#define RXE_PARITY (RXEMEMPARITYERR_EAGERTID|RXEMEMPARITYERR_EXPTID)
+
+#define QLOGIC_IB_E_PKTERRS (\
+		ERR_MASK(SendPktLenErr) |				\
+		ERR_MASK(SendDroppedDataPktErr) |			\
+		ERR_MASK(RcvVCRCErr) |					\
+		ERR_MASK(RcvICRCErr) |					\
+		ERR_MASK(RcvShortPktLenErr) |				\
+		ERR_MASK(RcvEBPErr))
+
+/* Convenience for decoding Send DMA errors */
+#define QLOGIC_IB_E_SDMAERRS ( \
+		ERR_MASK(SDmaGenMismatchErr) |				\
+		ERR_MASK(SDmaOutOfBoundErr) |				\
+		ERR_MASK(SDmaTailOutOfBoundErr) | ERR_MASK(SDmaBaseErr) | \
+		ERR_MASK(SDma1stDescErr) | ERR_MASK(SDmaRpyTagErr) |	\
+		ERR_MASK(SDmaDwEnErr) | ERR_MASK(SDmaMissingDwErr) |	\
+		ERR_MASK(SDmaUnexpDataErr) |				\
+		ERR_MASK(SDmaDescAddrMisalignErr) |			\
+		ERR_MASK(SDmaDisabledErr) |				\
+		ERR_MASK(SendBufMisuseErr))
+
+/* These are all rcv-related errors which we want to count for stats */
+#define E_SUM_PKTERRS \
+	(ERR_MASK(RcvHdrLenErr) | ERR_MASK(RcvBadTidErr) |		\
+	 ERR_MASK(RcvBadVersionErr) | ERR_MASK(RcvHdrErr) |		\
+	 ERR_MASK(RcvLongPktLenErr) | ERR_MASK(RcvShortPktLenErr) |	\
+	 ERR_MASK(RcvMaxPktLenErr) | ERR_MASK(RcvMinPktLenErr) |	\
+	 ERR_MASK(RcvFormatErr) | ERR_MASK(RcvUnsupportedVLErr) |	\
+	 ERR_MASK(RcvUnexpectedCharErr) | ERR_MASK(RcvEBPErr))
+
+/* These are all send-related errors which we want to count for stats */
+#define E_SUM_ERRS \
+	(ERR_MASK(SendPioArmLaunchErr) | ERR_MASK(SendUnexpectedPktNumErr) | \
+	 ERR_MASK(SendDroppedDataPktErr) | ERR_MASK(SendDroppedSmpPktErr) | \
+	 ERR_MASK(SendMaxPktLenErr) | ERR_MASK(SendUnsupportedVLErr) |	\
+	 ERR_MASK(SendMinPktLenErr) | ERR_MASK(SendPktLenErr) |		\
+	 ERR_MASK(InvalidAddrErr))
+
+/*
+ * this is similar to E_SUM_ERRS, but can't ignore armlaunch, don't ignore
+ * errors not related to freeze and cancelling buffers.  Can't ignore
+ * armlaunch because could get more while still cleaning up, and need
+ * to cancel those as they happen.
+ */
+#define E_SPKT_ERRS_IGNORE \
+	(ERR_MASK(SendDroppedDataPktErr) | ERR_MASK(SendDroppedSmpPktErr) | \
+	 ERR_MASK(SendMaxPktLenErr) | ERR_MASK(SendMinPktLenErr) |	\
+	 ERR_MASK(SendPktLenErr))
+
+/*
+ * these are errors that can occur when the link changes state while
+ * a packet is being sent or received.  This doesn't cover things
+ * like EBP or VCRC that can be the result of a sending having the
+ * link change state, so we receive a "known bad" packet.
+ */
+#define E_SUM_LINK_PKTERRS \
+	(ERR_MASK(SendDroppedDataPktErr) | ERR_MASK(SendDroppedSmpPktErr) | \
+	 ERR_MASK(SendMinPktLenErr) | ERR_MASK(SendPktLenErr) |		\
+	 ERR_MASK(RcvShortPktLenErr) | ERR_MASK(RcvMinPktLenErr) |	\
+	 ERR_MASK(RcvUnexpectedCharErr))
+
+static void autoneg_7220_work(struct work_struct *);
+static u32 __iomem *qib_7220_getsendbuf(struct qib_pportdata *, u64, u32 *);
+
+/*
+ * Called when we might have an error that is specific to a particular
+ * PIO buffer, and may need to cancel that buffer, so it can be re-used.
+ * because we don't need to force the update of pioavail.
+ */
+static void qib_disarm_7220_senderrbufs(struct qib_pportdata *ppd)
+{
+	unsigned long sbuf[3];
+	struct qib_devdata *dd = ppd->dd;
+
+	/*
+	 * It's possible that sendbuffererror could have bits set; might
+	 * have already done this as a result of hardware error handling.
+	 */
+	/* read these before writing errorclear */
+	sbuf[0] = qib_read_kreg64(dd, kr_sendbuffererror);
+	sbuf[1] = qib_read_kreg64(dd, kr_sendbuffererror + 1);
+	sbuf[2] = qib_read_kreg64(dd, kr_sendbuffererror + 2);
+
+	if (sbuf[0] || sbuf[1] || sbuf[2])
+		qib_disarm_piobufs_set(dd, sbuf,
+				       dd->piobcnt2k + dd->piobcnt4k);
+}
+
+static void qib_7220_txe_recover(struct qib_devdata *dd)
+{
+	qib_devinfo(dd->pcidev, "Recovering from TXE PIO parity error\n");
+	qib_disarm_7220_senderrbufs(dd->pport);
+}
+
+/*
+ * This is called with interrupts disabled and sdma_lock held.
+ */
+static void qib_7220_sdma_sendctrl(struct qib_pportdata *ppd, unsigned op)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 set_sendctrl = 0;
+	u64 clr_sendctrl = 0;
+
+	if (op & QIB_SDMA_SENDCTRL_OP_ENABLE)
+		set_sendctrl |= SYM_MASK(SendCtrl, SDmaEnable);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl, SDmaEnable);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_INTENABLE)
+		set_sendctrl |= SYM_MASK(SendCtrl, SDmaIntEnable);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl, SDmaIntEnable);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_HALT)
+		set_sendctrl |= SYM_MASK(SendCtrl, SDmaHalt);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl, SDmaHalt);
+
+	spin_lock(&dd->sendctrl_lock);
+
+	dd->sendctrl |= set_sendctrl;
+	dd->sendctrl &= ~clr_sendctrl;
+
+	qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	spin_unlock(&dd->sendctrl_lock);
+}
+
+static void qib_decode_7220_sdma_errs(struct qib_pportdata *ppd,
+				      u64 err, char *buf, size_t blen)
+{
+	static const struct {
+		u64 err;
+		const char *msg;
+	} errs[] = {
+		{ ERR_MASK(SDmaGenMismatchErr),
+		  "SDmaGenMismatch" },
+		{ ERR_MASK(SDmaOutOfBoundErr),
+		  "SDmaOutOfBound" },
+		{ ERR_MASK(SDmaTailOutOfBoundErr),
+		  "SDmaTailOutOfBound" },
+		{ ERR_MASK(SDmaBaseErr),
+		  "SDmaBase" },
+		{ ERR_MASK(SDma1stDescErr),
+		  "SDma1stDesc" },
+		{ ERR_MASK(SDmaRpyTagErr),
+		  "SDmaRpyTag" },
+		{ ERR_MASK(SDmaDwEnErr),
+		  "SDmaDwEn" },
+		{ ERR_MASK(SDmaMissingDwErr),
+		  "SDmaMissingDw" },
+		{ ERR_MASK(SDmaUnexpDataErr),
+		  "SDmaUnexpData" },
+		{ ERR_MASK(SDmaDescAddrMisalignErr),
+		  "SDmaDescAddrMisalign" },
+		{ ERR_MASK(SendBufMisuseErr),
+		  "SendBufMisuse" },
+		{ ERR_MASK(SDmaDisabledErr),
+		  "SDmaDisabled" },
+	};
+	int i;
+	size_t bidx = 0;
+
+	for (i = 0; i < ARRAY_SIZE(errs); i++) {
+		if (err & errs[i].err)
+			bidx += scnprintf(buf + bidx, blen - bidx,
+					 "%s ", errs[i].msg);
+	}
+}
+
+/*
+ * This is called as part of link down clean up so disarm and flush
+ * all send buffers so that SMP packets can be sent.
+ */
+static void qib_7220_sdma_hw_clean_up(struct qib_pportdata *ppd)
+{
+	/* This will trigger the Abort interrupt */
+	sendctrl_7220_mod(ppd, QIB_SENDCTRL_DISARM_ALL | QIB_SENDCTRL_FLUSH |
+			  QIB_SENDCTRL_AVAIL_BLIP);
+	ppd->dd->upd_pio_shadow  = 1; /* update our idea of what's busy */
+}
+
+static void qib_sdma_7220_setlengen(struct qib_pportdata *ppd)
+{
+	/*
+	 * Set SendDmaLenGen and clear and set
+	 * the MSB of the generation count to enable generation checking
+	 * and load the internal generation counter.
+	 */
+	qib_write_kreg(ppd->dd, kr_senddmalengen, ppd->sdma_descq_cnt);
+	qib_write_kreg(ppd->dd, kr_senddmalengen,
+		       ppd->sdma_descq_cnt |
+		       (1ULL << QIB_7220_SendDmaLenGen_Generation_MSB));
+}
+
+static void qib_7220_sdma_hw_start_up(struct qib_pportdata *ppd)
+{
+	qib_sdma_7220_setlengen(ppd);
+	qib_sdma_update_7220_tail(ppd, 0); /* Set SendDmaTail */
+	ppd->sdma_head_dma[0] = 0;
+}
+
+#define DISABLES_SDMA (							\
+		ERR_MASK(SDmaDisabledErr) |				\
+		ERR_MASK(SDmaBaseErr) |					\
+		ERR_MASK(SDmaTailOutOfBoundErr) |			\
+		ERR_MASK(SDmaOutOfBoundErr) |				\
+		ERR_MASK(SDma1stDescErr) |				\
+		ERR_MASK(SDmaRpyTagErr) |				\
+		ERR_MASK(SDmaGenMismatchErr) |				\
+		ERR_MASK(SDmaDescAddrMisalignErr) |			\
+		ERR_MASK(SDmaMissingDwErr) |				\
+		ERR_MASK(SDmaDwEnErr))
+
+static void sdma_7220_errors(struct qib_pportdata *ppd, u64 errs)
+{
+	unsigned long flags;
+	struct qib_devdata *dd = ppd->dd;
+	char *msg;
+
+	errs &= QLOGIC_IB_E_SDMAERRS;
+
+	msg = dd->cspec->sdmamsgbuf;
+	qib_decode_7220_sdma_errs(ppd, errs, msg, sizeof dd->cspec->sdmamsgbuf);
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	if (errs & ERR_MASK(SendBufMisuseErr)) {
+		unsigned long sbuf[3];
+
+		sbuf[0] = qib_read_kreg64(dd, kr_sendbuffererror);
+		sbuf[1] = qib_read_kreg64(dd, kr_sendbuffererror + 1);
+		sbuf[2] = qib_read_kreg64(dd, kr_sendbuffererror + 2);
+
+		qib_dev_err(ppd->dd,
+			    "IB%u:%u SendBufMisuse: %04lx %016lx %016lx\n",
+			    ppd->dd->unit, ppd->port, sbuf[2], sbuf[1],
+			    sbuf[0]);
+	}
+
+	if (errs & ERR_MASK(SDmaUnexpDataErr))
+		qib_dev_err(dd, "IB%u:%u SDmaUnexpData\n", ppd->dd->unit,
+			    ppd->port);
+
+	switch (ppd->sdma_state.current_state) {
+	case qib_sdma_state_s00_hw_down:
+		/* not expecting any interrupts */
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		/* handled in intr path */
+		break;
+
+	case qib_sdma_state_s20_idle:
+		/* not expecting any interrupts */
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		/* not expecting any interrupts */
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		if (errs & ERR_MASK(SDmaDisabledErr))
+			__qib_sdma_process_event(ppd,
+				qib_sdma_event_e50_hw_cleaned);
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		/* handled in intr path */
+		break;
+
+	case qib_sdma_state_s99_running:
+		if (errs & DISABLES_SDMA)
+			__qib_sdma_process_event(ppd,
+				qib_sdma_event_e7220_err_halted);
+		break;
+	}
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+/*
+ * Decode the error status into strings, deciding whether to always
+ * print * it or not depending on "normal packet errors" vs everything
+ * else.   Return 1 if "real" errors, otherwise 0 if only packet
+ * errors, so caller can decide what to print with the string.
+ */
+static int qib_decode_7220_err(struct qib_devdata *dd, char *buf, size_t blen,
+			       u64 err)
+{
+	int iserr = 1;
+
+	*buf = '\0';
+	if (err & QLOGIC_IB_E_PKTERRS) {
+		if (!(err & ~QLOGIC_IB_E_PKTERRS))
+			iserr = 0;
+		if ((err & ERR_MASK(RcvICRCErr)) &&
+		    !(err & (ERR_MASK(RcvVCRCErr) | ERR_MASK(RcvEBPErr))))
+			strlcat(buf, "CRC ", blen);
+		if (!iserr)
+			goto done;
+	}
+	if (err & ERR_MASK(RcvHdrLenErr))
+		strlcat(buf, "rhdrlen ", blen);
+	if (err & ERR_MASK(RcvBadTidErr))
+		strlcat(buf, "rbadtid ", blen);
+	if (err & ERR_MASK(RcvBadVersionErr))
+		strlcat(buf, "rbadversion ", blen);
+	if (err & ERR_MASK(RcvHdrErr))
+		strlcat(buf, "rhdr ", blen);
+	if (err & ERR_MASK(SendSpecialTriggerErr))
+		strlcat(buf, "sendspecialtrigger ", blen);
+	if (err & ERR_MASK(RcvLongPktLenErr))
+		strlcat(buf, "rlongpktlen ", blen);
+	if (err & ERR_MASK(RcvMaxPktLenErr))
+		strlcat(buf, "rmaxpktlen ", blen);
+	if (err & ERR_MASK(RcvMinPktLenErr))
+		strlcat(buf, "rminpktlen ", blen);
+	if (err & ERR_MASK(SendMinPktLenErr))
+		strlcat(buf, "sminpktlen ", blen);
+	if (err & ERR_MASK(RcvFormatErr))
+		strlcat(buf, "rformaterr ", blen);
+	if (err & ERR_MASK(RcvUnsupportedVLErr))
+		strlcat(buf, "runsupvl ", blen);
+	if (err & ERR_MASK(RcvUnexpectedCharErr))
+		strlcat(buf, "runexpchar ", blen);
+	if (err & ERR_MASK(RcvIBFlowErr))
+		strlcat(buf, "ribflow ", blen);
+	if (err & ERR_MASK(SendUnderRunErr))
+		strlcat(buf, "sunderrun ", blen);
+	if (err & ERR_MASK(SendPioArmLaunchErr))
+		strlcat(buf, "spioarmlaunch ", blen);
+	if (err & ERR_MASK(SendUnexpectedPktNumErr))
+		strlcat(buf, "sunexperrpktnum ", blen);
+	if (err & ERR_MASK(SendDroppedSmpPktErr))
+		strlcat(buf, "sdroppedsmppkt ", blen);
+	if (err & ERR_MASK(SendMaxPktLenErr))
+		strlcat(buf, "smaxpktlen ", blen);
+	if (err & ERR_MASK(SendUnsupportedVLErr))
+		strlcat(buf, "sunsupVL ", blen);
+	if (err & ERR_MASK(InvalidAddrErr))
+		strlcat(buf, "invalidaddr ", blen);
+	if (err & ERR_MASK(RcvEgrFullErr))
+		strlcat(buf, "rcvegrfull ", blen);
+	if (err & ERR_MASK(RcvHdrFullErr))
+		strlcat(buf, "rcvhdrfull ", blen);
+	if (err & ERR_MASK(IBStatusChanged))
+		strlcat(buf, "ibcstatuschg ", blen);
+	if (err & ERR_MASK(RcvIBLostLinkErr))
+		strlcat(buf, "riblostlink ", blen);
+	if (err & ERR_MASK(HardwareErr))
+		strlcat(buf, "hardware ", blen);
+	if (err & ERR_MASK(ResetNegated))
+		strlcat(buf, "reset ", blen);
+	if (err & QLOGIC_IB_E_SDMAERRS)
+		qib_decode_7220_sdma_errs(dd->pport, err, buf, blen);
+	if (err & ERR_MASK(InvalidEEPCmd))
+		strlcat(buf, "invalideepromcmd ", blen);
+done:
+	return iserr;
+}
+
+static void reenable_7220_chase(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+	ppd->cpspec->chase_timer.expires = 0;
+	qib_set_ib_7220_lstate(ppd, QLOGIC_IB_IBCC_LINKCMD_DOWN,
+		QLOGIC_IB_IBCC_LINKINITCMD_POLL);
+}
+
+static void handle_7220_chase(struct qib_pportdata *ppd, u64 ibcst)
+{
+	u8 ibclt;
+	unsigned long tnow;
+
+	ibclt = (u8)SYM_FIELD(ibcst, IBCStatus, LinkTrainingState);
+
+	/*
+	 * Detect and handle the state chase issue, where we can
+	 * get stuck if we are unlucky on timing on both sides of
+	 * the link.   If we are, we disable, set a timer, and
+	 * then re-enable.
+	 */
+	switch (ibclt) {
+	case IB_7220_LT_STATE_CFGRCVFCFG:
+	case IB_7220_LT_STATE_CFGWAITRMT:
+	case IB_7220_LT_STATE_TXREVLANES:
+	case IB_7220_LT_STATE_CFGENH:
+		tnow = jiffies;
+		if (ppd->cpspec->chase_end &&
+		    time_after(tnow, ppd->cpspec->chase_end)) {
+			ppd->cpspec->chase_end = 0;
+			qib_set_ib_7220_lstate(ppd,
+				QLOGIC_IB_IBCC_LINKCMD_DOWN,
+				QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+			ppd->cpspec->chase_timer.expires = jiffies +
+				QIB_CHASE_DIS_TIME;
+			add_timer(&ppd->cpspec->chase_timer);
+		} else if (!ppd->cpspec->chase_end)
+			ppd->cpspec->chase_end = tnow + QIB_CHASE_TIME;
+		break;
+
+	default:
+		ppd->cpspec->chase_end = 0;
+		break;
+	}
+}
+
+static void handle_7220_errors(struct qib_devdata *dd, u64 errs)
+{
+	char *msg;
+	u64 ignore_this_time = 0;
+	u64 iserr = 0;
+	int log_idx;
+	struct qib_pportdata *ppd = dd->pport;
+	u64 mask;
+
+	/* don't report errors that are masked */
+	errs &= dd->cspec->errormask;
+	msg = dd->cspec->emsgbuf;
+
+	/* do these first, they are most important */
+	if (errs & ERR_MASK(HardwareErr))
+		qib_7220_handle_hwerrors(dd, msg, sizeof dd->cspec->emsgbuf);
+	else
+		for (log_idx = 0; log_idx < QIB_EEP_LOG_CNT; ++log_idx)
+			if (errs & dd->eep_st_masks[log_idx].errs_to_log)
+				qib_inc_eeprom_err(dd, log_idx, 1);
+
+	if (errs & QLOGIC_IB_E_SDMAERRS)
+		sdma_7220_errors(ppd, errs);
+
+	if (errs & ~IB_E_BITSEXTANT)
+		qib_dev_err(dd,
+			"error interrupt with unknown errors %llx set\n",
+			(unsigned long long) (errs & ~IB_E_BITSEXTANT));
+
+	if (errs & E_SUM_ERRS) {
+		qib_disarm_7220_senderrbufs(ppd);
+		if ((errs & E_SUM_LINK_PKTERRS) &&
+		    !(ppd->lflags & QIBL_LINKACTIVE)) {
+			/*
+			 * This can happen when trying to bring the link
+			 * up, but the IB link changes state at the "wrong"
+			 * time. The IB logic then complains that the packet
+			 * isn't valid.  We don't want to confuse people, so
+			 * we just don't print them, except at debug
+			 */
+			ignore_this_time = errs & E_SUM_LINK_PKTERRS;
+		}
+	} else if ((errs & E_SUM_LINK_PKTERRS) &&
+		   !(ppd->lflags & QIBL_LINKACTIVE)) {
+		/*
+		 * This can happen when SMA is trying to bring the link
+		 * up, but the IB link changes state at the "wrong" time.
+		 * The IB logic then complains that the packet isn't
+		 * valid.  We don't want to confuse people, so we just
+		 * don't print them, except at debug
+		 */
+		ignore_this_time = errs & E_SUM_LINK_PKTERRS;
+	}
+
+	qib_write_kreg(dd, kr_errclear, errs);
+
+	errs &= ~ignore_this_time;
+	if (!errs)
+		goto done;
+
+	/*
+	 * The ones we mask off are handled specially below
+	 * or above.  Also mask SDMADISABLED by default as it
+	 * is too chatty.
+	 */
+	mask = ERR_MASK(IBStatusChanged) |
+		ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr) |
+		ERR_MASK(HardwareErr) | ERR_MASK(SDmaDisabledErr);
+
+	qib_decode_7220_err(dd, msg, sizeof dd->cspec->emsgbuf, errs & ~mask);
+
+	if (errs & E_SUM_PKTERRS)
+		qib_stats.sps_rcverrs++;
+	if (errs & E_SUM_ERRS)
+		qib_stats.sps_txerrs++;
+	iserr = errs & ~(E_SUM_PKTERRS | QLOGIC_IB_E_PKTERRS |
+			 ERR_MASK(SDmaDisabledErr));
+
+	if (errs & ERR_MASK(IBStatusChanged)) {
+		u64 ibcs;
+
+		ibcs = qib_read_kreg64(dd, kr_ibcstatus);
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+			handle_7220_chase(ppd, ibcs);
+
+		/* Update our picture of width and speed from chip */
+		ppd->link_width_active =
+			((ibcs >> IBA7220_LINKWIDTH_SHIFT) & 1) ?
+			    IB_WIDTH_4X : IB_WIDTH_1X;
+		ppd->link_speed_active =
+			((ibcs >> IBA7220_LINKSPEED_SHIFT) & 1) ?
+			    QIB_IB_DDR : QIB_IB_SDR;
+
+		/*
+		 * Since going into a recovery state causes the link state
+		 * to go down and since recovery is transitory, it is better
+		 * if we "miss" ever seeing the link training state go into
+		 * recovery (i.e., ignore this transition for link state
+		 * special handling purposes) without updating lastibcstat.
+		 */
+		if (qib_7220_phys_portstate(ibcs) !=
+					    IB_PHYSPORTSTATE_LINK_ERR_RECOVER)
+			qib_handle_e_ibstatuschanged(ppd, ibcs);
+	}
+
+	if (errs & ERR_MASK(ResetNegated)) {
+		qib_dev_err(dd,
+			"Got reset, requires re-init (unload and reload driver)\n");
+		dd->flags &= ~QIB_INITTED;  /* needs re-init */
+		/* mark as having had error */
+		*dd->devstatusp |= QIB_STATUS_HWERROR;
+		*dd->pport->statusp &= ~QIB_STATUS_IB_CONF;
+	}
+
+	if (*msg && iserr)
+		qib_dev_porterr(dd, ppd->port, "%s error\n", msg);
+
+	if (ppd->state_wanted & ppd->lflags)
+		wake_up_interruptible(&ppd->state_wait);
+
+	/*
+	 * If there were hdrq or egrfull errors, wake up any processes
+	 * waiting in poll.  We used to try to check which contexts had
+	 * the overflow, but given the cost of that and the chip reads
+	 * to support it, it's better to just wake everybody up if we
+	 * get an overflow; waiters can poll again if it's not them.
+	 */
+	if (errs & (ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr))) {
+		qib_handle_urcv(dd, ~0U);
+		if (errs & ERR_MASK(RcvEgrFullErr))
+			qib_stats.sps_buffull++;
+		else
+			qib_stats.sps_hdrfull++;
+	}
+done:
+	return;
+}
+
+/* enable/disable chip from delivering interrupts */
+static void qib_7220_set_intr_state(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		if (dd->flags & QIB_BADINTR)
+			return;
+		qib_write_kreg(dd, kr_intmask, ~0ULL);
+		/* force re-interrupt of any pending interrupts. */
+		qib_write_kreg(dd, kr_intclear, 0ULL);
+	} else
+		qib_write_kreg(dd, kr_intmask, 0ULL);
+}
+
+/*
+ * Try to cleanup as much as possible for anything that might have gone
+ * wrong while in freeze mode, such as pio buffers being written by user
+ * processes (causing armlaunch), send errors due to going into freeze mode,
+ * etc., and try to avoid causing extra interrupts while doing so.
+ * Forcibly update the in-memory pioavail register copies after cleanup
+ * because the chip won't do it while in freeze mode (the register values
+ * themselves are kept correct).
+ * Make sure that we don't lose any important interrupts by using the chip
+ * feature that says that writing 0 to a bit in *clear that is set in
+ * *status will cause an interrupt to be generated again (if allowed by
+ * the *mask value).
+ * This is in chip-specific code because of all of the register accesses,
+ * even though the details are similar on most chips.
+ */
+static void qib_7220_clear_freeze(struct qib_devdata *dd)
+{
+	/* disable error interrupts, to avoid confusion */
+	qib_write_kreg(dd, kr_errmask, 0ULL);
+
+	/* also disable interrupts; errormask is sometimes overwriten */
+	qib_7220_set_intr_state(dd, 0);
+
+	qib_cancel_sends(dd->pport);
+
+	/* clear the freeze, and be sure chip saw it */
+	qib_write_kreg(dd, kr_control, dd->control);
+	qib_read_kreg32(dd, kr_scratch);
+
+	/* force in-memory update now we are out of freeze */
+	qib_force_pio_avail_update(dd);
+
+	/*
+	 * force new interrupt if any hwerr, error or interrupt bits are
+	 * still set, and clear "safe" send packet errors related to freeze
+	 * and cancelling sends.  Re-enable error interrupts before possible
+	 * force of re-interrupt on pending interrupts.
+	 */
+	qib_write_kreg(dd, kr_hwerrclear, 0ULL);
+	qib_write_kreg(dd, kr_errclear, E_SPKT_ERRS_IGNORE);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+	qib_7220_set_intr_state(dd, 1);
+}
+
+/**
+ * qib_7220_handle_hwerrors - display hardware errors.
+ * @dd: the qlogic_ib device
+ * @msg: the output buffer
+ * @msgl: the size of the output buffer
+ *
+ * Use same msg buffer as regular errors to avoid excessive stack
+ * use.  Most hardware errors are catastrophic, but for right now,
+ * we'll print them and continue.  We reuse the same message buffer as
+ * handle_7220_errors() to avoid excessive stack usage.
+ */
+static void qib_7220_handle_hwerrors(struct qib_devdata *dd, char *msg,
+				     size_t msgl)
+{
+	u64 hwerrs;
+	u32 bits, ctrl;
+	int isfatal = 0;
+	char *bitsmsg;
+	int log_idx;
+
+	hwerrs = qib_read_kreg64(dd, kr_hwerrstatus);
+	if (!hwerrs)
+		goto bail;
+	if (hwerrs == ~0ULL) {
+		qib_dev_err(dd,
+			"Read of hardware error status failed (all bits set); ignoring\n");
+		goto bail;
+	}
+	qib_stats.sps_hwerrs++;
+
+	/*
+	 * Always clear the error status register, except MEMBISTFAIL,
+	 * regardless of whether we continue or stop using the chip.
+	 * We want that set so we know it failed, even across driver reload.
+	 * We'll still ignore it in the hwerrmask.  We do this partly for
+	 * diagnostics, but also for support.
+	 */
+	qib_write_kreg(dd, kr_hwerrclear,
+		       hwerrs & ~HWE_MASK(PowerOnBISTFailed));
+
+	hwerrs &= dd->cspec->hwerrmask;
+
+	/* We log some errors to EEPROM, check if we have any of those. */
+	for (log_idx = 0; log_idx < QIB_EEP_LOG_CNT; ++log_idx)
+		if (hwerrs & dd->eep_st_masks[log_idx].hwerrs_to_log)
+			qib_inc_eeprom_err(dd, log_idx, 1);
+	if (hwerrs & ~(TXEMEMPARITYERR_PIOBUF | TXEMEMPARITYERR_PIOPBC |
+		       RXE_PARITY))
+		qib_devinfo(dd->pcidev,
+			"Hardware error: hwerr=0x%llx (cleared)\n",
+			(unsigned long long) hwerrs);
+
+	if (hwerrs & ~IB_HWE_BITSEXTANT)
+		qib_dev_err(dd,
+			"hwerror interrupt with unknown errors %llx set\n",
+			(unsigned long long) (hwerrs & ~IB_HWE_BITSEXTANT));
+
+	if (hwerrs & QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR)
+		qib_sd7220_clr_ibpar(dd);
+
+	ctrl = qib_read_kreg32(dd, kr_control);
+	if ((ctrl & QLOGIC_IB_C_FREEZEMODE) && !dd->diag_client) {
+		/*
+		 * Parity errors in send memory are recoverable by h/w
+		 * just do housekeeping, exit freeze mode and continue.
+		 */
+		if (hwerrs & (TXEMEMPARITYERR_PIOBUF |
+			      TXEMEMPARITYERR_PIOPBC)) {
+			qib_7220_txe_recover(dd);
+			hwerrs &= ~(TXEMEMPARITYERR_PIOBUF |
+				    TXEMEMPARITYERR_PIOPBC);
+		}
+		if (hwerrs)
+			isfatal = 1;
+		else
+			qib_7220_clear_freeze(dd);
+	}
+
+	*msg = '\0';
+
+	if (hwerrs & HWE_MASK(PowerOnBISTFailed)) {
+		isfatal = 1;
+		strlcat(msg,
+			"[Memory BIST test failed, InfiniPath hardware unusable]",
+			msgl);
+		/* ignore from now on, so disable until driver reloaded */
+		dd->cspec->hwerrmask &= ~HWE_MASK(PowerOnBISTFailed);
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	qib_format_hwerrors(hwerrs, qib_7220_hwerror_msgs,
+			    ARRAY_SIZE(qib_7220_hwerror_msgs), msg, msgl);
+
+	bitsmsg = dd->cspec->bitsmsgbuf;
+	if (hwerrs & (QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK <<
+		      QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT)) {
+		bits = (u32) ((hwerrs >>
+			       QLOGIC_IB_HWE_PCIEMEMPARITYERR_SHIFT) &
+			      QLOGIC_IB_HWE_PCIEMEMPARITYERR_MASK);
+		snprintf(bitsmsg, sizeof dd->cspec->bitsmsgbuf,
+			 "[PCIe Mem Parity Errs %x] ", bits);
+		strlcat(msg, bitsmsg, msgl);
+	}
+
+#define _QIB_PLL_FAIL (QLOGIC_IB_HWE_COREPLL_FBSLIP |   \
+			 QLOGIC_IB_HWE_COREPLL_RFSLIP)
+
+	if (hwerrs & _QIB_PLL_FAIL) {
+		isfatal = 1;
+		snprintf(bitsmsg, sizeof dd->cspec->bitsmsgbuf,
+			 "[PLL failed (%llx), InfiniPath hardware unusable]",
+			 (unsigned long long) hwerrs & _QIB_PLL_FAIL);
+		strlcat(msg, bitsmsg, msgl);
+		/* ignore from now on, so disable until driver reloaded */
+		dd->cspec->hwerrmask &= ~(hwerrs & _QIB_PLL_FAIL);
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	if (hwerrs & QLOGIC_IB_HWE_SERDESPLLFAILED) {
+		/*
+		 * If it occurs, it is left masked since the eternal
+		 * interface is unused.
+		 */
+		dd->cspec->hwerrmask &= ~QLOGIC_IB_HWE_SERDESPLLFAILED;
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	qib_dev_err(dd, "%s hardware error\n", msg);
+
+	if (isfatal && !dd->diag_client) {
+		qib_dev_err(dd,
+			"Fatal Hardware Error, no longer usable, SN %.16s\n",
+			dd->serial);
+		/*
+		 * For /sys status file and user programs to print; if no
+		 * trailing brace is copied, we'll know it was truncated.
+		 */
+		if (dd->freezemsg)
+			snprintf(dd->freezemsg, dd->freezelen,
+				 "{%s}", msg);
+		qib_disable_after_error(dd);
+	}
+bail:;
+}
+
+/**
+ * qib_7220_init_hwerrors - enable hardware errors
+ * @dd: the qlogic_ib device
+ *
+ * now that we have finished initializing everything that might reasonably
+ * cause a hardware error, and cleared those errors bits as they occur,
+ * we can enable hardware errors in the mask (potentially enabling
+ * freeze mode), and enable hardware errors as errors (along with
+ * everything else) in errormask
+ */
+static void qib_7220_init_hwerrors(struct qib_devdata *dd)
+{
+	u64 val;
+	u64 extsval;
+
+	extsval = qib_read_kreg64(dd, kr_extstatus);
+
+	if (!(extsval & (QLOGIC_IB_EXTS_MEMBIST_ENDTEST |
+			 QLOGIC_IB_EXTS_MEMBIST_DISABLED)))
+		qib_dev_err(dd, "MemBIST did not complete!\n");
+	if (extsval & QLOGIC_IB_EXTS_MEMBIST_DISABLED)
+		qib_devinfo(dd->pcidev, "MemBIST is disabled.\n");
+
+	val = ~0ULL;    /* default to all hwerrors become interrupts, */
+
+	val &= ~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR;
+	dd->cspec->hwerrmask = val;
+
+	qib_write_kreg(dd, kr_hwerrclear, ~HWE_MASK(PowerOnBISTFailed));
+	qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+
+	/* clear all */
+	qib_write_kreg(dd, kr_errclear, ~0ULL);
+	/* enable errors that are masked, at least this first time. */
+	qib_write_kreg(dd, kr_errmask, ~0ULL);
+	dd->cspec->errormask = qib_read_kreg64(dd, kr_errmask);
+	/* clear any interrupts up to this point (ints still not enabled) */
+	qib_write_kreg(dd, kr_intclear, ~0ULL);
+}
+
+/*
+ * Disable and enable the armlaunch error.  Used for PIO bandwidth testing
+ * on chips that are count-based, rather than trigger-based.  There is no
+ * reference counting, but that's also fine, given the intended use.
+ * Only chip-specific because it's all register accesses
+ */
+static void qib_set_7220_armlaunch(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		qib_write_kreg(dd, kr_errclear, ERR_MASK(SendPioArmLaunchErr));
+		dd->cspec->errormask |= ERR_MASK(SendPioArmLaunchErr);
+	} else
+		dd->cspec->errormask &= ~ERR_MASK(SendPioArmLaunchErr);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+}
+
+/*
+ * Formerly took parameter <which> in pre-shifted,
+ * pre-merged form with LinkCmd and LinkInitCmd
+ * together, and assuming the zero was NOP.
+ */
+static void qib_set_ib_7220_lstate(struct qib_pportdata *ppd, u16 linkcmd,
+				   u16 linitcmd)
+{
+	u64 mod_wd;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	if (linitcmd == QLOGIC_IB_IBCC_LINKINITCMD_DISABLE) {
+		/*
+		 * If we are told to disable, note that so link-recovery
+		 * code does not attempt to bring us back up.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	} else if (linitcmd || linkcmd == QLOGIC_IB_IBCC_LINKCMD_DOWN) {
+		/*
+		 * Any other linkinitcmd will lead to LINKDOWN and then
+		 * to INIT (if all is well), so clear flag to let
+		 * link-recovery code attempt to bring us back up.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+
+	mod_wd = (linkcmd << IBA7220_IBCC_LINKCMD_SHIFT) |
+		(linitcmd << QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+
+	qib_write_kreg(dd, kr_ibcctrl, ppd->cpspec->ibcctrl | mod_wd);
+	/* write to chip to prevent back-to-back writes of ibc reg */
+	qib_write_kreg(dd, kr_scratch, 0);
+}
+
+/*
+ * All detailed interaction with the SerDes has been moved to qib_sd7220.c
+ *
+ * The portion of IBA7220-specific bringup_serdes() that actually deals with
+ * registers and memory within the SerDes itself is qib_sd7220_init().
+ */
+
+/**
+ * qib_7220_bringup_serdes - bring up the serdes
+ * @ppd: physical port on the qlogic_ib device
+ */
+static int qib_7220_bringup_serdes(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 val, prev_val, guid, ibc;
+	int ret = 0;
+
+	/* Put IBC in reset, sends disabled */
+	dd->control &= ~QLOGIC_IB_C_LINKENABLE;
+	qib_write_kreg(dd, kr_control, 0ULL);
+
+	if (qib_compat_ddr_negotiate) {
+		ppd->cpspec->ibdeltainprog = 1;
+		ppd->cpspec->ibsymsnap = read_7220_creg32(dd, cr_ibsymbolerr);
+		ppd->cpspec->iblnkerrsnap =
+			read_7220_creg32(dd, cr_iblinkerrrecov);
+	}
+
+	/* flowcontrolwatermark is in units of KBytes */
+	ibc = 0x5ULL << SYM_LSB(IBCCtrl, FlowCtrlWaterMark);
+	/*
+	 * How often flowctrl sent.  More or less in usecs; balance against
+	 * watermark value, so that in theory senders always get a flow
+	 * control update in time to not let the IB link go idle.
+	 */
+	ibc |= 0x3ULL << SYM_LSB(IBCCtrl, FlowCtrlPeriod);
+	/* max error tolerance */
+	ibc |= 0xfULL << SYM_LSB(IBCCtrl, PhyerrThreshold);
+	/* use "real" buffer space for */
+	ibc |= 4ULL << SYM_LSB(IBCCtrl, CreditScale);
+	/* IB credit flow control. */
+	ibc |= 0xfULL << SYM_LSB(IBCCtrl, OverrunThreshold);
+	/*
+	 * set initial max size pkt IBC will send, including ICRC; it's the
+	 * PIO buffer size in dwords, less 1; also see qib_set_mtu()
+	 */
+	ibc |= ((u64)(ppd->ibmaxlen >> 2) + 1) << SYM_LSB(IBCCtrl, MaxPktLen);
+	ppd->cpspec->ibcctrl = ibc; /* without linkcmd or linkinitcmd! */
+
+	/* initially come up waiting for TS1, without sending anything. */
+	val = ppd->cpspec->ibcctrl | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE <<
+		QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+	qib_write_kreg(dd, kr_ibcctrl, val);
+
+	if (!ppd->cpspec->ibcddrctrl) {
+		/* not on re-init after reset */
+		ppd->cpspec->ibcddrctrl = qib_read_kreg64(dd, kr_ibcddrctrl);
+
+		if (ppd->link_speed_enabled == (QIB_IB_SDR | QIB_IB_DDR))
+			ppd->cpspec->ibcddrctrl |=
+				IBA7220_IBC_SPEED_AUTONEG_MASK |
+				IBA7220_IBC_IBTA_1_2_MASK;
+		else
+			ppd->cpspec->ibcddrctrl |=
+				ppd->link_speed_enabled == QIB_IB_DDR ?
+				IBA7220_IBC_SPEED_DDR : IBA7220_IBC_SPEED_SDR;
+		if ((ppd->link_width_enabled & (IB_WIDTH_1X | IB_WIDTH_4X)) ==
+		    (IB_WIDTH_1X | IB_WIDTH_4X))
+			ppd->cpspec->ibcddrctrl |= IBA7220_IBC_WIDTH_AUTONEG;
+		else
+			ppd->cpspec->ibcddrctrl |=
+				ppd->link_width_enabled == IB_WIDTH_4X ?
+				IBA7220_IBC_WIDTH_4X_ONLY :
+				IBA7220_IBC_WIDTH_1X_ONLY;
+
+		/* always enable these on driver reload, not sticky */
+		ppd->cpspec->ibcddrctrl |=
+			IBA7220_IBC_RXPOL_MASK << IBA7220_IBC_RXPOL_SHIFT;
+		ppd->cpspec->ibcddrctrl |=
+			IBA7220_IBC_HRTBT_MASK << IBA7220_IBC_HRTBT_SHIFT;
+
+		/* enable automatic lane reversal detection for receive */
+		ppd->cpspec->ibcddrctrl |= IBA7220_IBC_LANE_REV_SUPPORTED;
+	} else
+		/* write to chip to prevent back-to-back writes of ibc reg */
+		qib_write_kreg(dd, kr_scratch, 0);
+
+	qib_write_kreg(dd, kr_ibcddrctrl, ppd->cpspec->ibcddrctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	qib_write_kreg(dd, kr_ncmodectrl, 0Ull);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	ret = qib_sd7220_init(dd);
+
+	val = qib_read_kreg64(dd, kr_xgxs_cfg);
+	prev_val = val;
+	val |= QLOGIC_IB_XGXS_FC_SAFE;
+	if (val != prev_val) {
+		qib_write_kreg(dd, kr_xgxs_cfg, val);
+		qib_read_kreg32(dd, kr_scratch);
+	}
+	if (val & QLOGIC_IB_XGXS_RESET)
+		val &= ~QLOGIC_IB_XGXS_RESET;
+	if (val != prev_val)
+		qib_write_kreg(dd, kr_xgxs_cfg, val);
+
+	/* first time through, set port guid */
+	if (!ppd->guid)
+		ppd->guid = dd->base_guid;
+	guid = be64_to_cpu(ppd->guid);
+
+	qib_write_kreg(dd, kr_hrtbt_guid, guid);
+	if (!ret) {
+		dd->control |= QLOGIC_IB_C_LINKENABLE;
+		qib_write_kreg(dd, kr_control, dd->control);
+	} else
+		/* write to chip to prevent back-to-back writes of ibc reg */
+		qib_write_kreg(dd, kr_scratch, 0);
+	return ret;
+}
+
+/**
+ * qib_7220_quiet_serdes - set serdes to txidle
+ * @ppd: physical port of the qlogic_ib device
+ * Called when driver is being unloaded
+ */
+static void qib_7220_quiet_serdes(struct qib_pportdata *ppd)
+{
+	u64 val;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	/* disable IBC */
+	dd->control &= ~QLOGIC_IB_C_LINKENABLE;
+	qib_write_kreg(dd, kr_control,
+		       dd->control | QLOGIC_IB_C_FREEZEMODE);
+
+	ppd->cpspec->chase_end = 0;
+	if (ppd->cpspec->chase_timer.data) /* if initted */
+		del_timer_sync(&ppd->cpspec->chase_timer);
+
+	if (ppd->cpspec->ibsymdelta || ppd->cpspec->iblnkerrdelta ||
+	    ppd->cpspec->ibdeltainprog) {
+		u64 diagc;
+
+		/* enable counter writes */
+		diagc = qib_read_kreg64(dd, kr_hwdiagctrl);
+		qib_write_kreg(dd, kr_hwdiagctrl,
+			       diagc | SYM_MASK(HwDiagCtrl, CounterWrEnable));
+
+		if (ppd->cpspec->ibsymdelta || ppd->cpspec->ibdeltainprog) {
+			val = read_7220_creg32(dd, cr_ibsymbolerr);
+			if (ppd->cpspec->ibdeltainprog)
+				val -= val - ppd->cpspec->ibsymsnap;
+			val -= ppd->cpspec->ibsymdelta;
+			write_7220_creg(dd, cr_ibsymbolerr, val);
+		}
+		if (ppd->cpspec->iblnkerrdelta || ppd->cpspec->ibdeltainprog) {
+			val = read_7220_creg32(dd, cr_iblinkerrrecov);
+			if (ppd->cpspec->ibdeltainprog)
+				val -= val - ppd->cpspec->iblnkerrsnap;
+			val -= ppd->cpspec->iblnkerrdelta;
+			write_7220_creg(dd, cr_iblinkerrrecov, val);
+		}
+
+		/* and disable counter writes */
+		qib_write_kreg(dd, kr_hwdiagctrl, diagc);
+	}
+	qib_set_ib_7220_lstate(ppd, 0, QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags &= ~QIBL_IB_AUTONEG_INPROG;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	wake_up(&ppd->cpspec->autoneg_wait);
+	cancel_delayed_work_sync(&ppd->cpspec->autoneg_work);
+
+	shutdown_7220_relock_poll(ppd->dd);
+	val = qib_read_kreg64(ppd->dd, kr_xgxs_cfg);
+	val |= QLOGIC_IB_XGXS_RESET;
+	qib_write_kreg(ppd->dd, kr_xgxs_cfg, val);
+}
+
+/**
+ * qib_setup_7220_setextled - set the state of the two external LEDs
+ * @dd: the qlogic_ib device
+ * @on: whether the link is up or not
+ *
+ * The exact combo of LEDs if on is true is determined by looking
+ * at the ibcstatus.
+ *
+ * These LEDs indicate the physical and logical state of IB link.
+ * For this chip (at least with recommended board pinouts), LED1
+ * is Yellow (logical state) and LED2 is Green (physical state),
+ *
+ * Note:  We try to match the Mellanox HCA LED behavior as best
+ * we can.  Green indicates physical link state is OK (something is
+ * plugged in, and we can train).
+ * Amber indicates the link is logically up (ACTIVE).
+ * Mellanox further blinks the amber LED to indicate data packet
+ * activity, but we have no hardware support for that, so it would
+ * require waking up every 10-20 msecs and checking the counters
+ * on the chip, and then turning the LED off if appropriate.  That's
+ * visible overhead, so not something we will do.
+ *
+ */
+static void qib_setup_7220_setextled(struct qib_pportdata *ppd, u32 on)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 extctl, ledblink = 0, val, lst, ltst;
+	unsigned long flags;
+
+	/*
+	 * The diags use the LED to indicate diag info, so we leave
+	 * the external LED alone when the diags are running.
+	 */
+	if (dd->diag_client)
+		return;
+
+	if (ppd->led_override) {
+		ltst = (ppd->led_override & QIB_LED_PHYS) ?
+			IB_PHYSPORTSTATE_LINKUP : IB_PHYSPORTSTATE_DISABLED,
+		lst = (ppd->led_override & QIB_LED_LOG) ?
+			IB_PORT_ACTIVE : IB_PORT_DOWN;
+	} else if (on) {
+		val = qib_read_kreg64(dd, kr_ibcstatus);
+		ltst = qib_7220_phys_portstate(val);
+		lst = qib_7220_iblink_state(val);
+	} else {
+		ltst = 0;
+		lst = 0;
+	}
+
+	spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+	extctl = dd->cspec->extctrl & ~(SYM_MASK(EXTCtrl, LEDPriPortGreenOn) |
+				 SYM_MASK(EXTCtrl, LEDPriPortYellowOn));
+	if (ltst == IB_PHYSPORTSTATE_LINKUP) {
+		extctl |= SYM_MASK(EXTCtrl, LEDPriPortGreenOn);
+		/*
+		 * counts are in chip clock (4ns) periods.
+		 * This is 1/16 sec (66.6ms) on,
+		 * 3/16 sec (187.5 ms) off, with packets rcvd
+		 */
+		ledblink = ((66600 * 1000UL / 4) << IBA7220_LEDBLINK_ON_SHIFT)
+			| ((187500 * 1000UL / 4) << IBA7220_LEDBLINK_OFF_SHIFT);
+	}
+	if (lst == IB_PORT_ACTIVE)
+		extctl |= SYM_MASK(EXTCtrl, LEDPriPortYellowOn);
+	dd->cspec->extctrl = extctl;
+	qib_write_kreg(dd, kr_extctrl, extctl);
+	spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+
+	if (ledblink) /* blink the LED on packet receive */
+		qib_write_kreg(dd, kr_rcvpktledcnt, ledblink);
+}
+
+static void qib_7220_free_irq(struct qib_devdata *dd)
+{
+	if (dd->cspec->irq) {
+		free_irq(dd->cspec->irq, dd);
+		dd->cspec->irq = 0;
+	}
+	qib_nomsi(dd);
+}
+
+/*
+ * qib_setup_7220_cleanup - clean up any per-chip chip-specific stuff
+ * @dd: the qlogic_ib device
+ *
+ * This is called during driver unload.
+ *
+ */
+static void qib_setup_7220_cleanup(struct qib_devdata *dd)
+{
+	qib_7220_free_irq(dd);
+	kfree(dd->cspec->cntrs);
+	kfree(dd->cspec->portcntrs);
+}
+
+/*
+ * This is only called for SDmaInt.
+ * SDmaDisabled is handled on the error path.
+ */
+static void sdma_7220_intr(struct qib_pportdata *ppd, u64 istat)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	switch (ppd->sdma_state.current_state) {
+	case qib_sdma_state_s00_hw_down:
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		__qib_sdma_process_event(ppd, qib_sdma_event_e20_hw_started);
+		break;
+
+	case qib_sdma_state_s20_idle:
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		__qib_sdma_process_event(ppd, qib_sdma_event_e60_hw_halted);
+		break;
+
+	case qib_sdma_state_s99_running:
+		/* too chatty to print here */
+		__qib_sdma_intr(ppd);
+		break;
+	}
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+static void qib_wantpiobuf_7220_intr(struct qib_devdata *dd, u32 needint)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+	if (needint) {
+		if (!(dd->sendctrl & SYM_MASK(SendCtrl, SendBufAvailUpd)))
+			goto done;
+		/*
+		 * blip the availupd off, next write will be on, so
+		 * we ensure an avail update, regardless of threshold or
+		 * buffers becoming free, whenever we want an interrupt
+		 */
+		qib_write_kreg(dd, kr_sendctrl, dd->sendctrl &
+			~SYM_MASK(SendCtrl, SendBufAvailUpd));
+		qib_write_kreg(dd, kr_scratch, 0ULL);
+		dd->sendctrl |= SYM_MASK(SendCtrl, SendIntBufAvail);
+	} else
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, SendIntBufAvail);
+	qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+done:
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+}
+
+/*
+ * Handle errors and unusual events first, separate function
+ * to improve cache hits for fast path interrupt handling.
+ */
+static noinline void unlikely_7220_intr(struct qib_devdata *dd, u64 istat)
+{
+	if (unlikely(istat & ~QLOGIC_IB_I_BITSEXTANT))
+		qib_dev_err(dd,
+			    "interrupt with unknown interrupts %Lx set\n",
+			    istat & ~QLOGIC_IB_I_BITSEXTANT);
+
+	if (istat & QLOGIC_IB_I_GPIO) {
+		u32 gpiostatus;
+
+		/*
+		 * Boards for this chip currently don't use GPIO interrupts,
+		 * so clear by writing GPIOstatus to GPIOclear, and complain
+		 * to alert developer. To avoid endless repeats, clear
+		 * the bits in the mask, since there is some kind of
+		 * programming error or chip problem.
+		 */
+		gpiostatus = qib_read_kreg32(dd, kr_gpio_status);
+		/*
+		 * In theory, writing GPIOstatus to GPIOclear could
+		 * have a bad side-effect on some diagnostic that wanted
+		 * to poll for a status-change, but the various shadows
+		 * make that problematic at best. Diags will just suppress
+		 * all GPIO interrupts during such tests.
+		 */
+		qib_write_kreg(dd, kr_gpio_clear, gpiostatus);
+
+		if (gpiostatus) {
+			const u32 mask = qib_read_kreg32(dd, kr_gpio_mask);
+			u32 gpio_irq = mask & gpiostatus;
+
+			/*
+			 * A bit set in status and (chip) Mask register
+			 * would cause an interrupt. Since we are not
+			 * expecting any, report it. Also check that the
+			 * chip reflects our shadow, report issues,
+			 * and refresh from the shadow.
+			 */
+			/*
+			 * Clear any troublemakers, and update chip
+			 * from shadow
+			 */
+			dd->cspec->gpio_mask &= ~gpio_irq;
+			qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+		}
+	}
+
+	if (istat & QLOGIC_IB_I_ERROR) {
+		u64 estat;
+
+		qib_stats.sps_errints++;
+		estat = qib_read_kreg64(dd, kr_errstatus);
+		if (!estat)
+			qib_devinfo(dd->pcidev,
+				"error interrupt (%Lx), but no error bits set!\n",
+				istat);
+		else
+			handle_7220_errors(dd, estat);
+	}
+}
+
+static irqreturn_t qib_7220intr(int irq, void *data)
+{
+	struct qib_devdata *dd = data;
+	irqreturn_t ret;
+	u64 istat;
+	u64 ctxtrbits;
+	u64 rmask;
+	unsigned i;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT) {
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		ret = IRQ_HANDLED;
+		goto bail;
+	}
+
+	istat = qib_read_kreg64(dd, kr_intstatus);
+
+	if (unlikely(!istat)) {
+		ret = IRQ_NONE; /* not our interrupt, or already handled */
+		goto bail;
+	}
+	if (unlikely(istat == -1)) {
+		qib_bad_intrstatus(dd);
+		/* don't know if it was our interrupt or not */
+		ret = IRQ_NONE;
+		goto bail;
+	}
+
+	this_cpu_inc(*dd->int_counter);
+	if (unlikely(istat & (~QLOGIC_IB_I_BITSEXTANT |
+			      QLOGIC_IB_I_GPIO | QLOGIC_IB_I_ERROR)))
+		unlikely_7220_intr(dd, istat);
+
+	/*
+	 * Clear the interrupt bits we found set, relatively early, so we
+	 * "know" know the chip will have seen this by the time we process
+	 * the queue, and will re-interrupt if necessary.  The processor
+	 * itself won't take the interrupt again until we return.
+	 */
+	qib_write_kreg(dd, kr_intclear, istat);
+
+	/*
+	 * Handle kernel receive queues before checking for pio buffers
+	 * available since receives can overflow; piobuf waiters can afford
+	 * a few extra cycles, since they were waiting anyway.
+	 */
+	ctxtrbits = istat &
+		((QLOGIC_IB_I_RCVAVAIL_MASK << QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+		 (QLOGIC_IB_I_RCVURG_MASK << QLOGIC_IB_I_RCVURG_SHIFT));
+	if (ctxtrbits) {
+		rmask = (1ULL << QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+			(1ULL << QLOGIC_IB_I_RCVURG_SHIFT);
+		for (i = 0; i < dd->first_user_ctxt; i++) {
+			if (ctxtrbits & rmask) {
+				ctxtrbits &= ~rmask;
+				qib_kreceive(dd->rcd[i], NULL, NULL);
+			}
+			rmask <<= 1;
+		}
+		if (ctxtrbits) {
+			ctxtrbits =
+				(ctxtrbits >> QLOGIC_IB_I_RCVAVAIL_SHIFT) |
+				(ctxtrbits >> QLOGIC_IB_I_RCVURG_SHIFT);
+			qib_handle_urcv(dd, ctxtrbits);
+		}
+	}
+
+	/* only call for SDmaInt */
+	if (istat & QLOGIC_IB_I_SDMAINT)
+		sdma_7220_intr(dd->pport, istat);
+
+	if ((istat & QLOGIC_IB_I_SPIOBUFAVAIL) && (dd->flags & QIB_INITTED))
+		qib_ib_piobufavail(dd);
+
+	ret = IRQ_HANDLED;
+bail:
+	return ret;
+}
+
+/*
+ * Set up our chip-specific interrupt handler.
+ * The interrupt type has already been setup, so
+ * we just need to do the registration and error checking.
+ * If we are using MSI interrupts, we may fall back to
+ * INTx later, if the interrupt handler doesn't get called
+ * within 1/2 second (see verify_interrupt()).
+ */
+static void qib_setup_7220_interrupt(struct qib_devdata *dd)
+{
+	if (!dd->cspec->irq)
+		qib_dev_err(dd,
+			"irq is 0, BIOS error?  Interrupts won't work\n");
+	else {
+		int ret = request_irq(dd->cspec->irq, qib_7220intr,
+			dd->msi_lo ? 0 : IRQF_SHARED,
+			QIB_DRV_NAME, dd);
+
+		if (ret)
+			qib_dev_err(dd,
+				"Couldn't setup %s interrupt (irq=%d): %d\n",
+				dd->msi_lo ?  "MSI" : "INTx",
+				dd->cspec->irq, ret);
+	}
+}
+
+/**
+ * qib_7220_boardname - fill in the board name
+ * @dd: the qlogic_ib device
+ *
+ * info is based on the board revision register
+ */
+static void qib_7220_boardname(struct qib_devdata *dd)
+{
+	char *n;
+	u32 boardid, namelen;
+
+	boardid = SYM_FIELD(dd->revision, Revision,
+			    BoardID);
+
+	switch (boardid) {
+	case 1:
+		n = "InfiniPath_QLE7240";
+		break;
+	case 2:
+		n = "InfiniPath_QLE7280";
+		break;
+	default:
+		qib_dev_err(dd, "Unknown 7220 board with ID %u\n", boardid);
+		n = "Unknown_InfiniPath_7220";
+		break;
+	}
+
+	namelen = strlen(n) + 1;
+	dd->boardname = kmalloc(namelen, GFP_KERNEL);
+	if (!dd->boardname)
+		qib_dev_err(dd, "Failed allocation for board name: %s\n", n);
+	else
+		snprintf(dd->boardname, namelen, "%s", n);
+
+	if (dd->majrev != 5 || !dd->minrev || dd->minrev > 2)
+		qib_dev_err(dd,
+			"Unsupported InfiniPath hardware revision %u.%u!\n",
+			dd->majrev, dd->minrev);
+
+	snprintf(dd->boardversion, sizeof(dd->boardversion),
+		 "ChipABI %u.%u, %s, InfiniPath%u %u.%u, SW Compat %u\n",
+		 QIB_CHIP_VERS_MAJ, QIB_CHIP_VERS_MIN, dd->boardname,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, Arch),
+		 dd->majrev, dd->minrev,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, SW));
+}
+
+/*
+ * This routine sleeps, so it can only be called from user context, not
+ * from interrupt context.
+ */
+static int qib_setup_7220_reset(struct qib_devdata *dd)
+{
+	u64 val;
+	int i;
+	int ret;
+	u16 cmdval;
+	u8 int_line, clinesz;
+	unsigned long flags;
+
+	qib_pcie_getcmd(dd, &cmdval, &int_line, &clinesz);
+
+	/* Use dev_err so it shows up in logs, etc. */
+	qib_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->unit);
+
+	/* no interrupts till re-initted */
+	qib_7220_set_intr_state(dd, 0);
+
+	dd->pport->cpspec->ibdeltainprog = 0;
+	dd->pport->cpspec->ibsymdelta = 0;
+	dd->pport->cpspec->iblnkerrdelta = 0;
+
+	/*
+	 * Keep chip from being accessed until we are ready.  Use
+	 * writeq() directly, to allow the write even though QIB_PRESENT
+	 * isn't set.
+	 */
+	dd->flags &= ~(QIB_INITTED | QIB_PRESENT);
+	/* so we check interrupts work again */
+	dd->z_int_counter = qib_int_counter(dd);
+	val = dd->control | QLOGIC_IB_C_RESET;
+	writeq(val, &dd->kregbase[kr_control]);
+	mb(); /* prevent compiler reordering around actual reset */
+
+	for (i = 1; i <= 5; i++) {
+		/*
+		 * Allow MBIST, etc. to complete; longer on each retry.
+		 * We sometimes get machine checks from bus timeout if no
+		 * response, so for now, make it *really* long.
+		 */
+		msleep(1000 + (1 + i) * 2000);
+
+		qib_pcie_reenable(dd, cmdval, int_line, clinesz);
+
+		/*
+		 * Use readq directly, so we don't need to mark it as PRESENT
+		 * until we get a successful indication that all is well.
+		 */
+		val = readq(&dd->kregbase[kr_revision]);
+		if (val == dd->revision) {
+			dd->flags |= QIB_PRESENT; /* it's back */
+			ret = qib_reinit_intr(dd);
+			goto bail;
+		}
+	}
+	ret = 0; /* failed */
+
+bail:
+	if (ret) {
+		if (qib_pcie_params(dd, dd->lbus_width, NULL, NULL))
+			qib_dev_err(dd,
+				"Reset failed to setup PCIe or interrupts; continuing anyway\n");
+
+		/* hold IBC in reset, no sends, etc till later */
+		qib_write_kreg(dd, kr_control, 0ULL);
+
+		/* clear the reset error, init error/hwerror mask */
+		qib_7220_init_hwerrors(dd);
+
+		/* do setup similar to speed or link-width changes */
+		if (dd->pport->cpspec->ibcddrctrl & IBA7220_IBC_IBTA_1_2_MASK)
+			dd->cspec->presets_needed = 1;
+		spin_lock_irqsave(&dd->pport->lflags_lock, flags);
+		dd->pport->lflags |= QIBL_IB_FORCE_NOTIFY;
+		dd->pport->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+		spin_unlock_irqrestore(&dd->pport->lflags_lock, flags);
+	}
+
+	return ret;
+}
+
+/**
+ * qib_7220_put_tid - write a TID to the chip
+ * @dd: the qlogic_ib device
+ * @tidptr: pointer to the expected TID (in chip) to update
+ * @tidtype: 0 for eager, 1 for expected
+ * @pa: physical address of in memory buffer; tidinvalid if freeing
+ */
+static void qib_7220_put_tid(struct qib_devdata *dd, u64 __iomem *tidptr,
+			     u32 type, unsigned long pa)
+{
+	if (pa != dd->tidinvalid) {
+		u64 chippa = pa >> IBA7220_TID_PA_SHIFT;
+
+		/* paranoia checks */
+		if (pa != (chippa << IBA7220_TID_PA_SHIFT)) {
+			qib_dev_err(dd, "Physaddr %lx not 2KB aligned!\n",
+				    pa);
+			return;
+		}
+		if (chippa >= (1UL << IBA7220_TID_SZ_SHIFT)) {
+			qib_dev_err(dd,
+				"Physical page address 0x%lx larger than supported\n",
+				pa);
+			return;
+		}
+
+		if (type == RCVHQ_RCV_TYPE_EAGER)
+			chippa |= dd->tidtemplate;
+		else /* for now, always full 4KB page */
+			chippa |= IBA7220_TID_SZ_4K;
+		pa = chippa;
+	}
+	writeq(pa, tidptr);
+	mmiowb();
+}
+
+/**
+ * qib_7220_clear_tids - clear all TID entries for a ctxt, expected and eager
+ * @dd: the qlogic_ib device
+ * @ctxt: the ctxt
+ *
+ * clear all TID entries for a ctxt, expected and eager.
+ * Used from qib_close().  On this chip, TIDs are only 32 bits,
+ * not 64, but they are still on 64 bit boundaries, so tidbase
+ * is declared as u64 * for the pointer math, even though we write 32 bits
+ */
+static void qib_7220_clear_tids(struct qib_devdata *dd,
+				struct qib_ctxtdata *rcd)
+{
+	u64 __iomem *tidbase;
+	unsigned long tidinv;
+	u32 ctxt;
+	int i;
+
+	if (!dd->kregbase || !rcd)
+		return;
+
+	ctxt = rcd->ctxt;
+
+	tidinv = dd->tidinvalid;
+	tidbase = (u64 __iomem *)
+		((char __iomem *)(dd->kregbase) +
+		 dd->rcvtidbase +
+		 ctxt * dd->rcvtidcnt * sizeof(*tidbase));
+
+	for (i = 0; i < dd->rcvtidcnt; i++)
+		qib_7220_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
+				 tidinv);
+
+	tidbase = (u64 __iomem *)
+		((char __iomem *)(dd->kregbase) +
+		 dd->rcvegrbase +
+		 rcd->rcvegr_tid_base * sizeof(*tidbase));
+
+	for (i = 0; i < rcd->rcvegrcnt; i++)
+		qib_7220_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
+				 tidinv);
+}
+
+/**
+ * qib_7220_tidtemplate - setup constants for TID updates
+ * @dd: the qlogic_ib device
+ *
+ * We setup stuff that we use a lot, to avoid calculating each time
+ */
+static void qib_7220_tidtemplate(struct qib_devdata *dd)
+{
+	if (dd->rcvegrbufsize == 2048)
+		dd->tidtemplate = IBA7220_TID_SZ_2K;
+	else if (dd->rcvegrbufsize == 4096)
+		dd->tidtemplate = IBA7220_TID_SZ_4K;
+	dd->tidinvalid = 0;
+}
+
+/**
+ * qib_init_7220_get_base_info - set chip-specific flags for user code
+ * @rcd: the qlogic_ib ctxt
+ * @kbase: qib_base_info pointer
+ *
+ * We set the PCIE flag because the lower bandwidth on PCIe vs
+ * HyperTransport can affect some user packet algorithims.
+ */
+static int qib_7220_get_base_info(struct qib_ctxtdata *rcd,
+				  struct qib_base_info *kinfo)
+{
+	kinfo->spi_runtime_flags |= QIB_RUNTIME_PCIE |
+		QIB_RUNTIME_NODMA_RTAIL | QIB_RUNTIME_SDMA;
+
+	if (rcd->dd->flags & QIB_USE_SPCL_TRIG)
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_SPECIAL_TRIGGER;
+
+	return 0;
+}
+
+static struct qib_message_header *
+qib_7220_get_msgheader(struct qib_devdata *dd, __le32 *rhf_addr)
+{
+	u32 offset = qib_hdrget_offset(rhf_addr);
+
+	return (struct qib_message_header *)
+		(rhf_addr - dd->rhf_offset + offset);
+}
+
+static void qib_7220_config_ctxts(struct qib_devdata *dd)
+{
+	unsigned long flags;
+	u32 nchipctxts;
+
+	nchipctxts = qib_read_kreg32(dd, kr_portcnt);
+	dd->cspec->numctxts = nchipctxts;
+	if (qib_n_krcv_queues > 1) {
+		dd->qpn_mask = 0x3e;
+		dd->first_user_ctxt = qib_n_krcv_queues * dd->num_pports;
+		if (dd->first_user_ctxt > nchipctxts)
+			dd->first_user_ctxt = nchipctxts;
+	} else
+		dd->first_user_ctxt = dd->num_pports;
+	dd->n_krcv_queues = dd->first_user_ctxt;
+
+	if (!qib_cfgctxts) {
+		int nctxts = dd->first_user_ctxt + num_online_cpus();
+
+		if (nctxts <= 5)
+			dd->ctxtcnt = 5;
+		else if (nctxts <= 9)
+			dd->ctxtcnt = 9;
+		else if (nctxts <= nchipctxts)
+			dd->ctxtcnt = nchipctxts;
+	} else if (qib_cfgctxts <= nchipctxts)
+		dd->ctxtcnt = qib_cfgctxts;
+	if (!dd->ctxtcnt) /* none of the above, set to max */
+		dd->ctxtcnt = nchipctxts;
+
+	/*
+	 * Chip can be configured for 5, 9, or 17 ctxts, and choice
+	 * affects number of eager TIDs per ctxt (1K, 2K, 4K).
+	 * Lock to be paranoid about later motion, etc.
+	 */
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+	if (dd->ctxtcnt > 9)
+		dd->rcvctrl |= 2ULL << IBA7220_R_CTXTCFG_SHIFT;
+	else if (dd->ctxtcnt > 5)
+		dd->rcvctrl |= 1ULL << IBA7220_R_CTXTCFG_SHIFT;
+	/* else configure for default 5 receive ctxts */
+	if (dd->qpn_mask)
+		dd->rcvctrl |= 1ULL << QIB_7220_RcvCtrl_RcvQPMapEnable_LSB;
+	qib_write_kreg(dd, kr_rcvctrl, dd->rcvctrl);
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+
+	/* kr_rcvegrcnt changes based on the number of contexts enabled */
+	dd->cspec->rcvegrcnt = qib_read_kreg32(dd, kr_rcvegrcnt);
+	dd->rcvhdrcnt = max(dd->cspec->rcvegrcnt, IBA7220_KRCVEGRCNT);
+}
+
+static int qib_7220_get_ib_cfg(struct qib_pportdata *ppd, int which)
+{
+	int lsb, ret = 0;
+	u64 maskr; /* right-justified mask */
+
+	switch (which) {
+	case QIB_IB_CFG_LWID_ENB: /* Get allowed Link-width */
+		ret = ppd->link_width_enabled;
+		goto done;
+
+	case QIB_IB_CFG_LWID: /* Get currently active Link-width */
+		ret = ppd->link_width_active;
+		goto done;
+
+	case QIB_IB_CFG_SPD_ENB: /* Get allowed Link speeds */
+		ret = ppd->link_speed_enabled;
+		goto done;
+
+	case QIB_IB_CFG_SPD: /* Get current Link spd */
+		ret = ppd->link_speed_active;
+		goto done;
+
+	case QIB_IB_CFG_RXPOL_ENB: /* Get Auto-RX-polarity enable */
+		lsb = IBA7220_IBC_RXPOL_SHIFT;
+		maskr = IBA7220_IBC_RXPOL_MASK;
+		break;
+
+	case QIB_IB_CFG_LREV_ENB: /* Get Auto-Lane-reversal enable */
+		lsb = IBA7220_IBC_LREV_SHIFT;
+		maskr = IBA7220_IBC_LREV_MASK;
+		break;
+
+	case QIB_IB_CFG_LINKLATENCY:
+		ret = qib_read_kreg64(ppd->dd, kr_ibcddrstatus)
+			& IBA7220_DDRSTAT_LINKLAT_MASK;
+		goto done;
+
+	case QIB_IB_CFG_OP_VLS:
+		ret = ppd->vls_operational;
+		goto done;
+
+	case QIB_IB_CFG_VL_HIGH_CAP:
+		ret = 0;
+		goto done;
+
+	case QIB_IB_CFG_VL_LOW_CAP:
+		ret = 0;
+		goto done;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		ret = SYM_FIELD(ppd->cpspec->ibcctrl, IBCCtrl,
+				OverrunThreshold);
+		goto done;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		ret = SYM_FIELD(ppd->cpspec->ibcctrl, IBCCtrl,
+				PhyerrThreshold);
+		goto done;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		ret = (ppd->cpspec->ibcctrl &
+		       SYM_MASK(IBCCtrl, LinkDownDefaultState)) ?
+			IB_LINKINITCMD_SLEEP : IB_LINKINITCMD_POLL;
+		goto done;
+
+	case QIB_IB_CFG_HRTBT: /* Get Heartbeat off/enable/auto */
+		lsb = IBA7220_IBC_HRTBT_SHIFT;
+		maskr = IBA7220_IBC_HRTBT_MASK;
+		break;
+
+	case QIB_IB_CFG_PMA_TICKS:
+		/*
+		 * 0x00 = 10x link transfer rate or 4 nsec. for 2.5Gbs
+		 * Since the clock is always 250MHz, the value is 1 or 0.
+		 */
+		ret = (ppd->link_speed_active == QIB_IB_DDR);
+		goto done;
+
+	default:
+		ret = -EINVAL;
+		goto done;
+	}
+	ret = (int)((ppd->cpspec->ibcddrctrl >> lsb) & maskr);
+done:
+	return ret;
+}
+
+static int qib_7220_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 maskr; /* right-justified mask */
+	int lsb, ret = 0, setforce = 0;
+	u16 lcmd, licmd;
+	unsigned long flags;
+	u32 tmp = 0;
+
+	switch (which) {
+	case QIB_IB_CFG_LIDLMC:
+		/*
+		 * Set LID and LMC. Combined to avoid possible hazard
+		 * caller puts LMC in 16MSbits, DLID in 16LSbits of val
+		 */
+		lsb = IBA7220_IBC_DLIDLMC_SHIFT;
+		maskr = IBA7220_IBC_DLIDLMC_MASK;
+		break;
+
+	case QIB_IB_CFG_LWID_ENB: /* set allowed Link-width */
+		/*
+		 * As with speed, only write the actual register if
+		 * the link is currently down, otherwise takes effect
+		 * on next link change.
+		 */
+		ppd->link_width_enabled = val;
+		if (!(ppd->lflags & QIBL_LINKDOWN))
+			goto bail;
+		/*
+		 * We set the QIBL_IB_FORCE_NOTIFY bit so updown
+		 * will get called because we want update
+		 * link_width_active, and the change may not take
+		 * effect for some time (if we are in POLL), so this
+		 * flag will force the updown routine to be called
+		 * on the next ibstatuschange down interrupt, even
+		 * if it's not an down->up transition.
+		 */
+		val--; /* convert from IB to chip */
+		maskr = IBA7220_IBC_WIDTH_MASK;
+		lsb = IBA7220_IBC_WIDTH_SHIFT;
+		setforce = 1;
+		break;
+
+	case QIB_IB_CFG_SPD_ENB: /* set allowed Link speeds */
+		/*
+		 * If we turn off IB1.2, need to preset SerDes defaults,
+		 * but not right now. Set a flag for the next time
+		 * we command the link down.  As with width, only write the
+		 * actual register if the link is currently down, otherwise
+		 * takes effect on next link change.  Since setting is being
+		 * explicitly requested (via MAD or sysfs), clear autoneg
+		 * failure status if speed autoneg is enabled.
+		 */
+		ppd->link_speed_enabled = val;
+		if ((ppd->cpspec->ibcddrctrl & IBA7220_IBC_IBTA_1_2_MASK) &&
+		    !(val & (val - 1)))
+			dd->cspec->presets_needed = 1;
+		if (!(ppd->lflags & QIBL_LINKDOWN))
+			goto bail;
+		/*
+		 * We set the QIBL_IB_FORCE_NOTIFY bit so updown
+		 * will get called because we want update
+		 * link_speed_active, and the change may not take
+		 * effect for some time (if we are in POLL), so this
+		 * flag will force the updown routine to be called
+		 * on the next ibstatuschange down interrupt, even
+		 * if it's not an down->up transition.
+		 */
+		if (val == (QIB_IB_SDR | QIB_IB_DDR)) {
+			val = IBA7220_IBC_SPEED_AUTONEG_MASK |
+				IBA7220_IBC_IBTA_1_2_MASK;
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		} else
+			val = val == QIB_IB_DDR ?
+				IBA7220_IBC_SPEED_DDR : IBA7220_IBC_SPEED_SDR;
+		maskr = IBA7220_IBC_SPEED_AUTONEG_MASK |
+			IBA7220_IBC_IBTA_1_2_MASK;
+		/* IBTA 1.2 mode + speed bits are contiguous */
+		lsb = SYM_LSB(IBCDDRCtrl, IB_ENHANCED_MODE);
+		setforce = 1;
+		break;
+
+	case QIB_IB_CFG_RXPOL_ENB: /* set Auto-RX-polarity enable */
+		lsb = IBA7220_IBC_RXPOL_SHIFT;
+		maskr = IBA7220_IBC_RXPOL_MASK;
+		break;
+
+	case QIB_IB_CFG_LREV_ENB: /* set Auto-Lane-reversal enable */
+		lsb = IBA7220_IBC_LREV_SHIFT;
+		maskr = IBA7220_IBC_LREV_MASK;
+		break;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		maskr = SYM_FIELD(ppd->cpspec->ibcctrl, IBCCtrl,
+				  OverrunThreshold);
+		if (maskr != val) {
+			ppd->cpspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, OverrunThreshold);
+			ppd->cpspec->ibcctrl |= (u64) val <<
+				SYM_LSB(IBCCtrl, OverrunThreshold);
+			qib_write_kreg(dd, kr_ibcctrl, ppd->cpspec->ibcctrl);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		maskr = SYM_FIELD(ppd->cpspec->ibcctrl, IBCCtrl,
+				  PhyerrThreshold);
+		if (maskr != val) {
+			ppd->cpspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, PhyerrThreshold);
+			ppd->cpspec->ibcctrl |= (u64) val <<
+				SYM_LSB(IBCCtrl, PhyerrThreshold);
+			qib_write_kreg(dd, kr_ibcctrl, ppd->cpspec->ibcctrl);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_PKEYS: /* update pkeys */
+		maskr = (u64) ppd->pkeys[0] | ((u64) ppd->pkeys[1] << 16) |
+			((u64) ppd->pkeys[2] << 32) |
+			((u64) ppd->pkeys[3] << 48);
+		qib_write_kreg(dd, kr_partitionkey, maskr);
+		goto bail;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		if (val == IB_LINKINITCMD_POLL)
+			ppd->cpspec->ibcctrl &=
+				~SYM_MASK(IBCCtrl, LinkDownDefaultState);
+		else /* SLEEP */
+			ppd->cpspec->ibcctrl |=
+				SYM_MASK(IBCCtrl, LinkDownDefaultState);
+		qib_write_kreg(dd, kr_ibcctrl, ppd->cpspec->ibcctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+		goto bail;
+
+	case QIB_IB_CFG_MTU: /* update the MTU in IBC */
+		/*
+		 * Update our housekeeping variables, and set IBC max
+		 * size, same as init code; max IBC is max we allow in
+		 * buffer, less the qword pbc, plus 1 for ICRC, in dwords
+		 * Set even if it's unchanged, print debug message only
+		 * on changes.
+		 */
+		val = (ppd->ibmaxlen >> 2) + 1;
+		ppd->cpspec->ibcctrl &= ~SYM_MASK(IBCCtrl, MaxPktLen);
+		ppd->cpspec->ibcctrl |= (u64)val << SYM_LSB(IBCCtrl, MaxPktLen);
+		qib_write_kreg(dd, kr_ibcctrl, ppd->cpspec->ibcctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+		goto bail;
+
+	case QIB_IB_CFG_LSTATE: /* set the IB link state */
+		switch (val & 0xffff0000) {
+		case IB_LINKCMD_DOWN:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_DOWN;
+			if (!ppd->cpspec->ibdeltainprog &&
+			    qib_compat_ddr_negotiate) {
+				ppd->cpspec->ibdeltainprog = 1;
+				ppd->cpspec->ibsymsnap =
+					read_7220_creg32(dd, cr_ibsymbolerr);
+				ppd->cpspec->iblnkerrsnap =
+					read_7220_creg32(dd, cr_iblinkerrrecov);
+			}
+			break;
+
+		case IB_LINKCMD_ARMED:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ARMED;
+			break;
+
+		case IB_LINKCMD_ACTIVE:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ACTIVE;
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkcmd req 0x%x\n", val >> 16);
+			goto bail;
+		}
+		switch (val & 0xffff) {
+		case IB_LINKINITCMD_NOP:
+			licmd = 0;
+			break;
+
+		case IB_LINKINITCMD_POLL:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_POLL;
+			break;
+
+		case IB_LINKINITCMD_SLEEP:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_SLEEP;
+			break;
+
+		case IB_LINKINITCMD_DISABLE:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_DISABLE;
+			ppd->cpspec->chase_end = 0;
+			/*
+			 * stop state chase counter and timer, if running.
+			 * wait forpending timer, but don't clear .data (ppd)!
+			 */
+			if (ppd->cpspec->chase_timer.expires) {
+				del_timer_sync(&ppd->cpspec->chase_timer);
+				ppd->cpspec->chase_timer.expires = 0;
+			}
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkinitcmd req 0x%x\n",
+				    val & 0xffff);
+			goto bail;
+		}
+		qib_set_ib_7220_lstate(ppd, lcmd, licmd);
+
+		maskr = IBA7220_IBC_WIDTH_MASK;
+		lsb = IBA7220_IBC_WIDTH_SHIFT;
+		tmp = (ppd->cpspec->ibcddrctrl >> lsb) & maskr;
+		/* If the width active on the chip does not match the
+		 * width in the shadow register, write the new active
+		 * width to the chip.
+		 * We don't have to worry about speed as the speed is taken
+		 * care of by set_7220_ibspeed_fast called by ib_updown.
+		 */
+		if (ppd->link_width_enabled-1 != tmp) {
+			ppd->cpspec->ibcddrctrl &= ~(maskr << lsb);
+			ppd->cpspec->ibcddrctrl |=
+				(((u64)(ppd->link_width_enabled-1) & maskr) <<
+				 lsb);
+			qib_write_kreg(dd, kr_ibcddrctrl,
+				       ppd->cpspec->ibcddrctrl);
+			qib_write_kreg(dd, kr_scratch, 0);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_IB_FORCE_NOTIFY;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_HRTBT: /* set Heartbeat off/enable/auto */
+		if (val > IBA7220_IBC_HRTBT_MASK) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		lsb = IBA7220_IBC_HRTBT_SHIFT;
+		maskr = IBA7220_IBC_HRTBT_MASK;
+		break;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+	ppd->cpspec->ibcddrctrl &= ~(maskr << lsb);
+	ppd->cpspec->ibcddrctrl |= (((u64) val & maskr) << lsb);
+	qib_write_kreg(dd, kr_ibcddrctrl, ppd->cpspec->ibcddrctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+	if (setforce) {
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_FORCE_NOTIFY;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+bail:
+	return ret;
+}
+
+static int qib_7220_set_loopback(struct qib_pportdata *ppd, const char *what)
+{
+	int ret = 0;
+	u64 val, ddr;
+
+	if (!strncmp(what, "ibc", 3)) {
+		ppd->cpspec->ibcctrl |= SYM_MASK(IBCCtrl, Loopback);
+		val = 0; /* disable heart beat, so link will come up */
+		qib_devinfo(ppd->dd->pcidev, "Enabling IB%u:%u IBC loopback\n",
+			 ppd->dd->unit, ppd->port);
+	} else if (!strncmp(what, "off", 3)) {
+		ppd->cpspec->ibcctrl &= ~SYM_MASK(IBCCtrl, Loopback);
+		/* enable heart beat again */
+		val = IBA7220_IBC_HRTBT_MASK << IBA7220_IBC_HRTBT_SHIFT;
+		qib_devinfo(ppd->dd->pcidev,
+			"Disabling IB%u:%u IBC loopback (normal)\n",
+			ppd->dd->unit, ppd->port);
+	} else
+		ret = -EINVAL;
+	if (!ret) {
+		qib_write_kreg(ppd->dd, kr_ibcctrl, ppd->cpspec->ibcctrl);
+		ddr = ppd->cpspec->ibcddrctrl & ~(IBA7220_IBC_HRTBT_MASK
+					     << IBA7220_IBC_HRTBT_SHIFT);
+		ppd->cpspec->ibcddrctrl = ddr | val;
+		qib_write_kreg(ppd->dd, kr_ibcddrctrl,
+			       ppd->cpspec->ibcddrctrl);
+		qib_write_kreg(ppd->dd, kr_scratch, 0);
+	}
+	return ret;
+}
+
+static void qib_update_7220_usrhead(struct qib_ctxtdata *rcd, u64 hd,
+				    u32 updegr, u32 egrhd, u32 npkts)
+{
+	if (updegr)
+		qib_write_ureg(rcd->dd, ur_rcvegrindexhead, egrhd, rcd->ctxt);
+	mmiowb();
+	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
+	mmiowb();
+}
+
+static u32 qib_7220_hdrqempty(struct qib_ctxtdata *rcd)
+{
+	u32 head, tail;
+
+	head = qib_read_ureg32(rcd->dd, ur_rcvhdrhead, rcd->ctxt);
+	if (rcd->rcvhdrtail_kvaddr)
+		tail = qib_get_rcvhdrtail(rcd);
+	else
+		tail = qib_read_ureg32(rcd->dd, ur_rcvhdrtail, rcd->ctxt);
+	return head == tail;
+}
+
+/*
+ * Modify the RCVCTRL register in chip-specific way. This
+ * is a function because bit positions and (future) register
+ * location is chip-specifc, but the needed operations are
+ * generic. <op> is a bit-mask because we often want to
+ * do multiple modifications.
+ */
+static void rcvctrl_7220_mod(struct qib_pportdata *ppd, unsigned int op,
+			     int ctxt)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 mask, val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+	if (op & QIB_RCVCTRL_TAILUPD_ENB)
+		dd->rcvctrl |= (1ULL << IBA7220_R_TAILUPD_SHIFT);
+	if (op & QIB_RCVCTRL_TAILUPD_DIS)
+		dd->rcvctrl &= ~(1ULL << IBA7220_R_TAILUPD_SHIFT);
+	if (op & QIB_RCVCTRL_PKEY_ENB)
+		dd->rcvctrl &= ~(1ULL << IBA7220_R_PKEY_DIS_SHIFT);
+	if (op & QIB_RCVCTRL_PKEY_DIS)
+		dd->rcvctrl |= (1ULL << IBA7220_R_PKEY_DIS_SHIFT);
+	if (ctxt < 0)
+		mask = (1ULL << dd->ctxtcnt) - 1;
+	else
+		mask = (1ULL << ctxt);
+	if (op & QIB_RCVCTRL_CTXT_ENB) {
+		/* always done for specific ctxt */
+		dd->rcvctrl |= (mask << SYM_LSB(RcvCtrl, PortEnable));
+		if (!(dd->flags & QIB_NODMA_RTAIL))
+			dd->rcvctrl |= 1ULL << IBA7220_R_TAILUPD_SHIFT;
+		/* Write these registers before the context is enabled. */
+		qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr, ctxt,
+			dd->rcd[ctxt]->rcvhdrqtailaddr_phys);
+		qib_write_kreg_ctxt(dd, kr_rcvhdraddr, ctxt,
+			dd->rcd[ctxt]->rcvhdrq_phys);
+		dd->rcd[ctxt]->seq_cnt = 1;
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS)
+		dd->rcvctrl &= ~(mask << SYM_LSB(RcvCtrl, PortEnable));
+	if (op & QIB_RCVCTRL_INTRAVAIL_ENB)
+		dd->rcvctrl |= (mask << IBA7220_R_INTRAVAIL_SHIFT);
+	if (op & QIB_RCVCTRL_INTRAVAIL_DIS)
+		dd->rcvctrl &= ~(mask << IBA7220_R_INTRAVAIL_SHIFT);
+	qib_write_kreg(dd, kr_rcvctrl, dd->rcvctrl);
+	if ((op & QIB_RCVCTRL_INTRAVAIL_ENB) && dd->rhdrhead_intr_off) {
+		/* arm rcv interrupt */
+		val = qib_read_ureg32(dd, ur_rcvhdrhead, ctxt) |
+			dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	}
+	if (op & QIB_RCVCTRL_CTXT_ENB) {
+		/*
+		 * Init the context registers also; if we were
+		 * disabled, tail and head should both be zero
+		 * already from the enable, but since we don't
+		 * know, we have to do it explicitly.
+		 */
+		val = qib_read_ureg32(dd, ur_rcvegrindextail, ctxt);
+		qib_write_ureg(dd, ur_rcvegrindexhead, val, ctxt);
+
+		val = qib_read_ureg32(dd, ur_rcvhdrtail, ctxt);
+		dd->rcd[ctxt]->head = val;
+		/* If kctxt, interrupt on next receive. */
+		if (ctxt < dd->first_user_ctxt)
+			val |= dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS) {
+		if (ctxt >= 0) {
+			qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr, ctxt, 0);
+			qib_write_kreg_ctxt(dd, kr_rcvhdraddr, ctxt, 0);
+		} else {
+			unsigned i;
+
+			for (i = 0; i < dd->cfgctxts; i++) {
+				qib_write_kreg_ctxt(dd, kr_rcvhdrtailaddr,
+						    i, 0);
+				qib_write_kreg_ctxt(dd, kr_rcvhdraddr, i, 0);
+			}
+		}
+	}
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+}
+
+/*
+ * Modify the SENDCTRL register in chip-specific way. This
+ * is a function there may be multiple such registers with
+ * slightly different layouts. To start, we assume the
+ * "canonical" register layout of the first chips.
+ * Chip requires no back-back sendctrl writes, so write
+ * scratch register after writing sendctrl
+ */
+static void sendctrl_7220_mod(struct qib_pportdata *ppd, u32 op)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 tmp_dd_sendctrl;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+
+	/* First the ones that are "sticky", saved in shadow */
+	if (op & QIB_SENDCTRL_CLEAR)
+		dd->sendctrl = 0;
+	if (op & QIB_SENDCTRL_SEND_DIS)
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, SPioEnable);
+	else if (op & QIB_SENDCTRL_SEND_ENB) {
+		dd->sendctrl |= SYM_MASK(SendCtrl, SPioEnable);
+		if (dd->flags & QIB_USE_SPCL_TRIG)
+			dd->sendctrl |= SYM_MASK(SendCtrl,
+						 SSpecialTriggerEn);
+	}
+	if (op & QIB_SENDCTRL_AVAIL_DIS)
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, SendBufAvailUpd);
+	else if (op & QIB_SENDCTRL_AVAIL_ENB)
+		dd->sendctrl |= SYM_MASK(SendCtrl, SendBufAvailUpd);
+
+	if (op & QIB_SENDCTRL_DISARM_ALL) {
+		u32 i, last;
+
+		tmp_dd_sendctrl = dd->sendctrl;
+		/*
+		 * disarm any that are not yet launched, disabling sends
+		 * and updates until done.
+		 */
+		last = dd->piobcnt2k + dd->piobcnt4k;
+		tmp_dd_sendctrl &=
+			~(SYM_MASK(SendCtrl, SPioEnable) |
+			  SYM_MASK(SendCtrl, SendBufAvailUpd));
+		for (i = 0; i < last; i++) {
+			qib_write_kreg(dd, kr_sendctrl,
+				       tmp_dd_sendctrl |
+				       SYM_MASK(SendCtrl, Disarm) | i);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+	}
+
+	tmp_dd_sendctrl = dd->sendctrl;
+
+	if (op & QIB_SENDCTRL_FLUSH)
+		tmp_dd_sendctrl |= SYM_MASK(SendCtrl, Abort);
+	if (op & QIB_SENDCTRL_DISARM)
+		tmp_dd_sendctrl |= SYM_MASK(SendCtrl, Disarm) |
+			((op & QIB_7220_SendCtrl_DisarmPIOBuf_RMASK) <<
+			 SYM_LSB(SendCtrl, DisarmPIOBuf));
+	if ((op & QIB_SENDCTRL_AVAIL_BLIP) &&
+	    (dd->sendctrl & SYM_MASK(SendCtrl, SendBufAvailUpd)))
+		tmp_dd_sendctrl &= ~SYM_MASK(SendCtrl, SendBufAvailUpd);
+
+	qib_write_kreg(dd, kr_sendctrl, tmp_dd_sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	if (op & QIB_SENDCTRL_AVAIL_BLIP) {
+		qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+
+	if (op & QIB_SENDCTRL_FLUSH) {
+		u32 v;
+		/*
+		 * ensure writes have hit chip, then do a few
+		 * more reads, to allow DMA of pioavail registers
+		 * to occur, so in-memory copy is in sync with
+		 * the chip.  Not always safe to sleep.
+		 */
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		qib_read_kreg32(dd, kr_scratch);
+	}
+}
+
+/**
+ * qib_portcntr_7220 - read a per-port counter
+ * @dd: the qlogic_ib device
+ * @creg: the counter to snapshot
+ */
+static u64 qib_portcntr_7220(struct qib_pportdata *ppd, u32 reg)
+{
+	u64 ret = 0ULL;
+	struct qib_devdata *dd = ppd->dd;
+	u16 creg;
+	/* 0xffff for unimplemented or synthesized counters */
+	static const u16 xlator[] = {
+		[QIBPORTCNTR_PKTSEND] = cr_pktsend,
+		[QIBPORTCNTR_WORDSEND] = cr_wordsend,
+		[QIBPORTCNTR_PSXMITDATA] = cr_psxmitdatacount,
+		[QIBPORTCNTR_PSXMITPKTS] = cr_psxmitpktscount,
+		[QIBPORTCNTR_PSXMITWAIT] = cr_psxmitwaitcount,
+		[QIBPORTCNTR_SENDSTALL] = cr_sendstall,
+		[QIBPORTCNTR_PKTRCV] = cr_pktrcv,
+		[QIBPORTCNTR_PSRCVDATA] = cr_psrcvdatacount,
+		[QIBPORTCNTR_PSRCVPKTS] = cr_psrcvpktscount,
+		[QIBPORTCNTR_RCVEBP] = cr_rcvebp,
+		[QIBPORTCNTR_RCVOVFL] = cr_rcvovfl,
+		[QIBPORTCNTR_WORDRCV] = cr_wordrcv,
+		[QIBPORTCNTR_RXDROPPKT] = cr_rxdroppkt,
+		[QIBPORTCNTR_RXLOCALPHYERR] = cr_rxotherlocalphyerr,
+		[QIBPORTCNTR_RXVLERR] = cr_rxvlerr,
+		[QIBPORTCNTR_ERRICRC] = cr_erricrc,
+		[QIBPORTCNTR_ERRVCRC] = cr_errvcrc,
+		[QIBPORTCNTR_ERRLPCRC] = cr_errlpcrc,
+		[QIBPORTCNTR_BADFORMAT] = cr_badformat,
+		[QIBPORTCNTR_ERR_RLEN] = cr_err_rlen,
+		[QIBPORTCNTR_IBSYMBOLERR] = cr_ibsymbolerr,
+		[QIBPORTCNTR_INVALIDRLEN] = cr_invalidrlen,
+		[QIBPORTCNTR_UNSUPVL] = cr_txunsupvl,
+		[QIBPORTCNTR_EXCESSBUFOVFL] = cr_excessbufferovfl,
+		[QIBPORTCNTR_ERRLINK] = cr_errlink,
+		[QIBPORTCNTR_IBLINKDOWN] = cr_iblinkdown,
+		[QIBPORTCNTR_IBLINKERRRECOV] = cr_iblinkerrrecov,
+		[QIBPORTCNTR_LLI] = cr_locallinkintegrityerr,
+		[QIBPORTCNTR_PSINTERVAL] = cr_psinterval,
+		[QIBPORTCNTR_PSSTART] = cr_psstart,
+		[QIBPORTCNTR_PSSTAT] = cr_psstat,
+		[QIBPORTCNTR_VL15PKTDROP] = cr_vl15droppedpkt,
+		[QIBPORTCNTR_ERRPKEY] = cr_errpkey,
+		[QIBPORTCNTR_KHDROVFL] = 0xffff,
+	};
+
+	if (reg >= ARRAY_SIZE(xlator)) {
+		qib_devinfo(ppd->dd->pcidev,
+			 "Unimplemented portcounter %u\n", reg);
+		goto done;
+	}
+	creg = xlator[reg];
+
+	if (reg == QIBPORTCNTR_KHDROVFL) {
+		int i;
+
+		/* sum over all kernel contexts */
+		for (i = 0; i < dd->first_user_ctxt; i++)
+			ret += read_7220_creg32(dd, cr_portovfl + i);
+	}
+	if (creg == 0xffff)
+		goto done;
+
+	/*
+	 * only fast incrementing counters are 64bit; use 32 bit reads to
+	 * avoid two independent reads when on opteron
+	 */
+	if ((creg == cr_wordsend || creg == cr_wordrcv ||
+	     creg == cr_pktsend || creg == cr_pktrcv))
+		ret = read_7220_creg(dd, creg);
+	else
+		ret = read_7220_creg32(dd, creg);
+	if (creg == cr_ibsymbolerr) {
+		if (dd->pport->cpspec->ibdeltainprog)
+			ret -= ret - ppd->cpspec->ibsymsnap;
+		ret -= dd->pport->cpspec->ibsymdelta;
+	} else if (creg == cr_iblinkerrrecov) {
+		if (dd->pport->cpspec->ibdeltainprog)
+			ret -= ret - ppd->cpspec->iblnkerrsnap;
+		ret -= dd->pport->cpspec->iblnkerrdelta;
+	}
+done:
+	return ret;
+}
+
+/*
+ * Device counter names (not port-specific), one line per stat,
+ * single string.  Used by utilities like ipathstats to print the stats
+ * in a way which works for different versions of drivers, without changing
+ * the utility.  Names need to be 12 chars or less (w/o newline), for proper
+ * display by utility.
+ * Non-error counters are first.
+ * Start of "error" conters is indicated by a leading "E " on the first
+ * "error" counter, and doesn't count in label length.
+ * The EgrOvfl list needs to be last so we truncate them at the configured
+ * context count for the device.
+ * cntr7220indices contains the corresponding register indices.
+ */
+static const char cntr7220names[] =
+	"Interrupts\n"
+	"HostBusStall\n"
+	"E RxTIDFull\n"
+	"RxTIDInvalid\n"
+	"Ctxt0EgrOvfl\n"
+	"Ctxt1EgrOvfl\n"
+	"Ctxt2EgrOvfl\n"
+	"Ctxt3EgrOvfl\n"
+	"Ctxt4EgrOvfl\n"
+	"Ctxt5EgrOvfl\n"
+	"Ctxt6EgrOvfl\n"
+	"Ctxt7EgrOvfl\n"
+	"Ctxt8EgrOvfl\n"
+	"Ctxt9EgrOvfl\n"
+	"Ctx10EgrOvfl\n"
+	"Ctx11EgrOvfl\n"
+	"Ctx12EgrOvfl\n"
+	"Ctx13EgrOvfl\n"
+	"Ctx14EgrOvfl\n"
+	"Ctx15EgrOvfl\n"
+	"Ctx16EgrOvfl\n";
+
+static const size_t cntr7220indices[] = {
+	cr_lbint,
+	cr_lbflowstall,
+	cr_errtidfull,
+	cr_errtidvalid,
+	cr_portovfl + 0,
+	cr_portovfl + 1,
+	cr_portovfl + 2,
+	cr_portovfl + 3,
+	cr_portovfl + 4,
+	cr_portovfl + 5,
+	cr_portovfl + 6,
+	cr_portovfl + 7,
+	cr_portovfl + 8,
+	cr_portovfl + 9,
+	cr_portovfl + 10,
+	cr_portovfl + 11,
+	cr_portovfl + 12,
+	cr_portovfl + 13,
+	cr_portovfl + 14,
+	cr_portovfl + 15,
+	cr_portovfl + 16,
+};
+
+/*
+ * same as cntr7220names and cntr7220indices, but for port-specific counters.
+ * portcntr7220indices is somewhat complicated by some registers needing
+ * adjustments of various kinds, and those are ORed with _PORT_VIRT_FLAG
+ */
+static const char portcntr7220names[] =
+	"TxPkt\n"
+	"TxFlowPkt\n"
+	"TxWords\n"
+	"RxPkt\n"
+	"RxFlowPkt\n"
+	"RxWords\n"
+	"TxFlowStall\n"
+	"TxDmaDesc\n"  /* 7220 and 7322-only */
+	"E RxDlidFltr\n"  /* 7220 and 7322-only */
+	"IBStatusChng\n"
+	"IBLinkDown\n"
+	"IBLnkRecov\n"
+	"IBRxLinkErr\n"
+	"IBSymbolErr\n"
+	"RxLLIErr\n"
+	"RxBadFormat\n"
+	"RxBadLen\n"
+	"RxBufOvrfl\n"
+	"RxEBP\n"
+	"RxFlowCtlErr\n"
+	"RxICRCerr\n"
+	"RxLPCRCerr\n"
+	"RxVCRCerr\n"
+	"RxInvalLen\n"
+	"RxInvalPKey\n"
+	"RxPktDropped\n"
+	"TxBadLength\n"
+	"TxDropped\n"
+	"TxInvalLen\n"
+	"TxUnderrun\n"
+	"TxUnsupVL\n"
+	"RxLclPhyErr\n" /* 7220 and 7322-only */
+	"RxVL15Drop\n" /* 7220 and 7322-only */
+	"RxVlErr\n" /* 7220 and 7322-only */
+	"XcessBufOvfl\n" /* 7220 and 7322-only */
+	;
+
+#define _PORT_VIRT_FLAG 0x8000 /* "virtual", need adjustments */
+static const size_t portcntr7220indices[] = {
+	QIBPORTCNTR_PKTSEND | _PORT_VIRT_FLAG,
+	cr_pktsendflow,
+	QIBPORTCNTR_WORDSEND | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_PKTRCV | _PORT_VIRT_FLAG,
+	cr_pktrcvflowctrl,
+	QIBPORTCNTR_WORDRCV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_SENDSTALL | _PORT_VIRT_FLAG,
+	cr_txsdmadesc,
+	cr_rxdlidfltr,
+	cr_ibstatuschange,
+	QIBPORTCNTR_IBLINKDOWN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBLINKERRRECOV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLINK | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBSYMBOLERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_LLI | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_BADFORMAT | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERR_RLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVOVFL | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVEBP | _PORT_VIRT_FLAG,
+	cr_rcvflowctrl_err,
+	QIBPORTCNTR_ERRICRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLPCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRVCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_INVALIDRLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRPKEY | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RXDROPPKT | _PORT_VIRT_FLAG,
+	cr_invalidslen,
+	cr_senddropped,
+	cr_errslen,
+	cr_sendunderrun,
+	cr_txunsupvl,
+	QIBPORTCNTR_RXLOCALPHYERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_VL15PKTDROP | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RXVLERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_EXCESSBUFOVFL | _PORT_VIRT_FLAG,
+};
+
+/* do all the setup to make the counter reads efficient later */
+static void init_7220_cntrnames(struct qib_devdata *dd)
+{
+	int i, j = 0;
+	char *s;
+
+	for (i = 0, s = (char *)cntr7220names; s && j <= dd->cfgctxts;
+	     i++) {
+		/* we always have at least one counter before the egrovfl */
+		if (!j && !strncmp("Ctxt0EgrOvfl", s + 1, 12))
+			j = 1;
+		s = strchr(s + 1, '\n');
+		if (s && j)
+			j++;
+	}
+	dd->cspec->ncntrs = i;
+	if (!s)
+		/* full list; size is without terminating null */
+		dd->cspec->cntrnamelen = sizeof(cntr7220names) - 1;
+	else
+		dd->cspec->cntrnamelen = 1 + s - cntr7220names;
+	dd->cspec->cntrs = kmalloc(dd->cspec->ncntrs
+		* sizeof(u64), GFP_KERNEL);
+	if (!dd->cspec->cntrs)
+		qib_dev_err(dd, "Failed allocation for counters\n");
+
+	for (i = 0, s = (char *)portcntr7220names; s; i++)
+		s = strchr(s + 1, '\n');
+	dd->cspec->nportcntrs = i - 1;
+	dd->cspec->portcntrnamelen = sizeof(portcntr7220names) - 1;
+	dd->cspec->portcntrs = kmalloc(dd->cspec->nportcntrs
+		* sizeof(u64), GFP_KERNEL);
+	if (!dd->cspec->portcntrs)
+		qib_dev_err(dd, "Failed allocation for portcounters\n");
+}
+
+static u32 qib_read_7220cntrs(struct qib_devdata *dd, loff_t pos, char **namep,
+			      u64 **cntrp)
+{
+	u32 ret;
+
+	if (!dd->cspec->cntrs) {
+		ret = 0;
+		goto done;
+	}
+
+	if (namep) {
+		*namep = (char *)cntr7220names;
+		ret = dd->cspec->cntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+	} else {
+		u64 *cntr = dd->cspec->cntrs;
+		int i;
+
+		ret = dd->cspec->ncntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->ncntrs; i++)
+			*cntr++ = read_7220_creg32(dd, cntr7220indices[i]);
+	}
+done:
+	return ret;
+}
+
+static u32 qib_read_7220portcntrs(struct qib_devdata *dd, loff_t pos, u32 port,
+				  char **namep, u64 **cntrp)
+{
+	u32 ret;
+
+	if (!dd->cspec->portcntrs) {
+		ret = 0;
+		goto done;
+	}
+	if (namep) {
+		*namep = (char *)portcntr7220names;
+		ret = dd->cspec->portcntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+	} else {
+		u64 *cntr = dd->cspec->portcntrs;
+		struct qib_pportdata *ppd = &dd->pport[port];
+		int i;
+
+		ret = dd->cspec->nportcntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->nportcntrs; i++) {
+			if (portcntr7220indices[i] & _PORT_VIRT_FLAG)
+				*cntr++ = qib_portcntr_7220(ppd,
+					portcntr7220indices[i] &
+					~_PORT_VIRT_FLAG);
+			else
+				*cntr++ = read_7220_creg32(dd,
+					   portcntr7220indices[i]);
+		}
+	}
+done:
+	return ret;
+}
+
+/**
+ * qib_get_7220_faststats - get word counters from chip before they overflow
+ * @opaque - contains a pointer to the qlogic_ib device qib_devdata
+ *
+ * This needs more work; in particular, decision on whether we really
+ * need traffic_wds done the way it is
+ * called from add_timer
+ */
+static void qib_get_7220_faststats(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *) opaque;
+	struct qib_pportdata *ppd = dd->pport;
+	unsigned long flags;
+	u64 traffic_wds;
+
+	/*
+	 * don't access the chip while running diags, or memory diags can
+	 * fail
+	 */
+	if (!(dd->flags & QIB_INITTED) || dd->diag_client)
+		/* but re-arm the timer, for diags case; won't hurt other */
+		goto done;
+
+	/*
+	 * We now try to maintain an activity timer, based on traffic
+	 * exceeding a threshold, so we need to check the word-counts
+	 * even if they are 64-bit.
+	 */
+	traffic_wds = qib_portcntr_7220(ppd, cr_wordsend) +
+		qib_portcntr_7220(ppd, cr_wordrcv);
+	spin_lock_irqsave(&dd->eep_st_lock, flags);
+	traffic_wds -= dd->traffic_wds;
+	dd->traffic_wds += traffic_wds;
+	if (traffic_wds  >= QIB_TRAFFIC_ACTIVE_THRESHOLD)
+		atomic_add(5, &dd->active_time); /* S/B #define */
+	spin_unlock_irqrestore(&dd->eep_st_lock, flags);
+done:
+	mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
+}
+
+/*
+ * If we are using MSI, try to fallback to INTx.
+ */
+static int qib_7220_intr_fallback(struct qib_devdata *dd)
+{
+	if (!dd->msi_lo)
+		return 0;
+
+	qib_devinfo(dd->pcidev,
+		"MSI interrupt not detected, trying INTx interrupts\n");
+	qib_7220_free_irq(dd);
+	qib_enable_intx(dd->pcidev);
+	/*
+	 * Some newer kernels require free_irq before disable_msi,
+	 * and irq can be changed during disable and INTx enable
+	 * and we need to therefore use the pcidev->irq value,
+	 * not our saved MSI value.
+	 */
+	dd->cspec->irq = dd->pcidev->irq;
+	qib_setup_7220_interrupt(dd);
+	return 1;
+}
+
+/*
+ * Reset the XGXS (between serdes and IBC).  Slightly less intrusive
+ * than resetting the IBC or external link state, and useful in some
+ * cases to cause some retraining.  To do this right, we reset IBC
+ * as well.
+ */
+static void qib_7220_xgxs_reset(struct qib_pportdata *ppd)
+{
+	u64 val, prev_val;
+	struct qib_devdata *dd = ppd->dd;
+
+	prev_val = qib_read_kreg64(dd, kr_xgxs_cfg);
+	val = prev_val | QLOGIC_IB_XGXS_RESET;
+	prev_val &= ~QLOGIC_IB_XGXS_RESET; /* be sure */
+	qib_write_kreg(dd, kr_control,
+		       dd->control & ~QLOGIC_IB_C_LINKENABLE);
+	qib_write_kreg(dd, kr_xgxs_cfg, val);
+	qib_read_kreg32(dd, kr_scratch);
+	qib_write_kreg(dd, kr_xgxs_cfg, prev_val);
+	qib_write_kreg(dd, kr_control, dd->control);
+}
+
+/*
+ * For this chip, we want to use the same buffer every time
+ * when we are trying to bring the link up (they are always VL15
+ * packets).  At that link state the packet should always go out immediately
+ * (or at least be discarded at the tx interface if the link is down).
+ * If it doesn't, and the buffer isn't available, that means some other
+ * sender has gotten ahead of us, and is preventing our packet from going
+ * out.  In that case, we flush all packets, and try again.  If that still
+ * fails, we fail the request, and hope things work the next time around.
+ *
+ * We don't need very complicated heuristics on whether the packet had
+ * time to go out or not, since even at SDR 1X, it goes out in very short
+ * time periods, covered by the chip reads done here and as part of the
+ * flush.
+ */
+static u32 __iomem *get_7220_link_buf(struct qib_pportdata *ppd, u32 *bnum)
+{
+	u32 __iomem *buf;
+	u32 lbuf = ppd->dd->cspec->lastbuf_for_pio;
+	int do_cleanup;
+	unsigned long flags;
+
+	/*
+	 * always blip to get avail list updated, since it's almost
+	 * always needed, and is fairly cheap.
+	 */
+	sendctrl_7220_mod(ppd->dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+	qib_read_kreg64(ppd->dd, kr_scratch); /* extra chip flush */
+	buf = qib_getsendbuf_range(ppd->dd, bnum, lbuf, lbuf);
+	if (buf)
+		goto done;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	if (ppd->sdma_state.current_state == qib_sdma_state_s20_idle &&
+	    ppd->sdma_state.current_state != qib_sdma_state_s00_hw_down) {
+		__qib_sdma_process_event(ppd, qib_sdma_event_e00_go_hw_down);
+		do_cleanup = 0;
+	} else {
+		do_cleanup = 1;
+		qib_7220_sdma_hw_clean_up(ppd);
+	}
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	if (do_cleanup) {
+		qib_read_kreg64(ppd->dd, kr_scratch); /* extra chip flush */
+		buf = qib_getsendbuf_range(ppd->dd, bnum, lbuf, lbuf);
+	}
+done:
+	return buf;
+}
+
+/*
+ * This code for non-IBTA-compliant IB speed negotiation is only known to
+ * work for the SDR to DDR transition, and only between an HCA and a switch
+ * with recent firmware.  It is based on observed heuristics, rather than
+ * actual knowledge of the non-compliant speed negotiation.
+ * It has a number of hard-coded fields, since the hope is to rewrite this
+ * when a spec is available on how the negoation is intended to work.
+ */
+static void autoneg_7220_sendpkt(struct qib_pportdata *ppd, u32 *hdr,
+				 u32 dcnt, u32 *data)
+{
+	int i;
+	u64 pbc;
+	u32 __iomem *piobuf;
+	u32 pnum;
+	struct qib_devdata *dd = ppd->dd;
+
+	i = 0;
+	pbc = 7 + dcnt + 1; /* 7 dword header, dword data, icrc */
+	pbc |= PBC_7220_VL15_SEND;
+	while (!(piobuf = get_7220_link_buf(ppd, &pnum))) {
+		if (i++ > 5)
+			return;
+		udelay(2);
+	}
+	sendctrl_7220_mod(dd->pport, QIB_SENDCTRL_DISARM_BUF(pnum));
+	writeq(pbc, piobuf);
+	qib_flush_wc();
+	qib_pio_copy(piobuf + 2, hdr, 7);
+	qib_pio_copy(piobuf + 9, data, dcnt);
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pnum >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pnum);
+}
+
+/*
+ * _start packet gets sent twice at start, _done gets sent twice at end
+ */
+static void autoneg_7220_send(struct qib_pportdata *ppd, int which)
+{
+	struct qib_devdata *dd = ppd->dd;
+	static u32 swapped;
+	u32 dw, i, hcnt, dcnt, *data;
+	static u32 hdr[7] = { 0xf002ffff, 0x48ffff, 0x6400abba };
+	static u32 madpayload_start[0x40] = {
+		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
+		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+		0x1, 0x1388, 0x15e, 0x1, /* rest 0's */
+		};
+	static u32 madpayload_done[0x40] = {
+		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
+		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+		0x40000001, 0x1388, 0x15e, /* rest 0's */
+		};
+
+	dcnt = ARRAY_SIZE(madpayload_start);
+	hcnt = ARRAY_SIZE(hdr);
+	if (!swapped) {
+		/* for maintainability, do it at runtime */
+		for (i = 0; i < hcnt; i++) {
+			dw = (__force u32) cpu_to_be32(hdr[i]);
+			hdr[i] = dw;
+		}
+		for (i = 0; i < dcnt; i++) {
+			dw = (__force u32) cpu_to_be32(madpayload_start[i]);
+			madpayload_start[i] = dw;
+			dw = (__force u32) cpu_to_be32(madpayload_done[i]);
+			madpayload_done[i] = dw;
+		}
+		swapped = 1;
+	}
+
+	data = which ? madpayload_done : madpayload_start;
+
+	autoneg_7220_sendpkt(ppd, hdr, dcnt, data);
+	qib_read_kreg64(dd, kr_scratch);
+	udelay(2);
+	autoneg_7220_sendpkt(ppd, hdr, dcnt, data);
+	qib_read_kreg64(dd, kr_scratch);
+	udelay(2);
+}
+
+/*
+ * Do the absolute minimum to cause an IB speed change, and make it
+ * ready, but don't actually trigger the change.   The caller will
+ * do that when ready (if link is in Polling training state, it will
+ * happen immediately, otherwise when link next goes down)
+ *
+ * This routine should only be used as part of the DDR autonegotation
+ * code for devices that are not compliant with IB 1.2 (or code that
+ * fixes things up for same).
+ *
+ * When link has gone down, and autoneg enabled, or autoneg has
+ * failed and we give up until next time we set both speeds, and
+ * then we want IBTA enabled as well as "use max enabled speed.
+ */
+static void set_7220_ibspeed_fast(struct qib_pportdata *ppd, u32 speed)
+{
+	ppd->cpspec->ibcddrctrl &= ~(IBA7220_IBC_SPEED_AUTONEG_MASK |
+		IBA7220_IBC_IBTA_1_2_MASK);
+
+	if (speed == (QIB_IB_SDR | QIB_IB_DDR))
+		ppd->cpspec->ibcddrctrl |= IBA7220_IBC_SPEED_AUTONEG_MASK |
+			IBA7220_IBC_IBTA_1_2_MASK;
+	else
+		ppd->cpspec->ibcddrctrl |= speed == QIB_IB_DDR ?
+			IBA7220_IBC_SPEED_DDR : IBA7220_IBC_SPEED_SDR;
+
+	qib_write_kreg(ppd->dd, kr_ibcddrctrl, ppd->cpspec->ibcddrctrl);
+	qib_write_kreg(ppd->dd, kr_scratch, 0);
+}
+
+/*
+ * This routine is only used when we are not talking to another
+ * IB 1.2-compliant device that we think can do DDR.
+ * (This includes all existing switch chips as of Oct 2007.)
+ * 1.2-compliant devices go directly to DDR prior to reaching INIT
+ */
+static void try_7220_autoneg(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+
+	/*
+	 * Required for older non-IB1.2 DDR switches.  Newer
+	 * non-IB-compliant switches don't need it, but so far,
+	 * aren't bothered by it either.  "Magic constant"
+	 */
+	qib_write_kreg(ppd->dd, kr_ncmodectrl, 0x3b9dc07);
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags |= QIBL_IB_AUTONEG_INPROG;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	autoneg_7220_send(ppd, 0);
+	set_7220_ibspeed_fast(ppd, QIB_IB_DDR);
+
+	toggle_7220_rclkrls(ppd->dd);
+	/* 2 msec is minimum length of a poll cycle */
+	queue_delayed_work(ib_wq, &ppd->cpspec->autoneg_work,
+			   msecs_to_jiffies(2));
+}
+
+/*
+ * Handle the empirically determined mechanism for auto-negotiation
+ * of DDR speed with switches.
+ */
+static void autoneg_7220_work(struct work_struct *work)
+{
+	struct qib_pportdata *ppd;
+	struct qib_devdata *dd;
+	u64 startms;
+	u32 i;
+	unsigned long flags;
+
+	ppd = &container_of(work, struct qib_chippport_specific,
+			    autoneg_work.work)->pportdata;
+	dd = ppd->dd;
+
+	startms = jiffies_to_msecs(jiffies);
+
+	/*
+	 * Busy wait for this first part, it should be at most a
+	 * few hundred usec, since we scheduled ourselves for 2msec.
+	 */
+	for (i = 0; i < 25; i++) {
+		if (SYM_FIELD(ppd->lastibcstat, IBCStatus, LinkTrainingState)
+		     == IB_7220_LT_STATE_POLLQUIET) {
+			qib_set_linkstate(ppd, QIB_IB_LINKDOWN_DISABLE);
+			break;
+		}
+		udelay(100);
+	}
+
+	if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+		goto done; /* we got there early or told to stop */
+
+	/* we expect this to timeout */
+	if (wait_event_timeout(ppd->cpspec->autoneg_wait,
+			       !(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+			       msecs_to_jiffies(90)))
+		goto done;
+
+	toggle_7220_rclkrls(dd);
+
+	/* we expect this to timeout */
+	if (wait_event_timeout(ppd->cpspec->autoneg_wait,
+			       !(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+			       msecs_to_jiffies(1700)))
+		goto done;
+
+	set_7220_ibspeed_fast(ppd, QIB_IB_SDR);
+	toggle_7220_rclkrls(dd);
+
+	/*
+	 * Wait up to 250 msec for link to train and get to INIT at DDR;
+	 * this should terminate early.
+	 */
+	wait_event_timeout(ppd->cpspec->autoneg_wait,
+		!(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+		msecs_to_jiffies(250));
+done:
+	if (ppd->lflags & QIBL_IB_AUTONEG_INPROG) {
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_AUTONEG_INPROG;
+		if (dd->cspec->autoneg_tries == AUTONEG_TRIES) {
+			ppd->lflags |= QIBL_IB_AUTONEG_FAILED;
+			dd->cspec->autoneg_tries = 0;
+		}
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		set_7220_ibspeed_fast(ppd, ppd->link_speed_enabled);
+	}
+}
+
+static u32 qib_7220_iblink_state(u64 ibcs)
+{
+	u32 state = (u32)SYM_FIELD(ibcs, IBCStatus, LinkState);
+
+	switch (state) {
+	case IB_7220_L_STATE_INIT:
+		state = IB_PORT_INIT;
+		break;
+	case IB_7220_L_STATE_ARM:
+		state = IB_PORT_ARMED;
+		break;
+	case IB_7220_L_STATE_ACTIVE:
+		/* fall through */
+	case IB_7220_L_STATE_ACT_DEFER:
+		state = IB_PORT_ACTIVE;
+		break;
+	default: /* fall through */
+	case IB_7220_L_STATE_DOWN:
+		state = IB_PORT_DOWN;
+		break;
+	}
+	return state;
+}
+
+/* returns the IBTA port state, rather than the IBC link training state */
+static u8 qib_7220_phys_portstate(u64 ibcs)
+{
+	u8 state = (u8)SYM_FIELD(ibcs, IBCStatus, LinkTrainingState);
+	return qib_7220_physportstate[state];
+}
+
+static int qib_7220_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
+{
+	int ret = 0, symadj = 0;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags &= ~QIBL_IB_FORCE_NOTIFY;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	if (!ibup) {
+		/*
+		 * When the link goes down we don't want AEQ running, so it
+		 * won't interfere with IBC training, etc., and we need
+		 * to go back to the static SerDes preset values.
+		 */
+		if (!(ppd->lflags & (QIBL_IB_AUTONEG_FAILED |
+				     QIBL_IB_AUTONEG_INPROG)))
+			set_7220_ibspeed_fast(ppd, ppd->link_speed_enabled);
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+			qib_sd7220_presets(dd);
+			qib_cancel_sends(ppd); /* initial disarm, etc. */
+			spin_lock_irqsave(&ppd->sdma_lock, flags);
+			if (__qib_sdma_running(ppd))
+				__qib_sdma_process_event(ppd,
+					qib_sdma_event_e70_go_idle);
+			spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+		}
+		/* this might better in qib_sd7220_presets() */
+		set_7220_relock_poll(dd, ibup);
+	} else {
+		if (qib_compat_ddr_negotiate &&
+		    !(ppd->lflags & (QIBL_IB_AUTONEG_FAILED |
+				     QIBL_IB_AUTONEG_INPROG)) &&
+		    ppd->link_speed_active == QIB_IB_SDR &&
+		    (ppd->link_speed_enabled & (QIB_IB_DDR | QIB_IB_SDR)) ==
+		    (QIB_IB_DDR | QIB_IB_SDR) &&
+		    dd->cspec->autoneg_tries < AUTONEG_TRIES) {
+			/* we are SDR, and DDR auto-negotiation enabled */
+			++dd->cspec->autoneg_tries;
+			if (!ppd->cpspec->ibdeltainprog) {
+				ppd->cpspec->ibdeltainprog = 1;
+				ppd->cpspec->ibsymsnap = read_7220_creg32(dd,
+					cr_ibsymbolerr);
+				ppd->cpspec->iblnkerrsnap = read_7220_creg32(dd,
+					cr_iblinkerrrecov);
+			}
+			try_7220_autoneg(ppd);
+			ret = 1; /* no other IB status change processing */
+		} else if ((ppd->lflags & QIBL_IB_AUTONEG_INPROG) &&
+			   ppd->link_speed_active == QIB_IB_SDR) {
+			autoneg_7220_send(ppd, 1);
+			set_7220_ibspeed_fast(ppd, QIB_IB_DDR);
+			udelay(2);
+			toggle_7220_rclkrls(dd);
+			ret = 1; /* no other IB status change processing */
+		} else {
+			if ((ppd->lflags & QIBL_IB_AUTONEG_INPROG) &&
+			    (ppd->link_speed_active & QIB_IB_DDR)) {
+				spin_lock_irqsave(&ppd->lflags_lock, flags);
+				ppd->lflags &= ~(QIBL_IB_AUTONEG_INPROG |
+						 QIBL_IB_AUTONEG_FAILED);
+				spin_unlock_irqrestore(&ppd->lflags_lock,
+						       flags);
+				dd->cspec->autoneg_tries = 0;
+				/* re-enable SDR, for next link down */
+				set_7220_ibspeed_fast(ppd,
+						      ppd->link_speed_enabled);
+				wake_up(&ppd->cpspec->autoneg_wait);
+				symadj = 1;
+			} else if (ppd->lflags & QIBL_IB_AUTONEG_FAILED) {
+				/*
+				 * Clear autoneg failure flag, and do setup
+				 * so we'll try next time link goes down and
+				 * back to INIT (possibly connected to a
+				 * different device).
+				 */
+				spin_lock_irqsave(&ppd->lflags_lock, flags);
+				ppd->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+				spin_unlock_irqrestore(&ppd->lflags_lock,
+						       flags);
+				ppd->cpspec->ibcddrctrl |=
+					IBA7220_IBC_IBTA_1_2_MASK;
+				qib_write_kreg(dd, kr_ncmodectrl, 0);
+				symadj = 1;
+			}
+		}
+
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+			symadj = 1;
+
+		if (!ret) {
+			ppd->delay_mult = rate_to_delay
+			    [(ibcs >> IBA7220_LINKSPEED_SHIFT) & 1]
+			    [(ibcs >> IBA7220_LINKWIDTH_SHIFT) & 1];
+
+			set_7220_relock_poll(dd, ibup);
+			spin_lock_irqsave(&ppd->sdma_lock, flags);
+			/*
+			 * Unlike 7322, the 7220 needs this, due to lack of
+			 * interrupt in some cases when we have sdma active
+			 * when the link goes down.
+			 */
+			if (ppd->sdma_state.current_state !=
+			    qib_sdma_state_s20_idle)
+				__qib_sdma_process_event(ppd,
+					qib_sdma_event_e00_go_hw_down);
+			spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+		}
+	}
+
+	if (symadj) {
+		if (ppd->cpspec->ibdeltainprog) {
+			ppd->cpspec->ibdeltainprog = 0;
+			ppd->cpspec->ibsymdelta += read_7220_creg32(ppd->dd,
+				cr_ibsymbolerr) - ppd->cpspec->ibsymsnap;
+			ppd->cpspec->iblnkerrdelta += read_7220_creg32(ppd->dd,
+				cr_iblinkerrrecov) - ppd->cpspec->iblnkerrsnap;
+		}
+	} else if (!ibup && qib_compat_ddr_negotiate &&
+		   !ppd->cpspec->ibdeltainprog &&
+			!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+		ppd->cpspec->ibdeltainprog = 1;
+		ppd->cpspec->ibsymsnap = read_7220_creg32(ppd->dd,
+							  cr_ibsymbolerr);
+		ppd->cpspec->iblnkerrsnap = read_7220_creg32(ppd->dd,
+						     cr_iblinkerrrecov);
+	}
+
+	if (!ret)
+		qib_setup_7220_setextled(ppd, ibup);
+	return ret;
+}
+
+/*
+ * Does read/modify/write to appropriate registers to
+ * set output and direction bits selected by mask.
+ * these are in their canonical postions (e.g. lsb of
+ * dir will end up in D48 of extctrl on existing chips).
+ * returns contents of GP Inputs.
+ */
+static int gpio_7220_mod(struct qib_devdata *dd, u32 out, u32 dir, u32 mask)
+{
+	u64 read_val, new_out;
+	unsigned long flags;
+
+	if (mask) {
+		/* some bits being written, lock access to GPIO */
+		dir &= mask;
+		out &= mask;
+		spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+		dd->cspec->extctrl &= ~((u64)mask << SYM_LSB(EXTCtrl, GPIOOe));
+		dd->cspec->extctrl |= ((u64) dir << SYM_LSB(EXTCtrl, GPIOOe));
+		new_out = (dd->cspec->gpio_out & ~mask) | out;
+
+		qib_write_kreg(dd, kr_extctrl, dd->cspec->extctrl);
+		qib_write_kreg(dd, kr_gpio_out, new_out);
+		dd->cspec->gpio_out = new_out;
+		spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+	}
+	/*
+	 * It is unlikely that a read at this time would get valid
+	 * data on a pin whose direction line was set in the same
+	 * call to this function. We include the read here because
+	 * that allows us to potentially combine a change on one pin with
+	 * a read on another, and because the old code did something like
+	 * this.
+	 */
+	read_val = qib_read_kreg64(dd, kr_extstatus);
+	return SYM_FIELD(read_val, EXTStatus, GPIOIn);
+}
+
+/*
+ * Read fundamental info we need to use the chip.  These are
+ * the registers that describe chip capabilities, and are
+ * saved in shadow registers.
+ */
+static void get_7220_chip_params(struct qib_devdata *dd)
+{
+	u64 val;
+	u32 piobufs;
+	int mtu;
+
+	dd->uregbase = qib_read_kreg32(dd, kr_userregbase);
+
+	dd->rcvtidcnt = qib_read_kreg32(dd, kr_rcvtidcnt);
+	dd->rcvtidbase = qib_read_kreg32(dd, kr_rcvtidbase);
+	dd->rcvegrbase = qib_read_kreg32(dd, kr_rcvegrbase);
+	dd->palign = qib_read_kreg32(dd, kr_palign);
+	dd->piobufbase = qib_read_kreg64(dd, kr_sendpiobufbase);
+	dd->pio2k_bufbase = dd->piobufbase & 0xffffffff;
+
+	val = qib_read_kreg64(dd, kr_sendpiosize);
+	dd->piosize2k = val & ~0U;
+	dd->piosize4k = val >> 32;
+
+	mtu = ib_mtu_enum_to_int(qib_ibmtu);
+	if (mtu == -1)
+		mtu = QIB_DEFAULT_MTU;
+	dd->pport->ibmtu = (u32)mtu;
+
+	val = qib_read_kreg64(dd, kr_sendpiobufcnt);
+	dd->piobcnt2k = val & ~0U;
+	dd->piobcnt4k = val >> 32;
+	/* these may be adjusted in init_chip_wc_pat() */
+	dd->pio2kbase = (u32 __iomem *)
+		((char __iomem *) dd->kregbase + dd->pio2k_bufbase);
+	if (dd->piobcnt4k) {
+		dd->pio4kbase = (u32 __iomem *)
+			((char __iomem *) dd->kregbase +
+			 (dd->piobufbase >> 32));
+		/*
+		 * 4K buffers take 2 pages; we use roundup just to be
+		 * paranoid; we calculate it once here, rather than on
+		 * ever buf allocate
+		 */
+		dd->align4k = ALIGN(dd->piosize4k, dd->palign);
+	}
+
+	piobufs = dd->piobcnt4k + dd->piobcnt2k;
+
+	dd->pioavregs = ALIGN(piobufs, sizeof(u64) * BITS_PER_BYTE / 2) /
+		(sizeof(u64) * BITS_PER_BYTE / 2);
+}
+
+/*
+ * The chip base addresses in cspec and cpspec have to be set
+ * after possible init_chip_wc_pat(), rather than in
+ * qib_get_7220_chip_params(), so split out as separate function
+ */
+static void set_7220_baseaddrs(struct qib_devdata *dd)
+{
+	u32 cregbase;
+	/* init after possible re-map in init_chip_wc_pat() */
+	cregbase = qib_read_kreg32(dd, kr_counterregbase);
+	dd->cspec->cregbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase + cregbase);
+
+	dd->egrtidbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase + dd->rcvegrbase);
+}
+
+
+#define SENDCTRL_SHADOWED (SYM_MASK(SendCtrl, SendIntBufAvail) |	\
+			   SYM_MASK(SendCtrl, SPioEnable) |		\
+			   SYM_MASK(SendCtrl, SSpecialTriggerEn) |	\
+			   SYM_MASK(SendCtrl, SendBufAvailUpd) |	\
+			   SYM_MASK(SendCtrl, AvailUpdThld) |		\
+			   SYM_MASK(SendCtrl, SDmaEnable) |		\
+			   SYM_MASK(SendCtrl, SDmaIntEnable) |		\
+			   SYM_MASK(SendCtrl, SDmaHalt) |		\
+			   SYM_MASK(SendCtrl, SDmaSingleDescriptor))
+
+static int sendctrl_hook(struct qib_devdata *dd,
+			 const struct diag_observer *op,
+			 u32 offs, u64 *data, u64 mask, int only_32)
+{
+	unsigned long flags;
+	unsigned idx = offs / sizeof(u64);
+	u64 local_data, all_bits;
+
+	if (idx != kr_sendctrl) {
+		qib_dev_err(dd, "SendCtrl Hook called with offs %X, %s-bit\n",
+			    offs, only_32 ? "32" : "64");
+		return 0;
+	}
+
+	all_bits = ~0ULL;
+	if (only_32)
+		all_bits >>= 32;
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+	if ((mask & all_bits) != all_bits) {
+		/*
+		 * At least some mask bits are zero, so we need
+		 * to read. The judgement call is whether from
+		 * reg or shadow. First-cut: read reg, and complain
+		 * if any bits which should be shadowed are different
+		 * from their shadowed value.
+		 */
+		if (only_32)
+			local_data = (u64)qib_read_kreg32(dd, idx);
+		else
+			local_data = qib_read_kreg64(dd, idx);
+		qib_dev_err(dd, "Sendctrl -> %X, Shad -> %X\n",
+			    (u32)local_data, (u32)dd->sendctrl);
+		if ((local_data & SENDCTRL_SHADOWED) !=
+		    (dd->sendctrl & SENDCTRL_SHADOWED))
+			qib_dev_err(dd, "Sendctrl read: %X shadow is %X\n",
+				(u32)local_data, (u32) dd->sendctrl);
+		*data = (local_data & ~mask) | (*data & mask);
+	}
+	if (mask) {
+		/*
+		 * At least some mask bits are one, so we need
+		 * to write, but only shadow some bits.
+		 */
+		u64 sval, tval; /* Shadowed, transient */
+
+		/*
+		 * New shadow val is bits we don't want to touch,
+		 * ORed with bits we do, that are intended for shadow.
+		 */
+		sval = (dd->sendctrl & ~mask);
+		sval |= *data & SENDCTRL_SHADOWED & mask;
+		dd->sendctrl = sval;
+		tval = sval | (*data & ~SENDCTRL_SHADOWED & mask);
+		qib_dev_err(dd, "Sendctrl <- %X, Shad <- %X\n",
+			    (u32)tval, (u32)sval);
+		qib_write_kreg(dd, kr_sendctrl, tval);
+		qib_write_kreg(dd, kr_scratch, 0Ull);
+	}
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+
+	return only_32 ? 4 : 8;
+}
+
+static const struct diag_observer sendctrl_observer = {
+	sendctrl_hook, kr_sendctrl * sizeof(u64),
+	kr_sendctrl * sizeof(u64)
+};
+
+/*
+ * write the final few registers that depend on some of the
+ * init setup.  Done late in init, just before bringing up
+ * the serdes.
+ */
+static int qib_late_7220_initreg(struct qib_devdata *dd)
+{
+	int ret = 0;
+	u64 val;
+
+	qib_write_kreg(dd, kr_rcvhdrentsize, dd->rcvhdrentsize);
+	qib_write_kreg(dd, kr_rcvhdrsize, dd->rcvhdrsize);
+	qib_write_kreg(dd, kr_rcvhdrcnt, dd->rcvhdrcnt);
+	qib_write_kreg(dd, kr_sendpioavailaddr, dd->pioavailregs_phys);
+	val = qib_read_kreg64(dd, kr_sendpioavailaddr);
+	if (val != dd->pioavailregs_phys) {
+		qib_dev_err(dd,
+			"Catastrophic software error, SendPIOAvailAddr written as %lx, read back as %llx\n",
+			(unsigned long) dd->pioavailregs_phys,
+			(unsigned long long) val);
+		ret = -EINVAL;
+	}
+	qib_register_observer(dd, &sendctrl_observer);
+	return ret;
+}
+
+static int qib_init_7220_variables(struct qib_devdata *dd)
+{
+	struct qib_chippport_specific *cpspec;
+	struct qib_pportdata *ppd;
+	int ret = 0;
+	u32 sbufs, updthresh;
+
+	cpspec = (struct qib_chippport_specific *)(dd + 1);
+	ppd = &cpspec->pportdata;
+	dd->pport = ppd;
+	dd->num_pports = 1;
+
+	dd->cspec = (struct qib_chip_specific *)(cpspec + dd->num_pports);
+	ppd->cpspec = cpspec;
+
+	spin_lock_init(&dd->cspec->sdepb_lock);
+	spin_lock_init(&dd->cspec->rcvmod_lock);
+	spin_lock_init(&dd->cspec->gpio_lock);
+
+	/* we haven't yet set QIB_PRESENT, so use read directly */
+	dd->revision = readq(&dd->kregbase[kr_revision]);
+
+	if ((dd->revision & 0xffffffffU) == 0xffffffffU) {
+		qib_dev_err(dd,
+			"Revision register read failure, giving up initialization\n");
+		ret = -ENODEV;
+		goto bail;
+	}
+	dd->flags |= QIB_PRESENT;  /* now register routines work */
+
+	dd->majrev = (u8) SYM_FIELD(dd->revision, Revision_R,
+				    ChipRevMajor);
+	dd->minrev = (u8) SYM_FIELD(dd->revision, Revision_R,
+				    ChipRevMinor);
+
+	get_7220_chip_params(dd);
+	qib_7220_boardname(dd);
+
+	/*
+	 * GPIO bits for TWSI data and clock,
+	 * used for serial EEPROM.
+	 */
+	dd->gpio_sda_num = _QIB_GPIO_SDA_NUM;
+	dd->gpio_scl_num = _QIB_GPIO_SCL_NUM;
+	dd->twsi_eeprom_dev = QIB_TWSI_EEPROM_DEV;
+
+	dd->flags |= QIB_HAS_INTX | QIB_HAS_LINK_LATENCY |
+		QIB_NODMA_RTAIL | QIB_HAS_THRESH_UPDATE;
+	dd->flags |= qib_special_trigger ?
+		QIB_USE_SPCL_TRIG : QIB_HAS_SEND_DMA;
+
+	/*
+	 * EEPROM error log 0 is TXE Parity errors. 1 is RXE Parity.
+	 * 2 is Some Misc, 3 is reserved for future.
+	 */
+	dd->eep_st_masks[0].hwerrs_to_log = HWE_MASK(TXEMemParityErr);
+
+	dd->eep_st_masks[1].hwerrs_to_log = HWE_MASK(RXEMemParityErr);
+
+	dd->eep_st_masks[2].errs_to_log = ERR_MASK(ResetNegated);
+
+	init_waitqueue_head(&cpspec->autoneg_wait);
+	INIT_DELAYED_WORK(&cpspec->autoneg_work, autoneg_7220_work);
+
+	ret = qib_init_pportdata(ppd, dd, 0, 1);
+	if (ret)
+		goto bail;
+	ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
+	ppd->link_speed_supported = QIB_IB_SDR | QIB_IB_DDR;
+
+	ppd->link_width_enabled = ppd->link_width_supported;
+	ppd->link_speed_enabled = ppd->link_speed_supported;
+	/*
+	 * Set the initial values to reasonable default, will be set
+	 * for real when link is up.
+	 */
+	ppd->link_width_active = IB_WIDTH_4X;
+	ppd->link_speed_active = QIB_IB_SDR;
+	ppd->delay_mult = rate_to_delay[0][1];
+	ppd->vls_supported = IB_VL_VL0;
+	ppd->vls_operational = ppd->vls_supported;
+
+	if (!qib_mini_init)
+		qib_write_kreg(dd, kr_rcvbthqp, QIB_KD_QP);
+
+	init_timer(&ppd->cpspec->chase_timer);
+	ppd->cpspec->chase_timer.function = reenable_7220_chase;
+	ppd->cpspec->chase_timer.data = (unsigned long)ppd;
+
+	qib_num_cfg_vls = 1; /* if any 7220's, only one VL */
+
+	dd->rcvhdrentsize = QIB_RCVHDR_ENTSIZE;
+	dd->rcvhdrsize = QIB_DFLT_RCVHDRSIZE;
+	dd->rhf_offset =
+		dd->rcvhdrentsize - sizeof(u64) / sizeof(u32);
+
+	/* we always allocate at least 2048 bytes for eager buffers */
+	ret = ib_mtu_enum_to_int(qib_ibmtu);
+	dd->rcvegrbufsize = ret != -1 ? max(ret, 2048) : QIB_DEFAULT_MTU;
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
+
+	qib_7220_tidtemplate(dd);
+
+	/*
+	 * We can request a receive interrupt for 1 or
+	 * more packets from current offset.  For now, we set this
+	 * up for a single packet.
+	 */
+	dd->rhdrhead_intr_off = 1ULL << 32;
+
+	/* setup the stats timer; the add_timer is done at end of init */
+	init_timer(&dd->stats_timer);
+	dd->stats_timer.function = qib_get_7220_faststats;
+	dd->stats_timer.data = (unsigned long) dd;
+	dd->stats_timer.expires = jiffies + ACTIVITY_TIMER * HZ;
+
+	/*
+	 * Control[4] has been added to change the arbitration within
+	 * the SDMA engine between favoring data fetches over descriptor
+	 * fetches.  qib_sdma_fetch_arb==0 gives data fetches priority.
+	 */
+	if (qib_sdma_fetch_arb)
+		dd->control |= 1 << 4;
+
+	dd->ureg_align = 0x10000;  /* 64KB alignment */
+
+	dd->piosize2kmax_dwords = (dd->piosize2k >> 2)-1;
+	qib_7220_config_ctxts(dd);
+	qib_set_ctxtcnt(dd);  /* needed for PAT setup */
+
+	if (qib_wc_pat) {
+		ret = init_chip_wc_pat(dd, 0);
+		if (ret)
+			goto bail;
+	}
+	set_7220_baseaddrs(dd); /* set chip access pointers now */
+
+	ret = 0;
+	if (qib_mini_init)
+		goto bail;
+
+	ret = qib_create_ctxts(dd);
+	init_7220_cntrnames(dd);
+
+	/* use all of 4KB buffers for the kernel SDMA, zero if !SDMA.
+	 * reserve the update threshold amount for other kernel use, such
+	 * as sending SMI, MAD, and ACKs, or 3, whichever is greater,
+	 * unless we aren't enabling SDMA, in which case we want to use
+	 * all the 4k bufs for the kernel.
+	 * if this was less than the update threshold, we could wait
+	 * a long time for an update.  Coded this way because we
+	 * sometimes change the update threshold for various reasons,
+	 * and we want this to remain robust.
+	 */
+	updthresh = 8U; /* update threshold */
+	if (dd->flags & QIB_HAS_SEND_DMA) {
+		dd->cspec->sdmabufcnt =  dd->piobcnt4k;
+		sbufs = updthresh > 3 ? updthresh : 3;
+	} else {
+		dd->cspec->sdmabufcnt = 0;
+		sbufs = dd->piobcnt4k;
+	}
+
+	dd->cspec->lastbuf_for_pio = dd->piobcnt2k + dd->piobcnt4k -
+		dd->cspec->sdmabufcnt;
+	dd->lastctxt_piobuf = dd->cspec->lastbuf_for_pio - sbufs;
+	dd->cspec->lastbuf_for_pio--; /* range is <= , not < */
+	dd->last_pio = dd->cspec->lastbuf_for_pio;
+	dd->pbufsctxt = dd->lastctxt_piobuf /
+		(dd->cfgctxts - dd->first_user_ctxt);
+
+	/*
+	 * if we are at 16 user contexts, we will have one 7 sbufs
+	 * per context, so drop the update threshold to match.  We
+	 * want to update before we actually run out, at low pbufs/ctxt
+	 * so give ourselves some margin
+	 */
+	if ((dd->pbufsctxt - 2) < updthresh)
+		updthresh = dd->pbufsctxt - 2;
+
+	dd->cspec->updthresh_dflt = updthresh;
+	dd->cspec->updthresh = updthresh;
+
+	/* before full enable, no interrupts, no locking needed */
+	dd->sendctrl |= (updthresh & SYM_RMASK(SendCtrl, AvailUpdThld))
+			     << SYM_LSB(SendCtrl, AvailUpdThld);
+
+	dd->psxmitwait_supported = 1;
+	dd->psxmitwait_check_rate = QIB_7220_PSXMITWAIT_CHECK_RATE;
+bail:
+	return ret;
+}
+
+static u32 __iomem *qib_7220_getsendbuf(struct qib_pportdata *ppd, u64 pbc,
+					u32 *pbufnum)
+{
+	u32 first, last, plen = pbc & QIB_PBC_LENGTH_MASK;
+	struct qib_devdata *dd = ppd->dd;
+	u32 __iomem *buf;
+
+	if (((pbc >> 32) & PBC_7220_VL15_SEND_CTRL) &&
+		!(ppd->lflags & (QIBL_IB_AUTONEG_INPROG | QIBL_LINKACTIVE)))
+		buf = get_7220_link_buf(ppd, pbufnum);
+	else {
+		if ((plen + 1) > dd->piosize2kmax_dwords)
+			first = dd->piobcnt2k;
+		else
+			first = 0;
+		/* try 4k if all 2k busy, so same last for both sizes */
+		last = dd->cspec->lastbuf_for_pio;
+		buf = qib_getsendbuf_range(dd, pbufnum, first, last);
+	}
+	return buf;
+}
+
+/* these 2 "counters" are really control registers, and are always RW */
+static void qib_set_cntr_7220_sample(struct qib_pportdata *ppd, u32 intv,
+				     u32 start)
+{
+	write_7220_creg(ppd->dd, cr_psinterval, intv);
+	write_7220_creg(ppd->dd, cr_psstart, start);
+}
+
+/*
+ * NOTE: no real attempt is made to generalize the SDMA stuff.
+ * At some point "soon" we will have a new more generalized
+ * set of sdma interface, and then we'll clean this up.
+ */
+
+/* Must be called with sdma_lock held, or before init finished */
+static void qib_sdma_update_7220_tail(struct qib_pportdata *ppd, u16 tail)
+{
+	/* Commit writes to memory and advance the tail on the chip */
+	wmb();
+	ppd->sdma_descq_tail = tail;
+	qib_write_kreg(ppd->dd, kr_senddmatail, tail);
+}
+
+static void qib_sdma_set_7220_desc_cnt(struct qib_pportdata *ppd, unsigned cnt)
+{
+}
+
+static struct sdma_set_state_action sdma_7220_action_table[] = {
+	[qib_sdma_state_s00_hw_down] = {
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.go_s99_running_tofalse = 1,
+	},
+	[qib_sdma_state_s10_hw_start_up_wait] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+	},
+	[qib_sdma_state_s20_idle] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+	},
+	[qib_sdma_state_s30_sw_clean_up_wait] = {
+		.op_enable = 0,
+		.op_intenable = 1,
+		.op_halt = 0,
+	},
+	[qib_sdma_state_s40_hw_clean_up_wait] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+	},
+	[qib_sdma_state_s50_hw_halt_wait] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+	},
+	[qib_sdma_state_s99_running] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 0,
+		.go_s99_running_totrue = 1,
+	},
+};
+
+static void qib_7220_sdma_init_early(struct qib_pportdata *ppd)
+{
+	ppd->sdma_state.set_state_action = sdma_7220_action_table;
+}
+
+static int init_sdma_7220_regs(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned i, n;
+	u64 senddmabufmask[3] = { 0 };
+
+	/* Set SendDmaBase */
+	qib_write_kreg(dd, kr_senddmabase, ppd->sdma_descq_phys);
+	qib_sdma_7220_setlengen(ppd);
+	qib_sdma_update_7220_tail(ppd, 0); /* Set SendDmaTail */
+	/* Set SendDmaHeadAddr */
+	qib_write_kreg(dd, kr_senddmaheadaddr, ppd->sdma_head_phys);
+
+	/*
+	 * Reserve all the former "kernel" piobufs, using high number range
+	 * so we get as many 4K buffers as possible
+	 */
+	n = dd->piobcnt2k + dd->piobcnt4k;
+	i = n - dd->cspec->sdmabufcnt;
+
+	for (; i < n; ++i) {
+		unsigned word = i / 64;
+		unsigned bit = i & 63;
+
+		BUG_ON(word >= 3);
+		senddmabufmask[word] |= 1ULL << bit;
+	}
+	qib_write_kreg(dd, kr_senddmabufmask0, senddmabufmask[0]);
+	qib_write_kreg(dd, kr_senddmabufmask1, senddmabufmask[1]);
+	qib_write_kreg(dd, kr_senddmabufmask2, senddmabufmask[2]);
+
+	ppd->sdma_state.first_sendbuf = i;
+	ppd->sdma_state.last_sendbuf = n;
+
+	return 0;
+}
+
+/* sdma_lock must be held */
+static u16 qib_sdma_7220_gethead(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int sane;
+	int use_dmahead;
+	u16 swhead;
+	u16 swtail;
+	u16 cnt;
+	u16 hwhead;
+
+	use_dmahead = __qib_sdma_running(ppd) &&
+		(dd->flags & QIB_HAS_SDMA_TIMEOUT);
+retry:
+	hwhead = use_dmahead ?
+		(u16)le64_to_cpu(*ppd->sdma_head_dma) :
+		(u16)qib_read_kreg32(dd, kr_senddmahead);
+
+	swhead = ppd->sdma_descq_head;
+	swtail = ppd->sdma_descq_tail;
+	cnt = ppd->sdma_descq_cnt;
+
+	if (swhead < swtail) {
+		/* not wrapped */
+		sane = (hwhead >= swhead) & (hwhead <= swtail);
+	} else if (swhead > swtail) {
+		/* wrapped around */
+		sane = ((hwhead >= swhead) && (hwhead < cnt)) ||
+			(hwhead <= swtail);
+	} else {
+		/* empty */
+		sane = (hwhead == swhead);
+	}
+
+	if (unlikely(!sane)) {
+		if (use_dmahead) {
+			/* try one more time, directly from the register */
+			use_dmahead = 0;
+			goto retry;
+		}
+		/* assume no progress */
+		hwhead = swhead;
+	}
+
+	return hwhead;
+}
+
+static int qib_sdma_7220_busy(struct qib_pportdata *ppd)
+{
+	u64 hwstatus = qib_read_kreg64(ppd->dd, kr_senddmastatus);
+
+	return (hwstatus & SYM_MASK(SendDmaStatus, ScoreBoardDrainInProg)) ||
+	       (hwstatus & SYM_MASK(SendDmaStatus, AbortInProg)) ||
+	       (hwstatus & SYM_MASK(SendDmaStatus, InternalSDmaEnable)) ||
+	       !(hwstatus & SYM_MASK(SendDmaStatus, ScbEmpty));
+}
+
+/*
+ * Compute the amount of delay before sending the next packet if the
+ * port's send rate differs from the static rate set for the QP.
+ * Since the delay affects this packet but the amount of the delay is
+ * based on the length of the previous packet, use the last delay computed
+ * and save the delay count for this packet to be used next time
+ * we get here.
+ */
+static u32 qib_7220_setpbc_control(struct qib_pportdata *ppd, u32 plen,
+				   u8 srate, u8 vl)
+{
+	u8 snd_mult = ppd->delay_mult;
+	u8 rcv_mult = ib_rate_to_delay[srate];
+	u32 ret = ppd->cpspec->last_delay_mult;
+
+	ppd->cpspec->last_delay_mult = (rcv_mult > snd_mult) ?
+		(plen * (rcv_mult - snd_mult) + 1) >> 1 : 0;
+
+	/* Indicate VL15, if necessary */
+	if (vl == 15)
+		ret |= PBC_7220_VL15_SEND_CTRL;
+	return ret;
+}
+
+static void qib_7220_initvl15_bufs(struct qib_devdata *dd)
+{
+}
+
+static void qib_7220_init_ctxt(struct qib_ctxtdata *rcd)
+{
+	if (!rcd->ctxt) {
+		rcd->rcvegrcnt = IBA7220_KRCVEGRCNT;
+		rcd->rcvegr_tid_base = 0;
+	} else {
+		rcd->rcvegrcnt = rcd->dd->cspec->rcvegrcnt;
+		rcd->rcvegr_tid_base = IBA7220_KRCVEGRCNT +
+			(rcd->ctxt - 1) * rcd->rcvegrcnt;
+	}
+}
+
+static void qib_7220_txchk_change(struct qib_devdata *dd, u32 start,
+				  u32 len, u32 which, struct qib_ctxtdata *rcd)
+{
+	int i;
+	unsigned long flags;
+
+	switch (which) {
+	case TXCHK_CHG_TYPE_KERN:
+		/* see if we need to raise avail update threshold */
+		spin_lock_irqsave(&dd->uctxt_lock, flags);
+		for (i = dd->first_user_ctxt;
+		     dd->cspec->updthresh != dd->cspec->updthresh_dflt
+		     && i < dd->cfgctxts; i++)
+			if (dd->rcd[i] && dd->rcd[i]->subctxt_cnt &&
+			   ((dd->rcd[i]->piocnt / dd->rcd[i]->subctxt_cnt) - 1)
+			   < dd->cspec->updthresh_dflt)
+				break;
+		spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+		if (i == dd->cfgctxts) {
+			spin_lock_irqsave(&dd->sendctrl_lock, flags);
+			dd->cspec->updthresh = dd->cspec->updthresh_dflt;
+			dd->sendctrl &= ~SYM_MASK(SendCtrl, AvailUpdThld);
+			dd->sendctrl |= (dd->cspec->updthresh &
+					 SYM_RMASK(SendCtrl, AvailUpdThld)) <<
+					   SYM_LSB(SendCtrl, AvailUpdThld);
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+			sendctrl_7220_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+		}
+		break;
+	case TXCHK_CHG_TYPE_USER:
+		spin_lock_irqsave(&dd->sendctrl_lock, flags);
+		if (rcd && rcd->subctxt_cnt && ((rcd->piocnt
+			/ rcd->subctxt_cnt) - 1) < dd->cspec->updthresh) {
+			dd->cspec->updthresh = (rcd->piocnt /
+						rcd->subctxt_cnt) - 1;
+			dd->sendctrl &= ~SYM_MASK(SendCtrl, AvailUpdThld);
+			dd->sendctrl |= (dd->cspec->updthresh &
+					SYM_RMASK(SendCtrl, AvailUpdThld))
+					<< SYM_LSB(SendCtrl, AvailUpdThld);
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+			sendctrl_7220_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+		} else
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+		break;
+	}
+}
+
+static void writescratch(struct qib_devdata *dd, u32 val)
+{
+	qib_write_kreg(dd, kr_scratch, val);
+}
+
+#define VALID_TS_RD_REG_MASK 0xBF
+/**
+ * qib_7220_tempsense_read - read register of temp sensor via TWSI
+ * @dd: the qlogic_ib device
+ * @regnum: register to read from
+ *
+ * returns reg contents (0..255) or < 0 for error
+ */
+static int qib_7220_tempsense_rd(struct qib_devdata *dd, int regnum)
+{
+	int ret;
+	u8 rdata;
+
+	if (regnum > 7) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/* return a bogus value for (the one) register we do not have */
+	if (!((1 << regnum) & VALID_TS_RD_REG_MASK)) {
+		ret = 0;
+		goto bail;
+	}
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto bail;
+
+	ret = qib_twsi_blk_rd(dd, QIB_TWSI_TEMP_DEV, regnum, &rdata, 1);
+	if (!ret)
+		ret = rdata;
+
+	mutex_unlock(&dd->eep_lock);
+
+	/*
+	 * There are three possibilities here:
+	 * ret is actual value (0..255)
+	 * ret is -ENXIO or -EINVAL from twsi code or this file
+	 * ret is -EINTR from mutex_lock_interruptible.
+	 */
+bail:
+	return ret;
+}
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+static int qib_7220_notify_dca(struct qib_devdata *dd, unsigned long event)
+{
+	return 0;
+}
+#endif
+
+/* Dummy function, as 7220 boards never disable EEPROM Write */
+static int qib_7220_eeprom_wen(struct qib_devdata *dd, int wen)
+{
+	return 1;
+}
+
+/**
+ * qib_init_iba7220_funcs - set up the chip-specific function pointers
+ * @dev: the pci_dev for qlogic_ib device
+ * @ent: pci_device_id struct for this dev
+ *
+ * This is global, and is called directly at init to set up the
+ * chip-specific function pointers for later use.
+ */
+struct qib_devdata *qib_init_iba7220_funcs(struct pci_dev *pdev,
+					   const struct pci_device_id *ent)
+{
+	struct qib_devdata *dd;
+	int ret;
+	u32 boardid, minwidth;
+
+	dd = qib_alloc_devdata(pdev, sizeof(struct qib_chip_specific) +
+		sizeof(struct qib_chippport_specific));
+	if (IS_ERR(dd))
+		goto bail;
+
+	dd->f_bringup_serdes    = qib_7220_bringup_serdes;
+	dd->f_cleanup           = qib_setup_7220_cleanup;
+	dd->f_clear_tids        = qib_7220_clear_tids;
+	dd->f_free_irq          = qib_7220_free_irq;
+	dd->f_get_base_info     = qib_7220_get_base_info;
+	dd->f_get_msgheader     = qib_7220_get_msgheader;
+	dd->f_getsendbuf        = qib_7220_getsendbuf;
+	dd->f_gpio_mod          = gpio_7220_mod;
+	dd->f_eeprom_wen        = qib_7220_eeprom_wen;
+	dd->f_hdrqempty         = qib_7220_hdrqempty;
+	dd->f_ib_updown         = qib_7220_ib_updown;
+	dd->f_init_ctxt         = qib_7220_init_ctxt;
+	dd->f_initvl15_bufs     = qib_7220_initvl15_bufs;
+	dd->f_intr_fallback     = qib_7220_intr_fallback;
+	dd->f_late_initreg      = qib_late_7220_initreg;
+	dd->f_setpbc_control    = qib_7220_setpbc_control;
+	dd->f_portcntr          = qib_portcntr_7220;
+	dd->f_put_tid           = qib_7220_put_tid;
+	dd->f_quiet_serdes      = qib_7220_quiet_serdes;
+	dd->f_rcvctrl           = rcvctrl_7220_mod;
+	dd->f_read_cntrs        = qib_read_7220cntrs;
+	dd->f_read_portcntrs    = qib_read_7220portcntrs;
+	dd->f_reset             = qib_setup_7220_reset;
+	dd->f_init_sdma_regs    = init_sdma_7220_regs;
+	dd->f_sdma_busy         = qib_sdma_7220_busy;
+	dd->f_sdma_gethead      = qib_sdma_7220_gethead;
+	dd->f_sdma_sendctrl     = qib_7220_sdma_sendctrl;
+	dd->f_sdma_set_desc_cnt = qib_sdma_set_7220_desc_cnt;
+	dd->f_sdma_update_tail  = qib_sdma_update_7220_tail;
+	dd->f_sdma_hw_clean_up  = qib_7220_sdma_hw_clean_up;
+	dd->f_sdma_hw_start_up  = qib_7220_sdma_hw_start_up;
+	dd->f_sdma_init_early   = qib_7220_sdma_init_early;
+	dd->f_sendctrl          = sendctrl_7220_mod;
+	dd->f_set_armlaunch     = qib_set_7220_armlaunch;
+	dd->f_set_cntr_sample   = qib_set_cntr_7220_sample;
+	dd->f_iblink_state      = qib_7220_iblink_state;
+	dd->f_ibphys_portstate  = qib_7220_phys_portstate;
+	dd->f_get_ib_cfg        = qib_7220_get_ib_cfg;
+	dd->f_set_ib_cfg        = qib_7220_set_ib_cfg;
+	dd->f_set_ib_loopback   = qib_7220_set_loopback;
+	dd->f_set_intr_state    = qib_7220_set_intr_state;
+	dd->f_setextled         = qib_setup_7220_setextled;
+	dd->f_txchk_change      = qib_7220_txchk_change;
+	dd->f_update_usrhead    = qib_update_7220_usrhead;
+	dd->f_wantpiobuf_intr   = qib_wantpiobuf_7220_intr;
+	dd->f_xgxs_reset        = qib_7220_xgxs_reset;
+	dd->f_writescratch      = writescratch;
+	dd->f_tempsense_rd	= qib_7220_tempsense_rd;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dd->f_notify_dca = qib_7220_notify_dca;
+#endif
+	/*
+	 * Do remaining pcie setup and save pcie values in dd.
+	 * Any error printing is already done by the init code.
+	 * On return, we have the chip mapped, but chip registers
+	 * are not set up until start of qib_init_7220_variables.
+	 */
+	ret = qib_pcie_ddinit(dd, pdev, ent);
+	if (ret < 0)
+		goto bail_free;
+
+	/* initialize chip-specific variables */
+	ret = qib_init_7220_variables(dd);
+	if (ret)
+		goto bail_cleanup;
+
+	if (qib_mini_init)
+		goto bail;
+
+	boardid = SYM_FIELD(dd->revision, Revision,
+			    BoardID);
+	switch (boardid) {
+	case 0:
+	case 2:
+	case 10:
+	case 12:
+		minwidth = 16; /* x16 capable boards */
+		break;
+	default:
+		minwidth = 8; /* x8 capable boards */
+		break;
+	}
+	if (qib_pcie_params(dd, minwidth, NULL, NULL))
+		qib_dev_err(dd,
+			"Failed to setup PCIe or interrupts; continuing anyway\n");
+
+	/* save IRQ for possible later use */
+	dd->cspec->irq = pdev->irq;
+
+	if (qib_read_kreg64(dd, kr_hwerrstatus) &
+	    QLOGIC_IB_HWE_SERDESPLLFAILED)
+		qib_write_kreg(dd, kr_hwerrclear,
+			       QLOGIC_IB_HWE_SERDESPLLFAILED);
+
+	/* setup interrupt handler (interrupt type handled above) */
+	qib_setup_7220_interrupt(dd);
+	qib_7220_init_hwerrors(dd);
+
+	/* clear diagctrl register, in case diags were running and crashed */
+	qib_write_kreg(dd, kr_hwdiagctrl, 0);
+
+	goto bail;
+
+bail_cleanup:
+	qib_pcie_ddcleanup(dd);
+bail_free:
+	qib_free_devdata(dd);
+	dd = ERR_PTR(ret);
+bail:
+	return dd;
+}
diff --git a/drivers/infiniband/hw/qib/qib_iba7322.c b/drivers/infiniband/hw/qib/qib_iba7322.c
new file mode 100644
index 00000000000..a7eb32517a0
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_iba7322.c
@@ -0,0 +1,8554 @@
+/*
+ * Copyright (c) 2012 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2008 - 2012 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file contains all of the code that is specific to the
+ * InfiniPath 7322 chip
+ */
+
+#include <linux/interrupt.h>
+#include <linux/pci.h>
+#include <linux/delay.h>
+#include <linux/io.h>
+#include <linux/jiffies.h>
+#include <linux/module.h>
+#include <rdma/ib_verbs.h>
+#include <rdma/ib_smi.h>
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+#include <linux/dca.h>
+#endif
+
+#include "qib.h"
+#include "qib_7322_regs.h"
+#include "qib_qsfp.h"
+
+#include "qib_mad.h"
+#include "qib_verbs.h"
+
+#undef pr_fmt
+#define pr_fmt(fmt) QIB_DRV_NAME " " fmt
+
+static void qib_setup_7322_setextled(struct qib_pportdata *, u32);
+static void qib_7322_handle_hwerrors(struct qib_devdata *, char *, size_t);
+static void sendctrl_7322_mod(struct qib_pportdata *ppd, u32 op);
+static irqreturn_t qib_7322intr(int irq, void *data);
+static irqreturn_t qib_7322bufavail(int irq, void *data);
+static irqreturn_t sdma_intr(int irq, void *data);
+static irqreturn_t sdma_idle_intr(int irq, void *data);
+static irqreturn_t sdma_progress_intr(int irq, void *data);
+static irqreturn_t sdma_cleanup_intr(int irq, void *data);
+static void qib_7322_txchk_change(struct qib_devdata *, u32, u32, u32,
+				  struct qib_ctxtdata *rcd);
+static u8 qib_7322_phys_portstate(u64);
+static u32 qib_7322_iblink_state(u64);
+static void qib_set_ib_7322_lstate(struct qib_pportdata *ppd, u16 linkcmd,
+				   u16 linitcmd);
+static void force_h1(struct qib_pportdata *);
+static void adj_tx_serdes(struct qib_pportdata *);
+static u32 qib_7322_setpbc_control(struct qib_pportdata *, u32, u8, u8);
+static void qib_7322_mini_pcs_reset(struct qib_pportdata *);
+
+static u32 ahb_mod(struct qib_devdata *, int, int, int, u32, u32);
+static void ibsd_wr_allchans(struct qib_pportdata *, int, unsigned, unsigned);
+static void serdes_7322_los_enable(struct qib_pportdata *, int);
+static int serdes_7322_init_old(struct qib_pportdata *);
+static int serdes_7322_init_new(struct qib_pportdata *);
+static void dump_sdma_7322_state(struct qib_pportdata *);
+
+#define BMASK(msb, lsb) (((1 << ((msb) + 1 - (lsb))) - 1) << (lsb))
+
+/* LE2 serdes values for different cases */
+#define LE2_DEFAULT 5
+#define LE2_5m 4
+#define LE2_QME 0
+
+/* Below is special-purpose, so only really works for the IB SerDes blocks. */
+#define IBSD(hw_pidx) (hw_pidx + 2)
+
+/* these are variables for documentation and experimentation purposes */
+static const unsigned rcv_int_timeout = 375;
+static const unsigned rcv_int_count = 16;
+static const unsigned sdma_idle_cnt = 64;
+
+/* Time to stop altering Rx Equalization parameters, after link up. */
+#define RXEQ_DISABLE_MSECS 2500
+
+/*
+ * Number of VLs we are configured to use (to allow for more
+ * credits per vl, etc.)
+ */
+ushort qib_num_cfg_vls = 2;
+module_param_named(num_vls, qib_num_cfg_vls, ushort, S_IRUGO);
+MODULE_PARM_DESC(num_vls, "Set number of Virtual Lanes to use (1-8)");
+
+static ushort qib_chase = 1;
+module_param_named(chase, qib_chase, ushort, S_IRUGO);
+MODULE_PARM_DESC(chase, "Enable state chase handling");
+
+static ushort qib_long_atten = 10; /* 10 dB ~= 5m length */
+module_param_named(long_attenuation, qib_long_atten, ushort, S_IRUGO);
+MODULE_PARM_DESC(long_attenuation, \
+		 "attenuation cutoff (dB) for long copper cable setup");
+
+static ushort qib_singleport;
+module_param_named(singleport, qib_singleport, ushort, S_IRUGO);
+MODULE_PARM_DESC(singleport, "Use only IB port 1; more per-port buffer space");
+
+static ushort qib_krcvq01_no_msi;
+module_param_named(krcvq01_no_msi, qib_krcvq01_no_msi, ushort, S_IRUGO);
+MODULE_PARM_DESC(krcvq01_no_msi, "No MSI for kctx < 2");
+
+/*
+ * Receive header queue sizes
+ */
+static unsigned qib_rcvhdrcnt;
+module_param_named(rcvhdrcnt, qib_rcvhdrcnt, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrcnt, "receive header count");
+
+static unsigned qib_rcvhdrsize;
+module_param_named(rcvhdrsize, qib_rcvhdrsize, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrsize, "receive header size in 32-bit words");
+
+static unsigned qib_rcvhdrentsize;
+module_param_named(rcvhdrentsize, qib_rcvhdrentsize, uint, S_IRUGO);
+MODULE_PARM_DESC(rcvhdrentsize, "receive header entry size in 32-bit words");
+
+#define MAX_ATTEN_LEN 64 /* plenty for any real system */
+/* for read back, default index is ~5m copper cable */
+static char txselect_list[MAX_ATTEN_LEN] = "10";
+static struct kparam_string kp_txselect = {
+	.string = txselect_list,
+	.maxlen = MAX_ATTEN_LEN
+};
+static int  setup_txselect(const char *, struct kernel_param *);
+module_param_call(txselect, setup_txselect, param_get_string,
+		  &kp_txselect, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(txselect, \
+		 "Tx serdes indices (for no QSFP or invalid QSFP data)");
+
+#define BOARD_QME7342 5
+#define BOARD_QMH7342 6
+#define IS_QMH(dd) (SYM_FIELD((dd)->revision, Revision, BoardID) == \
+		    BOARD_QMH7342)
+#define IS_QME(dd) (SYM_FIELD((dd)->revision, Revision, BoardID) == \
+		    BOARD_QME7342)
+
+#define KREG_IDX(regname)     (QIB_7322_##regname##_OFFS / sizeof(u64))
+
+#define KREG_IBPORT_IDX(regname) ((QIB_7322_##regname##_0_OFFS / sizeof(u64)))
+
+#define MASK_ACROSS(lsb, msb) \
+	(((1ULL << ((msb) + 1 - (lsb))) - 1) << (lsb))
+
+#define SYM_RMASK(regname, fldname) ((u64)              \
+	QIB_7322_##regname##_##fldname##_RMASK)
+
+#define SYM_MASK(regname, fldname) ((u64)               \
+	QIB_7322_##regname##_##fldname##_RMASK <<       \
+	 QIB_7322_##regname##_##fldname##_LSB)
+
+#define SYM_FIELD(value, regname, fldname) ((u64)	\
+	(((value) >> SYM_LSB(regname, fldname)) &	\
+	 SYM_RMASK(regname, fldname)))
+
+/* useful for things like LaFifoEmpty_0...7, TxCreditOK_0...7, etc. */
+#define SYM_FIELD_ACROSS(value, regname, fldname, nbits) \
+	(((value) >> SYM_LSB(regname, fldname)) & MASK_ACROSS(0, nbits))
+
+#define HWE_MASK(fldname) SYM_MASK(HwErrMask, fldname##Mask)
+#define ERR_MASK(fldname) SYM_MASK(ErrMask, fldname##Mask)
+#define ERR_MASK_N(fldname) SYM_MASK(ErrMask_0, fldname##Mask)
+#define INT_MASK(fldname) SYM_MASK(IntMask, fldname##IntMask)
+#define INT_MASK_P(fldname, port) SYM_MASK(IntMask, fldname##IntMask##_##port)
+/* Below because most, but not all, fields of IntMask have that full suffix */
+#define INT_MASK_PM(fldname, port) SYM_MASK(IntMask, fldname##Mask##_##port)
+
+
+#define SYM_LSB(regname, fldname) (QIB_7322_##regname##_##fldname##_LSB)
+
+/*
+ * the size bits give us 2^N, in KB units.  0 marks as invalid,
+ * and 7 is reserved.  We currently use only 2KB and 4KB
+ */
+#define IBA7322_TID_SZ_SHIFT QIB_7322_RcvTIDArray0_RT_BufSize_LSB
+#define IBA7322_TID_SZ_2K (1UL<<IBA7322_TID_SZ_SHIFT) /* 2KB */
+#define IBA7322_TID_SZ_4K (2UL<<IBA7322_TID_SZ_SHIFT) /* 4KB */
+#define IBA7322_TID_PA_SHIFT 11U /* TID addr in chip stored w/o low bits */
+
+#define SendIBSLIDAssignMask \
+	QIB_7322_SendIBSLIDAssign_0_SendIBSLIDAssign_15_0_RMASK
+#define SendIBSLMCMask \
+	QIB_7322_SendIBSLIDMask_0_SendIBSLIDMask_15_0_RMASK
+
+#define ExtLED_IB1_YEL SYM_MASK(EXTCtrl, LEDPort0YellowOn)
+#define ExtLED_IB1_GRN SYM_MASK(EXTCtrl, LEDPort0GreenOn)
+#define ExtLED_IB2_YEL SYM_MASK(EXTCtrl, LEDPort1YellowOn)
+#define ExtLED_IB2_GRN SYM_MASK(EXTCtrl, LEDPort1GreenOn)
+#define ExtLED_IB1_MASK (ExtLED_IB1_YEL | ExtLED_IB1_GRN)
+#define ExtLED_IB2_MASK (ExtLED_IB2_YEL | ExtLED_IB2_GRN)
+
+#define _QIB_GPIO_SDA_NUM 1
+#define _QIB_GPIO_SCL_NUM 0
+#define QIB_EEPROM_WEN_NUM 14
+#define QIB_TWSI_EEPROM_DEV 0xA2 /* All Production 7322 cards. */
+
+/* HW counter clock is at 4nsec */
+#define QIB_7322_PSXMITWAIT_CHECK_RATE 4000
+
+/* full speed IB port 1 only */
+#define PORT_SPD_CAP (QIB_IB_SDR | QIB_IB_DDR | QIB_IB_QDR)
+#define PORT_SPD_CAP_SHIFT 3
+
+/* full speed featuremask, both ports */
+#define DUAL_PORT_CAP (PORT_SPD_CAP | (PORT_SPD_CAP << PORT_SPD_CAP_SHIFT))
+
+/*
+ * This file contains almost all the chip-specific register information and
+ * access functions for the FAKED QLogic InfiniPath 7322 PCI-Express chip.
+ */
+
+/* Use defines to tie machine-generated names to lower-case names */
+#define kr_contextcnt KREG_IDX(ContextCnt)
+#define kr_control KREG_IDX(Control)
+#define kr_counterregbase KREG_IDX(CntrRegBase)
+#define kr_errclear KREG_IDX(ErrClear)
+#define kr_errmask KREG_IDX(ErrMask)
+#define kr_errstatus KREG_IDX(ErrStatus)
+#define kr_extctrl KREG_IDX(EXTCtrl)
+#define kr_extstatus KREG_IDX(EXTStatus)
+#define kr_gpio_clear KREG_IDX(GPIOClear)
+#define kr_gpio_mask KREG_IDX(GPIOMask)
+#define kr_gpio_out KREG_IDX(GPIOOut)
+#define kr_gpio_status KREG_IDX(GPIOStatus)
+#define kr_hwdiagctrl KREG_IDX(HwDiagCtrl)
+#define kr_debugportval KREG_IDX(DebugPortValueReg)
+#define kr_fmask KREG_IDX(feature_mask)
+#define kr_act_fmask KREG_IDX(active_feature_mask)
+#define kr_hwerrclear KREG_IDX(HwErrClear)
+#define kr_hwerrmask KREG_IDX(HwErrMask)
+#define kr_hwerrstatus KREG_IDX(HwErrStatus)
+#define kr_intclear KREG_IDX(IntClear)
+#define kr_intmask KREG_IDX(IntMask)
+#define kr_intredirect KREG_IDX(IntRedirect0)
+#define kr_intstatus KREG_IDX(IntStatus)
+#define kr_pagealign KREG_IDX(PageAlign)
+#define kr_rcvavailtimeout KREG_IDX(RcvAvailTimeOut0)
+#define kr_rcvctrl KREG_IDX(RcvCtrl) /* Common, but chip also has per-port */
+#define kr_rcvegrbase KREG_IDX(RcvEgrBase)
+#define kr_rcvegrcnt KREG_IDX(RcvEgrCnt)
+#define kr_rcvhdrcnt KREG_IDX(RcvHdrCnt)
+#define kr_rcvhdrentsize KREG_IDX(RcvHdrEntSize)
+#define kr_rcvhdrsize KREG_IDX(RcvHdrSize)
+#define kr_rcvtidbase KREG_IDX(RcvTIDBase)
+#define kr_rcvtidcnt KREG_IDX(RcvTIDCnt)
+#define kr_revision KREG_IDX(Revision)
+#define kr_scratch KREG_IDX(Scratch)
+#define kr_sendbuffererror KREG_IDX(SendBufErr0) /* and base for 1 and 2 */
+#define kr_sendcheckmask KREG_IDX(SendCheckMask0) /* and 1, 2 */
+#define kr_sendctrl KREG_IDX(SendCtrl)
+#define kr_sendgrhcheckmask KREG_IDX(SendGRHCheckMask0) /* and 1, 2 */
+#define kr_sendibpktmask KREG_IDX(SendIBPacketMask0) /* and 1, 2 */
+#define kr_sendpioavailaddr KREG_IDX(SendBufAvailAddr)
+#define kr_sendpiobufbase KREG_IDX(SendBufBase)
+#define kr_sendpiobufcnt KREG_IDX(SendBufCnt)
+#define kr_sendpiosize KREG_IDX(SendBufSize)
+#define kr_sendregbase KREG_IDX(SendRegBase)
+#define kr_sendbufavail0 KREG_IDX(SendBufAvail0)
+#define kr_userregbase KREG_IDX(UserRegBase)
+#define kr_intgranted KREG_IDX(Int_Granted)
+#define kr_vecclr_wo_int KREG_IDX(vec_clr_without_int)
+#define kr_intblocked KREG_IDX(IntBlocked)
+#define kr_r_access KREG_IDX(SPC_JTAG_ACCESS_REG)
+
+/*
+ * per-port kernel registers.  Access only with qib_read_kreg_port()
+ * or qib_write_kreg_port()
+ */
+#define krp_errclear KREG_IBPORT_IDX(ErrClear)
+#define krp_errmask KREG_IBPORT_IDX(ErrMask)
+#define krp_errstatus KREG_IBPORT_IDX(ErrStatus)
+#define krp_highprio_0 KREG_IBPORT_IDX(HighPriority0)
+#define krp_highprio_limit KREG_IBPORT_IDX(HighPriorityLimit)
+#define krp_hrtbt_guid KREG_IBPORT_IDX(HRTBT_GUID)
+#define krp_ib_pcsconfig KREG_IBPORT_IDX(IBPCSConfig)
+#define krp_ibcctrl_a KREG_IBPORT_IDX(IBCCtrlA)
+#define krp_ibcctrl_b KREG_IBPORT_IDX(IBCCtrlB)
+#define krp_ibcctrl_c KREG_IBPORT_IDX(IBCCtrlC)
+#define krp_ibcstatus_a KREG_IBPORT_IDX(IBCStatusA)
+#define krp_ibcstatus_b KREG_IBPORT_IDX(IBCStatusB)
+#define krp_txestatus KREG_IBPORT_IDX(TXEStatus)
+#define krp_lowprio_0 KREG_IBPORT_IDX(LowPriority0)
+#define krp_ncmodectrl KREG_IBPORT_IDX(IBNCModeCtrl)
+#define krp_partitionkey KREG_IBPORT_IDX(RcvPartitionKey)
+#define krp_psinterval KREG_IBPORT_IDX(PSInterval)
+#define krp_psstart KREG_IBPORT_IDX(PSStart)
+#define krp_psstat KREG_IBPORT_IDX(PSStat)
+#define krp_rcvbthqp KREG_IBPORT_IDX(RcvBTHQP)
+#define krp_rcvctrl KREG_IBPORT_IDX(RcvCtrl)
+#define krp_rcvpktledcnt KREG_IBPORT_IDX(RcvPktLEDCnt)
+#define krp_rcvqpmaptable KREG_IBPORT_IDX(RcvQPMapTableA)
+#define krp_rxcreditvl0 KREG_IBPORT_IDX(RxCreditVL0)
+#define krp_rxcreditvl15 (KREG_IBPORT_IDX(RxCreditVL0)+15)
+#define krp_sendcheckcontrol KREG_IBPORT_IDX(SendCheckControl)
+#define krp_sendctrl KREG_IBPORT_IDX(SendCtrl)
+#define krp_senddmabase KREG_IBPORT_IDX(SendDmaBase)
+#define krp_senddmabufmask0 KREG_IBPORT_IDX(SendDmaBufMask0)
+#define krp_senddmabufmask1 (KREG_IBPORT_IDX(SendDmaBufMask0) + 1)
+#define krp_senddmabufmask2 (KREG_IBPORT_IDX(SendDmaBufMask0) + 2)
+#define krp_senddmabuf_use0 KREG_IBPORT_IDX(SendDmaBufUsed0)
+#define krp_senddmabuf_use1 (KREG_IBPORT_IDX(SendDmaBufUsed0) + 1)
+#define krp_senddmabuf_use2 (KREG_IBPORT_IDX(SendDmaBufUsed0) + 2)
+#define krp_senddmadesccnt KREG_IBPORT_IDX(SendDmaDescCnt)
+#define krp_senddmahead KREG_IBPORT_IDX(SendDmaHead)
+#define krp_senddmaheadaddr KREG_IBPORT_IDX(SendDmaHeadAddr)
+#define krp_senddmaidlecnt KREG_IBPORT_IDX(SendDmaIdleCnt)
+#define krp_senddmalengen KREG_IBPORT_IDX(SendDmaLenGen)
+#define krp_senddmaprioritythld KREG_IBPORT_IDX(SendDmaPriorityThld)
+#define krp_senddmareloadcnt KREG_IBPORT_IDX(SendDmaReloadCnt)
+#define krp_senddmastatus KREG_IBPORT_IDX(SendDmaStatus)
+#define krp_senddmatail KREG_IBPORT_IDX(SendDmaTail)
+#define krp_sendhdrsymptom KREG_IBPORT_IDX(SendHdrErrSymptom)
+#define krp_sendslid KREG_IBPORT_IDX(SendIBSLIDAssign)
+#define krp_sendslidmask KREG_IBPORT_IDX(SendIBSLIDMask)
+#define krp_ibsdtestiftx KREG_IBPORT_IDX(IB_SDTEST_IF_TX)
+#define krp_adapt_dis_timer KREG_IBPORT_IDX(ADAPT_DISABLE_TIMER_THRESHOLD)
+#define krp_tx_deemph_override KREG_IBPORT_IDX(IBSD_TX_DEEMPHASIS_OVERRIDE)
+#define krp_serdesctrl KREG_IBPORT_IDX(IBSerdesCtrl)
+
+/*
+ * Per-context kernel registers.  Access only with qib_read_kreg_ctxt()
+ * or qib_write_kreg_ctxt()
+ */
+#define krc_rcvhdraddr KREG_IDX(RcvHdrAddr0)
+#define krc_rcvhdrtailaddr KREG_IDX(RcvHdrTailAddr0)
+
+/*
+ * TID Flow table, per context.  Reduces
+ * number of hdrq updates to one per flow (or on errors).
+ * context 0 and 1 share same memory, but have distinct
+ * addresses.  Since for now, we never use expected sends
+ * on kernel contexts, we don't worry about that (we initialize
+ * those entries for ctxt 0/1 on driver load twice, for example).
+ */
+#define NUM_TIDFLOWS_CTXT 0x20 /* 0x20 per context; have to hardcode */
+#define ur_rcvflowtable (KREG_IDX(RcvTIDFlowTable0) - KREG_IDX(RcvHdrTail0))
+
+/* these are the error bits in the tid flows, and are W1C */
+#define TIDFLOW_ERRBITS  ( \
+	(SYM_MASK(RcvTIDFlowTable0, GenMismatch) << \
+	SYM_LSB(RcvTIDFlowTable0, GenMismatch)) | \
+	(SYM_MASK(RcvTIDFlowTable0, SeqMismatch) << \
+	SYM_LSB(RcvTIDFlowTable0, SeqMismatch)))
+
+/* Most (not all) Counters are per-IBport.
+ * Requires LBIntCnt is at offset 0 in the group
+ */
+#define CREG_IDX(regname) \
+((QIB_7322_##regname##_0_OFFS - QIB_7322_LBIntCnt_OFFS) / sizeof(u64))
+
+#define crp_badformat CREG_IDX(RxVersionErrCnt)
+#define crp_err_rlen CREG_IDX(RxLenErrCnt)
+#define crp_erricrc CREG_IDX(RxICRCErrCnt)
+#define crp_errlink CREG_IDX(RxLinkMalformCnt)
+#define crp_errlpcrc CREG_IDX(RxLPCRCErrCnt)
+#define crp_errpkey CREG_IDX(RxPKeyMismatchCnt)
+#define crp_errvcrc CREG_IDX(RxVCRCErrCnt)
+#define crp_excessbufferovfl CREG_IDX(ExcessBufferOvflCnt)
+#define crp_iblinkdown CREG_IDX(IBLinkDownedCnt)
+#define crp_iblinkerrrecov CREG_IDX(IBLinkErrRecoveryCnt)
+#define crp_ibstatuschange CREG_IDX(IBStatusChangeCnt)
+#define crp_ibsymbolerr CREG_IDX(IBSymbolErrCnt)
+#define crp_invalidrlen CREG_IDX(RxMaxMinLenErrCnt)
+#define crp_locallinkintegrityerr CREG_IDX(LocalLinkIntegrityErrCnt)
+#define crp_pktrcv CREG_IDX(RxDataPktCnt)
+#define crp_pktrcvflowctrl CREG_IDX(RxFlowPktCnt)
+#define crp_pktsend CREG_IDX(TxDataPktCnt)
+#define crp_pktsendflow CREG_IDX(TxFlowPktCnt)
+#define crp_psrcvdatacount CREG_IDX(PSRcvDataCount)
+#define crp_psrcvpktscount CREG_IDX(PSRcvPktsCount)
+#define crp_psxmitdatacount CREG_IDX(PSXmitDataCount)
+#define crp_psxmitpktscount CREG_IDX(PSXmitPktsCount)
+#define crp_psxmitwaitcount CREG_IDX(PSXmitWaitCount)
+#define crp_rcvebp CREG_IDX(RxEBPCnt)
+#define crp_rcvflowctrlviol CREG_IDX(RxFlowCtrlViolCnt)
+#define crp_rcvovfl CREG_IDX(RxBufOvflCnt)
+#define crp_rxdlidfltr CREG_IDX(RxDlidFltrCnt)
+#define crp_rxdroppkt CREG_IDX(RxDroppedPktCnt)
+#define crp_rxotherlocalphyerr CREG_IDX(RxOtherLocalPhyErrCnt)
+#define crp_rxqpinvalidctxt CREG_IDX(RxQPInvalidContextCnt)
+#define crp_rxvlerr CREG_IDX(RxVlErrCnt)
+#define crp_sendstall CREG_IDX(TxFlowStallCnt)
+#define crp_txdroppedpkt CREG_IDX(TxDroppedPktCnt)
+#define crp_txhdrerr CREG_IDX(TxHeadersErrCnt)
+#define crp_txlenerr CREG_IDX(TxLenErrCnt)
+#define crp_txminmaxlenerr CREG_IDX(TxMaxMinLenErrCnt)
+#define crp_txsdmadesc CREG_IDX(TxSDmaDescCnt)
+#define crp_txunderrun CREG_IDX(TxUnderrunCnt)
+#define crp_txunsupvl CREG_IDX(TxUnsupVLErrCnt)
+#define crp_vl15droppedpkt CREG_IDX(RxVL15DroppedPktCnt)
+#define crp_wordrcv CREG_IDX(RxDwordCnt)
+#define crp_wordsend CREG_IDX(TxDwordCnt)
+#define crp_tx_creditstalls CREG_IDX(TxCreditUpToDateTimeOut)
+
+/* these are the (few) counters that are not port-specific */
+#define CREG_DEVIDX(regname) ((QIB_7322_##regname##_OFFS - \
+			QIB_7322_LBIntCnt_OFFS) / sizeof(u64))
+#define cr_base_egrovfl CREG_DEVIDX(RxP0HdrEgrOvflCnt)
+#define cr_lbint CREG_DEVIDX(LBIntCnt)
+#define cr_lbstall CREG_DEVIDX(LBFlowStallCnt)
+#define cr_pcieretrydiag CREG_DEVIDX(PcieRetryBufDiagQwordCnt)
+#define cr_rxtidflowdrop CREG_DEVIDX(RxTidFlowDropCnt)
+#define cr_tidfull CREG_DEVIDX(RxTIDFullErrCnt)
+#define cr_tidinvalid CREG_DEVIDX(RxTIDValidErrCnt)
+
+/* no chip register for # of IB ports supported, so define */
+#define NUM_IB_PORTS 2
+
+/* 1 VL15 buffer per hardware IB port, no register for this, so define */
+#define NUM_VL15_BUFS NUM_IB_PORTS
+
+/*
+ * context 0 and 1 are special, and there is no chip register that
+ * defines this value, so we have to define it here.
+ * These are all allocated to either 0 or 1 for single port
+ * hardware configuration, otherwise each gets half
+ */
+#define KCTXT0_EGRCNT 2048
+
+/* values for vl and port fields in PBC, 7322-specific */
+#define PBC_PORT_SEL_LSB 26
+#define PBC_PORT_SEL_RMASK 1
+#define PBC_VL_NUM_LSB 27
+#define PBC_VL_NUM_RMASK 7
+#define PBC_7322_VL15_SEND (1ULL << 63) /* pbc; VL15, no credit check */
+#define PBC_7322_VL15_SEND_CTRL (1ULL << 31) /* control version of same */
+
+static u8 ib_rate_to_delay[IB_RATE_120_GBPS + 1] = {
+	[IB_RATE_2_5_GBPS] = 16,
+	[IB_RATE_5_GBPS] = 8,
+	[IB_RATE_10_GBPS] = 4,
+	[IB_RATE_20_GBPS] = 2,
+	[IB_RATE_30_GBPS] = 2,
+	[IB_RATE_40_GBPS] = 1
+};
+
+#define IBA7322_LINKSPEED_SHIFT SYM_LSB(IBCStatusA_0, LinkSpeedActive)
+#define IBA7322_LINKWIDTH_SHIFT SYM_LSB(IBCStatusA_0, LinkWidthActive)
+
+/* link training states, from IBC */
+#define IB_7322_LT_STATE_DISABLED        0x00
+#define IB_7322_LT_STATE_LINKUP          0x01
+#define IB_7322_LT_STATE_POLLACTIVE      0x02
+#define IB_7322_LT_STATE_POLLQUIET       0x03
+#define IB_7322_LT_STATE_SLEEPDELAY      0x04
+#define IB_7322_LT_STATE_SLEEPQUIET      0x05
+#define IB_7322_LT_STATE_CFGDEBOUNCE     0x08
+#define IB_7322_LT_STATE_CFGRCVFCFG      0x09
+#define IB_7322_LT_STATE_CFGWAITRMT      0x0a
+#define IB_7322_LT_STATE_CFGIDLE         0x0b
+#define IB_7322_LT_STATE_RECOVERRETRAIN  0x0c
+#define IB_7322_LT_STATE_TXREVLANES      0x0d
+#define IB_7322_LT_STATE_RECOVERWAITRMT  0x0e
+#define IB_7322_LT_STATE_RECOVERIDLE     0x0f
+#define IB_7322_LT_STATE_CFGENH          0x10
+#define IB_7322_LT_STATE_CFGTEST         0x11
+#define IB_7322_LT_STATE_CFGWAITRMTTEST  0x12
+#define IB_7322_LT_STATE_CFGWAITENH      0x13
+
+/* link state machine states from IBC */
+#define IB_7322_L_STATE_DOWN             0x0
+#define IB_7322_L_STATE_INIT             0x1
+#define IB_7322_L_STATE_ARM              0x2
+#define IB_7322_L_STATE_ACTIVE           0x3
+#define IB_7322_L_STATE_ACT_DEFER        0x4
+
+static const u8 qib_7322_physportstate[0x20] = {
+	[IB_7322_LT_STATE_DISABLED] = IB_PHYSPORTSTATE_DISABLED,
+	[IB_7322_LT_STATE_LINKUP] = IB_PHYSPORTSTATE_LINKUP,
+	[IB_7322_LT_STATE_POLLACTIVE] = IB_PHYSPORTSTATE_POLL,
+	[IB_7322_LT_STATE_POLLQUIET] = IB_PHYSPORTSTATE_POLL,
+	[IB_7322_LT_STATE_SLEEPDELAY] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_7322_LT_STATE_SLEEPQUIET] = IB_PHYSPORTSTATE_SLEEP,
+	[IB_7322_LT_STATE_CFGDEBOUNCE] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7322_LT_STATE_CFGRCVFCFG] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7322_LT_STATE_CFGWAITRMT] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7322_LT_STATE_CFGIDLE] = IB_PHYSPORTSTATE_CFG_IDLE,
+	[IB_7322_LT_STATE_RECOVERRETRAIN] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_7322_LT_STATE_RECOVERWAITRMT] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_7322_LT_STATE_RECOVERIDLE] =
+		IB_PHYSPORTSTATE_LINK_ERR_RECOVER,
+	[IB_7322_LT_STATE_CFGENH] = IB_PHYSPORTSTATE_CFG_ENH,
+	[IB_7322_LT_STATE_CFGTEST] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7322_LT_STATE_CFGWAITRMTTEST] =
+		IB_PHYSPORTSTATE_CFG_TRAIN,
+	[IB_7322_LT_STATE_CFGWAITENH] =
+		IB_PHYSPORTSTATE_CFG_WAIT_ENH,
+	[0x14] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x15] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x16] = IB_PHYSPORTSTATE_CFG_TRAIN,
+	[0x17] = IB_PHYSPORTSTATE_CFG_TRAIN
+};
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+struct qib_irq_notify {
+	int rcv;
+	void *arg;
+	struct irq_affinity_notify notify;
+};
+#endif
+
+struct qib_chip_specific {
+	u64 __iomem *cregbase;
+	u64 *cntrs;
+	spinlock_t rcvmod_lock; /* protect rcvctrl shadow changes */
+	spinlock_t gpio_lock; /* RMW of shadows/regs for ExtCtrl and GPIO */
+	u64 main_int_mask;      /* clear bits which have dedicated handlers */
+	u64 int_enable_mask;  /* for per port interrupts in single port mode */
+	u64 errormask;
+	u64 hwerrmask;
+	u64 gpio_out; /* shadow of kr_gpio_out, for rmw ops */
+	u64 gpio_mask; /* shadow the gpio mask register */
+	u64 extctrl; /* shadow the gpio output enable, etc... */
+	u32 ncntrs;
+	u32 nportcntrs;
+	u32 cntrnamelen;
+	u32 portcntrnamelen;
+	u32 numctxts;
+	u32 rcvegrcnt;
+	u32 updthresh; /* current AvailUpdThld */
+	u32 updthresh_dflt; /* default AvailUpdThld */
+	u32 r1;
+	int irq;
+	u32 num_msix_entries;
+	u32 sdmabufcnt;
+	u32 lastbuf_for_pio;
+	u32 stay_in_freeze;
+	u32 recovery_ports_initted;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	u32 dca_ctrl;
+	int rhdr_cpu[18];
+	int sdma_cpu[2];
+	u64 dca_rcvhdr_ctrl[5]; /* B, C, D, E, F */
+#endif
+	struct qib_msix_entry *msix_entries;
+	unsigned long *sendchkenable;
+	unsigned long *sendgrhchk;
+	unsigned long *sendibchk;
+	u32 rcvavail_timeout[18];
+	char emsgbuf[128]; /* for device error interrupt msg buffer */
+};
+
+/* Table of entries in "human readable" form Tx Emphasis. */
+struct txdds_ent {
+	u8 amp;
+	u8 pre;
+	u8 main;
+	u8 post;
+};
+
+struct vendor_txdds_ent {
+	u8 oui[QSFP_VOUI_LEN];
+	u8 *partnum;
+	struct txdds_ent sdr;
+	struct txdds_ent ddr;
+	struct txdds_ent qdr;
+};
+
+static void write_tx_serdes_param(struct qib_pportdata *, struct txdds_ent *);
+
+#define TXDDS_TABLE_SZ 16 /* number of entries per speed in onchip table */
+#define TXDDS_EXTRA_SZ 18 /* number of extra tx settings entries */
+#define TXDDS_MFG_SZ 2    /* number of mfg tx settings entries */
+#define SERDES_CHANS 4 /* yes, it's obvious, but one less magic number */
+
+#define H1_FORCE_VAL 8
+#define H1_FORCE_QME 1 /*  may be overridden via setup_txselect() */
+#define H1_FORCE_QMH 7 /*  may be overridden via setup_txselect() */
+
+/* The static and dynamic registers are paired, and the pairs indexed by spd */
+#define krp_static_adapt_dis(spd) (KREG_IBPORT_IDX(ADAPT_DISABLE_STATIC_SDR) \
+	+ ((spd) * 2))
+
+#define QDR_DFE_DISABLE_DELAY 4000 /* msec after LINKUP */
+#define QDR_STATIC_ADAPT_DOWN 0xf0f0f0f0ULL /* link down, H1-H4 QDR adapts */
+#define QDR_STATIC_ADAPT_DOWN_R1 0ULL /* r1 link down, H1-H4 QDR adapts */
+#define QDR_STATIC_ADAPT_INIT 0xffffffffffULL /* up, disable H0,H1-8, LE */
+#define QDR_STATIC_ADAPT_INIT_R1 0xf0ffffffffULL /* r1 up, disable H0,H1-8 */
+
+struct qib_chippport_specific {
+	u64 __iomem *kpregbase;
+	u64 __iomem *cpregbase;
+	u64 *portcntrs;
+	struct qib_pportdata *ppd;
+	wait_queue_head_t autoneg_wait;
+	struct delayed_work autoneg_work;
+	struct delayed_work ipg_work;
+	struct timer_list chase_timer;
+	/*
+	 * these 5 fields are used to establish deltas for IB symbol
+	 * errors and linkrecovery errors.  They can be reported on
+	 * some chips during link negotiation prior to INIT, and with
+	 * DDR when faking DDR negotiations with non-IBTA switches.
+	 * The chip counters are adjusted at driver unload if there is
+	 * a non-zero delta.
+	 */
+	u64 ibdeltainprog;
+	u64 ibsymdelta;
+	u64 ibsymsnap;
+	u64 iblnkerrdelta;
+	u64 iblnkerrsnap;
+	u64 iblnkdownsnap;
+	u64 iblnkdowndelta;
+	u64 ibmalfdelta;
+	u64 ibmalfsnap;
+	u64 ibcctrl_a; /* krp_ibcctrl_a shadow */
+	u64 ibcctrl_b; /* krp_ibcctrl_b shadow */
+	unsigned long qdr_dfe_time;
+	unsigned long chase_end;
+	u32 autoneg_tries;
+	u32 recovery_init;
+	u32 qdr_dfe_on;
+	u32 qdr_reforce;
+	/*
+	 * Per-bay per-channel rcv QMH H1 values and Tx values for QDR.
+	 * entry zero is unused, to simplify indexing
+	 */
+	u8 h1_val;
+	u8 no_eep;  /* txselect table index to use if no qsfp info */
+	u8 ipg_tries;
+	u8 ibmalfusesnap;
+	struct qib_qsfp_data qsfp_data;
+	char epmsgbuf[192]; /* for port error interrupt msg buffer */
+	char sdmamsgbuf[192]; /* for per-port sdma error messages */
+};
+
+static struct {
+	const char *name;
+	irq_handler_t handler;
+	int lsb;
+	int port; /* 0 if not port-specific, else port # */
+	int dca;
+} irq_table[] = {
+	{ "", qib_7322intr, -1, 0, 0 },
+	{ " (buf avail)", qib_7322bufavail,
+		SYM_LSB(IntStatus, SendBufAvail), 0, 0},
+	{ " (sdma 0)", sdma_intr,
+		SYM_LSB(IntStatus, SDmaInt_0), 1, 1 },
+	{ " (sdma 1)", sdma_intr,
+		SYM_LSB(IntStatus, SDmaInt_1), 2, 1 },
+	{ " (sdmaI 0)", sdma_idle_intr,
+		SYM_LSB(IntStatus, SDmaIdleInt_0), 1, 1},
+	{ " (sdmaI 1)", sdma_idle_intr,
+		SYM_LSB(IntStatus, SDmaIdleInt_1), 2, 1},
+	{ " (sdmaP 0)", sdma_progress_intr,
+		SYM_LSB(IntStatus, SDmaProgressInt_0), 1, 1 },
+	{ " (sdmaP 1)", sdma_progress_intr,
+		SYM_LSB(IntStatus, SDmaProgressInt_1), 2, 1 },
+	{ " (sdmaC 0)", sdma_cleanup_intr,
+		SYM_LSB(IntStatus, SDmaCleanupDone_0), 1, 0 },
+	{ " (sdmaC 1)", sdma_cleanup_intr,
+		SYM_LSB(IntStatus, SDmaCleanupDone_1), 2 , 0},
+};
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+
+static const struct dca_reg_map {
+	int     shadow_inx;
+	int     lsb;
+	u64     mask;
+	u16     regno;
+} dca_rcvhdr_reg_map[] = {
+	{ 0, SYM_LSB(DCACtrlB, RcvHdrq0DCAOPH),
+	   ~SYM_MASK(DCACtrlB, RcvHdrq0DCAOPH) , KREG_IDX(DCACtrlB) },
+	{ 0, SYM_LSB(DCACtrlB, RcvHdrq1DCAOPH),
+	   ~SYM_MASK(DCACtrlB, RcvHdrq1DCAOPH) , KREG_IDX(DCACtrlB) },
+	{ 0, SYM_LSB(DCACtrlB, RcvHdrq2DCAOPH),
+	   ~SYM_MASK(DCACtrlB, RcvHdrq2DCAOPH) , KREG_IDX(DCACtrlB) },
+	{ 0, SYM_LSB(DCACtrlB, RcvHdrq3DCAOPH),
+	   ~SYM_MASK(DCACtrlB, RcvHdrq3DCAOPH) , KREG_IDX(DCACtrlB) },
+	{ 1, SYM_LSB(DCACtrlC, RcvHdrq4DCAOPH),
+	   ~SYM_MASK(DCACtrlC, RcvHdrq4DCAOPH) , KREG_IDX(DCACtrlC) },
+	{ 1, SYM_LSB(DCACtrlC, RcvHdrq5DCAOPH),
+	   ~SYM_MASK(DCACtrlC, RcvHdrq5DCAOPH) , KREG_IDX(DCACtrlC) },
+	{ 1, SYM_LSB(DCACtrlC, RcvHdrq6DCAOPH),
+	   ~SYM_MASK(DCACtrlC, RcvHdrq6DCAOPH) , KREG_IDX(DCACtrlC) },
+	{ 1, SYM_LSB(DCACtrlC, RcvHdrq7DCAOPH),
+	   ~SYM_MASK(DCACtrlC, RcvHdrq7DCAOPH) , KREG_IDX(DCACtrlC) },
+	{ 2, SYM_LSB(DCACtrlD, RcvHdrq8DCAOPH),
+	   ~SYM_MASK(DCACtrlD, RcvHdrq8DCAOPH) , KREG_IDX(DCACtrlD) },
+	{ 2, SYM_LSB(DCACtrlD, RcvHdrq9DCAOPH),
+	   ~SYM_MASK(DCACtrlD, RcvHdrq9DCAOPH) , KREG_IDX(DCACtrlD) },
+	{ 2, SYM_LSB(DCACtrlD, RcvHdrq10DCAOPH),
+	   ~SYM_MASK(DCACtrlD, RcvHdrq10DCAOPH) , KREG_IDX(DCACtrlD) },
+	{ 2, SYM_LSB(DCACtrlD, RcvHdrq11DCAOPH),
+	   ~SYM_MASK(DCACtrlD, RcvHdrq11DCAOPH) , KREG_IDX(DCACtrlD) },
+	{ 3, SYM_LSB(DCACtrlE, RcvHdrq12DCAOPH),
+	   ~SYM_MASK(DCACtrlE, RcvHdrq12DCAOPH) , KREG_IDX(DCACtrlE) },
+	{ 3, SYM_LSB(DCACtrlE, RcvHdrq13DCAOPH),
+	   ~SYM_MASK(DCACtrlE, RcvHdrq13DCAOPH) , KREG_IDX(DCACtrlE) },
+	{ 3, SYM_LSB(DCACtrlE, RcvHdrq14DCAOPH),
+	   ~SYM_MASK(DCACtrlE, RcvHdrq14DCAOPH) , KREG_IDX(DCACtrlE) },
+	{ 3, SYM_LSB(DCACtrlE, RcvHdrq15DCAOPH),
+	   ~SYM_MASK(DCACtrlE, RcvHdrq15DCAOPH) , KREG_IDX(DCACtrlE) },
+	{ 4, SYM_LSB(DCACtrlF, RcvHdrq16DCAOPH),
+	   ~SYM_MASK(DCACtrlF, RcvHdrq16DCAOPH) , KREG_IDX(DCACtrlF) },
+	{ 4, SYM_LSB(DCACtrlF, RcvHdrq17DCAOPH),
+	   ~SYM_MASK(DCACtrlF, RcvHdrq17DCAOPH) , KREG_IDX(DCACtrlF) },
+};
+#endif
+
+/* ibcctrl bits */
+#define QLOGIC_IB_IBCC_LINKINITCMD_DISABLE 1
+/* cycle through TS1/TS2 till OK */
+#define QLOGIC_IB_IBCC_LINKINITCMD_POLL 2
+/* wait for TS1, then go on */
+#define QLOGIC_IB_IBCC_LINKINITCMD_SLEEP 3
+#define QLOGIC_IB_IBCC_LINKINITCMD_SHIFT 16
+
+#define QLOGIC_IB_IBCC_LINKCMD_DOWN 1           /* move to 0x11 */
+#define QLOGIC_IB_IBCC_LINKCMD_ARMED 2          /* move to 0x21 */
+#define QLOGIC_IB_IBCC_LINKCMD_ACTIVE 3 /* move to 0x31 */
+
+#define BLOB_7322_IBCHG 0x101
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u32 regno, u64 value);
+static inline u32 qib_read_kreg32(const struct qib_devdata *, const u32);
+static void write_7322_initregs(struct qib_devdata *);
+static void write_7322_init_portregs(struct qib_pportdata *);
+static void setup_7322_link_recovery(struct qib_pportdata *, u32);
+static void check_7322_rxe_status(struct qib_pportdata *);
+static u32 __iomem *qib_7322_getsendbuf(struct qib_pportdata *, u64, u32 *);
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+static void qib_setup_dca(struct qib_devdata *dd);
+static void setup_dca_notifier(struct qib_devdata *dd,
+			       struct qib_msix_entry *m);
+static void reset_dca_notifier(struct qib_devdata *dd,
+			       struct qib_msix_entry *m);
+#endif
+
+/**
+ * qib_read_ureg32 - read 32-bit virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @ctxt: context number
+ *
+ * Return the contents of a register that is virtualized to be per context.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
+ */
+static inline u32 qib_read_ureg32(const struct qib_devdata *dd,
+				  enum qib_ureg regno, int ctxt)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readl(regno + (u64 __iomem *)(
+		(dd->ureg_align * ctxt) + (dd->userbase ?
+		 (char __iomem *)dd->userbase :
+		 (char __iomem *)dd->kregbase + dd->uregbase)));
+}
+
+/**
+ * qib_read_ureg - read virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @ctxt: context number
+ *
+ * Return the contents of a register that is virtualized to be per context.
+ * Returns -1 on errors (not distinguishable from valid contents at
+ * runtime; we may add a separate error variable at some point).
+ */
+static inline u64 qib_read_ureg(const struct qib_devdata *dd,
+				enum qib_ureg regno, int ctxt)
+{
+
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readq(regno + (u64 __iomem *)(
+		(dd->ureg_align * ctxt) + (dd->userbase ?
+		 (char __iomem *)dd->userbase :
+		 (char __iomem *)dd->kregbase + dd->uregbase)));
+}
+
+/**
+ * qib_write_ureg - write virtualized per-context register
+ * @dd: device
+ * @regno: register number
+ * @value: value
+ * @ctxt: context
+ *
+ * Write the contents of a register that is virtualized to be per context.
+ */
+static inline void qib_write_ureg(const struct qib_devdata *dd,
+				  enum qib_ureg regno, u64 value, int ctxt)
+{
+	u64 __iomem *ubase;
+	if (dd->userbase)
+		ubase = (u64 __iomem *)
+			((char __iomem *) dd->userbase +
+			 dd->ureg_align * ctxt);
+	else
+		ubase = (u64 __iomem *)
+			(dd->uregbase +
+			 (char __iomem *) dd->kregbase +
+			 dd->ureg_align * ctxt);
+
+	if (dd->kregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &ubase[regno]);
+}
+
+static inline u32 qib_read_kreg32(const struct qib_devdata *dd,
+				  const u32 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readl((u32 __iomem *) &dd->kregbase[regno]);
+}
+
+static inline u64 qib_read_kreg64(const struct qib_devdata *dd,
+				  const u32 regno)
+{
+	if (!dd->kregbase || !(dd->flags & QIB_PRESENT))
+		return -1;
+	return readq(&dd->kregbase[regno]);
+}
+
+static inline void qib_write_kreg(const struct qib_devdata *dd,
+				  const u32 regno, u64 value)
+{
+	if (dd->kregbase && (dd->flags & QIB_PRESENT))
+		writeq(value, &dd->kregbase[regno]);
+}
+
+/*
+ * not many sanity checks for the port-specific kernel register routines,
+ * since they are only used when it's known to be safe.
+*/
+static inline u64 qib_read_kreg_port(const struct qib_pportdata *ppd,
+				     const u16 regno)
+{
+	if (!ppd->cpspec->kpregbase || !(ppd->dd->flags & QIB_PRESENT))
+		return 0ULL;
+	return readq(&ppd->cpspec->kpregbase[regno]);
+}
+
+static inline void qib_write_kreg_port(const struct qib_pportdata *ppd,
+				       const u16 regno, u64 value)
+{
+	if (ppd->cpspec && ppd->dd && ppd->cpspec->kpregbase &&
+	    (ppd->dd->flags & QIB_PRESENT))
+		writeq(value, &ppd->cpspec->kpregbase[regno]);
+}
+
+/**
+ * qib_write_kreg_ctxt - write a device's per-ctxt 64-bit kernel register
+ * @dd: the qlogic_ib device
+ * @regno: the register number to write
+ * @ctxt: the context containing the register
+ * @value: the value to write
+ */
+static inline void qib_write_kreg_ctxt(const struct qib_devdata *dd,
+				       const u16 regno, unsigned ctxt,
+				       u64 value)
+{
+	qib_write_kreg(dd, regno + ctxt, value);
+}
+
+static inline u64 read_7322_creg(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readq(&dd->cspec->cregbase[regno]);
+
+
+}
+
+static inline u32 read_7322_creg32(const struct qib_devdata *dd, u16 regno)
+{
+	if (!dd->cspec->cregbase || !(dd->flags & QIB_PRESENT))
+		return 0;
+	return readl(&dd->cspec->cregbase[regno]);
+
+
+}
+
+static inline void write_7322_creg_port(const struct qib_pportdata *ppd,
+					u16 regno, u64 value)
+{
+	if (ppd->cpspec && ppd->cpspec->cpregbase &&
+	    (ppd->dd->flags & QIB_PRESENT))
+		writeq(value, &ppd->cpspec->cpregbase[regno]);
+}
+
+static inline u64 read_7322_creg_port(const struct qib_pportdata *ppd,
+				      u16 regno)
+{
+	if (!ppd->cpspec || !ppd->cpspec->cpregbase ||
+	    !(ppd->dd->flags & QIB_PRESENT))
+		return 0;
+	return readq(&ppd->cpspec->cpregbase[regno]);
+}
+
+static inline u32 read_7322_creg32_port(const struct qib_pportdata *ppd,
+					u16 regno)
+{
+	if (!ppd->cpspec || !ppd->cpspec->cpregbase ||
+	    !(ppd->dd->flags & QIB_PRESENT))
+		return 0;
+	return readl(&ppd->cpspec->cpregbase[regno]);
+}
+
+/* bits in Control register */
+#define QLOGIC_IB_C_RESET SYM_MASK(Control, SyncReset)
+#define QLOGIC_IB_C_SDMAFETCHPRIOEN SYM_MASK(Control, SDmaDescFetchPriorityEn)
+
+/* bits in general interrupt regs */
+#define QIB_I_RCVURG_LSB SYM_LSB(IntMask, RcvUrg0IntMask)
+#define QIB_I_RCVURG_RMASK MASK_ACROSS(0, 17)
+#define QIB_I_RCVURG_MASK (QIB_I_RCVURG_RMASK << QIB_I_RCVURG_LSB)
+#define QIB_I_RCVAVAIL_LSB SYM_LSB(IntMask, RcvAvail0IntMask)
+#define QIB_I_RCVAVAIL_RMASK MASK_ACROSS(0, 17)
+#define QIB_I_RCVAVAIL_MASK (QIB_I_RCVAVAIL_RMASK << QIB_I_RCVAVAIL_LSB)
+#define QIB_I_C_ERROR INT_MASK(Err)
+
+#define QIB_I_SPIOSENT (INT_MASK_P(SendDone, 0) | INT_MASK_P(SendDone, 1))
+#define QIB_I_SPIOBUFAVAIL INT_MASK(SendBufAvail)
+#define QIB_I_GPIO INT_MASK(AssertGPIO)
+#define QIB_I_P_SDMAINT(pidx) \
+	(INT_MASK_P(SDma, pidx) | INT_MASK_P(SDmaIdle, pidx) | \
+	 INT_MASK_P(SDmaProgress, pidx) | \
+	 INT_MASK_PM(SDmaCleanupDone, pidx))
+
+/* Interrupt bits that are "per port" */
+#define QIB_I_P_BITSEXTANT(pidx) \
+	(INT_MASK_P(Err, pidx) | INT_MASK_P(SendDone, pidx) | \
+	INT_MASK_P(SDma, pidx) | INT_MASK_P(SDmaIdle, pidx) | \
+	INT_MASK_P(SDmaProgress, pidx) | \
+	INT_MASK_PM(SDmaCleanupDone, pidx))
+
+/* Interrupt bits that are common to a device */
+/* currently unused: QIB_I_SPIOSENT */
+#define QIB_I_C_BITSEXTANT \
+	(QIB_I_RCVURG_MASK | QIB_I_RCVAVAIL_MASK | \
+	QIB_I_SPIOSENT | \
+	QIB_I_C_ERROR | QIB_I_SPIOBUFAVAIL | QIB_I_GPIO)
+
+#define QIB_I_BITSEXTANT (QIB_I_C_BITSEXTANT | \
+	QIB_I_P_BITSEXTANT(0) | QIB_I_P_BITSEXTANT(1))
+
+/*
+ * Error bits that are "per port".
+ */
+#define QIB_E_P_IBSTATUSCHANGED ERR_MASK_N(IBStatusChanged)
+#define QIB_E_P_SHDR ERR_MASK_N(SHeadersErr)
+#define QIB_E_P_VL15_BUF_MISUSE ERR_MASK_N(VL15BufMisuseErr)
+#define QIB_E_P_SND_BUF_MISUSE ERR_MASK_N(SendBufMisuseErr)
+#define QIB_E_P_SUNSUPVL ERR_MASK_N(SendUnsupportedVLErr)
+#define QIB_E_P_SUNEXP_PKTNUM ERR_MASK_N(SendUnexpectedPktNumErr)
+#define QIB_E_P_SDROP_DATA ERR_MASK_N(SendDroppedDataPktErr)
+#define QIB_E_P_SDROP_SMP ERR_MASK_N(SendDroppedSmpPktErr)
+#define QIB_E_P_SPKTLEN ERR_MASK_N(SendPktLenErr)
+#define QIB_E_P_SUNDERRUN ERR_MASK_N(SendUnderRunErr)
+#define QIB_E_P_SMAXPKTLEN ERR_MASK_N(SendMaxPktLenErr)
+#define QIB_E_P_SMINPKTLEN ERR_MASK_N(SendMinPktLenErr)
+#define QIB_E_P_RIBLOSTLINK ERR_MASK_N(RcvIBLostLinkErr)
+#define QIB_E_P_RHDR ERR_MASK_N(RcvHdrErr)
+#define QIB_E_P_RHDRLEN ERR_MASK_N(RcvHdrLenErr)
+#define QIB_E_P_RBADTID ERR_MASK_N(RcvBadTidErr)
+#define QIB_E_P_RBADVERSION ERR_MASK_N(RcvBadVersionErr)
+#define QIB_E_P_RIBFLOW ERR_MASK_N(RcvIBFlowErr)
+#define QIB_E_P_REBP ERR_MASK_N(RcvEBPErr)
+#define QIB_E_P_RUNSUPVL ERR_MASK_N(RcvUnsupportedVLErr)
+#define QIB_E_P_RUNEXPCHAR ERR_MASK_N(RcvUnexpectedCharErr)
+#define QIB_E_P_RSHORTPKTLEN ERR_MASK_N(RcvShortPktLenErr)
+#define QIB_E_P_RLONGPKTLEN ERR_MASK_N(RcvLongPktLenErr)
+#define QIB_E_P_RMAXPKTLEN ERR_MASK_N(RcvMaxPktLenErr)
+#define QIB_E_P_RMINPKTLEN ERR_MASK_N(RcvMinPktLenErr)
+#define QIB_E_P_RICRC ERR_MASK_N(RcvICRCErr)
+#define QIB_E_P_RVCRC ERR_MASK_N(RcvVCRCErr)
+#define QIB_E_P_RFORMATERR ERR_MASK_N(RcvFormatErr)
+
+#define QIB_E_P_SDMA1STDESC ERR_MASK_N(SDma1stDescErr)
+#define QIB_E_P_SDMABASE ERR_MASK_N(SDmaBaseErr)
+#define QIB_E_P_SDMADESCADDRMISALIGN ERR_MASK_N(SDmaDescAddrMisalignErr)
+#define QIB_E_P_SDMADWEN ERR_MASK_N(SDmaDwEnErr)
+#define QIB_E_P_SDMAGENMISMATCH ERR_MASK_N(SDmaGenMismatchErr)
+#define QIB_E_P_SDMAHALT ERR_MASK_N(SDmaHaltErr)
+#define QIB_E_P_SDMAMISSINGDW ERR_MASK_N(SDmaMissingDwErr)
+#define QIB_E_P_SDMAOUTOFBOUND ERR_MASK_N(SDmaOutOfBoundErr)
+#define QIB_E_P_SDMARPYTAG ERR_MASK_N(SDmaRpyTagErr)
+#define QIB_E_P_SDMATAILOUTOFBOUND ERR_MASK_N(SDmaTailOutOfBoundErr)
+#define QIB_E_P_SDMAUNEXPDATA ERR_MASK_N(SDmaUnexpDataErr)
+
+/* Error bits that are common to a device */
+#define QIB_E_RESET ERR_MASK(ResetNegated)
+#define QIB_E_HARDWARE ERR_MASK(HardwareErr)
+#define QIB_E_INVALIDADDR ERR_MASK(InvalidAddrErr)
+
+
+/*
+ * Per chip (rather than per-port) errors.  Most either do
+ * nothing but trigger a print (because they self-recover, or
+ * always occur in tandem with other errors that handle the
+ * issue), or because they indicate errors with no recovery,
+ * but we want to know that they happened.
+ */
+#define QIB_E_SBUF_VL15_MISUSE ERR_MASK(SBufVL15MisUseErr)
+#define QIB_E_BADEEP ERR_MASK(InvalidEEPCmd)
+#define QIB_E_VLMISMATCH ERR_MASK(SendVLMismatchErr)
+#define QIB_E_ARMLAUNCH ERR_MASK(SendArmLaunchErr)
+#define QIB_E_SPCLTRIG ERR_MASK(SendSpecialTriggerErr)
+#define QIB_E_RRCVHDRFULL ERR_MASK(RcvHdrFullErr)
+#define QIB_E_RRCVEGRFULL ERR_MASK(RcvEgrFullErr)
+#define QIB_E_RCVCTXTSHARE ERR_MASK(RcvContextShareErr)
+
+/* SDMA chip errors (not per port)
+ * QIB_E_SDMA_BUF_DUP needs no special handling, because we will also get
+ * the SDMAHALT error immediately, so we just print the dup error via the
+ * E_AUTO mechanism.  This is true of most of the per-port fatal errors
+ * as well, but since this is port-independent, by definition, it's
+ * handled a bit differently.  SDMA_VL15 and SDMA_WRONG_PORT are per
+ * packet send errors, and so are handled in the same manner as other
+ * per-packet errors.
+ */
+#define QIB_E_SDMA_VL15 ERR_MASK(SDmaVL15Err)
+#define QIB_E_SDMA_WRONG_PORT ERR_MASK(SDmaWrongPortErr)
+#define QIB_E_SDMA_BUF_DUP ERR_MASK(SDmaBufMaskDuplicateErr)
+
+/*
+ * Below functionally equivalent to legacy QLOGIC_IB_E_PKTERRS
+ * it is used to print "common" packet errors.
+ */
+#define QIB_E_P_PKTERRS (QIB_E_P_SPKTLEN |\
+	QIB_E_P_SDROP_DATA | QIB_E_P_RVCRC |\
+	QIB_E_P_RICRC | QIB_E_P_RSHORTPKTLEN |\
+	QIB_E_P_VL15_BUF_MISUSE | QIB_E_P_SHDR | \
+	QIB_E_P_REBP)
+
+/* Error Bits that Packet-related (Receive, per-port) */
+#define QIB_E_P_RPKTERRS (\
+	QIB_E_P_RHDRLEN | QIB_E_P_RBADTID | \
+	QIB_E_P_RBADVERSION | QIB_E_P_RHDR | \
+	QIB_E_P_RLONGPKTLEN | QIB_E_P_RSHORTPKTLEN |\
+	QIB_E_P_RMAXPKTLEN | QIB_E_P_RMINPKTLEN | \
+	QIB_E_P_RFORMATERR | QIB_E_P_RUNSUPVL | \
+	QIB_E_P_RUNEXPCHAR | QIB_E_P_RIBFLOW | QIB_E_P_REBP)
+
+/*
+ * Error bits that are Send-related (per port)
+ * (ARMLAUNCH excluded from E_SPKTERRS because it gets special handling).
+ * All of these potentially need to have a buffer disarmed
+ */
+#define QIB_E_P_SPKTERRS (\
+	QIB_E_P_SUNEXP_PKTNUM |\
+	QIB_E_P_SDROP_DATA | QIB_E_P_SDROP_SMP |\
+	QIB_E_P_SMAXPKTLEN |\
+	QIB_E_P_VL15_BUF_MISUSE | QIB_E_P_SHDR | \
+	QIB_E_P_SMINPKTLEN | QIB_E_P_SPKTLEN | \
+	QIB_E_P_SND_BUF_MISUSE | QIB_E_P_SUNSUPVL)
+
+#define QIB_E_SPKTERRS ( \
+		QIB_E_SBUF_VL15_MISUSE | QIB_E_VLMISMATCH | \
+		ERR_MASK_N(SendUnsupportedVLErr) |			\
+		QIB_E_SPCLTRIG | QIB_E_SDMA_VL15 | QIB_E_SDMA_WRONG_PORT)
+
+#define QIB_E_P_SDMAERRS ( \
+	QIB_E_P_SDMAHALT | \
+	QIB_E_P_SDMADESCADDRMISALIGN | \
+	QIB_E_P_SDMAUNEXPDATA | \
+	QIB_E_P_SDMAMISSINGDW | \
+	QIB_E_P_SDMADWEN | \
+	QIB_E_P_SDMARPYTAG | \
+	QIB_E_P_SDMA1STDESC | \
+	QIB_E_P_SDMABASE | \
+	QIB_E_P_SDMATAILOUTOFBOUND | \
+	QIB_E_P_SDMAOUTOFBOUND | \
+	QIB_E_P_SDMAGENMISMATCH)
+
+/*
+ * This sets some bits more than once, but makes it more obvious which
+ * bits are not handled under other categories, and the repeat definition
+ * is not a problem.
+ */
+#define QIB_E_P_BITSEXTANT ( \
+	QIB_E_P_SPKTERRS | QIB_E_P_PKTERRS | QIB_E_P_RPKTERRS | \
+	QIB_E_P_RIBLOSTLINK | QIB_E_P_IBSTATUSCHANGED | \
+	QIB_E_P_SND_BUF_MISUSE | QIB_E_P_SUNDERRUN | \
+	QIB_E_P_SHDR | QIB_E_P_VL15_BUF_MISUSE | QIB_E_P_SDMAERRS \
+	)
+
+/*
+ * These are errors that can occur when the link
+ * changes state while a packet is being sent or received.  This doesn't
+ * cover things like EBP or VCRC that can be the result of a sending
+ * having the link change state, so we receive a "known bad" packet.
+ * All of these are "per port", so renamed:
+ */
+#define QIB_E_P_LINK_PKTERRS (\
+	QIB_E_P_SDROP_DATA | QIB_E_P_SDROP_SMP |\
+	QIB_E_P_SMINPKTLEN | QIB_E_P_SPKTLEN |\
+	QIB_E_P_RSHORTPKTLEN | QIB_E_P_RMINPKTLEN |\
+	QIB_E_P_RUNEXPCHAR)
+
+/*
+ * This sets some bits more than once, but makes it more obvious which
+ * bits are not handled under other categories (such as QIB_E_SPKTERRS),
+ * and the repeat definition is not a problem.
+ */
+#define QIB_E_C_BITSEXTANT (\
+	QIB_E_HARDWARE | QIB_E_INVALIDADDR | QIB_E_BADEEP |\
+	QIB_E_ARMLAUNCH | QIB_E_VLMISMATCH | QIB_E_RRCVHDRFULL |\
+	QIB_E_RRCVEGRFULL | QIB_E_RESET | QIB_E_SBUF_VL15_MISUSE)
+
+/* Likewise Neuter E_SPKT_ERRS_IGNORE */
+#define E_SPKT_ERRS_IGNORE 0
+
+#define QIB_EXTS_MEMBIST_DISABLED \
+	SYM_MASK(EXTStatus, MemBISTDisabled)
+#define QIB_EXTS_MEMBIST_ENDTEST \
+	SYM_MASK(EXTStatus, MemBISTEndTest)
+
+#define QIB_E_SPIOARMLAUNCH \
+	ERR_MASK(SendArmLaunchErr)
+
+#define IBA7322_IBCC_LINKINITCMD_MASK SYM_RMASK(IBCCtrlA_0, LinkInitCmd)
+#define IBA7322_IBCC_LINKCMD_SHIFT SYM_LSB(IBCCtrlA_0, LinkCmd)
+
+/*
+ * IBTA_1_2 is set when multiple speeds are enabled (normal),
+ * and also if forced QDR (only QDR enabled).  It's enabled for the
+ * forced QDR case so that scrambling will be enabled by the TS3
+ * exchange, when supported by both sides of the link.
+ */
+#define IBA7322_IBC_IBTA_1_2_MASK SYM_MASK(IBCCtrlB_0, IB_ENHANCED_MODE)
+#define IBA7322_IBC_MAX_SPEED_MASK SYM_MASK(IBCCtrlB_0, SD_SPEED)
+#define IBA7322_IBC_SPEED_QDR SYM_MASK(IBCCtrlB_0, SD_SPEED_QDR)
+#define IBA7322_IBC_SPEED_DDR SYM_MASK(IBCCtrlB_0, SD_SPEED_DDR)
+#define IBA7322_IBC_SPEED_SDR SYM_MASK(IBCCtrlB_0, SD_SPEED_SDR)
+#define IBA7322_IBC_SPEED_MASK (SYM_MASK(IBCCtrlB_0, SD_SPEED_SDR) | \
+	SYM_MASK(IBCCtrlB_0, SD_SPEED_DDR) | SYM_MASK(IBCCtrlB_0, SD_SPEED_QDR))
+#define IBA7322_IBC_SPEED_LSB SYM_LSB(IBCCtrlB_0, SD_SPEED_SDR)
+
+#define IBA7322_LEDBLINK_OFF_SHIFT SYM_LSB(RcvPktLEDCnt_0, OFFperiod)
+#define IBA7322_LEDBLINK_ON_SHIFT SYM_LSB(RcvPktLEDCnt_0, ONperiod)
+
+#define IBA7322_IBC_WIDTH_AUTONEG SYM_MASK(IBCCtrlB_0, IB_NUM_CHANNELS)
+#define IBA7322_IBC_WIDTH_4X_ONLY (1<<SYM_LSB(IBCCtrlB_0, IB_NUM_CHANNELS))
+#define IBA7322_IBC_WIDTH_1X_ONLY (0<<SYM_LSB(IBCCtrlB_0, IB_NUM_CHANNELS))
+
+#define IBA7322_IBC_RXPOL_MASK SYM_MASK(IBCCtrlB_0, IB_POLARITY_REV_SUPP)
+#define IBA7322_IBC_RXPOL_LSB SYM_LSB(IBCCtrlB_0, IB_POLARITY_REV_SUPP)
+#define IBA7322_IBC_HRTBT_MASK (SYM_MASK(IBCCtrlB_0, HRTBT_AUTO) | \
+	SYM_MASK(IBCCtrlB_0, HRTBT_ENB))
+#define IBA7322_IBC_HRTBT_RMASK (IBA7322_IBC_HRTBT_MASK >> \
+	SYM_LSB(IBCCtrlB_0, HRTBT_ENB))
+#define IBA7322_IBC_HRTBT_LSB SYM_LSB(IBCCtrlB_0, HRTBT_ENB)
+
+#define IBA7322_REDIRECT_VEC_PER_REG 12
+
+#define IBA7322_SENDCHK_PKEY SYM_MASK(SendCheckControl_0, PKey_En)
+#define IBA7322_SENDCHK_BTHQP SYM_MASK(SendCheckControl_0, BTHQP_En)
+#define IBA7322_SENDCHK_SLID SYM_MASK(SendCheckControl_0, SLID_En)
+#define IBA7322_SENDCHK_RAW_IPV6 SYM_MASK(SendCheckControl_0, RawIPV6_En)
+#define IBA7322_SENDCHK_MINSZ SYM_MASK(SendCheckControl_0, PacketTooSmall_En)
+
+#define AUTONEG_TRIES 3 /* sequential retries to negotiate DDR */
+
+#define HWE_AUTO(fldname) { .mask = SYM_MASK(HwErrMask, fldname##Mask), \
+	.msg = #fldname , .sz = sizeof(#fldname) }
+#define HWE_AUTO_P(fldname, port) { .mask = SYM_MASK(HwErrMask, \
+	fldname##Mask##_##port), .msg = #fldname , .sz = sizeof(#fldname) }
+static const struct qib_hwerror_msgs qib_7322_hwerror_msgs[] = {
+	HWE_AUTO_P(IBSerdesPClkNotDetect, 1),
+	HWE_AUTO_P(IBSerdesPClkNotDetect, 0),
+	HWE_AUTO(PCIESerdesPClkNotDetect),
+	HWE_AUTO(PowerOnBISTFailed),
+	HWE_AUTO(TempsenseTholdReached),
+	HWE_AUTO(MemoryErr),
+	HWE_AUTO(PCIeBusParityErr),
+	HWE_AUTO(PcieCplTimeout),
+	HWE_AUTO(PciePoisonedTLP),
+	HWE_AUTO_P(SDmaMemReadErr, 1),
+	HWE_AUTO_P(SDmaMemReadErr, 0),
+	HWE_AUTO_P(IBCBusFromSPCParityErr, 1),
+	HWE_AUTO_P(IBCBusToSPCParityErr, 1),
+	HWE_AUTO_P(IBCBusFromSPCParityErr, 0),
+	HWE_AUTO(statusValidNoEop),
+	HWE_AUTO(LATriggered),
+	{ .mask = 0, .sz = 0 }
+};
+
+#define E_AUTO(fldname) { .mask = SYM_MASK(ErrMask, fldname##Mask), \
+	.msg = #fldname, .sz = sizeof(#fldname) }
+#define E_P_AUTO(fldname) { .mask = SYM_MASK(ErrMask_0, fldname##Mask), \
+	.msg = #fldname, .sz = sizeof(#fldname) }
+static const struct qib_hwerror_msgs qib_7322error_msgs[] = {
+	E_AUTO(RcvEgrFullErr),
+	E_AUTO(RcvHdrFullErr),
+	E_AUTO(ResetNegated),
+	E_AUTO(HardwareErr),
+	E_AUTO(InvalidAddrErr),
+	E_AUTO(SDmaVL15Err),
+	E_AUTO(SBufVL15MisUseErr),
+	E_AUTO(InvalidEEPCmd),
+	E_AUTO(RcvContextShareErr),
+	E_AUTO(SendVLMismatchErr),
+	E_AUTO(SendArmLaunchErr),
+	E_AUTO(SendSpecialTriggerErr),
+	E_AUTO(SDmaWrongPortErr),
+	E_AUTO(SDmaBufMaskDuplicateErr),
+	{ .mask = 0, .sz = 0 }
+};
+
+static const struct  qib_hwerror_msgs qib_7322p_error_msgs[] = {
+	E_P_AUTO(IBStatusChanged),
+	E_P_AUTO(SHeadersErr),
+	E_P_AUTO(VL15BufMisuseErr),
+	/*
+	 * SDmaHaltErr is not really an error, make it clearer;
+	 */
+	{.mask = SYM_MASK(ErrMask_0, SDmaHaltErrMask), .msg = "SDmaHalted",
+		.sz = 11},
+	E_P_AUTO(SDmaDescAddrMisalignErr),
+	E_P_AUTO(SDmaUnexpDataErr),
+	E_P_AUTO(SDmaMissingDwErr),
+	E_P_AUTO(SDmaDwEnErr),
+	E_P_AUTO(SDmaRpyTagErr),
+	E_P_AUTO(SDma1stDescErr),
+	E_P_AUTO(SDmaBaseErr),
+	E_P_AUTO(SDmaTailOutOfBoundErr),
+	E_P_AUTO(SDmaOutOfBoundErr),
+	E_P_AUTO(SDmaGenMismatchErr),
+	E_P_AUTO(SendBufMisuseErr),
+	E_P_AUTO(SendUnsupportedVLErr),
+	E_P_AUTO(SendUnexpectedPktNumErr),
+	E_P_AUTO(SendDroppedDataPktErr),
+	E_P_AUTO(SendDroppedSmpPktErr),
+	E_P_AUTO(SendPktLenErr),
+	E_P_AUTO(SendUnderRunErr),
+	E_P_AUTO(SendMaxPktLenErr),
+	E_P_AUTO(SendMinPktLenErr),
+	E_P_AUTO(RcvIBLostLinkErr),
+	E_P_AUTO(RcvHdrErr),
+	E_P_AUTO(RcvHdrLenErr),
+	E_P_AUTO(RcvBadTidErr),
+	E_P_AUTO(RcvBadVersionErr),
+	E_P_AUTO(RcvIBFlowErr),
+	E_P_AUTO(RcvEBPErr),
+	E_P_AUTO(RcvUnsupportedVLErr),
+	E_P_AUTO(RcvUnexpectedCharErr),
+	E_P_AUTO(RcvShortPktLenErr),
+	E_P_AUTO(RcvLongPktLenErr),
+	E_P_AUTO(RcvMaxPktLenErr),
+	E_P_AUTO(RcvMinPktLenErr),
+	E_P_AUTO(RcvICRCErr),
+	E_P_AUTO(RcvVCRCErr),
+	E_P_AUTO(RcvFormatErr),
+	{ .mask = 0, .sz = 0 }
+};
+
+/*
+ * Below generates "auto-message" for interrupts not specific to any port or
+ * context
+ */
+#define INTR_AUTO(fldname) { .mask = SYM_MASK(IntMask, fldname##Mask), \
+	.msg = #fldname, .sz = sizeof(#fldname) }
+/* Below generates "auto-message" for interrupts specific to a port */
+#define INTR_AUTO_P(fldname) { .mask = MASK_ACROSS(\
+	SYM_LSB(IntMask, fldname##Mask##_0), \
+	SYM_LSB(IntMask, fldname##Mask##_1)), \
+	.msg = #fldname "_P", .sz = sizeof(#fldname "_P") }
+/* For some reason, the SerDesTrimDone bits are reversed */
+#define INTR_AUTO_PI(fldname) { .mask = MASK_ACROSS(\
+	SYM_LSB(IntMask, fldname##Mask##_1), \
+	SYM_LSB(IntMask, fldname##Mask##_0)), \
+	.msg = #fldname "_P", .sz = sizeof(#fldname "_P") }
+/*
+ * Below generates "auto-message" for interrupts specific to a context,
+ * with ctxt-number appended
+ */
+#define INTR_AUTO_C(fldname) { .mask = MASK_ACROSS(\
+	SYM_LSB(IntMask, fldname##0IntMask), \
+	SYM_LSB(IntMask, fldname##17IntMask)), \
+	.msg = #fldname "_C", .sz = sizeof(#fldname "_C") }
+
+static const struct  qib_hwerror_msgs qib_7322_intr_msgs[] = {
+	INTR_AUTO_P(SDmaInt),
+	INTR_AUTO_P(SDmaProgressInt),
+	INTR_AUTO_P(SDmaIdleInt),
+	INTR_AUTO_P(SDmaCleanupDone),
+	INTR_AUTO_C(RcvUrg),
+	INTR_AUTO_P(ErrInt),
+	INTR_AUTO(ErrInt),      /* non-port-specific errs */
+	INTR_AUTO(AssertGPIOInt),
+	INTR_AUTO_P(SendDoneInt),
+	INTR_AUTO(SendBufAvailInt),
+	INTR_AUTO_C(RcvAvail),
+	{ .mask = 0, .sz = 0 }
+};
+
+#define TXSYMPTOM_AUTO_P(fldname) \
+	{ .mask = SYM_MASK(SendHdrErrSymptom_0, fldname), \
+	.msg = #fldname, .sz = sizeof(#fldname) }
+static const struct  qib_hwerror_msgs hdrchk_msgs[] = {
+	TXSYMPTOM_AUTO_P(NonKeyPacket),
+	TXSYMPTOM_AUTO_P(GRHFail),
+	TXSYMPTOM_AUTO_P(PkeyFail),
+	TXSYMPTOM_AUTO_P(QPFail),
+	TXSYMPTOM_AUTO_P(SLIDFail),
+	TXSYMPTOM_AUTO_P(RawIPV6),
+	TXSYMPTOM_AUTO_P(PacketTooSmall),
+	{ .mask = 0, .sz = 0 }
+};
+
+#define IBA7322_HDRHEAD_PKTINT_SHIFT 32 /* interrupt cnt in upper 32 bits */
+
+/*
+ * Called when we might have an error that is specific to a particular
+ * PIO buffer, and may need to cancel that buffer, so it can be re-used,
+ * because we don't need to force the update of pioavail
+ */
+static void qib_disarm_7322_senderrbufs(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 i;
+	int any;
+	u32 piobcnt = dd->piobcnt2k + dd->piobcnt4k + NUM_VL15_BUFS;
+	u32 regcnt = (piobcnt + BITS_PER_LONG - 1) / BITS_PER_LONG;
+	unsigned long sbuf[4];
+
+	/*
+	 * It's possible that sendbuffererror could have bits set; might
+	 * have already done this as a result of hardware error handling.
+	 */
+	any = 0;
+	for (i = 0; i < regcnt; ++i) {
+		sbuf[i] = qib_read_kreg64(dd, kr_sendbuffererror + i);
+		if (sbuf[i]) {
+			any = 1;
+			qib_write_kreg(dd, kr_sendbuffererror + i, sbuf[i]);
+		}
+	}
+
+	if (any)
+		qib_disarm_piobufs_set(dd, sbuf, piobcnt);
+}
+
+/* No txe_recover yet, if ever */
+
+/* No decode__errors yet */
+static void err_decode(char *msg, size_t len, u64 errs,
+		       const struct qib_hwerror_msgs *msp)
+{
+	u64 these, lmask;
+	int took, multi, n = 0;
+
+	while (errs && msp && msp->mask) {
+		multi = (msp->mask & (msp->mask - 1));
+		while (errs & msp->mask) {
+			these = (errs & msp->mask);
+			lmask = (these & (these - 1)) ^ these;
+			if (len) {
+				if (n++) {
+					/* separate the strings */
+					*msg++ = ',';
+					len--;
+				}
+				BUG_ON(!msp->sz);
+				/* msp->sz counts the nul */
+				took = min_t(size_t, msp->sz - (size_t)1, len);
+				memcpy(msg,  msp->msg, took);
+				len -= took;
+				msg += took;
+				if (len)
+					*msg = '\0';
+			}
+			errs &= ~lmask;
+			if (len && multi) {
+				/* More than one bit this mask */
+				int idx = -1;
+
+				while (lmask & msp->mask) {
+					++idx;
+					lmask >>= 1;
+				}
+				took = scnprintf(msg, len, "_%d", idx);
+				len -= took;
+				msg += took;
+			}
+		}
+		++msp;
+	}
+	/* If some bits are left, show in hex. */
+	if (len && errs)
+		snprintf(msg, len, "%sMORE:%llX", n ? "," : "",
+			(unsigned long long) errs);
+}
+
+/* only called if r1 set */
+static void flush_fifo(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 __iomem *piobuf;
+	u32 bufn;
+	u32 *hdr;
+	u64 pbc;
+	const unsigned hdrwords = 7;
+	static struct qib_ib_header ibhdr = {
+		.lrh[0] = cpu_to_be16(0xF000 | QIB_LRH_BTH),
+		.lrh[1] = IB_LID_PERMISSIVE,
+		.lrh[2] = cpu_to_be16(hdrwords + SIZE_OF_CRC),
+		.lrh[3] = IB_LID_PERMISSIVE,
+		.u.oth.bth[0] = cpu_to_be32(
+			(IB_OPCODE_UD_SEND_ONLY << 24) | QIB_DEFAULT_P_KEY),
+		.u.oth.bth[1] = cpu_to_be32(0),
+		.u.oth.bth[2] = cpu_to_be32(0),
+		.u.oth.u.ud.deth[0] = cpu_to_be32(0),
+		.u.oth.u.ud.deth[1] = cpu_to_be32(0),
+	};
+
+	/*
+	 * Send a dummy VL15 packet to flush the launch FIFO.
+	 * This will not actually be sent since the TxeBypassIbc bit is set.
+	 */
+	pbc = PBC_7322_VL15_SEND |
+		(((u64)ppd->hw_pidx) << (PBC_PORT_SEL_LSB + 32)) |
+		(hdrwords + SIZE_OF_CRC);
+	piobuf = qib_7322_getsendbuf(ppd, pbc, &bufn);
+	if (!piobuf)
+		return;
+	writeq(pbc, piobuf);
+	hdr = (u32 *) &ibhdr;
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, hdr, hdrwords - 1);
+		qib_flush_wc();
+		__raw_writel(hdr[hdrwords - 1], piobuf + hdrwords + 1);
+		qib_flush_wc();
+	} else
+		qib_pio_copy(piobuf + 2, hdr, hdrwords);
+	qib_sendbuf_done(dd, bufn);
+}
+
+/*
+ * This is called with interrupts disabled and sdma_lock held.
+ */
+static void qib_7322_sdma_sendctrl(struct qib_pportdata *ppd, unsigned op)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 set_sendctrl = 0;
+	u64 clr_sendctrl = 0;
+
+	if (op & QIB_SDMA_SENDCTRL_OP_ENABLE)
+		set_sendctrl |= SYM_MASK(SendCtrl_0, SDmaEnable);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl_0, SDmaEnable);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_INTENABLE)
+		set_sendctrl |= SYM_MASK(SendCtrl_0, SDmaIntEnable);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl_0, SDmaIntEnable);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_HALT)
+		set_sendctrl |= SYM_MASK(SendCtrl_0, SDmaHalt);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl_0, SDmaHalt);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_DRAIN)
+		set_sendctrl |= SYM_MASK(SendCtrl_0, TxeBypassIbc) |
+				SYM_MASK(SendCtrl_0, TxeAbortIbc) |
+				SYM_MASK(SendCtrl_0, TxeDrainRmFifo);
+	else
+		clr_sendctrl |= SYM_MASK(SendCtrl_0, TxeBypassIbc) |
+				SYM_MASK(SendCtrl_0, TxeAbortIbc) |
+				SYM_MASK(SendCtrl_0, TxeDrainRmFifo);
+
+	spin_lock(&dd->sendctrl_lock);
+
+	/* If we are draining everything, block sends first */
+	if (op & QIB_SDMA_SENDCTRL_OP_DRAIN) {
+		ppd->p_sendctrl &= ~SYM_MASK(SendCtrl_0, SendEnable);
+		qib_write_kreg_port(ppd, krp_sendctrl, ppd->p_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	ppd->p_sendctrl |= set_sendctrl;
+	ppd->p_sendctrl &= ~clr_sendctrl;
+
+	if (op & QIB_SDMA_SENDCTRL_OP_CLEANUP)
+		qib_write_kreg_port(ppd, krp_sendctrl,
+				    ppd->p_sendctrl |
+				    SYM_MASK(SendCtrl_0, SDmaCleanup));
+	else
+		qib_write_kreg_port(ppd, krp_sendctrl, ppd->p_sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	if (op & QIB_SDMA_SENDCTRL_OP_DRAIN) {
+		ppd->p_sendctrl |= SYM_MASK(SendCtrl_0, SendEnable);
+		qib_write_kreg_port(ppd, krp_sendctrl, ppd->p_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	spin_unlock(&dd->sendctrl_lock);
+
+	if ((op & QIB_SDMA_SENDCTRL_OP_DRAIN) && ppd->dd->cspec->r1)
+		flush_fifo(ppd);
+}
+
+static void qib_7322_sdma_hw_clean_up(struct qib_pportdata *ppd)
+{
+	__qib_sdma_process_event(ppd, qib_sdma_event_e50_hw_cleaned);
+}
+
+static void qib_sdma_7322_setlengen(struct qib_pportdata *ppd)
+{
+	/*
+	 * Set SendDmaLenGen and clear and set
+	 * the MSB of the generation count to enable generation checking
+	 * and load the internal generation counter.
+	 */
+	qib_write_kreg_port(ppd, krp_senddmalengen, ppd->sdma_descq_cnt);
+	qib_write_kreg_port(ppd, krp_senddmalengen,
+			    ppd->sdma_descq_cnt |
+			    (1ULL << QIB_7322_SendDmaLenGen_0_Generation_MSB));
+}
+
+/*
+ * Must be called with sdma_lock held, or before init finished.
+ */
+static void qib_sdma_update_7322_tail(struct qib_pportdata *ppd, u16 tail)
+{
+	/* Commit writes to memory and advance the tail on the chip */
+	wmb();
+	ppd->sdma_descq_tail = tail;
+	qib_write_kreg_port(ppd, krp_senddmatail, tail);
+}
+
+/*
+ * This is called with interrupts disabled and sdma_lock held.
+ */
+static void qib_7322_sdma_hw_start_up(struct qib_pportdata *ppd)
+{
+	/*
+	 * Drain all FIFOs.
+	 * The hardware doesn't require this but we do it so that verbs
+	 * and user applications don't wait for link active to send stale
+	 * data.
+	 */
+	sendctrl_7322_mod(ppd, QIB_SENDCTRL_FLUSH);
+
+	qib_sdma_7322_setlengen(ppd);
+	qib_sdma_update_7322_tail(ppd, 0); /* Set SendDmaTail */
+	ppd->sdma_head_dma[0] = 0;
+	qib_7322_sdma_sendctrl(ppd,
+		ppd->sdma_state.current_op | QIB_SDMA_SENDCTRL_OP_CLEANUP);
+}
+
+#define DISABLES_SDMA ( \
+	QIB_E_P_SDMAHALT | \
+	QIB_E_P_SDMADESCADDRMISALIGN | \
+	QIB_E_P_SDMAMISSINGDW | \
+	QIB_E_P_SDMADWEN | \
+	QIB_E_P_SDMARPYTAG | \
+	QIB_E_P_SDMA1STDESC | \
+	QIB_E_P_SDMABASE | \
+	QIB_E_P_SDMATAILOUTOFBOUND | \
+	QIB_E_P_SDMAOUTOFBOUND | \
+	QIB_E_P_SDMAGENMISMATCH)
+
+static void sdma_7322_p_errors(struct qib_pportdata *ppd, u64 errs)
+{
+	unsigned long flags;
+	struct qib_devdata *dd = ppd->dd;
+
+	errs &= QIB_E_P_SDMAERRS;
+	err_decode(ppd->cpspec->sdmamsgbuf, sizeof(ppd->cpspec->sdmamsgbuf),
+		   errs, qib_7322p_error_msgs);
+
+	if (errs & QIB_E_P_SDMAUNEXPDATA)
+		qib_dev_err(dd, "IB%u:%u SDmaUnexpData\n", dd->unit,
+			    ppd->port);
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	if (errs != QIB_E_P_SDMAHALT) {
+		/* SDMA errors have QIB_E_P_SDMAHALT and another bit set */
+		qib_dev_porterr(dd, ppd->port,
+			"SDMA %s 0x%016llx %s\n",
+			qib_sdma_state_names[ppd->sdma_state.current_state],
+			errs, ppd->cpspec->sdmamsgbuf);
+		dump_sdma_7322_state(ppd);
+	}
+
+	switch (ppd->sdma_state.current_state) {
+	case qib_sdma_state_s00_hw_down:
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		if (errs & QIB_E_P_SDMAHALT)
+			__qib_sdma_process_event(ppd,
+				qib_sdma_event_e20_hw_started);
+		break;
+
+	case qib_sdma_state_s20_idle:
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		if (errs & QIB_E_P_SDMAHALT)
+			__qib_sdma_process_event(ppd,
+				qib_sdma_event_e50_hw_cleaned);
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		if (errs & QIB_E_P_SDMAHALT)
+			__qib_sdma_process_event(ppd,
+				qib_sdma_event_e60_hw_halted);
+		break;
+
+	case qib_sdma_state_s99_running:
+		__qib_sdma_process_event(ppd, qib_sdma_event_e7322_err_halted);
+		__qib_sdma_process_event(ppd, qib_sdma_event_e60_hw_halted);
+		break;
+	}
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+/*
+ * handle per-device errors (not per-port errors)
+ */
+static noinline void handle_7322_errors(struct qib_devdata *dd)
+{
+	char *msg;
+	u64 iserr = 0;
+	u64 errs;
+	u64 mask;
+	int log_idx;
+
+	qib_stats.sps_errints++;
+	errs = qib_read_kreg64(dd, kr_errstatus);
+	if (!errs) {
+		qib_devinfo(dd->pcidev,
+			"device error interrupt, but no error bits set!\n");
+		goto done;
+	}
+
+	/* don't report errors that are masked */
+	errs &= dd->cspec->errormask;
+	msg = dd->cspec->emsgbuf;
+
+	/* do these first, they are most important */
+	if (errs & QIB_E_HARDWARE) {
+		*msg = '\0';
+		qib_7322_handle_hwerrors(dd, msg, sizeof dd->cspec->emsgbuf);
+	} else
+		for (log_idx = 0; log_idx < QIB_EEP_LOG_CNT; ++log_idx)
+			if (errs & dd->eep_st_masks[log_idx].errs_to_log)
+				qib_inc_eeprom_err(dd, log_idx, 1);
+
+	if (errs & QIB_E_SPKTERRS) {
+		qib_disarm_7322_senderrbufs(dd->pport);
+		qib_stats.sps_txerrs++;
+	} else if (errs & QIB_E_INVALIDADDR)
+		qib_stats.sps_txerrs++;
+	else if (errs & QIB_E_ARMLAUNCH) {
+		qib_stats.sps_txerrs++;
+		qib_disarm_7322_senderrbufs(dd->pport);
+	}
+	qib_write_kreg(dd, kr_errclear, errs);
+
+	/*
+	 * The ones we mask off are handled specially below
+	 * or above.  Also mask SDMADISABLED by default as it
+	 * is too chatty.
+	 */
+	mask = QIB_E_HARDWARE;
+	*msg = '\0';
+
+	err_decode(msg, sizeof dd->cspec->emsgbuf, errs & ~mask,
+		   qib_7322error_msgs);
+
+	/*
+	 * Getting reset is a tragedy for all ports. Mark the device
+	 * _and_ the ports as "offline" in way meaningful to each.
+	 */
+	if (errs & QIB_E_RESET) {
+		int pidx;
+
+		qib_dev_err(dd,
+			"Got reset, requires re-init (unload and reload driver)\n");
+		dd->flags &= ~QIB_INITTED;  /* needs re-init */
+		/* mark as having had error */
+		*dd->devstatusp |= QIB_STATUS_HWERROR;
+		for (pidx = 0; pidx < dd->num_pports; ++pidx)
+			if (dd->pport[pidx].link_speed_supported)
+				*dd->pport[pidx].statusp &= ~QIB_STATUS_IB_CONF;
+	}
+
+	if (*msg && iserr)
+		qib_dev_err(dd, "%s error\n", msg);
+
+	/*
+	 * If there were hdrq or egrfull errors, wake up any processes
+	 * waiting in poll.  We used to try to check which contexts had
+	 * the overflow, but given the cost of that and the chip reads
+	 * to support it, it's better to just wake everybody up if we
+	 * get an overflow; waiters can poll again if it's not them.
+	 */
+	if (errs & (ERR_MASK(RcvEgrFullErr) | ERR_MASK(RcvHdrFullErr))) {
+		qib_handle_urcv(dd, ~0U);
+		if (errs & ERR_MASK(RcvEgrFullErr))
+			qib_stats.sps_buffull++;
+		else
+			qib_stats.sps_hdrfull++;
+	}
+
+done:
+	return;
+}
+
+static void qib_error_tasklet(unsigned long data)
+{
+	struct qib_devdata *dd = (struct qib_devdata *)data;
+
+	handle_7322_errors(dd);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+}
+
+static void reenable_chase(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	ppd->cpspec->chase_timer.expires = 0;
+	qib_set_ib_7322_lstate(ppd, QLOGIC_IB_IBCC_LINKCMD_DOWN,
+		QLOGIC_IB_IBCC_LINKINITCMD_POLL);
+}
+
+static void disable_chase(struct qib_pportdata *ppd, unsigned long tnow,
+		u8 ibclt)
+{
+	ppd->cpspec->chase_end = 0;
+
+	if (!qib_chase)
+		return;
+
+	qib_set_ib_7322_lstate(ppd, QLOGIC_IB_IBCC_LINKCMD_DOWN,
+		QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+	ppd->cpspec->chase_timer.expires = jiffies + QIB_CHASE_DIS_TIME;
+	add_timer(&ppd->cpspec->chase_timer);
+}
+
+static void handle_serdes_issues(struct qib_pportdata *ppd, u64 ibcst)
+{
+	u8 ibclt;
+	unsigned long tnow;
+
+	ibclt = (u8)SYM_FIELD(ibcst, IBCStatusA_0, LinkTrainingState);
+
+	/*
+	 * Detect and handle the state chase issue, where we can
+	 * get stuck if we are unlucky on timing on both sides of
+	 * the link.   If we are, we disable, set a timer, and
+	 * then re-enable.
+	 */
+	switch (ibclt) {
+	case IB_7322_LT_STATE_CFGRCVFCFG:
+	case IB_7322_LT_STATE_CFGWAITRMT:
+	case IB_7322_LT_STATE_TXREVLANES:
+	case IB_7322_LT_STATE_CFGENH:
+		tnow = jiffies;
+		if (ppd->cpspec->chase_end &&
+		     time_after(tnow, ppd->cpspec->chase_end))
+			disable_chase(ppd, tnow, ibclt);
+		else if (!ppd->cpspec->chase_end)
+			ppd->cpspec->chase_end = tnow + QIB_CHASE_TIME;
+		break;
+	default:
+		ppd->cpspec->chase_end = 0;
+		break;
+	}
+
+	if (((ibclt >= IB_7322_LT_STATE_CFGTEST &&
+	      ibclt <= IB_7322_LT_STATE_CFGWAITENH) ||
+	     ibclt == IB_7322_LT_STATE_LINKUP) &&
+	    (ibcst & SYM_MASK(IBCStatusA_0, LinkSpeedQDR))) {
+		force_h1(ppd);
+		ppd->cpspec->qdr_reforce = 1;
+		if (!ppd->dd->cspec->r1)
+			serdes_7322_los_enable(ppd, 0);
+	} else if (ppd->cpspec->qdr_reforce &&
+		(ibcst & SYM_MASK(IBCStatusA_0, LinkSpeedQDR)) &&
+		 (ibclt == IB_7322_LT_STATE_CFGENH ||
+		ibclt == IB_7322_LT_STATE_CFGIDLE ||
+		ibclt == IB_7322_LT_STATE_LINKUP))
+		force_h1(ppd);
+
+	if ((IS_QMH(ppd->dd) || IS_QME(ppd->dd)) &&
+	    ppd->link_speed_enabled == QIB_IB_QDR &&
+	    (ibclt == IB_7322_LT_STATE_CFGTEST ||
+	     ibclt == IB_7322_LT_STATE_CFGENH ||
+	     (ibclt >= IB_7322_LT_STATE_POLLACTIVE &&
+	      ibclt <= IB_7322_LT_STATE_SLEEPQUIET)))
+		adj_tx_serdes(ppd);
+
+	if (ibclt != IB_7322_LT_STATE_LINKUP) {
+		u8 ltstate = qib_7322_phys_portstate(ibcst);
+		u8 pibclt = (u8)SYM_FIELD(ppd->lastibcstat, IBCStatusA_0,
+					  LinkTrainingState);
+		if (!ppd->dd->cspec->r1 &&
+		    pibclt == IB_7322_LT_STATE_LINKUP &&
+		    ltstate != IB_PHYSPORTSTATE_LINK_ERR_RECOVER &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_RETRAIN &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_WAITRMT &&
+		    ltstate != IB_PHYSPORTSTATE_RECOVERY_IDLE)
+			/* If the link went down (but no into recovery,
+			 * turn LOS back on */
+			serdes_7322_los_enable(ppd, 1);
+		if (!ppd->cpspec->qdr_dfe_on &&
+		    ibclt <= IB_7322_LT_STATE_SLEEPQUIET) {
+			ppd->cpspec->qdr_dfe_on = 1;
+			ppd->cpspec->qdr_dfe_time = 0;
+			/* On link down, reenable QDR adaptation */
+			qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+					    ppd->dd->cspec->r1 ?
+					    QDR_STATIC_ADAPT_DOWN_R1 :
+					    QDR_STATIC_ADAPT_DOWN);
+			pr_info(
+				"IB%u:%u re-enabled QDR adaptation ibclt %x\n",
+				ppd->dd->unit, ppd->port, ibclt);
+		}
+	}
+}
+
+static int qib_7322_set_ib_cfg(struct qib_pportdata *, int, u32);
+
+/*
+ * This is per-pport error handling.
+ * will likely get it's own MSIx interrupt (one for each port,
+ * although just a single handler).
+ */
+static noinline void handle_7322_p_errors(struct qib_pportdata *ppd)
+{
+	char *msg;
+	u64 ignore_this_time = 0, iserr = 0, errs, fmask;
+	struct qib_devdata *dd = ppd->dd;
+
+	/* do this as soon as possible */
+	fmask = qib_read_kreg64(dd, kr_act_fmask);
+	if (!fmask)
+		check_7322_rxe_status(ppd);
+
+	errs = qib_read_kreg_port(ppd, krp_errstatus);
+	if (!errs)
+		qib_devinfo(dd->pcidev,
+			 "Port%d error interrupt, but no error bits set!\n",
+			 ppd->port);
+	if (!fmask)
+		errs &= ~QIB_E_P_IBSTATUSCHANGED;
+	if (!errs)
+		goto done;
+
+	msg = ppd->cpspec->epmsgbuf;
+	*msg = '\0';
+
+	if (errs & ~QIB_E_P_BITSEXTANT) {
+		err_decode(msg, sizeof ppd->cpspec->epmsgbuf,
+			   errs & ~QIB_E_P_BITSEXTANT, qib_7322p_error_msgs);
+		if (!*msg)
+			snprintf(msg, sizeof ppd->cpspec->epmsgbuf,
+				 "no others");
+		qib_dev_porterr(dd, ppd->port,
+			"error interrupt with unknown errors 0x%016Lx set (and %s)\n",
+			(errs & ~QIB_E_P_BITSEXTANT), msg);
+		*msg = '\0';
+	}
+
+	if (errs & QIB_E_P_SHDR) {
+		u64 symptom;
+
+		/* determine cause, then write to clear */
+		symptom = qib_read_kreg_port(ppd, krp_sendhdrsymptom);
+		qib_write_kreg_port(ppd, krp_sendhdrsymptom, 0);
+		err_decode(msg, sizeof ppd->cpspec->epmsgbuf, symptom,
+			   hdrchk_msgs);
+		*msg = '\0';
+		/* senderrbuf cleared in SPKTERRS below */
+	}
+
+	if (errs & QIB_E_P_SPKTERRS) {
+		if ((errs & QIB_E_P_LINK_PKTERRS) &&
+		    !(ppd->lflags & QIBL_LINKACTIVE)) {
+			/*
+			 * This can happen when trying to bring the link
+			 * up, but the IB link changes state at the "wrong"
+			 * time. The IB logic then complains that the packet
+			 * isn't valid.  We don't want to confuse people, so
+			 * we just don't print them, except at debug
+			 */
+			err_decode(msg, sizeof ppd->cpspec->epmsgbuf,
+				   (errs & QIB_E_P_LINK_PKTERRS),
+				   qib_7322p_error_msgs);
+			*msg = '\0';
+			ignore_this_time = errs & QIB_E_P_LINK_PKTERRS;
+		}
+		qib_disarm_7322_senderrbufs(ppd);
+	} else if ((errs & QIB_E_P_LINK_PKTERRS) &&
+		   !(ppd->lflags & QIBL_LINKACTIVE)) {
+		/*
+		 * This can happen when SMA is trying to bring the link
+		 * up, but the IB link changes state at the "wrong" time.
+		 * The IB logic then complains that the packet isn't
+		 * valid.  We don't want to confuse people, so we just
+		 * don't print them, except at debug
+		 */
+		err_decode(msg, sizeof ppd->cpspec->epmsgbuf, errs,
+			   qib_7322p_error_msgs);
+		ignore_this_time = errs & QIB_E_P_LINK_PKTERRS;
+		*msg = '\0';
+	}
+
+	qib_write_kreg_port(ppd, krp_errclear, errs);
+
+	errs &= ~ignore_this_time;
+	if (!errs)
+		goto done;
+
+	if (errs & QIB_E_P_RPKTERRS)
+		qib_stats.sps_rcverrs++;
+	if (errs & QIB_E_P_SPKTERRS)
+		qib_stats.sps_txerrs++;
+
+	iserr = errs & ~(QIB_E_P_RPKTERRS | QIB_E_P_PKTERRS);
+
+	if (errs & QIB_E_P_SDMAERRS)
+		sdma_7322_p_errors(ppd, errs);
+
+	if (errs & QIB_E_P_IBSTATUSCHANGED) {
+		u64 ibcs;
+		u8 ltstate;
+
+		ibcs = qib_read_kreg_port(ppd, krp_ibcstatus_a);
+		ltstate = qib_7322_phys_portstate(ibcs);
+
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+			handle_serdes_issues(ppd, ibcs);
+		if (!(ppd->cpspec->ibcctrl_a &
+		      SYM_MASK(IBCCtrlA_0, IBStatIntReductionEn))) {
+			/*
+			 * We got our interrupt, so init code should be
+			 * happy and not try alternatives. Now squelch
+			 * other "chatter" from link-negotiation (pre Init)
+			 */
+			ppd->cpspec->ibcctrl_a |=
+				SYM_MASK(IBCCtrlA_0, IBStatIntReductionEn);
+			qib_write_kreg_port(ppd, krp_ibcctrl_a,
+					    ppd->cpspec->ibcctrl_a);
+		}
+
+		/* Update our picture of width and speed from chip */
+		ppd->link_width_active =
+			(ibcs & SYM_MASK(IBCStatusA_0, LinkWidthActive)) ?
+			    IB_WIDTH_4X : IB_WIDTH_1X;
+		ppd->link_speed_active = (ibcs & SYM_MASK(IBCStatusA_0,
+			LinkSpeedQDR)) ? QIB_IB_QDR : (ibcs &
+			  SYM_MASK(IBCStatusA_0, LinkSpeedActive)) ?
+				   QIB_IB_DDR : QIB_IB_SDR;
+
+		if ((ppd->lflags & QIBL_IB_LINK_DISABLED) && ltstate !=
+		    IB_PHYSPORTSTATE_DISABLED)
+			qib_set_ib_7322_lstate(ppd, 0,
+			       QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+		else
+			/*
+			 * Since going into a recovery state causes the link
+			 * state to go down and since recovery is transitory,
+			 * it is better if we "miss" ever seeing the link
+			 * training state go into recovery (i.e., ignore this
+			 * transition for link state special handling purposes)
+			 * without updating lastibcstat.
+			 */
+			if (ltstate != IB_PHYSPORTSTATE_LINK_ERR_RECOVER &&
+			    ltstate != IB_PHYSPORTSTATE_RECOVERY_RETRAIN &&
+			    ltstate != IB_PHYSPORTSTATE_RECOVERY_WAITRMT &&
+			    ltstate != IB_PHYSPORTSTATE_RECOVERY_IDLE)
+				qib_handle_e_ibstatuschanged(ppd, ibcs);
+	}
+	if (*msg && iserr)
+		qib_dev_porterr(dd, ppd->port, "%s error\n", msg);
+
+	if (ppd->state_wanted & ppd->lflags)
+		wake_up_interruptible(&ppd->state_wait);
+done:
+	return;
+}
+
+/* enable/disable chip from delivering interrupts */
+static void qib_7322_set_intr_state(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		if (dd->flags & QIB_BADINTR)
+			return;
+		qib_write_kreg(dd, kr_intmask, dd->cspec->int_enable_mask);
+		/* cause any pending enabled interrupts to be re-delivered */
+		qib_write_kreg(dd, kr_intclear, 0ULL);
+		if (dd->cspec->num_msix_entries) {
+			/* and same for MSIx */
+			u64 val = qib_read_kreg64(dd, kr_intgranted);
+			if (val)
+				qib_write_kreg(dd, kr_intgranted, val);
+		}
+	} else
+		qib_write_kreg(dd, kr_intmask, 0ULL);
+}
+
+/*
+ * Try to cleanup as much as possible for anything that might have gone
+ * wrong while in freeze mode, such as pio buffers being written by user
+ * processes (causing armlaunch), send errors due to going into freeze mode,
+ * etc., and try to avoid causing extra interrupts while doing so.
+ * Forcibly update the in-memory pioavail register copies after cleanup
+ * because the chip won't do it while in freeze mode (the register values
+ * themselves are kept correct).
+ * Make sure that we don't lose any important interrupts by using the chip
+ * feature that says that writing 0 to a bit in *clear that is set in
+ * *status will cause an interrupt to be generated again (if allowed by
+ * the *mask value).
+ * This is in chip-specific code because of all of the register accesses,
+ * even though the details are similar on most chips.
+ */
+static void qib_7322_clear_freeze(struct qib_devdata *dd)
+{
+	int pidx;
+
+	/* disable error interrupts, to avoid confusion */
+	qib_write_kreg(dd, kr_errmask, 0ULL);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		if (dd->pport[pidx].link_speed_supported)
+			qib_write_kreg_port(dd->pport + pidx, krp_errmask,
+					    0ULL);
+
+	/* also disable interrupts; errormask is sometimes overwriten */
+	qib_7322_set_intr_state(dd, 0);
+
+	/* clear the freeze, and be sure chip saw it */
+	qib_write_kreg(dd, kr_control, dd->control);
+	qib_read_kreg32(dd, kr_scratch);
+
+	/*
+	 * Force new interrupt if any hwerr, error or interrupt bits are
+	 * still set, and clear "safe" send packet errors related to freeze
+	 * and cancelling sends.  Re-enable error interrupts before possible
+	 * force of re-interrupt on pending interrupts.
+	 */
+	qib_write_kreg(dd, kr_hwerrclear, 0ULL);
+	qib_write_kreg(dd, kr_errclear, E_SPKT_ERRS_IGNORE);
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+	/* We need to purge per-port errs and reset mask, too */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		if (!dd->pport[pidx].link_speed_supported)
+			continue;
+		qib_write_kreg_port(dd->pport + pidx, krp_errclear, ~0Ull);
+		qib_write_kreg_port(dd->pport + pidx, krp_errmask, ~0Ull);
+	}
+	qib_7322_set_intr_state(dd, 1);
+}
+
+/* no error handling to speak of */
+/**
+ * qib_7322_handle_hwerrors - display hardware errors.
+ * @dd: the qlogic_ib device
+ * @msg: the output buffer
+ * @msgl: the size of the output buffer
+ *
+ * Use same msg buffer as regular errors to avoid excessive stack
+ * use.  Most hardware errors are catastrophic, but for right now,
+ * we'll print them and continue.  We reuse the same message buffer as
+ * qib_handle_errors() to avoid excessive stack usage.
+ */
+static void qib_7322_handle_hwerrors(struct qib_devdata *dd, char *msg,
+				     size_t msgl)
+{
+	u64 hwerrs;
+	u32 ctrl;
+	int isfatal = 0;
+
+	hwerrs = qib_read_kreg64(dd, kr_hwerrstatus);
+	if (!hwerrs)
+		goto bail;
+	if (hwerrs == ~0ULL) {
+		qib_dev_err(dd,
+			"Read of hardware error status failed (all bits set); ignoring\n");
+		goto bail;
+	}
+	qib_stats.sps_hwerrs++;
+
+	/* Always clear the error status register, except BIST fail */
+	qib_write_kreg(dd, kr_hwerrclear, hwerrs &
+		       ~HWE_MASK(PowerOnBISTFailed));
+
+	hwerrs &= dd->cspec->hwerrmask;
+
+	/* no EEPROM logging, yet */
+
+	if (hwerrs)
+		qib_devinfo(dd->pcidev,
+			"Hardware error: hwerr=0x%llx (cleared)\n",
+			(unsigned long long) hwerrs);
+
+	ctrl = qib_read_kreg32(dd, kr_control);
+	if ((ctrl & SYM_MASK(Control, FreezeMode)) && !dd->diag_client) {
+		/*
+		 * No recovery yet...
+		 */
+		if ((hwerrs & ~HWE_MASK(LATriggered)) ||
+		    dd->cspec->stay_in_freeze) {
+			/*
+			 * If any set that we aren't ignoring only make the
+			 * complaint once, in case it's stuck or recurring,
+			 * and we get here multiple times
+			 * Force link down, so switch knows, and
+			 * LEDs are turned off.
+			 */
+			if (dd->flags & QIB_INITTED)
+				isfatal = 1;
+		} else
+			qib_7322_clear_freeze(dd);
+	}
+
+	if (hwerrs & HWE_MASK(PowerOnBISTFailed)) {
+		isfatal = 1;
+		strlcpy(msg,
+			"[Memory BIST test failed, InfiniPath hardware unusable]",
+			msgl);
+		/* ignore from now on, so disable until driver reloaded */
+		dd->cspec->hwerrmask &= ~HWE_MASK(PowerOnBISTFailed);
+		qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+	}
+
+	err_decode(msg, msgl, hwerrs, qib_7322_hwerror_msgs);
+
+	/* Ignore esoteric PLL failures et al. */
+
+	qib_dev_err(dd, "%s hardware error\n", msg);
+
+	if (hwerrs &
+		   (SYM_MASK(HwErrMask, SDmaMemReadErrMask_0) |
+		    SYM_MASK(HwErrMask, SDmaMemReadErrMask_1))) {
+		int pidx = 0;
+		int err;
+		unsigned long flags;
+		struct qib_pportdata *ppd = dd->pport;
+		for (; pidx < dd->num_pports; ++pidx, ppd++) {
+			err = 0;
+			if (pidx == 0 && (hwerrs &
+				SYM_MASK(HwErrMask, SDmaMemReadErrMask_0)))
+				err++;
+			if (pidx == 1 && (hwerrs &
+				SYM_MASK(HwErrMask, SDmaMemReadErrMask_1)))
+				err++;
+			if (err) {
+				spin_lock_irqsave(&ppd->sdma_lock, flags);
+				dump_sdma_7322_state(ppd);
+				spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+			}
+		}
+	}
+
+	if (isfatal && !dd->diag_client) {
+		qib_dev_err(dd,
+			"Fatal Hardware Error, no longer usable, SN %.16s\n",
+			dd->serial);
+		/*
+		 * for /sys status file and user programs to print; if no
+		 * trailing brace is copied, we'll know it was truncated.
+		 */
+		if (dd->freezemsg)
+			snprintf(dd->freezemsg, dd->freezelen,
+				 "{%s}", msg);
+		qib_disable_after_error(dd);
+	}
+bail:;
+}
+
+/**
+ * qib_7322_init_hwerrors - enable hardware errors
+ * @dd: the qlogic_ib device
+ *
+ * now that we have finished initializing everything that might reasonably
+ * cause a hardware error, and cleared those errors bits as they occur,
+ * we can enable hardware errors in the mask (potentially enabling
+ * freeze mode), and enable hardware errors as errors (along with
+ * everything else) in errormask
+ */
+static void qib_7322_init_hwerrors(struct qib_devdata *dd)
+{
+	int pidx;
+	u64 extsval;
+
+	extsval = qib_read_kreg64(dd, kr_extstatus);
+	if (!(extsval & (QIB_EXTS_MEMBIST_DISABLED |
+			 QIB_EXTS_MEMBIST_ENDTEST)))
+		qib_dev_err(dd, "MemBIST did not complete!\n");
+
+	/* never clear BIST failure, so reported on each driver load */
+	qib_write_kreg(dd, kr_hwerrclear, ~HWE_MASK(PowerOnBISTFailed));
+	qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+
+	/* clear all */
+	qib_write_kreg(dd, kr_errclear, ~0ULL);
+	/* enable errors that are masked, at least this first time. */
+	qib_write_kreg(dd, kr_errmask, ~0ULL);
+	dd->cspec->errormask = qib_read_kreg64(dd, kr_errmask);
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		if (dd->pport[pidx].link_speed_supported)
+			qib_write_kreg_port(dd->pport + pidx, krp_errmask,
+					    ~0ULL);
+}
+
+/*
+ * Disable and enable the armlaunch error.  Used for PIO bandwidth testing
+ * on chips that are count-based, rather than trigger-based.  There is no
+ * reference counting, but that's also fine, given the intended use.
+ * Only chip-specific because it's all register accesses
+ */
+static void qib_set_7322_armlaunch(struct qib_devdata *dd, u32 enable)
+{
+	if (enable) {
+		qib_write_kreg(dd, kr_errclear, QIB_E_SPIOARMLAUNCH);
+		dd->cspec->errormask |= QIB_E_SPIOARMLAUNCH;
+	} else
+		dd->cspec->errormask &= ~QIB_E_SPIOARMLAUNCH;
+	qib_write_kreg(dd, kr_errmask, dd->cspec->errormask);
+}
+
+/*
+ * Formerly took parameter <which> in pre-shifted,
+ * pre-merged form with LinkCmd and LinkInitCmd
+ * together, and assuming the zero was NOP.
+ */
+static void qib_set_ib_7322_lstate(struct qib_pportdata *ppd, u16 linkcmd,
+				   u16 linitcmd)
+{
+	u64 mod_wd;
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+
+	if (linitcmd == QLOGIC_IB_IBCC_LINKINITCMD_DISABLE) {
+		/*
+		 * If we are told to disable, note that so link-recovery
+		 * code does not attempt to bring us back up.
+		 * Also reset everything that we can, so we start
+		 * completely clean when re-enabled (before we
+		 * actually issue the disable to the IBC)
+		 */
+		qib_7322_mini_pcs_reset(ppd);
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	} else if (linitcmd || linkcmd == QLOGIC_IB_IBCC_LINKCMD_DOWN) {
+		/*
+		 * Any other linkinitcmd will lead to LINKDOWN and then
+		 * to INIT (if all is well), so clear flag to let
+		 * link-recovery code attempt to bring us back up.
+		 */
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		/*
+		 * Clear status change interrupt reduction so the
+		 * new state is seen.
+		 */
+		ppd->cpspec->ibcctrl_a &=
+			~SYM_MASK(IBCCtrlA_0, IBStatIntReductionEn);
+	}
+
+	mod_wd = (linkcmd << IBA7322_IBCC_LINKCMD_SHIFT) |
+		(linitcmd << QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+
+	qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a |
+			    mod_wd);
+	/* write to chip to prevent back-to-back writes of ibc reg */
+	qib_write_kreg(dd, kr_scratch, 0);
+
+}
+
+/*
+ * The total RCV buffer memory is 64KB, used for both ports, and is
+ * in units of 64 bytes (same as IB flow control credit unit).
+ * The consumedVL unit in the same registers are in 32 byte units!
+ * So, a VL15 packet needs 4.50 IB credits, and 9 rx buffer chunks,
+ * and we can therefore allocate just 9 IB credits for 2 VL15 packets
+ * in krp_rxcreditvl15, rather than 10.
+ */
+#define RCV_BUF_UNITSZ 64
+#define NUM_RCV_BUF_UNITS(dd) ((64 * 1024) / (RCV_BUF_UNITSZ * dd->num_pports))
+
+static void set_vls(struct qib_pportdata *ppd)
+{
+	int i, numvls, totcred, cred_vl, vl0extra;
+	struct qib_devdata *dd = ppd->dd;
+	u64 val;
+
+	numvls = qib_num_vls(ppd->vls_operational);
+
+	/*
+	 * Set up per-VL credits. Below is kluge based on these assumptions:
+	 * 1) port is disabled at the time early_init is called.
+	 * 2) give VL15 17 credits, for two max-plausible packets.
+	 * 3) Give VL0-N the rest, with any rounding excess used for VL0
+	 */
+	/* 2 VL15 packets @ 288 bytes each (including IB headers) */
+	totcred = NUM_RCV_BUF_UNITS(dd);
+	cred_vl = (2 * 288 + RCV_BUF_UNITSZ - 1) / RCV_BUF_UNITSZ;
+	totcred -= cred_vl;
+	qib_write_kreg_port(ppd, krp_rxcreditvl15, (u64) cred_vl);
+	cred_vl = totcred / numvls;
+	vl0extra = totcred - cred_vl * numvls;
+	qib_write_kreg_port(ppd, krp_rxcreditvl0, cred_vl + vl0extra);
+	for (i = 1; i < numvls; i++)
+		qib_write_kreg_port(ppd, krp_rxcreditvl0 + i, cred_vl);
+	for (; i < 8; i++) /* no buffer space for other VLs */
+		qib_write_kreg_port(ppd, krp_rxcreditvl0 + i, 0);
+
+	/* Notify IBC that credits need to be recalculated */
+	val = qib_read_kreg_port(ppd, krp_ibsdtestiftx);
+	val |= SYM_MASK(IB_SDTEST_IF_TX_0, CREDIT_CHANGE);
+	qib_write_kreg_port(ppd, krp_ibsdtestiftx, val);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+	val &= ~SYM_MASK(IB_SDTEST_IF_TX_0, CREDIT_CHANGE);
+	qib_write_kreg_port(ppd, krp_ibsdtestiftx, val);
+
+	for (i = 0; i < numvls; i++)
+		val = qib_read_kreg_port(ppd, krp_rxcreditvl0 + i);
+	val = qib_read_kreg_port(ppd, krp_rxcreditvl15);
+
+	/* Change the number of operational VLs */
+	ppd->cpspec->ibcctrl_a = (ppd->cpspec->ibcctrl_a &
+				~SYM_MASK(IBCCtrlA_0, NumVLane)) |
+		((u64)(numvls - 1) << SYM_LSB(IBCCtrlA_0, NumVLane));
+	qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+}
+
+/*
+ * The code that deals with actual SerDes is in serdes_7322_init().
+ * Compared to the code for iba7220, it is minimal.
+ */
+static int serdes_7322_init(struct qib_pportdata *ppd);
+
+/**
+ * qib_7322_bringup_serdes - bring up the serdes
+ * @ppd: physical port on the qlogic_ib device
+ */
+static int qib_7322_bringup_serdes(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 val, guid, ibc;
+	unsigned long flags;
+	int ret = 0;
+
+	/*
+	 * SerDes model not in Pd, but still need to
+	 * set up much of IBCCtrl and IBCDDRCtrl; move elsewhere
+	 * eventually.
+	 */
+	/* Put IBC in reset, sends disabled (should be in reset already) */
+	ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0, IBLinkEn);
+	qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+
+	/* ensure previous Tx parameters are not still forced */
+	qib_write_kreg_port(ppd, krp_tx_deemph_override,
+		SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		reset_tx_deemphasis_override));
+
+	if (qib_compat_ddr_negotiate) {
+		ppd->cpspec->ibdeltainprog = 1;
+		ppd->cpspec->ibsymsnap = read_7322_creg32_port(ppd,
+						crp_ibsymbolerr);
+		ppd->cpspec->iblnkerrsnap = read_7322_creg32_port(ppd,
+						crp_iblinkerrrecov);
+	}
+
+	/* flowcontrolwatermark is in units of KBytes */
+	ibc = 0x5ULL << SYM_LSB(IBCCtrlA_0, FlowCtrlWaterMark);
+	/*
+	 * Flow control is sent this often, even if no changes in
+	 * buffer space occur.  Units are 128ns for this chip.
+	 * Set to 3usec.
+	 */
+	ibc |= 24ULL << SYM_LSB(IBCCtrlA_0, FlowCtrlPeriod);
+	/* max error tolerance */
+	ibc |= 0xfULL << SYM_LSB(IBCCtrlA_0, PhyerrThreshold);
+	/* IB credit flow control. */
+	ibc |= 0xfULL << SYM_LSB(IBCCtrlA_0, OverrunThreshold);
+	/*
+	 * set initial max size pkt IBC will send, including ICRC; it's the
+	 * PIO buffer size in dwords, less 1; also see qib_set_mtu()
+	 */
+	ibc |= ((u64)(ppd->ibmaxlen >> 2) + 1) <<
+		SYM_LSB(IBCCtrlA_0, MaxPktLen);
+	ppd->cpspec->ibcctrl_a = ibc; /* without linkcmd or linkinitcmd! */
+
+	/*
+	 * Reset the PCS interface to the serdes (and also ibc, which is still
+	 * in reset from above).  Writes new value of ibcctrl_a as last step.
+	 */
+	qib_7322_mini_pcs_reset(ppd);
+
+	if (!ppd->cpspec->ibcctrl_b) {
+		unsigned lse = ppd->link_speed_enabled;
+
+		/*
+		 * Not on re-init after reset, establish shadow
+		 * and force initial config.
+		 */
+		ppd->cpspec->ibcctrl_b = qib_read_kreg_port(ppd,
+							     krp_ibcctrl_b);
+		ppd->cpspec->ibcctrl_b &= ~(IBA7322_IBC_SPEED_QDR |
+				IBA7322_IBC_SPEED_DDR |
+				IBA7322_IBC_SPEED_SDR |
+				IBA7322_IBC_WIDTH_AUTONEG |
+				SYM_MASK(IBCCtrlB_0, IB_LANE_REV_SUPPORTED));
+		if (lse & (lse - 1)) /* Muliple speeds enabled */
+			ppd->cpspec->ibcctrl_b |=
+				(lse << IBA7322_IBC_SPEED_LSB) |
+				IBA7322_IBC_IBTA_1_2_MASK |
+				IBA7322_IBC_MAX_SPEED_MASK;
+		else
+			ppd->cpspec->ibcctrl_b |= (lse == QIB_IB_QDR) ?
+				IBA7322_IBC_SPEED_QDR |
+				 IBA7322_IBC_IBTA_1_2_MASK :
+				(lse == QIB_IB_DDR) ?
+					IBA7322_IBC_SPEED_DDR :
+					IBA7322_IBC_SPEED_SDR;
+		if ((ppd->link_width_enabled & (IB_WIDTH_1X | IB_WIDTH_4X)) ==
+		    (IB_WIDTH_1X | IB_WIDTH_4X))
+			ppd->cpspec->ibcctrl_b |= IBA7322_IBC_WIDTH_AUTONEG;
+		else
+			ppd->cpspec->ibcctrl_b |=
+				ppd->link_width_enabled == IB_WIDTH_4X ?
+				IBA7322_IBC_WIDTH_4X_ONLY :
+				IBA7322_IBC_WIDTH_1X_ONLY;
+
+		/* always enable these on driver reload, not sticky */
+		ppd->cpspec->ibcctrl_b |= (IBA7322_IBC_RXPOL_MASK |
+			IBA7322_IBC_HRTBT_MASK);
+	}
+	qib_write_kreg_port(ppd, krp_ibcctrl_b, ppd->cpspec->ibcctrl_b);
+
+	/* setup so we have more time at CFGTEST to change H1 */
+	val = qib_read_kreg_port(ppd, krp_ibcctrl_c);
+	val &= ~SYM_MASK(IBCCtrlC_0, IB_FRONT_PORCH);
+	val |= 0xfULL << SYM_LSB(IBCCtrlC_0, IB_FRONT_PORCH);
+	qib_write_kreg_port(ppd, krp_ibcctrl_c, val);
+
+	serdes_7322_init(ppd);
+
+	guid = be64_to_cpu(ppd->guid);
+	if (!guid) {
+		if (dd->base_guid)
+			guid = be64_to_cpu(dd->base_guid) + ppd->port - 1;
+		ppd->guid = cpu_to_be64(guid);
+	}
+
+	qib_write_kreg_port(ppd, krp_hrtbt_guid, guid);
+	/* write to chip to prevent back-to-back writes of ibc reg */
+	qib_write_kreg(dd, kr_scratch, 0);
+
+	/* Enable port */
+	ppd->cpspec->ibcctrl_a |= SYM_MASK(IBCCtrlA_0, IBLinkEn);
+	set_vls(ppd);
+
+	/* initially come up DISABLED, without sending anything. */
+	val = ppd->cpspec->ibcctrl_a | (QLOGIC_IB_IBCC_LINKINITCMD_DISABLE <<
+					QLOGIC_IB_IBCC_LINKINITCMD_SHIFT);
+	qib_write_kreg_port(ppd, krp_ibcctrl_a, val);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+	/* clear the linkinit cmds */
+	ppd->cpspec->ibcctrl_a = val & ~SYM_MASK(IBCCtrlA_0, LinkInitCmd);
+
+	/* be paranoid against later code motion, etc. */
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+	ppd->p_rcvctrl |= SYM_MASK(RcvCtrl_0, RcvIBPortEnable);
+	qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl);
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+
+	/* Also enable IBSTATUSCHG interrupt.  */
+	val = qib_read_kreg_port(ppd, krp_errmask);
+	qib_write_kreg_port(ppd, krp_errmask,
+		val | ERR_MASK_N(IBStatusChanged));
+
+	/* Always zero until we start messing with SerDes for real */
+	return ret;
+}
+
+/**
+ * qib_7322_quiet_serdes - set serdes to txidle
+ * @dd: the qlogic_ib device
+ * Called when driver is being unloaded
+ */
+static void qib_7322_mini_quiet_serdes(struct qib_pportdata *ppd)
+{
+	u64 val;
+	unsigned long flags;
+
+	qib_set_ib_7322_lstate(ppd, 0, QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags &= ~QIBL_IB_AUTONEG_INPROG;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	wake_up(&ppd->cpspec->autoneg_wait);
+	cancel_delayed_work_sync(&ppd->cpspec->autoneg_work);
+	if (ppd->dd->cspec->r1)
+		cancel_delayed_work_sync(&ppd->cpspec->ipg_work);
+
+	ppd->cpspec->chase_end = 0;
+	if (ppd->cpspec->chase_timer.data) /* if initted */
+		del_timer_sync(&ppd->cpspec->chase_timer);
+
+	/*
+	 * Despite the name, actually disables IBC as well. Do it when
+	 * we are as sure as possible that no more packets can be
+	 * received, following the down and the PCS reset.
+	 * The actual disabling happens in qib_7322_mini_pci_reset(),
+	 * along with the PCS being reset.
+	 */
+	ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0, IBLinkEn);
+	qib_7322_mini_pcs_reset(ppd);
+
+	/*
+	 * Update the adjusted counters so the adjustment persists
+	 * across driver reload.
+	 */
+	if (ppd->cpspec->ibsymdelta || ppd->cpspec->iblnkerrdelta ||
+	    ppd->cpspec->ibdeltainprog || ppd->cpspec->iblnkdowndelta) {
+		struct qib_devdata *dd = ppd->dd;
+		u64 diagc;
+
+		/* enable counter writes */
+		diagc = qib_read_kreg64(dd, kr_hwdiagctrl);
+		qib_write_kreg(dd, kr_hwdiagctrl,
+			       diagc | SYM_MASK(HwDiagCtrl, CounterWrEnable));
+
+		if (ppd->cpspec->ibsymdelta || ppd->cpspec->ibdeltainprog) {
+			val = read_7322_creg32_port(ppd, crp_ibsymbolerr);
+			if (ppd->cpspec->ibdeltainprog)
+				val -= val - ppd->cpspec->ibsymsnap;
+			val -= ppd->cpspec->ibsymdelta;
+			write_7322_creg_port(ppd, crp_ibsymbolerr, val);
+		}
+		if (ppd->cpspec->iblnkerrdelta || ppd->cpspec->ibdeltainprog) {
+			val = read_7322_creg32_port(ppd, crp_iblinkerrrecov);
+			if (ppd->cpspec->ibdeltainprog)
+				val -= val - ppd->cpspec->iblnkerrsnap;
+			val -= ppd->cpspec->iblnkerrdelta;
+			write_7322_creg_port(ppd, crp_iblinkerrrecov, val);
+		}
+		if (ppd->cpspec->iblnkdowndelta) {
+			val = read_7322_creg32_port(ppd, crp_iblinkdown);
+			val += ppd->cpspec->iblnkdowndelta;
+			write_7322_creg_port(ppd, crp_iblinkdown, val);
+		}
+		/*
+		 * No need to save ibmalfdelta since IB perfcounters
+		 * are cleared on driver reload.
+		 */
+
+		/* and disable counter writes */
+		qib_write_kreg(dd, kr_hwdiagctrl, diagc);
+	}
+}
+
+/**
+ * qib_setup_7322_setextled - set the state of the two external LEDs
+ * @ppd: physical port on the qlogic_ib device
+ * @on: whether the link is up or not
+ *
+ * The exact combo of LEDs if on is true is determined by looking
+ * at the ibcstatus.
+ *
+ * These LEDs indicate the physical and logical state of IB link.
+ * For this chip (at least with recommended board pinouts), LED1
+ * is Yellow (logical state) and LED2 is Green (physical state),
+ *
+ * Note:  We try to match the Mellanox HCA LED behavior as best
+ * we can.  Green indicates physical link state is OK (something is
+ * plugged in, and we can train).
+ * Amber indicates the link is logically up (ACTIVE).
+ * Mellanox further blinks the amber LED to indicate data packet
+ * activity, but we have no hardware support for that, so it would
+ * require waking up every 10-20 msecs and checking the counters
+ * on the chip, and then turning the LED off if appropriate.  That's
+ * visible overhead, so not something we will do.
+ */
+static void qib_setup_7322_setextled(struct qib_pportdata *ppd, u32 on)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 extctl, ledblink = 0, val;
+	unsigned long flags;
+	int yel, grn;
+
+	/*
+	 * The diags use the LED to indicate diag info, so we leave
+	 * the external LED alone when the diags are running.
+	 */
+	if (dd->diag_client)
+		return;
+
+	/* Allow override of LED display for, e.g. Locating system in rack */
+	if (ppd->led_override) {
+		grn = (ppd->led_override & QIB_LED_PHYS);
+		yel = (ppd->led_override & QIB_LED_LOG);
+	} else if (on) {
+		val = qib_read_kreg_port(ppd, krp_ibcstatus_a);
+		grn = qib_7322_phys_portstate(val) ==
+			IB_PHYSPORTSTATE_LINKUP;
+		yel = qib_7322_iblink_state(val) == IB_PORT_ACTIVE;
+	} else {
+		grn = 0;
+		yel = 0;
+	}
+
+	spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+	extctl = dd->cspec->extctrl & (ppd->port == 1 ?
+		~ExtLED_IB1_MASK : ~ExtLED_IB2_MASK);
+	if (grn) {
+		extctl |= ppd->port == 1 ? ExtLED_IB1_GRN : ExtLED_IB2_GRN;
+		/*
+		 * Counts are in chip clock (4ns) periods.
+		 * This is 1/16 sec (66.6ms) on,
+		 * 3/16 sec (187.5 ms) off, with packets rcvd.
+		 */
+		ledblink = ((66600 * 1000UL / 4) << IBA7322_LEDBLINK_ON_SHIFT) |
+			((187500 * 1000UL / 4) << IBA7322_LEDBLINK_OFF_SHIFT);
+	}
+	if (yel)
+		extctl |= ppd->port == 1 ? ExtLED_IB1_YEL : ExtLED_IB2_YEL;
+	dd->cspec->extctrl = extctl;
+	qib_write_kreg(dd, kr_extctrl, dd->cspec->extctrl);
+	spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+
+	if (ledblink) /* blink the LED on packet receive */
+		qib_write_kreg_port(ppd, krp_rcvpktledcnt, ledblink);
+}
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+
+static int qib_7322_notify_dca(struct qib_devdata *dd, unsigned long event)
+{
+	switch (event) {
+	case DCA_PROVIDER_ADD:
+		if (dd->flags & QIB_DCA_ENABLED)
+			break;
+		if (!dca_add_requester(&dd->pcidev->dev)) {
+			qib_devinfo(dd->pcidev, "DCA enabled\n");
+			dd->flags |= QIB_DCA_ENABLED;
+			qib_setup_dca(dd);
+		}
+		break;
+	case DCA_PROVIDER_REMOVE:
+		if (dd->flags & QIB_DCA_ENABLED) {
+			dca_remove_requester(&dd->pcidev->dev);
+			dd->flags &= ~QIB_DCA_ENABLED;
+			dd->cspec->dca_ctrl = 0;
+			qib_write_kreg(dd, KREG_IDX(DCACtrlA),
+				dd->cspec->dca_ctrl);
+		}
+		break;
+	}
+	return 0;
+}
+
+static void qib_update_rhdrq_dca(struct qib_ctxtdata *rcd, int cpu)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_chip_specific *cspec = dd->cspec;
+
+	if (!(dd->flags & QIB_DCA_ENABLED))
+		return;
+	if (cspec->rhdr_cpu[rcd->ctxt] != cpu) {
+		const struct dca_reg_map *rmp;
+
+		cspec->rhdr_cpu[rcd->ctxt] = cpu;
+		rmp = &dca_rcvhdr_reg_map[rcd->ctxt];
+		cspec->dca_rcvhdr_ctrl[rmp->shadow_inx] &= rmp->mask;
+		cspec->dca_rcvhdr_ctrl[rmp->shadow_inx] |=
+			(u64) dca3_get_tag(&dd->pcidev->dev, cpu) << rmp->lsb;
+		qib_devinfo(dd->pcidev,
+			"Ctxt %d cpu %d dca %llx\n", rcd->ctxt, cpu,
+			(long long) cspec->dca_rcvhdr_ctrl[rmp->shadow_inx]);
+		qib_write_kreg(dd, rmp->regno,
+			       cspec->dca_rcvhdr_ctrl[rmp->shadow_inx]);
+		cspec->dca_ctrl |= SYM_MASK(DCACtrlA, RcvHdrqDCAEnable);
+		qib_write_kreg(dd, KREG_IDX(DCACtrlA), cspec->dca_ctrl);
+	}
+}
+
+static void qib_update_sdma_dca(struct qib_pportdata *ppd, int cpu)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_chip_specific *cspec = dd->cspec;
+	unsigned pidx = ppd->port - 1;
+
+	if (!(dd->flags & QIB_DCA_ENABLED))
+		return;
+	if (cspec->sdma_cpu[pidx] != cpu) {
+		cspec->sdma_cpu[pidx] = cpu;
+		cspec->dca_rcvhdr_ctrl[4] &= ~(ppd->hw_pidx ?
+			SYM_MASK(DCACtrlF, SendDma1DCAOPH) :
+			SYM_MASK(DCACtrlF, SendDma0DCAOPH));
+		cspec->dca_rcvhdr_ctrl[4] |=
+			(u64) dca3_get_tag(&dd->pcidev->dev, cpu) <<
+				(ppd->hw_pidx ?
+					SYM_LSB(DCACtrlF, SendDma1DCAOPH) :
+					SYM_LSB(DCACtrlF, SendDma0DCAOPH));
+		qib_devinfo(dd->pcidev,
+			"sdma %d cpu %d dca %llx\n", ppd->hw_pidx, cpu,
+			(long long) cspec->dca_rcvhdr_ctrl[4]);
+		qib_write_kreg(dd, KREG_IDX(DCACtrlF),
+			       cspec->dca_rcvhdr_ctrl[4]);
+		cspec->dca_ctrl |= ppd->hw_pidx ?
+			SYM_MASK(DCACtrlA, SendDMAHead1DCAEnable) :
+			SYM_MASK(DCACtrlA, SendDMAHead0DCAEnable);
+		qib_write_kreg(dd, KREG_IDX(DCACtrlA), cspec->dca_ctrl);
+	}
+}
+
+static void qib_setup_dca(struct qib_devdata *dd)
+{
+	struct qib_chip_specific *cspec = dd->cspec;
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(cspec->rhdr_cpu); i++)
+		cspec->rhdr_cpu[i] = -1;
+	for (i = 0; i < ARRAY_SIZE(cspec->sdma_cpu); i++)
+		cspec->sdma_cpu[i] = -1;
+	cspec->dca_rcvhdr_ctrl[0] =
+		(1ULL << SYM_LSB(DCACtrlB, RcvHdrq0DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlB, RcvHdrq1DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlB, RcvHdrq2DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlB, RcvHdrq3DCAXfrCnt));
+	cspec->dca_rcvhdr_ctrl[1] =
+		(1ULL << SYM_LSB(DCACtrlC, RcvHdrq4DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlC, RcvHdrq5DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlC, RcvHdrq6DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlC, RcvHdrq7DCAXfrCnt));
+	cspec->dca_rcvhdr_ctrl[2] =
+		(1ULL << SYM_LSB(DCACtrlD, RcvHdrq8DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlD, RcvHdrq9DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlD, RcvHdrq10DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlD, RcvHdrq11DCAXfrCnt));
+	cspec->dca_rcvhdr_ctrl[3] =
+		(1ULL << SYM_LSB(DCACtrlE, RcvHdrq12DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlE, RcvHdrq13DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlE, RcvHdrq14DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlE, RcvHdrq15DCAXfrCnt));
+	cspec->dca_rcvhdr_ctrl[4] =
+		(1ULL << SYM_LSB(DCACtrlF, RcvHdrq16DCAXfrCnt)) |
+		(1ULL << SYM_LSB(DCACtrlF, RcvHdrq17DCAXfrCnt));
+	for (i = 0; i < ARRAY_SIZE(cspec->sdma_cpu); i++)
+		qib_write_kreg(dd, KREG_IDX(DCACtrlB) + i,
+			       cspec->dca_rcvhdr_ctrl[i]);
+	for (i = 0; i < cspec->num_msix_entries; i++)
+		setup_dca_notifier(dd, &cspec->msix_entries[i]);
+}
+
+static void qib_irq_notifier_notify(struct irq_affinity_notify *notify,
+			     const cpumask_t *mask)
+{
+	struct qib_irq_notify *n =
+		container_of(notify, struct qib_irq_notify, notify);
+	int cpu = cpumask_first(mask);
+
+	if (n->rcv) {
+		struct qib_ctxtdata *rcd = (struct qib_ctxtdata *)n->arg;
+		qib_update_rhdrq_dca(rcd, cpu);
+	} else {
+		struct qib_pportdata *ppd = (struct qib_pportdata *)n->arg;
+		qib_update_sdma_dca(ppd, cpu);
+	}
+}
+
+static void qib_irq_notifier_release(struct kref *ref)
+{
+	struct qib_irq_notify *n =
+		container_of(ref, struct qib_irq_notify, notify.kref);
+	struct qib_devdata *dd;
+
+	if (n->rcv) {
+		struct qib_ctxtdata *rcd = (struct qib_ctxtdata *)n->arg;
+		dd = rcd->dd;
+	} else {
+		struct qib_pportdata *ppd = (struct qib_pportdata *)n->arg;
+		dd = ppd->dd;
+	}
+	qib_devinfo(dd->pcidev,
+		"release on HCA notify 0x%p n 0x%p\n", ref, n);
+	kfree(n);
+}
+#endif
+
+/*
+ * Disable MSIx interrupt if enabled, call generic MSIx code
+ * to cleanup, and clear pending MSIx interrupts.
+ * Used for fallback to INTx, after reset, and when MSIx setup fails.
+ */
+static void qib_7322_nomsix(struct qib_devdata *dd)
+{
+	u64 intgranted;
+	int n;
+
+	dd->cspec->main_int_mask = ~0ULL;
+	n = dd->cspec->num_msix_entries;
+	if (n) {
+		int i;
+
+		dd->cspec->num_msix_entries = 0;
+		for (i = 0; i < n; i++) {
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+			reset_dca_notifier(dd, &dd->cspec->msix_entries[i]);
+#endif
+			irq_set_affinity_hint(
+			  dd->cspec->msix_entries[i].msix.vector, NULL);
+			free_cpumask_var(dd->cspec->msix_entries[i].mask);
+			free_irq(dd->cspec->msix_entries[i].msix.vector,
+			   dd->cspec->msix_entries[i].arg);
+		}
+		qib_nomsix(dd);
+	}
+	/* make sure no MSIx interrupts are left pending */
+	intgranted = qib_read_kreg64(dd, kr_intgranted);
+	if (intgranted)
+		qib_write_kreg(dd, kr_intgranted, intgranted);
+}
+
+static void qib_7322_free_irq(struct qib_devdata *dd)
+{
+	if (dd->cspec->irq) {
+		free_irq(dd->cspec->irq, dd);
+		dd->cspec->irq = 0;
+	}
+	qib_7322_nomsix(dd);
+}
+
+static void qib_setup_7322_cleanup(struct qib_devdata *dd)
+{
+	int i;
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	if (dd->flags & QIB_DCA_ENABLED) {
+		dca_remove_requester(&dd->pcidev->dev);
+		dd->flags &= ~QIB_DCA_ENABLED;
+		dd->cspec->dca_ctrl = 0;
+		qib_write_kreg(dd, KREG_IDX(DCACtrlA), dd->cspec->dca_ctrl);
+	}
+#endif
+
+	qib_7322_free_irq(dd);
+	kfree(dd->cspec->cntrs);
+	kfree(dd->cspec->sendchkenable);
+	kfree(dd->cspec->sendgrhchk);
+	kfree(dd->cspec->sendibchk);
+	kfree(dd->cspec->msix_entries);
+	for (i = 0; i < dd->num_pports; i++) {
+		unsigned long flags;
+		u32 mask = QSFP_GPIO_MOD_PRS_N |
+			(QSFP_GPIO_MOD_PRS_N << QSFP_GPIO_PORT2_SHIFT);
+
+		kfree(dd->pport[i].cpspec->portcntrs);
+		if (dd->flags & QIB_HAS_QSFP) {
+			spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+			dd->cspec->gpio_mask &= ~mask;
+			qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+			spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+			qib_qsfp_deinit(&dd->pport[i].cpspec->qsfp_data);
+		}
+		if (dd->pport[i].ibport_data.smi_ah)
+			ib_destroy_ah(&dd->pport[i].ibport_data.smi_ah->ibah);
+	}
+}
+
+/* handle SDMA interrupts */
+static void sdma_7322_intr(struct qib_devdata *dd, u64 istat)
+{
+	struct qib_pportdata *ppd0 = &dd->pport[0];
+	struct qib_pportdata *ppd1 = &dd->pport[1];
+	u64 intr0 = istat & (INT_MASK_P(SDma, 0) |
+		INT_MASK_P(SDmaIdle, 0) | INT_MASK_P(SDmaProgress, 0));
+	u64 intr1 = istat & (INT_MASK_P(SDma, 1) |
+		INT_MASK_P(SDmaIdle, 1) | INT_MASK_P(SDmaProgress, 1));
+
+	if (intr0)
+		qib_sdma_intr(ppd0);
+	if (intr1)
+		qib_sdma_intr(ppd1);
+
+	if (istat & INT_MASK_PM(SDmaCleanupDone, 0))
+		qib_sdma_process_event(ppd0, qib_sdma_event_e20_hw_started);
+	if (istat & INT_MASK_PM(SDmaCleanupDone, 1))
+		qib_sdma_process_event(ppd1, qib_sdma_event_e20_hw_started);
+}
+
+/*
+ * Set or clear the Send buffer available interrupt enable bit.
+ */
+static void qib_wantpiobuf_7322_intr(struct qib_devdata *dd, u32 needint)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+	if (needint)
+		dd->sendctrl |= SYM_MASK(SendCtrl, SendIntBufAvail);
+	else
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, SendIntBufAvail);
+	qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+}
+
+/*
+ * Somehow got an interrupt with reserved bits set in interrupt status.
+ * Print a message so we know it happened, then clear them.
+ * keep mainline interrupt handler cache-friendly
+ */
+static noinline void unknown_7322_ibits(struct qib_devdata *dd, u64 istat)
+{
+	u64 kills;
+	char msg[128];
+
+	kills = istat & ~QIB_I_BITSEXTANT;
+	qib_dev_err(dd,
+		"Clearing reserved interrupt(s) 0x%016llx: %s\n",
+		(unsigned long long) kills, msg);
+	qib_write_kreg(dd, kr_intmask, (dd->cspec->int_enable_mask & ~kills));
+}
+
+/* keep mainline interrupt handler cache-friendly */
+static noinline void unknown_7322_gpio_intr(struct qib_devdata *dd)
+{
+	u32 gpiostatus;
+	int handled = 0;
+	int pidx;
+
+	/*
+	 * Boards for this chip currently don't use GPIO interrupts,
+	 * so clear by writing GPIOstatus to GPIOclear, and complain
+	 * to developer.  To avoid endless repeats, clear
+	 * the bits in the mask, since there is some kind of
+	 * programming error or chip problem.
+	 */
+	gpiostatus = qib_read_kreg32(dd, kr_gpio_status);
+	/*
+	 * In theory, writing GPIOstatus to GPIOclear could
+	 * have a bad side-effect on some diagnostic that wanted
+	 * to poll for a status-change, but the various shadows
+	 * make that problematic at best. Diags will just suppress
+	 * all GPIO interrupts during such tests.
+	 */
+	qib_write_kreg(dd, kr_gpio_clear, gpiostatus);
+	/*
+	 * Check for QSFP MOD_PRS changes
+	 * only works for single port if IB1 != pidx1
+	 */
+	for (pidx = 0; pidx < dd->num_pports && (dd->flags & QIB_HAS_QSFP);
+	     ++pidx) {
+		struct qib_pportdata *ppd;
+		struct qib_qsfp_data *qd;
+		u32 mask;
+		if (!dd->pport[pidx].link_speed_supported)
+			continue;
+		mask = QSFP_GPIO_MOD_PRS_N;
+		ppd = dd->pport + pidx;
+		mask <<= (QSFP_GPIO_PORT2_SHIFT * ppd->hw_pidx);
+		if (gpiostatus & dd->cspec->gpio_mask & mask) {
+			u64 pins;
+			qd = &ppd->cpspec->qsfp_data;
+			gpiostatus &= ~mask;
+			pins = qib_read_kreg64(dd, kr_extstatus);
+			pins >>= SYM_LSB(EXTStatus, GPIOIn);
+			if (!(pins & mask)) {
+				++handled;
+				qd->t_insert = jiffies;
+				queue_work(ib_wq, &qd->work);
+			}
+		}
+	}
+
+	if (gpiostatus && !handled) {
+		const u32 mask = qib_read_kreg32(dd, kr_gpio_mask);
+		u32 gpio_irq = mask & gpiostatus;
+
+		/*
+		 * Clear any troublemakers, and update chip from shadow
+		 */
+		dd->cspec->gpio_mask &= ~gpio_irq;
+		qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+	}
+}
+
+/*
+ * Handle errors and unusual events first, separate function
+ * to improve cache hits for fast path interrupt handling.
+ */
+static noinline void unlikely_7322_intr(struct qib_devdata *dd, u64 istat)
+{
+	if (istat & ~QIB_I_BITSEXTANT)
+		unknown_7322_ibits(dd, istat);
+	if (istat & QIB_I_GPIO)
+		unknown_7322_gpio_intr(dd);
+	if (istat & QIB_I_C_ERROR) {
+		qib_write_kreg(dd, kr_errmask, 0ULL);
+		tasklet_schedule(&dd->error_tasklet);
+	}
+	if (istat & INT_MASK_P(Err, 0) && dd->rcd[0])
+		handle_7322_p_errors(dd->rcd[0]->ppd);
+	if (istat & INT_MASK_P(Err, 1) && dd->rcd[1])
+		handle_7322_p_errors(dd->rcd[1]->ppd);
+}
+
+/*
+ * Dynamically adjust the rcv int timeout for a context based on incoming
+ * packet rate.
+ */
+static void adjust_rcv_timeout(struct qib_ctxtdata *rcd, int npkts)
+{
+	struct qib_devdata *dd = rcd->dd;
+	u32 timeout = dd->cspec->rcvavail_timeout[rcd->ctxt];
+
+	/*
+	 * Dynamically adjust idle timeout on chip
+	 * based on number of packets processed.
+	 */
+	if (npkts < rcv_int_count && timeout > 2)
+		timeout >>= 1;
+	else if (npkts >= rcv_int_count && timeout < rcv_int_timeout)
+		timeout = min(timeout << 1, rcv_int_timeout);
+	else
+		return;
+
+	dd->cspec->rcvavail_timeout[rcd->ctxt] = timeout;
+	qib_write_kreg(dd, kr_rcvavailtimeout + rcd->ctxt, timeout);
+}
+
+/*
+ * This is the main interrupt handler.
+ * It will normally only be used for low frequency interrupts but may
+ * have to handle all interrupts if INTx is enabled or fewer than normal
+ * MSIx interrupts were allocated.
+ * This routine should ignore the interrupt bits for any of the
+ * dedicated MSIx handlers.
+ */
+static irqreturn_t qib_7322intr(int irq, void *data)
+{
+	struct qib_devdata *dd = data;
+	irqreturn_t ret;
+	u64 istat;
+	u64 ctxtrbits;
+	u64 rmask;
+	unsigned i;
+	u32 npkts;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT) {
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		ret = IRQ_HANDLED;
+		goto bail;
+	}
+
+	istat = qib_read_kreg64(dd, kr_intstatus);
+
+	if (unlikely(istat == ~0ULL)) {
+		qib_bad_intrstatus(dd);
+		qib_dev_err(dd, "Interrupt status all f's, skipping\n");
+		/* don't know if it was our interrupt or not */
+		ret = IRQ_NONE;
+		goto bail;
+	}
+
+	istat &= dd->cspec->main_int_mask;
+	if (unlikely(!istat)) {
+		/* already handled, or shared and not us */
+		ret = IRQ_NONE;
+		goto bail;
+	}
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* handle "errors" of various kinds first, device ahead of port */
+	if (unlikely(istat & (~QIB_I_BITSEXTANT | QIB_I_GPIO |
+			      QIB_I_C_ERROR | INT_MASK_P(Err, 0) |
+			      INT_MASK_P(Err, 1))))
+		unlikely_7322_intr(dd, istat);
+
+	/*
+	 * Clear the interrupt bits we found set, relatively early, so we
+	 * "know" know the chip will have seen this by the time we process
+	 * the queue, and will re-interrupt if necessary.  The processor
+	 * itself won't take the interrupt again until we return.
+	 */
+	qib_write_kreg(dd, kr_intclear, istat);
+
+	/*
+	 * Handle kernel receive queues before checking for pio buffers
+	 * available since receives can overflow; piobuf waiters can afford
+	 * a few extra cycles, since they were waiting anyway.
+	 */
+	ctxtrbits = istat & (QIB_I_RCVAVAIL_MASK | QIB_I_RCVURG_MASK);
+	if (ctxtrbits) {
+		rmask = (1ULL << QIB_I_RCVAVAIL_LSB) |
+			(1ULL << QIB_I_RCVURG_LSB);
+		for (i = 0; i < dd->first_user_ctxt; i++) {
+			if (ctxtrbits & rmask) {
+				ctxtrbits &= ~rmask;
+				if (dd->rcd[i])
+					qib_kreceive(dd->rcd[i], NULL, &npkts);
+			}
+			rmask <<= 1;
+		}
+		if (ctxtrbits) {
+			ctxtrbits = (ctxtrbits >> QIB_I_RCVAVAIL_LSB) |
+				(ctxtrbits >> QIB_I_RCVURG_LSB);
+			qib_handle_urcv(dd, ctxtrbits);
+		}
+	}
+
+	if (istat & (QIB_I_P_SDMAINT(0) | QIB_I_P_SDMAINT(1)))
+		sdma_7322_intr(dd, istat);
+
+	if ((istat & QIB_I_SPIOBUFAVAIL) && (dd->flags & QIB_INITTED))
+		qib_ib_piobufavail(dd);
+
+	ret = IRQ_HANDLED;
+bail:
+	return ret;
+}
+
+/*
+ * Dedicated receive packet available interrupt handler.
+ */
+static irqreturn_t qib_7322pintr(int irq, void *data)
+{
+	struct qib_ctxtdata *rcd = data;
+	struct qib_devdata *dd = rcd->dd;
+	u32 npkts;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, ((1ULL << QIB_I_RCVAVAIL_LSB) |
+		       (1ULL << QIB_I_RCVURG_LSB)) << rcd->ctxt);
+
+	qib_kreceive(rcd, NULL, &npkts);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Dedicated Send buffer available interrupt handler.
+ */
+static irqreturn_t qib_7322bufavail(int irq, void *data)
+{
+	struct qib_devdata *dd = data;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, QIB_I_SPIOBUFAVAIL);
+
+	/* qib_ib_piobufavail() will clear the want PIO interrupt if needed */
+	if (dd->flags & QIB_INITTED)
+		qib_ib_piobufavail(dd);
+	else
+		qib_wantpiobuf_7322_intr(dd, 0);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Dedicated Send DMA interrupt handler.
+ */
+static irqreturn_t sdma_intr(int irq, void *data)
+{
+	struct qib_pportdata *ppd = data;
+	struct qib_devdata *dd = ppd->dd;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
+		       INT_MASK_P(SDma, 1) : INT_MASK_P(SDma, 0));
+	qib_sdma_intr(ppd);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Dedicated Send DMA idle interrupt handler.
+ */
+static irqreturn_t sdma_idle_intr(int irq, void *data)
+{
+	struct qib_pportdata *ppd = data;
+	struct qib_devdata *dd = ppd->dd;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
+		       INT_MASK_P(SDmaIdle, 1) : INT_MASK_P(SDmaIdle, 0));
+	qib_sdma_intr(ppd);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Dedicated Send DMA progress interrupt handler.
+ */
+static irqreturn_t sdma_progress_intr(int irq, void *data)
+{
+	struct qib_pportdata *ppd = data;
+	struct qib_devdata *dd = ppd->dd;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
+		       INT_MASK_P(SDmaProgress, 1) :
+		       INT_MASK_P(SDmaProgress, 0));
+	qib_sdma_intr(ppd);
+
+	return IRQ_HANDLED;
+}
+
+/*
+ * Dedicated Send DMA cleanup interrupt handler.
+ */
+static irqreturn_t sdma_cleanup_intr(int irq, void *data)
+{
+	struct qib_pportdata *ppd = data;
+	struct qib_devdata *dd = ppd->dd;
+
+	if ((dd->flags & (QIB_PRESENT | QIB_BADINTR)) != QIB_PRESENT)
+		/*
+		 * This return value is not great, but we do not want the
+		 * interrupt core code to remove our interrupt handler
+		 * because we don't appear to be handling an interrupt
+		 * during a chip reset.
+		 */
+		return IRQ_HANDLED;
+
+	this_cpu_inc(*dd->int_counter);
+
+	/* Clear the interrupt bit we expect to be set. */
+	qib_write_kreg(dd, kr_intclear, ppd->hw_pidx ?
+		       INT_MASK_PM(SDmaCleanupDone, 1) :
+		       INT_MASK_PM(SDmaCleanupDone, 0));
+	qib_sdma_process_event(ppd, qib_sdma_event_e20_hw_started);
+
+	return IRQ_HANDLED;
+}
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+
+static void reset_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m)
+{
+	if (!m->dca)
+		return;
+	qib_devinfo(dd->pcidev,
+		"Disabling notifier on HCA %d irq %d\n",
+		dd->unit,
+		m->msix.vector);
+	irq_set_affinity_notifier(
+		m->msix.vector,
+		NULL);
+	m->notifier = NULL;
+}
+
+static void setup_dca_notifier(struct qib_devdata *dd, struct qib_msix_entry *m)
+{
+	struct qib_irq_notify *n;
+
+	if (!m->dca)
+		return;
+	n = kzalloc(sizeof(*n), GFP_KERNEL);
+	if (n) {
+		int ret;
+
+		m->notifier = n;
+		n->notify.irq = m->msix.vector;
+		n->notify.notify = qib_irq_notifier_notify;
+		n->notify.release = qib_irq_notifier_release;
+		n->arg = m->arg;
+		n->rcv = m->rcv;
+		qib_devinfo(dd->pcidev,
+			"set notifier irq %d rcv %d notify %p\n",
+			n->notify.irq, n->rcv, &n->notify);
+		ret = irq_set_affinity_notifier(
+				n->notify.irq,
+				&n->notify);
+		if (ret) {
+			m->notifier = NULL;
+			kfree(n);
+		}
+	}
+}
+
+#endif
+
+/*
+ * Set up our chip-specific interrupt handler.
+ * The interrupt type has already been setup, so
+ * we just need to do the registration and error checking.
+ * If we are using MSIx interrupts, we may fall back to
+ * INTx later, if the interrupt handler doesn't get called
+ * within 1/2 second (see verify_interrupt()).
+ */
+static void qib_setup_7322_interrupt(struct qib_devdata *dd, int clearpend)
+{
+	int ret, i, msixnum;
+	u64 redirect[6];
+	u64 mask;
+	const struct cpumask *local_mask;
+	int firstcpu, secondcpu = 0, currrcvcpu = 0;
+
+	if (!dd->num_pports)
+		return;
+
+	if (clearpend) {
+		/*
+		 * if not switching interrupt types, be sure interrupts are
+		 * disabled, and then clear anything pending at this point,
+		 * because we are starting clean.
+		 */
+		qib_7322_set_intr_state(dd, 0);
+
+		/* clear the reset error, init error/hwerror mask */
+		qib_7322_init_hwerrors(dd);
+
+		/* clear any interrupt bits that might be set */
+		qib_write_kreg(dd, kr_intclear, ~0ULL);
+
+		/* make sure no pending MSIx intr, and clear diag reg */
+		qib_write_kreg(dd, kr_intgranted, ~0ULL);
+		qib_write_kreg(dd, kr_vecclr_wo_int, ~0ULL);
+	}
+
+	if (!dd->cspec->num_msix_entries) {
+		/* Try to get INTx interrupt */
+try_intx:
+		if (!dd->pcidev->irq) {
+			qib_dev_err(dd,
+				"irq is 0, BIOS error?  Interrupts won't work\n");
+			goto bail;
+		}
+		ret = request_irq(dd->pcidev->irq, qib_7322intr,
+				  IRQF_SHARED, QIB_DRV_NAME, dd);
+		if (ret) {
+			qib_dev_err(dd,
+				"Couldn't setup INTx interrupt (irq=%d): %d\n",
+				dd->pcidev->irq, ret);
+			goto bail;
+		}
+		dd->cspec->irq = dd->pcidev->irq;
+		dd->cspec->main_int_mask = ~0ULL;
+		goto bail;
+	}
+
+	/* Try to get MSIx interrupts */
+	memset(redirect, 0, sizeof redirect);
+	mask = ~0ULL;
+	msixnum = 0;
+	local_mask = cpumask_of_pcibus(dd->pcidev->bus);
+	firstcpu = cpumask_first(local_mask);
+	if (firstcpu >= nr_cpu_ids ||
+			cpumask_weight(local_mask) == num_online_cpus()) {
+		local_mask = topology_core_cpumask(0);
+		firstcpu = cpumask_first(local_mask);
+	}
+	if (firstcpu < nr_cpu_ids) {
+		secondcpu = cpumask_next(firstcpu, local_mask);
+		if (secondcpu >= nr_cpu_ids)
+			secondcpu = firstcpu;
+		currrcvcpu = secondcpu;
+	}
+	for (i = 0; msixnum < dd->cspec->num_msix_entries; i++) {
+		irq_handler_t handler;
+		void *arg;
+		u64 val;
+		int lsb, reg, sh;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+		int dca = 0;
+#endif
+
+		dd->cspec->msix_entries[msixnum].
+			name[sizeof(dd->cspec->msix_entries[msixnum].name) - 1]
+			= '\0';
+		if (i < ARRAY_SIZE(irq_table)) {
+			if (irq_table[i].port) {
+				/* skip if for a non-configured port */
+				if (irq_table[i].port > dd->num_pports)
+					continue;
+				arg = dd->pport + irq_table[i].port - 1;
+			} else
+				arg = dd;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+			dca = irq_table[i].dca;
+#endif
+			lsb = irq_table[i].lsb;
+			handler = irq_table[i].handler;
+			snprintf(dd->cspec->msix_entries[msixnum].name,
+				sizeof(dd->cspec->msix_entries[msixnum].name)
+				 - 1,
+				QIB_DRV_NAME "%d%s", dd->unit,
+				irq_table[i].name);
+		} else {
+			unsigned ctxt;
+
+			ctxt = i - ARRAY_SIZE(irq_table);
+			/* per krcvq context receive interrupt */
+			arg = dd->rcd[ctxt];
+			if (!arg)
+				continue;
+			if (qib_krcvq01_no_msi && ctxt < 2)
+				continue;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+			dca = 1;
+#endif
+			lsb = QIB_I_RCVAVAIL_LSB + ctxt;
+			handler = qib_7322pintr;
+			snprintf(dd->cspec->msix_entries[msixnum].name,
+				sizeof(dd->cspec->msix_entries[msixnum].name)
+				 - 1,
+				QIB_DRV_NAME "%d (kctx)", dd->unit);
+		}
+		ret = request_irq(
+			dd->cspec->msix_entries[msixnum].msix.vector,
+			handler, 0, dd->cspec->msix_entries[msixnum].name,
+			arg);
+		if (ret) {
+			/*
+			 * Shouldn't happen since the enable said we could
+			 * have as many as we are trying to setup here.
+			 */
+			qib_dev_err(dd,
+				"Couldn't setup MSIx interrupt (vec=%d, irq=%d): %d\n",
+				msixnum,
+				dd->cspec->msix_entries[msixnum].msix.vector,
+				ret);
+			qib_7322_nomsix(dd);
+			goto try_intx;
+		}
+		dd->cspec->msix_entries[msixnum].arg = arg;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+		dd->cspec->msix_entries[msixnum].dca = dca;
+		dd->cspec->msix_entries[msixnum].rcv =
+			handler == qib_7322pintr;
+#endif
+		if (lsb >= 0) {
+			reg = lsb / IBA7322_REDIRECT_VEC_PER_REG;
+			sh = (lsb % IBA7322_REDIRECT_VEC_PER_REG) *
+				SYM_LSB(IntRedirect0, vec1);
+			mask &= ~(1ULL << lsb);
+			redirect[reg] |= ((u64) msixnum) << sh;
+		}
+		val = qib_read_kreg64(dd, 2 * msixnum + 1 +
+			(QIB_7322_MsixTable_OFFS / sizeof(u64)));
+		if (firstcpu < nr_cpu_ids &&
+			zalloc_cpumask_var(
+				&dd->cspec->msix_entries[msixnum].mask,
+				GFP_KERNEL)) {
+			if (handler == qib_7322pintr) {
+				cpumask_set_cpu(currrcvcpu,
+					dd->cspec->msix_entries[msixnum].mask);
+				currrcvcpu = cpumask_next(currrcvcpu,
+					local_mask);
+				if (currrcvcpu >= nr_cpu_ids)
+					currrcvcpu = secondcpu;
+			} else {
+				cpumask_set_cpu(firstcpu,
+					dd->cspec->msix_entries[msixnum].mask);
+			}
+			irq_set_affinity_hint(
+				dd->cspec->msix_entries[msixnum].msix.vector,
+				dd->cspec->msix_entries[msixnum].mask);
+		}
+		msixnum++;
+	}
+	/* Initialize the vector mapping */
+	for (i = 0; i < ARRAY_SIZE(redirect); i++)
+		qib_write_kreg(dd, kr_intredirect + i, redirect[i]);
+	dd->cspec->main_int_mask = mask;
+	tasklet_init(&dd->error_tasklet, qib_error_tasklet,
+		(unsigned long)dd);
+bail:;
+}
+
+/**
+ * qib_7322_boardname - fill in the board name and note features
+ * @dd: the qlogic_ib device
+ *
+ * info will be based on the board revision register
+ */
+static unsigned qib_7322_boardname(struct qib_devdata *dd)
+{
+	/* Will need enumeration of board-types here */
+	char *n;
+	u32 boardid, namelen;
+	unsigned features = DUAL_PORT_CAP;
+
+	boardid = SYM_FIELD(dd->revision, Revision, BoardID);
+
+	switch (boardid) {
+	case 0:
+		n = "InfiniPath_QLE7342_Emulation";
+		break;
+	case 1:
+		n = "InfiniPath_QLE7340";
+		dd->flags |= QIB_HAS_QSFP;
+		features = PORT_SPD_CAP;
+		break;
+	case 2:
+		n = "InfiniPath_QLE7342";
+		dd->flags |= QIB_HAS_QSFP;
+		break;
+	case 3:
+		n = "InfiniPath_QMI7342";
+		break;
+	case 4:
+		n = "InfiniPath_Unsupported7342";
+		qib_dev_err(dd, "Unsupported version of QMH7342\n");
+		features = 0;
+		break;
+	case BOARD_QMH7342:
+		n = "InfiniPath_QMH7342";
+		features = 0x24;
+		break;
+	case BOARD_QME7342:
+		n = "InfiniPath_QME7342";
+		break;
+	case 8:
+		n = "InfiniPath_QME7362";
+		dd->flags |= QIB_HAS_QSFP;
+		break;
+	case 15:
+		n = "InfiniPath_QLE7342_TEST";
+		dd->flags |= QIB_HAS_QSFP;
+		break;
+	default:
+		n = "InfiniPath_QLE73xy_UNKNOWN";
+		qib_dev_err(dd, "Unknown 7322 board type %u\n", boardid);
+		break;
+	}
+	dd->board_atten = 1; /* index into txdds_Xdr */
+
+	namelen = strlen(n) + 1;
+	dd->boardname = kmalloc(namelen, GFP_KERNEL);
+	if (!dd->boardname)
+		qib_dev_err(dd, "Failed allocation for board name: %s\n", n);
+	else
+		snprintf(dd->boardname, namelen, "%s", n);
+
+	snprintf(dd->boardversion, sizeof(dd->boardversion),
+		 "ChipABI %u.%u, %s, InfiniPath%u %u.%u, SW Compat %u\n",
+		 QIB_CHIP_VERS_MAJ, QIB_CHIP_VERS_MIN, dd->boardname,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, Arch),
+		 dd->majrev, dd->minrev,
+		 (unsigned)SYM_FIELD(dd->revision, Revision_R, SW));
+
+	if (qib_singleport && (features >> PORT_SPD_CAP_SHIFT) & PORT_SPD_CAP) {
+		qib_devinfo(dd->pcidev,
+			"IB%u: Forced to single port mode by module parameter\n",
+			dd->unit);
+		features &= PORT_SPD_CAP;
+	}
+
+	return features;
+}
+
+/*
+ * This routine sleeps, so it can only be called from user context, not
+ * from interrupt context.
+ */
+static int qib_do_7322_reset(struct qib_devdata *dd)
+{
+	u64 val;
+	u64 *msix_vecsave;
+	int i, msix_entries, ret = 1;
+	u16 cmdval;
+	u8 int_line, clinesz;
+	unsigned long flags;
+
+	/* Use dev_err so it shows up in logs, etc. */
+	qib_dev_err(dd, "Resetting InfiniPath unit %u\n", dd->unit);
+
+	qib_pcie_getcmd(dd, &cmdval, &int_line, &clinesz);
+
+	msix_entries = dd->cspec->num_msix_entries;
+
+	/* no interrupts till re-initted */
+	qib_7322_set_intr_state(dd, 0);
+
+	if (msix_entries) {
+		qib_7322_nomsix(dd);
+		/* can be up to 512 bytes, too big for stack */
+		msix_vecsave = kmalloc(2 * dd->cspec->num_msix_entries *
+			sizeof(u64), GFP_KERNEL);
+		if (!msix_vecsave)
+			qib_dev_err(dd, "No mem to save MSIx data\n");
+	} else
+		msix_vecsave = NULL;
+
+	/*
+	 * Core PCI (as of 2.6.18) doesn't save or rewrite the full vector
+	 * info that is set up by the BIOS, so we have to save and restore
+	 * it ourselves.   There is some risk something could change it,
+	 * after we save it, but since we have disabled the MSIx, it
+	 * shouldn't be touched...
+	 */
+	for (i = 0; i < msix_entries; i++) {
+		u64 vecaddr, vecdata;
+		vecaddr = qib_read_kreg64(dd, 2 * i +
+				  (QIB_7322_MsixTable_OFFS / sizeof(u64)));
+		vecdata = qib_read_kreg64(dd, 1 + 2 * i +
+				  (QIB_7322_MsixTable_OFFS / sizeof(u64)));
+		if (msix_vecsave) {
+			msix_vecsave[2 * i] = vecaddr;
+			/* save it without the masked bit set */
+			msix_vecsave[1 + 2 * i] = vecdata & ~0x100000000ULL;
+		}
+	}
+
+	dd->pport->cpspec->ibdeltainprog = 0;
+	dd->pport->cpspec->ibsymdelta = 0;
+	dd->pport->cpspec->iblnkerrdelta = 0;
+	dd->pport->cpspec->ibmalfdelta = 0;
+	/* so we check interrupts work again */
+	dd->z_int_counter = qib_int_counter(dd);
+
+	/*
+	 * Keep chip from being accessed until we are ready.  Use
+	 * writeq() directly, to allow the write even though QIB_PRESENT
+	 * isn't set.
+	 */
+	dd->flags &= ~(QIB_INITTED | QIB_PRESENT | QIB_BADINTR);
+	dd->flags |= QIB_DOING_RESET;
+	val = dd->control | QLOGIC_IB_C_RESET;
+	writeq(val, &dd->kregbase[kr_control]);
+
+	for (i = 1; i <= 5; i++) {
+		/*
+		 * Allow MBIST, etc. to complete; longer on each retry.
+		 * We sometimes get machine checks from bus timeout if no
+		 * response, so for now, make it *really* long.
+		 */
+		msleep(1000 + (1 + i) * 3000);
+
+		qib_pcie_reenable(dd, cmdval, int_line, clinesz);
+
+		/*
+		 * Use readq directly, so we don't need to mark it as PRESENT
+		 * until we get a successful indication that all is well.
+		 */
+		val = readq(&dd->kregbase[kr_revision]);
+		if (val == dd->revision)
+			break;
+		if (i == 5) {
+			qib_dev_err(dd,
+				"Failed to initialize after reset, unusable\n");
+			ret = 0;
+			goto  bail;
+		}
+	}
+
+	dd->flags |= QIB_PRESENT; /* it's back */
+
+	if (msix_entries) {
+		/* restore the MSIx vector address and data if saved above */
+		for (i = 0; i < msix_entries; i++) {
+			dd->cspec->msix_entries[i].msix.entry = i;
+			if (!msix_vecsave || !msix_vecsave[2 * i])
+				continue;
+			qib_write_kreg(dd, 2 * i +
+				(QIB_7322_MsixTable_OFFS / sizeof(u64)),
+				msix_vecsave[2 * i]);
+			qib_write_kreg(dd, 1 + 2 * i +
+				(QIB_7322_MsixTable_OFFS / sizeof(u64)),
+				msix_vecsave[1 + 2 * i]);
+		}
+	}
+
+	/* initialize the remaining registers.  */
+	for (i = 0; i < dd->num_pports; ++i)
+		write_7322_init_portregs(&dd->pport[i]);
+	write_7322_initregs(dd);
+
+	if (qib_pcie_params(dd, dd->lbus_width,
+			    &dd->cspec->num_msix_entries,
+			    dd->cspec->msix_entries))
+		qib_dev_err(dd,
+			"Reset failed to setup PCIe or interrupts; continuing anyway\n");
+
+	qib_setup_7322_interrupt(dd, 1);
+
+	for (i = 0; i < dd->num_pports; ++i) {
+		struct qib_pportdata *ppd = &dd->pport[i];
+
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_FORCE_NOTIFY;
+		ppd->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+
+bail:
+	dd->flags &= ~QIB_DOING_RESET; /* OK or not, no longer resetting */
+	kfree(msix_vecsave);
+	return ret;
+}
+
+/**
+ * qib_7322_put_tid - write a TID to the chip
+ * @dd: the qlogic_ib device
+ * @tidptr: pointer to the expected TID (in chip) to update
+ * @tidtype: 0 for eager, 1 for expected
+ * @pa: physical address of in memory buffer; tidinvalid if freeing
+ */
+static void qib_7322_put_tid(struct qib_devdata *dd, u64 __iomem *tidptr,
+			     u32 type, unsigned long pa)
+{
+	if (!(dd->flags & QIB_PRESENT))
+		return;
+	if (pa != dd->tidinvalid) {
+		u64 chippa = pa >> IBA7322_TID_PA_SHIFT;
+
+		/* paranoia checks */
+		if (pa != (chippa << IBA7322_TID_PA_SHIFT)) {
+			qib_dev_err(dd, "Physaddr %lx not 2KB aligned!\n",
+				    pa);
+			return;
+		}
+		if (chippa >= (1UL << IBA7322_TID_SZ_SHIFT)) {
+			qib_dev_err(dd,
+				"Physical page address 0x%lx larger than supported\n",
+				pa);
+			return;
+		}
+
+		if (type == RCVHQ_RCV_TYPE_EAGER)
+			chippa |= dd->tidtemplate;
+		else /* for now, always full 4KB page */
+			chippa |= IBA7322_TID_SZ_4K;
+		pa = chippa;
+	}
+	writeq(pa, tidptr);
+	mmiowb();
+}
+
+/**
+ * qib_7322_clear_tids - clear all TID entries for a ctxt, expected and eager
+ * @dd: the qlogic_ib device
+ * @ctxt: the ctxt
+ *
+ * clear all TID entries for a ctxt, expected and eager.
+ * Used from qib_close().
+ */
+static void qib_7322_clear_tids(struct qib_devdata *dd,
+				struct qib_ctxtdata *rcd)
+{
+	u64 __iomem *tidbase;
+	unsigned long tidinv;
+	u32 ctxt;
+	int i;
+
+	if (!dd->kregbase || !rcd)
+		return;
+
+	ctxt = rcd->ctxt;
+
+	tidinv = dd->tidinvalid;
+	tidbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase +
+		 dd->rcvtidbase +
+		 ctxt * dd->rcvtidcnt * sizeof(*tidbase));
+
+	for (i = 0; i < dd->rcvtidcnt; i++)
+		qib_7322_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EXPECTED,
+				 tidinv);
+
+	tidbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase +
+		 dd->rcvegrbase +
+		 rcd->rcvegr_tid_base * sizeof(*tidbase));
+
+	for (i = 0; i < rcd->rcvegrcnt; i++)
+		qib_7322_put_tid(dd, &tidbase[i], RCVHQ_RCV_TYPE_EAGER,
+				 tidinv);
+}
+
+/**
+ * qib_7322_tidtemplate - setup constants for TID updates
+ * @dd: the qlogic_ib device
+ *
+ * We setup stuff that we use a lot, to avoid calculating each time
+ */
+static void qib_7322_tidtemplate(struct qib_devdata *dd)
+{
+	/*
+	 * For now, we always allocate 4KB buffers (at init) so we can
+	 * receive max size packets.  We may want a module parameter to
+	 * specify 2KB or 4KB and/or make it per port instead of per device
+	 * for those who want to reduce memory footprint.  Note that the
+	 * rcvhdrentsize size must be large enough to hold the largest
+	 * IB header (currently 96 bytes) that we expect to handle (plus of
+	 * course the 2 dwords of RHF).
+	 */
+	if (dd->rcvegrbufsize == 2048)
+		dd->tidtemplate = IBA7322_TID_SZ_2K;
+	else if (dd->rcvegrbufsize == 4096)
+		dd->tidtemplate = IBA7322_TID_SZ_4K;
+	dd->tidinvalid = 0;
+}
+
+/**
+ * qib_init_7322_get_base_info - set chip-specific flags for user code
+ * @rcd: the qlogic_ib ctxt
+ * @kbase: qib_base_info pointer
+ *
+ * We set the PCIE flag because the lower bandwidth on PCIe vs
+ * HyperTransport can affect some user packet algorithims.
+ */
+
+static int qib_7322_get_base_info(struct qib_ctxtdata *rcd,
+				  struct qib_base_info *kinfo)
+{
+	kinfo->spi_runtime_flags |= QIB_RUNTIME_CTXT_MSB_IN_QP |
+		QIB_RUNTIME_PCIE | QIB_RUNTIME_NODMA_RTAIL |
+		QIB_RUNTIME_HDRSUPP | QIB_RUNTIME_SDMA;
+	if (rcd->dd->cspec->r1)
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_RCHK;
+	if (rcd->dd->flags & QIB_USE_SPCL_TRIG)
+		kinfo->spi_runtime_flags |= QIB_RUNTIME_SPECIAL_TRIGGER;
+
+	return 0;
+}
+
+static struct qib_message_header *
+qib_7322_get_msgheader(struct qib_devdata *dd, __le32 *rhf_addr)
+{
+	u32 offset = qib_hdrget_offset(rhf_addr);
+
+	return (struct qib_message_header *)
+		(rhf_addr - dd->rhf_offset + offset);
+}
+
+/*
+ * Configure number of contexts.
+ */
+static void qib_7322_config_ctxts(struct qib_devdata *dd)
+{
+	unsigned long flags;
+	u32 nchipctxts;
+
+	nchipctxts = qib_read_kreg32(dd, kr_contextcnt);
+	dd->cspec->numctxts = nchipctxts;
+	if (qib_n_krcv_queues > 1 && dd->num_pports) {
+		dd->first_user_ctxt = NUM_IB_PORTS +
+			(qib_n_krcv_queues - 1) * dd->num_pports;
+		if (dd->first_user_ctxt > nchipctxts)
+			dd->first_user_ctxt = nchipctxts;
+		dd->n_krcv_queues = dd->first_user_ctxt / dd->num_pports;
+	} else {
+		dd->first_user_ctxt = NUM_IB_PORTS;
+		dd->n_krcv_queues = 1;
+	}
+
+	if (!qib_cfgctxts) {
+		int nctxts = dd->first_user_ctxt + num_online_cpus();
+
+		if (nctxts <= 6)
+			dd->ctxtcnt = 6;
+		else if (nctxts <= 10)
+			dd->ctxtcnt = 10;
+		else if (nctxts <= nchipctxts)
+			dd->ctxtcnt = nchipctxts;
+	} else if (qib_cfgctxts < dd->num_pports)
+		dd->ctxtcnt = dd->num_pports;
+	else if (qib_cfgctxts <= nchipctxts)
+		dd->ctxtcnt = qib_cfgctxts;
+	if (!dd->ctxtcnt) /* none of the above, set to max */
+		dd->ctxtcnt = nchipctxts;
+
+	/*
+	 * Chip can be configured for 6, 10, or 18 ctxts, and choice
+	 * affects number of eager TIDs per ctxt (1K, 2K, 4K).
+	 * Lock to be paranoid about later motion, etc.
+	 */
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+	if (dd->ctxtcnt > 10)
+		dd->rcvctrl |= 2ULL << SYM_LSB(RcvCtrl, ContextCfg);
+	else if (dd->ctxtcnt > 6)
+		dd->rcvctrl |= 1ULL << SYM_LSB(RcvCtrl, ContextCfg);
+	/* else configure for default 6 receive ctxts */
+
+	/* The XRC opcode is 5. */
+	dd->rcvctrl |= 5ULL << SYM_LSB(RcvCtrl, XrcTypeCode);
+
+	/*
+	 * RcvCtrl *must* be written here so that the
+	 * chip understands how to change rcvegrcnt below.
+	 */
+	qib_write_kreg(dd, kr_rcvctrl, dd->rcvctrl);
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+
+	/* kr_rcvegrcnt changes based on the number of contexts enabled */
+	dd->cspec->rcvegrcnt = qib_read_kreg32(dd, kr_rcvegrcnt);
+	if (qib_rcvhdrcnt)
+		dd->rcvhdrcnt = max(dd->cspec->rcvegrcnt, qib_rcvhdrcnt);
+	else
+		dd->rcvhdrcnt = 2 * max(dd->cspec->rcvegrcnt,
+				    dd->num_pports > 1 ? 1024U : 2048U);
+}
+
+static int qib_7322_get_ib_cfg(struct qib_pportdata *ppd, int which)
+{
+
+	int lsb, ret = 0;
+	u64 maskr; /* right-justified mask */
+
+	switch (which) {
+
+	case QIB_IB_CFG_LWID_ENB: /* Get allowed Link-width */
+		ret = ppd->link_width_enabled;
+		goto done;
+
+	case QIB_IB_CFG_LWID: /* Get currently active Link-width */
+		ret = ppd->link_width_active;
+		goto done;
+
+	case QIB_IB_CFG_SPD_ENB: /* Get allowed Link speeds */
+		ret = ppd->link_speed_enabled;
+		goto done;
+
+	case QIB_IB_CFG_SPD: /* Get current Link spd */
+		ret = ppd->link_speed_active;
+		goto done;
+
+	case QIB_IB_CFG_RXPOL_ENB: /* Get Auto-RX-polarity enable */
+		lsb = SYM_LSB(IBCCtrlB_0, IB_POLARITY_REV_SUPP);
+		maskr = SYM_RMASK(IBCCtrlB_0, IB_POLARITY_REV_SUPP);
+		break;
+
+	case QIB_IB_CFG_LREV_ENB: /* Get Auto-Lane-reversal enable */
+		lsb = SYM_LSB(IBCCtrlB_0, IB_LANE_REV_SUPPORTED);
+		maskr = SYM_RMASK(IBCCtrlB_0, IB_LANE_REV_SUPPORTED);
+		break;
+
+	case QIB_IB_CFG_LINKLATENCY:
+		ret = qib_read_kreg_port(ppd, krp_ibcstatus_b) &
+			SYM_MASK(IBCStatusB_0, LinkRoundTripLatency);
+		goto done;
+
+	case QIB_IB_CFG_OP_VLS:
+		ret = ppd->vls_operational;
+		goto done;
+
+	case QIB_IB_CFG_VL_HIGH_CAP:
+		ret = 16;
+		goto done;
+
+	case QIB_IB_CFG_VL_LOW_CAP:
+		ret = 16;
+		goto done;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		ret = SYM_FIELD(ppd->cpspec->ibcctrl_a, IBCCtrlA_0,
+				OverrunThreshold);
+		goto done;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		ret = SYM_FIELD(ppd->cpspec->ibcctrl_a, IBCCtrlA_0,
+				PhyerrThreshold);
+		goto done;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		ret = (ppd->cpspec->ibcctrl_a &
+		       SYM_MASK(IBCCtrlA_0, LinkDownDefaultState)) ?
+			IB_LINKINITCMD_SLEEP : IB_LINKINITCMD_POLL;
+		goto done;
+
+	case QIB_IB_CFG_HRTBT: /* Get Heartbeat off/enable/auto */
+		lsb = IBA7322_IBC_HRTBT_LSB;
+		maskr = IBA7322_IBC_HRTBT_RMASK; /* OR of AUTO and ENB */
+		break;
+
+	case QIB_IB_CFG_PMA_TICKS:
+		/*
+		 * 0x00 = 10x link transfer rate or 4 nsec. for 2.5Gbs
+		 * Since the clock is always 250MHz, the value is 3, 1 or 0.
+		 */
+		if (ppd->link_speed_active == QIB_IB_QDR)
+			ret = 3;
+		else if (ppd->link_speed_active == QIB_IB_DDR)
+			ret = 1;
+		else
+			ret = 0;
+		goto done;
+
+	default:
+		ret = -EINVAL;
+		goto done;
+	}
+	ret = (int)((ppd->cpspec->ibcctrl_b >> lsb) & maskr);
+done:
+	return ret;
+}
+
+/*
+ * Below again cribbed liberally from older version. Do not lean
+ * heavily on it.
+ */
+#define IBA7322_IBC_DLIDLMC_SHIFT QIB_7322_IBCCtrlB_0_IB_DLID_LSB
+#define IBA7322_IBC_DLIDLMC_MASK (QIB_7322_IBCCtrlB_0_IB_DLID_RMASK \
+	| (QIB_7322_IBCCtrlB_0_IB_DLID_MASK_RMASK << 16))
+
+static int qib_7322_set_ib_cfg(struct qib_pportdata *ppd, int which, u32 val)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 maskr; /* right-justified mask */
+	int lsb, ret = 0;
+	u16 lcmd, licmd;
+	unsigned long flags;
+
+	switch (which) {
+	case QIB_IB_CFG_LIDLMC:
+		/*
+		 * Set LID and LMC. Combined to avoid possible hazard
+		 * caller puts LMC in 16MSbits, DLID in 16LSbits of val
+		 */
+		lsb = IBA7322_IBC_DLIDLMC_SHIFT;
+		maskr = IBA7322_IBC_DLIDLMC_MASK;
+		/*
+		 * For header-checking, the SLID in the packet will
+		 * be masked with SendIBSLMCMask, and compared
+		 * with SendIBSLIDAssignMask. Make sure we do not
+		 * set any bits not covered by the mask, or we get
+		 * false-positives.
+		 */
+		qib_write_kreg_port(ppd, krp_sendslid,
+				    val & (val >> 16) & SendIBSLIDAssignMask);
+		qib_write_kreg_port(ppd, krp_sendslidmask,
+				    (val >> 16) & SendIBSLMCMask);
+		break;
+
+	case QIB_IB_CFG_LWID_ENB: /* set allowed Link-width */
+		ppd->link_width_enabled = val;
+		/* convert IB value to chip register value */
+		if (val == IB_WIDTH_1X)
+			val = 0;
+		else if (val == IB_WIDTH_4X)
+			val = 1;
+		else
+			val = 3;
+		maskr = SYM_RMASK(IBCCtrlB_0, IB_NUM_CHANNELS);
+		lsb = SYM_LSB(IBCCtrlB_0, IB_NUM_CHANNELS);
+		break;
+
+	case QIB_IB_CFG_SPD_ENB: /* set allowed Link speeds */
+		/*
+		 * As with width, only write the actual register if the
+		 * link is currently down, otherwise takes effect on next
+		 * link change.  Since setting is being explicitly requested
+		 * (via MAD or sysfs), clear autoneg failure status if speed
+		 * autoneg is enabled.
+		 */
+		ppd->link_speed_enabled = val;
+		val <<= IBA7322_IBC_SPEED_LSB;
+		maskr = IBA7322_IBC_SPEED_MASK | IBA7322_IBC_IBTA_1_2_MASK |
+			IBA7322_IBC_MAX_SPEED_MASK;
+		if (val & (val - 1)) {
+			/* Muliple speeds enabled */
+			val |= IBA7322_IBC_IBTA_1_2_MASK |
+				IBA7322_IBC_MAX_SPEED_MASK;
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		} else if (val & IBA7322_IBC_SPEED_QDR)
+			val |= IBA7322_IBC_IBTA_1_2_MASK;
+		/* IBTA 1.2 mode + min/max + speed bits are contiguous */
+		lsb = SYM_LSB(IBCCtrlB_0, IB_ENHANCED_MODE);
+		break;
+
+	case QIB_IB_CFG_RXPOL_ENB: /* set Auto-RX-polarity enable */
+		lsb = SYM_LSB(IBCCtrlB_0, IB_POLARITY_REV_SUPP);
+		maskr = SYM_RMASK(IBCCtrlB_0, IB_POLARITY_REV_SUPP);
+		break;
+
+	case QIB_IB_CFG_LREV_ENB: /* set Auto-Lane-reversal enable */
+		lsb = SYM_LSB(IBCCtrlB_0, IB_LANE_REV_SUPPORTED);
+		maskr = SYM_RMASK(IBCCtrlB_0, IB_LANE_REV_SUPPORTED);
+		break;
+
+	case QIB_IB_CFG_OVERRUN_THRESH: /* IB overrun threshold */
+		maskr = SYM_FIELD(ppd->cpspec->ibcctrl_a, IBCCtrlA_0,
+				  OverrunThreshold);
+		if (maskr != val) {
+			ppd->cpspec->ibcctrl_a &=
+				~SYM_MASK(IBCCtrlA_0, OverrunThreshold);
+			ppd->cpspec->ibcctrl_a |= (u64) val <<
+				SYM_LSB(IBCCtrlA_0, OverrunThreshold);
+			qib_write_kreg_port(ppd, krp_ibcctrl_a,
+					    ppd->cpspec->ibcctrl_a);
+			qib_write_kreg(dd, kr_scratch, 0ULL);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_PHYERR_THRESH: /* IB PHY error threshold */
+		maskr = SYM_FIELD(ppd->cpspec->ibcctrl_a, IBCCtrlA_0,
+				  PhyerrThreshold);
+		if (maskr != val) {
+			ppd->cpspec->ibcctrl_a &=
+				~SYM_MASK(IBCCtrlA_0, PhyerrThreshold);
+			ppd->cpspec->ibcctrl_a |= (u64) val <<
+				SYM_LSB(IBCCtrlA_0, PhyerrThreshold);
+			qib_write_kreg_port(ppd, krp_ibcctrl_a,
+					    ppd->cpspec->ibcctrl_a);
+			qib_write_kreg(dd, kr_scratch, 0ULL);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_PKEYS: /* update pkeys */
+		maskr = (u64) ppd->pkeys[0] | ((u64) ppd->pkeys[1] << 16) |
+			((u64) ppd->pkeys[2] << 32) |
+			((u64) ppd->pkeys[3] << 48);
+		qib_write_kreg_port(ppd, krp_partitionkey, maskr);
+		goto bail;
+
+	case QIB_IB_CFG_LINKDEFAULT: /* IB link default (sleep/poll) */
+		/* will only take effect when the link state changes */
+		if (val == IB_LINKINITCMD_POLL)
+			ppd->cpspec->ibcctrl_a &=
+				~SYM_MASK(IBCCtrlA_0, LinkDownDefaultState);
+		else /* SLEEP */
+			ppd->cpspec->ibcctrl_a |=
+				SYM_MASK(IBCCtrlA_0, LinkDownDefaultState);
+		qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a);
+		qib_write_kreg(dd, kr_scratch, 0ULL);
+		goto bail;
+
+	case QIB_IB_CFG_MTU: /* update the MTU in IBC */
+		/*
+		 * Update our housekeeping variables, and set IBC max
+		 * size, same as init code; max IBC is max we allow in
+		 * buffer, less the qword pbc, plus 1 for ICRC, in dwords
+		 * Set even if it's unchanged, print debug message only
+		 * on changes.
+		 */
+		val = (ppd->ibmaxlen >> 2) + 1;
+		ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0, MaxPktLen);
+		ppd->cpspec->ibcctrl_a |= (u64)val <<
+			SYM_LSB(IBCCtrlA_0, MaxPktLen);
+		qib_write_kreg_port(ppd, krp_ibcctrl_a,
+				    ppd->cpspec->ibcctrl_a);
+		qib_write_kreg(dd, kr_scratch, 0ULL);
+		goto bail;
+
+	case QIB_IB_CFG_LSTATE: /* set the IB link state */
+		switch (val & 0xffff0000) {
+		case IB_LINKCMD_DOWN:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_DOWN;
+			ppd->cpspec->ibmalfusesnap = 1;
+			ppd->cpspec->ibmalfsnap = read_7322_creg32_port(ppd,
+				crp_errlink);
+			if (!ppd->cpspec->ibdeltainprog &&
+			    qib_compat_ddr_negotiate) {
+				ppd->cpspec->ibdeltainprog = 1;
+				ppd->cpspec->ibsymsnap =
+					read_7322_creg32_port(ppd,
+							      crp_ibsymbolerr);
+				ppd->cpspec->iblnkerrsnap =
+					read_7322_creg32_port(ppd,
+						      crp_iblinkerrrecov);
+			}
+			break;
+
+		case IB_LINKCMD_ARMED:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ARMED;
+			if (ppd->cpspec->ibmalfusesnap) {
+				ppd->cpspec->ibmalfusesnap = 0;
+				ppd->cpspec->ibmalfdelta +=
+					read_7322_creg32_port(ppd,
+							      crp_errlink) -
+					ppd->cpspec->ibmalfsnap;
+			}
+			break;
+
+		case IB_LINKCMD_ACTIVE:
+			lcmd = QLOGIC_IB_IBCC_LINKCMD_ACTIVE;
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkcmd req 0x%x\n", val >> 16);
+			goto bail;
+		}
+		switch (val & 0xffff) {
+		case IB_LINKINITCMD_NOP:
+			licmd = 0;
+			break;
+
+		case IB_LINKINITCMD_POLL:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_POLL;
+			break;
+
+		case IB_LINKINITCMD_SLEEP:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_SLEEP;
+			break;
+
+		case IB_LINKINITCMD_DISABLE:
+			licmd = QLOGIC_IB_IBCC_LINKINITCMD_DISABLE;
+			ppd->cpspec->chase_end = 0;
+			/*
+			 * stop state chase counter and timer, if running.
+			 * wait forpending timer, but don't clear .data (ppd)!
+			 */
+			if (ppd->cpspec->chase_timer.expires) {
+				del_timer_sync(&ppd->cpspec->chase_timer);
+				ppd->cpspec->chase_timer.expires = 0;
+			}
+			break;
+
+		default:
+			ret = -EINVAL;
+			qib_dev_err(dd, "bad linkinitcmd req 0x%x\n",
+				    val & 0xffff);
+			goto bail;
+		}
+		qib_set_ib_7322_lstate(ppd, lcmd, licmd);
+		goto bail;
+
+	case QIB_IB_CFG_OP_VLS:
+		if (ppd->vls_operational != val) {
+			ppd->vls_operational = val;
+			set_vls(ppd);
+		}
+		goto bail;
+
+	case QIB_IB_CFG_VL_HIGH_LIMIT:
+		qib_write_kreg_port(ppd, krp_highprio_limit, val);
+		goto bail;
+
+	case QIB_IB_CFG_HRTBT: /* set Heartbeat off/enable/auto */
+		if (val > 3) {
+			ret = -EINVAL;
+			goto bail;
+		}
+		lsb = IBA7322_IBC_HRTBT_LSB;
+		maskr = IBA7322_IBC_HRTBT_RMASK; /* OR of AUTO and ENB */
+		break;
+
+	case QIB_IB_CFG_PORT:
+		/* val is the port number of the switch we are connected to. */
+		if (ppd->dd->cspec->r1) {
+			cancel_delayed_work(&ppd->cpspec->ipg_work);
+			ppd->cpspec->ipg_tries = 0;
+		}
+		goto bail;
+
+	default:
+		ret = -EINVAL;
+		goto bail;
+	}
+	ppd->cpspec->ibcctrl_b &= ~(maskr << lsb);
+	ppd->cpspec->ibcctrl_b |= (((u64) val & maskr) << lsb);
+	qib_write_kreg_port(ppd, krp_ibcctrl_b, ppd->cpspec->ibcctrl_b);
+	qib_write_kreg(dd, kr_scratch, 0);
+bail:
+	return ret;
+}
+
+static int qib_7322_set_loopback(struct qib_pportdata *ppd, const char *what)
+{
+	int ret = 0;
+	u64 val, ctrlb;
+
+	/* only IBC loopback, may add serdes and xgxs loopbacks later */
+	if (!strncmp(what, "ibc", 3)) {
+		ppd->cpspec->ibcctrl_a |= SYM_MASK(IBCCtrlA_0,
+						       Loopback);
+		val = 0; /* disable heart beat, so link will come up */
+		qib_devinfo(ppd->dd->pcidev, "Enabling IB%u:%u IBC loopback\n",
+			 ppd->dd->unit, ppd->port);
+	} else if (!strncmp(what, "off", 3)) {
+		ppd->cpspec->ibcctrl_a &= ~SYM_MASK(IBCCtrlA_0,
+							Loopback);
+		/* enable heart beat again */
+		val = IBA7322_IBC_HRTBT_RMASK << IBA7322_IBC_HRTBT_LSB;
+		qib_devinfo(ppd->dd->pcidev,
+			"Disabling IB%u:%u IBC loopback (normal)\n",
+			ppd->dd->unit, ppd->port);
+	} else
+		ret = -EINVAL;
+	if (!ret) {
+		qib_write_kreg_port(ppd, krp_ibcctrl_a,
+				    ppd->cpspec->ibcctrl_a);
+		ctrlb = ppd->cpspec->ibcctrl_b & ~(IBA7322_IBC_HRTBT_MASK
+					     << IBA7322_IBC_HRTBT_LSB);
+		ppd->cpspec->ibcctrl_b = ctrlb | val;
+		qib_write_kreg_port(ppd, krp_ibcctrl_b,
+				    ppd->cpspec->ibcctrl_b);
+		qib_write_kreg(ppd->dd, kr_scratch, 0);
+	}
+	return ret;
+}
+
+static void get_vl_weights(struct qib_pportdata *ppd, unsigned regno,
+			   struct ib_vl_weight_elem *vl)
+{
+	unsigned i;
+
+	for (i = 0; i < 16; i++, regno++, vl++) {
+		u32 val = qib_read_kreg_port(ppd, regno);
+
+		vl->vl = (val >> SYM_LSB(LowPriority0_0, VirtualLane)) &
+			SYM_RMASK(LowPriority0_0, VirtualLane);
+		vl->weight = (val >> SYM_LSB(LowPriority0_0, Weight)) &
+			SYM_RMASK(LowPriority0_0, Weight);
+	}
+}
+
+static void set_vl_weights(struct qib_pportdata *ppd, unsigned regno,
+			   struct ib_vl_weight_elem *vl)
+{
+	unsigned i;
+
+	for (i = 0; i < 16; i++, regno++, vl++) {
+		u64 val;
+
+		val = ((vl->vl & SYM_RMASK(LowPriority0_0, VirtualLane)) <<
+			SYM_LSB(LowPriority0_0, VirtualLane)) |
+		      ((vl->weight & SYM_RMASK(LowPriority0_0, Weight)) <<
+			SYM_LSB(LowPriority0_0, Weight));
+		qib_write_kreg_port(ppd, regno, val);
+	}
+	if (!(ppd->p_sendctrl & SYM_MASK(SendCtrl_0, IBVLArbiterEn))) {
+		struct qib_devdata *dd = ppd->dd;
+		unsigned long flags;
+
+		spin_lock_irqsave(&dd->sendctrl_lock, flags);
+		ppd->p_sendctrl |= SYM_MASK(SendCtrl_0, IBVLArbiterEn);
+		qib_write_kreg_port(ppd, krp_sendctrl, ppd->p_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+		spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+	}
+}
+
+static int qib_7322_get_ib_table(struct qib_pportdata *ppd, int which, void *t)
+{
+	switch (which) {
+	case QIB_IB_TBL_VL_HIGH_ARB:
+		get_vl_weights(ppd, krp_highprio_0, t);
+		break;
+
+	case QIB_IB_TBL_VL_LOW_ARB:
+		get_vl_weights(ppd, krp_lowprio_0, t);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static int qib_7322_set_ib_table(struct qib_pportdata *ppd, int which, void *t)
+{
+	switch (which) {
+	case QIB_IB_TBL_VL_HIGH_ARB:
+		set_vl_weights(ppd, krp_highprio_0, t);
+		break;
+
+	case QIB_IB_TBL_VL_LOW_ARB:
+		set_vl_weights(ppd, krp_lowprio_0, t);
+		break;
+
+	default:
+		return -EINVAL;
+	}
+	return 0;
+}
+
+static void qib_update_7322_usrhead(struct qib_ctxtdata *rcd, u64 hd,
+				    u32 updegr, u32 egrhd, u32 npkts)
+{
+	/*
+	 * Need to write timeout register before updating rcvhdrhead to ensure
+	 * that the timer is enabled on reception of a packet.
+	 */
+	if (hd >> IBA7322_HDRHEAD_PKTINT_SHIFT)
+		adjust_rcv_timeout(rcd, npkts);
+	if (updegr)
+		qib_write_ureg(rcd->dd, ur_rcvegrindexhead, egrhd, rcd->ctxt);
+	mmiowb();
+	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
+	qib_write_ureg(rcd->dd, ur_rcvhdrhead, hd, rcd->ctxt);
+	mmiowb();
+}
+
+static u32 qib_7322_hdrqempty(struct qib_ctxtdata *rcd)
+{
+	u32 head, tail;
+
+	head = qib_read_ureg32(rcd->dd, ur_rcvhdrhead, rcd->ctxt);
+	if (rcd->rcvhdrtail_kvaddr)
+		tail = qib_get_rcvhdrtail(rcd);
+	else
+		tail = qib_read_ureg32(rcd->dd, ur_rcvhdrtail, rcd->ctxt);
+	return head == tail;
+}
+
+#define RCVCTRL_COMMON_MODS (QIB_RCVCTRL_CTXT_ENB | \
+	QIB_RCVCTRL_CTXT_DIS | \
+	QIB_RCVCTRL_TIDFLOW_ENB | \
+	QIB_RCVCTRL_TIDFLOW_DIS | \
+	QIB_RCVCTRL_TAILUPD_ENB | \
+	QIB_RCVCTRL_TAILUPD_DIS | \
+	QIB_RCVCTRL_INTRAVAIL_ENB | \
+	QIB_RCVCTRL_INTRAVAIL_DIS | \
+	QIB_RCVCTRL_BP_ENB | \
+	QIB_RCVCTRL_BP_DIS)
+
+#define RCVCTRL_PORT_MODS (QIB_RCVCTRL_CTXT_ENB | \
+	QIB_RCVCTRL_CTXT_DIS | \
+	QIB_RCVCTRL_PKEY_DIS | \
+	QIB_RCVCTRL_PKEY_ENB)
+
+/*
+ * Modify the RCVCTRL register in chip-specific way. This
+ * is a function because bit positions and (future) register
+ * location is chip-specifc, but the needed operations are
+ * generic. <op> is a bit-mask because we often want to
+ * do multiple modifications.
+ */
+static void rcvctrl_7322_mod(struct qib_pportdata *ppd, unsigned int op,
+			     int ctxt)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	u64 mask, val;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+
+	if (op & QIB_RCVCTRL_TIDFLOW_ENB)
+		dd->rcvctrl |= SYM_MASK(RcvCtrl, TidFlowEnable);
+	if (op & QIB_RCVCTRL_TIDFLOW_DIS)
+		dd->rcvctrl &= ~SYM_MASK(RcvCtrl, TidFlowEnable);
+	if (op & QIB_RCVCTRL_TAILUPD_ENB)
+		dd->rcvctrl |= SYM_MASK(RcvCtrl, TailUpd);
+	if (op & QIB_RCVCTRL_TAILUPD_DIS)
+		dd->rcvctrl &= ~SYM_MASK(RcvCtrl, TailUpd);
+	if (op & QIB_RCVCTRL_PKEY_ENB)
+		ppd->p_rcvctrl &= ~SYM_MASK(RcvCtrl_0, RcvPartitionKeyDisable);
+	if (op & QIB_RCVCTRL_PKEY_DIS)
+		ppd->p_rcvctrl |= SYM_MASK(RcvCtrl_0, RcvPartitionKeyDisable);
+	if (ctxt < 0) {
+		mask = (1ULL << dd->ctxtcnt) - 1;
+		rcd = NULL;
+	} else {
+		mask = (1ULL << ctxt);
+		rcd = dd->rcd[ctxt];
+	}
+	if ((op & QIB_RCVCTRL_CTXT_ENB) && rcd) {
+		ppd->p_rcvctrl |=
+			(mask << SYM_LSB(RcvCtrl_0, ContextEnableKernel));
+		if (!(dd->flags & QIB_NODMA_RTAIL)) {
+			op |= QIB_RCVCTRL_TAILUPD_ENB; /* need reg write */
+			dd->rcvctrl |= SYM_MASK(RcvCtrl, TailUpd);
+		}
+		/* Write these registers before the context is enabled. */
+		qib_write_kreg_ctxt(dd, krc_rcvhdrtailaddr, ctxt,
+				    rcd->rcvhdrqtailaddr_phys);
+		qib_write_kreg_ctxt(dd, krc_rcvhdraddr, ctxt,
+				    rcd->rcvhdrq_phys);
+		rcd->seq_cnt = 1;
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS)
+		ppd->p_rcvctrl &=
+			~(mask << SYM_LSB(RcvCtrl_0, ContextEnableKernel));
+	if (op & QIB_RCVCTRL_BP_ENB)
+		dd->rcvctrl |= mask << SYM_LSB(RcvCtrl, dontDropRHQFull);
+	if (op & QIB_RCVCTRL_BP_DIS)
+		dd->rcvctrl &= ~(mask << SYM_LSB(RcvCtrl, dontDropRHQFull));
+	if (op & QIB_RCVCTRL_INTRAVAIL_ENB)
+		dd->rcvctrl |= (mask << SYM_LSB(RcvCtrl, IntrAvail));
+	if (op & QIB_RCVCTRL_INTRAVAIL_DIS)
+		dd->rcvctrl &= ~(mask << SYM_LSB(RcvCtrl, IntrAvail));
+	/*
+	 * Decide which registers to write depending on the ops enabled.
+	 * Special case is "flush" (no bits set at all)
+	 * which needs to write both.
+	 */
+	if (op == 0 || (op & RCVCTRL_COMMON_MODS))
+		qib_write_kreg(dd, kr_rcvctrl, dd->rcvctrl);
+	if (op == 0 || (op & RCVCTRL_PORT_MODS))
+		qib_write_kreg_port(ppd, krp_rcvctrl, ppd->p_rcvctrl);
+	if ((op & QIB_RCVCTRL_CTXT_ENB) && dd->rcd[ctxt]) {
+		/*
+		 * Init the context registers also; if we were
+		 * disabled, tail and head should both be zero
+		 * already from the enable, but since we don't
+		 * know, we have to do it explicitly.
+		 */
+		val = qib_read_ureg32(dd, ur_rcvegrindextail, ctxt);
+		qib_write_ureg(dd, ur_rcvegrindexhead, val, ctxt);
+
+		/* be sure enabling write seen; hd/tl should be 0 */
+		(void) qib_read_kreg32(dd, kr_scratch);
+		val = qib_read_ureg32(dd, ur_rcvhdrtail, ctxt);
+		dd->rcd[ctxt]->head = val;
+		/* If kctxt, interrupt on next receive. */
+		if (ctxt < dd->first_user_ctxt)
+			val |= dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	} else if ((op & QIB_RCVCTRL_INTRAVAIL_ENB) &&
+		dd->rcd[ctxt] && dd->rhdrhead_intr_off) {
+		/* arm rcv interrupt */
+		val = dd->rcd[ctxt]->head | dd->rhdrhead_intr_off;
+		qib_write_ureg(dd, ur_rcvhdrhead, val, ctxt);
+	}
+	if (op & QIB_RCVCTRL_CTXT_DIS) {
+		unsigned f;
+
+		/* Now that the context is disabled, clear these registers. */
+		if (ctxt >= 0) {
+			qib_write_kreg_ctxt(dd, krc_rcvhdrtailaddr, ctxt, 0);
+			qib_write_kreg_ctxt(dd, krc_rcvhdraddr, ctxt, 0);
+			for (f = 0; f < NUM_TIDFLOWS_CTXT; f++)
+				qib_write_ureg(dd, ur_rcvflowtable + f,
+					       TIDFLOW_ERRBITS, ctxt);
+		} else {
+			unsigned i;
+
+			for (i = 0; i < dd->cfgctxts; i++) {
+				qib_write_kreg_ctxt(dd, krc_rcvhdrtailaddr,
+						    i, 0);
+				qib_write_kreg_ctxt(dd, krc_rcvhdraddr, i, 0);
+				for (f = 0; f < NUM_TIDFLOWS_CTXT; f++)
+					qib_write_ureg(dd, ur_rcvflowtable + f,
+						       TIDFLOW_ERRBITS, i);
+			}
+		}
+	}
+	spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+}
+
+/*
+ * Modify the SENDCTRL register in chip-specific way. This
+ * is a function where there are multiple such registers with
+ * slightly different layouts.
+ * The chip doesn't allow back-to-back sendctrl writes, so write
+ * the scratch register after writing sendctrl.
+ *
+ * Which register is written depends on the operation.
+ * Most operate on the common register, while
+ * SEND_ENB and SEND_DIS operate on the per-port ones.
+ * SEND_ENB is included in common because it can change SPCL_TRIG
+ */
+#define SENDCTRL_COMMON_MODS (\
+	QIB_SENDCTRL_CLEAR | \
+	QIB_SENDCTRL_AVAIL_DIS | \
+	QIB_SENDCTRL_AVAIL_ENB | \
+	QIB_SENDCTRL_AVAIL_BLIP | \
+	QIB_SENDCTRL_DISARM | \
+	QIB_SENDCTRL_DISARM_ALL | \
+	QIB_SENDCTRL_SEND_ENB)
+
+#define SENDCTRL_PORT_MODS (\
+	QIB_SENDCTRL_CLEAR | \
+	QIB_SENDCTRL_SEND_ENB | \
+	QIB_SENDCTRL_SEND_DIS | \
+	QIB_SENDCTRL_FLUSH)
+
+static void sendctrl_7322_mod(struct qib_pportdata *ppd, u32 op)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 tmp_dd_sendctrl;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+
+	/* First the dd ones that are "sticky", saved in shadow */
+	if (op & QIB_SENDCTRL_CLEAR)
+		dd->sendctrl = 0;
+	if (op & QIB_SENDCTRL_AVAIL_DIS)
+		dd->sendctrl &= ~SYM_MASK(SendCtrl, SendBufAvailUpd);
+	else if (op & QIB_SENDCTRL_AVAIL_ENB) {
+		dd->sendctrl |= SYM_MASK(SendCtrl, SendBufAvailUpd);
+		if (dd->flags & QIB_USE_SPCL_TRIG)
+			dd->sendctrl |= SYM_MASK(SendCtrl, SpecialTriggerEn);
+	}
+
+	/* Then the ppd ones that are "sticky", saved in shadow */
+	if (op & QIB_SENDCTRL_SEND_DIS)
+		ppd->p_sendctrl &= ~SYM_MASK(SendCtrl_0, SendEnable);
+	else if (op & QIB_SENDCTRL_SEND_ENB)
+		ppd->p_sendctrl |= SYM_MASK(SendCtrl_0, SendEnable);
+
+	if (op & QIB_SENDCTRL_DISARM_ALL) {
+		u32 i, last;
+
+		tmp_dd_sendctrl = dd->sendctrl;
+		last = dd->piobcnt2k + dd->piobcnt4k + NUM_VL15_BUFS;
+		/*
+		 * Disarm any buffers that are not yet launched,
+		 * disabling updates until done.
+		 */
+		tmp_dd_sendctrl &= ~SYM_MASK(SendCtrl, SendBufAvailUpd);
+		for (i = 0; i < last; i++) {
+			qib_write_kreg(dd, kr_sendctrl,
+				       tmp_dd_sendctrl |
+				       SYM_MASK(SendCtrl, Disarm) | i);
+			qib_write_kreg(dd, kr_scratch, 0);
+		}
+	}
+
+	if (op & QIB_SENDCTRL_FLUSH) {
+		u64 tmp_ppd_sendctrl = ppd->p_sendctrl;
+
+		/*
+		 * Now drain all the fifos.  The Abort bit should never be
+		 * needed, so for now, at least, we don't use it.
+		 */
+		tmp_ppd_sendctrl |=
+			SYM_MASK(SendCtrl_0, TxeDrainRmFifo) |
+			SYM_MASK(SendCtrl_0, TxeDrainLaFifo) |
+			SYM_MASK(SendCtrl_0, TxeBypassIbc);
+		qib_write_kreg_port(ppd, krp_sendctrl, tmp_ppd_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	tmp_dd_sendctrl = dd->sendctrl;
+
+	if (op & QIB_SENDCTRL_DISARM)
+		tmp_dd_sendctrl |= SYM_MASK(SendCtrl, Disarm) |
+			((op & QIB_7322_SendCtrl_DisarmSendBuf_RMASK) <<
+			 SYM_LSB(SendCtrl, DisarmSendBuf));
+	if ((op & QIB_SENDCTRL_AVAIL_BLIP) &&
+	    (dd->sendctrl & SYM_MASK(SendCtrl, SendBufAvailUpd)))
+		tmp_dd_sendctrl &= ~SYM_MASK(SendCtrl, SendBufAvailUpd);
+
+	if (op == 0 || (op & SENDCTRL_COMMON_MODS)) {
+		qib_write_kreg(dd, kr_sendctrl, tmp_dd_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	if (op == 0 || (op & SENDCTRL_PORT_MODS)) {
+		qib_write_kreg_port(ppd, krp_sendctrl, ppd->p_sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	if (op & QIB_SENDCTRL_AVAIL_BLIP) {
+		qib_write_kreg(dd, kr_sendctrl, dd->sendctrl);
+		qib_write_kreg(dd, kr_scratch, 0);
+	}
+
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+
+	if (op & QIB_SENDCTRL_FLUSH) {
+		u32 v;
+		/*
+		 * ensure writes have hit chip, then do a few
+		 * more reads, to allow DMA of pioavail registers
+		 * to occur, so in-memory copy is in sync with
+		 * the chip.  Not always safe to sleep.
+		 */
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		v = qib_read_kreg32(dd, kr_scratch);
+		qib_write_kreg(dd, kr_scratch, v);
+		qib_read_kreg32(dd, kr_scratch);
+	}
+}
+
+#define _PORT_VIRT_FLAG 0x8000U /* "virtual", need adjustments */
+#define _PORT_64BIT_FLAG 0x10000U /* not "virtual", but 64bit */
+#define _PORT_CNTR_IDXMASK 0x7fffU /* mask off flags above */
+
+/**
+ * qib_portcntr_7322 - read a per-port chip counter
+ * @ppd: the qlogic_ib pport
+ * @creg: the counter to read (not a chip offset)
+ */
+static u64 qib_portcntr_7322(struct qib_pportdata *ppd, u32 reg)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 ret = 0ULL;
+	u16 creg;
+	/* 0xffff for unimplemented or synthesized counters */
+	static const u32 xlator[] = {
+		[QIBPORTCNTR_PKTSEND] = crp_pktsend | _PORT_64BIT_FLAG,
+		[QIBPORTCNTR_WORDSEND] = crp_wordsend | _PORT_64BIT_FLAG,
+		[QIBPORTCNTR_PSXMITDATA] = crp_psxmitdatacount,
+		[QIBPORTCNTR_PSXMITPKTS] = crp_psxmitpktscount,
+		[QIBPORTCNTR_PSXMITWAIT] = crp_psxmitwaitcount,
+		[QIBPORTCNTR_SENDSTALL] = crp_sendstall,
+		[QIBPORTCNTR_PKTRCV] = crp_pktrcv | _PORT_64BIT_FLAG,
+		[QIBPORTCNTR_PSRCVDATA] = crp_psrcvdatacount,
+		[QIBPORTCNTR_PSRCVPKTS] = crp_psrcvpktscount,
+		[QIBPORTCNTR_RCVEBP] = crp_rcvebp,
+		[QIBPORTCNTR_RCVOVFL] = crp_rcvovfl,
+		[QIBPORTCNTR_WORDRCV] = crp_wordrcv | _PORT_64BIT_FLAG,
+		[QIBPORTCNTR_RXDROPPKT] = 0xffff, /* not needed  for 7322 */
+		[QIBPORTCNTR_RXLOCALPHYERR] = crp_rxotherlocalphyerr,
+		[QIBPORTCNTR_RXVLERR] = crp_rxvlerr,
+		[QIBPORTCNTR_ERRICRC] = crp_erricrc,
+		[QIBPORTCNTR_ERRVCRC] = crp_errvcrc,
+		[QIBPORTCNTR_ERRLPCRC] = crp_errlpcrc,
+		[QIBPORTCNTR_BADFORMAT] = crp_badformat,
+		[QIBPORTCNTR_ERR_RLEN] = crp_err_rlen,
+		[QIBPORTCNTR_IBSYMBOLERR] = crp_ibsymbolerr,
+		[QIBPORTCNTR_INVALIDRLEN] = crp_invalidrlen,
+		[QIBPORTCNTR_UNSUPVL] = crp_txunsupvl,
+		[QIBPORTCNTR_EXCESSBUFOVFL] = crp_excessbufferovfl,
+		[QIBPORTCNTR_ERRLINK] = crp_errlink,
+		[QIBPORTCNTR_IBLINKDOWN] = crp_iblinkdown,
+		[QIBPORTCNTR_IBLINKERRRECOV] = crp_iblinkerrrecov,
+		[QIBPORTCNTR_LLI] = crp_locallinkintegrityerr,
+		[QIBPORTCNTR_VL15PKTDROP] = crp_vl15droppedpkt,
+		[QIBPORTCNTR_ERRPKEY] = crp_errpkey,
+		/*
+		 * the next 3 aren't really counters, but were implemented
+		 * as counters in older chips, so still get accessed as
+		 * though they were counters from this code.
+		 */
+		[QIBPORTCNTR_PSINTERVAL] = krp_psinterval,
+		[QIBPORTCNTR_PSSTART] = krp_psstart,
+		[QIBPORTCNTR_PSSTAT] = krp_psstat,
+		/* pseudo-counter, summed for all ports */
+		[QIBPORTCNTR_KHDROVFL] = 0xffff,
+	};
+
+	if (reg >= ARRAY_SIZE(xlator)) {
+		qib_devinfo(ppd->dd->pcidev,
+			 "Unimplemented portcounter %u\n", reg);
+		goto done;
+	}
+	creg = xlator[reg] & _PORT_CNTR_IDXMASK;
+
+	/* handle non-counters and special cases first */
+	if (reg == QIBPORTCNTR_KHDROVFL) {
+		int i;
+
+		/* sum over all kernel contexts (skip if mini_init) */
+		for (i = 0; dd->rcd && i < dd->first_user_ctxt; i++) {
+			struct qib_ctxtdata *rcd = dd->rcd[i];
+
+			if (!rcd || rcd->ppd != ppd)
+				continue;
+			ret += read_7322_creg32(dd, cr_base_egrovfl + i);
+		}
+		goto done;
+	} else if (reg == QIBPORTCNTR_RXDROPPKT) {
+		/*
+		 * Used as part of the synthesis of port_rcv_errors
+		 * in the verbs code for IBTA counters.  Not needed for 7322,
+		 * because all the errors are already counted by other cntrs.
+		 */
+		goto done;
+	} else if (reg == QIBPORTCNTR_PSINTERVAL ||
+		   reg == QIBPORTCNTR_PSSTART || reg == QIBPORTCNTR_PSSTAT) {
+		/* were counters in older chips, now per-port kernel regs */
+		ret = qib_read_kreg_port(ppd, creg);
+		goto done;
+	}
+
+	/*
+	 * Only fast increment counters are 64 bits; use 32 bit reads to
+	 * avoid two independent reads when on Opteron.
+	 */
+	if (xlator[reg] & _PORT_64BIT_FLAG)
+		ret = read_7322_creg_port(ppd, creg);
+	else
+		ret = read_7322_creg32_port(ppd, creg);
+	if (creg == crp_ibsymbolerr) {
+		if (ppd->cpspec->ibdeltainprog)
+			ret -= ret - ppd->cpspec->ibsymsnap;
+		ret -= ppd->cpspec->ibsymdelta;
+	} else if (creg == crp_iblinkerrrecov) {
+		if (ppd->cpspec->ibdeltainprog)
+			ret -= ret - ppd->cpspec->iblnkerrsnap;
+		ret -= ppd->cpspec->iblnkerrdelta;
+	} else if (creg == crp_errlink)
+		ret -= ppd->cpspec->ibmalfdelta;
+	else if (creg == crp_iblinkdown)
+		ret += ppd->cpspec->iblnkdowndelta;
+done:
+	return ret;
+}
+
+/*
+ * Device counter names (not port-specific), one line per stat,
+ * single string.  Used by utilities like ipathstats to print the stats
+ * in a way which works for different versions of drivers, without changing
+ * the utility.  Names need to be 12 chars or less (w/o newline), for proper
+ * display by utility.
+ * Non-error counters are first.
+ * Start of "error" conters is indicated by a leading "E " on the first
+ * "error" counter, and doesn't count in label length.
+ * The EgrOvfl list needs to be last so we truncate them at the configured
+ * context count for the device.
+ * cntr7322indices contains the corresponding register indices.
+ */
+static const char cntr7322names[] =
+	"Interrupts\n"
+	"HostBusStall\n"
+	"E RxTIDFull\n"
+	"RxTIDInvalid\n"
+	"RxTIDFloDrop\n" /* 7322 only */
+	"Ctxt0EgrOvfl\n"
+	"Ctxt1EgrOvfl\n"
+	"Ctxt2EgrOvfl\n"
+	"Ctxt3EgrOvfl\n"
+	"Ctxt4EgrOvfl\n"
+	"Ctxt5EgrOvfl\n"
+	"Ctxt6EgrOvfl\n"
+	"Ctxt7EgrOvfl\n"
+	"Ctxt8EgrOvfl\n"
+	"Ctxt9EgrOvfl\n"
+	"Ctx10EgrOvfl\n"
+	"Ctx11EgrOvfl\n"
+	"Ctx12EgrOvfl\n"
+	"Ctx13EgrOvfl\n"
+	"Ctx14EgrOvfl\n"
+	"Ctx15EgrOvfl\n"
+	"Ctx16EgrOvfl\n"
+	"Ctx17EgrOvfl\n"
+	;
+
+static const u32 cntr7322indices[] = {
+	cr_lbint | _PORT_64BIT_FLAG,
+	cr_lbstall | _PORT_64BIT_FLAG,
+	cr_tidfull,
+	cr_tidinvalid,
+	cr_rxtidflowdrop,
+	cr_base_egrovfl + 0,
+	cr_base_egrovfl + 1,
+	cr_base_egrovfl + 2,
+	cr_base_egrovfl + 3,
+	cr_base_egrovfl + 4,
+	cr_base_egrovfl + 5,
+	cr_base_egrovfl + 6,
+	cr_base_egrovfl + 7,
+	cr_base_egrovfl + 8,
+	cr_base_egrovfl + 9,
+	cr_base_egrovfl + 10,
+	cr_base_egrovfl + 11,
+	cr_base_egrovfl + 12,
+	cr_base_egrovfl + 13,
+	cr_base_egrovfl + 14,
+	cr_base_egrovfl + 15,
+	cr_base_egrovfl + 16,
+	cr_base_egrovfl + 17,
+};
+
+/*
+ * same as cntr7322names and cntr7322indices, but for port-specific counters.
+ * portcntr7322indices is somewhat complicated by some registers needing
+ * adjustments of various kinds, and those are ORed with _PORT_VIRT_FLAG
+ */
+static const char portcntr7322names[] =
+	"TxPkt\n"
+	"TxFlowPkt\n"
+	"TxWords\n"
+	"RxPkt\n"
+	"RxFlowPkt\n"
+	"RxWords\n"
+	"TxFlowStall\n"
+	"TxDmaDesc\n"  /* 7220 and 7322-only */
+	"E RxDlidFltr\n"  /* 7220 and 7322-only */
+	"IBStatusChng\n"
+	"IBLinkDown\n"
+	"IBLnkRecov\n"
+	"IBRxLinkErr\n"
+	"IBSymbolErr\n"
+	"RxLLIErr\n"
+	"RxBadFormat\n"
+	"RxBadLen\n"
+	"RxBufOvrfl\n"
+	"RxEBP\n"
+	"RxFlowCtlErr\n"
+	"RxICRCerr\n"
+	"RxLPCRCerr\n"
+	"RxVCRCerr\n"
+	"RxInvalLen\n"
+	"RxInvalPKey\n"
+	"RxPktDropped\n"
+	"TxBadLength\n"
+	"TxDropped\n"
+	"TxInvalLen\n"
+	"TxUnderrun\n"
+	"TxUnsupVL\n"
+	"RxLclPhyErr\n" /* 7220 and 7322-only from here down */
+	"RxVL15Drop\n"
+	"RxVlErr\n"
+	"XcessBufOvfl\n"
+	"RxQPBadCtxt\n" /* 7322-only from here down */
+	"TXBadHeader\n"
+	;
+
+static const u32 portcntr7322indices[] = {
+	QIBPORTCNTR_PKTSEND | _PORT_VIRT_FLAG,
+	crp_pktsendflow,
+	QIBPORTCNTR_WORDSEND | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_PKTRCV | _PORT_VIRT_FLAG,
+	crp_pktrcvflowctrl,
+	QIBPORTCNTR_WORDRCV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_SENDSTALL | _PORT_VIRT_FLAG,
+	crp_txsdmadesc | _PORT_64BIT_FLAG,
+	crp_rxdlidfltr,
+	crp_ibstatuschange,
+	QIBPORTCNTR_IBLINKDOWN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBLINKERRRECOV | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLINK | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_IBSYMBOLERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_LLI | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_BADFORMAT | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERR_RLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVOVFL | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RCVEBP | _PORT_VIRT_FLAG,
+	crp_rcvflowctrlviol,
+	QIBPORTCNTR_ERRICRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRLPCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRVCRC | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_INVALIDRLEN | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_ERRPKEY | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RXDROPPKT | _PORT_VIRT_FLAG,
+	crp_txminmaxlenerr,
+	crp_txdroppedpkt,
+	crp_txlenerr,
+	crp_txunderrun,
+	crp_txunsupvl,
+	QIBPORTCNTR_RXLOCALPHYERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_VL15PKTDROP | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_RXVLERR | _PORT_VIRT_FLAG,
+	QIBPORTCNTR_EXCESSBUFOVFL | _PORT_VIRT_FLAG,
+	crp_rxqpinvalidctxt,
+	crp_txhdrerr,
+};
+
+/* do all the setup to make the counter reads efficient later */
+static void init_7322_cntrnames(struct qib_devdata *dd)
+{
+	int i, j = 0;
+	char *s;
+
+	for (i = 0, s = (char *)cntr7322names; s && j <= dd->cfgctxts;
+	     i++) {
+		/* we always have at least one counter before the egrovfl */
+		if (!j && !strncmp("Ctxt0EgrOvfl", s + 1, 12))
+			j = 1;
+		s = strchr(s + 1, '\n');
+		if (s && j)
+			j++;
+	}
+	dd->cspec->ncntrs = i;
+	if (!s)
+		/* full list; size is without terminating null */
+		dd->cspec->cntrnamelen = sizeof(cntr7322names) - 1;
+	else
+		dd->cspec->cntrnamelen = 1 + s - cntr7322names;
+	dd->cspec->cntrs = kmalloc(dd->cspec->ncntrs
+		* sizeof(u64), GFP_KERNEL);
+	if (!dd->cspec->cntrs)
+		qib_dev_err(dd, "Failed allocation for counters\n");
+
+	for (i = 0, s = (char *)portcntr7322names; s; i++)
+		s = strchr(s + 1, '\n');
+	dd->cspec->nportcntrs = i - 1;
+	dd->cspec->portcntrnamelen = sizeof(portcntr7322names) - 1;
+	for (i = 0; i < dd->num_pports; ++i) {
+		dd->pport[i].cpspec->portcntrs = kmalloc(dd->cspec->nportcntrs
+			* sizeof(u64), GFP_KERNEL);
+		if (!dd->pport[i].cpspec->portcntrs)
+			qib_dev_err(dd,
+				"Failed allocation for portcounters\n");
+	}
+}
+
+static u32 qib_read_7322cntrs(struct qib_devdata *dd, loff_t pos, char **namep,
+			      u64 **cntrp)
+{
+	u32 ret;
+
+	if (namep) {
+		ret = dd->cspec->cntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+		else
+			*namep = (char *) cntr7322names;
+	} else {
+		u64 *cntr = dd->cspec->cntrs;
+		int i;
+
+		ret = dd->cspec->ncntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->ncntrs; i++)
+			if (cntr7322indices[i] & _PORT_64BIT_FLAG)
+				*cntr++ = read_7322_creg(dd,
+							 cntr7322indices[i] &
+							 _PORT_CNTR_IDXMASK);
+			else
+				*cntr++ = read_7322_creg32(dd,
+							   cntr7322indices[i]);
+	}
+done:
+	return ret;
+}
+
+static u32 qib_read_7322portcntrs(struct qib_devdata *dd, loff_t pos, u32 port,
+				  char **namep, u64 **cntrp)
+{
+	u32 ret;
+
+	if (namep) {
+		ret = dd->cspec->portcntrnamelen;
+		if (pos >= ret)
+			ret = 0; /* final read after getting everything */
+		else
+			*namep = (char *)portcntr7322names;
+	} else {
+		struct qib_pportdata *ppd = &dd->pport[port];
+		u64 *cntr = ppd->cpspec->portcntrs;
+		int i;
+
+		ret = dd->cspec->nportcntrs * sizeof(u64);
+		if (!cntr || pos >= ret) {
+			/* everything read, or couldn't get memory */
+			ret = 0;
+			goto done;
+		}
+		*cntrp = cntr;
+		for (i = 0; i < dd->cspec->nportcntrs; i++) {
+			if (portcntr7322indices[i] & _PORT_VIRT_FLAG)
+				*cntr++ = qib_portcntr_7322(ppd,
+					portcntr7322indices[i] &
+					_PORT_CNTR_IDXMASK);
+			else if (portcntr7322indices[i] & _PORT_64BIT_FLAG)
+				*cntr++ = read_7322_creg_port(ppd,
+					   portcntr7322indices[i] &
+					    _PORT_CNTR_IDXMASK);
+			else
+				*cntr++ = read_7322_creg32_port(ppd,
+					   portcntr7322indices[i]);
+		}
+	}
+done:
+	return ret;
+}
+
+/**
+ * qib_get_7322_faststats - get word counters from chip before they overflow
+ * @opaque - contains a pointer to the qlogic_ib device qib_devdata
+ *
+ * VESTIGIAL IBA7322 has no "small fast counters", so the only
+ * real purpose of this function is to maintain the notion of
+ * "active time", which in turn is only logged into the eeprom,
+ * which we don;t have, yet, for 7322-based boards.
+ *
+ * called from add_timer
+ */
+static void qib_get_7322_faststats(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *) opaque;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+	u64 traffic_wds;
+	int pidx;
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+
+		/*
+		 * If port isn't enabled or not operational ports, or
+		 * diags is running (can cause memory diags to fail)
+		 * skip this port this time.
+		 */
+		if (!ppd->link_speed_supported || !(dd->flags & QIB_INITTED)
+		    || dd->diag_client)
+			continue;
+
+		/*
+		 * Maintain an activity timer, based on traffic
+		 * exceeding a threshold, so we need to check the word-counts
+		 * even if they are 64-bit.
+		 */
+		traffic_wds = qib_portcntr_7322(ppd, QIBPORTCNTR_WORDRCV) +
+			qib_portcntr_7322(ppd, QIBPORTCNTR_WORDSEND);
+		spin_lock_irqsave(&ppd->dd->eep_st_lock, flags);
+		traffic_wds -= ppd->dd->traffic_wds;
+		ppd->dd->traffic_wds += traffic_wds;
+		if (traffic_wds >= QIB_TRAFFIC_ACTIVE_THRESHOLD)
+			atomic_add(ACTIVITY_TIMER, &ppd->dd->active_time);
+		spin_unlock_irqrestore(&ppd->dd->eep_st_lock, flags);
+		if (ppd->cpspec->qdr_dfe_on && (ppd->link_speed_active &
+						QIB_IB_QDR) &&
+		    (ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED |
+				    QIBL_LINKACTIVE)) &&
+		    ppd->cpspec->qdr_dfe_time &&
+		    time_is_before_jiffies(ppd->cpspec->qdr_dfe_time)) {
+			ppd->cpspec->qdr_dfe_on = 0;
+
+			qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+					    ppd->dd->cspec->r1 ?
+					    QDR_STATIC_ADAPT_INIT_R1 :
+					    QDR_STATIC_ADAPT_INIT);
+			force_h1(ppd);
+		}
+	}
+	mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
+}
+
+/*
+ * If we were using MSIx, try to fallback to INTx.
+ */
+static int qib_7322_intr_fallback(struct qib_devdata *dd)
+{
+	if (!dd->cspec->num_msix_entries)
+		return 0; /* already using INTx */
+
+	qib_devinfo(dd->pcidev,
+		"MSIx interrupt not detected, trying INTx interrupts\n");
+	qib_7322_nomsix(dd);
+	qib_enable_intx(dd->pcidev);
+	qib_setup_7322_interrupt(dd, 0);
+	return 1;
+}
+
+/*
+ * Reset the XGXS (between serdes and IBC).  Slightly less intrusive
+ * than resetting the IBC or external link state, and useful in some
+ * cases to cause some retraining.  To do this right, we reset IBC
+ * as well, then return to previous state (which may be still in reset)
+ * NOTE: some callers of this "know" this writes the current value
+ * of cpspec->ibcctrl_a as part of it's operation, so if that changes,
+ * check all callers.
+ */
+static void qib_7322_mini_pcs_reset(struct qib_pportdata *ppd)
+{
+	u64 val;
+	struct qib_devdata *dd = ppd->dd;
+	const u64 reset_bits = SYM_MASK(IBPCSConfig_0, xcv_rreset) |
+		SYM_MASK(IBPCSConfig_0, xcv_treset) |
+		SYM_MASK(IBPCSConfig_0, tx_rx_reset);
+
+	val = qib_read_kreg_port(ppd, krp_ib_pcsconfig);
+	qib_write_kreg(dd, kr_hwerrmask,
+		       dd->cspec->hwerrmask & ~HWE_MASK(statusValidNoEop));
+	qib_write_kreg_port(ppd, krp_ibcctrl_a,
+			    ppd->cpspec->ibcctrl_a &
+			    ~SYM_MASK(IBCCtrlA_0, IBLinkEn));
+
+	qib_write_kreg_port(ppd, krp_ib_pcsconfig, val | reset_bits);
+	qib_read_kreg32(dd, kr_scratch);
+	qib_write_kreg_port(ppd, krp_ib_pcsconfig, val & ~reset_bits);
+	qib_write_kreg_port(ppd, krp_ibcctrl_a, ppd->cpspec->ibcctrl_a);
+	qib_write_kreg(dd, kr_scratch, 0ULL);
+	qib_write_kreg(dd, kr_hwerrclear,
+		       SYM_MASK(HwErrClear, statusValidNoEopClear));
+	qib_write_kreg(dd, kr_hwerrmask, dd->cspec->hwerrmask);
+}
+
+/*
+ * This code for non-IBTA-compliant IB speed negotiation is only known to
+ * work for the SDR to DDR transition, and only between an HCA and a switch
+ * with recent firmware.  It is based on observed heuristics, rather than
+ * actual knowledge of the non-compliant speed negotiation.
+ * It has a number of hard-coded fields, since the hope is to rewrite this
+ * when a spec is available on how the negoation is intended to work.
+ */
+static void autoneg_7322_sendpkt(struct qib_pportdata *ppd, u32 *hdr,
+				 u32 dcnt, u32 *data)
+{
+	int i;
+	u64 pbc;
+	u32 __iomem *piobuf;
+	u32 pnum, control, len;
+	struct qib_devdata *dd = ppd->dd;
+
+	i = 0;
+	len = 7 + dcnt + 1; /* 7 dword header, dword data, icrc */
+	control = qib_7322_setpbc_control(ppd, len, 0, 15);
+	pbc = ((u64) control << 32) | len;
+	while (!(piobuf = qib_7322_getsendbuf(ppd, pbc, &pnum))) {
+		if (i++ > 15)
+			return;
+		udelay(2);
+	}
+	/* disable header check on this packet, since it can't be valid */
+	dd->f_txchk_change(dd, pnum, 1, TXCHK_CHG_TYPE_DIS1, NULL);
+	writeq(pbc, piobuf);
+	qib_flush_wc();
+	qib_pio_copy(piobuf + 2, hdr, 7);
+	qib_pio_copy(piobuf + 9, data, dcnt);
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pnum >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pnum);
+	/* and re-enable hdr check */
+	dd->f_txchk_change(dd, pnum, 1, TXCHK_CHG_TYPE_ENAB1, NULL);
+}
+
+/*
+ * _start packet gets sent twice at start, _done gets sent twice at end
+ */
+static void qib_autoneg_7322_send(struct qib_pportdata *ppd, int which)
+{
+	struct qib_devdata *dd = ppd->dd;
+	static u32 swapped;
+	u32 dw, i, hcnt, dcnt, *data;
+	static u32 hdr[7] = { 0xf002ffff, 0x48ffff, 0x6400abba };
+	static u32 madpayload_start[0x40] = {
+		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
+		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+		0x1, 0x1388, 0x15e, 0x1, /* rest 0's */
+		};
+	static u32 madpayload_done[0x40] = {
+		0x1810103, 0x1, 0x0, 0x0, 0x2c90000, 0x2c9, 0x0, 0x0,
+		0xffffffff, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0, 0x0,
+		0x40000001, 0x1388, 0x15e, /* rest 0's */
+		};
+
+	dcnt = ARRAY_SIZE(madpayload_start);
+	hcnt = ARRAY_SIZE(hdr);
+	if (!swapped) {
+		/* for maintainability, do it at runtime */
+		for (i = 0; i < hcnt; i++) {
+			dw = (__force u32) cpu_to_be32(hdr[i]);
+			hdr[i] = dw;
+		}
+		for (i = 0; i < dcnt; i++) {
+			dw = (__force u32) cpu_to_be32(madpayload_start[i]);
+			madpayload_start[i] = dw;
+			dw = (__force u32) cpu_to_be32(madpayload_done[i]);
+			madpayload_done[i] = dw;
+		}
+		swapped = 1;
+	}
+
+	data = which ? madpayload_done : madpayload_start;
+
+	autoneg_7322_sendpkt(ppd, hdr, dcnt, data);
+	qib_read_kreg64(dd, kr_scratch);
+	udelay(2);
+	autoneg_7322_sendpkt(ppd, hdr, dcnt, data);
+	qib_read_kreg64(dd, kr_scratch);
+	udelay(2);
+}
+
+/*
+ * Do the absolute minimum to cause an IB speed change, and make it
+ * ready, but don't actually trigger the change.   The caller will
+ * do that when ready (if link is in Polling training state, it will
+ * happen immediately, otherwise when link next goes down)
+ *
+ * This routine should only be used as part of the DDR autonegotation
+ * code for devices that are not compliant with IB 1.2 (or code that
+ * fixes things up for same).
+ *
+ * When link has gone down, and autoneg enabled, or autoneg has
+ * failed and we give up until next time we set both speeds, and
+ * then we want IBTA enabled as well as "use max enabled speed.
+ */
+static void set_7322_ibspeed_fast(struct qib_pportdata *ppd, u32 speed)
+{
+	u64 newctrlb;
+	newctrlb = ppd->cpspec->ibcctrl_b & ~(IBA7322_IBC_SPEED_MASK |
+				    IBA7322_IBC_IBTA_1_2_MASK |
+				    IBA7322_IBC_MAX_SPEED_MASK);
+
+	if (speed & (speed - 1)) /* multiple speeds */
+		newctrlb |= (speed << IBA7322_IBC_SPEED_LSB) |
+				    IBA7322_IBC_IBTA_1_2_MASK |
+				    IBA7322_IBC_MAX_SPEED_MASK;
+	else
+		newctrlb |= speed == QIB_IB_QDR ?
+			IBA7322_IBC_SPEED_QDR | IBA7322_IBC_IBTA_1_2_MASK :
+			((speed == QIB_IB_DDR ?
+			  IBA7322_IBC_SPEED_DDR : IBA7322_IBC_SPEED_SDR));
+
+	if (newctrlb == ppd->cpspec->ibcctrl_b)
+		return;
+
+	ppd->cpspec->ibcctrl_b = newctrlb;
+	qib_write_kreg_port(ppd, krp_ibcctrl_b, ppd->cpspec->ibcctrl_b);
+	qib_write_kreg(ppd->dd, kr_scratch, 0);
+}
+
+/*
+ * This routine is only used when we are not talking to another
+ * IB 1.2-compliant device that we think can do DDR.
+ * (This includes all existing switch chips as of Oct 2007.)
+ * 1.2-compliant devices go directly to DDR prior to reaching INIT
+ */
+static void try_7322_autoneg(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags |= QIBL_IB_AUTONEG_INPROG;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	qib_autoneg_7322_send(ppd, 0);
+	set_7322_ibspeed_fast(ppd, QIB_IB_DDR);
+	qib_7322_mini_pcs_reset(ppd);
+	/* 2 msec is minimum length of a poll cycle */
+	queue_delayed_work(ib_wq, &ppd->cpspec->autoneg_work,
+			   msecs_to_jiffies(2));
+}
+
+/*
+ * Handle the empirically determined mechanism for auto-negotiation
+ * of DDR speed with switches.
+ */
+static void autoneg_7322_work(struct work_struct *work)
+{
+	struct qib_pportdata *ppd;
+	struct qib_devdata *dd;
+	u64 startms;
+	u32 i;
+	unsigned long flags;
+
+	ppd = container_of(work, struct qib_chippport_specific,
+			    autoneg_work.work)->ppd;
+	dd = ppd->dd;
+
+	startms = jiffies_to_msecs(jiffies);
+
+	/*
+	 * Busy wait for this first part, it should be at most a
+	 * few hundred usec, since we scheduled ourselves for 2msec.
+	 */
+	for (i = 0; i < 25; i++) {
+		if (SYM_FIELD(ppd->lastibcstat, IBCStatusA_0, LinkState)
+		     == IB_7322_LT_STATE_POLLQUIET) {
+			qib_set_linkstate(ppd, QIB_IB_LINKDOWN_DISABLE);
+			break;
+		}
+		udelay(100);
+	}
+
+	if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+		goto done; /* we got there early or told to stop */
+
+	/* we expect this to timeout */
+	if (wait_event_timeout(ppd->cpspec->autoneg_wait,
+			       !(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+			       msecs_to_jiffies(90)))
+		goto done;
+	qib_7322_mini_pcs_reset(ppd);
+
+	/* we expect this to timeout */
+	if (wait_event_timeout(ppd->cpspec->autoneg_wait,
+			       !(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+			       msecs_to_jiffies(1700)))
+		goto done;
+	qib_7322_mini_pcs_reset(ppd);
+
+	set_7322_ibspeed_fast(ppd, QIB_IB_SDR);
+
+	/*
+	 * Wait up to 250 msec for link to train and get to INIT at DDR;
+	 * this should terminate early.
+	 */
+	wait_event_timeout(ppd->cpspec->autoneg_wait,
+		!(ppd->lflags & QIBL_IB_AUTONEG_INPROG),
+		msecs_to_jiffies(250));
+done:
+	if (ppd->lflags & QIBL_IB_AUTONEG_INPROG) {
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_IB_AUTONEG_INPROG;
+		if (ppd->cpspec->autoneg_tries == AUTONEG_TRIES) {
+			ppd->lflags |= QIBL_IB_AUTONEG_FAILED;
+			ppd->cpspec->autoneg_tries = 0;
+		}
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled);
+	}
+}
+
+/*
+ * This routine is used to request IPG set in the QLogic switch.
+ * Only called if r1.
+ */
+static void try_7322_ipg(struct qib_pportdata *ppd)
+{
+	struct qib_ibport *ibp = &ppd->ibport_data;
+	struct ib_mad_send_buf *send_buf;
+	struct ib_mad_agent *agent;
+	struct ib_smp *smp;
+	unsigned delay;
+	int ret;
+
+	agent = ibp->send_agent;
+	if (!agent)
+		goto retry;
+
+	send_buf = ib_create_send_mad(agent, 0, 0, 0, IB_MGMT_MAD_HDR,
+				      IB_MGMT_MAD_DATA, GFP_ATOMIC);
+	if (IS_ERR(send_buf))
+		goto retry;
+
+	if (!ibp->smi_ah) {
+		struct ib_ah *ah;
+
+		ah = qib_create_qp0_ah(ibp, be16_to_cpu(IB_LID_PERMISSIVE));
+		if (IS_ERR(ah))
+			ret = PTR_ERR(ah);
+		else {
+			send_buf->ah = ah;
+			ibp->smi_ah = to_iah(ah);
+			ret = 0;
+		}
+	} else {
+		send_buf->ah = &ibp->smi_ah->ibah;
+		ret = 0;
+	}
+
+	smp = send_buf->mad;
+	smp->base_version = IB_MGMT_BASE_VERSION;
+	smp->mgmt_class = IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE;
+	smp->class_version = 1;
+	smp->method = IB_MGMT_METHOD_SEND;
+	smp->hop_cnt = 1;
+	smp->attr_id = QIB_VENDOR_IPG;
+	smp->attr_mod = 0;
+
+	if (!ret)
+		ret = ib_post_send_mad(send_buf, NULL);
+	if (ret)
+		ib_free_send_mad(send_buf);
+retry:
+	delay = 2 << ppd->cpspec->ipg_tries;
+	queue_delayed_work(ib_wq, &ppd->cpspec->ipg_work,
+			   msecs_to_jiffies(delay));
+}
+
+/*
+ * Timeout handler for setting IPG.
+ * Only called if r1.
+ */
+static void ipg_7322_work(struct work_struct *work)
+{
+	struct qib_pportdata *ppd;
+
+	ppd = container_of(work, struct qib_chippport_specific,
+			   ipg_work.work)->ppd;
+	if ((ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED | QIBL_LINKACTIVE))
+	    && ++ppd->cpspec->ipg_tries <= 10)
+		try_7322_ipg(ppd);
+}
+
+static u32 qib_7322_iblink_state(u64 ibcs)
+{
+	u32 state = (u32)SYM_FIELD(ibcs, IBCStatusA_0, LinkState);
+
+	switch (state) {
+	case IB_7322_L_STATE_INIT:
+		state = IB_PORT_INIT;
+		break;
+	case IB_7322_L_STATE_ARM:
+		state = IB_PORT_ARMED;
+		break;
+	case IB_7322_L_STATE_ACTIVE:
+		/* fall through */
+	case IB_7322_L_STATE_ACT_DEFER:
+		state = IB_PORT_ACTIVE;
+		break;
+	default: /* fall through */
+	case IB_7322_L_STATE_DOWN:
+		state = IB_PORT_DOWN;
+		break;
+	}
+	return state;
+}
+
+/* returns the IBTA port state, rather than the IBC link training state */
+static u8 qib_7322_phys_portstate(u64 ibcs)
+{
+	u8 state = (u8)SYM_FIELD(ibcs, IBCStatusA_0, LinkTrainingState);
+	return qib_7322_physportstate[state];
+}
+
+static int qib_7322_ib_updown(struct qib_pportdata *ppd, int ibup, u64 ibcs)
+{
+	int ret = 0, symadj = 0;
+	unsigned long flags;
+	int mult;
+
+	spin_lock_irqsave(&ppd->lflags_lock, flags);
+	ppd->lflags &= ~QIBL_IB_FORCE_NOTIFY;
+	spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+	/* Update our picture of width and speed from chip */
+	if (ibcs & SYM_MASK(IBCStatusA_0, LinkSpeedQDR)) {
+		ppd->link_speed_active = QIB_IB_QDR;
+		mult = 4;
+	} else if (ibcs & SYM_MASK(IBCStatusA_0, LinkSpeedActive)) {
+		ppd->link_speed_active = QIB_IB_DDR;
+		mult = 2;
+	} else {
+		ppd->link_speed_active = QIB_IB_SDR;
+		mult = 1;
+	}
+	if (ibcs & SYM_MASK(IBCStatusA_0, LinkWidthActive)) {
+		ppd->link_width_active = IB_WIDTH_4X;
+		mult *= 4;
+	} else
+		ppd->link_width_active = IB_WIDTH_1X;
+	ppd->delay_mult = ib_rate_to_delay[mult_to_ib_rate(mult)];
+
+	if (!ibup) {
+		u64 clr;
+
+		/* Link went down. */
+		/* do IPG MAD again after linkdown, even if last time failed */
+		ppd->cpspec->ipg_tries = 0;
+		clr = qib_read_kreg_port(ppd, krp_ibcstatus_b) &
+			(SYM_MASK(IBCStatusB_0, heartbeat_timed_out) |
+			 SYM_MASK(IBCStatusB_0, heartbeat_crosstalk));
+		if (clr)
+			qib_write_kreg_port(ppd, krp_ibcstatus_b, clr);
+		if (!(ppd->lflags & (QIBL_IB_AUTONEG_FAILED |
+				     QIBL_IB_AUTONEG_INPROG)))
+			set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled);
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+			struct qib_qsfp_data *qd =
+				&ppd->cpspec->qsfp_data;
+			/* unlock the Tx settings, speed may change */
+			qib_write_kreg_port(ppd, krp_tx_deemph_override,
+				SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+				reset_tx_deemphasis_override));
+			qib_cancel_sends(ppd);
+			/* on link down, ensure sane pcs state */
+			qib_7322_mini_pcs_reset(ppd);
+			/* schedule the qsfp refresh which should turn the link
+			   off */
+			if (ppd->dd->flags & QIB_HAS_QSFP) {
+				qd->t_insert = jiffies;
+				queue_work(ib_wq, &qd->work);
+			}
+			spin_lock_irqsave(&ppd->sdma_lock, flags);
+			if (__qib_sdma_running(ppd))
+				__qib_sdma_process_event(ppd,
+					qib_sdma_event_e70_go_idle);
+			spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+		}
+		clr = read_7322_creg32_port(ppd, crp_iblinkdown);
+		if (clr == ppd->cpspec->iblnkdownsnap)
+			ppd->cpspec->iblnkdowndelta++;
+	} else {
+		if (qib_compat_ddr_negotiate &&
+		    !(ppd->lflags & (QIBL_IB_AUTONEG_FAILED |
+				     QIBL_IB_AUTONEG_INPROG)) &&
+		    ppd->link_speed_active == QIB_IB_SDR &&
+		    (ppd->link_speed_enabled & QIB_IB_DDR)
+		    && ppd->cpspec->autoneg_tries < AUTONEG_TRIES) {
+			/* we are SDR, and auto-negotiation enabled */
+			++ppd->cpspec->autoneg_tries;
+			if (!ppd->cpspec->ibdeltainprog) {
+				ppd->cpspec->ibdeltainprog = 1;
+				ppd->cpspec->ibsymdelta +=
+					read_7322_creg32_port(ppd,
+						crp_ibsymbolerr) -
+						ppd->cpspec->ibsymsnap;
+				ppd->cpspec->iblnkerrdelta +=
+					read_7322_creg32_port(ppd,
+						crp_iblinkerrrecov) -
+						ppd->cpspec->iblnkerrsnap;
+			}
+			try_7322_autoneg(ppd);
+			ret = 1; /* no other IB status change processing */
+		} else if ((ppd->lflags & QIBL_IB_AUTONEG_INPROG) &&
+			   ppd->link_speed_active == QIB_IB_SDR) {
+			qib_autoneg_7322_send(ppd, 1);
+			set_7322_ibspeed_fast(ppd, QIB_IB_DDR);
+			qib_7322_mini_pcs_reset(ppd);
+			udelay(2);
+			ret = 1; /* no other IB status change processing */
+		} else if ((ppd->lflags & QIBL_IB_AUTONEG_INPROG) &&
+			   (ppd->link_speed_active & QIB_IB_DDR)) {
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags &= ~(QIBL_IB_AUTONEG_INPROG |
+					 QIBL_IB_AUTONEG_FAILED);
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			ppd->cpspec->autoneg_tries = 0;
+			/* re-enable SDR, for next link down */
+			set_7322_ibspeed_fast(ppd, ppd->link_speed_enabled);
+			wake_up(&ppd->cpspec->autoneg_wait);
+			symadj = 1;
+		} else if (ppd->lflags & QIBL_IB_AUTONEG_FAILED) {
+			/*
+			 * Clear autoneg failure flag, and do setup
+			 * so we'll try next time link goes down and
+			 * back to INIT (possibly connected to a
+			 * different device).
+			 */
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags &= ~QIBL_IB_AUTONEG_FAILED;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			ppd->cpspec->ibcctrl_b |= IBA7322_IBC_IBTA_1_2_MASK;
+			symadj = 1;
+		}
+		if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+			symadj = 1;
+			if (ppd->dd->cspec->r1 && ppd->cpspec->ipg_tries <= 10)
+				try_7322_ipg(ppd);
+			if (!ppd->cpspec->recovery_init)
+				setup_7322_link_recovery(ppd, 0);
+			ppd->cpspec->qdr_dfe_time = jiffies +
+				msecs_to_jiffies(QDR_DFE_DISABLE_DELAY);
+		}
+		ppd->cpspec->ibmalfusesnap = 0;
+		ppd->cpspec->ibmalfsnap = read_7322_creg32_port(ppd,
+			crp_errlink);
+	}
+	if (symadj) {
+		ppd->cpspec->iblnkdownsnap =
+			read_7322_creg32_port(ppd, crp_iblinkdown);
+		if (ppd->cpspec->ibdeltainprog) {
+			ppd->cpspec->ibdeltainprog = 0;
+			ppd->cpspec->ibsymdelta += read_7322_creg32_port(ppd,
+				crp_ibsymbolerr) - ppd->cpspec->ibsymsnap;
+			ppd->cpspec->iblnkerrdelta += read_7322_creg32_port(ppd,
+				crp_iblinkerrrecov) - ppd->cpspec->iblnkerrsnap;
+		}
+	} else if (!ibup && qib_compat_ddr_negotiate &&
+		   !ppd->cpspec->ibdeltainprog &&
+			!(ppd->lflags & QIBL_IB_AUTONEG_INPROG)) {
+		ppd->cpspec->ibdeltainprog = 1;
+		ppd->cpspec->ibsymsnap = read_7322_creg32_port(ppd,
+			crp_ibsymbolerr);
+		ppd->cpspec->iblnkerrsnap = read_7322_creg32_port(ppd,
+			crp_iblinkerrrecov);
+	}
+
+	if (!ret)
+		qib_setup_7322_setextled(ppd, ibup);
+	return ret;
+}
+
+/*
+ * Does read/modify/write to appropriate registers to
+ * set output and direction bits selected by mask.
+ * these are in their canonical postions (e.g. lsb of
+ * dir will end up in D48 of extctrl on existing chips).
+ * returns contents of GP Inputs.
+ */
+static int gpio_7322_mod(struct qib_devdata *dd, u32 out, u32 dir, u32 mask)
+{
+	u64 read_val, new_out;
+	unsigned long flags;
+
+	if (mask) {
+		/* some bits being written, lock access to GPIO */
+		dir &= mask;
+		out &= mask;
+		spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+		dd->cspec->extctrl &= ~((u64)mask << SYM_LSB(EXTCtrl, GPIOOe));
+		dd->cspec->extctrl |= ((u64) dir << SYM_LSB(EXTCtrl, GPIOOe));
+		new_out = (dd->cspec->gpio_out & ~mask) | out;
+
+		qib_write_kreg(dd, kr_extctrl, dd->cspec->extctrl);
+		qib_write_kreg(dd, kr_gpio_out, new_out);
+		dd->cspec->gpio_out = new_out;
+		spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+	}
+	/*
+	 * It is unlikely that a read at this time would get valid
+	 * data on a pin whose direction line was set in the same
+	 * call to this function. We include the read here because
+	 * that allows us to potentially combine a change on one pin with
+	 * a read on another, and because the old code did something like
+	 * this.
+	 */
+	read_val = qib_read_kreg64(dd, kr_extstatus);
+	return SYM_FIELD(read_val, EXTStatus, GPIOIn);
+}
+
+/* Enable writes to config EEPROM, if possible. Returns previous state */
+static int qib_7322_eeprom_wen(struct qib_devdata *dd, int wen)
+{
+	int prev_wen;
+	u32 mask;
+
+	mask = 1 << QIB_EEPROM_WEN_NUM;
+	prev_wen = ~gpio_7322_mod(dd, 0, 0, 0) >> QIB_EEPROM_WEN_NUM;
+	gpio_7322_mod(dd, wen ? 0 : mask, mask, mask);
+
+	return prev_wen & 1;
+}
+
+/*
+ * Read fundamental info we need to use the chip.  These are
+ * the registers that describe chip capabilities, and are
+ * saved in shadow registers.
+ */
+static void get_7322_chip_params(struct qib_devdata *dd)
+{
+	u64 val;
+	u32 piobufs;
+	int mtu;
+
+	dd->palign = qib_read_kreg32(dd, kr_pagealign);
+
+	dd->uregbase = qib_read_kreg32(dd, kr_userregbase);
+
+	dd->rcvtidcnt = qib_read_kreg32(dd, kr_rcvtidcnt);
+	dd->rcvtidbase = qib_read_kreg32(dd, kr_rcvtidbase);
+	dd->rcvegrbase = qib_read_kreg32(dd, kr_rcvegrbase);
+	dd->piobufbase = qib_read_kreg64(dd, kr_sendpiobufbase);
+	dd->pio2k_bufbase = dd->piobufbase & 0xffffffff;
+
+	val = qib_read_kreg64(dd, kr_sendpiobufcnt);
+	dd->piobcnt2k = val & ~0U;
+	dd->piobcnt4k = val >> 32;
+	val = qib_read_kreg64(dd, kr_sendpiosize);
+	dd->piosize2k = val & ~0U;
+	dd->piosize4k = val >> 32;
+
+	mtu = ib_mtu_enum_to_int(qib_ibmtu);
+	if (mtu == -1)
+		mtu = QIB_DEFAULT_MTU;
+	dd->pport[0].ibmtu = (u32)mtu;
+	dd->pport[1].ibmtu = (u32)mtu;
+
+	/* these may be adjusted in init_chip_wc_pat() */
+	dd->pio2kbase = (u32 __iomem *)
+		((char __iomem *) dd->kregbase + dd->pio2k_bufbase);
+	dd->pio4kbase = (u32 __iomem *)
+		((char __iomem *) dd->kregbase +
+		 (dd->piobufbase >> 32));
+	/*
+	 * 4K buffers take 2 pages; we use roundup just to be
+	 * paranoid; we calculate it once here, rather than on
+	 * ever buf allocate
+	 */
+	dd->align4k = ALIGN(dd->piosize4k, dd->palign);
+
+	piobufs = dd->piobcnt4k + dd->piobcnt2k + NUM_VL15_BUFS;
+
+	dd->pioavregs = ALIGN(piobufs, sizeof(u64) * BITS_PER_BYTE / 2) /
+		(sizeof(u64) * BITS_PER_BYTE / 2);
+}
+
+/*
+ * The chip base addresses in cspec and cpspec have to be set
+ * after possible init_chip_wc_pat(), rather than in
+ * get_7322_chip_params(), so split out as separate function
+ */
+static void qib_7322_set_baseaddrs(struct qib_devdata *dd)
+{
+	u32 cregbase;
+	cregbase = qib_read_kreg32(dd, kr_counterregbase);
+
+	dd->cspec->cregbase = (u64 __iomem *)(cregbase +
+		(char __iomem *)dd->kregbase);
+
+	dd->egrtidbase = (u64 __iomem *)
+		((char __iomem *) dd->kregbase + dd->rcvegrbase);
+
+	/* port registers are defined as relative to base of chip */
+	dd->pport[0].cpspec->kpregbase =
+		(u64 __iomem *)((char __iomem *)dd->kregbase);
+	dd->pport[1].cpspec->kpregbase =
+		(u64 __iomem *)(dd->palign +
+		(char __iomem *)dd->kregbase);
+	dd->pport[0].cpspec->cpregbase =
+		(u64 __iomem *)(qib_read_kreg_port(&dd->pport[0],
+		kr_counterregbase) + (char __iomem *)dd->kregbase);
+	dd->pport[1].cpspec->cpregbase =
+		(u64 __iomem *)(qib_read_kreg_port(&dd->pport[1],
+		kr_counterregbase) + (char __iomem *)dd->kregbase);
+}
+
+/*
+ * This is a fairly special-purpose observer, so we only support
+ * the port-specific parts of SendCtrl
+ */
+
+#define SENDCTRL_SHADOWED (SYM_MASK(SendCtrl_0, SendEnable) |		\
+			   SYM_MASK(SendCtrl_0, SDmaEnable) |		\
+			   SYM_MASK(SendCtrl_0, SDmaIntEnable) |	\
+			   SYM_MASK(SendCtrl_0, SDmaSingleDescriptor) | \
+			   SYM_MASK(SendCtrl_0, SDmaHalt) |		\
+			   SYM_MASK(SendCtrl_0, IBVLArbiterEn) |	\
+			   SYM_MASK(SendCtrl_0, ForceCreditUpToDate))
+
+static int sendctrl_hook(struct qib_devdata *dd,
+			 const struct diag_observer *op, u32 offs,
+			 u64 *data, u64 mask, int only_32)
+{
+	unsigned long flags;
+	unsigned idx;
+	unsigned pidx;
+	struct qib_pportdata *ppd = NULL;
+	u64 local_data, all_bits;
+
+	/*
+	 * The fixed correspondence between Physical ports and pports is
+	 * severed. We need to hunt for the ppd that corresponds
+	 * to the offset we got. And we have to do that without admitting
+	 * we know the stride, apparently.
+	 */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		u64 __iomem *psptr;
+		u32 psoffs;
+
+		ppd = dd->pport + pidx;
+		if (!ppd->cpspec->kpregbase)
+			continue;
+
+		psptr = ppd->cpspec->kpregbase + krp_sendctrl;
+		psoffs = (u32) (psptr - dd->kregbase) * sizeof(*psptr);
+		if (psoffs == offs)
+			break;
+	}
+
+	/* If pport is not being managed by driver, just avoid shadows. */
+	if (pidx >= dd->num_pports)
+		ppd = NULL;
+
+	/* In any case, "idx" is flat index in kreg space */
+	idx = offs / sizeof(u64);
+
+	all_bits = ~0ULL;
+	if (only_32)
+		all_bits >>= 32;
+
+	spin_lock_irqsave(&dd->sendctrl_lock, flags);
+	if (!ppd || (mask & all_bits) != all_bits) {
+		/*
+		 * At least some mask bits are zero, so we need
+		 * to read. The judgement call is whether from
+		 * reg or shadow. First-cut: read reg, and complain
+		 * if any bits which should be shadowed are different
+		 * from their shadowed value.
+		 */
+		if (only_32)
+			local_data = (u64)qib_read_kreg32(dd, idx);
+		else
+			local_data = qib_read_kreg64(dd, idx);
+		*data = (local_data & ~mask) | (*data & mask);
+	}
+	if (mask) {
+		/*
+		 * At least some mask bits are one, so we need
+		 * to write, but only shadow some bits.
+		 */
+		u64 sval, tval; /* Shadowed, transient */
+
+		/*
+		 * New shadow val is bits we don't want to touch,
+		 * ORed with bits we do, that are intended for shadow.
+		 */
+		if (ppd) {
+			sval = ppd->p_sendctrl & ~mask;
+			sval |= *data & SENDCTRL_SHADOWED & mask;
+			ppd->p_sendctrl = sval;
+		} else
+			sval = *data & SENDCTRL_SHADOWED & mask;
+		tval = sval | (*data & ~SENDCTRL_SHADOWED & mask);
+		qib_write_kreg(dd, idx, tval);
+		qib_write_kreg(dd, kr_scratch, 0Ull);
+	}
+	spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+	return only_32 ? 4 : 8;
+}
+
+static const struct diag_observer sendctrl_0_observer = {
+	sendctrl_hook, KREG_IDX(SendCtrl_0) * sizeof(u64),
+	KREG_IDX(SendCtrl_0) * sizeof(u64)
+};
+
+static const struct diag_observer sendctrl_1_observer = {
+	sendctrl_hook, KREG_IDX(SendCtrl_1) * sizeof(u64),
+	KREG_IDX(SendCtrl_1) * sizeof(u64)
+};
+
+static ushort sdma_fetch_prio = 8;
+module_param_named(sdma_fetch_prio, sdma_fetch_prio, ushort, S_IRUGO);
+MODULE_PARM_DESC(sdma_fetch_prio, "SDMA descriptor fetch priority");
+
+/* Besides logging QSFP events, we set appropriate TxDDS values */
+static void init_txdds_table(struct qib_pportdata *ppd, int override);
+
+static void qsfp_7322_event(struct work_struct *work)
+{
+	struct qib_qsfp_data *qd;
+	struct qib_pportdata *ppd;
+	unsigned long pwrup;
+	unsigned long flags;
+	int ret;
+	u32 le2;
+
+	qd = container_of(work, struct qib_qsfp_data, work);
+	ppd = qd->ppd;
+	pwrup = qd->t_insert +
+		msecs_to_jiffies(QSFP_PWR_LAG_MSEC - QSFP_MODPRS_LAG_MSEC);
+
+	/* Delay for 20 msecs to allow ModPrs resistor to setup */
+	mdelay(QSFP_MODPRS_LAG_MSEC);
+
+	if (!qib_qsfp_mod_present(ppd)) {
+		ppd->cpspec->qsfp_data.modpresent = 0;
+		/* Set the physical link to disabled */
+		qib_set_ib_7322_lstate(ppd, 0,
+				       QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	} else {
+		/*
+		 * Some QSFP's not only do not respond until the full power-up
+		 * time, but may behave badly if we try. So hold off responding
+		 * to insertion.
+		 */
+		while (1) {
+			if (time_is_before_jiffies(pwrup))
+				break;
+			msleep(20);
+		}
+
+		ret = qib_refresh_qsfp_cache(ppd, &qd->cache);
+
+		/*
+		 * Need to change LE2 back to defaults if we couldn't
+		 * read the cable type (to handle cable swaps), so do this
+		 * even on failure to read cable information.  We don't
+		 * get here for QME, so IS_QME check not needed here.
+		 */
+		if (!ret && !ppd->dd->cspec->r1) {
+			if (QSFP_IS_ACTIVE_FAR(qd->cache.tech))
+				le2 = LE2_QME;
+			else if (qd->cache.atten[1] >= qib_long_atten &&
+				 QSFP_IS_CU(qd->cache.tech))
+				le2 = LE2_5m;
+			else
+				le2 = LE2_DEFAULT;
+		} else
+			le2 = LE2_DEFAULT;
+		ibsd_wr_allchans(ppd, 13, (le2 << 7), BMASK(9, 7));
+		/*
+		 * We always change parameteters, since we can choose
+		 * values for cables without eeproms, and the cable may have
+		 * changed from a cable with full or partial eeprom content
+		 * to one with partial or no content.
+		 */
+		init_txdds_table(ppd, 0);
+		/* The physical link is being re-enabled only when the
+		 * previous state was DISABLED and the VALID bit is not
+		 * set. This should only happen when  the cable has been
+		 * physically pulled. */
+		if (!ppd->cpspec->qsfp_data.modpresent &&
+		    (ppd->lflags & (QIBL_LINKV | QIBL_IB_LINK_DISABLED))) {
+			ppd->cpspec->qsfp_data.modpresent = 1;
+			qib_set_ib_7322_lstate(ppd, 0,
+				QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_LINKV;
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		}
+	}
+}
+
+/*
+ * There is little we can do but complain to the user if QSFP
+ * initialization fails.
+ */
+static void qib_init_7322_qsfp(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+	struct qib_qsfp_data *qd = &ppd->cpspec->qsfp_data;
+	struct qib_devdata *dd = ppd->dd;
+	u64 mod_prs_bit = QSFP_GPIO_MOD_PRS_N;
+
+	mod_prs_bit <<= (QSFP_GPIO_PORT2_SHIFT * ppd->hw_pidx);
+	qd->ppd = ppd;
+	qib_qsfp_init(qd, qsfp_7322_event);
+	spin_lock_irqsave(&dd->cspec->gpio_lock, flags);
+	dd->cspec->extctrl |= (mod_prs_bit << SYM_LSB(EXTCtrl, GPIOInvert));
+	dd->cspec->gpio_mask |= mod_prs_bit;
+	qib_write_kreg(dd, kr_extctrl, dd->cspec->extctrl);
+	qib_write_kreg(dd, kr_gpio_mask, dd->cspec->gpio_mask);
+	spin_unlock_irqrestore(&dd->cspec->gpio_lock, flags);
+}
+
+/*
+ * called at device initialization time, and also if the txselect
+ * module parameter is changed.  This is used for cables that don't
+ * have valid QSFP EEPROMs (not present, or attenuation is zero).
+ * We initialize to the default, then if there is a specific
+ * unit,port match, we use that (and set it immediately, for the
+ * current speed, if the link is at INIT or better).
+ * String format is "default# unit#,port#=# ... u,p=#", separators must
+ * be a SPACE character.  A newline terminates.  The u,p=# tuples may
+ * optionally have "u,p=#,#", where the final # is the H1 value
+ * The last specific match is used (actually, all are used, but last
+ * one is the one that winds up set); if none at all, fall back on default.
+ */
+static void set_no_qsfp_atten(struct qib_devdata *dd, int change)
+{
+	char *nxt, *str;
+	u32 pidx, unit, port, deflt, h1;
+	unsigned long val;
+	int any = 0, seth1;
+	int txdds_size;
+
+	str = txselect_list;
+
+	/* default number is validated in setup_txselect() */
+	deflt = simple_strtoul(str, &nxt, 0);
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		dd->pport[pidx].cpspec->no_eep = deflt;
+
+	txdds_size = TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ;
+	if (IS_QME(dd) || IS_QMH(dd))
+		txdds_size += TXDDS_MFG_SZ;
+
+	while (*nxt && nxt[1]) {
+		str = ++nxt;
+		unit = simple_strtoul(str, &nxt, 0);
+		if (nxt == str || !*nxt || *nxt != ',') {
+			while (*nxt && *nxt++ != ' ') /* skip to next, if any */
+				;
+			continue;
+		}
+		str = ++nxt;
+		port = simple_strtoul(str, &nxt, 0);
+		if (nxt == str || *nxt != '=') {
+			while (*nxt && *nxt++ != ' ') /* skip to next, if any */
+				;
+			continue;
+		}
+		str = ++nxt;
+		val = simple_strtoul(str, &nxt, 0);
+		if (nxt == str) {
+			while (*nxt && *nxt++ != ' ') /* skip to next, if any */
+				;
+			continue;
+		}
+		if (val >= txdds_size)
+			continue;
+		seth1 = 0;
+		h1 = 0; /* gcc thinks it might be used uninitted */
+		if (*nxt == ',' && nxt[1]) {
+			str = ++nxt;
+			h1 = (u32)simple_strtoul(str, &nxt, 0);
+			if (nxt == str)
+				while (*nxt && *nxt++ != ' ') /* skip */
+					;
+			else
+				seth1 = 1;
+		}
+		for (pidx = 0; dd->unit == unit && pidx < dd->num_pports;
+		     ++pidx) {
+			struct qib_pportdata *ppd = &dd->pport[pidx];
+
+			if (ppd->port != port || !ppd->link_speed_supported)
+				continue;
+			ppd->cpspec->no_eep = val;
+			if (seth1)
+				ppd->cpspec->h1_val = h1;
+			/* now change the IBC and serdes, overriding generic */
+			init_txdds_table(ppd, 1);
+			/* Re-enable the physical state machine on mezz boards
+			 * now that the correct settings have been set.
+			 * QSFP boards are handles by the QSFP event handler */
+			if (IS_QMH(dd) || IS_QME(dd))
+				qib_set_ib_7322_lstate(ppd, 0,
+					    QLOGIC_IB_IBCC_LINKINITCMD_SLEEP);
+			any++;
+		}
+		if (*nxt == '\n')
+			break; /* done */
+	}
+	if (change && !any) {
+		/* no specific setting, use the default.
+		 * Change the IBC and serdes, but since it's
+		 * general, don't override specific settings.
+		 */
+		for (pidx = 0; pidx < dd->num_pports; ++pidx)
+			if (dd->pport[pidx].link_speed_supported)
+				init_txdds_table(&dd->pport[pidx], 0);
+	}
+}
+
+/* handle the txselect parameter changing */
+static int setup_txselect(const char *str, struct kernel_param *kp)
+{
+	struct qib_devdata *dd;
+	unsigned long val;
+	char *n;
+	if (strlen(str) >= MAX_ATTEN_LEN) {
+		pr_info("txselect_values string too long\n");
+		return -ENOSPC;
+	}
+	val = simple_strtoul(str, &n, 0);
+	if (n == str || val >= (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
+				TXDDS_MFG_SZ)) {
+		pr_info("txselect_values must start with a number < %d\n",
+			TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ + TXDDS_MFG_SZ);
+		return -EINVAL;
+	}
+	strcpy(txselect_list, str);
+
+	list_for_each_entry(dd, &qib_dev_list, list)
+		if (dd->deviceid == PCI_DEVICE_ID_QLOGIC_IB_7322)
+			set_no_qsfp_atten(dd, 1);
+	return 0;
+}
+
+/*
+ * Write the final few registers that depend on some of the
+ * init setup.  Done late in init, just before bringing up
+ * the serdes.
+ */
+static int qib_late_7322_initreg(struct qib_devdata *dd)
+{
+	int ret = 0, n;
+	u64 val;
+
+	qib_write_kreg(dd, kr_rcvhdrentsize, dd->rcvhdrentsize);
+	qib_write_kreg(dd, kr_rcvhdrsize, dd->rcvhdrsize);
+	qib_write_kreg(dd, kr_rcvhdrcnt, dd->rcvhdrcnt);
+	qib_write_kreg(dd, kr_sendpioavailaddr, dd->pioavailregs_phys);
+	val = qib_read_kreg64(dd, kr_sendpioavailaddr);
+	if (val != dd->pioavailregs_phys) {
+		qib_dev_err(dd,
+			"Catastrophic software error, SendPIOAvailAddr written as %lx, read back as %llx\n",
+			(unsigned long) dd->pioavailregs_phys,
+			(unsigned long long) val);
+		ret = -EINVAL;
+	}
+
+	n = dd->piobcnt2k + dd->piobcnt4k + NUM_VL15_BUFS;
+	qib_7322_txchk_change(dd, 0, n, TXCHK_CHG_TYPE_KERN, NULL);
+	/* driver sends get pkey, lid, etc. checking also, to catch bugs */
+	qib_7322_txchk_change(dd, 0, n, TXCHK_CHG_TYPE_ENAB1, NULL);
+
+	qib_register_observer(dd, &sendctrl_0_observer);
+	qib_register_observer(dd, &sendctrl_1_observer);
+
+	dd->control &= ~QLOGIC_IB_C_SDMAFETCHPRIOEN;
+	qib_write_kreg(dd, kr_control, dd->control);
+	/*
+	 * Set SendDmaFetchPriority and init Tx params, including
+	 * QSFP handler on boards that have QSFP.
+	 * First set our default attenuation entry for cables that
+	 * don't have valid attenuation.
+	 */
+	set_no_qsfp_atten(dd, 0);
+	for (n = 0; n < dd->num_pports; ++n) {
+		struct qib_pportdata *ppd = dd->pport + n;
+
+		qib_write_kreg_port(ppd, krp_senddmaprioritythld,
+				    sdma_fetch_prio & 0xf);
+		/* Initialize qsfp if present on board. */
+		if (dd->flags & QIB_HAS_QSFP)
+			qib_init_7322_qsfp(ppd);
+	}
+	dd->control |= QLOGIC_IB_C_SDMAFETCHPRIOEN;
+	qib_write_kreg(dd, kr_control, dd->control);
+
+	return ret;
+}
+
+/* per IB port errors.  */
+#define SENDCTRL_PIBP (MASK_ACROSS(0, 1) | MASK_ACROSS(3, 3) | \
+	MASK_ACROSS(8, 15))
+#define RCVCTRL_PIBP (MASK_ACROSS(0, 17) | MASK_ACROSS(39, 41))
+#define ERRS_PIBP (MASK_ACROSS(57, 58) | MASK_ACROSS(54, 54) | \
+	MASK_ACROSS(36, 49) | MASK_ACROSS(29, 34) | MASK_ACROSS(14, 17) | \
+	MASK_ACROSS(0, 11))
+
+/*
+ * Write the initialization per-port registers that need to be done at
+ * driver load and after reset completes (i.e., that aren't done as part
+ * of other init procedures called from qib_init.c).
+ * Some of these should be redundant on reset, but play safe.
+ */
+static void write_7322_init_portregs(struct qib_pportdata *ppd)
+{
+	u64 val;
+	int i;
+
+	if (!ppd->link_speed_supported) {
+		/* no buffer credits for this port */
+		for (i = 1; i < 8; i++)
+			qib_write_kreg_port(ppd, krp_rxcreditvl0 + i, 0);
+		qib_write_kreg_port(ppd, krp_ibcctrl_b, 0);
+		qib_write_kreg(ppd->dd, kr_scratch, 0);
+		return;
+	}
+
+	/*
+	 * Set the number of supported virtual lanes in IBC,
+	 * for flow control packet handling on unsupported VLs
+	 */
+	val = qib_read_kreg_port(ppd, krp_ibsdtestiftx);
+	val &= ~SYM_MASK(IB_SDTEST_IF_TX_0, VL_CAP);
+	val |= (u64)(ppd->vls_supported - 1) <<
+		SYM_LSB(IB_SDTEST_IF_TX_0, VL_CAP);
+	qib_write_kreg_port(ppd, krp_ibsdtestiftx, val);
+
+	qib_write_kreg_port(ppd, krp_rcvbthqp, QIB_KD_QP);
+
+	/* enable tx header checking */
+	qib_write_kreg_port(ppd, krp_sendcheckcontrol, IBA7322_SENDCHK_PKEY |
+			    IBA7322_SENDCHK_BTHQP | IBA7322_SENDCHK_SLID |
+			    IBA7322_SENDCHK_RAW_IPV6 | IBA7322_SENDCHK_MINSZ);
+
+	qib_write_kreg_port(ppd, krp_ncmodectrl,
+		SYM_MASK(IBNCModeCtrl_0, ScrambleCapLocal));
+
+	/*
+	 * Unconditionally clear the bufmask bits.  If SDMA is
+	 * enabled, we'll set them appropriately later.
+	 */
+	qib_write_kreg_port(ppd, krp_senddmabufmask0, 0);
+	qib_write_kreg_port(ppd, krp_senddmabufmask1, 0);
+	qib_write_kreg_port(ppd, krp_senddmabufmask2, 0);
+	if (ppd->dd->cspec->r1)
+		ppd->p_sendctrl |= SYM_MASK(SendCtrl_0, ForceCreditUpToDate);
+}
+
+/*
+ * Write the initialization per-device registers that need to be done at
+ * driver load and after reset completes (i.e., that aren't done as part
+ * of other init procedures called from qib_init.c).  Also write per-port
+ * registers that are affected by overall device config, such as QP mapping
+ * Some of these should be redundant on reset, but play safe.
+ */
+static void write_7322_initregs(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int i, pidx;
+	u64 val;
+
+	/* Set Multicast QPs received by port 2 to map to context one. */
+	qib_write_kreg(dd, KREG_IDX(RcvQPMulticastContext_1), 1);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		unsigned n, regno;
+		unsigned long flags;
+
+		if (dd->n_krcv_queues < 2 ||
+			!dd->pport[pidx].link_speed_supported)
+			continue;
+
+		ppd = &dd->pport[pidx];
+
+		/* be paranoid against later code motion, etc. */
+		spin_lock_irqsave(&dd->cspec->rcvmod_lock, flags);
+		ppd->p_rcvctrl |= SYM_MASK(RcvCtrl_0, RcvQPMapEnable);
+		spin_unlock_irqrestore(&dd->cspec->rcvmod_lock, flags);
+
+		/* Initialize QP to context mapping */
+		regno = krp_rcvqpmaptable;
+		val = 0;
+		if (dd->num_pports > 1)
+			n = dd->first_user_ctxt / dd->num_pports;
+		else
+			n = dd->first_user_ctxt - 1;
+		for (i = 0; i < 32; ) {
+			unsigned ctxt;
+
+			if (dd->num_pports > 1)
+				ctxt = (i % n) * dd->num_pports + pidx;
+			else if (i % n)
+				ctxt = (i % n) + 1;
+			else
+				ctxt = ppd->hw_pidx;
+			val |= ctxt << (5 * (i % 6));
+			i++;
+			if (i % 6 == 0) {
+				qib_write_kreg_port(ppd, regno, val);
+				val = 0;
+				regno++;
+			}
+		}
+		qib_write_kreg_port(ppd, regno, val);
+	}
+
+	/*
+	 * Setup up interrupt mitigation for kernel contexts, but
+	 * not user contexts (user contexts use interrupts when
+	 * stalled waiting for any packet, so want those interrupts
+	 * right away).
+	 */
+	for (i = 0; i < dd->first_user_ctxt; i++) {
+		dd->cspec->rcvavail_timeout[i] = rcv_int_timeout;
+		qib_write_kreg(dd, kr_rcvavailtimeout + i, rcv_int_timeout);
+	}
+
+	/*
+	 * Initialize  as (disabled) rcvflow tables.  Application code
+	 * will setup each flow as it uses the flow.
+	 * Doesn't clear any of the error bits that might be set.
+	 */
+	val = TIDFLOW_ERRBITS; /* these are W1C */
+	for (i = 0; i < dd->cfgctxts; i++) {
+		int flow;
+		for (flow = 0; flow < NUM_TIDFLOWS_CTXT; flow++)
+			qib_write_ureg(dd, ur_rcvflowtable+flow, val, i);
+	}
+
+	/*
+	 * dual cards init to dual port recovery, single port cards to
+	 * the one port.  Dual port cards may later adjust to 1 port,
+	 * and then back to dual port if both ports are connected
+	 * */
+	if (dd->num_pports)
+		setup_7322_link_recovery(dd->pport, dd->num_pports > 1);
+}
+
+static int qib_init_7322_variables(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	unsigned features, pidx, sbufcnt;
+	int ret, mtu;
+	u32 sbufs, updthresh;
+
+	/* pport structs are contiguous, allocated after devdata */
+	ppd = (struct qib_pportdata *)(dd + 1);
+	dd->pport = ppd;
+	ppd[0].dd = dd;
+	ppd[1].dd = dd;
+
+	dd->cspec = (struct qib_chip_specific *)(ppd + 2);
+
+	ppd[0].cpspec = (struct qib_chippport_specific *)(dd->cspec + 1);
+	ppd[1].cpspec = &ppd[0].cpspec[1];
+	ppd[0].cpspec->ppd = &ppd[0]; /* for autoneg_7322_work() */
+	ppd[1].cpspec->ppd = &ppd[1]; /* for autoneg_7322_work() */
+
+	spin_lock_init(&dd->cspec->rcvmod_lock);
+	spin_lock_init(&dd->cspec->gpio_lock);
+
+	/* we haven't yet set QIB_PRESENT, so use read directly */
+	dd->revision = readq(&dd->kregbase[kr_revision]);
+
+	if ((dd->revision & 0xffffffffU) == 0xffffffffU) {
+		qib_dev_err(dd,
+			"Revision register read failure, giving up initialization\n");
+		ret = -ENODEV;
+		goto bail;
+	}
+	dd->flags |= QIB_PRESENT;  /* now register routines work */
+
+	dd->majrev = (u8) SYM_FIELD(dd->revision, Revision_R, ChipRevMajor);
+	dd->minrev = (u8) SYM_FIELD(dd->revision, Revision_R, ChipRevMinor);
+	dd->cspec->r1 = dd->minrev == 1;
+
+	get_7322_chip_params(dd);
+	features = qib_7322_boardname(dd);
+
+	/* now that piobcnt2k and 4k set, we can allocate these */
+	sbufcnt = dd->piobcnt2k + dd->piobcnt4k +
+		NUM_VL15_BUFS + BITS_PER_LONG - 1;
+	sbufcnt /= BITS_PER_LONG;
+	dd->cspec->sendchkenable = kmalloc(sbufcnt *
+		sizeof(*dd->cspec->sendchkenable), GFP_KERNEL);
+	dd->cspec->sendgrhchk = kmalloc(sbufcnt *
+		sizeof(*dd->cspec->sendgrhchk), GFP_KERNEL);
+	dd->cspec->sendibchk = kmalloc(sbufcnt *
+		sizeof(*dd->cspec->sendibchk), GFP_KERNEL);
+	if (!dd->cspec->sendchkenable || !dd->cspec->sendgrhchk ||
+		!dd->cspec->sendibchk) {
+		qib_dev_err(dd, "Failed allocation for hdrchk bitmaps\n");
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	ppd = dd->pport;
+
+	/*
+	 * GPIO bits for TWSI data and clock,
+	 * used for serial EEPROM.
+	 */
+	dd->gpio_sda_num = _QIB_GPIO_SDA_NUM;
+	dd->gpio_scl_num = _QIB_GPIO_SCL_NUM;
+	dd->twsi_eeprom_dev = QIB_TWSI_EEPROM_DEV;
+
+	dd->flags |= QIB_HAS_INTX | QIB_HAS_LINK_LATENCY |
+		QIB_NODMA_RTAIL | QIB_HAS_VLSUPP | QIB_HAS_HDRSUPP |
+		QIB_HAS_THRESH_UPDATE |
+		(sdma_idle_cnt ? QIB_HAS_SDMA_TIMEOUT : 0);
+	dd->flags |= qib_special_trigger ?
+		QIB_USE_SPCL_TRIG : QIB_HAS_SEND_DMA;
+
+	/*
+	 * Setup initial values.  These may change when PAT is enabled, but
+	 * we need these to do initial chip register accesses.
+	 */
+	qib_7322_set_baseaddrs(dd);
+
+	mtu = ib_mtu_enum_to_int(qib_ibmtu);
+	if (mtu == -1)
+		mtu = QIB_DEFAULT_MTU;
+
+	dd->cspec->int_enable_mask = QIB_I_BITSEXTANT;
+	/* all hwerrors become interrupts, unless special purposed */
+	dd->cspec->hwerrmask = ~0ULL;
+	/*  link_recovery setup causes these errors, so ignore them,
+	 *  other than clearing them when they occur */
+	dd->cspec->hwerrmask &=
+		~(SYM_MASK(HwErrMask, IBSerdesPClkNotDetectMask_0) |
+		  SYM_MASK(HwErrMask, IBSerdesPClkNotDetectMask_1) |
+		  HWE_MASK(LATriggered));
+
+	for (pidx = 0; pidx < NUM_IB_PORTS; ++pidx) {
+		struct qib_chippport_specific *cp = ppd->cpspec;
+		ppd->link_speed_supported = features & PORT_SPD_CAP;
+		features >>=  PORT_SPD_CAP_SHIFT;
+		if (!ppd->link_speed_supported) {
+			/* single port mode (7340, or configured) */
+			dd->skip_kctxt_mask |= 1 << pidx;
+			if (pidx == 0) {
+				/* Make sure port is disabled. */
+				qib_write_kreg_port(ppd, krp_rcvctrl, 0);
+				qib_write_kreg_port(ppd, krp_ibcctrl_a, 0);
+				ppd[0] = ppd[1];
+				dd->cspec->hwerrmask &= ~(SYM_MASK(HwErrMask,
+						  IBSerdesPClkNotDetectMask_0)
+						  | SYM_MASK(HwErrMask,
+						  SDmaMemReadErrMask_0));
+				dd->cspec->int_enable_mask &= ~(
+				     SYM_MASK(IntMask, SDmaCleanupDoneMask_0) |
+				     SYM_MASK(IntMask, SDmaIdleIntMask_0) |
+				     SYM_MASK(IntMask, SDmaProgressIntMask_0) |
+				     SYM_MASK(IntMask, SDmaIntMask_0) |
+				     SYM_MASK(IntMask, ErrIntMask_0) |
+				     SYM_MASK(IntMask, SendDoneIntMask_0));
+			} else {
+				/* Make sure port is disabled. */
+				qib_write_kreg_port(ppd, krp_rcvctrl, 0);
+				qib_write_kreg_port(ppd, krp_ibcctrl_a, 0);
+				dd->cspec->hwerrmask &= ~(SYM_MASK(HwErrMask,
+						  IBSerdesPClkNotDetectMask_1)
+						  | SYM_MASK(HwErrMask,
+						  SDmaMemReadErrMask_1));
+				dd->cspec->int_enable_mask &= ~(
+				     SYM_MASK(IntMask, SDmaCleanupDoneMask_1) |
+				     SYM_MASK(IntMask, SDmaIdleIntMask_1) |
+				     SYM_MASK(IntMask, SDmaProgressIntMask_1) |
+				     SYM_MASK(IntMask, SDmaIntMask_1) |
+				     SYM_MASK(IntMask, ErrIntMask_1) |
+				     SYM_MASK(IntMask, SendDoneIntMask_1));
+			}
+			continue;
+		}
+
+		dd->num_pports++;
+		ret = qib_init_pportdata(ppd, dd, pidx, dd->num_pports);
+		if (ret) {
+			dd->num_pports--;
+			goto bail;
+		}
+
+		ppd->link_width_supported = IB_WIDTH_1X | IB_WIDTH_4X;
+		ppd->link_width_enabled = IB_WIDTH_4X;
+		ppd->link_speed_enabled = ppd->link_speed_supported;
+		/*
+		 * Set the initial values to reasonable default, will be set
+		 * for real when link is up.
+		 */
+		ppd->link_width_active = IB_WIDTH_4X;
+		ppd->link_speed_active = QIB_IB_SDR;
+		ppd->delay_mult = ib_rate_to_delay[IB_RATE_10_GBPS];
+		switch (qib_num_cfg_vls) {
+		case 1:
+			ppd->vls_supported = IB_VL_VL0;
+			break;
+		case 2:
+			ppd->vls_supported = IB_VL_VL0_1;
+			break;
+		default:
+			qib_devinfo(dd->pcidev,
+				    "Invalid num_vls %u, using 4 VLs\n",
+				    qib_num_cfg_vls);
+			qib_num_cfg_vls = 4;
+			/* fall through */
+		case 4:
+			ppd->vls_supported = IB_VL_VL0_3;
+			break;
+		case 8:
+			if (mtu <= 2048)
+				ppd->vls_supported = IB_VL_VL0_7;
+			else {
+				qib_devinfo(dd->pcidev,
+					    "Invalid num_vls %u for MTU %d "
+					    ", using 4 VLs\n",
+					    qib_num_cfg_vls, mtu);
+				ppd->vls_supported = IB_VL_VL0_3;
+				qib_num_cfg_vls = 4;
+			}
+			break;
+		}
+		ppd->vls_operational = ppd->vls_supported;
+
+		init_waitqueue_head(&cp->autoneg_wait);
+		INIT_DELAYED_WORK(&cp->autoneg_work,
+				  autoneg_7322_work);
+		if (ppd->dd->cspec->r1)
+			INIT_DELAYED_WORK(&cp->ipg_work, ipg_7322_work);
+
+		/*
+		 * For Mez and similar cards, no qsfp info, so do
+		 * the "cable info" setup here.  Can be overridden
+		 * in adapter-specific routines.
+		 */
+		if (!(dd->flags & QIB_HAS_QSFP)) {
+			if (!IS_QMH(dd) && !IS_QME(dd))
+				qib_devinfo(dd->pcidev,
+					"IB%u:%u: Unknown mezzanine card type\n",
+					dd->unit, ppd->port);
+			cp->h1_val = IS_QMH(dd) ? H1_FORCE_QMH : H1_FORCE_QME;
+			/*
+			 * Choose center value as default tx serdes setting
+			 * until changed through module parameter.
+			 */
+			ppd->cpspec->no_eep = IS_QMH(dd) ?
+				TXDDS_TABLE_SZ + 2 : TXDDS_TABLE_SZ + 4;
+		} else
+			cp->h1_val = H1_FORCE_VAL;
+
+		/* Avoid writes to chip for mini_init */
+		if (!qib_mini_init)
+			write_7322_init_portregs(ppd);
+
+		init_timer(&cp->chase_timer);
+		cp->chase_timer.function = reenable_chase;
+		cp->chase_timer.data = (unsigned long)ppd;
+
+		ppd++;
+	}
+
+	dd->rcvhdrentsize = qib_rcvhdrentsize ?
+		qib_rcvhdrentsize : QIB_RCVHDR_ENTSIZE;
+	dd->rcvhdrsize = qib_rcvhdrsize ?
+		qib_rcvhdrsize : QIB_DFLT_RCVHDRSIZE;
+	dd->rhf_offset = dd->rcvhdrentsize - sizeof(u64) / sizeof(u32);
+
+	/* we always allocate at least 2048 bytes for eager buffers */
+	dd->rcvegrbufsize = max(mtu, 2048);
+	BUG_ON(!is_power_of_2(dd->rcvegrbufsize));
+	dd->rcvegrbufsize_shift = ilog2(dd->rcvegrbufsize);
+
+	qib_7322_tidtemplate(dd);
+
+	/*
+	 * We can request a receive interrupt for 1 or
+	 * more packets from current offset.
+	 */
+	dd->rhdrhead_intr_off =
+		(u64) rcv_int_count << IBA7322_HDRHEAD_PKTINT_SHIFT;
+
+	/* setup the stats timer; the add_timer is done at end of init */
+	init_timer(&dd->stats_timer);
+	dd->stats_timer.function = qib_get_7322_faststats;
+	dd->stats_timer.data = (unsigned long) dd;
+
+	dd->ureg_align = 0x10000;  /* 64KB alignment */
+
+	dd->piosize2kmax_dwords = dd->piosize2k >> 2;
+
+	qib_7322_config_ctxts(dd);
+	qib_set_ctxtcnt(dd);
+
+	if (qib_wc_pat) {
+		resource_size_t vl15off;
+		/*
+		 * We do not set WC on the VL15 buffers to avoid
+		 * a rare problem with unaligned writes from
+		 * interrupt-flushed store buffers, so we need
+		 * to map those separately here.  We can't solve
+		 * this for the rarely used mtrr case.
+		 */
+		ret = init_chip_wc_pat(dd, 0);
+		if (ret)
+			goto bail;
+
+		/* vl15 buffers start just after the 4k buffers */
+		vl15off = dd->physaddr + (dd->piobufbase >> 32) +
+			dd->piobcnt4k * dd->align4k;
+		dd->piovl15base	= ioremap_nocache(vl15off,
+						  NUM_VL15_BUFS * dd->align4k);
+		if (!dd->piovl15base) {
+			ret = -ENOMEM;
+			goto bail;
+		}
+	}
+	qib_7322_set_baseaddrs(dd); /* set chip access pointers now */
+
+	ret = 0;
+	if (qib_mini_init)
+		goto bail;
+	if (!dd->num_pports) {
+		qib_dev_err(dd, "No ports enabled, giving up initialization\n");
+		goto bail; /* no error, so can still figure out why err */
+	}
+
+	write_7322_initregs(dd);
+	ret = qib_create_ctxts(dd);
+	init_7322_cntrnames(dd);
+
+	updthresh = 8U; /* update threshold */
+
+	/* use all of 4KB buffers for the kernel SDMA, zero if !SDMA.
+	 * reserve the update threshold amount for other kernel use, such
+	 * as sending SMI, MAD, and ACKs, or 3, whichever is greater,
+	 * unless we aren't enabling SDMA, in which case we want to use
+	 * all the 4k bufs for the kernel.
+	 * if this was less than the update threshold, we could wait
+	 * a long time for an update.  Coded this way because we
+	 * sometimes change the update threshold for various reasons,
+	 * and we want this to remain robust.
+	 */
+	if (dd->flags & QIB_HAS_SEND_DMA) {
+		dd->cspec->sdmabufcnt = dd->piobcnt4k;
+		sbufs = updthresh > 3 ? updthresh : 3;
+	} else {
+		dd->cspec->sdmabufcnt = 0;
+		sbufs = dd->piobcnt4k;
+	}
+	dd->cspec->lastbuf_for_pio = dd->piobcnt2k + dd->piobcnt4k -
+		dd->cspec->sdmabufcnt;
+	dd->lastctxt_piobuf = dd->cspec->lastbuf_for_pio - sbufs;
+	dd->cspec->lastbuf_for_pio--; /* range is <= , not < */
+	dd->last_pio = dd->cspec->lastbuf_for_pio;
+	dd->pbufsctxt = (dd->cfgctxts > dd->first_user_ctxt) ?
+		dd->lastctxt_piobuf / (dd->cfgctxts - dd->first_user_ctxt) : 0;
+
+	/*
+	 * If we have 16 user contexts, we will have 7 sbufs
+	 * per context, so reduce the update threshold to match.  We
+	 * want to update before we actually run out, at low pbufs/ctxt
+	 * so give ourselves some margin.
+	 */
+	if (dd->pbufsctxt >= 2 && dd->pbufsctxt - 2 < updthresh)
+		updthresh = dd->pbufsctxt - 2;
+	dd->cspec->updthresh_dflt = updthresh;
+	dd->cspec->updthresh = updthresh;
+
+	/* before full enable, no interrupts, no locking needed */
+	dd->sendctrl |= ((updthresh & SYM_RMASK(SendCtrl, AvailUpdThld))
+			     << SYM_LSB(SendCtrl, AvailUpdThld)) |
+			SYM_MASK(SendCtrl, SendBufAvailPad64Byte);
+
+	dd->psxmitwait_supported = 1;
+	dd->psxmitwait_check_rate = QIB_7322_PSXMITWAIT_CHECK_RATE;
+bail:
+	if (!dd->ctxtcnt)
+		dd->ctxtcnt = 1; /* for other initialization code */
+
+	return ret;
+}
+
+static u32 __iomem *qib_7322_getsendbuf(struct qib_pportdata *ppd, u64 pbc,
+					u32 *pbufnum)
+{
+	u32 first, last, plen = pbc & QIB_PBC_LENGTH_MASK;
+	struct qib_devdata *dd = ppd->dd;
+
+	/* last is same for 2k and 4k, because we use 4k if all 2k busy */
+	if (pbc & PBC_7322_VL15_SEND) {
+		first = dd->piobcnt2k + dd->piobcnt4k + ppd->hw_pidx;
+		last = first;
+	} else {
+		if ((plen + 1) > dd->piosize2kmax_dwords)
+			first = dd->piobcnt2k;
+		else
+			first = 0;
+		last = dd->cspec->lastbuf_for_pio;
+	}
+	return qib_getsendbuf_range(dd, pbufnum, first, last);
+}
+
+static void qib_set_cntr_7322_sample(struct qib_pportdata *ppd, u32 intv,
+				     u32 start)
+{
+	qib_write_kreg_port(ppd, krp_psinterval, intv);
+	qib_write_kreg_port(ppd, krp_psstart, start);
+}
+
+/*
+ * Must be called with sdma_lock held, or before init finished.
+ */
+static void qib_sdma_set_7322_desc_cnt(struct qib_pportdata *ppd, unsigned cnt)
+{
+	qib_write_kreg_port(ppd, krp_senddmadesccnt, cnt);
+}
+
+/*
+ * sdma_lock should be acquired before calling this routine
+ */
+static void dump_sdma_7322_state(struct qib_pportdata *ppd)
+{
+	u64 reg, reg1, reg2;
+
+	reg = qib_read_kreg_port(ppd, krp_senddmastatus);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmastatus: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_sendctrl);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA sendctrl: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmabase);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmabase: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmabufmask0);
+	reg1 = qib_read_kreg_port(ppd, krp_senddmabufmask1);
+	reg2 = qib_read_kreg_port(ppd, krp_senddmabufmask2);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmabufmask 0:%llx  1:%llx  2:%llx\n",
+		 reg, reg1, reg2);
+
+	/* get bufuse bits, clear them, and print them again if non-zero */
+	reg = qib_read_kreg_port(ppd, krp_senddmabuf_use0);
+	qib_write_kreg_port(ppd, krp_senddmabuf_use0, reg);
+	reg1 = qib_read_kreg_port(ppd, krp_senddmabuf_use1);
+	qib_write_kreg_port(ppd, krp_senddmabuf_use0, reg1);
+	reg2 = qib_read_kreg_port(ppd, krp_senddmabuf_use2);
+	qib_write_kreg_port(ppd, krp_senddmabuf_use0, reg2);
+	/* 0 and 1 should always be zero, so print as short form */
+	qib_dev_porterr(ppd->dd, ppd->port,
+		 "SDMA current senddmabuf_use 0:%llx  1:%llx  2:%llx\n",
+		 reg, reg1, reg2);
+	reg = qib_read_kreg_port(ppd, krp_senddmabuf_use0);
+	reg1 = qib_read_kreg_port(ppd, krp_senddmabuf_use1);
+	reg2 = qib_read_kreg_port(ppd, krp_senddmabuf_use2);
+	/* 0 and 1 should always be zero, so print as short form */
+	qib_dev_porterr(ppd->dd, ppd->port,
+		 "SDMA cleared senddmabuf_use 0:%llx  1:%llx  2:%llx\n",
+		 reg, reg1, reg2);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmatail);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmatail: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmahead);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmahead: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmaheadaddr);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmaheadaddr: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmalengen);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmalengen: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmadesccnt);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmadesccnt: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmaidlecnt);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmaidlecnt: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmaprioritythld);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmapriorityhld: 0x%016llx\n", reg);
+
+	reg = qib_read_kreg_port(ppd, krp_senddmareloadcnt);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA senddmareloadcnt: 0x%016llx\n", reg);
+
+	dump_sdma_state(ppd);
+}
+
+static struct sdma_set_state_action sdma_7322_action_table[] = {
+	[qib_sdma_state_s00_hw_down] = {
+		.go_s99_running_tofalse = 1,
+		.op_enable = 0,
+		.op_intenable = 0,
+		.op_halt = 0,
+		.op_drain = 0,
+	},
+	[qib_sdma_state_s10_hw_start_up_wait] = {
+		.op_enable = 0,
+		.op_intenable = 1,
+		.op_halt = 1,
+		.op_drain = 0,
+	},
+	[qib_sdma_state_s20_idle] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+		.op_drain = 0,
+	},
+	[qib_sdma_state_s30_sw_clean_up_wait] = {
+		.op_enable = 0,
+		.op_intenable = 1,
+		.op_halt = 1,
+		.op_drain = 0,
+	},
+	[qib_sdma_state_s40_hw_clean_up_wait] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+		.op_drain = 0,
+	},
+	[qib_sdma_state_s50_hw_halt_wait] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 1,
+		.op_drain = 1,
+	},
+	[qib_sdma_state_s99_running] = {
+		.op_enable = 1,
+		.op_intenable = 1,
+		.op_halt = 0,
+		.op_drain = 0,
+		.go_s99_running_totrue = 1,
+	},
+};
+
+static void qib_7322_sdma_init_early(struct qib_pportdata *ppd)
+{
+	ppd->sdma_state.set_state_action = sdma_7322_action_table;
+}
+
+static int init_sdma_7322_regs(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned lastbuf, erstbuf;
+	u64 senddmabufmask[3] = { 0 };
+	int n, ret = 0;
+
+	qib_write_kreg_port(ppd, krp_senddmabase, ppd->sdma_descq_phys);
+	qib_sdma_7322_setlengen(ppd);
+	qib_sdma_update_7322_tail(ppd, 0); /* Set SendDmaTail */
+	qib_write_kreg_port(ppd, krp_senddmareloadcnt, sdma_idle_cnt);
+	qib_write_kreg_port(ppd, krp_senddmadesccnt, 0);
+	qib_write_kreg_port(ppd, krp_senddmaheadaddr, ppd->sdma_head_phys);
+
+	if (dd->num_pports)
+		n = dd->cspec->sdmabufcnt / dd->num_pports; /* no remainder */
+	else
+		n = dd->cspec->sdmabufcnt; /* failsafe for init */
+	erstbuf = (dd->piobcnt2k + dd->piobcnt4k) -
+		((dd->num_pports == 1 || ppd->port == 2) ? n :
+		dd->cspec->sdmabufcnt);
+	lastbuf = erstbuf + n;
+
+	ppd->sdma_state.first_sendbuf = erstbuf;
+	ppd->sdma_state.last_sendbuf = lastbuf;
+	for (; erstbuf < lastbuf; ++erstbuf) {
+		unsigned word = erstbuf / BITS_PER_LONG;
+		unsigned bit = erstbuf & (BITS_PER_LONG - 1);
+
+		BUG_ON(word >= 3);
+		senddmabufmask[word] |= 1ULL << bit;
+	}
+	qib_write_kreg_port(ppd, krp_senddmabufmask0, senddmabufmask[0]);
+	qib_write_kreg_port(ppd, krp_senddmabufmask1, senddmabufmask[1]);
+	qib_write_kreg_port(ppd, krp_senddmabufmask2, senddmabufmask[2]);
+	return ret;
+}
+
+/* sdma_lock must be held */
+static u16 qib_sdma_7322_gethead(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int sane;
+	int use_dmahead;
+	u16 swhead;
+	u16 swtail;
+	u16 cnt;
+	u16 hwhead;
+
+	use_dmahead = __qib_sdma_running(ppd) &&
+		(dd->flags & QIB_HAS_SDMA_TIMEOUT);
+retry:
+	hwhead = use_dmahead ?
+		(u16) le64_to_cpu(*ppd->sdma_head_dma) :
+		(u16) qib_read_kreg_port(ppd, krp_senddmahead);
+
+	swhead = ppd->sdma_descq_head;
+	swtail = ppd->sdma_descq_tail;
+	cnt = ppd->sdma_descq_cnt;
+
+	if (swhead < swtail)
+		/* not wrapped */
+		sane = (hwhead >= swhead) & (hwhead <= swtail);
+	else if (swhead > swtail)
+		/* wrapped around */
+		sane = ((hwhead >= swhead) && (hwhead < cnt)) ||
+			(hwhead <= swtail);
+	else
+		/* empty */
+		sane = (hwhead == swhead);
+
+	if (unlikely(!sane)) {
+		if (use_dmahead) {
+			/* try one more time, directly from the register */
+			use_dmahead = 0;
+			goto retry;
+		}
+		/* proceed as if no progress */
+		hwhead = swhead;
+	}
+
+	return hwhead;
+}
+
+static int qib_sdma_7322_busy(struct qib_pportdata *ppd)
+{
+	u64 hwstatus = qib_read_kreg_port(ppd, krp_senddmastatus);
+
+	return (hwstatus & SYM_MASK(SendDmaStatus_0, ScoreBoardDrainInProg)) ||
+	       (hwstatus & SYM_MASK(SendDmaStatus_0, HaltInProg)) ||
+	       !(hwstatus & SYM_MASK(SendDmaStatus_0, InternalSDmaHalt)) ||
+	       !(hwstatus & SYM_MASK(SendDmaStatus_0, ScbEmpty));
+}
+
+/*
+ * Compute the amount of delay before sending the next packet if the
+ * port's send rate differs from the static rate set for the QP.
+ * The delay affects the next packet and the amount of the delay is
+ * based on the length of the this packet.
+ */
+static u32 qib_7322_setpbc_control(struct qib_pportdata *ppd, u32 plen,
+				   u8 srate, u8 vl)
+{
+	u8 snd_mult = ppd->delay_mult;
+	u8 rcv_mult = ib_rate_to_delay[srate];
+	u32 ret;
+
+	ret = rcv_mult > snd_mult ? ((plen + 1) >> 1) * snd_mult : 0;
+
+	/* Indicate VL15, else set the VL in the control word */
+	if (vl == 15)
+		ret |= PBC_7322_VL15_SEND_CTRL;
+	else
+		ret |= vl << PBC_VL_NUM_LSB;
+	ret |= ((u32)(ppd->hw_pidx)) << PBC_PORT_SEL_LSB;
+
+	return ret;
+}
+
+/*
+ * Enable the per-port VL15 send buffers for use.
+ * They follow the rest of the buffers, without a config parameter.
+ * This was in initregs, but that is done before the shadow
+ * is set up, and this has to be done after the shadow is
+ * set up.
+ */
+static void qib_7322_initvl15_bufs(struct qib_devdata *dd)
+{
+	unsigned vl15bufs;
+
+	vl15bufs = dd->piobcnt2k + dd->piobcnt4k;
+	qib_chg_pioavailkernel(dd, vl15bufs, NUM_VL15_BUFS,
+			       TXCHK_CHG_TYPE_KERN, NULL);
+}
+
+static void qib_7322_init_ctxt(struct qib_ctxtdata *rcd)
+{
+	if (rcd->ctxt < NUM_IB_PORTS) {
+		if (rcd->dd->num_pports > 1) {
+			rcd->rcvegrcnt = KCTXT0_EGRCNT / 2;
+			rcd->rcvegr_tid_base = rcd->ctxt ? rcd->rcvegrcnt : 0;
+		} else {
+			rcd->rcvegrcnt = KCTXT0_EGRCNT;
+			rcd->rcvegr_tid_base = 0;
+		}
+	} else {
+		rcd->rcvegrcnt = rcd->dd->cspec->rcvegrcnt;
+		rcd->rcvegr_tid_base = KCTXT0_EGRCNT +
+			(rcd->ctxt - NUM_IB_PORTS) * rcd->rcvegrcnt;
+	}
+}
+
+#define QTXSLEEPS 5000
+static void qib_7322_txchk_change(struct qib_devdata *dd, u32 start,
+				  u32 len, u32 which, struct qib_ctxtdata *rcd)
+{
+	int i;
+	const int last = start + len - 1;
+	const int lastr = last / BITS_PER_LONG;
+	u32 sleeps = 0;
+	int wait = rcd != NULL;
+	unsigned long flags;
+
+	while (wait) {
+		unsigned long shadow;
+		int cstart, previ = -1;
+
+		/*
+		 * when flipping from kernel to user, we can't change
+		 * the checking type if the buffer is allocated to the
+		 * driver.   It's OK the other direction, because it's
+		 * from close, and we have just disarm'ed all the
+		 * buffers.  All the kernel to kernel changes are also
+		 * OK.
+		 */
+		for (cstart = start; cstart <= last; cstart++) {
+			i = ((2 * cstart) + QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT)
+				/ BITS_PER_LONG;
+			if (i != previ) {
+				shadow = (unsigned long)
+					le64_to_cpu(dd->pioavailregs_dma[i]);
+				previ = i;
+			}
+			if (test_bit(((2 * cstart) +
+				      QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT)
+				     % BITS_PER_LONG, &shadow))
+				break;
+		}
+
+		if (cstart > last)
+			break;
+
+		if (sleeps == QTXSLEEPS)
+			break;
+		/* make sure we see an updated copy next time around */
+		sendctrl_7322_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+		sleeps++;
+		msleep(20);
+	}
+
+	switch (which) {
+	case TXCHK_CHG_TYPE_DIS1:
+		/*
+		 * disable checking on a range; used by diags; just
+		 * one buffer, but still written generically
+		 */
+		for (i = start; i <= last; i++)
+			clear_bit(i, dd->cspec->sendchkenable);
+		break;
+
+	case TXCHK_CHG_TYPE_ENAB1:
+		/*
+		 * (re)enable checking on a range; used by diags; just
+		 * one buffer, but still written generically; read
+		 * scratch to be sure buffer actually triggered, not
+		 * just flushed from processor.
+		 */
+		qib_read_kreg32(dd, kr_scratch);
+		for (i = start; i <= last; i++)
+			set_bit(i, dd->cspec->sendchkenable);
+		break;
+
+	case TXCHK_CHG_TYPE_KERN:
+		/* usable by kernel */
+		for (i = start; i <= last; i++) {
+			set_bit(i, dd->cspec->sendibchk);
+			clear_bit(i, dd->cspec->sendgrhchk);
+		}
+		spin_lock_irqsave(&dd->uctxt_lock, flags);
+		/* see if we need to raise avail update threshold */
+		for (i = dd->first_user_ctxt;
+		     dd->cspec->updthresh != dd->cspec->updthresh_dflt
+		     && i < dd->cfgctxts; i++)
+			if (dd->rcd[i] && dd->rcd[i]->subctxt_cnt &&
+			   ((dd->rcd[i]->piocnt / dd->rcd[i]->subctxt_cnt) - 1)
+			   < dd->cspec->updthresh_dflt)
+				break;
+		spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+		if (i == dd->cfgctxts) {
+			spin_lock_irqsave(&dd->sendctrl_lock, flags);
+			dd->cspec->updthresh = dd->cspec->updthresh_dflt;
+			dd->sendctrl &= ~SYM_MASK(SendCtrl, AvailUpdThld);
+			dd->sendctrl |= (dd->cspec->updthresh &
+					 SYM_RMASK(SendCtrl, AvailUpdThld)) <<
+					   SYM_LSB(SendCtrl, AvailUpdThld);
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+			sendctrl_7322_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+		}
+		break;
+
+	case TXCHK_CHG_TYPE_USER:
+		/* for user process */
+		for (i = start; i <= last; i++) {
+			clear_bit(i, dd->cspec->sendibchk);
+			set_bit(i, dd->cspec->sendgrhchk);
+		}
+		spin_lock_irqsave(&dd->sendctrl_lock, flags);
+		if (rcd && rcd->subctxt_cnt && ((rcd->piocnt
+			/ rcd->subctxt_cnt) - 1) < dd->cspec->updthresh) {
+			dd->cspec->updthresh = (rcd->piocnt /
+						rcd->subctxt_cnt) - 1;
+			dd->sendctrl &= ~SYM_MASK(SendCtrl, AvailUpdThld);
+			dd->sendctrl |= (dd->cspec->updthresh &
+					SYM_RMASK(SendCtrl, AvailUpdThld))
+					<< SYM_LSB(SendCtrl, AvailUpdThld);
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+			sendctrl_7322_mod(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+		} else
+			spin_unlock_irqrestore(&dd->sendctrl_lock, flags);
+		break;
+
+	default:
+		break;
+	}
+
+	for (i = start / BITS_PER_LONG; which >= 2 && i <= lastr; ++i)
+		qib_write_kreg(dd, kr_sendcheckmask + i,
+			       dd->cspec->sendchkenable[i]);
+
+	for (i = start / BITS_PER_LONG; which < 2 && i <= lastr; ++i) {
+		qib_write_kreg(dd, kr_sendgrhcheckmask + i,
+			       dd->cspec->sendgrhchk[i]);
+		qib_write_kreg(dd, kr_sendibpktmask + i,
+			       dd->cspec->sendibchk[i]);
+	}
+
+	/*
+	 * Be sure whatever we did was seen by the chip and acted upon,
+	 * before we return.  Mostly important for which >= 2.
+	 */
+	qib_read_kreg32(dd, kr_scratch);
+}
+
+
+/* useful for trigger analyzers, etc. */
+static void writescratch(struct qib_devdata *dd, u32 val)
+{
+	qib_write_kreg(dd, kr_scratch, val);
+}
+
+/* Dummy for now, use chip regs soon */
+static int qib_7322_tempsense_rd(struct qib_devdata *dd, int regnum)
+{
+	return -ENXIO;
+}
+
+/**
+ * qib_init_iba7322_funcs - set up the chip-specific function pointers
+ * @dev: the pci_dev for qlogic_ib device
+ * @ent: pci_device_id struct for this dev
+ *
+ * Also allocates, inits, and returns the devdata struct for this
+ * device instance
+ *
+ * This is global, and is called directly at init to set up the
+ * chip-specific function pointers for later use.
+ */
+struct qib_devdata *qib_init_iba7322_funcs(struct pci_dev *pdev,
+					   const struct pci_device_id *ent)
+{
+	struct qib_devdata *dd;
+	int ret, i;
+	u32 tabsize, actual_cnt = 0;
+
+	dd = qib_alloc_devdata(pdev,
+		NUM_IB_PORTS * sizeof(struct qib_pportdata) +
+		sizeof(struct qib_chip_specific) +
+		NUM_IB_PORTS * sizeof(struct qib_chippport_specific));
+	if (IS_ERR(dd))
+		goto bail;
+
+	dd->f_bringup_serdes    = qib_7322_bringup_serdes;
+	dd->f_cleanup           = qib_setup_7322_cleanup;
+	dd->f_clear_tids        = qib_7322_clear_tids;
+	dd->f_free_irq          = qib_7322_free_irq;
+	dd->f_get_base_info     = qib_7322_get_base_info;
+	dd->f_get_msgheader     = qib_7322_get_msgheader;
+	dd->f_getsendbuf        = qib_7322_getsendbuf;
+	dd->f_gpio_mod          = gpio_7322_mod;
+	dd->f_eeprom_wen        = qib_7322_eeprom_wen;
+	dd->f_hdrqempty         = qib_7322_hdrqempty;
+	dd->f_ib_updown         = qib_7322_ib_updown;
+	dd->f_init_ctxt         = qib_7322_init_ctxt;
+	dd->f_initvl15_bufs     = qib_7322_initvl15_bufs;
+	dd->f_intr_fallback     = qib_7322_intr_fallback;
+	dd->f_late_initreg      = qib_late_7322_initreg;
+	dd->f_setpbc_control    = qib_7322_setpbc_control;
+	dd->f_portcntr          = qib_portcntr_7322;
+	dd->f_put_tid           = qib_7322_put_tid;
+	dd->f_quiet_serdes      = qib_7322_mini_quiet_serdes;
+	dd->f_rcvctrl           = rcvctrl_7322_mod;
+	dd->f_read_cntrs        = qib_read_7322cntrs;
+	dd->f_read_portcntrs    = qib_read_7322portcntrs;
+	dd->f_reset             = qib_do_7322_reset;
+	dd->f_init_sdma_regs    = init_sdma_7322_regs;
+	dd->f_sdma_busy         = qib_sdma_7322_busy;
+	dd->f_sdma_gethead      = qib_sdma_7322_gethead;
+	dd->f_sdma_sendctrl     = qib_7322_sdma_sendctrl;
+	dd->f_sdma_set_desc_cnt = qib_sdma_set_7322_desc_cnt;
+	dd->f_sdma_update_tail  = qib_sdma_update_7322_tail;
+	dd->f_sendctrl          = sendctrl_7322_mod;
+	dd->f_set_armlaunch     = qib_set_7322_armlaunch;
+	dd->f_set_cntr_sample   = qib_set_cntr_7322_sample;
+	dd->f_iblink_state      = qib_7322_iblink_state;
+	dd->f_ibphys_portstate  = qib_7322_phys_portstate;
+	dd->f_get_ib_cfg        = qib_7322_get_ib_cfg;
+	dd->f_set_ib_cfg        = qib_7322_set_ib_cfg;
+	dd->f_set_ib_loopback   = qib_7322_set_loopback;
+	dd->f_get_ib_table      = qib_7322_get_ib_table;
+	dd->f_set_ib_table      = qib_7322_set_ib_table;
+	dd->f_set_intr_state    = qib_7322_set_intr_state;
+	dd->f_setextled         = qib_setup_7322_setextled;
+	dd->f_txchk_change      = qib_7322_txchk_change;
+	dd->f_update_usrhead    = qib_update_7322_usrhead;
+	dd->f_wantpiobuf_intr   = qib_wantpiobuf_7322_intr;
+	dd->f_xgxs_reset        = qib_7322_mini_pcs_reset;
+	dd->f_sdma_hw_clean_up  = qib_7322_sdma_hw_clean_up;
+	dd->f_sdma_hw_start_up  = qib_7322_sdma_hw_start_up;
+	dd->f_sdma_init_early   = qib_7322_sdma_init_early;
+	dd->f_writescratch      = writescratch;
+	dd->f_tempsense_rd	= qib_7322_tempsense_rd;
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dd->f_notify_dca	= qib_7322_notify_dca;
+#endif
+	/*
+	 * Do remaining PCIe setup and save PCIe values in dd.
+	 * Any error printing is already done by the init code.
+	 * On return, we have the chip mapped, but chip registers
+	 * are not set up until start of qib_init_7322_variables.
+	 */
+	ret = qib_pcie_ddinit(dd, pdev, ent);
+	if (ret < 0)
+		goto bail_free;
+
+	/* initialize chip-specific variables */
+	ret = qib_init_7322_variables(dd);
+	if (ret)
+		goto bail_cleanup;
+
+	if (qib_mini_init || !dd->num_pports)
+		goto bail;
+
+	/*
+	 * Determine number of vectors we want; depends on port count
+	 * and number of configured kernel receive queues actually used.
+	 * Should also depend on whether sdma is enabled or not, but
+	 * that's such a rare testing case it's not worth worrying about.
+	 */
+	tabsize = dd->first_user_ctxt + ARRAY_SIZE(irq_table);
+	for (i = 0; i < tabsize; i++)
+		if ((i < ARRAY_SIZE(irq_table) &&
+		     irq_table[i].port <= dd->num_pports) ||
+		    (i >= ARRAY_SIZE(irq_table) &&
+		     dd->rcd[i - ARRAY_SIZE(irq_table)]))
+			actual_cnt++;
+	/* reduce by ctxt's < 2 */
+	if (qib_krcvq01_no_msi)
+		actual_cnt -= dd->num_pports;
+
+	tabsize = actual_cnt;
+	dd->cspec->msix_entries = kzalloc(tabsize *
+			sizeof(struct qib_msix_entry), GFP_KERNEL);
+	if (!dd->cspec->msix_entries) {
+		qib_dev_err(dd, "No memory for MSIx table\n");
+		tabsize = 0;
+	}
+	for (i = 0; i < tabsize; i++)
+		dd->cspec->msix_entries[i].msix.entry = i;
+
+	if (qib_pcie_params(dd, 8, &tabsize, dd->cspec->msix_entries))
+		qib_dev_err(dd,
+			"Failed to setup PCIe or interrupts; continuing anyway\n");
+	/* may be less than we wanted, if not enough available */
+	dd->cspec->num_msix_entries = tabsize;
+
+	/* setup interrupt handler */
+	qib_setup_7322_interrupt(dd, 1);
+
+	/* clear diagctrl register, in case diags were running and crashed */
+	qib_write_kreg(dd, kr_hwdiagctrl, 0);
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	if (!dca_add_requester(&pdev->dev)) {
+		qib_devinfo(dd->pcidev, "DCA enabled\n");
+		dd->flags |= QIB_DCA_ENABLED;
+		qib_setup_dca(dd);
+	}
+#endif
+	goto bail;
+
+bail_cleanup:
+	qib_pcie_ddcleanup(dd);
+bail_free:
+	qib_free_devdata(dd);
+	dd = ERR_PTR(ret);
+bail:
+	return dd;
+}
+
+/*
+ * Set the table entry at the specified index from the table specifed.
+ * There are 3 * TXDDS_TABLE_SZ entries in all per port, with the first
+ * TXDDS_TABLE_SZ for SDR, the next for DDR, and the last for QDR.
+ * 'idx' below addresses the correct entry, while its 4 LSBs select the
+ * corresponding entry (one of TXDDS_TABLE_SZ) from the selected table.
+ */
+#define DDS_ENT_AMP_LSB 14
+#define DDS_ENT_MAIN_LSB 9
+#define DDS_ENT_POST_LSB 5
+#define DDS_ENT_PRE_XTRA_LSB 3
+#define DDS_ENT_PRE_LSB 0
+
+/*
+ * Set one entry in the TxDDS table for spec'd port
+ * ridx picks one of the entries, while tp points
+ * to the appropriate table entry.
+ */
+static void set_txdds(struct qib_pportdata *ppd, int ridx,
+		      const struct txdds_ent *tp)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 pack_ent;
+	int regidx;
+
+	/* Get correct offset in chip-space, and in source table */
+	regidx = KREG_IBPORT_IDX(IBSD_DDS_MAP_TABLE) + ridx;
+	/*
+	 * We do not use qib_write_kreg_port() because it was intended
+	 * only for registers in the lower "port specific" pages.
+	 * So do index calculation  by hand.
+	 */
+	if (ppd->hw_pidx)
+		regidx += (dd->palign / sizeof(u64));
+
+	pack_ent = tp->amp << DDS_ENT_AMP_LSB;
+	pack_ent |= tp->main << DDS_ENT_MAIN_LSB;
+	pack_ent |= tp->pre << DDS_ENT_PRE_LSB;
+	pack_ent |= tp->post << DDS_ENT_POST_LSB;
+	qib_write_kreg(dd, regidx, pack_ent);
+	/* Prevent back-to-back writes by hitting scratch */
+	qib_write_kreg(ppd->dd, kr_scratch, 0);
+}
+
+static const struct vendor_txdds_ent vendor_txdds[] = {
+	{ /* Amphenol 1m 30awg NoEq */
+		{ 0x41, 0x50, 0x48 }, "584470002       ",
+		{ 10,  0,  0,  5 }, { 10,  0,  0,  9 }, {  7,  1,  0, 13 },
+	},
+	{ /* Amphenol 3m 28awg NoEq */
+		{ 0x41, 0x50, 0x48 }, "584470004       ",
+		{  0,  0,  0,  8 }, {  0,  0,  0, 11 }, {  0,  1,  7, 15 },
+	},
+	{ /* Finisar 3m OM2 Optical */
+		{ 0x00, 0x90, 0x65 }, "FCBG410QB1C03-QL",
+		{  0,  0,  0,  3 }, {  0,  0,  0,  4 }, {  0,  0,  0, 13 },
+	},
+	{ /* Finisar 30m OM2 Optical */
+		{ 0x00, 0x90, 0x65 }, "FCBG410QB1C30-QL",
+		{  0,  0,  0,  1 }, {  0,  0,  0,  5 }, {  0,  0,  0, 11 },
+	},
+	{ /* Finisar Default OM2 Optical */
+		{ 0x00, 0x90, 0x65 }, NULL,
+		{  0,  0,  0,  2 }, {  0,  0,  0,  5 }, {  0,  0,  0, 12 },
+	},
+	{ /* Gore 1m 30awg NoEq */
+		{ 0x00, 0x21, 0x77 }, "QSN3300-1       ",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  9 }, {  0,  1,  0, 15 },
+	},
+	{ /* Gore 2m 30awg NoEq */
+		{ 0x00, 0x21, 0x77 }, "QSN3300-2       ",
+		{  0,  0,  0,  8 }, {  0,  0,  0, 10 }, {  0,  1,  7, 15 },
+	},
+	{ /* Gore 1m 28awg NoEq */
+		{ 0x00, 0x21, 0x77 }, "QSN3800-1       ",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  8 }, {  0,  1,  0, 15 },
+	},
+	{ /* Gore 3m 28awg NoEq */
+		{ 0x00, 0x21, 0x77 }, "QSN3800-3       ",
+		{  0,  0,  0,  9 }, {  0,  0,  0, 13 }, {  0,  1,  7, 15 },
+	},
+	{ /* Gore 5m 24awg Eq */
+		{ 0x00, 0x21, 0x77 }, "QSN7000-5       ",
+		{  0,  0,  0,  7 }, {  0,  0,  0,  9 }, {  0,  1,  3, 15 },
+	},
+	{ /* Gore 7m 24awg Eq */
+		{ 0x00, 0x21, 0x77 }, "QSN7000-7       ",
+		{  0,  0,  0,  9 }, {  0,  0,  0, 11 }, {  0,  2,  6, 15 },
+	},
+	{ /* Gore 5m 26awg Eq */
+		{ 0x00, 0x21, 0x77 }, "QSN7600-5       ",
+		{  0,  0,  0,  8 }, {  0,  0,  0, 11 }, {  0,  1,  9, 13 },
+	},
+	{ /* Gore 7m 26awg Eq */
+		{ 0x00, 0x21, 0x77 }, "QSN7600-7       ",
+		{  0,  0,  0,  8 }, {  0,  0,  0, 11 }, {  10,  1,  8, 15 },
+	},
+	{ /* Intersil 12m 24awg Active */
+		{ 0x00, 0x30, 0xB4 }, "QLX4000CQSFP1224",
+		{  0,  0,  0,  2 }, {  0,  0,  0,  5 }, {  0,  3,  0,  9 },
+	},
+	{ /* Intersil 10m 28awg Active */
+		{ 0x00, 0x30, 0xB4 }, "QLX4000CQSFP1028",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  4 }, {  0,  2,  0,  2 },
+	},
+	{ /* Intersil 7m 30awg Active */
+		{ 0x00, 0x30, 0xB4 }, "QLX4000CQSFP0730",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  4 }, {  0,  1,  0,  3 },
+	},
+	{ /* Intersil 5m 32awg Active */
+		{ 0x00, 0x30, 0xB4 }, "QLX4000CQSFP0532",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  6 }, {  0,  2,  0,  8 },
+	},
+	{ /* Intersil Default Active */
+		{ 0x00, 0x30, 0xB4 }, NULL,
+		{  0,  0,  0,  6 }, {  0,  0,  0,  5 }, {  0,  2,  0,  5 },
+	},
+	{ /* Luxtera 20m Active Optical */
+		{ 0x00, 0x25, 0x63 }, NULL,
+		{  0,  0,  0,  5 }, {  0,  0,  0,  8 }, {  0,  2,  0,  12 },
+	},
+	{ /* Molex 1M Cu loopback */
+		{ 0x00, 0x09, 0x3A }, "74763-0025      ",
+		{  2,  2,  6, 15 }, {  2,  2,  6, 15 }, {  2,  2,  6, 15 },
+	},
+	{ /* Molex 2m 28awg NoEq */
+		{ 0x00, 0x09, 0x3A }, "74757-2201      ",
+		{  0,  0,  0,  6 }, {  0,  0,  0,  9 }, {  0,  1,  1, 15 },
+	},
+};
+
+static const struct txdds_ent txdds_sdr[TXDDS_TABLE_SZ] = {
+	/* amp, pre, main, post */
+	{  2, 2, 15,  6 },	/* Loopback */
+	{  0, 0,  0,  1 },	/*  2 dB */
+	{  0, 0,  0,  2 },	/*  3 dB */
+	{  0, 0,  0,  3 },	/*  4 dB */
+	{  0, 0,  0,  4 },	/*  5 dB */
+	{  0, 0,  0,  5 },	/*  6 dB */
+	{  0, 0,  0,  6 },	/*  7 dB */
+	{  0, 0,  0,  7 },	/*  8 dB */
+	{  0, 0,  0,  8 },	/*  9 dB */
+	{  0, 0,  0,  9 },	/* 10 dB */
+	{  0, 0,  0, 10 },	/* 11 dB */
+	{  0, 0,  0, 11 },	/* 12 dB */
+	{  0, 0,  0, 12 },	/* 13 dB */
+	{  0, 0,  0, 13 },	/* 14 dB */
+	{  0, 0,  0, 14 },	/* 15 dB */
+	{  0, 0,  0, 15 },	/* 16 dB */
+};
+
+static const struct txdds_ent txdds_ddr[TXDDS_TABLE_SZ] = {
+	/* amp, pre, main, post */
+	{  2, 2, 15,  6 },	/* Loopback */
+	{  0, 0,  0,  8 },	/*  2 dB */
+	{  0, 0,  0,  8 },	/*  3 dB */
+	{  0, 0,  0,  9 },	/*  4 dB */
+	{  0, 0,  0,  9 },	/*  5 dB */
+	{  0, 0,  0, 10 },	/*  6 dB */
+	{  0, 0,  0, 10 },	/*  7 dB */
+	{  0, 0,  0, 11 },	/*  8 dB */
+	{  0, 0,  0, 11 },	/*  9 dB */
+	{  0, 0,  0, 12 },	/* 10 dB */
+	{  0, 0,  0, 12 },	/* 11 dB */
+	{  0, 0,  0, 13 },	/* 12 dB */
+	{  0, 0,  0, 13 },	/* 13 dB */
+	{  0, 0,  0, 14 },	/* 14 dB */
+	{  0, 0,  0, 14 },	/* 15 dB */
+	{  0, 0,  0, 15 },	/* 16 dB */
+};
+
+static const struct txdds_ent txdds_qdr[TXDDS_TABLE_SZ] = {
+	/* amp, pre, main, post */
+	{  2, 2, 15,  6 },	/* Loopback */
+	{  0, 1,  0,  7 },	/*  2 dB (also QMH7342) */
+	{  0, 1,  0,  9 },	/*  3 dB (also QMH7342) */
+	{  0, 1,  0, 11 },	/*  4 dB */
+	{  0, 1,  0, 13 },	/*  5 dB */
+	{  0, 1,  0, 15 },	/*  6 dB */
+	{  0, 1,  3, 15 },	/*  7 dB */
+	{  0, 1,  7, 15 },	/*  8 dB */
+	{  0, 1,  7, 15 },	/*  9 dB */
+	{  0, 1,  8, 15 },	/* 10 dB */
+	{  0, 1,  9, 15 },	/* 11 dB */
+	{  0, 1, 10, 15 },	/* 12 dB */
+	{  0, 2,  6, 15 },	/* 13 dB */
+	{  0, 2,  7, 15 },	/* 14 dB */
+	{  0, 2,  8, 15 },	/* 15 dB */
+	{  0, 2,  9, 15 },	/* 16 dB */
+};
+
+/*
+ * extra entries for use with txselect, for indices >= TXDDS_TABLE_SZ.
+ * These are mostly used for mez cards going through connectors
+ * and backplane traces, but can be used to add other "unusual"
+ * table values as well.
+ */
+static const struct txdds_ent txdds_extra_sdr[TXDDS_EXTRA_SZ] = {
+	/* amp, pre, main, post */
+	{  0, 0, 0,  1 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  1 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  2 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  2 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  3 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  4 },	/* QMH7342 backplane settings */
+	{  0, 1, 4, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 3, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 12 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 11 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0,  9 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 14 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 2, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 11 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  7 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  9 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  6 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  8 },       /* QME7342 backplane settings 1.1 */
+};
+
+static const struct txdds_ent txdds_extra_ddr[TXDDS_EXTRA_SZ] = {
+	/* amp, pre, main, post */
+	{  0, 0, 0,  7 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  7 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  8 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  8 },	/* QMH7342 backplane settings */
+	{  0, 0, 0,  9 },	/* QMH7342 backplane settings */
+	{  0, 0, 0, 10 },	/* QMH7342 backplane settings */
+	{  0, 1, 4, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 3, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 12 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 11 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0,  9 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 14 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 2, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1, 0, 11 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  7 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  9 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  6 },       /* QME7342 backplane settings 1.1 */
+	{  0, 1, 0,  8 },       /* QME7342 backplane settings 1.1 */
+};
+
+static const struct txdds_ent txdds_extra_qdr[TXDDS_EXTRA_SZ] = {
+	/* amp, pre, main, post */
+	{  0, 1,  0,  4 },	/* QMH7342 backplane settings */
+	{  0, 1,  0,  5 },	/* QMH7342 backplane settings */
+	{  0, 1,  0,  6 },	/* QMH7342 backplane settings */
+	{  0, 1,  0,  8 },	/* QMH7342 backplane settings */
+	{  0, 1,  0, 10 },	/* QMH7342 backplane settings */
+	{  0, 1,  0, 12 },	/* QMH7342 backplane settings */
+	{  0, 1,  4, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  3, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  0, 12 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  0, 11 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  0,  9 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  0, 14 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  2, 15 },	/* QME7342 backplane settings 1.0 */
+	{  0, 1,  0, 11 },      /* QME7342 backplane settings 1.1 */
+	{  0, 1,  0,  7 },      /* QME7342 backplane settings 1.1 */
+	{  0, 1,  0,  9 },      /* QME7342 backplane settings 1.1 */
+	{  0, 1,  0,  6 },      /* QME7342 backplane settings 1.1 */
+	{  0, 1,  0,  8 },      /* QME7342 backplane settings 1.1 */
+};
+
+static const struct txdds_ent txdds_extra_mfg[TXDDS_MFG_SZ] = {
+	/* amp, pre, main, post */
+	{ 0, 0, 0, 0 },         /* QME7342 mfg settings */
+	{ 0, 0, 0, 6 },         /* QME7342 P2 mfg settings */
+};
+
+static const struct txdds_ent *get_atten_table(const struct txdds_ent *txdds,
+					       unsigned atten)
+{
+	/*
+	 * The attenuation table starts at 2dB for entry 1,
+	 * with entry 0 being the loopback entry.
+	 */
+	if (atten <= 2)
+		atten = 1;
+	else if (atten > TXDDS_TABLE_SZ)
+		atten = TXDDS_TABLE_SZ - 1;
+	else
+		atten--;
+	return txdds + atten;
+}
+
+/*
+ * if override is set, the module parameter txselect has a value
+ * for this specific port, so use it, rather than our normal mechanism.
+ */
+static void find_best_ent(struct qib_pportdata *ppd,
+			  const struct txdds_ent **sdr_dds,
+			  const struct txdds_ent **ddr_dds,
+			  const struct txdds_ent **qdr_dds, int override)
+{
+	struct qib_qsfp_cache *qd = &ppd->cpspec->qsfp_data.cache;
+	int idx;
+
+	/* Search table of known cables */
+	for (idx = 0; !override && idx < ARRAY_SIZE(vendor_txdds); ++idx) {
+		const struct vendor_txdds_ent *v = vendor_txdds + idx;
+
+		if (!memcmp(v->oui, qd->oui, QSFP_VOUI_LEN) &&
+		    (!v->partnum ||
+		     !memcmp(v->partnum, qd->partnum, QSFP_PN_LEN))) {
+			*sdr_dds = &v->sdr;
+			*ddr_dds = &v->ddr;
+			*qdr_dds = &v->qdr;
+			return;
+		}
+	}
+
+	/* Active cables don't have attenuation so we only set SERDES
+	 * settings to account for the attenuation of the board traces. */
+	if (!override && QSFP_IS_ACTIVE(qd->tech)) {
+		*sdr_dds = txdds_sdr + ppd->dd->board_atten;
+		*ddr_dds = txdds_ddr + ppd->dd->board_atten;
+		*qdr_dds = txdds_qdr + ppd->dd->board_atten;
+		return;
+	}
+
+	if (!override && QSFP_HAS_ATTEN(qd->tech) && (qd->atten[0] ||
+						      qd->atten[1])) {
+		*sdr_dds = get_atten_table(txdds_sdr, qd->atten[0]);
+		*ddr_dds = get_atten_table(txdds_ddr, qd->atten[0]);
+		*qdr_dds = get_atten_table(txdds_qdr, qd->atten[1]);
+		return;
+	} else if (ppd->cpspec->no_eep < TXDDS_TABLE_SZ) {
+		/*
+		 * If we have no (or incomplete) data from the cable
+		 * EEPROM, or no QSFP, or override is set, use the
+		 * module parameter value to index into the attentuation
+		 * table.
+		 */
+		idx = ppd->cpspec->no_eep;
+		*sdr_dds = &txdds_sdr[idx];
+		*ddr_dds = &txdds_ddr[idx];
+		*qdr_dds = &txdds_qdr[idx];
+	} else if (ppd->cpspec->no_eep < (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ)) {
+		/* similar to above, but index into the "extra" table. */
+		idx = ppd->cpspec->no_eep - TXDDS_TABLE_SZ;
+		*sdr_dds = &txdds_extra_sdr[idx];
+		*ddr_dds = &txdds_extra_ddr[idx];
+		*qdr_dds = &txdds_extra_qdr[idx];
+	} else if ((IS_QME(ppd->dd) || IS_QMH(ppd->dd)) &&
+		   ppd->cpspec->no_eep < (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ +
+					  TXDDS_MFG_SZ)) {
+		idx = ppd->cpspec->no_eep - (TXDDS_TABLE_SZ + TXDDS_EXTRA_SZ);
+		pr_info("IB%u:%u use idx %u into txdds_mfg\n",
+			ppd->dd->unit, ppd->port, idx);
+		*sdr_dds = &txdds_extra_mfg[idx];
+		*ddr_dds = &txdds_extra_mfg[idx];
+		*qdr_dds = &txdds_extra_mfg[idx];
+	} else {
+		/* this shouldn't happen, it's range checked */
+		*sdr_dds = txdds_sdr + qib_long_atten;
+		*ddr_dds = txdds_ddr + qib_long_atten;
+		*qdr_dds = txdds_qdr + qib_long_atten;
+	}
+}
+
+static void init_txdds_table(struct qib_pportdata *ppd, int override)
+{
+	const struct txdds_ent *sdr_dds, *ddr_dds, *qdr_dds;
+	struct txdds_ent *dds;
+	int idx;
+	int single_ent = 0;
+
+	find_best_ent(ppd, &sdr_dds, &ddr_dds, &qdr_dds, override);
+
+	/* for mez cards or override, use the selected value for all entries */
+	if (!(ppd->dd->flags & QIB_HAS_QSFP) || override)
+		single_ent = 1;
+
+	/* Fill in the first entry with the best entry found. */
+	set_txdds(ppd, 0, sdr_dds);
+	set_txdds(ppd, TXDDS_TABLE_SZ, ddr_dds);
+	set_txdds(ppd, 2 * TXDDS_TABLE_SZ, qdr_dds);
+	if (ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED |
+		QIBL_LINKACTIVE)) {
+		dds = (struct txdds_ent *)(ppd->link_speed_active ==
+					   QIB_IB_QDR ?  qdr_dds :
+					   (ppd->link_speed_active ==
+					    QIB_IB_DDR ? ddr_dds : sdr_dds));
+		write_tx_serdes_param(ppd, dds);
+	}
+
+	/* Fill in the remaining entries with the default table values. */
+	for (idx = 1; idx < ARRAY_SIZE(txdds_sdr); ++idx) {
+		set_txdds(ppd, idx, single_ent ? sdr_dds : txdds_sdr + idx);
+		set_txdds(ppd, idx + TXDDS_TABLE_SZ,
+			  single_ent ? ddr_dds : txdds_ddr + idx);
+		set_txdds(ppd, idx + 2 * TXDDS_TABLE_SZ,
+			  single_ent ? qdr_dds : txdds_qdr + idx);
+	}
+}
+
+#define KR_AHB_ACC KREG_IDX(ahb_access_ctrl)
+#define KR_AHB_TRANS KREG_IDX(ahb_transaction_reg)
+#define AHB_TRANS_RDY SYM_MASK(ahb_transaction_reg, ahb_rdy)
+#define AHB_ADDR_LSB SYM_LSB(ahb_transaction_reg, ahb_address)
+#define AHB_DATA_LSB SYM_LSB(ahb_transaction_reg, ahb_data)
+#define AHB_WR SYM_MASK(ahb_transaction_reg, write_not_read)
+#define AHB_TRANS_TRIES 10
+
+/*
+ * The chan argument is 0=chan0, 1=chan1, 2=pll, 3=chan2, 4=chan4,
+ * 5=subsystem which is why most calls have "chan + chan >> 1"
+ * for the channel argument.
+ */
+static u32 ahb_mod(struct qib_devdata *dd, int quad, int chan, int addr,
+		    u32 data, u32 mask)
+{
+	u32 rd_data, wr_data, sz_mask;
+	u64 trans, acc, prev_acc;
+	u32 ret = 0xBAD0BAD;
+	int tries;
+
+	prev_acc = qib_read_kreg64(dd, KR_AHB_ACC);
+	/* From this point on, make sure we return access */
+	acc = (quad << 1) | 1;
+	qib_write_kreg(dd, KR_AHB_ACC, acc);
+
+	for (tries = 1; tries < AHB_TRANS_TRIES; ++tries) {
+		trans = qib_read_kreg64(dd, KR_AHB_TRANS);
+		if (trans & AHB_TRANS_RDY)
+			break;
+	}
+	if (tries >= AHB_TRANS_TRIES) {
+		qib_dev_err(dd, "No ahb_rdy in %d tries\n", AHB_TRANS_TRIES);
+		goto bail;
+	}
+
+	/* If mask is not all 1s, we need to read, but different SerDes
+	 * entities have different sizes
+	 */
+	sz_mask = (1UL << ((quad == 1) ? 32 : 16)) - 1;
+	wr_data = data & mask & sz_mask;
+	if ((~mask & sz_mask) != 0) {
+		trans = ((chan << 6) | addr) << (AHB_ADDR_LSB + 1);
+		qib_write_kreg(dd, KR_AHB_TRANS, trans);
+
+		for (tries = 1; tries < AHB_TRANS_TRIES; ++tries) {
+			trans = qib_read_kreg64(dd, KR_AHB_TRANS);
+			if (trans & AHB_TRANS_RDY)
+				break;
+		}
+		if (tries >= AHB_TRANS_TRIES) {
+			qib_dev_err(dd, "No Rd ahb_rdy in %d tries\n",
+				    AHB_TRANS_TRIES);
+			goto bail;
+		}
+		/* Re-read in case host split reads and read data first */
+		trans = qib_read_kreg64(dd, KR_AHB_TRANS);
+		rd_data = (uint32_t)(trans >> AHB_DATA_LSB);
+		wr_data |= (rd_data & ~mask & sz_mask);
+	}
+
+	/* If mask is not zero, we need to write. */
+	if (mask & sz_mask) {
+		trans = ((chan << 6) | addr) << (AHB_ADDR_LSB + 1);
+		trans |= ((uint64_t)wr_data << AHB_DATA_LSB);
+		trans |= AHB_WR;
+		qib_write_kreg(dd, KR_AHB_TRANS, trans);
+
+		for (tries = 1; tries < AHB_TRANS_TRIES; ++tries) {
+			trans = qib_read_kreg64(dd, KR_AHB_TRANS);
+			if (trans & AHB_TRANS_RDY)
+				break;
+		}
+		if (tries >= AHB_TRANS_TRIES) {
+			qib_dev_err(dd, "No Wr ahb_rdy in %d tries\n",
+				    AHB_TRANS_TRIES);
+			goto bail;
+		}
+	}
+	ret = wr_data;
+bail:
+	qib_write_kreg(dd, KR_AHB_ACC, prev_acc);
+	return ret;
+}
+
+static void ibsd_wr_allchans(struct qib_pportdata *ppd, int addr, unsigned data,
+			     unsigned mask)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int chan;
+	u32 rbc;
+
+	for (chan = 0; chan < SERDES_CHANS; ++chan) {
+		ahb_mod(dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)), addr,
+			data, mask);
+		rbc = ahb_mod(dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+			      addr, 0, 0);
+	}
+}
+
+static void serdes_7322_los_enable(struct qib_pportdata *ppd, int enable)
+{
+	u64 data = qib_read_kreg_port(ppd, krp_serdesctrl);
+	u8 state = SYM_FIELD(data, IBSerdesCtrl_0, RXLOSEN);
+
+	if (enable && !state) {
+		pr_info("IB%u:%u Turning LOS on\n",
+			ppd->dd->unit, ppd->port);
+		data |= SYM_MASK(IBSerdesCtrl_0, RXLOSEN);
+	} else if (!enable && state) {
+		pr_info("IB%u:%u Turning LOS off\n",
+			ppd->dd->unit, ppd->port);
+		data &= ~SYM_MASK(IBSerdesCtrl_0, RXLOSEN);
+	}
+	qib_write_kreg_port(ppd, krp_serdesctrl, data);
+}
+
+static int serdes_7322_init(struct qib_pportdata *ppd)
+{
+	int ret = 0;
+	if (ppd->dd->cspec->r1)
+		ret = serdes_7322_init_old(ppd);
+	else
+		ret = serdes_7322_init_new(ppd);
+	return ret;
+}
+
+static int serdes_7322_init_old(struct qib_pportdata *ppd)
+{
+	u32 le_val;
+
+	/*
+	 * Initialize the Tx DDS tables.  Also done every QSFP event,
+	 * for adapters with QSFP
+	 */
+	init_txdds_table(ppd, 0);
+
+	/* ensure no tx overrides from earlier driver loads */
+	qib_write_kreg_port(ppd, krp_tx_deemph_override,
+		SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		reset_tx_deemphasis_override));
+
+	/* Patch some SerDes defaults to "Better for IB" */
+	/* Timing Loop Bandwidth: cdr_timing[11:9] = 0 */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(11, 9));
+
+	/* Termination: rxtermctrl_r2d addr 11 bits [12:11] = 1 */
+	ibsd_wr_allchans(ppd, 11, (1 << 11), BMASK(12, 11));
+	/* Enable LE2: rxle2en_r2a addr 13 bit [6] = 1 */
+	ibsd_wr_allchans(ppd, 13, (1 << 6), (1 << 6));
+
+	/* May be overridden in qsfp_7322_event */
+	le_val = IS_QME(ppd->dd) ? LE2_QME : LE2_DEFAULT;
+	ibsd_wr_allchans(ppd, 13, (le_val << 7), BMASK(9, 7));
+
+	/* enable LE1 adaptation for all but QME, which is disabled */
+	le_val = IS_QME(ppd->dd) ? 0 : 1;
+	ibsd_wr_allchans(ppd, 13, (le_val << 5), (1 << 5));
+
+	/* Clear cmode-override, may be set from older driver */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14);
+
+	/* Timing Recovery: rxtapsel addr 5 bits [9:8] = 0 */
+	ibsd_wr_allchans(ppd, 5, (0 << 8), BMASK(9, 8));
+
+	/* setup LoS params; these are subsystem, so chan == 5 */
+	/* LoS filter threshold_count on, ch 0-3, set to 8 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 5, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 8 << 4, BMASK(7, 4));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 8, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 8 << 4, BMASK(7, 4));
+
+	/* LoS filter threshold_count off, ch 0-3, set to 4 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 6, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 4 << 8, BMASK(11, 8));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 4 << 8, BMASK(11, 8));
+
+	/* LoS filter select enabled */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 1 << 15, 1 << 15);
+
+	/* LoS target data:  SDR=4, DDR=2, QDR=1 */
+	ibsd_wr_allchans(ppd, 14, (1 << 3), BMASK(5, 3)); /* QDR */
+	ibsd_wr_allchans(ppd, 20, (2 << 10), BMASK(12, 10)); /* DDR */
+	ibsd_wr_allchans(ppd, 20, (4 << 13), BMASK(15, 13)); /* SDR */
+
+	serdes_7322_los_enable(ppd, 1);
+
+	/* rxbistena; set 0 to avoid effects of it switch later */
+	ibsd_wr_allchans(ppd, 9, 0 << 15, 1 << 15);
+
+	/* Configure 4 DFE taps, and only they adapt */
+	ibsd_wr_allchans(ppd, 16, 0 << 0, BMASK(1, 0));
+
+	/* gain hi stop 32 (22) (6:1) lo stop 7 (10:7) target 22 (13) (15:11) */
+	le_val = (ppd->dd->cspec->r1 || IS_QME(ppd->dd)) ? 0xb6c0 : 0x6bac;
+	ibsd_wr_allchans(ppd, 21, le_val, 0xfffe);
+
+	/*
+	 * Set receive adaptation mode.  SDR and DDR adaptation are
+	 * always on, and QDR is initially enabled; later disabled.
+	 */
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(0), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(1), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+			    ppd->dd->cspec->r1 ?
+			    QDR_STATIC_ADAPT_DOWN_R1 : QDR_STATIC_ADAPT_DOWN);
+	ppd->cpspec->qdr_dfe_on = 1;
+
+	/* FLoop LOS gate: PPM filter  enabled */
+	ibsd_wr_allchans(ppd, 38, 0 << 10, 1 << 10);
+
+	/* rx offset center enabled */
+	ibsd_wr_allchans(ppd, 12, 1 << 4, 1 << 4);
+
+	if (!ppd->dd->cspec->r1) {
+		ibsd_wr_allchans(ppd, 12, 1 << 12, 1 << 12);
+		ibsd_wr_allchans(ppd, 12, 2 << 8, 0x0f << 8);
+	}
+
+	/* Set the frequency loop bandwidth to 15 */
+	ibsd_wr_allchans(ppd, 2, 15 << 5, BMASK(8, 5));
+
+	return 0;
+}
+
+static int serdes_7322_init_new(struct qib_pportdata *ppd)
+{
+	unsigned long tend;
+	u32 le_val, rxcaldone;
+	int chan, chan_done = (1 << SERDES_CHANS) - 1;
+
+	/* Clear cmode-override, may be set from older driver */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 0 << 14, 1 << 14);
+
+	/* ensure no tx overrides from earlier driver loads */
+	qib_write_kreg_port(ppd, krp_tx_deemph_override,
+		SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		reset_tx_deemphasis_override));
+
+	/* START OF LSI SUGGESTED SERDES BRINGUP */
+	/* Reset - Calibration Setup */
+	/*       Stop DFE adaptaion */
+	ibsd_wr_allchans(ppd, 1, 0, BMASK(9, 1));
+	/*       Disable LE1 */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(5, 5));
+	/*       Disable autoadapt for LE1 */
+	ibsd_wr_allchans(ppd, 1, 0, BMASK(15, 15));
+	/*       Disable LE2 */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(6, 6));
+	/*       Disable VGA */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(0, 0));
+	/*       Disable AFE Offset Cancel */
+	ibsd_wr_allchans(ppd, 12, 0, BMASK(12, 12));
+	/*       Disable Timing Loop */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(3, 3));
+	/*       Disable Frequency Loop */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(4, 4));
+	/*       Disable Baseline Wander Correction */
+	ibsd_wr_allchans(ppd, 13, 0, BMASK(13, 13));
+	/*       Disable RX Calibration */
+	ibsd_wr_allchans(ppd, 4, 0, BMASK(10, 10));
+	/*       Disable RX Offset Calibration */
+	ibsd_wr_allchans(ppd, 12, 0, BMASK(4, 4));
+	/*       Select BB CDR */
+	ibsd_wr_allchans(ppd, 2, (1 << 15), BMASK(15, 15));
+	/*       CDR Step Size */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(9, 8));
+	/*       Enable phase Calibration */
+	ibsd_wr_allchans(ppd, 12, (1 << 5), BMASK(5, 5));
+	/*       DFE Bandwidth [2:14-12] */
+	ibsd_wr_allchans(ppd, 2, (4 << 12), BMASK(14, 12));
+	/*       DFE Config (4 taps only) */
+	ibsd_wr_allchans(ppd, 16, 0, BMASK(1, 0));
+	/*       Gain Loop Bandwidth */
+	if (!ppd->dd->cspec->r1) {
+		ibsd_wr_allchans(ppd, 12, 1 << 12, BMASK(12, 12));
+		ibsd_wr_allchans(ppd, 12, 2 << 8, BMASK(11, 8));
+	} else {
+		ibsd_wr_allchans(ppd, 19, (3 << 11), BMASK(13, 11));
+	}
+	/*       Baseline Wander Correction Gain [13:4-0] (leave as default) */
+	/*       Baseline Wander Correction Gain [3:7-5] (leave as default) */
+	/*       Data Rate Select [5:7-6] (leave as default) */
+	/*       RX Parallel Word Width [3:10-8] (leave as default) */
+
+	/* RX REST */
+	/*       Single- or Multi-channel reset */
+	/*       RX Analog reset */
+	/*       RX Digital reset */
+	ibsd_wr_allchans(ppd, 0, 0, BMASK(15, 13));
+	msleep(20);
+	/*       RX Analog reset */
+	ibsd_wr_allchans(ppd, 0, (1 << 14), BMASK(14, 14));
+	msleep(20);
+	/*       RX Digital reset */
+	ibsd_wr_allchans(ppd, 0, (1 << 13), BMASK(13, 13));
+	msleep(20);
+
+	/* setup LoS params; these are subsystem, so chan == 5 */
+	/* LoS filter threshold_count on, ch 0-3, set to 8 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 5, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 8 << 4, BMASK(7, 4));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 8, 8 << 11, BMASK(14, 11));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 8 << 4, BMASK(7, 4));
+
+	/* LoS filter threshold_count off, ch 0-3, set to 4 */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 6, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 7, 4 << 8, BMASK(11, 8));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 4 << 0, BMASK(3, 0));
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 10, 4 << 8, BMASK(11, 8));
+
+	/* LoS filter select enabled */
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), 5, 9, 1 << 15, 1 << 15);
+
+	/* LoS target data:  SDR=4, DDR=2, QDR=1 */
+	ibsd_wr_allchans(ppd, 14, (1 << 3), BMASK(5, 3)); /* QDR */
+	ibsd_wr_allchans(ppd, 20, (2 << 10), BMASK(12, 10)); /* DDR */
+	ibsd_wr_allchans(ppd, 20, (4 << 13), BMASK(15, 13)); /* SDR */
+
+	/* Turn on LOS on initial SERDES init */
+	serdes_7322_los_enable(ppd, 1);
+	/* FLoop LOS gate: PPM filter  enabled */
+	ibsd_wr_allchans(ppd, 38, 0 << 10, 1 << 10);
+
+	/* RX LATCH CALIBRATION */
+	/*       Enable Eyefinder Phase Calibration latch */
+	ibsd_wr_allchans(ppd, 15, 1, BMASK(0, 0));
+	/*       Enable RX Offset Calibration latch */
+	ibsd_wr_allchans(ppd, 12, (1 << 4), BMASK(4, 4));
+	msleep(20);
+	/*       Start Calibration */
+	ibsd_wr_allchans(ppd, 4, (1 << 10), BMASK(10, 10));
+	tend = jiffies + msecs_to_jiffies(500);
+	while (chan_done && !time_is_before_jiffies(tend)) {
+		msleep(20);
+		for (chan = 0; chan < SERDES_CHANS; ++chan) {
+			rxcaldone = ahb_mod(ppd->dd, IBSD(ppd->hw_pidx),
+					    (chan + (chan >> 1)),
+					    25, 0, 0);
+			if ((~rxcaldone & (u32)BMASK(9, 9)) == 0 &&
+			    (~chan_done & (1 << chan)) == 0)
+				chan_done &= ~(1 << chan);
+		}
+	}
+	if (chan_done) {
+		pr_info("Serdes %d calibration not done after .5 sec: 0x%x\n",
+			 IBSD(ppd->hw_pidx), chan_done);
+	} else {
+		for (chan = 0; chan < SERDES_CHANS; ++chan) {
+			rxcaldone = ahb_mod(ppd->dd, IBSD(ppd->hw_pidx),
+					    (chan + (chan >> 1)),
+					    25, 0, 0);
+			if ((~rxcaldone & (u32)BMASK(10, 10)) == 0)
+				pr_info("Serdes %d chan %d calibration failed\n",
+					IBSD(ppd->hw_pidx), chan);
+		}
+	}
+
+	/*       Turn off Calibration */
+	ibsd_wr_allchans(ppd, 4, 0, BMASK(10, 10));
+	msleep(20);
+
+	/* BRING RX UP */
+	/*       Set LE2 value (May be overridden in qsfp_7322_event) */
+	le_val = IS_QME(ppd->dd) ? LE2_QME : LE2_DEFAULT;
+	ibsd_wr_allchans(ppd, 13, (le_val << 7), BMASK(9, 7));
+	/*       Set LE2 Loop bandwidth */
+	ibsd_wr_allchans(ppd, 3, (7 << 5), BMASK(7, 5));
+	/*       Enable LE2 */
+	ibsd_wr_allchans(ppd, 13, (1 << 6), BMASK(6, 6));
+	msleep(20);
+	/*       Enable H0 only */
+	ibsd_wr_allchans(ppd, 1, 1, BMASK(9, 1));
+	/* gain hi stop 32 (22) (6:1) lo stop 7 (10:7) target 22 (13) (15:11) */
+	le_val = (ppd->dd->cspec->r1 || IS_QME(ppd->dd)) ? 0xb6c0 : 0x6bac;
+	ibsd_wr_allchans(ppd, 21, le_val, 0xfffe);
+	/*       Enable VGA */
+	ibsd_wr_allchans(ppd, 5, 0, BMASK(0, 0));
+	msleep(20);
+	/*       Set Frequency Loop Bandwidth */
+	ibsd_wr_allchans(ppd, 2, (15 << 5), BMASK(8, 5));
+	/*       Enable Frequency Loop */
+	ibsd_wr_allchans(ppd, 2, (1 << 4), BMASK(4, 4));
+	/*       Set Timing Loop Bandwidth */
+	ibsd_wr_allchans(ppd, 2, 0, BMASK(11, 9));
+	/*       Enable Timing Loop */
+	ibsd_wr_allchans(ppd, 2, (1 << 3), BMASK(3, 3));
+	msleep(50);
+	/*       Enable DFE
+	 *       Set receive adaptation mode.  SDR and DDR adaptation are
+	 *       always on, and QDR is initially enabled; later disabled.
+	 */
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(0), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(1), 0ULL);
+	qib_write_kreg_port(ppd, krp_static_adapt_dis(2),
+			    ppd->dd->cspec->r1 ?
+			    QDR_STATIC_ADAPT_DOWN_R1 : QDR_STATIC_ADAPT_DOWN);
+	ppd->cpspec->qdr_dfe_on = 1;
+	/*       Disable LE1  */
+	ibsd_wr_allchans(ppd, 13, (0 << 5), (1 << 5));
+	/*       Disable auto adapt for LE1 */
+	ibsd_wr_allchans(ppd, 1, (0 << 15), BMASK(15, 15));
+	msleep(20);
+	/*       Enable AFE Offset Cancel */
+	ibsd_wr_allchans(ppd, 12, (1 << 12), BMASK(12, 12));
+	/*       Enable Baseline Wander Correction */
+	ibsd_wr_allchans(ppd, 12, (1 << 13), BMASK(13, 13));
+	/* Termination: rxtermctrl_r2d addr 11 bits [12:11] = 1 */
+	ibsd_wr_allchans(ppd, 11, (1 << 11), BMASK(12, 11));
+	/* VGA output common mode */
+	ibsd_wr_allchans(ppd, 12, (3 << 2), BMASK(3, 2));
+
+	/*
+	 * Initialize the Tx DDS tables.  Also done every QSFP event,
+	 * for adapters with QSFP
+	 */
+	init_txdds_table(ppd, 0);
+
+	return 0;
+}
+
+/* start adjust QMH serdes parameters */
+
+static void set_man_code(struct qib_pportdata *ppd, int chan, int code)
+{
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+		9, code << 9, 0x3f << 9);
+}
+
+static void set_man_mode_h1(struct qib_pportdata *ppd, int chan,
+	int enable, u32 tapenable)
+{
+	if (enable)
+		ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+			1, 3 << 10, 0x1f << 10);
+	else
+		ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+			1, 0, 0x1f << 10);
+}
+
+/* Set clock to 1, 0, 1, 0 */
+static void clock_man(struct qib_pportdata *ppd, int chan)
+{
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+		4, 0x4000, 0x4000);
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+		4, 0, 0x4000);
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+		4, 0x4000, 0x4000);
+	ahb_mod(ppd->dd, IBSD(ppd->hw_pidx), (chan + (chan >> 1)),
+		4, 0, 0x4000);
+}
+
+/*
+ * write the current Tx serdes pre,post,main,amp settings into the serdes.
+ * The caller must pass the settings appropriate for the current speed,
+ * or not care if they are correct for the current speed.
+ */
+static void write_tx_serdes_param(struct qib_pportdata *ppd,
+				  struct txdds_ent *txdds)
+{
+	u64 deemph;
+
+	deemph = qib_read_kreg_port(ppd, krp_tx_deemph_override);
+	/* field names for amp, main, post, pre, respectively */
+	deemph &= ~(SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, txampcntl_d2a) |
+		    SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, txc0_ena) |
+		    SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, txcp1_ena) |
+		    SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0, txcn1_ena));
+
+	deemph |= SYM_MASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+			   tx_override_deemphasis_select);
+	deemph |= (txdds->amp & SYM_RMASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		    txampcntl_d2a)) << SYM_LSB(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+				       txampcntl_d2a);
+	deemph |= (txdds->main & SYM_RMASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		     txc0_ena)) << SYM_LSB(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+				   txc0_ena);
+	deemph |= (txdds->post & SYM_RMASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		     txcp1_ena)) << SYM_LSB(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+				    txcp1_ena);
+	deemph |= (txdds->pre & SYM_RMASK(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+		     txcn1_ena)) << SYM_LSB(IBSD_TX_DEEMPHASIS_OVERRIDE_0,
+				    txcn1_ena);
+	qib_write_kreg_port(ppd, krp_tx_deemph_override, deemph);
+}
+
+/*
+ * Set the parameters for mez cards on link bounce, so they are
+ * always exactly what was requested.  Similar logic to init_txdds
+ * but does just the serdes.
+ */
+static void adj_tx_serdes(struct qib_pportdata *ppd)
+{
+	const struct txdds_ent *sdr_dds, *ddr_dds, *qdr_dds;
+	struct txdds_ent *dds;
+
+	find_best_ent(ppd, &sdr_dds, &ddr_dds, &qdr_dds, 1);
+	dds = (struct txdds_ent *)(ppd->link_speed_active == QIB_IB_QDR ?
+		qdr_dds : (ppd->link_speed_active == QIB_IB_DDR ?
+				ddr_dds : sdr_dds));
+	write_tx_serdes_param(ppd, dds);
+}
+
+/* set QDR forced value for H1, if needed */
+static void force_h1(struct qib_pportdata *ppd)
+{
+	int chan;
+
+	ppd->cpspec->qdr_reforce = 0;
+	if (!ppd->dd->cspec->r1)
+		return;
+
+	for (chan = 0; chan < SERDES_CHANS; chan++) {
+		set_man_mode_h1(ppd, chan, 1, 0);
+		set_man_code(ppd, chan, ppd->cpspec->h1_val);
+		clock_man(ppd, chan);
+		set_man_mode_h1(ppd, chan, 0, 0);
+	}
+}
+
+#define SJA_EN SYM_MASK(SPC_JTAG_ACCESS_REG, SPC_JTAG_ACCESS_EN)
+#define BISTEN_LSB SYM_LSB(SPC_JTAG_ACCESS_REG, bist_en)
+
+#define R_OPCODE_LSB 3
+#define R_OP_NOP 0
+#define R_OP_SHIFT 2
+#define R_OP_UPDATE 3
+#define R_TDI_LSB 2
+#define R_TDO_LSB 1
+#define R_RDY 1
+
+static int qib_r_grab(struct qib_devdata *dd)
+{
+	u64 val;
+	val = SJA_EN;
+	qib_write_kreg(dd, kr_r_access, val);
+	qib_read_kreg32(dd, kr_scratch);
+	return 0;
+}
+
+/* qib_r_wait_for_rdy() not only waits for the ready bit, it
+ * returns the current state of R_TDO
+ */
+static int qib_r_wait_for_rdy(struct qib_devdata *dd)
+{
+	u64 val;
+	int timeout;
+	for (timeout = 0; timeout < 100 ; ++timeout) {
+		val = qib_read_kreg32(dd, kr_r_access);
+		if (val & R_RDY)
+			return (val >> R_TDO_LSB) & 1;
+	}
+	return -1;
+}
+
+static int qib_r_shift(struct qib_devdata *dd, int bisten,
+		       int len, u8 *inp, u8 *outp)
+{
+	u64 valbase, val;
+	int ret, pos;
+
+	valbase = SJA_EN | (bisten << BISTEN_LSB) |
+		(R_OP_SHIFT << R_OPCODE_LSB);
+	ret = qib_r_wait_for_rdy(dd);
+	if (ret < 0)
+		goto bail;
+	for (pos = 0; pos < len; ++pos) {
+		val = valbase;
+		if (outp) {
+			outp[pos >> 3] &= ~(1 << (pos & 7));
+			outp[pos >> 3] |= (ret << (pos & 7));
+		}
+		if (inp) {
+			int tdi = inp[pos >> 3] >> (pos & 7);
+			val |= ((tdi & 1) << R_TDI_LSB);
+		}
+		qib_write_kreg(dd, kr_r_access, val);
+		qib_read_kreg32(dd, kr_scratch);
+		ret = qib_r_wait_for_rdy(dd);
+		if (ret < 0)
+			break;
+	}
+	/* Restore to NOP between operations. */
+	val =  SJA_EN | (bisten << BISTEN_LSB);
+	qib_write_kreg(dd, kr_r_access, val);
+	qib_read_kreg32(dd, kr_scratch);
+	ret = qib_r_wait_for_rdy(dd);
+
+	if (ret >= 0)
+		ret = pos;
+bail:
+	return ret;
+}
+
+static int qib_r_update(struct qib_devdata *dd, int bisten)
+{
+	u64 val;
+	int ret;
+
+	val = SJA_EN | (bisten << BISTEN_LSB) | (R_OP_UPDATE << R_OPCODE_LSB);
+	ret = qib_r_wait_for_rdy(dd);
+	if (ret >= 0) {
+		qib_write_kreg(dd, kr_r_access, val);
+		qib_read_kreg32(dd, kr_scratch);
+	}
+	return ret;
+}
+
+#define BISTEN_PORT_SEL 15
+#define LEN_PORT_SEL 625
+#define BISTEN_AT 17
+#define LEN_AT 156
+#define BISTEN_ETM 16
+#define LEN_ETM 632
+
+#define BIT2BYTE(x) (((x) +  BITS_PER_BYTE - 1) / BITS_PER_BYTE)
+
+/* these are common for all IB port use cases. */
+static u8 reset_at[BIT2BYTE(LEN_AT)] = {
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00,
+};
+static u8 reset_atetm[BIT2BYTE(LEN_ETM)] = {
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x80, 0xe3, 0x81, 0x73, 0x3c, 0x70, 0x8e,
+	0x07, 0xce, 0xf1, 0xc0, 0x39, 0x1e, 0x38, 0xc7, 0x03, 0xe7,
+	0x78, 0xe0, 0x1c, 0x0f, 0x9c, 0x7f, 0x80, 0x73, 0x0f, 0x70,
+	0xde, 0x01, 0xce, 0x39, 0xc0, 0xf9, 0x06, 0x38, 0xd7, 0x00,
+	0xe7, 0x19, 0xe0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0xff, 0xff, 0xff, 0xff, 0x00, 0x00, 0x00, 0x00,
+};
+static u8 at[BIT2BYTE(LEN_AT)] = {
+	0x00, 0x00, 0x18, 0x00, 0x00, 0x00, 0x18, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x20, 0x00,
+};
+
+/* used for IB1 or IB2, only one in use */
+static u8 atetm_1port[BIT2BYTE(LEN_ETM)] = {
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x10, 0xf2, 0x80, 0x83, 0x1e, 0x38, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x50, 0xf4, 0x41, 0x00, 0x18, 0x78, 0xc8, 0x03,
+	0x07, 0x7b, 0xa0, 0x3e, 0x00, 0x02, 0x00, 0x00, 0x18, 0x00,
+	0x18, 0x00, 0x00, 0x00, 0x00, 0x4b, 0x00, 0x00, 0x00,
+};
+
+/* used when both IB1 and IB2 are in use */
+static u8 atetm_2port[BIT2BYTE(LEN_ETM)] = {
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x79,
+	0xc0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00,
+	0x00, 0x00, 0xf8, 0x80, 0x83, 0x1e, 0x38, 0xe0, 0x03, 0x05,
+	0x7b, 0xa0, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x80,
+	0xa2, 0x0f, 0x50, 0xf4, 0x41, 0x00, 0x18, 0x78, 0xd1, 0x07,
+	0x02, 0x7c, 0x80, 0x3e, 0x00, 0x02, 0x00, 0x00, 0x3e, 0x00,
+	0x02, 0x00, 0x00, 0x00, 0x00, 0x64, 0x00, 0x00, 0x00,
+};
+
+/* used when only IB1 is in use */
+static u8 portsel_port1[BIT2BYTE(LEN_PORT_SEL)] = {
+	0x32, 0x65, 0xa4, 0x7b, 0x10, 0x98, 0xdc, 0xfe, 0x13, 0x13,
+	0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x73, 0x0c, 0x0c, 0x0c,
+	0x0c, 0x0c, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13,
+	0x13, 0x78, 0x78, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13,
+	0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x74, 0x32,
+	0x32, 0x32, 0x32, 0x32, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14,
+	0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14,
+	0x14, 0x14, 0x9f, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
+};
+
+/* used when only IB2 is in use */
+static u8 portsel_port2[BIT2BYTE(LEN_PORT_SEL)] = {
+	0x32, 0x65, 0xa4, 0x7b, 0x10, 0x98, 0xdc, 0xfe, 0x39, 0x39,
+	0x39, 0x39, 0x39, 0x39, 0x39, 0x39, 0x73, 0x32, 0x32, 0x32,
+	0x32, 0x32, 0x39, 0x39, 0x39, 0x39, 0x39, 0x39, 0x39, 0x39,
+	0x39, 0x78, 0x78, 0x39, 0x39, 0x39, 0x39, 0x39, 0x39, 0x39,
+	0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x74, 0x32,
+	0x32, 0x32, 0x32, 0x32, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a,
+	0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a, 0x3a,
+	0x3a, 0x3a, 0x9f, 0x01, 0x00, 0x00, 0x00, 0x00, 0x01,
+};
+
+/* used when both IB1 and IB2 are in use */
+static u8 portsel_2port[BIT2BYTE(LEN_PORT_SEL)] = {
+	0x32, 0xba, 0x54, 0x76, 0x10, 0x98, 0xdc, 0xfe, 0x13, 0x13,
+	0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x73, 0x0c, 0x0c, 0x0c,
+	0x0c, 0x0c, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13,
+	0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13, 0x13,
+	0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x74, 0x32,
+	0x32, 0x32, 0x32, 0x32, 0x14, 0x14, 0x14, 0x14, 0x14, 0x3a,
+	0x3a, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14, 0x14,
+	0x14, 0x14, 0x9f, 0x01, 0x00, 0x00, 0x00, 0x00, 0x00,
+};
+
+/*
+ * Do setup to properly handle IB link recovery; if port is zero, we
+ * are initializing to cover both ports; otherwise we are initializing
+ * to cover a single port card, or the port has reached INIT and we may
+ * need to switch coverage types.
+ */
+static void setup_7322_link_recovery(struct qib_pportdata *ppd, u32 both)
+{
+	u8 *portsel, *etm;
+	struct qib_devdata *dd = ppd->dd;
+
+	if (!ppd->dd->cspec->r1)
+		return;
+	if (!both) {
+		dd->cspec->recovery_ports_initted++;
+		ppd->cpspec->recovery_init = 1;
+	}
+	if (!both && dd->cspec->recovery_ports_initted == 1) {
+		portsel = ppd->port == 1 ? portsel_port1 : portsel_port2;
+		etm = atetm_1port;
+	} else {
+		portsel = portsel_2port;
+		etm = atetm_2port;
+	}
+
+	if (qib_r_grab(dd) < 0 ||
+		qib_r_shift(dd, BISTEN_ETM, LEN_ETM, reset_atetm, NULL) < 0 ||
+		qib_r_update(dd, BISTEN_ETM) < 0 ||
+		qib_r_shift(dd, BISTEN_AT, LEN_AT, reset_at, NULL) < 0 ||
+		qib_r_update(dd, BISTEN_AT) < 0 ||
+		qib_r_shift(dd, BISTEN_PORT_SEL, LEN_PORT_SEL,
+			    portsel, NULL) < 0 ||
+		qib_r_update(dd, BISTEN_PORT_SEL) < 0 ||
+		qib_r_shift(dd, BISTEN_AT, LEN_AT, at, NULL) < 0 ||
+		qib_r_update(dd, BISTEN_AT) < 0 ||
+		qib_r_shift(dd, BISTEN_ETM, LEN_ETM, etm, NULL) < 0 ||
+		qib_r_update(dd, BISTEN_ETM) < 0)
+		qib_dev_err(dd, "Failed IB link recovery setup\n");
+}
+
+static void check_7322_rxe_status(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u64 fmask;
+
+	if (dd->cspec->recovery_ports_initted != 1)
+		return; /* rest doesn't apply to dualport */
+	qib_write_kreg(dd, kr_control, dd->control |
+		       SYM_MASK(Control, FreezeMode));
+	(void)qib_read_kreg64(dd, kr_scratch);
+	udelay(3); /* ibcreset asserted 400ns, be sure that's over */
+	fmask = qib_read_kreg64(dd, kr_act_fmask);
+	if (!fmask) {
+		/*
+		 * require a powercycle before we'll work again, and make
+		 * sure we get no more interrupts, and don't turn off
+		 * freeze.
+		 */
+		ppd->dd->cspec->stay_in_freeze = 1;
+		qib_7322_set_intr_state(ppd->dd, 0);
+		qib_write_kreg(dd, kr_fmask, 0ULL);
+		qib_dev_err(dd, "HCA unusable until powercycled\n");
+		return; /* eventually reset */
+	}
+
+	qib_write_kreg(ppd->dd, kr_hwerrclear,
+	    SYM_MASK(HwErrClear, IBSerdesPClkNotDetectClear_1));
+
+	/* don't do the full clear_freeze(), not needed for this */
+	qib_write_kreg(dd, kr_control, dd->control);
+	qib_read_kreg32(dd, kr_scratch);
+	/* take IBC out of reset */
+	if (ppd->link_speed_supported) {
+		ppd->cpspec->ibcctrl_a &=
+			~SYM_MASK(IBCCtrlA_0, IBStatIntReductionEn);
+		qib_write_kreg_port(ppd, krp_ibcctrl_a,
+				    ppd->cpspec->ibcctrl_a);
+		qib_read_kreg32(dd, kr_scratch);
+		if (ppd->lflags & QIBL_IB_LINK_DISABLED)
+			qib_set_ib_7322_lstate(ppd, 0,
+				QLOGIC_IB_IBCC_LINKINITCMD_DISABLE);
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_init.c b/drivers/infiniband/hw/qib/qib_init.c
new file mode 100644
index 00000000000..8d3c78ddc90
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_init.c
@@ -0,0 +1,1857 @@
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+#include <linux/delay.h>
+#include <linux/idr.h>
+#include <linux/module.h>
+#include <linux/printk.h>
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+#include <linux/dca.h>
+#endif
+
+#include "qib.h"
+#include "qib_common.h"
+#include "qib_mad.h"
+#ifdef CONFIG_DEBUG_FS
+#include "qib_debugfs.h"
+#include "qib_verbs.h"
+#endif
+
+#undef pr_fmt
+#define pr_fmt(fmt) QIB_DRV_NAME ": " fmt
+
+/*
+ * min buffers we want to have per context, after driver
+ */
+#define QIB_MIN_USER_CTXT_BUFCNT 7
+
+#define QLOGIC_IB_R_SOFTWARE_MASK 0xFF
+#define QLOGIC_IB_R_SOFTWARE_SHIFT 24
+#define QLOGIC_IB_R_EMULATOR_MASK (1ULL<<62)
+
+/*
+ * Number of ctxts we are configured to use (to allow for more pio
+ * buffers per ctxt, etc.)  Zero means use chip value.
+ */
+ushort qib_cfgctxts;
+module_param_named(cfgctxts, qib_cfgctxts, ushort, S_IRUGO);
+MODULE_PARM_DESC(cfgctxts, "Set max number of contexts to use");
+
+unsigned qib_numa_aware;
+module_param_named(numa_aware, qib_numa_aware, uint, S_IRUGO);
+MODULE_PARM_DESC(numa_aware,
+	"0 -> PSM allocation close to HCA, 1 -> PSM allocation local to process");
+
+/*
+ * If set, do not write to any regs if avoidable, hack to allow
+ * check for deranged default register values.
+ */
+ushort qib_mini_init;
+module_param_named(mini_init, qib_mini_init, ushort, S_IRUGO);
+MODULE_PARM_DESC(mini_init, "If set, do minimal diag init");
+
+unsigned qib_n_krcv_queues;
+module_param_named(krcvqs, qib_n_krcv_queues, uint, S_IRUGO);
+MODULE_PARM_DESC(krcvqs, "number of kernel receive queues per IB port");
+
+unsigned qib_cc_table_size;
+module_param_named(cc_table_size, qib_cc_table_size, uint, S_IRUGO);
+MODULE_PARM_DESC(cc_table_size, "Congestion control table entries 0 (CCA disabled - default), min = 128, max = 1984");
+/*
+ * qib_wc_pat parameter:
+ *      0 is WC via MTRR
+ *      1 is WC via PAT
+ *      If PAT initialization fails, code reverts back to MTRR
+ */
+unsigned qib_wc_pat = 1; /* default (1) is to use PAT, not MTRR */
+module_param_named(wc_pat, qib_wc_pat, uint, S_IRUGO);
+MODULE_PARM_DESC(wc_pat, "enable write-combining via PAT mechanism");
+
+static void verify_interrupt(unsigned long);
+
+static struct idr qib_unit_table;
+u32 qib_cpulist_count;
+unsigned long *qib_cpulist;
+
+/* set number of contexts we'll actually use */
+void qib_set_ctxtcnt(struct qib_devdata *dd)
+{
+	if (!qib_cfgctxts) {
+		dd->cfgctxts = dd->first_user_ctxt + num_online_cpus();
+		if (dd->cfgctxts > dd->ctxtcnt)
+			dd->cfgctxts = dd->ctxtcnt;
+	} else if (qib_cfgctxts < dd->num_pports)
+		dd->cfgctxts = dd->ctxtcnt;
+	else if (qib_cfgctxts <= dd->ctxtcnt)
+		dd->cfgctxts = qib_cfgctxts;
+	else
+		dd->cfgctxts = dd->ctxtcnt;
+	dd->freectxts = (dd->first_user_ctxt > dd->cfgctxts) ? 0 :
+		dd->cfgctxts - dd->first_user_ctxt;
+}
+
+/*
+ * Common code for creating the receive context array.
+ */
+int qib_create_ctxts(struct qib_devdata *dd)
+{
+	unsigned i;
+	int local_node_id = pcibus_to_node(dd->pcidev->bus);
+
+	if (local_node_id < 0)
+		local_node_id = numa_node_id();
+	dd->assigned_node_id = local_node_id;
+
+	/*
+	 * Allocate full ctxtcnt array, rather than just cfgctxts, because
+	 * cleanup iterates across all possible ctxts.
+	 */
+	dd->rcd = kzalloc(sizeof(*dd->rcd) * dd->ctxtcnt, GFP_KERNEL);
+	if (!dd->rcd) {
+		qib_dev_err(dd,
+			"Unable to allocate ctxtdata array, failing\n");
+		return -ENOMEM;
+	}
+
+	/* create (one or more) kctxt */
+	for (i = 0; i < dd->first_user_ctxt; ++i) {
+		struct qib_pportdata *ppd;
+		struct qib_ctxtdata *rcd;
+
+		if (dd->skip_kctxt_mask & (1 << i))
+			continue;
+
+		ppd = dd->pport + (i % dd->num_pports);
+
+		rcd = qib_create_ctxtdata(ppd, i, dd->assigned_node_id);
+		if (!rcd) {
+			qib_dev_err(dd,
+				"Unable to allocate ctxtdata for Kernel ctxt, failing\n");
+			kfree(dd->rcd);
+			dd->rcd = NULL;
+			return -ENOMEM;
+		}
+		rcd->pkeys[0] = QIB_DEFAULT_P_KEY;
+		rcd->seq_cnt = 1;
+	}
+	return 0;
+}
+
+/*
+ * Common code for user and kernel context setup.
+ */
+struct qib_ctxtdata *qib_create_ctxtdata(struct qib_pportdata *ppd, u32 ctxt,
+	int node_id)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+
+	rcd = kzalloc_node(sizeof(*rcd), GFP_KERNEL, node_id);
+	if (rcd) {
+		INIT_LIST_HEAD(&rcd->qp_wait_list);
+		rcd->node_id = node_id;
+		rcd->ppd = ppd;
+		rcd->dd = dd;
+		rcd->cnt = 1;
+		rcd->ctxt = ctxt;
+		dd->rcd[ctxt] = rcd;
+#ifdef CONFIG_DEBUG_FS
+		if (ctxt < dd->first_user_ctxt) { /* N/A for PSM contexts */
+			rcd->opstats = kzalloc_node(sizeof(*rcd->opstats),
+				GFP_KERNEL, node_id);
+			if (!rcd->opstats) {
+				kfree(rcd);
+				qib_dev_err(dd,
+					"Unable to allocate per ctxt stats buffer\n");
+				return NULL;
+			}
+		}
+#endif
+		dd->f_init_ctxt(rcd);
+
+		/*
+		 * To avoid wasting a lot of memory, we allocate 32KB chunks
+		 * of physically contiguous memory, advance through it until
+		 * used up and then allocate more.  Of course, we need
+		 * memory to store those extra pointers, now.  32KB seems to
+		 * be the most that is "safe" under memory pressure
+		 * (creating large files and then copying them over
+		 * NFS while doing lots of MPI jobs).  The OOM killer can
+		 * get invoked, even though we say we can sleep and this can
+		 * cause significant system problems....
+		 */
+		rcd->rcvegrbuf_size = 0x8000;
+		rcd->rcvegrbufs_perchunk =
+			rcd->rcvegrbuf_size / dd->rcvegrbufsize;
+		rcd->rcvegrbuf_chunks = (rcd->rcvegrcnt +
+			rcd->rcvegrbufs_perchunk - 1) /
+			rcd->rcvegrbufs_perchunk;
+		BUG_ON(!is_power_of_2(rcd->rcvegrbufs_perchunk));
+		rcd->rcvegrbufs_perchunk_shift =
+			ilog2(rcd->rcvegrbufs_perchunk);
+	}
+	return rcd;
+}
+
+/*
+ * Common code for initializing the physical port structure.
+ */
+int qib_init_pportdata(struct qib_pportdata *ppd, struct qib_devdata *dd,
+			u8 hw_pidx, u8 port)
+{
+	int size;
+	ppd->dd = dd;
+	ppd->hw_pidx = hw_pidx;
+	ppd->port = port; /* IB port number, not index */
+
+	spin_lock_init(&ppd->sdma_lock);
+	spin_lock_init(&ppd->lflags_lock);
+	spin_lock_init(&ppd->cc_shadow_lock);
+	init_waitqueue_head(&ppd->state_wait);
+
+	init_timer(&ppd->symerr_clear_timer);
+	ppd->symerr_clear_timer.function = qib_clear_symerror_on_linkup;
+	ppd->symerr_clear_timer.data = (unsigned long)ppd;
+
+	ppd->qib_wq = NULL;
+	ppd->ibport_data.pmastats =
+		alloc_percpu(struct qib_pma_counters);
+	if (!ppd->ibport_data.pmastats)
+		return -ENOMEM;
+
+	if (qib_cc_table_size < IB_CCT_MIN_ENTRIES)
+		goto bail;
+
+	ppd->cc_supported_table_entries = min(max_t(int, qib_cc_table_size,
+		IB_CCT_MIN_ENTRIES), IB_CCT_ENTRIES*IB_CC_TABLE_CAP_DEFAULT);
+
+	ppd->cc_max_table_entries =
+		ppd->cc_supported_table_entries/IB_CCT_ENTRIES;
+
+	size = IB_CC_TABLE_CAP_DEFAULT * sizeof(struct ib_cc_table_entry)
+		* IB_CCT_ENTRIES;
+	ppd->ccti_entries = kzalloc(size, GFP_KERNEL);
+	if (!ppd->ccti_entries) {
+		qib_dev_err(dd,
+		  "failed to allocate congestion control table for port %d!\n",
+		  port);
+		goto bail;
+	}
+
+	size = IB_CC_CCS_ENTRIES * sizeof(struct ib_cc_congestion_entry);
+	ppd->congestion_entries = kzalloc(size, GFP_KERNEL);
+	if (!ppd->congestion_entries) {
+		qib_dev_err(dd,
+		 "failed to allocate congestion setting list for port %d!\n",
+		 port);
+		goto bail_1;
+	}
+
+	size = sizeof(struct cc_table_shadow);
+	ppd->ccti_entries_shadow = kzalloc(size, GFP_KERNEL);
+	if (!ppd->ccti_entries_shadow) {
+		qib_dev_err(dd,
+		 "failed to allocate shadow ccti list for port %d!\n",
+		 port);
+		goto bail_2;
+	}
+
+	size = sizeof(struct ib_cc_congestion_setting_attr);
+	ppd->congestion_entries_shadow = kzalloc(size, GFP_KERNEL);
+	if (!ppd->congestion_entries_shadow) {
+		qib_dev_err(dd,
+		 "failed to allocate shadow congestion setting list for port %d!\n",
+		 port);
+		goto bail_3;
+	}
+
+	return 0;
+
+bail_3:
+	kfree(ppd->ccti_entries_shadow);
+	ppd->ccti_entries_shadow = NULL;
+bail_2:
+	kfree(ppd->congestion_entries);
+	ppd->congestion_entries = NULL;
+bail_1:
+	kfree(ppd->ccti_entries);
+	ppd->ccti_entries = NULL;
+bail:
+	/* User is intentionally disabling the congestion control agent */
+	if (!qib_cc_table_size)
+		return 0;
+
+	if (qib_cc_table_size < IB_CCT_MIN_ENTRIES) {
+		qib_cc_table_size = 0;
+		qib_dev_err(dd,
+		 "Congestion Control table size %d less than minimum %d for port %d\n",
+		 qib_cc_table_size, IB_CCT_MIN_ENTRIES, port);
+	}
+
+	qib_dev_err(dd, "Congestion Control Agent disabled for port %d\n",
+		port);
+	return 0;
+}
+
+static int init_pioavailregs(struct qib_devdata *dd)
+{
+	int ret, pidx;
+	u64 *status_page;
+
+	dd->pioavailregs_dma = dma_alloc_coherent(
+		&dd->pcidev->dev, PAGE_SIZE, &dd->pioavailregs_phys,
+		GFP_KERNEL);
+	if (!dd->pioavailregs_dma) {
+		qib_dev_err(dd,
+			"failed to allocate PIOavail reg area in memory\n");
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	/*
+	 * We really want L2 cache aligned, but for current CPUs of
+	 * interest, they are the same.
+	 */
+	status_page = (u64 *)
+		((char *) dd->pioavailregs_dma +
+		 ((2 * L1_CACHE_BYTES +
+		   dd->pioavregs * sizeof(u64)) & ~L1_CACHE_BYTES));
+	/* device status comes first, for backwards compatibility */
+	dd->devstatusp = status_page;
+	*status_page++ = 0;
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		dd->pport[pidx].statusp = status_page;
+		*status_page++ = 0;
+	}
+
+	/*
+	 * Setup buffer to hold freeze and other messages, accessible to
+	 * apps, following statusp.  This is per-unit, not per port.
+	 */
+	dd->freezemsg = (char *) status_page;
+	*dd->freezemsg = 0;
+	/* length of msg buffer is "whatever is left" */
+	ret = (char *) status_page - (char *) dd->pioavailregs_dma;
+	dd->freezelen = PAGE_SIZE - ret;
+
+	ret = 0;
+
+done:
+	return ret;
+}
+
+/**
+ * init_shadow_tids - allocate the shadow TID array
+ * @dd: the qlogic_ib device
+ *
+ * allocate the shadow TID array, so we can qib_munlock previous
+ * entries.  It may make more sense to move the pageshadow to the
+ * ctxt data structure, so we only allocate memory for ctxts actually
+ * in use, since we at 8k per ctxt, now.
+ * We don't want failures here to prevent use of the driver/chip,
+ * so no return value.
+ */
+static void init_shadow_tids(struct qib_devdata *dd)
+{
+	struct page **pages;
+	dma_addr_t *addrs;
+
+	pages = vzalloc(dd->cfgctxts * dd->rcvtidcnt * sizeof(struct page *));
+	if (!pages) {
+		qib_dev_err(dd,
+			"failed to allocate shadow page * array, no expected sends!\n");
+		goto bail;
+	}
+
+	addrs = vzalloc(dd->cfgctxts * dd->rcvtidcnt * sizeof(dma_addr_t));
+	if (!addrs) {
+		qib_dev_err(dd,
+			"failed to allocate shadow dma handle array, no expected sends!\n");
+		goto bail_free;
+	}
+
+	dd->pageshadow = pages;
+	dd->physshadow = addrs;
+	return;
+
+bail_free:
+	vfree(pages);
+bail:
+	dd->pageshadow = NULL;
+}
+
+/*
+ * Do initialization for device that is only needed on
+ * first detect, not on resets.
+ */
+static int loadtime_init(struct qib_devdata *dd)
+{
+	int ret = 0;
+
+	if (((dd->revision >> QLOGIC_IB_R_SOFTWARE_SHIFT) &
+	     QLOGIC_IB_R_SOFTWARE_MASK) != QIB_CHIP_SWVERSION) {
+		qib_dev_err(dd,
+			"Driver only handles version %d, chip swversion is %d (%llx), failng\n",
+			QIB_CHIP_SWVERSION,
+			(int)(dd->revision >>
+				QLOGIC_IB_R_SOFTWARE_SHIFT) &
+				QLOGIC_IB_R_SOFTWARE_MASK,
+			(unsigned long long) dd->revision);
+		ret = -ENOSYS;
+		goto done;
+	}
+
+	if (dd->revision & QLOGIC_IB_R_EMULATOR_MASK)
+		qib_devinfo(dd->pcidev, "%s", dd->boardversion);
+
+	spin_lock_init(&dd->pioavail_lock);
+	spin_lock_init(&dd->sendctrl_lock);
+	spin_lock_init(&dd->uctxt_lock);
+	spin_lock_init(&dd->qib_diag_trans_lock);
+	spin_lock_init(&dd->eep_st_lock);
+	mutex_init(&dd->eep_lock);
+
+	if (qib_mini_init)
+		goto done;
+
+	ret = init_pioavailregs(dd);
+	init_shadow_tids(dd);
+
+	qib_get_eeprom_info(dd);
+
+	/* setup time (don't start yet) to verify we got interrupt */
+	init_timer(&dd->intrchk_timer);
+	dd->intrchk_timer.function = verify_interrupt;
+	dd->intrchk_timer.data = (unsigned long) dd;
+
+	ret = qib_cq_init(dd);
+done:
+	return ret;
+}
+
+/**
+ * init_after_reset - re-initialize after a reset
+ * @dd: the qlogic_ib device
+ *
+ * sanity check at least some of the values after reset, and
+ * ensure no receive or transmit (explicitly, in case reset
+ * failed
+ */
+static int init_after_reset(struct qib_devdata *dd)
+{
+	int i;
+
+	/*
+	 * Ensure chip does no sends or receives, tail updates, or
+	 * pioavail updates while we re-initialize.  This is mostly
+	 * for the driver data structures, not chip registers.
+	 */
+	for (i = 0; i < dd->num_pports; ++i) {
+		/*
+		 * ctxt == -1 means "all contexts". Only really safe for
+		 * _dis_abling things, as here.
+		 */
+		dd->f_rcvctrl(dd->pport + i, QIB_RCVCTRL_CTXT_DIS |
+				  QIB_RCVCTRL_INTRAVAIL_DIS |
+				  QIB_RCVCTRL_TAILUPD_DIS, -1);
+		/* Redundant across ports for some, but no big deal.  */
+		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_DIS |
+			QIB_SENDCTRL_AVAIL_DIS);
+	}
+
+	return 0;
+}
+
+static void enable_chip(struct qib_devdata *dd)
+{
+	u64 rcvmask;
+	int i;
+
+	/*
+	 * Enable PIO send, and update of PIOavail regs to memory.
+	 */
+	for (i = 0; i < dd->num_pports; ++i)
+		dd->f_sendctrl(dd->pport + i, QIB_SENDCTRL_SEND_ENB |
+			QIB_SENDCTRL_AVAIL_ENB);
+	/*
+	 * Enable kernel ctxts' receive and receive interrupt.
+	 * Other ctxts done as user opens and inits them.
+	 */
+	rcvmask = QIB_RCVCTRL_CTXT_ENB | QIB_RCVCTRL_INTRAVAIL_ENB;
+	rcvmask |= (dd->flags & QIB_NODMA_RTAIL) ?
+		  QIB_RCVCTRL_TAILUPD_DIS : QIB_RCVCTRL_TAILUPD_ENB;
+	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
+		struct qib_ctxtdata *rcd = dd->rcd[i];
+
+		if (rcd)
+			dd->f_rcvctrl(rcd->ppd, rcvmask, i);
+	}
+}
+
+static void verify_interrupt(unsigned long opaque)
+{
+	struct qib_devdata *dd = (struct qib_devdata *) opaque;
+	u64 int_counter;
+
+	if (!dd)
+		return; /* being torn down */
+
+	/*
+	 * If we don't have a lid or any interrupts, let the user know and
+	 * don't bother checking again.
+	 */
+	int_counter = qib_int_counter(dd) - dd->z_int_counter;
+	if (int_counter == 0) {
+		if (!dd->f_intr_fallback(dd))
+			dev_err(&dd->pcidev->dev,
+				"No interrupts detected, not usable.\n");
+		else /* re-arm the timer to see if fallback works */
+			mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
+	}
+}
+
+static void init_piobuf_state(struct qib_devdata *dd)
+{
+	int i, pidx;
+	u32 uctxts;
+
+	/*
+	 * Ensure all buffers are free, and fifos empty.  Buffers
+	 * are common, so only do once for port 0.
+	 *
+	 * After enable and qib_chg_pioavailkernel so we can safely
+	 * enable pioavail updates and PIOENABLE.  After this, packets
+	 * are ready and able to go out.
+	 */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_ALL);
+	for (pidx = 0; pidx < dd->num_pports; ++pidx)
+		dd->f_sendctrl(dd->pport + pidx, QIB_SENDCTRL_FLUSH);
+
+	/*
+	 * If not all sendbufs are used, add the one to each of the lower
+	 * numbered contexts.  pbufsctxt and lastctxt_piobuf are
+	 * calculated in chip-specific code because it may cause some
+	 * chip-specific adjustments to be made.
+	 */
+	uctxts = dd->cfgctxts - dd->first_user_ctxt;
+	dd->ctxts_extrabuf = dd->pbufsctxt ?
+		dd->lastctxt_piobuf - (dd->pbufsctxt * uctxts) : 0;
+
+	/*
+	 * Set up the shadow copies of the piobufavail registers,
+	 * which we compare against the chip registers for now, and
+	 * the in memory DMA'ed copies of the registers.
+	 * By now pioavail updates to memory should have occurred, so
+	 * copy them into our working/shadow registers; this is in
+	 * case something went wrong with abort, but mostly to get the
+	 * initial values of the generation bit correct.
+	 */
+	for (i = 0; i < dd->pioavregs; i++) {
+		__le64 tmp;
+
+		tmp = dd->pioavailregs_dma[i];
+		/*
+		 * Don't need to worry about pioavailkernel here
+		 * because we will call qib_chg_pioavailkernel() later
+		 * in initialization, to busy out buffers as needed.
+		 */
+		dd->pioavailshadow[i] = le64_to_cpu(tmp);
+	}
+	while (i < ARRAY_SIZE(dd->pioavailshadow))
+		dd->pioavailshadow[i++] = 0; /* for debugging sanity */
+
+	/* after pioavailshadow is setup */
+	qib_chg_pioavailkernel(dd, 0, dd->piobcnt2k + dd->piobcnt4k,
+			       TXCHK_CHG_TYPE_KERN, NULL);
+	dd->f_initvl15_bufs(dd);
+}
+
+/**
+ * qib_create_workqueues - create per port workqueues
+ * @dd: the qlogic_ib device
+ */
+static int qib_create_workqueues(struct qib_devdata *dd)
+{
+	int pidx;
+	struct qib_pportdata *ppd;
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (!ppd->qib_wq) {
+			char wq_name[8]; /* 3 + 2 + 1 + 1 + 1 */
+			snprintf(wq_name, sizeof(wq_name), "qib%d_%d",
+				dd->unit, pidx);
+			ppd->qib_wq =
+				create_singlethread_workqueue(wq_name);
+			if (!ppd->qib_wq)
+				goto wq_error;
+		}
+	}
+	return 0;
+wq_error:
+	pr_err("create_singlethread_workqueue failed for port %d\n",
+		pidx + 1);
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (ppd->qib_wq) {
+			destroy_workqueue(ppd->qib_wq);
+			ppd->qib_wq = NULL;
+		}
+	}
+	return -ENOMEM;
+}
+
+static void qib_free_pportdata(struct qib_pportdata *ppd)
+{
+	free_percpu(ppd->ibport_data.pmastats);
+	ppd->ibport_data.pmastats = NULL;
+}
+
+/**
+ * qib_init - do the actual initialization sequence on the chip
+ * @dd: the qlogic_ib device
+ * @reinit: reinitializing, so don't allocate new memory
+ *
+ * Do the actual initialization sequence on the chip.  This is done
+ * both from the init routine called from the PCI infrastructure, and
+ * when we reset the chip, or detect that it was reset internally,
+ * or it's administratively re-enabled.
+ *
+ * Memory allocation here and in called routines is only done in
+ * the first case (reinit == 0).  We have to be careful, because even
+ * without memory allocation, we need to re-write all the chip registers
+ * TIDs, etc. after the reset or enable has completed.
+ */
+int qib_init(struct qib_devdata *dd, int reinit)
+{
+	int ret = 0, pidx, lastfail = 0;
+	u32 portok = 0;
+	unsigned i;
+	struct qib_ctxtdata *rcd;
+	struct qib_pportdata *ppd;
+	unsigned long flags;
+
+	/* Set linkstate to unknown, so we can watch for a transition. */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~(QIBL_LINKACTIVE | QIBL_LINKARMED |
+				 QIBL_LINKDOWN | QIBL_LINKINIT |
+				 QIBL_LINKV);
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+	}
+
+	if (reinit)
+		ret = init_after_reset(dd);
+	else
+		ret = loadtime_init(dd);
+	if (ret)
+		goto done;
+
+	/* Bypass most chip-init, to get to device creation */
+	if (qib_mini_init)
+		return 0;
+
+	ret = dd->f_late_initreg(dd);
+	if (ret)
+		goto done;
+
+	/* dd->rcd can be NULL if early init failed */
+	for (i = 0; dd->rcd && i < dd->first_user_ctxt; ++i) {
+		/*
+		 * Set up the (kernel) rcvhdr queue and egr TIDs.  If doing
+		 * re-init, the simplest way to handle this is to free
+		 * existing, and re-allocate.
+		 * Need to re-create rest of ctxt 0 ctxtdata as well.
+		 */
+		rcd = dd->rcd[i];
+		if (!rcd)
+			continue;
+
+		lastfail = qib_create_rcvhdrq(dd, rcd);
+		if (!lastfail)
+			lastfail = qib_setup_eagerbufs(rcd);
+		if (lastfail) {
+			qib_dev_err(dd,
+				"failed to allocate kernel ctxt's rcvhdrq and/or egr bufs\n");
+			continue;
+		}
+	}
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		int mtu;
+		if (lastfail)
+			ret = lastfail;
+		ppd = dd->pport + pidx;
+		mtu = ib_mtu_enum_to_int(qib_ibmtu);
+		if (mtu == -1) {
+			mtu = QIB_DEFAULT_MTU;
+			qib_ibmtu = 0; /* don't leave invalid value */
+		}
+		/* set max we can ever have for this driver load */
+		ppd->init_ibmaxlen = min(mtu > 2048 ?
+					 dd->piosize4k : dd->piosize2k,
+					 dd->rcvegrbufsize +
+					 (dd->rcvhdrentsize << 2));
+		/*
+		 * Have to initialize ibmaxlen, but this will normally
+		 * change immediately in qib_set_mtu().
+		 */
+		ppd->ibmaxlen = ppd->init_ibmaxlen;
+		qib_set_mtu(ppd, mtu);
+
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_IB_LINK_DISABLED;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+
+		lastfail = dd->f_bringup_serdes(ppd);
+		if (lastfail) {
+			qib_devinfo(dd->pcidev,
+				 "Failed to bringup IB port %u\n", ppd->port);
+			lastfail = -ENETDOWN;
+			continue;
+		}
+
+		portok++;
+	}
+
+	if (!portok) {
+		/* none of the ports initialized */
+		if (!ret && lastfail)
+			ret = lastfail;
+		else if (!ret)
+			ret = -ENETDOWN;
+		/* but continue on, so we can debug cause */
+	}
+
+	enable_chip(dd);
+
+	init_piobuf_state(dd);
+
+done:
+	if (!ret) {
+		/* chip is OK for user apps; mark it as initialized */
+		for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+			ppd = dd->pport + pidx;
+			/*
+			 * Set status even if port serdes is not initialized
+			 * so that diags will work.
+			 */
+			*ppd->statusp |= QIB_STATUS_CHIP_PRESENT |
+				QIB_STATUS_INITTED;
+			if (!ppd->link_speed_enabled)
+				continue;
+			if (dd->flags & QIB_HAS_SEND_DMA)
+				ret = qib_setup_sdma(ppd);
+			init_timer(&ppd->hol_timer);
+			ppd->hol_timer.function = qib_hol_event;
+			ppd->hol_timer.data = (unsigned long)ppd;
+			ppd->hol_state = QIB_HOL_UP;
+		}
+
+		/* now we can enable all interrupts from the chip */
+		dd->f_set_intr_state(dd, 1);
+
+		/*
+		 * Setup to verify we get an interrupt, and fallback
+		 * to an alternate if necessary and possible.
+		 */
+		mod_timer(&dd->intrchk_timer, jiffies + HZ/2);
+		/* start stats retrieval timer */
+		mod_timer(&dd->stats_timer, jiffies + HZ * ACTIVITY_TIMER);
+	}
+
+	/* if ret is non-zero, we probably should do some cleanup here... */
+	return ret;
+}
+
+/*
+ * These next two routines are placeholders in case we don't have per-arch
+ * code for controlling write combining.  If explicit control of write
+ * combining is not available, performance will probably be awful.
+ */
+
+int __attribute__((weak)) qib_enable_wc(struct qib_devdata *dd)
+{
+	return -EOPNOTSUPP;
+}
+
+void __attribute__((weak)) qib_disable_wc(struct qib_devdata *dd)
+{
+}
+
+static inline struct qib_devdata *__qib_lookup(int unit)
+{
+	return idr_find(&qib_unit_table, unit);
+}
+
+struct qib_devdata *qib_lookup(int unit)
+{
+	struct qib_devdata *dd;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	dd = __qib_lookup(unit);
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+	return dd;
+}
+
+/*
+ * Stop the timers during unit shutdown, or after an error late
+ * in initialization.
+ */
+static void qib_stop_timers(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int pidx;
+
+	if (dd->stats_timer.data) {
+		del_timer_sync(&dd->stats_timer);
+		dd->stats_timer.data = 0;
+	}
+	if (dd->intrchk_timer.data) {
+		del_timer_sync(&dd->intrchk_timer);
+		dd->intrchk_timer.data = 0;
+	}
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		if (ppd->hol_timer.data)
+			del_timer_sync(&ppd->hol_timer);
+		if (ppd->led_override_timer.data) {
+			del_timer_sync(&ppd->led_override_timer);
+			atomic_set(&ppd->led_override_timer_active, 0);
+		}
+		if (ppd->symerr_clear_timer.data)
+			del_timer_sync(&ppd->symerr_clear_timer);
+	}
+}
+
+/**
+ * qib_shutdown_device - shut down a device
+ * @dd: the qlogic_ib device
+ *
+ * This is called to make the device quiet when we are about to
+ * unload the driver, and also when the device is administratively
+ * disabled.   It does not free any data structures.
+ * Everything it does has to be setup again by qib_init(dd, 1)
+ */
+static void qib_shutdown_device(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	unsigned pidx;
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+
+		spin_lock_irq(&ppd->lflags_lock);
+		ppd->lflags &= ~(QIBL_LINKDOWN | QIBL_LINKINIT |
+				 QIBL_LINKARMED | QIBL_LINKACTIVE |
+				 QIBL_LINKV);
+		spin_unlock_irq(&ppd->lflags_lock);
+		*ppd->statusp &= ~(QIB_STATUS_IB_CONF | QIB_STATUS_IB_READY);
+	}
+	dd->flags &= ~QIB_INITTED;
+
+	/* mask interrupts, but not errors */
+	dd->f_set_intr_state(dd, 0);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		dd->f_rcvctrl(ppd, QIB_RCVCTRL_TAILUPD_DIS |
+				   QIB_RCVCTRL_CTXT_DIS |
+				   QIB_RCVCTRL_INTRAVAIL_DIS |
+				   QIB_RCVCTRL_PKEY_ENB, -1);
+		/*
+		 * Gracefully stop all sends allowing any in progress to
+		 * trickle out first.
+		 */
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_CLEAR);
+	}
+
+	/*
+	 * Enough for anything that's going to trickle out to have actually
+	 * done so.
+	 */
+	udelay(20);
+
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		ppd = dd->pport + pidx;
+		dd->f_setextled(ppd, 0); /* make sure LEDs are off */
+
+		if (dd->flags & QIB_HAS_SEND_DMA)
+			qib_teardown_sdma(ppd);
+
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_AVAIL_DIS |
+				    QIB_SENDCTRL_SEND_DIS);
+		/*
+		 * Clear SerdesEnable.
+		 * We can't count on interrupts since we are stopping.
+		 */
+		dd->f_quiet_serdes(ppd);
+
+		if (ppd->qib_wq) {
+			destroy_workqueue(ppd->qib_wq);
+			ppd->qib_wq = NULL;
+		}
+		qib_free_pportdata(ppd);
+	}
+
+	qib_update_eeprom_log(dd);
+}
+
+/**
+ * qib_free_ctxtdata - free a context's allocated data
+ * @dd: the qlogic_ib device
+ * @rcd: the ctxtdata structure
+ *
+ * free up any allocated data for a context
+ * This should not touch anything that would affect a simultaneous
+ * re-allocation of context data, because it is called after qib_mutex
+ * is released (and can be called from reinit as well).
+ * It should never change any chip state, or global driver state.
+ */
+void qib_free_ctxtdata(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
+{
+	if (!rcd)
+		return;
+
+	if (rcd->rcvhdrq) {
+		dma_free_coherent(&dd->pcidev->dev, rcd->rcvhdrq_size,
+				  rcd->rcvhdrq, rcd->rcvhdrq_phys);
+		rcd->rcvhdrq = NULL;
+		if (rcd->rcvhdrtail_kvaddr) {
+			dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+					  rcd->rcvhdrtail_kvaddr,
+					  rcd->rcvhdrqtailaddr_phys);
+			rcd->rcvhdrtail_kvaddr = NULL;
+		}
+	}
+	if (rcd->rcvegrbuf) {
+		unsigned e;
+
+		for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
+			void *base = rcd->rcvegrbuf[e];
+			size_t size = rcd->rcvegrbuf_size;
+
+			dma_free_coherent(&dd->pcidev->dev, size,
+					  base, rcd->rcvegrbuf_phys[e]);
+		}
+		kfree(rcd->rcvegrbuf);
+		rcd->rcvegrbuf = NULL;
+		kfree(rcd->rcvegrbuf_phys);
+		rcd->rcvegrbuf_phys = NULL;
+		rcd->rcvegrbuf_chunks = 0;
+	}
+
+	kfree(rcd->tid_pg_list);
+	vfree(rcd->user_event_mask);
+	vfree(rcd->subctxt_uregbase);
+	vfree(rcd->subctxt_rcvegrbuf);
+	vfree(rcd->subctxt_rcvhdr_base);
+#ifdef CONFIG_DEBUG_FS
+	kfree(rcd->opstats);
+	rcd->opstats = NULL;
+#endif
+	kfree(rcd);
+}
+
+/*
+ * Perform a PIO buffer bandwidth write test, to verify proper system
+ * configuration.  Even when all the setup calls work, occasionally
+ * BIOS or other issues can prevent write combining from working, or
+ * can cause other bandwidth problems to the chip.
+ *
+ * This test simply writes the same buffer over and over again, and
+ * measures close to the peak bandwidth to the chip (not testing
+ * data bandwidth to the wire).   On chips that use an address-based
+ * trigger to send packets to the wire, this is easy.  On chips that
+ * use a count to trigger, we want to make sure that the packet doesn't
+ * go out on the wire, or trigger flow control checks.
+ */
+static void qib_verify_pioperf(struct qib_devdata *dd)
+{
+	u32 pbnum, cnt, lcnt;
+	u32 __iomem *piobuf;
+	u32 *addr;
+	u64 msecs, emsecs;
+
+	piobuf = dd->f_getsendbuf(dd->pport, 0ULL, &pbnum);
+	if (!piobuf) {
+		qib_devinfo(dd->pcidev,
+			 "No PIObufs for checking perf, skipping\n");
+		return;
+	}
+
+	/*
+	 * Enough to give us a reasonable test, less than piobuf size, and
+	 * likely multiple of store buffer length.
+	 */
+	cnt = 1024;
+
+	addr = vmalloc(cnt);
+	if (!addr) {
+		qib_devinfo(dd->pcidev,
+			 "Couldn't get memory for checking PIO perf,"
+			 " skipping\n");
+		goto done;
+	}
+
+	preempt_disable();  /* we want reasonably accurate elapsed time */
+	msecs = 1 + jiffies_to_msecs(jiffies);
+	for (lcnt = 0; lcnt < 10000U; lcnt++) {
+		/* wait until we cross msec boundary */
+		if (jiffies_to_msecs(jiffies) >= msecs)
+			break;
+		udelay(1);
+	}
+
+	dd->f_set_armlaunch(dd, 0);
+
+	/*
+	 * length 0, no dwords actually sent
+	 */
+	writeq(0, piobuf);
+	qib_flush_wc();
+
+	/*
+	 * This is only roughly accurate, since even with preempt we
+	 * still take interrupts that could take a while.   Running for
+	 * >= 5 msec seems to get us "close enough" to accurate values.
+	 */
+	msecs = jiffies_to_msecs(jiffies);
+	for (emsecs = lcnt = 0; emsecs <= 5UL; lcnt++) {
+		qib_pio_copy(piobuf + 64, addr, cnt >> 2);
+		emsecs = jiffies_to_msecs(jiffies) - msecs;
+	}
+
+	/* 1 GiB/sec, slightly over IB SDR line rate */
+	if (lcnt < (emsecs * 1024U))
+		qib_dev_err(dd,
+			    "Performance problem: bandwidth to PIO buffers is only %u MiB/sec\n",
+			    lcnt / (u32) emsecs);
+
+	preempt_enable();
+
+	vfree(addr);
+
+done:
+	/* disarm piobuf, so it's available again */
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(pbnum));
+	qib_sendbuf_done(dd, pbnum);
+	dd->f_set_armlaunch(dd, 1);
+}
+
+void qib_free_devdata(struct qib_devdata *dd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	idr_remove(&qib_unit_table, dd->unit);
+	list_del(&dd->list);
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+
+#ifdef CONFIG_DEBUG_FS
+	qib_dbg_ibdev_exit(&dd->verbs_dev);
+#endif
+	free_percpu(dd->int_counter);
+	ib_dealloc_device(&dd->verbs_dev.ibdev);
+}
+
+u64 qib_int_counter(struct qib_devdata *dd)
+{
+	int cpu;
+	u64 int_counter = 0;
+
+	for_each_possible_cpu(cpu)
+		int_counter += *per_cpu_ptr(dd->int_counter, cpu);
+	return int_counter;
+}
+
+u64 qib_sps_ints(void)
+{
+	unsigned long flags;
+	struct qib_devdata *dd;
+	u64 sps_ints = 0;
+
+	spin_lock_irqsave(&qib_devs_lock, flags);
+	list_for_each_entry(dd, &qib_dev_list, list) {
+		sps_ints += qib_int_counter(dd);
+	}
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+	return sps_ints;
+}
+
+/*
+ * Allocate our primary per-unit data structure.  Must be done via verbs
+ * allocator, because the verbs cleanup process both does cleanup and
+ * free of the data structure.
+ * "extra" is for chip-specific data.
+ *
+ * Use the idr mechanism to get a unit number for this unit.
+ */
+struct qib_devdata *qib_alloc_devdata(struct pci_dev *pdev, size_t extra)
+{
+	unsigned long flags;
+	struct qib_devdata *dd;
+	int ret;
+
+	dd = (struct qib_devdata *) ib_alloc_device(sizeof(*dd) + extra);
+	if (!dd)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&dd->list);
+
+	idr_preload(GFP_KERNEL);
+	spin_lock_irqsave(&qib_devs_lock, flags);
+
+	ret = idr_alloc(&qib_unit_table, dd, 0, 0, GFP_NOWAIT);
+	if (ret >= 0) {
+		dd->unit = ret;
+		list_add(&dd->list, &qib_dev_list);
+	}
+
+	spin_unlock_irqrestore(&qib_devs_lock, flags);
+	idr_preload_end();
+
+	if (ret < 0) {
+		qib_early_err(&pdev->dev,
+			      "Could not allocate unit ID: error %d\n", -ret);
+		goto bail;
+	}
+	dd->int_counter = alloc_percpu(u64);
+	if (!dd->int_counter) {
+		ret = -ENOMEM;
+		qib_early_err(&pdev->dev,
+			      "Could not allocate per-cpu int_counter\n");
+		goto bail;
+	}
+
+	if (!qib_cpulist_count) {
+		u32 count = num_online_cpus();
+		qib_cpulist = kzalloc(BITS_TO_LONGS(count) *
+				      sizeof(long), GFP_KERNEL);
+		if (qib_cpulist)
+			qib_cpulist_count = count;
+		else
+			qib_early_err(&pdev->dev,
+				"Could not alloc cpulist info, cpu affinity might be wrong\n");
+	}
+#ifdef CONFIG_DEBUG_FS
+	qib_dbg_ibdev_init(&dd->verbs_dev);
+#endif
+	return dd;
+bail:
+	if (!list_empty(&dd->list))
+		list_del_init(&dd->list);
+	ib_dealloc_device(&dd->verbs_dev.ibdev);
+	return ERR_PTR(ret);;
+}
+
+/*
+ * Called from freeze mode handlers, and from PCI error
+ * reporting code.  Should be paranoid about state of
+ * system and data structures.
+ */
+void qib_disable_after_error(struct qib_devdata *dd)
+{
+	if (dd->flags & QIB_INITTED) {
+		u32 pidx;
+
+		dd->flags &= ~QIB_INITTED;
+		if (dd->pport)
+			for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+				struct qib_pportdata *ppd;
+
+				ppd = dd->pport + pidx;
+				if (dd->flags & QIB_PRESENT) {
+					qib_set_linkstate(ppd,
+						QIB_IB_LINKDOWN_DISABLE);
+					dd->f_setextled(ppd, 0);
+				}
+				*ppd->statusp &= ~QIB_STATUS_IB_READY;
+			}
+	}
+
+	/*
+	 * Mark as having had an error for driver, and also
+	 * for /sys and status word mapped to user programs.
+	 * This marks unit as not usable, until reset.
+	 */
+	if (dd->devstatusp)
+		*dd->devstatusp |= QIB_STATUS_HWERROR;
+}
+
+static void qib_remove_one(struct pci_dev *);
+static int qib_init_one(struct pci_dev *, const struct pci_device_id *);
+
+#define DRIVER_LOAD_MSG "Intel " QIB_DRV_NAME " loaded: "
+#define PFX QIB_DRV_NAME ": "
+
+static DEFINE_PCI_DEVICE_TABLE(qib_pci_tbl) = {
+	{ PCI_DEVICE(PCI_VENDOR_ID_PATHSCALE, PCI_DEVICE_ID_QLOGIC_IB_6120) },
+	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7220) },
+	{ PCI_DEVICE(PCI_VENDOR_ID_QLOGIC, PCI_DEVICE_ID_QLOGIC_IB_7322) },
+	{ 0, }
+};
+
+MODULE_DEVICE_TABLE(pci, qib_pci_tbl);
+
+static struct pci_driver qib_driver = {
+	.name = QIB_DRV_NAME,
+	.probe = qib_init_one,
+	.remove = qib_remove_one,
+	.id_table = qib_pci_tbl,
+	.err_handler = &qib_pci_err_handler,
+};
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+
+static int qib_notify_dca(struct notifier_block *, unsigned long, void *);
+static struct notifier_block dca_notifier = {
+	.notifier_call  = qib_notify_dca,
+	.next           = NULL,
+	.priority       = 0
+};
+
+static int qib_notify_dca_device(struct device *device, void *data)
+{
+	struct qib_devdata *dd = dev_get_drvdata(device);
+	unsigned long event = *(unsigned long *)data;
+
+	return dd->f_notify_dca(dd, event);
+}
+
+static int qib_notify_dca(struct notifier_block *nb, unsigned long event,
+					  void *p)
+{
+	int rval;
+
+	rval = driver_for_each_device(&qib_driver.driver, NULL,
+				      &event, qib_notify_dca_device);
+	return rval ? NOTIFY_BAD : NOTIFY_DONE;
+}
+
+#endif
+
+/*
+ * Do all the generic driver unit- and chip-independent memory
+ * allocation and initialization.
+ */
+static int __init qib_ib_init(void)
+{
+	int ret;
+
+	ret = qib_dev_init();
+	if (ret)
+		goto bail;
+
+	/*
+	 * These must be called before the driver is registered with
+	 * the PCI subsystem.
+	 */
+	idr_init(&qib_unit_table);
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dca_register_notify(&dca_notifier);
+#endif
+#ifdef CONFIG_DEBUG_FS
+	qib_dbg_init();
+#endif
+	ret = pci_register_driver(&qib_driver);
+	if (ret < 0) {
+		pr_err("Unable to register driver: error %d\n", -ret);
+		goto bail_dev;
+	}
+
+	/* not fatal if it doesn't work */
+	if (qib_init_qibfs())
+		pr_err("Unable to register ipathfs\n");
+	goto bail; /* all OK */
+
+bail_dev:
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dca_unregister_notify(&dca_notifier);
+#endif
+#ifdef CONFIG_DEBUG_FS
+	qib_dbg_exit();
+#endif
+	idr_destroy(&qib_unit_table);
+	qib_dev_cleanup();
+bail:
+	return ret;
+}
+
+module_init(qib_ib_init);
+
+/*
+ * Do the non-unit driver cleanup, memory free, etc. at unload.
+ */
+static void __exit qib_ib_cleanup(void)
+{
+	int ret;
+
+	ret = qib_exit_qibfs();
+	if (ret)
+		pr_err(
+			"Unable to cleanup counter filesystem: error %d\n",
+			-ret);
+
+#ifdef CONFIG_INFINIBAND_QIB_DCA
+	dca_unregister_notify(&dca_notifier);
+#endif
+	pci_unregister_driver(&qib_driver);
+#ifdef CONFIG_DEBUG_FS
+	qib_dbg_exit();
+#endif
+
+	qib_cpulist_count = 0;
+	kfree(qib_cpulist);
+
+	idr_destroy(&qib_unit_table);
+	qib_dev_cleanup();
+}
+
+module_exit(qib_ib_cleanup);
+
+/* this can only be called after a successful initialization */
+static void cleanup_device_data(struct qib_devdata *dd)
+{
+	int ctxt;
+	int pidx;
+	struct qib_ctxtdata **tmp;
+	unsigned long flags;
+
+	/* users can't do anything more with chip */
+	for (pidx = 0; pidx < dd->num_pports; ++pidx) {
+		if (dd->pport[pidx].statusp)
+			*dd->pport[pidx].statusp &= ~QIB_STATUS_CHIP_PRESENT;
+
+		spin_lock(&dd->pport[pidx].cc_shadow_lock);
+
+		kfree(dd->pport[pidx].congestion_entries);
+		dd->pport[pidx].congestion_entries = NULL;
+		kfree(dd->pport[pidx].ccti_entries);
+		dd->pport[pidx].ccti_entries = NULL;
+		kfree(dd->pport[pidx].ccti_entries_shadow);
+		dd->pport[pidx].ccti_entries_shadow = NULL;
+		kfree(dd->pport[pidx].congestion_entries_shadow);
+		dd->pport[pidx].congestion_entries_shadow = NULL;
+
+		spin_unlock(&dd->pport[pidx].cc_shadow_lock);
+	}
+
+	if (!qib_wc_pat)
+		qib_disable_wc(dd);
+
+	if (dd->pioavailregs_dma) {
+		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+				  (void *) dd->pioavailregs_dma,
+				  dd->pioavailregs_phys);
+		dd->pioavailregs_dma = NULL;
+	}
+
+	if (dd->pageshadow) {
+		struct page **tmpp = dd->pageshadow;
+		dma_addr_t *tmpd = dd->physshadow;
+		int i;
+
+		for (ctxt = 0; ctxt < dd->cfgctxts; ctxt++) {
+			int ctxt_tidbase = ctxt * dd->rcvtidcnt;
+			int maxtid = ctxt_tidbase + dd->rcvtidcnt;
+
+			for (i = ctxt_tidbase; i < maxtid; i++) {
+				if (!tmpp[i])
+					continue;
+				pci_unmap_page(dd->pcidev, tmpd[i],
+					       PAGE_SIZE, PCI_DMA_FROMDEVICE);
+				qib_release_user_pages(&tmpp[i], 1);
+				tmpp[i] = NULL;
+			}
+		}
+
+		dd->pageshadow = NULL;
+		vfree(tmpp);
+		dd->physshadow = NULL;
+		vfree(tmpd);
+	}
+
+	/*
+	 * Free any resources still in use (usually just kernel contexts)
+	 * at unload; we do for ctxtcnt, because that's what we allocate.
+	 * We acquire lock to be really paranoid that rcd isn't being
+	 * accessed from some interrupt-related code (that should not happen,
+	 * but best to be sure).
+	 */
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	tmp = dd->rcd;
+	dd->rcd = NULL;
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+	for (ctxt = 0; tmp && ctxt < dd->ctxtcnt; ctxt++) {
+		struct qib_ctxtdata *rcd = tmp[ctxt];
+
+		tmp[ctxt] = NULL; /* debugging paranoia */
+		qib_free_ctxtdata(dd, rcd);
+	}
+	kfree(tmp);
+	kfree(dd->boardname);
+	qib_cq_exit(dd);
+}
+
+/*
+ * Clean up on unit shutdown, or error during unit load after
+ * successful initialization.
+ */
+static void qib_postinit_cleanup(struct qib_devdata *dd)
+{
+	/*
+	 * Clean up chip-specific stuff.
+	 * We check for NULL here, because it's outside
+	 * the kregbase check, and we need to call it
+	 * after the free_irq.  Thus it's possible that
+	 * the function pointers were never initialized.
+	 */
+	if (dd->f_cleanup)
+		dd->f_cleanup(dd);
+
+	qib_pcie_ddcleanup(dd);
+
+	cleanup_device_data(dd);
+
+	qib_free_devdata(dd);
+}
+
+static int qib_init_one(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+	int ret, j, pidx, initfail;
+	struct qib_devdata *dd = NULL;
+
+	ret = qib_pcie_init(pdev, ent);
+	if (ret)
+		goto bail;
+
+	/*
+	 * Do device-specific initialiation, function table setup, dd
+	 * allocation, etc.
+	 */
+	switch (ent->device) {
+	case PCI_DEVICE_ID_QLOGIC_IB_6120:
+#ifdef CONFIG_PCI_MSI
+		dd = qib_init_iba6120_funcs(pdev, ent);
+#else
+		qib_early_err(&pdev->dev,
+			"Intel PCIE device 0x%x cannot work if CONFIG_PCI_MSI is not enabled\n",
+			ent->device);
+		dd = ERR_PTR(-ENODEV);
+#endif
+		break;
+
+	case PCI_DEVICE_ID_QLOGIC_IB_7220:
+		dd = qib_init_iba7220_funcs(pdev, ent);
+		break;
+
+	case PCI_DEVICE_ID_QLOGIC_IB_7322:
+		dd = qib_init_iba7322_funcs(pdev, ent);
+		break;
+
+	default:
+		qib_early_err(&pdev->dev,
+			"Failing on unknown Intel deviceid 0x%x\n",
+			ent->device);
+		ret = -ENODEV;
+	}
+
+	if (IS_ERR(dd))
+		ret = PTR_ERR(dd);
+	if (ret)
+		goto bail; /* error already printed */
+
+	ret = qib_create_workqueues(dd);
+	if (ret)
+		goto bail;
+
+	/* do the generic initialization */
+	initfail = qib_init(dd, 0);
+
+	ret = qib_register_ib_device(dd);
+
+	/*
+	 * Now ready for use.  this should be cleared whenever we
+	 * detect a reset, or initiate one.  If earlier failure,
+	 * we still create devices, so diags, etc. can be used
+	 * to determine cause of problem.
+	 */
+	if (!qib_mini_init && !initfail && !ret)
+		dd->flags |= QIB_INITTED;
+
+	j = qib_device_create(dd);
+	if (j)
+		qib_dev_err(dd, "Failed to create /dev devices: %d\n", -j);
+	j = qibfs_add(dd);
+	if (j)
+		qib_dev_err(dd, "Failed filesystem setup for counters: %d\n",
+			    -j);
+
+	if (qib_mini_init || initfail || ret) {
+		qib_stop_timers(dd);
+		flush_workqueue(ib_wq);
+		for (pidx = 0; pidx < dd->num_pports; ++pidx)
+			dd->f_quiet_serdes(dd->pport + pidx);
+		if (qib_mini_init)
+			goto bail;
+		if (!j) {
+			(void) qibfs_remove(dd);
+			qib_device_remove(dd);
+		}
+		if (!ret)
+			qib_unregister_ib_device(dd);
+		qib_postinit_cleanup(dd);
+		if (initfail)
+			ret = initfail;
+		goto bail;
+	}
+
+	if (!qib_wc_pat) {
+		ret = qib_enable_wc(dd);
+		if (ret) {
+			qib_dev_err(dd,
+				"Write combining not enabled (err %d): performance may be poor\n",
+				-ret);
+			ret = 0;
+		}
+	}
+
+	qib_verify_pioperf(dd);
+bail:
+	return ret;
+}
+
+static void qib_remove_one(struct pci_dev *pdev)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	int ret;
+
+	/* unregister from IB core */
+	qib_unregister_ib_device(dd);
+
+	/*
+	 * Disable the IB link, disable interrupts on the device,
+	 * clear dma engines, etc.
+	 */
+	if (!qib_mini_init)
+		qib_shutdown_device(dd);
+
+	qib_stop_timers(dd);
+
+	/* wait until all of our (qsfp) queue_work() calls complete */
+	flush_workqueue(ib_wq);
+
+	ret = qibfs_remove(dd);
+	if (ret)
+		qib_dev_err(dd, "Failed counters filesystem cleanup: %d\n",
+			    -ret);
+
+	qib_device_remove(dd);
+
+	qib_postinit_cleanup(dd);
+}
+
+/**
+ * qib_create_rcvhdrq - create a receive header queue
+ * @dd: the qlogic_ib device
+ * @rcd: the context data
+ *
+ * This must be contiguous memory (from an i/o perspective), and must be
+ * DMA'able (which means for some systems, it will go through an IOMMU,
+ * or be forced into a low address range).
+ */
+int qib_create_rcvhdrq(struct qib_devdata *dd, struct qib_ctxtdata *rcd)
+{
+	unsigned amt;
+	int old_node_id;
+
+	if (!rcd->rcvhdrq) {
+		dma_addr_t phys_hdrqtail;
+		gfp_t gfp_flags;
+
+		amt = ALIGN(dd->rcvhdrcnt * dd->rcvhdrentsize *
+			    sizeof(u32), PAGE_SIZE);
+		gfp_flags = (rcd->ctxt >= dd->first_user_ctxt) ?
+			GFP_USER : GFP_KERNEL;
+
+		old_node_id = dev_to_node(&dd->pcidev->dev);
+		set_dev_node(&dd->pcidev->dev, rcd->node_id);
+		rcd->rcvhdrq = dma_alloc_coherent(
+			&dd->pcidev->dev, amt, &rcd->rcvhdrq_phys,
+			gfp_flags | __GFP_COMP);
+		set_dev_node(&dd->pcidev->dev, old_node_id);
+
+		if (!rcd->rcvhdrq) {
+			qib_dev_err(dd,
+				"attempt to allocate %d bytes for ctxt %u rcvhdrq failed\n",
+				amt, rcd->ctxt);
+			goto bail;
+		}
+
+		if (rcd->ctxt >= dd->first_user_ctxt) {
+			rcd->user_event_mask = vmalloc_user(PAGE_SIZE);
+			if (!rcd->user_event_mask)
+				goto bail_free_hdrq;
+		}
+
+		if (!(dd->flags & QIB_NODMA_RTAIL)) {
+			set_dev_node(&dd->pcidev->dev, rcd->node_id);
+			rcd->rcvhdrtail_kvaddr = dma_alloc_coherent(
+				&dd->pcidev->dev, PAGE_SIZE, &phys_hdrqtail,
+				gfp_flags);
+			set_dev_node(&dd->pcidev->dev, old_node_id);
+			if (!rcd->rcvhdrtail_kvaddr)
+				goto bail_free;
+			rcd->rcvhdrqtailaddr_phys = phys_hdrqtail;
+		}
+
+		rcd->rcvhdrq_size = amt;
+	}
+
+	/* clear for security and sanity on each use */
+	memset(rcd->rcvhdrq, 0, rcd->rcvhdrq_size);
+	if (rcd->rcvhdrtail_kvaddr)
+		memset(rcd->rcvhdrtail_kvaddr, 0, PAGE_SIZE);
+	return 0;
+
+bail_free:
+	qib_dev_err(dd,
+		"attempt to allocate 1 page for ctxt %u rcvhdrqtailaddr failed\n",
+		rcd->ctxt);
+	vfree(rcd->user_event_mask);
+	rcd->user_event_mask = NULL;
+bail_free_hdrq:
+	dma_free_coherent(&dd->pcidev->dev, amt, rcd->rcvhdrq,
+			  rcd->rcvhdrq_phys);
+	rcd->rcvhdrq = NULL;
+bail:
+	return -ENOMEM;
+}
+
+/**
+ * allocate eager buffers, both kernel and user contexts.
+ * @rcd: the context we are setting up.
+ *
+ * Allocate the eager TID buffers and program them into hip.
+ * They are no longer completely contiguous, we do multiple allocation
+ * calls.  Otherwise we get the OOM code involved, by asking for too
+ * much per call, with disastrous results on some kernels.
+ */
+int qib_setup_eagerbufs(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned e, egrcnt, egrperchunk, chunk, egrsize, egroff;
+	size_t size;
+	gfp_t gfp_flags;
+	int old_node_id;
+
+	/*
+	 * GFP_USER, but without GFP_FS, so buffer cache can be
+	 * coalesced (we hope); otherwise, even at order 4,
+	 * heavy filesystem activity makes these fail, and we can
+	 * use compound pages.
+	 */
+	gfp_flags = __GFP_WAIT | __GFP_IO | __GFP_COMP;
+
+	egrcnt = rcd->rcvegrcnt;
+	egroff = rcd->rcvegr_tid_base;
+	egrsize = dd->rcvegrbufsize;
+
+	chunk = rcd->rcvegrbuf_chunks;
+	egrperchunk = rcd->rcvegrbufs_perchunk;
+	size = rcd->rcvegrbuf_size;
+	if (!rcd->rcvegrbuf) {
+		rcd->rcvegrbuf =
+			kzalloc_node(chunk * sizeof(rcd->rcvegrbuf[0]),
+				GFP_KERNEL, rcd->node_id);
+		if (!rcd->rcvegrbuf)
+			goto bail;
+	}
+	if (!rcd->rcvegrbuf_phys) {
+		rcd->rcvegrbuf_phys =
+			kmalloc_node(chunk * sizeof(rcd->rcvegrbuf_phys[0]),
+				GFP_KERNEL, rcd->node_id);
+		if (!rcd->rcvegrbuf_phys)
+			goto bail_rcvegrbuf;
+	}
+	for (e = 0; e < rcd->rcvegrbuf_chunks; e++) {
+		if (rcd->rcvegrbuf[e])
+			continue;
+
+		old_node_id = dev_to_node(&dd->pcidev->dev);
+		set_dev_node(&dd->pcidev->dev, rcd->node_id);
+		rcd->rcvegrbuf[e] =
+			dma_alloc_coherent(&dd->pcidev->dev, size,
+					   &rcd->rcvegrbuf_phys[e],
+					   gfp_flags);
+		set_dev_node(&dd->pcidev->dev, old_node_id);
+		if (!rcd->rcvegrbuf[e])
+			goto bail_rcvegrbuf_phys;
+	}
+
+	rcd->rcvegr_phys = rcd->rcvegrbuf_phys[0];
+
+	for (e = chunk = 0; chunk < rcd->rcvegrbuf_chunks; chunk++) {
+		dma_addr_t pa = rcd->rcvegrbuf_phys[chunk];
+		unsigned i;
+
+		/* clear for security and sanity on each use */
+		memset(rcd->rcvegrbuf[chunk], 0, size);
+
+		for (i = 0; e < egrcnt && i < egrperchunk; e++, i++) {
+			dd->f_put_tid(dd, e + egroff +
+					  (u64 __iomem *)
+					  ((char __iomem *)
+					   dd->kregbase +
+					   dd->rcvegrbase),
+					  RCVHQ_RCV_TYPE_EAGER, pa);
+			pa += egrsize;
+		}
+		cond_resched(); /* don't hog the cpu */
+	}
+
+	return 0;
+
+bail_rcvegrbuf_phys:
+	for (e = 0; e < rcd->rcvegrbuf_chunks && rcd->rcvegrbuf[e]; e++)
+		dma_free_coherent(&dd->pcidev->dev, size,
+				  rcd->rcvegrbuf[e], rcd->rcvegrbuf_phys[e]);
+	kfree(rcd->rcvegrbuf_phys);
+	rcd->rcvegrbuf_phys = NULL;
+bail_rcvegrbuf:
+	kfree(rcd->rcvegrbuf);
+	rcd->rcvegrbuf = NULL;
+bail:
+	return -ENOMEM;
+}
+
+/*
+ * Note: Changes to this routine should be mirrored
+ * for the diagnostics routine qib_remap_ioaddr32().
+ * There is also related code for VL15 buffers in qib_init_7322_variables().
+ * The teardown code that unmaps is in qib_pcie_ddcleanup()
+ */
+int init_chip_wc_pat(struct qib_devdata *dd, u32 vl15buflen)
+{
+	u64 __iomem *qib_kregbase = NULL;
+	void __iomem *qib_piobase = NULL;
+	u64 __iomem *qib_userbase = NULL;
+	u64 qib_kreglen;
+	u64 qib_pio2koffset = dd->piobufbase & 0xffffffff;
+	u64 qib_pio4koffset = dd->piobufbase >> 32;
+	u64 qib_pio2klen = dd->piobcnt2k * dd->palign;
+	u64 qib_pio4klen = dd->piobcnt4k * dd->align4k;
+	u64 qib_physaddr = dd->physaddr;
+	u64 qib_piolen;
+	u64 qib_userlen = 0;
+
+	/*
+	 * Free the old mapping because the kernel will try to reuse the
+	 * old mapping and not create a new mapping with the
+	 * write combining attribute.
+	 */
+	iounmap(dd->kregbase);
+	dd->kregbase = NULL;
+
+	/*
+	 * Assumes chip address space looks like:
+	 *	- kregs + sregs + cregs + uregs (in any order)
+	 *	- piobufs (2K and 4K bufs in either order)
+	 * or:
+	 *	- kregs + sregs + cregs (in any order)
+	 *	- piobufs (2K and 4K bufs in either order)
+	 *	- uregs
+	 */
+	if (dd->piobcnt4k == 0) {
+		qib_kreglen = qib_pio2koffset;
+		qib_piolen = qib_pio2klen;
+	} else if (qib_pio2koffset < qib_pio4koffset) {
+		qib_kreglen = qib_pio2koffset;
+		qib_piolen = qib_pio4koffset + qib_pio4klen - qib_kreglen;
+	} else {
+		qib_kreglen = qib_pio4koffset;
+		qib_piolen = qib_pio2koffset + qib_pio2klen - qib_kreglen;
+	}
+	qib_piolen += vl15buflen;
+	/* Map just the configured ports (not all hw ports) */
+	if (dd->uregbase > qib_kreglen)
+		qib_userlen = dd->ureg_align * dd->cfgctxts;
+
+	/* Sanity checks passed, now create the new mappings */
+	qib_kregbase = ioremap_nocache(qib_physaddr, qib_kreglen);
+	if (!qib_kregbase)
+		goto bail;
+
+	qib_piobase = ioremap_wc(qib_physaddr + qib_kreglen, qib_piolen);
+	if (!qib_piobase)
+		goto bail_kregbase;
+
+	if (qib_userlen) {
+		qib_userbase = ioremap_nocache(qib_physaddr + dd->uregbase,
+					       qib_userlen);
+		if (!qib_userbase)
+			goto bail_piobase;
+	}
+
+	dd->kregbase = qib_kregbase;
+	dd->kregend = (u64 __iomem *)
+		((char __iomem *) qib_kregbase + qib_kreglen);
+	dd->piobase = qib_piobase;
+	dd->pio2kbase = (void __iomem *)
+		(((char __iomem *) dd->piobase) +
+		 qib_pio2koffset - qib_kreglen);
+	if (dd->piobcnt4k)
+		dd->pio4kbase = (void __iomem *)
+			(((char __iomem *) dd->piobase) +
+			 qib_pio4koffset - qib_kreglen);
+	if (qib_userlen)
+		/* ureg will now be accessed relative to dd->userbase */
+		dd->userbase = qib_userbase;
+	return 0;
+
+bail_piobase:
+	iounmap(qib_piobase);
+bail_kregbase:
+	iounmap(qib_kregbase);
+bail:
+	return -ENOMEM;
+}
diff --git a/drivers/infiniband/hw/qib/qib_intr.c b/drivers/infiniband/hw/qib/qib_intr.c
new file mode 100644
index 00000000000..f4918f2165e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_intr.c
@@ -0,0 +1,241 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/**
+ * qib_format_hwmsg - format a single hwerror message
+ * @msg message buffer
+ * @msgl length of message buffer
+ * @hwmsg message to add to message buffer
+ */
+static void qib_format_hwmsg(char *msg, size_t msgl, const char *hwmsg)
+{
+	strlcat(msg, "[", msgl);
+	strlcat(msg, hwmsg, msgl);
+	strlcat(msg, "]", msgl);
+}
+
+/**
+ * qib_format_hwerrors - format hardware error messages for display
+ * @hwerrs hardware errors bit vector
+ * @hwerrmsgs hardware error descriptions
+ * @nhwerrmsgs number of hwerrmsgs
+ * @msg message buffer
+ * @msgl message buffer length
+ */
+void qib_format_hwerrors(u64 hwerrs, const struct qib_hwerror_msgs *hwerrmsgs,
+			 size_t nhwerrmsgs, char *msg, size_t msgl)
+{
+	int i;
+
+	for (i = 0; i < nhwerrmsgs; i++)
+		if (hwerrs & hwerrmsgs[i].mask)
+			qib_format_hwmsg(msg, msgl, hwerrmsgs[i].msg);
+}
+
+static void signal_ib_event(struct qib_pportdata *ppd, enum ib_event_type ev)
+{
+	struct ib_event event;
+	struct qib_devdata *dd = ppd->dd;
+
+	event.device = &dd->verbs_dev.ibdev;
+	event.element.port_num = ppd->port;
+	event.event = ev;
+	ib_dispatch_event(&event);
+}
+
+void qib_handle_e_ibstatuschanged(struct qib_pportdata *ppd, u64 ibcs)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	u32 lstate;
+	u8 ltstate;
+	enum ib_event_type ev = 0;
+
+	lstate = dd->f_iblink_state(ibcs); /* linkstate */
+	ltstate = dd->f_ibphys_portstate(ibcs);
+
+	/*
+	 * If linkstate transitions into INIT from any of the various down
+	 * states, or if it transitions from any of the up (INIT or better)
+	 * states into any of the down states (except link recovery), then
+	 * call the chip-specific code to take appropriate actions.
+	 *
+	 * ppd->lflags could be 0 if this is the first time the interrupt
+	 * handlers has been called but the link is already up.
+	 */
+	if (lstate >= IB_PORT_INIT &&
+	    (!ppd->lflags || (ppd->lflags & QIBL_LINKDOWN)) &&
+	    ltstate == IB_PHYSPORTSTATE_LINKUP) {
+		/* transitioned to UP */
+		if (dd->f_ib_updown(ppd, 1, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+	} else if (ppd->lflags & (QIBL_LINKINIT | QIBL_LINKARMED |
+		   QIBL_LINKACTIVE | QIBL_IB_FORCE_NOTIFY)) {
+		if (ltstate != IB_PHYSPORTSTATE_LINKUP &&
+		    ltstate <= IB_PHYSPORTSTATE_CFG_TRAIN &&
+		    dd->f_ib_updown(ppd, 0, ibcs))
+			goto skip_ibchange; /* chip-code handled */
+		qib_set_uevent_bits(ppd, _QIB_EVENT_LINKDOWN_BIT);
+	}
+
+	if (lstate != IB_PORT_DOWN) {
+		/* lstate is INIT, ARMED, or ACTIVE */
+		if (lstate != IB_PORT_ACTIVE) {
+			*ppd->statusp &= ~QIB_STATUS_IB_READY;
+			if (ppd->lflags & QIBL_LINKACTIVE)
+				ev = IB_EVENT_PORT_ERR;
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			if (lstate == IB_PORT_ARMED) {
+				ppd->lflags |= QIBL_LINKARMED | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKINIT |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			} else {
+				ppd->lflags |= QIBL_LINKINIT | QIBL_LINKV;
+				ppd->lflags &= ~(QIBL_LINKARMED |
+					QIBL_LINKDOWN | QIBL_LINKACTIVE);
+			}
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			/* start a 75msec timer to clear symbol errors */
+			mod_timer(&ppd->symerr_clear_timer,
+				  msecs_to_jiffies(75));
+		} else if (ltstate == IB_PHYSPORTSTATE_LINKUP &&
+			   !(ppd->lflags & QIBL_LINKACTIVE)) {
+			/* active, but not active defered */
+			qib_hol_up(ppd); /* useful only for 6120 now */
+			*ppd->statusp |=
+				QIB_STATUS_IB_READY | QIB_STATUS_IB_CONF;
+			qib_clear_symerror_on_linkup((unsigned long)ppd);
+			spin_lock_irqsave(&ppd->lflags_lock, flags);
+			ppd->lflags |= QIBL_LINKACTIVE | QIBL_LINKV;
+			ppd->lflags &= ~(QIBL_LINKINIT |
+				QIBL_LINKDOWN | QIBL_LINKARMED);
+			spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+			if (dd->flags & QIB_HAS_SEND_DMA)
+				qib_sdma_process_event(ppd,
+					qib_sdma_event_e30_go_running);
+			ev = IB_EVENT_PORT_ACTIVE;
+			dd->f_setextled(ppd, 1);
+		}
+	} else { /* down */
+		if (ppd->lflags & QIBL_LINKACTIVE)
+			ev = IB_EVENT_PORT_ERR;
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags |= QIBL_LINKDOWN | QIBL_LINKV;
+		ppd->lflags &= ~(QIBL_LINKINIT |
+				 QIBL_LINKACTIVE | QIBL_LINKARMED);
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		*ppd->statusp &= ~QIB_STATUS_IB_READY;
+	}
+
+skip_ibchange:
+	ppd->lastibcstat = ibcs;
+	if (ev)
+		signal_ib_event(ppd, ev);
+	return;
+}
+
+void qib_clear_symerror_on_linkup(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	if (ppd->lflags & QIBL_LINKACTIVE)
+		return;
+
+	ppd->ibport_data.z_symbol_error_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBSYMBOLERR);
+}
+
+/*
+ * Handle receive interrupts for user ctxts; this means a user
+ * process was waiting for a packet to arrive, and didn't want
+ * to poll.
+ */
+void qib_handle_urcv(struct qib_devdata *dd, u64 ctxtr)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	int i;
+
+	spin_lock_irqsave(&dd->uctxt_lock, flags);
+	for (i = dd->first_user_ctxt; dd->rcd && i < dd->cfgctxts; i++) {
+		if (!(ctxtr & (1ULL << i)))
+			continue;
+		rcd = dd->rcd[i];
+		if (!rcd || !rcd->cnt)
+			continue;
+
+		if (test_and_clear_bit(QIB_CTXT_WAITING_RCV, &rcd->flag)) {
+			wake_up_interruptible(&rcd->wait);
+			dd->f_rcvctrl(rcd->ppd, QIB_RCVCTRL_INTRAVAIL_DIS,
+				      rcd->ctxt);
+		} else if (test_and_clear_bit(QIB_CTXT_WAITING_URG,
+					      &rcd->flag)) {
+			rcd->urgent++;
+			wake_up_interruptible(&rcd->wait);
+		}
+	}
+	spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+}
+
+void qib_bad_intrstatus(struct qib_devdata *dd)
+{
+	static int allbits;
+
+	/* separate routine, for better optimization of qib_intr() */
+
+	/*
+	 * We print the message and disable interrupts, in hope of
+	 * having a better chance of debugging the problem.
+	 */
+	qib_dev_err(dd,
+		"Read of chip interrupt status failed disabling interrupts\n");
+	if (allbits++) {
+		/* disable interrupt delivery, something is very wrong */
+		if (allbits == 2)
+			dd->f_set_intr_state(dd, 0);
+		if (allbits == 3) {
+			qib_dev_err(dd,
+				"2nd bad interrupt status, unregistering interrupts\n");
+			dd->flags |= QIB_BADINTR;
+			dd->flags &= ~QIB_INITTED;
+			dd->f_free_irq(dd);
+		}
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_keys.c b/drivers/infiniband/hw/qib/qib_keys.c
new file mode 100644
index 00000000000..3b9afccaaad
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_keys.c
@@ -0,0 +1,387 @@
+/*
+ * Copyright (c) 2006, 2007, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_alloc_lkey - allocate an lkey
+ * @mr: memory region that this lkey protects
+ * @dma_region: 0->normal key, 1->restricted DMA key
+ *
+ * Returns 0 if successful, otherwise returns -errno.
+ *
+ * Increments mr reference count as required.
+ *
+ * Sets the lkey field mr for non-dma regions.
+ *
+ */
+
+int qib_alloc_lkey(struct qib_mregion *mr, int dma_region)
+{
+	unsigned long flags;
+	u32 r;
+	u32 n;
+	int ret = 0;
+	struct qib_ibdev *dev = to_idev(mr->pd->device);
+	struct qib_lkey_table *rkt = &dev->lk_table;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+
+	/* special case for dma_mr lkey == 0 */
+	if (dma_region) {
+		struct qib_mregion *tmr;
+
+		tmr = rcu_access_pointer(dev->dma_mr);
+		if (!tmr) {
+			qib_get_mr(mr);
+			rcu_assign_pointer(dev->dma_mr, mr);
+			mr->lkey_published = 1;
+		}
+		goto success;
+	}
+
+	/* Find the next available LKEY */
+	r = rkt->next;
+	n = r;
+	for (;;) {
+		if (rkt->table[r] == NULL)
+			break;
+		r = (r + 1) & (rkt->max - 1);
+		if (r == n)
+			goto bail;
+	}
+	rkt->next = (r + 1) & (rkt->max - 1);
+	/*
+	 * Make sure lkey is never zero which is reserved to indicate an
+	 * unrestricted LKEY.
+	 */
+	rkt->gen++;
+	mr->lkey = (r << (32 - ib_qib_lkey_table_size)) |
+		((((1 << (24 - ib_qib_lkey_table_size)) - 1) & rkt->gen)
+		 << 8);
+	if (mr->lkey == 0) {
+		mr->lkey |= 1 << 8;
+		rkt->gen++;
+	}
+	qib_get_mr(mr);
+	rcu_assign_pointer(rkt->table[r], mr);
+	mr->lkey_published = 1;
+success:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+out:
+	return ret;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	ret = -ENOMEM;
+	goto out;
+}
+
+/**
+ * qib_free_lkey - free an lkey
+ * @mr: mr to free from tables
+ */
+void qib_free_lkey(struct qib_mregion *mr)
+{
+	unsigned long flags;
+	u32 lkey = mr->lkey;
+	u32 r;
+	struct qib_ibdev *dev = to_idev(mr->pd->device);
+	struct qib_lkey_table *rkt = &dev->lk_table;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (!mr->lkey_published)
+		goto out;
+	if (lkey == 0)
+		rcu_assign_pointer(dev->dma_mr, NULL);
+	else {
+		r = lkey >> (32 - ib_qib_lkey_table_size);
+		rcu_assign_pointer(rkt->table[r], NULL);
+	}
+	qib_put_mr(mr);
+	mr->lkey_published = 0;
+out:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+}
+
+/**
+ * qib_lkey_ok - check IB SGE for validity and initialize
+ * @rkt: table containing lkey to check SGE against
+ * @pd: protection domain
+ * @isge: outgoing internal SGE
+ * @sge: SGE to check
+ * @acc: access flags
+ *
+ * Return 1 if valid and successful, otherwise returns 0.
+ *
+ * increments the reference count upon success
+ *
+ * Check the IB SGE for validity and initialize our internal version
+ * of it.
+ */
+int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
+		struct qib_sge *isge, struct ib_sge *sge, int acc)
+{
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+
+	/*
+	 * We use LKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	rcu_read_lock();
+	if (sge->lkey == 0) {
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		mr = rcu_dereference(dev->dma_mr);
+		if (!mr)
+			goto bail;
+		if (unlikely(!atomic_inc_not_zero(&mr->refcount)))
+			goto bail;
+		rcu_read_unlock();
+
+		isge->mr = mr;
+		isge->vaddr = (void *) sge->addr;
+		isge->length = sge->length;
+		isge->sge_length = sge->length;
+		isge->m = 0;
+		isge->n = 0;
+		goto ok;
+	}
+	mr = rcu_dereference(
+		rkt->table[(sge->lkey >> (32 - ib_qib_lkey_table_size))]);
+	if (unlikely(!mr || mr->lkey != sge->lkey || mr->pd != &pd->ibpd))
+		goto bail;
+
+	off = sge->addr - mr->user_base;
+	if (unlikely(sge->addr < mr->user_base ||
+		     off + sge->length > mr->length ||
+		     (mr->access_flags & acc) != acc))
+		goto bail;
+	if (unlikely(!atomic_inc_not_zero(&mr->refcount)))
+		goto bail;
+	rcu_read_unlock();
+
+	off += mr->offset;
+	if (mr->page_shift) {
+		/*
+		page sizes are uniform power of 2 so no loop is necessary
+		entries_spanned_by_off is the number of times the loop below
+		would have executed.
+		*/
+		size_t entries_spanned_by_off;
+
+		entries_spanned_by_off = off >> mr->page_shift;
+		off -= (entries_spanned_by_off << mr->page_shift);
+		m = entries_spanned_by_off/QIB_SEGSZ;
+		n = entries_spanned_by_off%QIB_SEGSZ;
+	} else {
+		m = 0;
+		n = 0;
+		while (off >= mr->map[m]->segs[n].length) {
+			off -= mr->map[m]->segs[n].length;
+			n++;
+			if (n >= QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
+		}
+	}
+	isge->mr = mr;
+	isge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	isge->length = mr->map[m]->segs[n].length - off;
+	isge->sge_length = sge->length;
+	isge->m = m;
+	isge->n = n;
+ok:
+	return 1;
+bail:
+	rcu_read_unlock();
+	return 0;
+}
+
+/**
+ * qib_rkey_ok - check the IB virtual address, length, and RKEY
+ * @qp: qp for validation
+ * @sge: SGE state
+ * @len: length of data
+ * @vaddr: virtual address to place data
+ * @rkey: rkey to check
+ * @acc: access flags
+ *
+ * Return 1 if successful, otherwise 0.
+ *
+ * increments the reference count upon success
+ */
+int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
+		u32 len, u64 vaddr, u32 rkey, int acc)
+{
+	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
+	struct qib_mregion *mr;
+	unsigned n, m;
+	size_t off;
+
+	/*
+	 * We use RKEY == zero for kernel virtual addresses
+	 * (see qib_get_dma_mr and qib_dma.c).
+	 */
+	rcu_read_lock();
+	if (rkey == 0) {
+		struct qib_pd *pd = to_ipd(qp->ibqp.pd);
+		struct qib_ibdev *dev = to_idev(pd->ibpd.device);
+
+		if (pd->user)
+			goto bail;
+		mr = rcu_dereference(dev->dma_mr);
+		if (!mr)
+			goto bail;
+		if (unlikely(!atomic_inc_not_zero(&mr->refcount)))
+			goto bail;
+		rcu_read_unlock();
+
+		sge->mr = mr;
+		sge->vaddr = (void *) vaddr;
+		sge->length = len;
+		sge->sge_length = len;
+		sge->m = 0;
+		sge->n = 0;
+		goto ok;
+	}
+
+	mr = rcu_dereference(
+		rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))]);
+	if (unlikely(!mr || mr->lkey != rkey || qp->ibqp.pd != mr->pd))
+		goto bail;
+
+	off = vaddr - mr->iova;
+	if (unlikely(vaddr < mr->iova || off + len > mr->length ||
+		     (mr->access_flags & acc) == 0))
+		goto bail;
+	if (unlikely(!atomic_inc_not_zero(&mr->refcount)))
+		goto bail;
+	rcu_read_unlock();
+
+	off += mr->offset;
+	if (mr->page_shift) {
+		/*
+		page sizes are uniform power of 2 so no loop is necessary
+		entries_spanned_by_off is the number of times the loop below
+		would have executed.
+		*/
+		size_t entries_spanned_by_off;
+
+		entries_spanned_by_off = off >> mr->page_shift;
+		off -= (entries_spanned_by_off << mr->page_shift);
+		m = entries_spanned_by_off/QIB_SEGSZ;
+		n = entries_spanned_by_off%QIB_SEGSZ;
+	} else {
+		m = 0;
+		n = 0;
+		while (off >= mr->map[m]->segs[n].length) {
+			off -= mr->map[m]->segs[n].length;
+			n++;
+			if (n >= QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
+		}
+	}
+	sge->mr = mr;
+	sge->vaddr = mr->map[m]->segs[n].vaddr + off;
+	sge->length = mr->map[m]->segs[n].length - off;
+	sge->sge_length = len;
+	sge->m = m;
+	sge->n = n;
+ok:
+	return 1;
+bail:
+	rcu_read_unlock();
+	return 0;
+}
+
+/*
+ * Initialize the memory region specified by the work reqeust.
+ */
+int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr)
+{
+	struct qib_lkey_table *rkt = &to_idev(qp->ibqp.device)->lk_table;
+	struct qib_pd *pd = to_ipd(qp->ibqp.pd);
+	struct qib_mregion *mr;
+	u32 rkey = wr->wr.fast_reg.rkey;
+	unsigned i, n, m;
+	int ret = -EINVAL;
+	unsigned long flags;
+	u64 *page_list;
+	size_t ps;
+
+	spin_lock_irqsave(&rkt->lock, flags);
+	if (pd->user || rkey == 0)
+		goto bail;
+
+	mr = rcu_dereference_protected(
+		rkt->table[(rkey >> (32 - ib_qib_lkey_table_size))],
+		lockdep_is_held(&rkt->lock));
+	if (unlikely(mr == NULL || qp->ibqp.pd != mr->pd))
+		goto bail;
+
+	if (wr->wr.fast_reg.page_list_len > mr->max_segs)
+		goto bail;
+
+	ps = 1UL << wr->wr.fast_reg.page_shift;
+	if (wr->wr.fast_reg.length > ps * wr->wr.fast_reg.page_list_len)
+		goto bail;
+
+	mr->user_base = wr->wr.fast_reg.iova_start;
+	mr->iova = wr->wr.fast_reg.iova_start;
+	mr->lkey = rkey;
+	mr->length = wr->wr.fast_reg.length;
+	mr->access_flags = wr->wr.fast_reg.access_flags;
+	page_list = wr->wr.fast_reg.page_list->page_list;
+	m = 0;
+	n = 0;
+	for (i = 0; i < wr->wr.fast_reg.page_list_len; i++) {
+		mr->map[m]->segs[n].vaddr = (void *) page_list[i];
+		mr->map[m]->segs[n].length = ps;
+		if (++n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+
+	ret = 0;
+bail:
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	return ret;
+}
diff --git a/drivers/infiniband/hw/qib/qib_mad.c b/drivers/infiniband/hw/qib/qib_mad.c
new file mode 100644
index 00000000000..22c720e5740
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mad.c
@@ -0,0 +1,2533 @@
+/*
+ * Copyright (c) 2012 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+static int reply(struct ib_smp *smp)
+{
+	/*
+	 * The verbs framework will handle the directed/LID route
+	 * packet changes.
+	 */
+	smp->method = IB_MGMT_METHOD_GET_RESP;
+	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		smp->status |= IB_SMP_DIRECTION;
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_REPLY;
+}
+
+static int reply_failure(struct ib_smp *smp)
+{
+	/*
+	 * The verbs framework will handle the directed/LID route
+	 * packet changes.
+	 */
+	smp->method = IB_MGMT_METHOD_GET_RESP;
+	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+		smp->status |= IB_SMP_DIRECTION;
+	return IB_MAD_RESULT_FAILURE | IB_MAD_RESULT_REPLY;
+}
+
+static void qib_send_trap(struct qib_ibport *ibp, void *data, unsigned len)
+{
+	struct ib_mad_send_buf *send_buf;
+	struct ib_mad_agent *agent;
+	struct ib_smp *smp;
+	int ret;
+	unsigned long flags;
+	unsigned long timeout;
+
+	agent = ibp->send_agent;
+	if (!agent)
+		return;
+
+	/* o14-3.2.1 */
+	if (!(ppd_from_ibp(ibp)->lflags & QIBL_LINKACTIVE))
+		return;
+
+	/* o14-2 */
+	if (ibp->trap_timeout && time_before(jiffies, ibp->trap_timeout))
+		return;
+
+	send_buf = ib_create_send_mad(agent, 0, 0, 0, IB_MGMT_MAD_HDR,
+				      IB_MGMT_MAD_DATA, GFP_ATOMIC);
+	if (IS_ERR(send_buf))
+		return;
+
+	smp = send_buf->mad;
+	smp->base_version = IB_MGMT_BASE_VERSION;
+	smp->mgmt_class = IB_MGMT_CLASS_SUBN_LID_ROUTED;
+	smp->class_version = 1;
+	smp->method = IB_MGMT_METHOD_TRAP;
+	ibp->tid++;
+	smp->tid = cpu_to_be64(ibp->tid);
+	smp->attr_id = IB_SMP_ATTR_NOTICE;
+	/* o14-1: smp->mkey = 0; */
+	memcpy(smp->data, data, len);
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	if (!ibp->sm_ah) {
+		if (ibp->sm_lid != be16_to_cpu(IB_LID_PERMISSIVE)) {
+			struct ib_ah *ah;
+
+			ah = qib_create_qp0_ah(ibp, ibp->sm_lid);
+			if (IS_ERR(ah))
+				ret = PTR_ERR(ah);
+			else {
+				send_buf->ah = ah;
+				ibp->sm_ah = to_iah(ah);
+				ret = 0;
+			}
+		} else
+			ret = -EINVAL;
+	} else {
+		send_buf->ah = &ibp->sm_ah->ibah;
+		ret = 0;
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	if (!ret)
+		ret = ib_post_send_mad(send_buf, NULL);
+	if (!ret) {
+		/* 4.096 usec. */
+		timeout = (4096 * (1UL << ibp->subnet_timeout)) / 1000;
+		ibp->trap_timeout = jiffies + usecs_to_jiffies(timeout);
+	} else {
+		ib_free_send_mad(send_buf);
+		ibp->trap_timeout = 0;
+	}
+}
+
+/*
+ * Send a bad [PQ]_Key trap (ch. 14.3.8).
+ */
+void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
+		   u32 qp1, u32 qp2, __be16 lid1, __be16 lid2)
+{
+	struct ib_mad_notice_attr data;
+
+	if (trap_num == IB_NOTICE_TRAP_BAD_PKEY)
+		ibp->pkey_violations++;
+	else
+		ibp->qkey_violations++;
+	ibp->n_pkt_drops++;
+
+	/* Send violation trap */
+	data.generic_type = IB_NOTICE_TYPE_SECURITY;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = trap_num;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_257_258.lid1 = lid1;
+	data.details.ntc_257_258.lid2 = lid2;
+	data.details.ntc_257_258.key = cpu_to_be32(key);
+	data.details.ntc_257_258.sl_qp1 = cpu_to_be32((sl << 28) | qp1);
+	data.details.ntc_257_258.qp2 = cpu_to_be32(qp2);
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a bad M_Key trap (ch. 14.3.9).
+ */
+static void qib_bad_mkey(struct qib_ibport *ibp, struct ib_smp *smp)
+{
+	struct ib_mad_notice_attr data;
+
+	/* Send violation trap */
+	data.generic_type = IB_NOTICE_TYPE_SECURITY;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_BAD_MKEY;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_256.lid = data.issuer_lid;
+	data.details.ntc_256.method = smp->method;
+	data.details.ntc_256.attr_id = smp->attr_id;
+	data.details.ntc_256.attr_mod = smp->attr_mod;
+	data.details.ntc_256.mkey = smp->mkey;
+	if (smp->mgmt_class == IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE) {
+		u8 hop_cnt;
+
+		data.details.ntc_256.dr_slid = smp->dr_slid;
+		data.details.ntc_256.dr_trunc_hop = IB_NOTICE_TRAP_DR_NOTICE;
+		hop_cnt = smp->hop_cnt;
+		if (hop_cnt > ARRAY_SIZE(data.details.ntc_256.dr_rtn_path)) {
+			data.details.ntc_256.dr_trunc_hop |=
+				IB_NOTICE_TRAP_DR_TRUNC;
+			hop_cnt = ARRAY_SIZE(data.details.ntc_256.dr_rtn_path);
+		}
+		data.details.ntc_256.dr_trunc_hop |= hop_cnt;
+		memcpy(data.details.ntc_256.dr_rtn_path, smp->return_path,
+		       hop_cnt);
+	}
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a Port Capability Mask Changed trap (ch. 14.3.11).
+ */
+void qib_cap_mask_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_144.lid = data.issuer_lid;
+	data.details.ntc_144.new_cap_mask = cpu_to_be32(ibp->port_cap_flags);
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a System Image GUID Changed trap (ch. 14.3.12).
+ */
+void qib_sys_guid_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_SYS_GUID_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_145.lid = data.issuer_lid;
+	data.details.ntc_145.new_sys_guid = ib_qib_sys_image_guid;
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+/*
+ * Send a Node Description Changed trap (ch. 14.3.13).
+ */
+void qib_node_desc_chg(struct qib_ibport *ibp)
+{
+	struct ib_mad_notice_attr data;
+
+	data.generic_type = IB_NOTICE_TYPE_INFO;
+	data.prod_type_msb = 0;
+	data.prod_type_lsb = IB_NOTICE_PROD_CA;
+	data.trap_num = IB_NOTICE_TRAP_CAP_MASK_CHG;
+	data.issuer_lid = cpu_to_be16(ppd_from_ibp(ibp)->lid);
+	data.toggle_count = 0;
+	memset(&data.details, 0, sizeof data.details);
+	data.details.ntc_144.lid = data.issuer_lid;
+	data.details.ntc_144.local_changes = 1;
+	data.details.ntc_144.change_flags = IB_NOTICE_TRAP_NODE_DESC_CHG;
+
+	qib_send_trap(ibp, &data, sizeof data);
+}
+
+static int subn_get_nodedescription(struct ib_smp *smp,
+				    struct ib_device *ibdev)
+{
+	if (smp->attr_mod)
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	memcpy(smp->data, ibdev->node_desc, sizeof(smp->data));
+
+	return reply(smp);
+}
+
+static int subn_get_nodeinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct ib_node_info *nip = (struct ib_node_info *)&smp->data;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 vendor, majrev, minrev;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* GUID 0 is illegal */
+	if (smp->attr_mod || pidx >= dd->num_pports ||
+	    dd->pport[pidx].guid == 0)
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else
+		nip->port_guid = dd->pport[pidx].guid;
+
+	nip->base_version = 1;
+	nip->class_version = 1;
+	nip->node_type = 1;     /* channel adapter */
+	nip->num_ports = ibdev->phys_port_cnt;
+	/* This is already in network order */
+	nip->sys_guid = ib_qib_sys_image_guid;
+	nip->node_guid = dd->pport->guid; /* Use first-port GUID as node */
+	nip->partition_cap = cpu_to_be16(qib_get_npkeys(dd));
+	nip->device_id = cpu_to_be16(dd->deviceid);
+	majrev = dd->majrev;
+	minrev = dd->minrev;
+	nip->revision = cpu_to_be32((majrev << 16) | minrev);
+	nip->local_port_num = port;
+	vendor = dd->vendorid;
+	nip->vendor_id[0] = QIB_SRC_OUI_1;
+	nip->vendor_id[1] = QIB_SRC_OUI_2;
+	nip->vendor_id[2] = QIB_SRC_OUI_3;
+
+	return reply(smp);
+}
+
+static int subn_get_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
+	__be64 *p = (__be64 *) smp->data;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* 32 blocks of 8 64-bit GUIDs per block */
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (startgx == 0 && pidx < dd->num_pports) {
+		struct qib_pportdata *ppd = dd->pport + pidx;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		__be64 g = ppd->guid;
+		unsigned i;
+
+		/* GUID 0 is illegal */
+		if (g == 0)
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else {
+			/* The first is a copy of the read-only HW GUID. */
+			p[0] = g;
+			for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
+				p[i] = ibp->guids[i - 1];
+		}
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static void set_link_width_enabled(struct qib_pportdata *ppd, u32 w)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LWID_ENB, w);
+}
+
+static void set_link_speed_enabled(struct qib_pportdata *ppd, u32 s)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_SPD_ENB, s);
+}
+
+static int get_overrunthreshold(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH);
+}
+
+/**
+ * set_overrunthreshold - set the overrun threshold
+ * @ppd: the physical port data
+ * @n: the new threshold
+ *
+ * Note that this will only take effect when the link state changes.
+ */
+static int set_overrunthreshold(struct qib_pportdata *ppd, unsigned n)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OVERRUN_THRESH,
+					 (u32)n);
+	return 0;
+}
+
+static int get_phyerrthreshold(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH);
+}
+
+/**
+ * set_phyerrthreshold - set the physical error threshold
+ * @ppd: the physical port data
+ * @n: the new threshold
+ *
+ * Note that this will only take effect when the link state changes.
+ */
+static int set_phyerrthreshold(struct qib_pportdata *ppd, unsigned n)
+{
+	(void) ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PHYERR_THRESH,
+					 (u32)n);
+	return 0;
+}
+
+/**
+ * get_linkdowndefaultstate - get the default linkdown state
+ * @ppd: the physical port data
+ *
+ * Returns zero if the default is POLL, 1 if the default is SLEEP.
+ */
+static int get_linkdowndefaultstate(struct qib_pportdata *ppd)
+{
+	return ppd->dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT) ==
+		IB_LINKINITCMD_SLEEP;
+}
+
+static int check_mkey(struct qib_ibport *ibp, struct ib_smp *smp, int mad_flags)
+{
+	int valid_mkey = 0;
+	int ret = 0;
+
+	/* Is the mkey in the process of expiring? */
+	if (ibp->mkey_lease_timeout &&
+	    time_after_eq(jiffies, ibp->mkey_lease_timeout)) {
+		/* Clear timeout and mkey protection field. */
+		ibp->mkey_lease_timeout = 0;
+		ibp->mkeyprot = 0;
+	}
+
+	if ((mad_flags & IB_MAD_IGNORE_MKEY) ||  ibp->mkey == 0 ||
+	    ibp->mkey == smp->mkey)
+		valid_mkey = 1;
+
+	/* Unset lease timeout on any valid Get/Set/TrapRepress */
+	if (valid_mkey && ibp->mkey_lease_timeout &&
+	    (smp->method == IB_MGMT_METHOD_GET ||
+	     smp->method == IB_MGMT_METHOD_SET ||
+	     smp->method == IB_MGMT_METHOD_TRAP_REPRESS))
+		ibp->mkey_lease_timeout = 0;
+
+	if (!valid_mkey) {
+		switch (smp->method) {
+		case IB_MGMT_METHOD_GET:
+			/* Bad mkey not a violation below level 2 */
+			if (ibp->mkeyprot < 2)
+				break;
+		case IB_MGMT_METHOD_SET:
+		case IB_MGMT_METHOD_TRAP_REPRESS:
+			if (ibp->mkey_violations != 0xFFFF)
+				++ibp->mkey_violations;
+			if (!ibp->mkey_lease_timeout && ibp->mkey_lease_period)
+				ibp->mkey_lease_timeout = jiffies +
+					ibp->mkey_lease_period * HZ;
+			/* Generate a trap notice. */
+			qib_bad_mkey(ibp, smp);
+			ret = 1;
+		}
+	}
+
+	return ret;
+}
+
+static int subn_get_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
+	u8 mtu;
+	int ret;
+	u32 state;
+	u32 port_num = be32_to_cpu(smp->attr_mod);
+
+	if (port_num == 0)
+		port_num = port;
+	else {
+		if (port_num > ibdev->phys_port_cnt) {
+			smp->status |= IB_SMP_INVALID_FIELD;
+			ret = reply(smp);
+			goto bail;
+		}
+		if (port_num != port) {
+			ibp = to_iport(ibdev, port_num);
+			ret = check_mkey(ibp, smp, 0);
+			if (ret) {
+				ret = IB_MAD_RESULT_FAILURE;
+				goto bail;
+			}
+		}
+	}
+
+	dd = dd_from_ibdev(ibdev);
+	/* IB numbers ports from 1, hdw from 0 */
+	ppd = dd->pport + (port_num - 1);
+	ibp = &ppd->ibport_data;
+
+	/* Clear all fields.  Only set the non-zero fields. */
+	memset(smp->data, 0, sizeof(smp->data));
+
+	/* Only return the mkey if the protection field allows it. */
+	if (!(smp->method == IB_MGMT_METHOD_GET &&
+	      ibp->mkey != smp->mkey &&
+	      ibp->mkeyprot == 1))
+		pip->mkey = ibp->mkey;
+	pip->gid_prefix = ibp->gid_prefix;
+	pip->lid = cpu_to_be16(ppd->lid);
+	pip->sm_lid = cpu_to_be16(ibp->sm_lid);
+	pip->cap_mask = cpu_to_be32(ibp->port_cap_flags);
+	/* pip->diag_code; */
+	pip->mkey_lease_period = cpu_to_be16(ibp->mkey_lease_period);
+	pip->local_port_num = port;
+	pip->link_width_enabled = ppd->link_width_enabled;
+	pip->link_width_supported = ppd->link_width_supported;
+	pip->link_width_active = ppd->link_width_active;
+	state = dd->f_iblink_state(ppd->lastibcstat);
+	pip->linkspeed_portstate = ppd->link_speed_supported << 4 | state;
+
+	pip->portphysstate_linkdown =
+		(dd->f_ibphys_portstate(ppd->lastibcstat) << 4) |
+		(get_linkdowndefaultstate(ppd) ? 1 : 2);
+	pip->mkeyprot_resv_lmc = (ibp->mkeyprot << 6) | ppd->lmc;
+	pip->linkspeedactive_enabled = (ppd->link_speed_active << 4) |
+		ppd->link_speed_enabled;
+	switch (ppd->ibmtu) {
+	default: /* something is wrong; fall through */
+	case 4096:
+		mtu = IB_MTU_4096;
+		break;
+	case 2048:
+		mtu = IB_MTU_2048;
+		break;
+	case 1024:
+		mtu = IB_MTU_1024;
+		break;
+	case 512:
+		mtu = IB_MTU_512;
+		break;
+	case 256:
+		mtu = IB_MTU_256;
+		break;
+	}
+	pip->neighbormtu_mastersmsl = (mtu << 4) | ibp->sm_sl;
+	pip->vlcap_inittype = ppd->vls_supported << 4;  /* InitType = 0 */
+	pip->vl_high_limit = ibp->vl_high_limit;
+	pip->vl_arb_high_cap =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_CAP);
+	pip->vl_arb_low_cap =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_VL_LOW_CAP);
+	/* InitTypeReply = 0 */
+	pip->inittypereply_mtucap = qib_ibmtu ? qib_ibmtu : IB_MTU_4096;
+	/* HCAs ignore VLStallCount and HOQLife */
+	/* pip->vlstallcnt_hoqlife; */
+	pip->operationalvl_pei_peo_fpi_fpo =
+		dd->f_get_ib_cfg(ppd, QIB_IB_CFG_OP_VLS) << 4;
+	pip->mkey_violations = cpu_to_be16(ibp->mkey_violations);
+	/* P_KeyViolations are counted by hardware. */
+	pip->pkey_violations = cpu_to_be16(ibp->pkey_violations);
+	pip->qkey_violations = cpu_to_be16(ibp->qkey_violations);
+	/* Only the hardware GUID is supported for now */
+	pip->guid_cap = QIB_GUIDS_PER_PORT;
+	pip->clientrereg_resv_subnetto = ibp->subnet_timeout;
+	/* 32.768 usec. response time (guessing) */
+	pip->resv_resptimevalue = 3;
+	pip->localphyerrors_overrunerrors =
+		(get_phyerrthreshold(ppd) << 4) |
+		get_overrunthreshold(ppd);
+	/* pip->max_credit_hint; */
+	if (ibp->port_cap_flags & IB_PORT_LINK_LATENCY_SUP) {
+		u32 v;
+
+		v = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_LINKLATENCY);
+		pip->link_roundtrip_latency[0] = v >> 16;
+		pip->link_roundtrip_latency[1] = v >> 8;
+		pip->link_roundtrip_latency[2] = v;
+	}
+
+	ret = reply(smp);
+
+bail:
+	return ret;
+}
+
+/**
+ * get_pkeys - return the PKEY table
+ * @dd: the qlogic_ib device
+ * @port: the IB port number
+ * @pkeys: the pkey table is placed here
+ */
+static int get_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
+{
+	struct qib_pportdata *ppd = dd->pport + port - 1;
+	/*
+	 * always a kernel context, no locking needed.
+	 * If we get here with ppd setup, no need to check
+	 * that pd is valid.
+	 */
+	struct qib_ctxtdata *rcd = dd->rcd[ppd->hw_pidx];
+
+	memcpy(pkeys, rcd->pkeys, sizeof(rcd->pkeys));
+
+	return 0;
+}
+
+static int subn_get_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
+			      u8 port)
+{
+	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
+	u16 *p = (u16 *) smp->data;
+	__be16 *q = (__be16 *) smp->data;
+
+	/* 64 blocks of 32 16-bit P_Key entries */
+
+	memset(smp->data, 0, sizeof(smp->data));
+	if (startpx == 0) {
+		struct qib_devdata *dd = dd_from_ibdev(ibdev);
+		unsigned i, n = qib_get_npkeys(dd);
+
+		get_pkeys(dd, port, p);
+
+		for (i = 0; i < n; i++)
+			q[i] = cpu_to_be16(p[i]);
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static int subn_set_guidinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	u32 startgx = 8 * be32_to_cpu(smp->attr_mod);
+	__be64 *p = (__be64 *) smp->data;
+	unsigned pidx = port - 1; /* IB number port from 1, hdw from 0 */
+
+	/* 32 blocks of 8 64-bit GUIDs per block */
+
+	if (startgx == 0 && pidx < dd->num_pports) {
+		struct qib_pportdata *ppd = dd->pport + pidx;
+		struct qib_ibport *ibp = &ppd->ibport_data;
+		unsigned i;
+
+		/* The first entry is read-only. */
+		for (i = 1; i < QIB_GUIDS_PER_PORT; i++)
+			ibp->guids[i - 1] = p[i];
+	} else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	/* The only GUID we support is the first read-only entry. */
+	return subn_get_guidinfo(smp, ibdev, port);
+}
+
+/**
+ * subn_set_portinfo - set port information
+ * @smp: the incoming SM packet
+ * @ibdev: the infiniband device
+ * @port: the port on the device
+ *
+ * Set Portinfo (see ch. 14.2.5.6).
+ */
+static int subn_set_portinfo(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct ib_port_info *pip = (struct ib_port_info *)smp->data;
+	struct ib_event event;
+	struct qib_devdata *dd;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	u8 clientrereg = (pip->clientrereg_resv_subnetto & 0x80);
+	unsigned long flags;
+	u16 lid, smlid;
+	u8 lwe;
+	u8 lse;
+	u8 state;
+	u8 vls;
+	u8 msl;
+	u16 lstate;
+	int ret, ore, mtu;
+	u32 port_num = be32_to_cpu(smp->attr_mod);
+
+	if (port_num == 0)
+		port_num = port;
+	else {
+		if (port_num > ibdev->phys_port_cnt)
+			goto err;
+		/* Port attributes can only be set on the receiving port */
+		if (port_num != port)
+			goto get_only;
+	}
+
+	dd = dd_from_ibdev(ibdev);
+	/* IB numbers ports from 1, hdw from 0 */
+	ppd = dd->pport + (port_num - 1);
+	ibp = &ppd->ibport_data;
+	event.device = ibdev;
+	event.element.port_num = port;
+
+	ibp->mkey = pip->mkey;
+	ibp->gid_prefix = pip->gid_prefix;
+	ibp->mkey_lease_period = be16_to_cpu(pip->mkey_lease_period);
+
+	lid = be16_to_cpu(pip->lid);
+	/* Must be a valid unicast LID address. */
+	if (lid == 0 || lid >= QIB_MULTICAST_LID_BASE)
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else if (ppd->lid != lid || ppd->lmc != (pip->mkeyprot_resv_lmc & 7)) {
+		if (ppd->lid != lid)
+			qib_set_uevent_bits(ppd, _QIB_EVENT_LID_CHANGE_BIT);
+		if (ppd->lmc != (pip->mkeyprot_resv_lmc & 7))
+			qib_set_uevent_bits(ppd, _QIB_EVENT_LMC_CHANGE_BIT);
+		qib_set_lid(ppd, lid, pip->mkeyprot_resv_lmc & 7);
+		event.event = IB_EVENT_LID_CHANGE;
+		ib_dispatch_event(&event);
+	}
+
+	smlid = be16_to_cpu(pip->sm_lid);
+	msl = pip->neighbormtu_mastersmsl & 0xF;
+	/* Must be a valid unicast LID address. */
+	if (smlid == 0 || smlid >= QIB_MULTICAST_LID_BASE)
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else if (smlid != ibp->sm_lid || msl != ibp->sm_sl) {
+		spin_lock_irqsave(&ibp->lock, flags);
+		if (ibp->sm_ah) {
+			if (smlid != ibp->sm_lid)
+				ibp->sm_ah->attr.dlid = smlid;
+			if (msl != ibp->sm_sl)
+				ibp->sm_ah->attr.sl = msl;
+		}
+		spin_unlock_irqrestore(&ibp->lock, flags);
+		if (smlid != ibp->sm_lid)
+			ibp->sm_lid = smlid;
+		if (msl != ibp->sm_sl)
+			ibp->sm_sl = msl;
+		event.event = IB_EVENT_SM_CHANGE;
+		ib_dispatch_event(&event);
+	}
+
+	/* Allow 1x or 4x to be set (see 14.2.6.6). */
+	lwe = pip->link_width_enabled;
+	if (lwe) {
+		if (lwe == 0xFF)
+			set_link_width_enabled(ppd, ppd->link_width_supported);
+		else if (lwe >= 16 || (lwe & ~ppd->link_width_supported))
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else if (lwe != ppd->link_width_enabled)
+			set_link_width_enabled(ppd, lwe);
+	}
+
+	lse = pip->linkspeedactive_enabled & 0xF;
+	if (lse) {
+		/*
+		 * The IB 1.2 spec. only allows link speed values
+		 * 1, 3, 5, 7, 15.  1.2.1 extended to allow specific
+		 * speeds.
+		 */
+		if (lse == 15)
+			set_link_speed_enabled(ppd,
+					       ppd->link_speed_supported);
+		else if (lse >= 8 || (lse & ~ppd->link_speed_supported))
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else if (lse != ppd->link_speed_enabled)
+			set_link_speed_enabled(ppd, lse);
+	}
+
+	/* Set link down default state. */
+	switch (pip->portphysstate_linkdown & 0xF) {
+	case 0: /* NOP */
+		break;
+	case 1: /* SLEEP */
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
+					IB_LINKINITCMD_SLEEP);
+		break;
+	case 2: /* POLL */
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_LINKDEFAULT,
+					IB_LINKINITCMD_POLL);
+		break;
+	default:
+		smp->status |= IB_SMP_INVALID_FIELD;
+	}
+
+	ibp->mkeyprot = pip->mkeyprot_resv_lmc >> 6;
+	ibp->vl_high_limit = pip->vl_high_limit;
+	(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_VL_HIGH_LIMIT,
+				    ibp->vl_high_limit);
+
+	mtu = ib_mtu_enum_to_int((pip->neighbormtu_mastersmsl >> 4) & 0xF);
+	if (mtu == -1)
+		smp->status |= IB_SMP_INVALID_FIELD;
+	else
+		qib_set_mtu(ppd, mtu);
+
+	/* Set operational VLs */
+	vls = (pip->operationalvl_pei_peo_fpi_fpo >> 4) & 0xF;
+	if (vls) {
+		if (vls > ppd->vls_supported)
+			smp->status |= IB_SMP_INVALID_FIELD;
+		else
+			(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_OP_VLS, vls);
+	}
+
+	if (pip->mkey_violations == 0)
+		ibp->mkey_violations = 0;
+
+	if (pip->pkey_violations == 0)
+		ibp->pkey_violations = 0;
+
+	if (pip->qkey_violations == 0)
+		ibp->qkey_violations = 0;
+
+	ore = pip->localphyerrors_overrunerrors;
+	if (set_phyerrthreshold(ppd, (ore >> 4) & 0xF))
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	if (set_overrunthreshold(ppd, (ore & 0xF)))
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	ibp->subnet_timeout = pip->clientrereg_resv_subnetto & 0x1F;
+
+	/*
+	 * Do the port state change now that the other link parameters
+	 * have been set.
+	 * Changing the port physical state only makes sense if the link
+	 * is down or is being set to down.
+	 */
+	state = pip->linkspeed_portstate & 0xF;
+	lstate = (pip->portphysstate_linkdown >> 4) & 0xF;
+	if (lstate && !(state == IB_PORT_DOWN || state == IB_PORT_NOP))
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	/*
+	 * Only state changes of DOWN, ARM, and ACTIVE are valid
+	 * and must be in the correct state to take effect (see 7.2.6).
+	 */
+	switch (state) {
+	case IB_PORT_NOP:
+		if (lstate == 0)
+			break;
+		/* FALLTHROUGH */
+	case IB_PORT_DOWN:
+		if (lstate == 0)
+			lstate = QIB_IB_LINKDOWN_ONLY;
+		else if (lstate == 1)
+			lstate = QIB_IB_LINKDOWN_SLEEP;
+		else if (lstate == 2)
+			lstate = QIB_IB_LINKDOWN;
+		else if (lstate == 3)
+			lstate = QIB_IB_LINKDOWN_DISABLE;
+		else {
+			smp->status |= IB_SMP_INVALID_FIELD;
+			break;
+		}
+		spin_lock_irqsave(&ppd->lflags_lock, flags);
+		ppd->lflags &= ~QIBL_LINKV;
+		spin_unlock_irqrestore(&ppd->lflags_lock, flags);
+		qib_set_linkstate(ppd, lstate);
+		/*
+		 * Don't send a reply if the response would be sent
+		 * through the disabled port.
+		 */
+		if (lstate == QIB_IB_LINKDOWN_DISABLE && smp->hop_cnt) {
+			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+			goto done;
+		}
+		qib_wait_linkstate(ppd, QIBL_LINKV, 10);
+		break;
+	case IB_PORT_ARMED:
+		qib_set_linkstate(ppd, QIB_IB_LINKARM);
+		break;
+	case IB_PORT_ACTIVE:
+		qib_set_linkstate(ppd, QIB_IB_LINKACTIVE);
+		break;
+	default:
+		smp->status |= IB_SMP_INVALID_FIELD;
+	}
+
+	if (clientrereg) {
+		event.event = IB_EVENT_CLIENT_REREGISTER;
+		ib_dispatch_event(&event);
+	}
+
+	ret = subn_get_portinfo(smp, ibdev, port);
+
+	/* restore re-reg bit per o14-12.2.1 */
+	pip->clientrereg_resv_subnetto |= clientrereg;
+
+	goto get_only;
+
+err:
+	smp->status |= IB_SMP_INVALID_FIELD;
+get_only:
+	ret = subn_get_portinfo(smp, ibdev, port);
+done:
+	return ret;
+}
+
+/**
+ * rm_pkey - decrecment the reference count for the given PKEY
+ * @dd: the qlogic_ib device
+ * @key: the PKEY index
+ *
+ * Return true if this was the last reference and the hardware table entry
+ * needs to be changed.
+ */
+static int rm_pkey(struct qib_pportdata *ppd, u16 key)
+{
+	int i;
+	int ret;
+
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (ppd->pkeys[i] != key)
+			continue;
+		if (atomic_dec_and_test(&ppd->pkeyrefs[i])) {
+			ppd->pkeys[i] = 0;
+			ret = 1;
+			goto bail;
+		}
+		break;
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * add_pkey - add the given PKEY to the hardware table
+ * @dd: the qlogic_ib device
+ * @key: the PKEY
+ *
+ * Return an error code if unable to add the entry, zero if no change,
+ * or 1 if the hardware PKEY register needs to be updated.
+ */
+static int add_pkey(struct qib_pportdata *ppd, u16 key)
+{
+	int i;
+	u16 lkey = key & 0x7FFF;
+	int any = 0;
+	int ret;
+
+	if (lkey == 0x7FFF) {
+		ret = 0;
+		goto bail;
+	}
+
+	/* Look for an empty slot or a matching PKEY. */
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i]) {
+			any++;
+			continue;
+		}
+		/* If it matches exactly, try to increment the ref count */
+		if (ppd->pkeys[i] == key) {
+			if (atomic_inc_return(&ppd->pkeyrefs[i]) > 1) {
+				ret = 0;
+				goto bail;
+			}
+			/* Lost the race. Look for an empty slot below. */
+			atomic_dec(&ppd->pkeyrefs[i]);
+			any++;
+		}
+		/*
+		 * It makes no sense to have both the limited and unlimited
+		 * PKEY set at the same time since the unlimited one will
+		 * disable the limited one.
+		 */
+		if ((ppd->pkeys[i] & 0x7FFF) == lkey) {
+			ret = -EEXIST;
+			goto bail;
+		}
+	}
+	if (!any) {
+		ret = -EBUSY;
+		goto bail;
+	}
+	for (i = 0; i < ARRAY_SIZE(ppd->pkeys); i++) {
+		if (!ppd->pkeys[i] &&
+		    atomic_inc_return(&ppd->pkeyrefs[i]) == 1) {
+			/* for qibstats, etc. */
+			ppd->pkeys[i] = key;
+			ret = 1;
+			goto bail;
+		}
+	}
+	ret = -EBUSY;
+
+bail:
+	return ret;
+}
+
+/**
+ * set_pkeys - set the PKEY table for ctxt 0
+ * @dd: the qlogic_ib device
+ * @port: the IB port number
+ * @pkeys: the PKEY table
+ */
+static int set_pkeys(struct qib_devdata *dd, u8 port, u16 *pkeys)
+{
+	struct qib_pportdata *ppd;
+	struct qib_ctxtdata *rcd;
+	int i;
+	int changed = 0;
+
+	/*
+	 * IB port one/two always maps to context zero/one,
+	 * always a kernel context, no locking needed
+	 * If we get here with ppd setup, no need to check
+	 * that rcd is valid.
+	 */
+	ppd = dd->pport + (port - 1);
+	rcd = dd->rcd[ppd->hw_pidx];
+
+	for (i = 0; i < ARRAY_SIZE(rcd->pkeys); i++) {
+		u16 key = pkeys[i];
+		u16 okey = rcd->pkeys[i];
+
+		if (key == okey)
+			continue;
+		/*
+		 * The value of this PKEY table entry is changing.
+		 * Remove the old entry in the hardware's array of PKEYs.
+		 */
+		if (okey & 0x7FFF)
+			changed |= rm_pkey(ppd, okey);
+		if (key & 0x7FFF) {
+			int ret = add_pkey(ppd, key);
+
+			if (ret < 0)
+				key = 0;
+			else
+				changed |= ret;
+		}
+		rcd->pkeys[i] = key;
+	}
+	if (changed) {
+		struct ib_event event;
+
+		(void) dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PKEYS, 0);
+
+		event.event = IB_EVENT_PKEY_CHANGE;
+		event.device = &dd->verbs_dev.ibdev;
+		event.element.port_num = port;
+		ib_dispatch_event(&event);
+	}
+	return 0;
+}
+
+static int subn_set_pkeytable(struct ib_smp *smp, struct ib_device *ibdev,
+			      u8 port)
+{
+	u32 startpx = 32 * (be32_to_cpu(smp->attr_mod) & 0xffff);
+	__be16 *p = (__be16 *) smp->data;
+	u16 *q = (u16 *) smp->data;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	unsigned i, n = qib_get_npkeys(dd);
+
+	for (i = 0; i < n; i++)
+		q[i] = be16_to_cpu(p[i]);
+
+	if (startpx != 0 || set_pkeys(dd, port, q) != 0)
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return subn_get_pkeytable(smp, ibdev, port);
+}
+
+static int subn_get_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	u8 *p = (u8 *) smp->data;
+	unsigned i;
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (!(ibp->port_cap_flags & IB_PORT_SL_MAP_SUP))
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else
+		for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2)
+			*p++ = (ibp->sl_to_vl[i] << 4) | ibp->sl_to_vl[i + 1];
+
+	return reply(smp);
+}
+
+static int subn_set_sl_to_vl(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	u8 *p = (u8 *) smp->data;
+	unsigned i;
+
+	if (!(ibp->port_cap_flags & IB_PORT_SL_MAP_SUP)) {
+		smp->status |= IB_SMP_UNSUP_METHOD;
+		return reply(smp);
+	}
+
+	for (i = 0; i < ARRAY_SIZE(ibp->sl_to_vl); i += 2, p++) {
+		ibp->sl_to_vl[i] = *p >> 4;
+		ibp->sl_to_vl[i + 1] = *p & 0xF;
+	}
+	qib_set_uevent_bits(ppd_from_ibp(to_iport(ibdev, port)),
+			    _QIB_EVENT_SL2VL_CHANGE_BIT);
+
+	return subn_get_sl_to_vl(smp, ibdev, port);
+}
+
+static int subn_get_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
+			   u8 port)
+{
+	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
+	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
+
+	memset(smp->data, 0, sizeof(smp->data));
+
+	if (ppd->vls_supported == IB_VL_VL0)
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else if (which == IB_VLARB_LOWPRI_0_31)
+		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
+						   smp->data);
+	else if (which == IB_VLARB_HIGHPRI_0_31)
+		(void) ppd->dd->f_get_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
+						   smp->data);
+	else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return reply(smp);
+}
+
+static int subn_set_vl_arb(struct ib_smp *smp, struct ib_device *ibdev,
+			   u8 port)
+{
+	unsigned which = be32_to_cpu(smp->attr_mod) >> 16;
+	struct qib_pportdata *ppd = ppd_from_ibp(to_iport(ibdev, port));
+
+	if (ppd->vls_supported == IB_VL_VL0)
+		smp->status |= IB_SMP_UNSUP_METHOD;
+	else if (which == IB_VLARB_LOWPRI_0_31)
+		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_LOW_ARB,
+						   smp->data);
+	else if (which == IB_VLARB_HIGHPRI_0_31)
+		(void) ppd->dd->f_set_ib_table(ppd, QIB_IB_TBL_VL_HIGH_ARB,
+						   smp->data);
+	else
+		smp->status |= IB_SMP_INVALID_FIELD;
+
+	return subn_get_vl_arb(smp, ibdev, port);
+}
+
+static int subn_trap_repress(struct ib_smp *smp, struct ib_device *ibdev,
+			     u8 port)
+{
+	/*
+	 * For now, we only send the trap once so no need to process this.
+	 * o13-6, o13-7,
+	 * o14-3.a4 The SMA shall not send any message in response to a valid
+	 * SubnTrapRepress() message.
+	 */
+	return IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+}
+
+static int pma_get_classportinfo(struct ib_pma_mad *pmp,
+				 struct ib_device *ibdev)
+{
+	struct ib_class_port_info *p =
+		(struct ib_class_port_info *)pmp->data;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	if (pmp->mad_hdr.attr_mod != 0)
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+
+	/* Note that AllPortSelect is not valid */
+	p->base_version = 1;
+	p->class_version = 1;
+	p->capability_mask = IB_PMA_CLASS_CAP_EXT_WIDTH;
+	/*
+	 * Set the most significant bit of CM2 to indicate support for
+	 * congestion statistics
+	 */
+	p->reserved[0] = dd->psxmitwait_supported << 7;
+	/*
+	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
+	 */
+	p->resp_time_value = 18;
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portsamplescontrol(struct ib_pma_mad *pmp,
+				      struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplescontrol *p =
+		(struct ib_pma_portsamplescontrol *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 port_select = p->port_select;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->mad_hdr.attr_mod != 0 || port_select != port) {
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+		goto bail;
+	}
+	spin_lock_irqsave(&ibp->lock, flags);
+	p->tick = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_PMA_TICKS);
+	p->sample_status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	p->counter_width = 4;   /* 32 bit counters */
+	p->counter_mask0_9 = COUNTER_MASK0_9;
+	p->sample_start = cpu_to_be32(ibp->pma_sample_start);
+	p->sample_interval = cpu_to_be32(ibp->pma_sample_interval);
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	p->counter_select[0] = ibp->pma_counter_select[0];
+	p->counter_select[1] = ibp->pma_counter_select[1];
+	p->counter_select[2] = ibp->pma_counter_select[2];
+	p->counter_select[3] = ibp->pma_counter_select[3];
+	p->counter_select[4] = ibp->pma_counter_select[4];
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+bail:
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_set_portsamplescontrol(struct ib_pma_mad *pmp,
+				      struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplescontrol *p =
+		(struct ib_pma_portsamplescontrol *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 status, xmit_flags;
+	int ret;
+
+	if (pmp->mad_hdr.attr_mod != 0 || p->port_select != port) {
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+		ret = reply((struct ib_smp *) pmp);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&ibp->lock, flags);
+
+	/* Port Sampling code owns the PS* HW counters */
+	xmit_flags = ppd->cong_stats.flags;
+	ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_SAMPLE;
+	status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+	if (status == IB_PMA_SAMPLE_STATUS_DONE ||
+	    (status == IB_PMA_SAMPLE_STATUS_RUNNING &&
+	     xmit_flags == IB_PMA_CONG_HW_CONTROL_TIMER)) {
+		ibp->pma_sample_start = be32_to_cpu(p->sample_start);
+		ibp->pma_sample_interval = be32_to_cpu(p->sample_interval);
+		ibp->pma_tag = be16_to_cpu(p->tag);
+		ibp->pma_counter_select[0] = p->counter_select[0];
+		ibp->pma_counter_select[1] = p->counter_select[1];
+		ibp->pma_counter_select[2] = p->counter_select[2];
+		ibp->pma_counter_select[3] = p->counter_select[3];
+		ibp->pma_counter_select[4] = p->counter_select[4];
+		dd->f_set_cntr_sample(ppd, ibp->pma_sample_interval,
+				      ibp->pma_sample_start);
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	ret = pma_get_portsamplescontrol(pmp, ibdev, port);
+
+bail:
+	return ret;
+}
+
+static u64 get_counter(struct qib_ibport *ibp, struct qib_pportdata *ppd,
+		       __be16 sel)
+{
+	u64 ret;
+
+	switch (sel) {
+	case IB_PMA_PORT_XMIT_DATA:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITDATA);
+		break;
+	case IB_PMA_PORT_RCV_DATA:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVDATA);
+		break;
+	case IB_PMA_PORT_XMIT_PKTS:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITPKTS);
+		break;
+	case IB_PMA_PORT_RCV_PKTS:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSRCVPKTS);
+		break;
+	case IB_PMA_PORT_XMIT_WAIT:
+		ret = ppd->dd->f_portcntr(ppd, QIBPORTCNTR_PSXMITWAIT);
+		break;
+	default:
+		ret = 0;
+	}
+
+	return ret;
+}
+
+/* This function assumes that the xmit_wait lock is already held */
+static u64 xmit_wait_get_value_delta(struct qib_pportdata *ppd)
+{
+	u32 delta;
+
+	delta = get_counter(&ppd->ibport_data, ppd,
+			    IB_PMA_PORT_XMIT_WAIT);
+	return ppd->cong_stats.counter + delta;
+}
+
+static void cache_hw_sample_counters(struct qib_pportdata *ppd)
+{
+	struct qib_ibport *ibp = &ppd->ibport_data;
+
+	ppd->cong_stats.counter_cache.psxmitdata =
+		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_DATA);
+	ppd->cong_stats.counter_cache.psrcvdata =
+		get_counter(ibp, ppd, IB_PMA_PORT_RCV_DATA);
+	ppd->cong_stats.counter_cache.psxmitpkts =
+		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_PKTS);
+	ppd->cong_stats.counter_cache.psrcvpkts =
+		get_counter(ibp, ppd, IB_PMA_PORT_RCV_PKTS);
+	ppd->cong_stats.counter_cache.psxmitwait =
+		get_counter(ibp, ppd, IB_PMA_PORT_XMIT_WAIT);
+}
+
+static u64 get_cache_hw_sample_counters(struct qib_pportdata *ppd,
+					__be16 sel)
+{
+	u64 ret;
+
+	switch (sel) {
+	case IB_PMA_PORT_XMIT_DATA:
+		ret = ppd->cong_stats.counter_cache.psxmitdata;
+		break;
+	case IB_PMA_PORT_RCV_DATA:
+		ret = ppd->cong_stats.counter_cache.psrcvdata;
+		break;
+	case IB_PMA_PORT_XMIT_PKTS:
+		ret = ppd->cong_stats.counter_cache.psxmitpkts;
+		break;
+	case IB_PMA_PORT_RCV_PKTS:
+		ret = ppd->cong_stats.counter_cache.psrcvpkts;
+		break;
+	case IB_PMA_PORT_XMIT_WAIT:
+		ret = ppd->cong_stats.counter_cache.psxmitwait;
+		break;
+	default:
+		ret = 0;
+	}
+
+	return ret;
+}
+
+static int pma_get_portsamplesresult(struct ib_pma_mad *pmp,
+				     struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplesresult *p =
+		(struct ib_pma_portsamplesresult *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 status;
+	int i;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+	spin_lock_irqsave(&ibp->lock, flags);
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_TIMER)
+		p->sample_status = IB_PMA_SAMPLE_STATUS_DONE;
+	else {
+		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+		p->sample_status = cpu_to_be16(status);
+		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
+			cache_hw_sample_counters(ppd);
+			ppd->cong_stats.counter =
+				xmit_wait_get_value_delta(ppd);
+			dd->f_set_cntr_sample(ppd,
+					      QIB_CONG_TIMER_PSINTERVAL, 0);
+			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
+		}
+	}
+	for (i = 0; i < ARRAY_SIZE(ibp->pma_counter_select); i++)
+		p->counter[i] = cpu_to_be32(
+			get_cache_hw_sample_counters(
+				ppd, ibp->pma_counter_select[i]));
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portsamplesresult_ext(struct ib_pma_mad *pmp,
+					 struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portsamplesresult_ext *p =
+		(struct ib_pma_portsamplesresult_ext *)pmp->data;
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	unsigned long flags;
+	u8 status;
+	int i;
+
+	/* Port Sampling code owns the PS* HW counters */
+	memset(pmp->data, 0, sizeof(pmp->data));
+	spin_lock_irqsave(&ibp->lock, flags);
+	p->tag = cpu_to_be16(ibp->pma_tag);
+	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_TIMER)
+		p->sample_status = IB_PMA_SAMPLE_STATUS_DONE;
+	else {
+		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+		p->sample_status = cpu_to_be16(status);
+		/* 64 bits */
+		p->extended_width = cpu_to_be32(0x80000000);
+		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
+			cache_hw_sample_counters(ppd);
+			ppd->cong_stats.counter =
+				xmit_wait_get_value_delta(ppd);
+			dd->f_set_cntr_sample(ppd,
+					      QIB_CONG_TIMER_PSINTERVAL, 0);
+			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
+		}
+	}
+	for (i = 0; i < ARRAY_SIZE(ibp->pma_counter_select); i++)
+		p->counter[i] = cpu_to_be64(
+			get_cache_hw_sample_counters(
+				ppd, ibp->pma_counter_select[i]));
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portcounters(struct ib_pma_mad *pmp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_counters cntrs;
+	u8 port_select = p->port_select;
+
+	qib_get_counters(ppd, &cntrs);
+
+	/* Adjust counters for any resets done. */
+	cntrs.symbol_error_counter -= ibp->z_symbol_error_counter;
+	cntrs.link_error_recovery_counter -=
+		ibp->z_link_error_recovery_counter;
+	cntrs.link_downed_counter -= ibp->z_link_downed_counter;
+	cntrs.port_rcv_errors -= ibp->z_port_rcv_errors;
+	cntrs.port_rcv_remphys_errors -= ibp->z_port_rcv_remphys_errors;
+	cntrs.port_xmit_discards -= ibp->z_port_xmit_discards;
+	cntrs.port_xmit_data -= ibp->z_port_xmit_data;
+	cntrs.port_rcv_data -= ibp->z_port_rcv_data;
+	cntrs.port_xmit_packets -= ibp->z_port_xmit_packets;
+	cntrs.port_rcv_packets -= ibp->z_port_rcv_packets;
+	cntrs.local_link_integrity_errors -=
+		ibp->z_local_link_integrity_errors;
+	cntrs.excessive_buffer_overrun_errors -=
+		ibp->z_excessive_buffer_overrun_errors;
+	cntrs.vl15_dropped -= ibp->z_vl15_dropped;
+	cntrs.vl15_dropped += ibp->n_vl15_dropped;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->mad_hdr.attr_mod != 0 || port_select != port)
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+
+	if (cntrs.symbol_error_counter > 0xFFFFUL)
+		p->symbol_error_counter = cpu_to_be16(0xFFFF);
+	else
+		p->symbol_error_counter =
+			cpu_to_be16((u16)cntrs.symbol_error_counter);
+	if (cntrs.link_error_recovery_counter > 0xFFUL)
+		p->link_error_recovery_counter = 0xFF;
+	else
+		p->link_error_recovery_counter =
+			(u8)cntrs.link_error_recovery_counter;
+	if (cntrs.link_downed_counter > 0xFFUL)
+		p->link_downed_counter = 0xFF;
+	else
+		p->link_downed_counter = (u8)cntrs.link_downed_counter;
+	if (cntrs.port_rcv_errors > 0xFFFFUL)
+		p->port_rcv_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_errors =
+			cpu_to_be16((u16) cntrs.port_rcv_errors);
+	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
+		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_remphys_errors =
+			cpu_to_be16((u16)cntrs.port_rcv_remphys_errors);
+	if (cntrs.port_xmit_discards > 0xFFFFUL)
+		p->port_xmit_discards = cpu_to_be16(0xFFFF);
+	else
+		p->port_xmit_discards =
+			cpu_to_be16((u16)cntrs.port_xmit_discards);
+	if (cntrs.local_link_integrity_errors > 0xFUL)
+		cntrs.local_link_integrity_errors = 0xFUL;
+	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
+		cntrs.excessive_buffer_overrun_errors = 0xFUL;
+	p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
+		cntrs.excessive_buffer_overrun_errors;
+	if (cntrs.vl15_dropped > 0xFFFFUL)
+		p->vl15_dropped = cpu_to_be16(0xFFFF);
+	else
+		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
+	if (cntrs.port_xmit_data > 0xFFFFFFFFUL)
+		p->port_xmit_data = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_xmit_data = cpu_to_be32((u32)cntrs.port_xmit_data);
+	if (cntrs.port_rcv_data > 0xFFFFFFFFUL)
+		p->port_rcv_data = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_rcv_data = cpu_to_be32((u32)cntrs.port_rcv_data);
+	if (cntrs.port_xmit_packets > 0xFFFFFFFFUL)
+		p->port_xmit_packets = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_xmit_packets =
+			cpu_to_be32((u32)cntrs.port_xmit_packets);
+	if (cntrs.port_rcv_packets > 0xFFFFFFFFUL)
+		p->port_rcv_packets = cpu_to_be32(0xFFFFFFFF);
+	else
+		p->port_rcv_packets =
+			cpu_to_be32((u32) cntrs.port_rcv_packets);
+
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_get_portcounters_cong(struct ib_pma_mad *pmp,
+				     struct ib_device *ibdev, u8 port)
+{
+	/* Congestion PMA packets start at offset 24 not 64 */
+	struct ib_pma_portcounters_cong *p =
+		(struct ib_pma_portcounters_cong *)pmp->reserved;
+	struct qib_verbs_counters cntrs;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = dd_from_ppd(ppd);
+	u32 port_select = be32_to_cpu(pmp->mad_hdr.attr_mod) & 0xFF;
+	u64 xmit_wait_counter;
+	unsigned long flags;
+
+	/*
+	 * This check is performed only in the GET method because the
+	 * SET method ends up calling this anyway.
+	 */
+	if (!dd->psxmitwait_supported)
+		pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
+	if (port_select != port)
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+
+	qib_get_counters(ppd, &cntrs);
+	spin_lock_irqsave(&ppd->ibport_data.lock, flags);
+	xmit_wait_counter = xmit_wait_get_value_delta(ppd);
+	spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
+
+	/* Adjust counters for any resets done. */
+	cntrs.symbol_error_counter -= ibp->z_symbol_error_counter;
+	cntrs.link_error_recovery_counter -=
+		ibp->z_link_error_recovery_counter;
+	cntrs.link_downed_counter -= ibp->z_link_downed_counter;
+	cntrs.port_rcv_errors -= ibp->z_port_rcv_errors;
+	cntrs.port_rcv_remphys_errors -=
+		ibp->z_port_rcv_remphys_errors;
+	cntrs.port_xmit_discards -= ibp->z_port_xmit_discards;
+	cntrs.local_link_integrity_errors -=
+		ibp->z_local_link_integrity_errors;
+	cntrs.excessive_buffer_overrun_errors -=
+		ibp->z_excessive_buffer_overrun_errors;
+	cntrs.vl15_dropped -= ibp->z_vl15_dropped;
+	cntrs.vl15_dropped += ibp->n_vl15_dropped;
+	cntrs.port_xmit_data -= ibp->z_port_xmit_data;
+	cntrs.port_rcv_data -= ibp->z_port_rcv_data;
+	cntrs.port_xmit_packets -= ibp->z_port_xmit_packets;
+	cntrs.port_rcv_packets -= ibp->z_port_rcv_packets;
+
+	memset(pmp->reserved, 0, sizeof(pmp->reserved) +
+	       sizeof(pmp->data));
+
+	/*
+	 * Set top 3 bits to indicate interval in picoseconds in
+	 * remaining bits.
+	 */
+	p->port_check_rate =
+		cpu_to_be16((QIB_XMIT_RATE_PICO << 13) |
+			    (dd->psxmitwait_check_rate &
+			     ~(QIB_XMIT_RATE_PICO << 13)));
+	p->port_adr_events = cpu_to_be64(0);
+	p->port_xmit_wait = cpu_to_be64(xmit_wait_counter);
+	p->port_xmit_data = cpu_to_be64(cntrs.port_xmit_data);
+	p->port_rcv_data = cpu_to_be64(cntrs.port_rcv_data);
+	p->port_xmit_packets =
+		cpu_to_be64(cntrs.port_xmit_packets);
+	p->port_rcv_packets =
+		cpu_to_be64(cntrs.port_rcv_packets);
+	if (cntrs.symbol_error_counter > 0xFFFFUL)
+		p->symbol_error_counter = cpu_to_be16(0xFFFF);
+	else
+		p->symbol_error_counter =
+			cpu_to_be16(
+				(u16)cntrs.symbol_error_counter);
+	if (cntrs.link_error_recovery_counter > 0xFFUL)
+		p->link_error_recovery_counter = 0xFF;
+	else
+		p->link_error_recovery_counter =
+			(u8)cntrs.link_error_recovery_counter;
+	if (cntrs.link_downed_counter > 0xFFUL)
+		p->link_downed_counter = 0xFF;
+	else
+		p->link_downed_counter =
+			(u8)cntrs.link_downed_counter;
+	if (cntrs.port_rcv_errors > 0xFFFFUL)
+		p->port_rcv_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_errors =
+			cpu_to_be16((u16) cntrs.port_rcv_errors);
+	if (cntrs.port_rcv_remphys_errors > 0xFFFFUL)
+		p->port_rcv_remphys_errors = cpu_to_be16(0xFFFF);
+	else
+		p->port_rcv_remphys_errors =
+			cpu_to_be16(
+				(u16)cntrs.port_rcv_remphys_errors);
+	if (cntrs.port_xmit_discards > 0xFFFFUL)
+		p->port_xmit_discards = cpu_to_be16(0xFFFF);
+	else
+		p->port_xmit_discards =
+			cpu_to_be16((u16)cntrs.port_xmit_discards);
+	if (cntrs.local_link_integrity_errors > 0xFUL)
+		cntrs.local_link_integrity_errors = 0xFUL;
+	if (cntrs.excessive_buffer_overrun_errors > 0xFUL)
+		cntrs.excessive_buffer_overrun_errors = 0xFUL;
+	p->link_overrun_errors = (cntrs.local_link_integrity_errors << 4) |
+		cntrs.excessive_buffer_overrun_errors;
+	if (cntrs.vl15_dropped > 0xFFFFUL)
+		p->vl15_dropped = cpu_to_be16(0xFFFF);
+	else
+		p->vl15_dropped = cpu_to_be16((u16)cntrs.vl15_dropped);
+
+	return reply((struct ib_smp *)pmp);
+}
+
+static void qib_snapshot_pmacounters(
+	struct qib_ibport *ibp,
+	struct qib_pma_counters *pmacounters)
+{
+	struct qib_pma_counters *p;
+	int cpu;
+
+	memset(pmacounters, 0, sizeof(*pmacounters));
+	for_each_possible_cpu(cpu) {
+		p = per_cpu_ptr(ibp->pmastats, cpu);
+		pmacounters->n_unicast_xmit += p->n_unicast_xmit;
+		pmacounters->n_unicast_rcv += p->n_unicast_rcv;
+		pmacounters->n_multicast_xmit += p->n_multicast_xmit;
+		pmacounters->n_multicast_rcv += p->n_multicast_rcv;
+	}
+}
+
+static int pma_get_portcounters_ext(struct ib_pma_mad *pmp,
+				    struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters_ext *p =
+		(struct ib_pma_portcounters_ext *)pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 swords, rwords, spkts, rpkts, xwait;
+	struct qib_pma_counters pma;
+	u8 port_select = p->port_select;
+
+	memset(pmp->data, 0, sizeof(pmp->data));
+
+	p->port_select = port_select;
+	if (pmp->mad_hdr.attr_mod != 0 || port_select != port) {
+		pmp->mad_hdr.status |= IB_SMP_INVALID_FIELD;
+		goto bail;
+	}
+
+	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
+
+	/* Adjust counters for any resets done. */
+	swords -= ibp->z_port_xmit_data;
+	rwords -= ibp->z_port_rcv_data;
+	spkts -= ibp->z_port_xmit_packets;
+	rpkts -= ibp->z_port_rcv_packets;
+
+	p->port_xmit_data = cpu_to_be64(swords);
+	p->port_rcv_data = cpu_to_be64(rwords);
+	p->port_xmit_packets = cpu_to_be64(spkts);
+	p->port_rcv_packets = cpu_to_be64(rpkts);
+
+	qib_snapshot_pmacounters(ibp, &pma);
+
+	p->port_unicast_xmit_packets = cpu_to_be64(pma.n_unicast_xmit
+		- ibp->z_unicast_xmit);
+	p->port_unicast_rcv_packets = cpu_to_be64(pma.n_unicast_rcv
+		- ibp->z_unicast_rcv);
+	p->port_multicast_xmit_packets = cpu_to_be64(pma.n_multicast_xmit
+		- ibp->z_multicast_xmit);
+	p->port_multicast_rcv_packets = cpu_to_be64(pma.n_multicast_rcv
+		- ibp->z_multicast_rcv);
+
+bail:
+	return reply((struct ib_smp *) pmp);
+}
+
+static int pma_set_portcounters(struct ib_pma_mad *pmp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_counters cntrs;
+
+	/*
+	 * Since the HW doesn't support clearing counters, we save the
+	 * current count and subtract it from future responses.
+	 */
+	qib_get_counters(ppd, &cntrs);
+
+	if (p->counter_select & IB_PMA_SEL_SYMBOL_ERROR)
+		ibp->z_symbol_error_counter = cntrs.symbol_error_counter;
+
+	if (p->counter_select & IB_PMA_SEL_LINK_ERROR_RECOVERY)
+		ibp->z_link_error_recovery_counter =
+			cntrs.link_error_recovery_counter;
+
+	if (p->counter_select & IB_PMA_SEL_LINK_DOWNED)
+		ibp->z_link_downed_counter = cntrs.link_downed_counter;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_ERRORS)
+		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_REMPHYS_ERRORS)
+		ibp->z_port_rcv_remphys_errors =
+			cntrs.port_rcv_remphys_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DISCARDS)
+		ibp->z_port_xmit_discards = cntrs.port_xmit_discards;
+
+	if (p->counter_select & IB_PMA_SEL_LOCAL_LINK_INTEGRITY_ERRORS)
+		ibp->z_local_link_integrity_errors =
+			cntrs.local_link_integrity_errors;
+
+	if (p->counter_select & IB_PMA_SEL_EXCESSIVE_BUFFER_OVERRUNS)
+		ibp->z_excessive_buffer_overrun_errors =
+			cntrs.excessive_buffer_overrun_errors;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_VL15_DROPPED) {
+		ibp->n_vl15_dropped = 0;
+		ibp->z_vl15_dropped = cntrs.vl15_dropped;
+	}
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_DATA)
+		ibp->z_port_xmit_data = cntrs.port_xmit_data;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_DATA)
+		ibp->z_port_rcv_data = cntrs.port_rcv_data;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_XMIT_PACKETS)
+		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+
+	if (p->counter_select & IB_PMA_SEL_PORT_RCV_PACKETS)
+		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+
+	return pma_get_portcounters(pmp, ibdev, port);
+}
+
+static int pma_set_portcounters_cong(struct ib_pma_mad *pmp,
+				     struct ib_device *ibdev, u8 port)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = dd_from_ppd(ppd);
+	struct qib_verbs_counters cntrs;
+	u32 counter_select = (be32_to_cpu(pmp->mad_hdr.attr_mod) >> 24) & 0xFF;
+	int ret = 0;
+	unsigned long flags;
+
+	qib_get_counters(ppd, &cntrs);
+	/* Get counter values before we save them */
+	ret = pma_get_portcounters_cong(pmp, ibdev, port);
+
+	if (counter_select & IB_PMA_SEL_CONG_XMIT) {
+		spin_lock_irqsave(&ppd->ibport_data.lock, flags);
+		ppd->cong_stats.counter = 0;
+		dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL,
+				      0x0);
+		spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
+	}
+	if (counter_select & IB_PMA_SEL_CONG_PORT_DATA) {
+		ibp->z_port_xmit_data = cntrs.port_xmit_data;
+		ibp->z_port_rcv_data = cntrs.port_rcv_data;
+		ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+		ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+	}
+	if (counter_select & IB_PMA_SEL_CONG_ALL) {
+		ibp->z_symbol_error_counter =
+			cntrs.symbol_error_counter;
+		ibp->z_link_error_recovery_counter =
+			cntrs.link_error_recovery_counter;
+		ibp->z_link_downed_counter =
+			cntrs.link_downed_counter;
+		ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+		ibp->z_port_rcv_remphys_errors =
+			cntrs.port_rcv_remphys_errors;
+		ibp->z_port_xmit_discards =
+			cntrs.port_xmit_discards;
+		ibp->z_local_link_integrity_errors =
+			cntrs.local_link_integrity_errors;
+		ibp->z_excessive_buffer_overrun_errors =
+			cntrs.excessive_buffer_overrun_errors;
+		ibp->n_vl15_dropped = 0;
+		ibp->z_vl15_dropped = cntrs.vl15_dropped;
+	}
+
+	return ret;
+}
+
+static int pma_set_portcounters_ext(struct ib_pma_mad *pmp,
+				    struct ib_device *ibdev, u8 port)
+{
+	struct ib_pma_portcounters *p = (struct ib_pma_portcounters *)
+		pmp->data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 swords, rwords, spkts, rpkts, xwait;
+	struct qib_pma_counters pma;
+
+	qib_snapshot_counters(ppd, &swords, &rwords, &spkts, &rpkts, &xwait);
+
+	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_DATA)
+		ibp->z_port_xmit_data = swords;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_RCV_DATA)
+		ibp->z_port_rcv_data = rwords;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_XMIT_PACKETS)
+		ibp->z_port_xmit_packets = spkts;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_RCV_PACKETS)
+		ibp->z_port_rcv_packets = rpkts;
+
+	qib_snapshot_pmacounters(ibp, &pma);
+
+	if (p->counter_select & IB_PMA_SELX_PORT_UNI_XMIT_PACKETS)
+		ibp->z_unicast_xmit = pma.n_unicast_xmit;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_UNI_RCV_PACKETS)
+		ibp->z_unicast_rcv = pma.n_unicast_rcv;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_XMIT_PACKETS)
+		ibp->z_multicast_xmit = pma.n_multicast_xmit;
+
+	if (p->counter_select & IB_PMA_SELX_PORT_MULTI_RCV_PACKETS)
+		ibp->z_multicast_rcv = pma.n_multicast_rcv;
+
+	return pma_get_portcounters_ext(pmp, ibdev, port);
+}
+
+static int process_subn(struct ib_device *ibdev, int mad_flags,
+			u8 port, struct ib_mad *in_mad,
+			struct ib_mad *out_mad)
+{
+	struct ib_smp *smp = (struct ib_smp *)out_mad;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int ret;
+
+	*out_mad = *in_mad;
+	if (smp->class_version != 1) {
+		smp->status |= IB_SMP_UNSUP_VERSION;
+		ret = reply(smp);
+		goto bail;
+	}
+
+	ret = check_mkey(ibp, smp, mad_flags);
+	if (ret) {
+		u32 port_num = be32_to_cpu(smp->attr_mod);
+
+		/*
+		 * If this is a get/set portinfo, we already check the
+		 * M_Key if the MAD is for another port and the M_Key
+		 * is OK on the receiving port. This check is needed
+		 * to increment the error counters when the M_Key
+		 * fails to match on *both* ports.
+		 */
+		if (in_mad->mad_hdr.attr_id == IB_SMP_ATTR_PORT_INFO &&
+		    (smp->method == IB_MGMT_METHOD_GET ||
+		     smp->method == IB_MGMT_METHOD_SET) &&
+		    port_num && port_num <= ibdev->phys_port_cnt &&
+		    port != port_num)
+			(void) check_mkey(to_iport(ibdev, port_num), smp, 0);
+		ret = IB_MAD_RESULT_FAILURE;
+		goto bail;
+	}
+
+	switch (smp->method) {
+	case IB_MGMT_METHOD_GET:
+		switch (smp->attr_id) {
+		case IB_SMP_ATTR_NODE_DESC:
+			ret = subn_get_nodedescription(smp, ibdev);
+			goto bail;
+		case IB_SMP_ATTR_NODE_INFO:
+			ret = subn_get_nodeinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_GUID_INFO:
+			ret = subn_get_guidinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PORT_INFO:
+			ret = subn_get_portinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PKEY_TABLE:
+			ret = subn_get_pkeytable(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SL_TO_VL_TABLE:
+			ret = subn_get_sl_to_vl(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_VL_ARB_TABLE:
+			ret = subn_get_vl_arb(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SM_INFO:
+			if (ibp->port_cap_flags & IB_PORT_SM_DISABLED) {
+				ret = IB_MAD_RESULT_SUCCESS |
+					IB_MAD_RESULT_CONSUMED;
+				goto bail;
+			}
+			if (ibp->port_cap_flags & IB_PORT_SM) {
+				ret = IB_MAD_RESULT_SUCCESS;
+				goto bail;
+			}
+			/* FALLTHROUGH */
+		default:
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_SET:
+		switch (smp->attr_id) {
+		case IB_SMP_ATTR_GUID_INFO:
+			ret = subn_set_guidinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PORT_INFO:
+			ret = subn_set_portinfo(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_PKEY_TABLE:
+			ret = subn_set_pkeytable(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SL_TO_VL_TABLE:
+			ret = subn_set_sl_to_vl(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_VL_ARB_TABLE:
+			ret = subn_set_vl_arb(smp, ibdev, port);
+			goto bail;
+		case IB_SMP_ATTR_SM_INFO:
+			if (ibp->port_cap_flags & IB_PORT_SM_DISABLED) {
+				ret = IB_MAD_RESULT_SUCCESS |
+					IB_MAD_RESULT_CONSUMED;
+				goto bail;
+			}
+			if (ibp->port_cap_flags & IB_PORT_SM) {
+				ret = IB_MAD_RESULT_SUCCESS;
+				goto bail;
+			}
+			/* FALLTHROUGH */
+		default:
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_TRAP_REPRESS:
+		if (smp->attr_id == IB_SMP_ATTR_NOTICE)
+			ret = subn_trap_repress(smp, ibdev, port);
+		else {
+			smp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply(smp);
+		}
+		goto bail;
+
+	case IB_MGMT_METHOD_TRAP:
+	case IB_MGMT_METHOD_REPORT:
+	case IB_MGMT_METHOD_REPORT_RESP:
+	case IB_MGMT_METHOD_GET_RESP:
+		/*
+		 * The ib_mad module will call us to process responses
+		 * before checking for other consumers.
+		 * Just tell the caller to process it normally.
+		 */
+		ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	case IB_MGMT_METHOD_SEND:
+		if (ib_get_smp_direction(smp) &&
+		    smp->attr_id == QIB_VENDOR_IPG) {
+			ppd->dd->f_set_ib_cfg(ppd, QIB_IB_CFG_PORT,
+					      smp->data[0]);
+			ret = IB_MAD_RESULT_SUCCESS | IB_MAD_RESULT_CONSUMED;
+		} else
+			ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	default:
+		smp->status |= IB_SMP_UNSUP_METHOD;
+		ret = reply(smp);
+	}
+
+bail:
+	return ret;
+}
+
+static int process_perf(struct ib_device *ibdev, u8 port,
+			struct ib_mad *in_mad,
+			struct ib_mad *out_mad)
+{
+	struct ib_pma_mad *pmp = (struct ib_pma_mad *)out_mad;
+	int ret;
+
+	*out_mad = *in_mad;
+	if (pmp->mad_hdr.class_version != 1) {
+		pmp->mad_hdr.status |= IB_SMP_UNSUP_VERSION;
+		ret = reply((struct ib_smp *) pmp);
+		goto bail;
+	}
+
+	switch (pmp->mad_hdr.method) {
+	case IB_MGMT_METHOD_GET:
+		switch (pmp->mad_hdr.attr_id) {
+		case IB_PMA_CLASS_PORT_INFO:
+			ret = pma_get_classportinfo(pmp, ibdev);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_CONTROL:
+			ret = pma_get_portsamplescontrol(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_RESULT:
+			ret = pma_get_portsamplesresult(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_SAMPLES_RESULT_EXT:
+			ret = pma_get_portsamplesresult_ext(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS:
+			ret = pma_get_portcounters(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_EXT:
+			ret = pma_get_portcounters_ext(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_CONG:
+			ret = pma_get_portcounters_cong(pmp, ibdev, port);
+			goto bail;
+		default:
+			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) pmp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_SET:
+		switch (pmp->mad_hdr.attr_id) {
+		case IB_PMA_PORT_SAMPLES_CONTROL:
+			ret = pma_set_portsamplescontrol(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS:
+			ret = pma_set_portcounters(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_EXT:
+			ret = pma_set_portcounters_ext(pmp, ibdev, port);
+			goto bail;
+		case IB_PMA_PORT_COUNTERS_CONG:
+			ret = pma_set_portcounters_cong(pmp, ibdev, port);
+			goto bail;
+		default:
+			pmp->mad_hdr.status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) pmp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_TRAP:
+	case IB_MGMT_METHOD_GET_RESP:
+		/*
+		 * The ib_mad module will call us to process responses
+		 * before checking for other consumers.
+		 * Just tell the caller to process it normally.
+		 */
+		ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	default:
+		pmp->mad_hdr.status |= IB_SMP_UNSUP_METHOD;
+		ret = reply((struct ib_smp *) pmp);
+	}
+
+bail:
+	return ret;
+}
+
+static int cc_get_classportinfo(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev)
+{
+	struct ib_cc_classportinfo_attr *p =
+		(struct ib_cc_classportinfo_attr *)ccp->mgmt_data;
+
+	memset(ccp->mgmt_data, 0, sizeof(ccp->mgmt_data));
+
+	p->base_version = 1;
+	p->class_version = 1;
+	p->cap_mask = 0;
+
+	/*
+	 * Expected response time is 4.096 usec. * 2^18 == 1.073741824 sec.
+	 */
+	p->resp_time_value = 18;
+
+	return reply((struct ib_smp *) ccp);
+}
+
+static int cc_get_congestion_info(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_cc_info_attr *p =
+		(struct ib_cc_info_attr *)ccp->mgmt_data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+	memset(ccp->mgmt_data, 0, sizeof(ccp->mgmt_data));
+
+	p->congestion_info = 0;
+	p->control_table_cap = ppd->cc_max_table_entries;
+
+	return reply((struct ib_smp *) ccp);
+}
+
+static int cc_get_congestion_setting(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev, u8 port)
+{
+	int i;
+	struct ib_cc_congestion_setting_attr *p =
+		(struct ib_cc_congestion_setting_attr *)ccp->mgmt_data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct ib_cc_congestion_entry_shadow *entries;
+
+	memset(ccp->mgmt_data, 0, sizeof(ccp->mgmt_data));
+
+	spin_lock(&ppd->cc_shadow_lock);
+
+	entries = ppd->congestion_entries_shadow->entries;
+	p->port_control = cpu_to_be16(
+		ppd->congestion_entries_shadow->port_control);
+	p->control_map = cpu_to_be16(
+		ppd->congestion_entries_shadow->control_map);
+	for (i = 0; i < IB_CC_CCS_ENTRIES; i++) {
+		p->entries[i].ccti_increase = entries[i].ccti_increase;
+		p->entries[i].ccti_timer = cpu_to_be16(entries[i].ccti_timer);
+		p->entries[i].trigger_threshold = entries[i].trigger_threshold;
+		p->entries[i].ccti_min = entries[i].ccti_min;
+	}
+
+	spin_unlock(&ppd->cc_shadow_lock);
+
+	return reply((struct ib_smp *) ccp);
+}
+
+static int cc_get_congestion_control_table(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_cc_table_attr *p =
+		(struct ib_cc_table_attr *)ccp->mgmt_data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u32 cct_block_index = be32_to_cpu(ccp->attr_mod);
+	u32 max_cct_block;
+	u32 cct_entry;
+	struct ib_cc_table_entry_shadow *entries;
+	int i;
+
+	/* Is the table index more than what is supported? */
+	if (cct_block_index > IB_CC_TABLE_CAP_DEFAULT - 1)
+		goto bail;
+
+	memset(ccp->mgmt_data, 0, sizeof(ccp->mgmt_data));
+
+	spin_lock(&ppd->cc_shadow_lock);
+
+	max_cct_block =
+		(ppd->ccti_entries_shadow->ccti_last_entry + 1)/IB_CCT_ENTRIES;
+	max_cct_block = max_cct_block ? max_cct_block - 1 : 0;
+
+	if (cct_block_index > max_cct_block) {
+		spin_unlock(&ppd->cc_shadow_lock);
+		goto bail;
+	}
+
+	ccp->attr_mod = cpu_to_be32(cct_block_index);
+
+	cct_entry = IB_CCT_ENTRIES * (cct_block_index + 1);
+
+	cct_entry--;
+
+	p->ccti_limit = cpu_to_be16(cct_entry);
+
+	entries = &ppd->ccti_entries_shadow->
+			entries[IB_CCT_ENTRIES * cct_block_index];
+	cct_entry %= IB_CCT_ENTRIES;
+
+	for (i = 0; i <= cct_entry; i++)
+		p->ccti_entries[i].entry = cpu_to_be16(entries[i].entry);
+
+	spin_unlock(&ppd->cc_shadow_lock);
+
+	return reply((struct ib_smp *) ccp);
+
+bail:
+	return reply_failure((struct ib_smp *) ccp);
+}
+
+static int cc_set_congestion_setting(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_cc_congestion_setting_attr *p =
+		(struct ib_cc_congestion_setting_attr *)ccp->mgmt_data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int i;
+
+	ppd->cc_sl_control_map = be16_to_cpu(p->control_map);
+
+	for (i = 0; i < IB_CC_CCS_ENTRIES; i++) {
+		ppd->congestion_entries[i].ccti_increase =
+			p->entries[i].ccti_increase;
+
+		ppd->congestion_entries[i].ccti_timer =
+			be16_to_cpu(p->entries[i].ccti_timer);
+
+		ppd->congestion_entries[i].trigger_threshold =
+			p->entries[i].trigger_threshold;
+
+		ppd->congestion_entries[i].ccti_min =
+			p->entries[i].ccti_min;
+	}
+
+	return reply((struct ib_smp *) ccp);
+}
+
+static int cc_set_congestion_control_table(struct ib_cc_mad *ccp,
+				struct ib_device *ibdev, u8 port)
+{
+	struct ib_cc_table_attr *p =
+		(struct ib_cc_table_attr *)ccp->mgmt_data;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u32 cct_block_index = be32_to_cpu(ccp->attr_mod);
+	u32 cct_entry;
+	struct ib_cc_table_entry_shadow *entries;
+	int i;
+
+	/* Is the table index more than what is supported? */
+	if (cct_block_index > IB_CC_TABLE_CAP_DEFAULT - 1)
+		goto bail;
+
+	/* If this packet is the first in the sequence then
+	 * zero the total table entry count.
+	 */
+	if (be16_to_cpu(p->ccti_limit) < IB_CCT_ENTRIES)
+		ppd->total_cct_entry = 0;
+
+	cct_entry = (be16_to_cpu(p->ccti_limit))%IB_CCT_ENTRIES;
+
+	/* ccti_limit is 0 to 63 */
+	ppd->total_cct_entry += (cct_entry + 1);
+
+	if (ppd->total_cct_entry > ppd->cc_supported_table_entries)
+		goto bail;
+
+	ppd->ccti_limit = be16_to_cpu(p->ccti_limit);
+
+	entries = ppd->ccti_entries + (IB_CCT_ENTRIES * cct_block_index);
+
+	for (i = 0; i <= cct_entry; i++)
+		entries[i].entry = be16_to_cpu(p->ccti_entries[i].entry);
+
+	spin_lock(&ppd->cc_shadow_lock);
+
+	ppd->ccti_entries_shadow->ccti_last_entry = ppd->total_cct_entry - 1;
+	memcpy(ppd->ccti_entries_shadow->entries, ppd->ccti_entries,
+		(ppd->total_cct_entry * sizeof(struct ib_cc_table_entry)));
+
+	ppd->congestion_entries_shadow->port_control = IB_CC_CCS_PC_SL_BASED;
+	ppd->congestion_entries_shadow->control_map = ppd->cc_sl_control_map;
+	memcpy(ppd->congestion_entries_shadow->entries, ppd->congestion_entries,
+		IB_CC_CCS_ENTRIES * sizeof(struct ib_cc_congestion_entry));
+
+	spin_unlock(&ppd->cc_shadow_lock);
+
+	return reply((struct ib_smp *) ccp);
+
+bail:
+	return reply_failure((struct ib_smp *) ccp);
+}
+
+static int check_cc_key(struct qib_ibport *ibp,
+			struct ib_cc_mad *ccp, int mad_flags)
+{
+	return 0;
+}
+
+static int process_cc(struct ib_device *ibdev, int mad_flags,
+			u8 port, struct ib_mad *in_mad,
+			struct ib_mad *out_mad)
+{
+	struct ib_cc_mad *ccp = (struct ib_cc_mad *)out_mad;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	int ret;
+
+	*out_mad = *in_mad;
+
+	if (ccp->class_version != 2) {
+		ccp->status |= IB_SMP_UNSUP_VERSION;
+		ret = reply((struct ib_smp *)ccp);
+		goto bail;
+	}
+
+	ret = check_cc_key(ibp, ccp, mad_flags);
+	if (ret)
+		goto bail;
+
+	switch (ccp->method) {
+	case IB_MGMT_METHOD_GET:
+		switch (ccp->attr_id) {
+		case IB_CC_ATTR_CLASSPORTINFO:
+			ret = cc_get_classportinfo(ccp, ibdev);
+			goto bail;
+
+		case IB_CC_ATTR_CONGESTION_INFO:
+			ret = cc_get_congestion_info(ccp, ibdev, port);
+			goto bail;
+
+		case IB_CC_ATTR_CA_CONGESTION_SETTING:
+			ret = cc_get_congestion_setting(ccp, ibdev, port);
+			goto bail;
+
+		case IB_CC_ATTR_CONGESTION_CONTROL_TABLE:
+			ret = cc_get_congestion_control_table(ccp, ibdev, port);
+			goto bail;
+
+			/* FALLTHROUGH */
+		default:
+			ccp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) ccp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_SET:
+		switch (ccp->attr_id) {
+		case IB_CC_ATTR_CA_CONGESTION_SETTING:
+			ret = cc_set_congestion_setting(ccp, ibdev, port);
+			goto bail;
+
+		case IB_CC_ATTR_CONGESTION_CONTROL_TABLE:
+			ret = cc_set_congestion_control_table(ccp, ibdev, port);
+			goto bail;
+
+			/* FALLTHROUGH */
+		default:
+			ccp->status |= IB_SMP_UNSUP_METH_ATTR;
+			ret = reply((struct ib_smp *) ccp);
+			goto bail;
+		}
+
+	case IB_MGMT_METHOD_GET_RESP:
+		/*
+		 * The ib_mad module will call us to process responses
+		 * before checking for other consumers.
+		 * Just tell the caller to process it normally.
+		 */
+		ret = IB_MAD_RESULT_SUCCESS;
+		goto bail;
+
+	case IB_MGMT_METHOD_TRAP:
+	default:
+		ccp->status |= IB_SMP_UNSUP_METHOD;
+		ret = reply((struct ib_smp *) ccp);
+	}
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_process_mad - process an incoming MAD packet
+ * @ibdev: the infiniband device this packet came in on
+ * @mad_flags: MAD flags
+ * @port: the port number this packet came in on
+ * @in_wc: the work completion entry for this packet
+ * @in_grh: the global route header for this packet
+ * @in_mad: the incoming MAD
+ * @out_mad: any outgoing MAD reply
+ *
+ * Returns IB_MAD_RESULT_SUCCESS if this is a MAD that we are not
+ * interested in processing.
+ *
+ * Note that the verbs framework has already done the MAD sanity checks,
+ * and hop count/pointer updating for IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE
+ * MADs.
+ *
+ * This is called by the ib_mad module.
+ */
+int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port,
+		    struct ib_wc *in_wc, struct ib_grh *in_grh,
+		    struct ib_mad *in_mad, struct ib_mad *out_mad)
+{
+	int ret;
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+	switch (in_mad->mad_hdr.mgmt_class) {
+	case IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE:
+	case IB_MGMT_CLASS_SUBN_LID_ROUTED:
+		ret = process_subn(ibdev, mad_flags, port, in_mad, out_mad);
+		goto bail;
+
+	case IB_MGMT_CLASS_PERF_MGMT:
+		ret = process_perf(ibdev, port, in_mad, out_mad);
+		goto bail;
+
+	case IB_MGMT_CLASS_CONG_MGMT:
+		if (!ppd->congestion_entries_shadow ||
+			 !qib_cc_table_size) {
+			ret = IB_MAD_RESULT_SUCCESS;
+			goto bail;
+		}
+		ret = process_cc(ibdev, mad_flags, port, in_mad, out_mad);
+		goto bail;
+
+	default:
+		ret = IB_MAD_RESULT_SUCCESS;
+	}
+
+bail:
+	return ret;
+}
+
+static void send_handler(struct ib_mad_agent *agent,
+			 struct ib_mad_send_wc *mad_send_wc)
+{
+	ib_free_send_mad(mad_send_wc->send_buf);
+}
+
+static void xmit_wait_timer_func(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+	struct qib_devdata *dd = dd_from_ppd(ppd);
+	unsigned long flags;
+	u8 status;
+
+	spin_lock_irqsave(&ppd->ibport_data.lock, flags);
+	if (ppd->cong_stats.flags == IB_PMA_CONG_HW_CONTROL_SAMPLE) {
+		status = dd->f_portcntr(ppd, QIBPORTCNTR_PSSTAT);
+		if (status == IB_PMA_SAMPLE_STATUS_DONE) {
+			/* save counter cache */
+			cache_hw_sample_counters(ppd);
+			ppd->cong_stats.flags = IB_PMA_CONG_HW_CONTROL_TIMER;
+		} else
+			goto done;
+	}
+	ppd->cong_stats.counter = xmit_wait_get_value_delta(ppd);
+	dd->f_set_cntr_sample(ppd, QIB_CONG_TIMER_PSINTERVAL, 0x0);
+done:
+	spin_unlock_irqrestore(&ppd->ibport_data.lock, flags);
+	mod_timer(&ppd->cong_stats.timer, jiffies + HZ);
+}
+
+int qib_create_agents(struct qib_ibdev *dev)
+{
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct ib_mad_agent *agent;
+	struct qib_ibport *ibp;
+	int p;
+	int ret;
+
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		agent = ib_register_mad_agent(&dev->ibdev, p + 1, IB_QPT_SMI,
+					      NULL, 0, send_handler,
+					      NULL, NULL);
+		if (IS_ERR(agent)) {
+			ret = PTR_ERR(agent);
+			goto err;
+		}
+
+		/* Initialize xmit_wait structure */
+		dd->pport[p].cong_stats.counter = 0;
+		init_timer(&dd->pport[p].cong_stats.timer);
+		dd->pport[p].cong_stats.timer.function = xmit_wait_timer_func;
+		dd->pport[p].cong_stats.timer.data =
+			(unsigned long)(&dd->pport[p]);
+		dd->pport[p].cong_stats.timer.expires = 0;
+		add_timer(&dd->pport[p].cong_stats.timer);
+
+		ibp->send_agent = agent;
+	}
+
+	return 0;
+
+err:
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		if (ibp->send_agent) {
+			agent = ibp->send_agent;
+			ibp->send_agent = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+	}
+
+	return ret;
+}
+
+void qib_free_agents(struct qib_ibdev *dev)
+{
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct ib_mad_agent *agent;
+	struct qib_ibport *ibp;
+	int p;
+
+	for (p = 0; p < dd->num_pports; p++) {
+		ibp = &dd->pport[p].ibport_data;
+		if (ibp->send_agent) {
+			agent = ibp->send_agent;
+			ibp->send_agent = NULL;
+			ib_unregister_mad_agent(agent);
+		}
+		if (ibp->sm_ah) {
+			ib_destroy_ah(&ibp->sm_ah->ibah);
+			ibp->sm_ah = NULL;
+		}
+		if (dd->pport[p].cong_stats.timer.data)
+			del_timer_sync(&dd->pport[p].cong_stats.timer);
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_mad.h b/drivers/infiniband/hw/qib/qib_mad.h
new file mode 100644
index 00000000000..941d4d50d8e
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mad.h
@@ -0,0 +1,431 @@
+/*
+ * Copyright (c) 2012 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#ifndef _QIB_MAD_H
+#define _QIB_MAD_H
+
+#include <rdma/ib_pma.h>
+
+#define IB_SMP_UNSUP_VERSION    cpu_to_be16(0x0004)
+#define IB_SMP_UNSUP_METHOD     cpu_to_be16(0x0008)
+#define IB_SMP_UNSUP_METH_ATTR  cpu_to_be16(0x000C)
+#define IB_SMP_INVALID_FIELD    cpu_to_be16(0x001C)
+
+struct ib_node_info {
+	u8 base_version;
+	u8 class_version;
+	u8 node_type;
+	u8 num_ports;
+	__be64 sys_guid;
+	__be64 node_guid;
+	__be64 port_guid;
+	__be16 partition_cap;
+	__be16 device_id;
+	__be32 revision;
+	u8 local_port_num;
+	u8 vendor_id[3];
+} __packed;
+
+struct ib_mad_notice_attr {
+	u8 generic_type;
+	u8 prod_type_msb;
+	__be16 prod_type_lsb;
+	__be16 trap_num;
+	__be16 issuer_lid;
+	__be16 toggle_count;
+
+	union {
+		struct {
+			u8	details[54];
+		} raw_data;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* where violation happened */
+			u8	port_num;	/* where violation happened */
+		} __packed ntc_129_131;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* LID where change occurred */
+			u8	reserved2;
+			u8	local_changes;	/* low bit - local changes */
+			__be32	new_cap_mask;	/* new capability mask */
+			u8	reserved3;
+			u8	change_flags;	/* low 3 bits only */
+		} __packed ntc_144;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;		/* lid where sys guid changed */
+			__be16	reserved2;
+			__be64	new_sys_guid;
+		} __packed ntc_145;
+
+		struct {
+			__be16	reserved;
+			__be16	lid;
+			__be16	dr_slid;
+			u8	method;
+			u8	reserved2;
+			__be16	attr_id;
+			__be32	attr_mod;
+			__be64	mkey;
+			u8	reserved3;
+			u8	dr_trunc_hop;
+			u8	dr_rtn_path[30];
+		} __packed ntc_256;
+
+		struct {
+			__be16		reserved;
+			__be16		lid1;
+			__be16		lid2;
+			__be32		key;
+			__be32		sl_qp1;	/* SL: high 4 bits */
+			__be32		qp2;	/* high 8 bits reserved */
+			union ib_gid	gid1;
+			union ib_gid	gid2;
+		} __packed ntc_257_258;
+
+	} details;
+};
+
+/*
+ * Generic trap/notice types
+ */
+#define IB_NOTICE_TYPE_FATAL	0x80
+#define IB_NOTICE_TYPE_URGENT	0x81
+#define IB_NOTICE_TYPE_SECURITY	0x82
+#define IB_NOTICE_TYPE_SM	0x83
+#define IB_NOTICE_TYPE_INFO	0x84
+
+/*
+ * Generic trap/notice producers
+ */
+#define IB_NOTICE_PROD_CA		cpu_to_be16(1)
+#define IB_NOTICE_PROD_SWITCH		cpu_to_be16(2)
+#define IB_NOTICE_PROD_ROUTER		cpu_to_be16(3)
+#define IB_NOTICE_PROD_CLASS_MGR	cpu_to_be16(4)
+
+/*
+ * Generic trap/notice numbers
+ */
+#define IB_NOTICE_TRAP_LLI_THRESH	cpu_to_be16(129)
+#define IB_NOTICE_TRAP_EBO_THRESH	cpu_to_be16(130)
+#define IB_NOTICE_TRAP_FLOW_UPDATE	cpu_to_be16(131)
+#define IB_NOTICE_TRAP_CAP_MASK_CHG	cpu_to_be16(144)
+#define IB_NOTICE_TRAP_SYS_GUID_CHG	cpu_to_be16(145)
+#define IB_NOTICE_TRAP_BAD_MKEY		cpu_to_be16(256)
+#define IB_NOTICE_TRAP_BAD_PKEY		cpu_to_be16(257)
+#define IB_NOTICE_TRAP_BAD_QKEY		cpu_to_be16(258)
+
+/*
+ * Repress trap/notice flags
+ */
+#define IB_NOTICE_REPRESS_LLI_THRESH	(1 << 0)
+#define IB_NOTICE_REPRESS_EBO_THRESH	(1 << 1)
+#define IB_NOTICE_REPRESS_FLOW_UPDATE	(1 << 2)
+#define IB_NOTICE_REPRESS_CAP_MASK_CHG	(1 << 3)
+#define IB_NOTICE_REPRESS_SYS_GUID_CHG	(1 << 4)
+#define IB_NOTICE_REPRESS_BAD_MKEY	(1 << 5)
+#define IB_NOTICE_REPRESS_BAD_PKEY	(1 << 6)
+#define IB_NOTICE_REPRESS_BAD_QKEY	(1 << 7)
+
+/*
+ * Generic trap/notice other local changes flags (trap 144).
+ */
+#define IB_NOTICE_TRAP_LSE_CHG		0x04	/* Link Speed Enable changed */
+#define IB_NOTICE_TRAP_LWE_CHG		0x02	/* Link Width Enable changed */
+#define IB_NOTICE_TRAP_NODE_DESC_CHG	0x01
+
+/*
+ * Generic trap/notice M_Key volation flags in dr_trunc_hop (trap 256).
+ */
+#define IB_NOTICE_TRAP_DR_NOTICE	0x80
+#define IB_NOTICE_TRAP_DR_TRUNC		0x40
+
+struct ib_vl_weight_elem {
+	u8      vl;     /* Only low 4 bits, upper 4 bits reserved */
+	u8      weight;
+};
+
+#define IB_VLARB_LOWPRI_0_31    1
+#define IB_VLARB_LOWPRI_32_63   2
+#define IB_VLARB_HIGHPRI_0_31   3
+#define IB_VLARB_HIGHPRI_32_63  4
+
+#define IB_PMA_PORT_COUNTERS_CONG       cpu_to_be16(0xFF00)
+
+struct ib_pma_portcounters_cong {
+	u8 reserved;
+	u8 reserved1;
+	__be16 port_check_rate;
+	__be16 symbol_error_counter;
+	u8 link_error_recovery_counter;
+	u8 link_downed_counter;
+	__be16 port_rcv_errors;
+	__be16 port_rcv_remphys_errors;
+	__be16 port_rcv_switch_relay_errors;
+	__be16 port_xmit_discards;
+	u8 port_xmit_constraint_errors;
+	u8 port_rcv_constraint_errors;
+	u8 reserved2;
+	u8 link_overrun_errors; /* LocalLink: 7:4, BufferOverrun: 3:0 */
+	__be16 reserved3;
+	__be16 vl15_dropped;
+	__be64 port_xmit_data;
+	__be64 port_rcv_data;
+	__be64 port_xmit_packets;
+	__be64 port_rcv_packets;
+	__be64 port_xmit_wait;
+	__be64 port_adr_events;
+} __packed;
+
+#define IB_PMA_CONG_HW_CONTROL_TIMER            0x00
+#define IB_PMA_CONG_HW_CONTROL_SAMPLE           0x01
+
+#define QIB_XMIT_RATE_UNSUPPORTED               0x0
+#define QIB_XMIT_RATE_PICO                      0x7
+/* number of 4nsec cycles equaling 2secs */
+#define QIB_CONG_TIMER_PSINTERVAL               0x1DCD64EC
+
+#define IB_PMA_SEL_CONG_ALL                     0x01
+#define IB_PMA_SEL_CONG_PORT_DATA               0x02
+#define IB_PMA_SEL_CONG_XMIT                    0x04
+#define IB_PMA_SEL_CONG_ROUTING                 0x08
+
+/*
+ * Congestion control class attributes
+ */
+#define IB_CC_ATTR_CLASSPORTINFO			cpu_to_be16(0x0001)
+#define IB_CC_ATTR_NOTICE				cpu_to_be16(0x0002)
+#define IB_CC_ATTR_CONGESTION_INFO			cpu_to_be16(0x0011)
+#define IB_CC_ATTR_CONGESTION_KEY_INFO			cpu_to_be16(0x0012)
+#define IB_CC_ATTR_CONGESTION_LOG			cpu_to_be16(0x0013)
+#define IB_CC_ATTR_SWITCH_CONGESTION_SETTING		cpu_to_be16(0x0014)
+#define IB_CC_ATTR_SWITCH_PORT_CONGESTION_SETTING	cpu_to_be16(0x0015)
+#define IB_CC_ATTR_CA_CONGESTION_SETTING		cpu_to_be16(0x0016)
+#define IB_CC_ATTR_CONGESTION_CONTROL_TABLE		cpu_to_be16(0x0017)
+#define IB_CC_ATTR_TIME_STAMP				cpu_to_be16(0x0018)
+
+/* generalizations for threshold values */
+#define IB_CC_THRESHOLD_NONE 0x0
+#define IB_CC_THRESHOLD_MIN  0x1
+#define IB_CC_THRESHOLD_MAX  0xf
+
+/* CCA MAD header constants */
+#define IB_CC_MAD_LOGDATA_LEN 32
+#define IB_CC_MAD_MGMTDATA_LEN 192
+
+struct ib_cc_mad {
+	u8	base_version;
+	u8	mgmt_class;
+	u8	class_version;
+	u8	method;
+	__be16	status;
+	__be16	class_specific;
+	__be64	tid;
+	__be16	attr_id;
+	__be16	resv;
+	__be32	attr_mod;
+	__be64 cckey;
+
+	/* For CongestionLog attribute only */
+	u8 log_data[IB_CC_MAD_LOGDATA_LEN];
+
+	u8 mgmt_data[IB_CC_MAD_MGMTDATA_LEN];
+} __packed;
+
+/*
+ * Congestion Control class portinfo capability mask bits
+ */
+#define IB_CC_CPI_CM_TRAP_GEN		cpu_to_be16(1 << 0)
+#define IB_CC_CPI_CM_GET_SET_NOTICE	cpu_to_be16(1 << 1)
+#define IB_CC_CPI_CM_CAP2		cpu_to_be16(1 << 2)
+#define IB_CC_CPI_CM_ENHANCEDPORT0_CC	cpu_to_be16(1 << 8)
+
+struct ib_cc_classportinfo_attr {
+	u8 base_version;
+	u8 class_version;
+	__be16 cap_mask;
+	u8 reserved[3];
+	u8 resp_time_value;     /* only lower 5 bits */
+	union ib_gid redirect_gid;
+	__be32 redirect_tc_sl_fl;       /* 8, 4, 20 bits respectively */
+	__be16 redirect_lid;
+	__be16 redirect_pkey;
+	__be32 redirect_qp;     /* only lower 24 bits */
+	__be32 redirect_qkey;
+	union ib_gid trap_gid;
+	__be32 trap_tc_sl_fl;   /* 8, 4, 20 bits respectively */
+	__be16 trap_lid;
+	__be16 trap_pkey;
+	__be32 trap_hl_qp;      /* 8, 24 bits respectively */
+	__be32 trap_qkey;
+} __packed;
+
+/* Congestion control traps */
+#define IB_CC_TRAP_KEY_VIOLATION 0x0000
+
+struct ib_cc_trap_key_violation_attr {
+	__be16 source_lid;
+	u8 method;
+	u8 reserved1;
+	__be16 attrib_id;
+	__be32 attrib_mod;
+	__be32 qp;
+	__be64 cckey;
+	u8 sgid[16];
+	u8 padding[24];
+} __packed;
+
+/* Congestion info flags */
+#define IB_CC_CI_FLAGS_CREDIT_STARVATION 0x1
+#define IB_CC_TABLE_CAP_DEFAULT 31
+
+struct ib_cc_info_attr {
+	__be16 congestion_info;
+	u8  control_table_cap; /* Multiple of 64 entry unit CCTs */
+} __packed;
+
+struct ib_cc_key_info_attr {
+	__be64 cckey;
+	u8  protect;
+	__be16 lease_period;
+	__be16 violations;
+} __packed;
+
+#define IB_CC_CL_CA_LOGEVENTS_LEN 208
+
+struct ib_cc_log_attr {
+	u8 log_type;
+	u8 congestion_flags;
+	__be16 threshold_event_counter;
+	__be16 threshold_congestion_event_map;
+	__be16 current_time_stamp;
+	u8 log_events[IB_CC_CL_CA_LOGEVENTS_LEN];
+} __packed;
+
+#define IB_CC_CLEC_SERVICETYPE_RC 0x0
+#define IB_CC_CLEC_SERVICETYPE_UC 0x1
+#define IB_CC_CLEC_SERVICETYPE_RD 0x2
+#define IB_CC_CLEC_SERVICETYPE_UD 0x3
+
+struct ib_cc_log_event {
+	u8 local_qp_cn_entry;
+	u8 remote_qp_number_cn_entry[3];
+	u8  sl_cn_entry:4;
+	u8  service_type_cn_entry:4;
+	__be32 remote_lid_cn_entry;
+	__be32 timestamp_cn_entry;
+} __packed;
+
+/* Sixteen congestion entries */
+#define IB_CC_CCS_ENTRIES 16
+
+/* Port control flags */
+#define IB_CC_CCS_PC_SL_BASED 0x01
+
+struct ib_cc_congestion_entry {
+	u8 ccti_increase;
+	__be16 ccti_timer;
+	u8 trigger_threshold;
+	u8 ccti_min; /* min CCTI for cc table */
+} __packed;
+
+struct ib_cc_congestion_entry_shadow {
+	u8 ccti_increase;
+	u16 ccti_timer;
+	u8 trigger_threshold;
+	u8 ccti_min; /* min CCTI for cc table */
+} __packed;
+
+struct ib_cc_congestion_setting_attr {
+	__be16 port_control;
+	__be16 control_map;
+	struct ib_cc_congestion_entry entries[IB_CC_CCS_ENTRIES];
+} __packed;
+
+struct ib_cc_congestion_setting_attr_shadow {
+	u16 port_control;
+	u16 control_map;
+	struct ib_cc_congestion_entry_shadow entries[IB_CC_CCS_ENTRIES];
+} __packed;
+
+#define IB_CC_TABLE_ENTRY_INCREASE_DEFAULT 1
+#define IB_CC_TABLE_ENTRY_TIMER_DEFAULT 1
+
+/* 64 Congestion Control table entries in a single MAD */
+#define IB_CCT_ENTRIES 64
+#define IB_CCT_MIN_ENTRIES (IB_CCT_ENTRIES * 2)
+
+struct ib_cc_table_entry {
+	__be16 entry; /* shift:2, multiplier:14 */
+};
+
+struct ib_cc_table_entry_shadow {
+	u16 entry; /* shift:2, multiplier:14 */
+};
+
+struct ib_cc_table_attr {
+	__be16 ccti_limit; /* max CCTI for cc table */
+	struct ib_cc_table_entry ccti_entries[IB_CCT_ENTRIES];
+} __packed;
+
+struct ib_cc_table_attr_shadow {
+	u16 ccti_limit; /* max CCTI for cc table */
+	struct ib_cc_table_entry_shadow ccti_entries[IB_CCT_ENTRIES];
+} __packed;
+
+#define CC_TABLE_SHADOW_MAX \
+	(IB_CC_TABLE_CAP_DEFAULT * IB_CCT_ENTRIES)
+
+struct cc_table_shadow {
+	u16 ccti_last_entry;
+	struct ib_cc_table_entry_shadow entries[CC_TABLE_SHADOW_MAX];
+} __packed;
+
+/*
+ * The PortSamplesControl.CounterMasks field is an array of 3 bit fields
+ * which specify the N'th counter's capabilities. See ch. 16.1.3.2.
+ * We support 5 counters which only count the mandatory quantities.
+ */
+#define COUNTER_MASK(q, n) (q << ((9 - n) * 3))
+#define COUNTER_MASK0_9 \
+	cpu_to_be32(COUNTER_MASK(1, 0) | \
+		    COUNTER_MASK(1, 1) | \
+		    COUNTER_MASK(1, 2) | \
+		    COUNTER_MASK(1, 3) | \
+		    COUNTER_MASK(1, 4))
+
+#endif				/* _QIB_MAD_H */
diff --git a/drivers/infiniband/hw/qib/qib_mmap.c b/drivers/infiniband/hw/qib/qib_mmap.c
new file mode 100644
index 00000000000..8b73a11d571
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mmap.c
@@ -0,0 +1,174 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+#include <linux/mm.h>
+#include <linux/errno.h>
+#include <asm/pgtable.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_release_mmap_info - free mmap info structure
+ * @ref: a pointer to the kref within struct qib_mmap_info
+ */
+void qib_release_mmap_info(struct kref *ref)
+{
+	struct qib_mmap_info *ip =
+		container_of(ref, struct qib_mmap_info, ref);
+	struct qib_ibdev *dev = to_idev(ip->context->device);
+
+	spin_lock_irq(&dev->pending_lock);
+	list_del(&ip->pending_mmaps);
+	spin_unlock_irq(&dev->pending_lock);
+
+	vfree(ip->obj);
+	kfree(ip);
+}
+
+/*
+ * open and close keep track of how many times the CQ is mapped,
+ * to avoid releasing it.
+ */
+static void qib_vma_open(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_get(&ip->ref);
+}
+
+static void qib_vma_close(struct vm_area_struct *vma)
+{
+	struct qib_mmap_info *ip = vma->vm_private_data;
+
+	kref_put(&ip->ref, qib_release_mmap_info);
+}
+
+static struct vm_operations_struct qib_vm_ops = {
+	.open =     qib_vma_open,
+	.close =    qib_vma_close,
+};
+
+/**
+ * qib_mmap - create a new mmap region
+ * @context: the IB user context of the process making the mmap() call
+ * @vma: the VMA to be initialized
+ * Return zero if the mmap is OK. Otherwise, return an errno.
+ */
+int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma)
+{
+	struct qib_ibdev *dev = to_idev(context->device);
+	unsigned long offset = vma->vm_pgoff << PAGE_SHIFT;
+	unsigned long size = vma->vm_end - vma->vm_start;
+	struct qib_mmap_info *ip, *pp;
+	int ret = -EINVAL;
+
+	/*
+	 * Search the device's list of objects waiting for a mmap call.
+	 * Normally, this list is very short since a call to create a
+	 * CQ, QP, or SRQ is soon followed by a call to mmap().
+	 */
+	spin_lock_irq(&dev->pending_lock);
+	list_for_each_entry_safe(ip, pp, &dev->pending_mmaps,
+				 pending_mmaps) {
+		/* Only the creator is allowed to mmap the object */
+		if (context != ip->context || (__u64) offset != ip->offset)
+			continue;
+		/* Don't allow a mmap larger than the object. */
+		if (size > ip->size)
+			break;
+
+		list_del_init(&ip->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+
+		ret = remap_vmalloc_range(vma, ip->obj, 0);
+		if (ret)
+			goto done;
+		vma->vm_ops = &qib_vm_ops;
+		vma->vm_private_data = ip;
+		qib_vma_open(vma);
+		goto done;
+	}
+	spin_unlock_irq(&dev->pending_lock);
+done:
+	return ret;
+}
+
+/*
+ * Allocate information for qib_mmap
+ */
+struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev,
+					   u32 size,
+					   struct ib_ucontext *context,
+					   void *obj) {
+	struct qib_mmap_info *ip;
+
+	ip = kmalloc(sizeof *ip, GFP_KERNEL);
+	if (!ip)
+		goto bail;
+
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	INIT_LIST_HEAD(&ip->pending_mmaps);
+	ip->size = size;
+	ip->context = context;
+	ip->obj = obj;
+	kref_init(&ip->ref);
+
+bail:
+	return ip;
+}
+
+void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
+			  u32 size, void *obj)
+{
+	size = PAGE_ALIGN(size);
+
+	spin_lock_irq(&dev->mmap_offset_lock);
+	if (dev->mmap_offset == 0)
+		dev->mmap_offset = PAGE_SIZE;
+	ip->offset = dev->mmap_offset;
+	dev->mmap_offset += size;
+	spin_unlock_irq(&dev->mmap_offset_lock);
+
+	ip->size = size;
+	ip->obj = obj;
+}
diff --git a/drivers/infiniband/hw/qib/qib_mr.c b/drivers/infiniband/hw/qib/qib_mr.c
new file mode 100644
index 00000000000..9bbb55347cc
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_mr.c
@@ -0,0 +1,532 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_umem.h>
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+
+/* Fast memory region */
+struct qib_fmr {
+	struct ib_fmr ibfmr;
+	struct qib_mregion mr;        /* must be last */
+};
+
+static inline struct qib_fmr *to_ifmr(struct ib_fmr *ibfmr)
+{
+	return container_of(ibfmr, struct qib_fmr, ibfmr);
+}
+
+static int init_qib_mregion(struct qib_mregion *mr, struct ib_pd *pd,
+	int count)
+{
+	int m, i = 0;
+	int rval = 0;
+
+	m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	for (; i < m; i++) {
+		mr->map[i] = kzalloc(sizeof *mr->map[0], GFP_KERNEL);
+		if (!mr->map[i])
+			goto bail;
+	}
+	mr->mapsz = m;
+	init_completion(&mr->comp);
+	/* count returning the ptr to user */
+	atomic_set(&mr->refcount, 1);
+	mr->pd = pd;
+	mr->max_segs = count;
+out:
+	return rval;
+bail:
+	while (i)
+		kfree(mr->map[--i]);
+	rval = -ENOMEM;
+	goto out;
+}
+
+static void deinit_qib_mregion(struct qib_mregion *mr)
+{
+	int i = mr->mapsz;
+
+	mr->mapsz = 0;
+	while (i)
+		kfree(mr->map[--i]);
+}
+
+
+/**
+ * qib_get_dma_mr - get a DMA memory region
+ * @pd: protection domain for this memory region
+ * @acc: access flags
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ * Note that all DMA addresses should be created via the
+ * struct ib_dma_mapping_ops functions (see qib_dma.c).
+ */
+struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	struct qib_mr *mr = NULL;
+	struct ib_mr *ret;
+	int rval;
+
+	if (to_ipd(pd)->user) {
+		ret = ERR_PTR(-EPERM);
+		goto bail;
+	}
+
+	mr = kzalloc(sizeof *mr, GFP_KERNEL);
+	if (!mr) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	rval = init_qib_mregion(&mr->mr, pd, 0);
+	if (rval) {
+		ret = ERR_PTR(rval);
+		goto bail;
+	}
+
+
+	rval = qib_alloc_lkey(&mr->mr, 1);
+	if (rval) {
+		ret = ERR_PTR(rval);
+		goto bail_mregion;
+	}
+
+	mr->mr.access_flags = acc;
+	ret = &mr->ibmr;
+done:
+	return ret;
+
+bail_mregion:
+	deinit_qib_mregion(&mr->mr);
+bail:
+	kfree(mr);
+	goto done;
+}
+
+static struct qib_mr *alloc_mr(int count, struct ib_pd *pd)
+{
+	struct qib_mr *mr;
+	int rval = -ENOMEM;
+	int m;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (count + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	mr = kzalloc(sizeof *mr + m * sizeof mr->mr.map[0], GFP_KERNEL);
+	if (!mr)
+		goto bail;
+
+	rval = init_qib_mregion(&mr->mr, pd, count);
+	if (rval)
+		goto bail;
+	/*
+	 * ib_reg_phys_mr() will initialize mr->ibmr except for
+	 * lkey and rkey.
+	 */
+	rval = qib_alloc_lkey(&mr->mr, 0);
+	if (rval)
+		goto bail_mregion;
+	mr->ibmr.lkey = mr->mr.lkey;
+	mr->ibmr.rkey = mr->mr.lkey;
+done:
+	return mr;
+
+bail_mregion:
+	deinit_qib_mregion(&mr->mr);
+bail:
+	kfree(mr);
+	mr = ERR_PTR(rval);
+	goto done;
+}
+
+/**
+ * qib_reg_phys_mr - register a physical memory region
+ * @pd: protection domain for this memory region
+ * @buffer_list: pointer to the list of physical buffers to register
+ * @num_phys_buf: the number of physical buffers to register
+ * @iova_start: the starting address passed over IB which maps to this MR
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
+			      struct ib_phys_buf *buffer_list,
+			      int num_phys_buf, int acc, u64 *iova_start)
+{
+	struct qib_mr *mr;
+	int n, m, i;
+	struct ib_mr *ret;
+
+	mr = alloc_mr(num_phys_buf, pd);
+	if (IS_ERR(mr)) {
+		ret = (struct ib_mr *)mr;
+		goto bail;
+	}
+
+	mr->mr.user_base = *iova_start;
+	mr->mr.iova = *iova_start;
+	mr->mr.access_flags = acc;
+
+	m = 0;
+	n = 0;
+	for (i = 0; i < num_phys_buf; i++) {
+		mr->mr.map[m]->segs[n].vaddr = (void *) buffer_list[i].addr;
+		mr->mr.map[m]->segs[n].length = buffer_list[i].size;
+		mr->mr.length += buffer_list[i].size;
+		n++;
+		if (n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_reg_user_mr - register a userspace memory region
+ * @pd: protection domain for this memory region
+ * @start: starting userspace address
+ * @length: length of region to register
+ * @mr_access_flags: access flags for this memory region
+ * @udata: unused by the QLogic_IB driver
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			      u64 virt_addr, int mr_access_flags,
+			      struct ib_udata *udata)
+{
+	struct qib_mr *mr;
+	struct ib_umem *umem;
+	struct scatterlist *sg;
+	int n, m, entry;
+	struct ib_mr *ret;
+
+	if (length == 0) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	umem = ib_umem_get(pd->uobject->context, start, length,
+			   mr_access_flags, 0);
+	if (IS_ERR(umem))
+		return (void *) umem;
+
+	n = umem->nmap;
+
+	mr = alloc_mr(n, pd);
+	if (IS_ERR(mr)) {
+		ret = (struct ib_mr *)mr;
+		ib_umem_release(umem);
+		goto bail;
+	}
+
+	mr->mr.user_base = start;
+	mr->mr.iova = virt_addr;
+	mr->mr.length = length;
+	mr->mr.offset = umem->offset;
+	mr->mr.access_flags = mr_access_flags;
+	mr->umem = umem;
+
+	if (is_power_of_2(umem->page_size))
+		mr->mr.page_shift = ilog2(umem->page_size);
+	m = 0;
+	n = 0;
+	for_each_sg(umem->sg_head.sgl, sg, umem->nmap, entry) {
+			void *vaddr;
+
+			vaddr = page_address(sg_page(sg));
+			if (!vaddr) {
+				ret = ERR_PTR(-EINVAL);
+				goto bail;
+			}
+			mr->mr.map[m]->segs[n].vaddr = vaddr;
+			mr->mr.map[m]->segs[n].length = umem->page_size;
+			n++;
+			if (n == QIB_SEGSZ) {
+				m++;
+				n = 0;
+			}
+	}
+	ret = &mr->ibmr;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_dereg_mr - unregister and free a memory region
+ * @ibmr: the memory region to free
+ *
+ * Returns 0 on success.
+ *
+ * Note that this is called to free MRs created by qib_get_dma_mr()
+ * or qib_reg_user_mr().
+ */
+int qib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct qib_mr *mr = to_imr(ibmr);
+	int ret = 0;
+	unsigned long timeout;
+
+	qib_free_lkey(&mr->mr);
+
+	qib_put_mr(&mr->mr); /* will set completion if last */
+	timeout = wait_for_completion_timeout(&mr->mr.comp,
+		5 * HZ);
+	if (!timeout) {
+		qib_get_mr(&mr->mr);
+		ret = -EBUSY;
+		goto out;
+	}
+	deinit_qib_mregion(&mr->mr);
+	if (mr->umem)
+		ib_umem_release(mr->umem);
+	kfree(mr);
+out:
+	return ret;
+}
+
+/*
+ * Allocate a memory region usable with the
+ * IB_WR_FAST_REG_MR send work request.
+ *
+ * Return the memory region on success, otherwise return an errno.
+ */
+struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len)
+{
+	struct qib_mr *mr;
+
+	mr = alloc_mr(max_page_list_len, pd);
+	if (IS_ERR(mr))
+		return (struct ib_mr *)mr;
+
+	return &mr->ibmr;
+}
+
+struct ib_fast_reg_page_list *
+qib_alloc_fast_reg_page_list(struct ib_device *ibdev, int page_list_len)
+{
+	unsigned size = page_list_len * sizeof(u64);
+	struct ib_fast_reg_page_list *pl;
+
+	if (size > PAGE_SIZE)
+		return ERR_PTR(-EINVAL);
+
+	pl = kzalloc(sizeof *pl, GFP_KERNEL);
+	if (!pl)
+		return ERR_PTR(-ENOMEM);
+
+	pl->page_list = kzalloc(size, GFP_KERNEL);
+	if (!pl->page_list)
+		goto err_free;
+
+	return pl;
+
+err_free:
+	kfree(pl);
+	return ERR_PTR(-ENOMEM);
+}
+
+void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl)
+{
+	kfree(pl->page_list);
+	kfree(pl);
+}
+
+/**
+ * qib_alloc_fmr - allocate a fast memory region
+ * @pd: the protection domain for this memory region
+ * @mr_access_flags: access flags for this memory region
+ * @fmr_attr: fast memory region attributes
+ *
+ * Returns the memory region on success, otherwise returns an errno.
+ */
+struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
+			     struct ib_fmr_attr *fmr_attr)
+{
+	struct qib_fmr *fmr;
+	int m;
+	struct ib_fmr *ret;
+	int rval = -ENOMEM;
+
+	/* Allocate struct plus pointers to first level page tables. */
+	m = (fmr_attr->max_pages + QIB_SEGSZ - 1) / QIB_SEGSZ;
+	fmr = kzalloc(sizeof *fmr + m * sizeof fmr->mr.map[0], GFP_KERNEL);
+	if (!fmr)
+		goto bail;
+
+	rval = init_qib_mregion(&fmr->mr, pd, fmr_attr->max_pages);
+	if (rval)
+		goto bail;
+
+	/*
+	 * ib_alloc_fmr() will initialize fmr->ibfmr except for lkey &
+	 * rkey.
+	 */
+	rval = qib_alloc_lkey(&fmr->mr, 0);
+	if (rval)
+		goto bail_mregion;
+	fmr->ibfmr.rkey = fmr->mr.lkey;
+	fmr->ibfmr.lkey = fmr->mr.lkey;
+	/*
+	 * Resources are allocated but no valid mapping (RKEY can't be
+	 * used).
+	 */
+	fmr->mr.access_flags = mr_access_flags;
+	fmr->mr.max_segs = fmr_attr->max_pages;
+	fmr->mr.page_shift = fmr_attr->page_shift;
+
+	ret = &fmr->ibfmr;
+done:
+	return ret;
+
+bail_mregion:
+	deinit_qib_mregion(&fmr->mr);
+bail:
+	kfree(fmr);
+	ret = ERR_PTR(rval);
+	goto done;
+}
+
+/**
+ * qib_map_phys_fmr - set up a fast memory region
+ * @ibmfr: the fast memory region to set up
+ * @page_list: the list of pages to associate with the fast memory region
+ * @list_len: the number of pages to associate with the fast memory region
+ * @iova: the virtual address of the start of the fast memory region
+ *
+ * This may be called from interrupt context.
+ */
+
+int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		     int list_len, u64 iova)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+	int m, n, i;
+	u32 ps;
+	int ret;
+
+	i = atomic_read(&fmr->mr.refcount);
+	if (i > 2)
+		return -EBUSY;
+
+	if (list_len > fmr->mr.max_segs) {
+		ret = -EINVAL;
+		goto bail;
+	}
+	rkt = &to_idev(ibfmr->device)->lk_table;
+	spin_lock_irqsave(&rkt->lock, flags);
+	fmr->mr.user_base = iova;
+	fmr->mr.iova = iova;
+	ps = 1 << fmr->mr.page_shift;
+	fmr->mr.length = list_len * ps;
+	m = 0;
+	n = 0;
+	for (i = 0; i < list_len; i++) {
+		fmr->mr.map[m]->segs[n].vaddr = (void *) page_list[i];
+		fmr->mr.map[m]->segs[n].length = ps;
+		if (++n == QIB_SEGSZ) {
+			m++;
+			n = 0;
+		}
+	}
+	spin_unlock_irqrestore(&rkt->lock, flags);
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_unmap_fmr - unmap fast memory regions
+ * @fmr_list: the list of fast memory regions to unmap
+ *
+ * Returns 0 on success.
+ */
+int qib_unmap_fmr(struct list_head *fmr_list)
+{
+	struct qib_fmr *fmr;
+	struct qib_lkey_table *rkt;
+	unsigned long flags;
+
+	list_for_each_entry(fmr, fmr_list, ibfmr.list) {
+		rkt = &to_idev(fmr->ibfmr.device)->lk_table;
+		spin_lock_irqsave(&rkt->lock, flags);
+		fmr->mr.user_base = 0;
+		fmr->mr.iova = 0;
+		fmr->mr.length = 0;
+		spin_unlock_irqrestore(&rkt->lock, flags);
+	}
+	return 0;
+}
+
+/**
+ * qib_dealloc_fmr - deallocate a fast memory region
+ * @ibfmr: the fast memory region to deallocate
+ *
+ * Returns 0 on success.
+ */
+int qib_dealloc_fmr(struct ib_fmr *ibfmr)
+{
+	struct qib_fmr *fmr = to_ifmr(ibfmr);
+	int ret = 0;
+	unsigned long timeout;
+
+	qib_free_lkey(&fmr->mr);
+	qib_put_mr(&fmr->mr); /* will set completion if last */
+	timeout = wait_for_completion_timeout(&fmr->mr.comp,
+		5 * HZ);
+	if (!timeout) {
+		qib_get_mr(&fmr->mr);
+		ret = -EBUSY;
+		goto out;
+	}
+	deinit_qib_mregion(&fmr->mr);
+	kfree(fmr);
+out:
+	return ret;
+}
+
+void mr_rcu_callback(struct rcu_head *list)
+{
+	struct qib_mregion *mr = container_of(list, struct qib_mregion, list);
+
+	complete(&mr->comp);
+}
diff --git a/drivers/infiniband/hw/qib/qib_pcie.c b/drivers/infiniband/hw/qib/qib_pcie.c
new file mode 100644
index 00000000000..61a0046efb7
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_pcie.c
@@ -0,0 +1,715 @@
+/*
+ * Copyright (c) 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/vmalloc.h>
+#include <linux/aer.h>
+#include <linux/module.h>
+
+#include "qib.h"
+
+/*
+ * This file contains PCIe utility routines that are common to the
+ * various QLogic InfiniPath adapters
+ */
+
+/*
+ * Code to adjust PCIe capabilities.
+ * To minimize the change footprint, we call it
+ * from qib_pcie_params, which every chip-specific
+ * file calls, even though this violates some
+ * expectations of harmlessness.
+ */
+static void qib_tune_pcie_caps(struct qib_devdata *);
+static void qib_tune_pcie_coalesce(struct qib_devdata *);
+
+/*
+ * Do all the common PCIe setup and initialization.
+ * devdata is not yet allocated, and is not allocated until after this
+ * routine returns success.  Therefore qib_dev_err() can't be used for error
+ * printing.
+ */
+int qib_pcie_init(struct pci_dev *pdev, const struct pci_device_id *ent)
+{
+	int ret;
+
+	ret = pci_enable_device(pdev);
+	if (ret) {
+		/*
+		 * This can happen (in theory) iff:
+		 * We did a chip reset, and then failed to reprogram the
+		 * BAR, or the chip reset due to an internal error.  We then
+		 * unloaded the driver and reloaded it.
+		 *
+		 * Both reset cases set the BAR back to initial state.  For
+		 * the latter case, the AER sticky error bit at offset 0x718
+		 * should be set, but the Linux kernel doesn't yet know
+		 * about that, it appears.  If the original BAR was retained
+		 * in the kernel data structures, this may be OK.
+		 */
+		qib_early_err(&pdev->dev, "pci enable failed: error %d\n",
+			      -ret);
+		goto done;
+	}
+
+	ret = pci_request_regions(pdev, QIB_DRV_NAME);
+	if (ret) {
+		qib_devinfo(pdev, "pci_request_regions fails: err %d\n", -ret);
+		goto bail;
+	}
+
+	ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret) {
+		/*
+		 * If the 64 bit setup fails, try 32 bit.  Some systems
+		 * do not setup 64 bit maps on systems with 2GB or less
+		 * memory installed.
+		 */
+		ret = pci_set_dma_mask(pdev, DMA_BIT_MASK(32));
+		if (ret) {
+			qib_devinfo(pdev, "Unable to set DMA mask: %d\n", ret);
+			goto bail;
+		}
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(32));
+	} else
+		ret = pci_set_consistent_dma_mask(pdev, DMA_BIT_MASK(64));
+	if (ret) {
+		qib_early_err(&pdev->dev,
+			      "Unable to set DMA consistent mask: %d\n", ret);
+		goto bail;
+	}
+
+	pci_set_master(pdev);
+	ret = pci_enable_pcie_error_reporting(pdev);
+	if (ret) {
+		qib_early_err(&pdev->dev,
+			      "Unable to enable pcie error reporting: %d\n",
+			      ret);
+		ret = 0;
+	}
+	goto done;
+
+bail:
+	pci_disable_device(pdev);
+	pci_release_regions(pdev);
+done:
+	return ret;
+}
+
+/*
+ * Do remaining PCIe setup, once dd is allocated, and save away
+ * fields required to re-initialize after a chip reset, or for
+ * various other purposes
+ */
+int qib_pcie_ddinit(struct qib_devdata *dd, struct pci_dev *pdev,
+		    const struct pci_device_id *ent)
+{
+	unsigned long len;
+	resource_size_t addr;
+
+	dd->pcidev = pdev;
+	pci_set_drvdata(pdev, dd);
+
+	addr = pci_resource_start(pdev, 0);
+	len = pci_resource_len(pdev, 0);
+
+#if defined(__powerpc__)
+	/* There isn't a generic way to specify writethrough mappings */
+	dd->kregbase = __ioremap(addr, len, _PAGE_NO_CACHE | _PAGE_WRITETHRU);
+#else
+	dd->kregbase = ioremap_nocache(addr, len);
+#endif
+
+	if (!dd->kregbase)
+		return -ENOMEM;
+
+	dd->kregend = (u64 __iomem *)((void __iomem *) dd->kregbase + len);
+	dd->physaddr = addr;        /* used for io_remap, etc. */
+
+	/*
+	 * Save BARs to rewrite after device reset.  Save all 64 bits of
+	 * BAR, just in case.
+	 */
+	dd->pcibar0 = addr;
+	dd->pcibar1 = addr >> 32;
+	dd->deviceid = ent->device; /* save for later use */
+	dd->vendorid = ent->vendor;
+
+	return 0;
+}
+
+/*
+ * Do PCIe cleanup, after chip-specific cleanup, etc.  Just prior
+ * to releasing the dd memory.
+ * void because none of the core pcie cleanup returns are void
+ */
+void qib_pcie_ddcleanup(struct qib_devdata *dd)
+{
+	u64 __iomem *base = (void __iomem *) dd->kregbase;
+
+	dd->kregbase = NULL;
+	iounmap(base);
+	if (dd->piobase)
+		iounmap(dd->piobase);
+	if (dd->userbase)
+		iounmap(dd->userbase);
+	if (dd->piovl15base)
+		iounmap(dd->piovl15base);
+
+	pci_disable_device(dd->pcidev);
+	pci_release_regions(dd->pcidev);
+
+	pci_set_drvdata(dd->pcidev, NULL);
+}
+
+static void qib_msix_setup(struct qib_devdata *dd, int pos, u32 *msixcnt,
+			   struct qib_msix_entry *qib_msix_entry)
+{
+	int ret;
+	int nvec = *msixcnt;
+	struct msix_entry *msix_entry;
+	int i;
+
+	ret = pci_msix_vec_count(dd->pcidev);
+	if (ret < 0)
+		goto do_intx;
+
+	nvec = min(nvec, ret);
+
+	/* We can't pass qib_msix_entry array to qib_msix_setup
+	 * so use a dummy msix_entry array and copy the allocated
+	 * irq back to the qib_msix_entry array. */
+	msix_entry = kmalloc(nvec * sizeof(*msix_entry), GFP_KERNEL);
+	if (!msix_entry)
+		goto do_intx;
+
+	for (i = 0; i < nvec; i++)
+		msix_entry[i] = qib_msix_entry[i].msix;
+
+	ret = pci_enable_msix_range(dd->pcidev, msix_entry, 1, nvec);
+	if (ret < 0)
+		goto free_msix_entry;
+	else
+		nvec = ret;
+
+	for (i = 0; i < nvec; i++)
+		qib_msix_entry[i].msix = msix_entry[i];
+
+	kfree(msix_entry);
+	*msixcnt = nvec;
+	return;
+
+free_msix_entry:
+	kfree(msix_entry);
+
+do_intx:
+	qib_dev_err(dd, "pci_enable_msix_range %d vectors failed: %d, "
+			"falling back to INTx\n", nvec, ret);
+	*msixcnt = 0;
+	qib_enable_intx(dd->pcidev);
+}
+
+/**
+ * We save the msi lo and hi values, so we can restore them after
+ * chip reset (the kernel PCI infrastructure doesn't yet handle that
+ * correctly.
+ */
+static int qib_msi_setup(struct qib_devdata *dd, int pos)
+{
+	struct pci_dev *pdev = dd->pcidev;
+	u16 control;
+	int ret;
+
+	ret = pci_enable_msi(pdev);
+	if (ret)
+		qib_dev_err(dd,
+			"pci_enable_msi failed: %d, interrupts may not work\n",
+			ret);
+	/* continue even if it fails, we may still be OK... */
+
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_LO,
+			      &dd->msi_lo);
+	pci_read_config_dword(pdev, pos + PCI_MSI_ADDRESS_HI,
+			      &dd->msi_hi);
+	pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &control);
+	/* now save the data (vector) info */
+	pci_read_config_word(pdev, pos + ((control & PCI_MSI_FLAGS_64BIT)
+				    ? 12 : 8),
+			     &dd->msi_data);
+	return ret;
+}
+
+int qib_pcie_params(struct qib_devdata *dd, u32 minw, u32 *nent,
+		    struct qib_msix_entry *entry)
+{
+	u16 linkstat, speed;
+	int pos = 0, ret = 1;
+
+	if (!pci_is_pcie(dd->pcidev)) {
+		qib_dev_err(dd, "Can't find PCI Express capability!\n");
+		/* set up something... */
+		dd->lbus_width = 1;
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		goto bail;
+	}
+
+	pos = dd->pcidev->msix_cap;
+	if (nent && *nent && pos) {
+		qib_msix_setup(dd, pos, nent, entry);
+		ret = 0; /* did it, either MSIx or INTx */
+	} else {
+		pos = dd->pcidev->msi_cap;
+		if (pos)
+			ret = qib_msi_setup(dd, pos);
+		else
+			qib_dev_err(dd, "No PCI MSI or MSIx capability!\n");
+	}
+	if (!pos)
+		qib_enable_intx(dd->pcidev);
+
+	pcie_capability_read_word(dd->pcidev, PCI_EXP_LNKSTA, &linkstat);
+	/*
+	 * speed is bits 0-3, linkwidth is bits 4-8
+	 * no defines for them in headers
+	 */
+	speed = linkstat & 0xf;
+	linkstat >>= 4;
+	linkstat &= 0x1f;
+	dd->lbus_width = linkstat;
+
+	switch (speed) {
+	case 1:
+		dd->lbus_speed = 2500; /* Gen1, 2.5GHz */
+		break;
+	case 2:
+		dd->lbus_speed = 5000; /* Gen1, 5GHz */
+		break;
+	default: /* not defined, assume gen1 */
+		dd->lbus_speed = 2500;
+		break;
+	}
+
+	/*
+	 * Check against expected pcie width and complain if "wrong"
+	 * on first initialization, not afterwards (i.e., reset).
+	 */
+	if (minw && linkstat < minw)
+		qib_dev_err(dd,
+			    "PCIe width %u (x%u HCA), performance reduced\n",
+			    linkstat, minw);
+
+	qib_tune_pcie_caps(dd);
+
+	qib_tune_pcie_coalesce(dd);
+
+bail:
+	/* fill in string, even on errors */
+	snprintf(dd->lbus_info, sizeof(dd->lbus_info),
+		 "PCIe,%uMHz,x%u\n", dd->lbus_speed, dd->lbus_width);
+	return ret;
+}
+
+/*
+ * Setup pcie interrupt stuff again after a reset.  I'd like to just call
+ * pci_enable_msi() again for msi, but when I do that,
+ * the MSI enable bit doesn't get set in the command word, and
+ * we switch to to a different interrupt vector, which is confusing,
+ * so I instead just do it all inline.  Perhaps somehow can tie this
+ * into the PCIe hotplug support at some point
+ */
+int qib_reinit_intr(struct qib_devdata *dd)
+{
+	int pos;
+	u16 control;
+	int ret = 0;
+
+	/* If we aren't using MSI, don't restore it */
+	if (!dd->msi_lo)
+		goto bail;
+
+	pos = dd->pcidev->msi_cap;
+	if (!pos) {
+		qib_dev_err(dd,
+			"Can't find MSI capability, can't restore MSI settings\n");
+		ret = 0;
+		/* nothing special for MSIx, just MSI */
+		goto bail;
+	}
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_LO,
+			       dd->msi_lo);
+	pci_write_config_dword(dd->pcidev, pos + PCI_MSI_ADDRESS_HI,
+			       dd->msi_hi);
+	pci_read_config_word(dd->pcidev, pos + PCI_MSI_FLAGS, &control);
+	if (!(control & PCI_MSI_FLAGS_ENABLE)) {
+		control |= PCI_MSI_FLAGS_ENABLE;
+		pci_write_config_word(dd->pcidev, pos + PCI_MSI_FLAGS,
+				      control);
+	}
+	/* now rewrite the data (vector) info */
+	pci_write_config_word(dd->pcidev, pos +
+			      ((control & PCI_MSI_FLAGS_64BIT) ? 12 : 8),
+			      dd->msi_data);
+	ret = 1;
+bail:
+	if (!ret && (dd->flags & QIB_HAS_INTX)) {
+		qib_enable_intx(dd->pcidev);
+		ret = 1;
+	}
+
+	/* and now set the pci master bit again */
+	pci_set_master(dd->pcidev);
+
+	return ret;
+}
+
+/*
+ * Disable msi interrupt if enabled, and clear msi_lo.
+ * This is used primarily for the fallback to INTx, but
+ * is also used in reinit after reset, and during cleanup.
+ */
+void qib_nomsi(struct qib_devdata *dd)
+{
+	dd->msi_lo = 0;
+	pci_disable_msi(dd->pcidev);
+}
+
+/*
+ * Same as qib_nosmi, but for MSIx.
+ */
+void qib_nomsix(struct qib_devdata *dd)
+{
+	pci_disable_msix(dd->pcidev);
+}
+
+/*
+ * Similar to pci_intx(pdev, 1), except that we make sure
+ * msi(x) is off.
+ */
+void qib_enable_intx(struct pci_dev *pdev)
+{
+	u16 cw, new;
+	int pos;
+
+	/* first, turn on INTx */
+	pci_read_config_word(pdev, PCI_COMMAND, &cw);
+	new = cw & ~PCI_COMMAND_INTX_DISABLE;
+	if (new != cw)
+		pci_write_config_word(pdev, PCI_COMMAND, new);
+
+	pos = pdev->msi_cap;
+	if (pos) {
+		/* then turn off MSI */
+		pci_read_config_word(pdev, pos + PCI_MSI_FLAGS, &cw);
+		new = cw & ~PCI_MSI_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSI_FLAGS, new);
+	}
+	pos = pdev->msix_cap;
+	if (pos) {
+		/* then turn off MSIx */
+		pci_read_config_word(pdev, pos + PCI_MSIX_FLAGS, &cw);
+		new = cw & ~PCI_MSIX_FLAGS_ENABLE;
+		if (new != cw)
+			pci_write_config_word(pdev, pos + PCI_MSIX_FLAGS, new);
+	}
+}
+
+/*
+ * These two routines are helper routines for the device reset code
+ * to move all the pcie code out of the chip-specific driver code.
+ */
+void qib_pcie_getcmd(struct qib_devdata *dd, u16 *cmd, u8 *iline, u8 *cline)
+{
+	pci_read_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_read_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_read_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+}
+
+void qib_pcie_reenable(struct qib_devdata *dd, u16 cmd, u8 iline, u8 cline)
+{
+	int r;
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_0,
+				   dd->pcibar0);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR0 failed: %d\n", r);
+	r = pci_write_config_dword(dd->pcidev, PCI_BASE_ADDRESS_1,
+				   dd->pcibar1);
+	if (r)
+		qib_dev_err(dd, "rewrite of BAR1 failed: %d\n", r);
+	/* now re-enable memory access, and restore cosmetic settings */
+	pci_write_config_word(dd->pcidev, PCI_COMMAND, cmd);
+	pci_write_config_byte(dd->pcidev, PCI_INTERRUPT_LINE, iline);
+	pci_write_config_byte(dd->pcidev, PCI_CACHE_LINE_SIZE, cline);
+	r = pci_enable_device(dd->pcidev);
+	if (r)
+		qib_dev_err(dd,
+			"pci_enable_device failed after reset: %d\n", r);
+}
+
+
+static int qib_pcie_coalesce;
+module_param_named(pcie_coalesce, qib_pcie_coalesce, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_coalesce, "tune PCIe colescing on some Intel chipsets");
+
+/*
+ * Enable PCIe completion and data coalescing, on Intel 5x00 and 7300
+ * chipsets.   This is known to be unsafe for some revisions of some
+ * of these chipsets, with some BIOS settings, and enabling it on those
+ * systems may result in the system crashing, and/or data corruption.
+ */
+static void qib_tune_pcie_coalesce(struct qib_devdata *dd)
+{
+	int r;
+	struct pci_dev *parent;
+	u16 devid;
+	u32 mask, bits, val;
+
+	if (!qib_pcie_coalesce)
+		return;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (parent->bus->parent) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		return;
+	}
+	if (!pci_is_pcie(parent))
+		return;
+	if (parent->vendor != 0x8086)
+		return;
+
+	/*
+	 *  - bit 12: Max_rdcmp_Imt_EN: need to set to 1
+	 *  - bit 11: COALESCE_FORCE: need to set to 0
+	 *  - bit 10: COALESCE_EN: need to set to 1
+	 *  (but limitations on some on some chipsets)
+	 *
+	 *  On the Intel 5000, 5100, and 7300 chipsets, there is
+	 *  also: - bit 25:24: COALESCE_MODE, need to set to 0
+	 */
+	devid = parent->device;
+	if (devid >= 0x25e2 && devid <= 0x25fa) {
+		/* 5000 P/V/X/Z */
+		if (parent->revision <= 0xb2)
+			bits = 1U << 10;
+		else
+			bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x65e2 && devid <= 0x65fa) {
+		/* 5100 */
+		bits = 1U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else if (devid >= 0x4021 && devid <= 0x402e) {
+		/* 5400 */
+		bits = 7U << 10;
+		mask = 7U << 10;
+	} else if (devid >= 0x3604 && devid <= 0x360a) {
+		/* 7300 */
+		bits = 7U << 10;
+		mask = (3U << 24) | (7U << 10);
+	} else {
+		/* not one of the chipsets that we know about */
+		return;
+	}
+	pci_read_config_dword(parent, 0x48, &val);
+	val &= ~mask;
+	val |= bits;
+	r = pci_write_config_dword(parent, 0x48, val);
+}
+
+/*
+ * BIOS may not set PCIe bus-utilization parameters for best performance.
+ * Check and optionally adjust them to maximize our throughput.
+ */
+static int qib_pcie_caps;
+module_param_named(pcie_caps, qib_pcie_caps, int, S_IRUGO);
+MODULE_PARM_DESC(pcie_caps, "Max PCIe tuning: Payload (0..3), ReadReq (4..7)");
+
+static void qib_tune_pcie_caps(struct qib_devdata *dd)
+{
+	struct pci_dev *parent;
+	u16 rc_mpss, rc_mps, ep_mpss, ep_mps;
+	u16 rc_mrrs, ep_mrrs, max_mrrs;
+
+	/* Find out supported and configured values for parent (root) */
+	parent = dd->pcidev->bus->self;
+	if (!pci_is_root_bus(parent->bus)) {
+		qib_devinfo(dd->pcidev, "Parent not root\n");
+		return;
+	}
+
+	if (!pci_is_pcie(parent) || !pci_is_pcie(dd->pcidev))
+		return;
+
+	rc_mpss = parent->pcie_mpss;
+	rc_mps = ffs(pcie_get_mps(parent)) - 8;
+	/* Find out supported and configured values for endpoint (us) */
+	ep_mpss = dd->pcidev->pcie_mpss;
+	ep_mps = ffs(pcie_get_mps(dd->pcidev)) - 8;
+
+	/* Find max payload supported by root, endpoint */
+	if (rc_mpss > ep_mpss)
+		rc_mpss = ep_mpss;
+
+	/* If Supported greater than limit in module param, limit it */
+	if (rc_mpss > (qib_pcie_caps & 7))
+		rc_mpss = qib_pcie_caps & 7;
+	/* If less than (allowed, supported), bump root payload */
+	if (rc_mpss > rc_mps) {
+		rc_mps = rc_mpss;
+		pcie_set_mps(parent, 128 << rc_mps);
+	}
+	/* If less than (allowed, supported), bump endpoint payload */
+	if (rc_mpss > ep_mps) {
+		ep_mps = rc_mpss;
+		pcie_set_mps(dd->pcidev, 128 << ep_mps);
+	}
+
+	/*
+	 * Now the Read Request size.
+	 * No field for max supported, but PCIe spec limits it to 4096,
+	 * which is code '5' (log2(4096) - 7)
+	 */
+	max_mrrs = 5;
+	if (max_mrrs > ((qib_pcie_caps >> 4) & 7))
+		max_mrrs = (qib_pcie_caps >> 4) & 7;
+
+	max_mrrs = 128 << max_mrrs;
+	rc_mrrs = pcie_get_readrq(parent);
+	ep_mrrs = pcie_get_readrq(dd->pcidev);
+
+	if (max_mrrs > rc_mrrs) {
+		rc_mrrs = max_mrrs;
+		pcie_set_readrq(parent, rc_mrrs);
+	}
+	if (max_mrrs > ep_mrrs) {
+		ep_mrrs = max_mrrs;
+		pcie_set_readrq(dd->pcidev, ep_mrrs);
+	}
+}
+/* End of PCIe capability tuning */
+
+/*
+ * From here through qib_pci_err_handler definition is invoked via
+ * PCI error infrastructure, registered via pci
+ */
+static pci_ers_result_t
+qib_pci_error_detected(struct pci_dev *pdev, pci_channel_state_t state)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	switch (state) {
+	case pci_channel_io_normal:
+		qib_devinfo(pdev, "State Normal, ignoring\n");
+		break;
+
+	case pci_channel_io_frozen:
+		qib_devinfo(pdev, "State Frozen, requesting reset\n");
+		pci_disable_device(pdev);
+		ret = PCI_ERS_RESULT_NEED_RESET;
+		break;
+
+	case pci_channel_io_perm_failure:
+		qib_devinfo(pdev, "State Permanent Failure, disabling\n");
+		if (dd) {
+			/* no more register accesses! */
+			dd->flags &= ~QIB_PRESENT;
+			qib_disable_after_error(dd);
+		}
+		 /* else early, or other problem */
+		ret =  PCI_ERS_RESULT_DISCONNECT;
+		break;
+
+	default: /* shouldn't happen */
+		qib_devinfo(pdev, "QIB PCI errors detected (state %d)\n",
+			state);
+		break;
+	}
+	return ret;
+}
+
+static pci_ers_result_t
+qib_pci_mmio_enabled(struct pci_dev *pdev)
+{
+	u64 words = 0U;
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	pci_ers_result_t ret = PCI_ERS_RESULT_RECOVERED;
+
+	if (dd && dd->pport) {
+		words = dd->f_portcntr(dd->pport, QIBPORTCNTR_WORDRCV);
+		if (words == ~0ULL)
+			ret = PCI_ERS_RESULT_NEED_RESET;
+	}
+	qib_devinfo(pdev,
+		"QIB mmio_enabled function called, read wordscntr %Lx, returning %d\n",
+		words, ret);
+	return  ret;
+}
+
+static pci_ers_result_t
+qib_pci_slot_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB slot_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static pci_ers_result_t
+qib_pci_link_reset(struct pci_dev *pdev)
+{
+	qib_devinfo(pdev, "QIB link_reset function called, ignored\n");
+	return PCI_ERS_RESULT_CAN_RECOVER;
+}
+
+static void
+qib_pci_resume(struct pci_dev *pdev)
+{
+	struct qib_devdata *dd = pci_get_drvdata(pdev);
+	qib_devinfo(pdev, "QIB resume function called\n");
+	pci_cleanup_aer_uncorrect_error_status(pdev);
+	/*
+	 * Running jobs will fail, since it's asynchronous
+	 * unlike sysfs-requested reset.   Better than
+	 * doing nothing.
+	 */
+	qib_init(dd, 1); /* same as re-init after reset */
+}
+
+const struct pci_error_handlers qib_pci_err_handler = {
+	.error_detected = qib_pci_error_detected,
+	.mmio_enabled = qib_pci_mmio_enabled,
+	.link_reset = qib_pci_link_reset,
+	.slot_reset = qib_pci_slot_reset,
+	.resume = qib_pci_resume,
+};
diff --git a/drivers/infiniband/hw/qib/qib_pio_copy.c b/drivers/infiniband/hw/qib/qib_pio_copy.c
new file mode 100644
index 00000000000..10b8c444dd3
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_pio_copy.c
@@ -0,0 +1,64 @@
+/*
+ * Copyright (c) 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/**
+ * qib_pio_copy - copy data to MMIO space, in multiples of 32-bits
+ * @to: destination, in MMIO space (must be 64-bit aligned)
+ * @from: source (must be 64-bit aligned)
+ * @count: number of 32-bit quantities to copy
+ *
+ * Copy data from kernel space to MMIO space, in multiples of 32 bits at a
+ * time.  Order of access is not guaranteed, nor is a memory barrier
+ * performed afterwards.
+ */
+void qib_pio_copy(void __iomem *to, const void *from, size_t count)
+{
+#ifdef CONFIG_64BIT
+	u64 __iomem *dst = to;
+	const u64 *src = from;
+	const u64 *end = src + (count >> 1);
+
+	while (src < end)
+		__raw_writeq(*src++, dst++);
+	if (count & 1)
+		__raw_writel(*(const u32 *)src, dst);
+#else
+	u32 __iomem *dst = to;
+	const u32 *src = from;
+	const u32 *end = src + count;
+
+	while (src < end)
+		__raw_writel(*src++, dst++);
+#endif
+}
diff --git a/drivers/infiniband/hw/qib/qib_qp.c b/drivers/infiniband/hw/qib/qib_qp.c
new file mode 100644
index 00000000000..7fcc150d603
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qp.c
@@ -0,0 +1,1384 @@
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation.  * All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/vmalloc.h>
+#include <linux/jhash.h>
+#ifdef CONFIG_DEBUG_FS
+#include <linux/seq_file.h>
+#endif
+
+#include "qib.h"
+
+#define BITS_PER_PAGE           (PAGE_SIZE*BITS_PER_BYTE)
+#define BITS_PER_PAGE_MASK      (BITS_PER_PAGE-1)
+
+static inline unsigned mk_qpn(struct qib_qpn_table *qpt,
+			      struct qpn_map *map, unsigned off)
+{
+	return (map - qpt->map) * BITS_PER_PAGE + off;
+}
+
+static inline unsigned find_next_offset(struct qib_qpn_table *qpt,
+					struct qpn_map *map, unsigned off,
+					unsigned n)
+{
+	if (qpt->mask) {
+		off++;
+		if (((off & qpt->mask) >> 1) >= n)
+			off = (off | qpt->mask) + 2;
+	} else
+		off = find_next_zero_bit(map->page, BITS_PER_PAGE, off);
+	return off;
+}
+
+/*
+ * Convert the AETH credit code into the number of credits.
+ */
+static u32 credit_table[31] = {
+	0,                      /* 0 */
+	1,                      /* 1 */
+	2,                      /* 2 */
+	3,                      /* 3 */
+	4,                      /* 4 */
+	6,                      /* 5 */
+	8,                      /* 6 */
+	12,                     /* 7 */
+	16,                     /* 8 */
+	24,                     /* 9 */
+	32,                     /* A */
+	48,                     /* B */
+	64,                     /* C */
+	96,                     /* D */
+	128,                    /* E */
+	192,                    /* F */
+	256,                    /* 10 */
+	384,                    /* 11 */
+	512,                    /* 12 */
+	768,                    /* 13 */
+	1024,                   /* 14 */
+	1536,                   /* 15 */
+	2048,                   /* 16 */
+	3072,                   /* 17 */
+	4096,                   /* 18 */
+	6144,                   /* 19 */
+	8192,                   /* 1A */
+	12288,                  /* 1B */
+	16384,                  /* 1C */
+	24576,                  /* 1D */
+	32768                   /* 1E */
+};
+
+static void get_map_page(struct qib_qpn_table *qpt, struct qpn_map *map)
+{
+	unsigned long page = get_zeroed_page(GFP_KERNEL);
+
+	/*
+	 * Free the page if someone raced with us installing it.
+	 */
+
+	spin_lock(&qpt->lock);
+	if (map->page)
+		free_page(page);
+	else
+		map->page = (void *)page;
+	spin_unlock(&qpt->lock);
+}
+
+/*
+ * Allocate the next available QPN or
+ * zero/one for QP type IB_QPT_SMI/IB_QPT_GSI.
+ */
+static int alloc_qpn(struct qib_devdata *dd, struct qib_qpn_table *qpt,
+		     enum ib_qp_type type, u8 port)
+{
+	u32 i, offset, max_scan, qpn;
+	struct qpn_map *map;
+	u32 ret;
+
+	if (type == IB_QPT_SMI || type == IB_QPT_GSI) {
+		unsigned n;
+
+		ret = type == IB_QPT_GSI;
+		n = 1 << (ret + 2 * (port - 1));
+		spin_lock(&qpt->lock);
+		if (qpt->flags & n)
+			ret = -EINVAL;
+		else
+			qpt->flags |= n;
+		spin_unlock(&qpt->lock);
+		goto bail;
+	}
+
+	qpn = qpt->last + 2;
+	if (qpn >= QPN_MAX)
+		qpn = 2;
+	if (qpt->mask && ((qpn & qpt->mask) >> 1) >= dd->n_krcv_queues)
+		qpn = (qpn | qpt->mask) + 2;
+	offset = qpn & BITS_PER_PAGE_MASK;
+	map = &qpt->map[qpn / BITS_PER_PAGE];
+	max_scan = qpt->nmaps - !offset;
+	for (i = 0;;) {
+		if (unlikely(!map->page)) {
+			get_map_page(qpt, map);
+			if (unlikely(!map->page))
+				break;
+		}
+		do {
+			if (!test_and_set_bit(offset, map->page)) {
+				qpt->last = qpn;
+				ret = qpn;
+				goto bail;
+			}
+			offset = find_next_offset(qpt, map, offset,
+				dd->n_krcv_queues);
+			qpn = mk_qpn(qpt, map, offset);
+			/*
+			 * This test differs from alloc_pidmap().
+			 * If find_next_offset() does find a zero
+			 * bit, we don't need to check for QPN
+			 * wrapping around past our starting QPN.
+			 * We just need to be sure we don't loop
+			 * forever.
+			 */
+		} while (offset < BITS_PER_PAGE && qpn < QPN_MAX);
+		/*
+		 * In order to keep the number of pages allocated to a
+		 * minimum, we scan the all existing pages before increasing
+		 * the size of the bitmap table.
+		 */
+		if (++i > max_scan) {
+			if (qpt->nmaps == QPNMAP_ENTRIES)
+				break;
+			map = &qpt->map[qpt->nmaps++];
+			offset = 0;
+		} else if (map < &qpt->map[qpt->nmaps]) {
+			++map;
+			offset = 0;
+		} else {
+			map = &qpt->map[0];
+			offset = 2;
+		}
+		qpn = mk_qpn(qpt, map, offset);
+	}
+
+	ret = -ENOMEM;
+
+bail:
+	return ret;
+}
+
+static void free_qpn(struct qib_qpn_table *qpt, u32 qpn)
+{
+	struct qpn_map *map;
+
+	map = qpt->map + qpn / BITS_PER_PAGE;
+	if (map->page)
+		clear_bit(qpn & BITS_PER_PAGE_MASK, map->page);
+}
+
+static inline unsigned qpn_hash(struct qib_ibdev *dev, u32 qpn)
+{
+	return jhash_1word(qpn, dev->qp_rnd) &
+		(dev->qp_table_size - 1);
+}
+
+
+/*
+ * Put the QP into the hash table.
+ * The hash table holds a reference to the QP.
+ */
+static void insert_qp(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	unsigned long flags;
+	unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
+
+	atomic_inc(&qp->refcount);
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+
+	if (qp->ibqp.qp_num == 0)
+		rcu_assign_pointer(ibp->qp0, qp);
+	else if (qp->ibqp.qp_num == 1)
+		rcu_assign_pointer(ibp->qp1, qp);
+	else {
+		qp->next = dev->qp_table[n];
+		rcu_assign_pointer(dev->qp_table[n], qp);
+	}
+
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+}
+
+/*
+ * Remove the QP from the table so it can't be found asynchronously by
+ * the receive interrupt routine.
+ */
+static void remove_qp(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	unsigned n = qpn_hash(dev, qp->ibqp.qp_num);
+	unsigned long flags;
+	int removed = 1;
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+
+	if (rcu_dereference_protected(ibp->qp0,
+			lockdep_is_held(&dev->qpt_lock)) == qp) {
+		rcu_assign_pointer(ibp->qp0, NULL);
+	} else if (rcu_dereference_protected(ibp->qp1,
+			lockdep_is_held(&dev->qpt_lock)) == qp) {
+		rcu_assign_pointer(ibp->qp1, NULL);
+	} else {
+		struct qib_qp *q;
+		struct qib_qp __rcu **qpp;
+
+		removed = 0;
+		qpp = &dev->qp_table[n];
+		for (; (q = rcu_dereference_protected(*qpp,
+				lockdep_is_held(&dev->qpt_lock))) != NULL;
+				qpp = &q->next)
+			if (q == qp) {
+				rcu_assign_pointer(*qpp,
+					rcu_dereference_protected(qp->next,
+					 lockdep_is_held(&dev->qpt_lock)));
+				removed = 1;
+				break;
+			}
+	}
+
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	if (removed) {
+		synchronize_rcu();
+		atomic_dec(&qp->refcount);
+	}
+}
+
+/**
+ * qib_free_all_qps - check for QPs still in use
+ * @qpt: the QP table to empty
+ *
+ * There should not be any QPs still in use.
+ * Free memory for table.
+ */
+unsigned qib_free_all_qps(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	unsigned long flags;
+	struct qib_qp *qp;
+	unsigned n, qp_inuse = 0;
+
+	for (n = 0; n < dd->num_pports; n++) {
+		struct qib_ibport *ibp = &dd->pport[n].ibport_data;
+
+		if (!qib_mcast_tree_empty(ibp))
+			qp_inuse++;
+		rcu_read_lock();
+		if (rcu_dereference(ibp->qp0))
+			qp_inuse++;
+		if (rcu_dereference(ibp->qp1))
+			qp_inuse++;
+		rcu_read_unlock();
+	}
+
+	spin_lock_irqsave(&dev->qpt_lock, flags);
+	for (n = 0; n < dev->qp_table_size; n++) {
+		qp = rcu_dereference_protected(dev->qp_table[n],
+			lockdep_is_held(&dev->qpt_lock));
+		rcu_assign_pointer(dev->qp_table[n], NULL);
+
+		for (; qp; qp = rcu_dereference_protected(qp->next,
+					lockdep_is_held(&dev->qpt_lock)))
+			qp_inuse++;
+	}
+	spin_unlock_irqrestore(&dev->qpt_lock, flags);
+	synchronize_rcu();
+
+	return qp_inuse;
+}
+
+/**
+ * qib_lookup_qpn - return the QP with the given QPN
+ * @qpt: the QP table
+ * @qpn: the QP number to look up
+ *
+ * The caller is responsible for decrementing the QP reference count
+ * when done.
+ */
+struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn)
+{
+	struct qib_qp *qp = NULL;
+
+	rcu_read_lock();
+	if (unlikely(qpn <= 1)) {
+		if (qpn == 0)
+			qp = rcu_dereference(ibp->qp0);
+		else
+			qp = rcu_dereference(ibp->qp1);
+		if (qp)
+			atomic_inc(&qp->refcount);
+	} else {
+		struct qib_ibdev *dev = &ppd_from_ibp(ibp)->dd->verbs_dev;
+		unsigned n = qpn_hash(dev, qpn);
+
+		for (qp = rcu_dereference(dev->qp_table[n]); qp;
+			qp = rcu_dereference(qp->next))
+			if (qp->ibqp.qp_num == qpn) {
+				atomic_inc(&qp->refcount);
+				break;
+			}
+	}
+	rcu_read_unlock();
+	return qp;
+}
+
+/**
+ * qib_reset_qp - initialize the QP state to the reset state
+ * @qp: the QP to reset
+ * @type: the QP type
+ */
+static void qib_reset_qp(struct qib_qp *qp, enum ib_qp_type type)
+{
+	qp->remote_qpn = 0;
+	qp->qkey = 0;
+	qp->qp_access_flags = 0;
+	atomic_set(&qp->s_dma_busy, 0);
+	qp->s_flags &= QIB_S_SIGNAL_REQ_WR;
+	qp->s_hdrwords = 0;
+	qp->s_wqe = NULL;
+	qp->s_draining = 0;
+	qp->s_next_psn = 0;
+	qp->s_last_psn = 0;
+	qp->s_sending_psn = 0;
+	qp->s_sending_hpsn = 0;
+	qp->s_psn = 0;
+	qp->r_psn = 0;
+	qp->r_msn = 0;
+	if (type == IB_QPT_RC) {
+		qp->s_state = IB_OPCODE_RC_SEND_LAST;
+		qp->r_state = IB_OPCODE_RC_SEND_LAST;
+	} else {
+		qp->s_state = IB_OPCODE_UC_SEND_LAST;
+		qp->r_state = IB_OPCODE_UC_SEND_LAST;
+	}
+	qp->s_ack_state = IB_OPCODE_RC_ACKNOWLEDGE;
+	qp->r_nak_state = 0;
+	qp->r_aflags = 0;
+	qp->r_flags = 0;
+	qp->s_head = 0;
+	qp->s_tail = 0;
+	qp->s_cur = 0;
+	qp->s_acked = 0;
+	qp->s_last = 0;
+	qp->s_ssn = 1;
+	qp->s_lsn = 0;
+	qp->s_mig_state = IB_MIG_MIGRATED;
+	memset(qp->s_ack_queue, 0, sizeof(qp->s_ack_queue));
+	qp->r_head_ack_queue = 0;
+	qp->s_tail_ack_queue = 0;
+	qp->s_num_rd_atomic = 0;
+	if (qp->r_rq.wq) {
+		qp->r_rq.wq->head = 0;
+		qp->r_rq.wq->tail = 0;
+	}
+	qp->r_sge.num_sge = 0;
+}
+
+static void clear_mr_refs(struct qib_qp *qp, int clr_sends)
+{
+	unsigned n;
+
+	if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
+		qib_put_ss(&qp->s_rdma_read_sge);
+
+	qib_put_ss(&qp->r_sge);
+
+	if (clr_sends) {
+		while (qp->s_last != qp->s_head) {
+			struct qib_swqe *wqe = get_swqe_ptr(qp, qp->s_last);
+			unsigned i;
+
+			for (i = 0; i < wqe->wr.num_sge; i++) {
+				struct qib_sge *sge = &wqe->sg_list[i];
+
+				qib_put_mr(sge->mr);
+			}
+			if (qp->ibqp.qp_type == IB_QPT_UD ||
+			    qp->ibqp.qp_type == IB_QPT_SMI ||
+			    qp->ibqp.qp_type == IB_QPT_GSI)
+				atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+			if (++qp->s_last >= qp->s_size)
+				qp->s_last = 0;
+		}
+		if (qp->s_rdma_mr) {
+			qib_put_mr(qp->s_rdma_mr);
+			qp->s_rdma_mr = NULL;
+		}
+	}
+
+	if (qp->ibqp.qp_type != IB_QPT_RC)
+		return;
+
+	for (n = 0; n < ARRAY_SIZE(qp->s_ack_queue); n++) {
+		struct qib_ack_entry *e = &qp->s_ack_queue[n];
+
+		if (e->opcode == IB_OPCODE_RC_RDMA_READ_REQUEST &&
+		    e->rdma_sge.mr) {
+			qib_put_mr(e->rdma_sge.mr);
+			e->rdma_sge.mr = NULL;
+		}
+	}
+}
+
+/**
+ * qib_error_qp - put a QP into the error state
+ * @qp: the QP to put into the error state
+ * @err: the receive completion error to signal if a RWQE is active
+ *
+ * Flushes both send and receive work queues.
+ * Returns true if last WQE event should be generated.
+ * The QP r_lock and s_lock should be held and interrupts disabled.
+ * If we are already in error state, just return.
+ */
+int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct ib_wc wc;
+	int ret = 0;
+
+	if (qp->state == IB_QPS_ERR || qp->state == IB_QPS_RESET)
+		goto bail;
+
+	qp->state = IB_QPS_ERR;
+
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+
+	if (qp->s_flags & QIB_S_ANY_WAIT_SEND)
+		qp->s_flags &= ~QIB_S_ANY_WAIT_SEND;
+
+	spin_lock(&dev->pending_lock);
+	if (!list_empty(&qp->iowait) && !(qp->s_flags & QIB_S_BUSY)) {
+		qp->s_flags &= ~QIB_S_ANY_WAIT_IO;
+		list_del_init(&qp->iowait);
+	}
+	spin_unlock(&dev->pending_lock);
+
+	if (!(qp->s_flags & QIB_S_BUSY)) {
+		qp->s_hdrwords = 0;
+		if (qp->s_rdma_mr) {
+			qib_put_mr(qp->s_rdma_mr);
+			qp->s_rdma_mr = NULL;
+		}
+		if (qp->s_tx) {
+			qib_put_txreq(qp->s_tx);
+			qp->s_tx = NULL;
+		}
+	}
+
+	/* Schedule the sending tasklet to drain the send work queue. */
+	if (qp->s_last != qp->s_head)
+		qib_schedule_send(qp);
+
+	clear_mr_refs(qp, 0);
+
+	memset(&wc, 0, sizeof(wc));
+	wc.qp = &qp->ibqp;
+	wc.opcode = IB_WC_RECV;
+
+	if (test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags)) {
+		wc.wr_id = qp->r_wr_id;
+		wc.status = err;
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	}
+	wc.status = IB_WC_WR_FLUSH_ERR;
+
+	if (qp->r_rq.wq) {
+		struct qib_rwq *wq;
+		u32 head;
+		u32 tail;
+
+		spin_lock(&qp->r_rq.lock);
+
+		/* sanity check pointers before trusting them */
+		wq = qp->r_rq.wq;
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
+		while (tail != head) {
+			wc.wr_id = get_rwqe_ptr(&qp->r_rq, tail)->wr_id;
+			if (++tail >= qp->r_rq.size)
+				tail = 0;
+			qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+		}
+		wq->tail = tail;
+
+		spin_unlock(&qp->r_rq.lock);
+	} else if (qp->ibqp.event_handler)
+		ret = 1;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_modify_qp - modify the attributes of a queue pair
+ * @ibqp: the queue pair who's attributes we're modifying
+ * @attr: the new attributes
+ * @attr_mask: the mask of attributes to modify
+ * @udata: user data for libibverbs.so
+ *
+ * Returns 0 on success, otherwise returns an errno.
+ */
+int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		  int attr_mask, struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_qp *qp = to_iqp(ibqp);
+	enum ib_qp_state cur_state, new_state;
+	struct ib_event ev;
+	int lastwqe = 0;
+	int mig = 0;
+	int ret;
+	u32 pmtu = 0; /* for gcc warning only */
+
+	spin_lock_irq(&qp->r_lock);
+	spin_lock(&qp->s_lock);
+
+	cur_state = attr_mask & IB_QP_CUR_STATE ?
+		attr->cur_qp_state : qp->state;
+	new_state = attr_mask & IB_QP_STATE ? attr->qp_state : cur_state;
+
+	if (!ib_modify_qp_is_ok(cur_state, new_state, ibqp->qp_type,
+				attr_mask, IB_LINK_LAYER_UNSPECIFIED))
+		goto inval;
+
+	if (attr_mask & IB_QP_AV) {
+		if (attr->ah_attr.dlid >= QIB_MULTICAST_LID_BASE)
+			goto inval;
+		if (qib_check_ah(qp->ibqp.device, &attr->ah_attr))
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		if (attr->alt_ah_attr.dlid >= QIB_MULTICAST_LID_BASE)
+			goto inval;
+		if (qib_check_ah(qp->ibqp.device, &attr->alt_ah_attr))
+			goto inval;
+		if (attr->alt_pkey_index >= qib_get_npkeys(dd_from_dev(dev)))
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		if (attr->pkey_index >= qib_get_npkeys(dd_from_dev(dev)))
+			goto inval;
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		if (attr->min_rnr_timer > 31)
+			goto inval;
+
+	if (attr_mask & IB_QP_PORT)
+		if (qp->ibqp.qp_type == IB_QPT_SMI ||
+		    qp->ibqp.qp_type == IB_QPT_GSI ||
+		    attr->port_num == 0 ||
+		    attr->port_num > ibqp->device->phys_port_cnt)
+			goto inval;
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		if (attr->dest_qp_num > QIB_QPN_MASK)
+			goto inval;
+
+	if (attr_mask & IB_QP_RETRY_CNT)
+		if (attr->retry_cnt > 7)
+			goto inval;
+
+	if (attr_mask & IB_QP_RNR_RETRY)
+		if (attr->rnr_retry > 7)
+			goto inval;
+
+	/*
+	 * Don't allow invalid path_mtu values.  OK to set greater
+	 * than the active mtu (or even the max_cap, if we have tuned
+	 * that to a small mtu.  We'll set qp->path_mtu
+	 * to the lesser of requested attribute mtu and active,
+	 * for packetizing messages.
+	 * Note that the QP port has to be set in INIT and MTU in RTR.
+	 */
+	if (attr_mask & IB_QP_PATH_MTU) {
+		struct qib_devdata *dd = dd_from_dev(dev);
+		int mtu, pidx = qp->port_num - 1;
+
+		mtu = ib_mtu_enum_to_int(attr->path_mtu);
+		if (mtu == -1)
+			goto inval;
+		if (mtu > dd->pport[pidx].ibmtu) {
+			switch (dd->pport[pidx].ibmtu) {
+			case 4096:
+				pmtu = IB_MTU_4096;
+				break;
+			case 2048:
+				pmtu = IB_MTU_2048;
+				break;
+			case 1024:
+				pmtu = IB_MTU_1024;
+				break;
+			case 512:
+				pmtu = IB_MTU_512;
+				break;
+			case 256:
+				pmtu = IB_MTU_256;
+				break;
+			default:
+				pmtu = IB_MTU_2048;
+			}
+		} else
+			pmtu = attr->path_mtu;
+	}
+
+	if (attr_mask & IB_QP_PATH_MIG_STATE) {
+		if (attr->path_mig_state == IB_MIG_REARM) {
+			if (qp->s_mig_state == IB_MIG_ARMED)
+				goto inval;
+			if (new_state != IB_QPS_RTS)
+				goto inval;
+		} else if (attr->path_mig_state == IB_MIG_MIGRATED) {
+			if (qp->s_mig_state == IB_MIG_REARM)
+				goto inval;
+			if (new_state != IB_QPS_RTS && new_state != IB_QPS_SQD)
+				goto inval;
+			if (qp->s_mig_state == IB_MIG_ARMED)
+				mig = 1;
+		} else
+			goto inval;
+	}
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		if (attr->max_dest_rd_atomic > QIB_MAX_RDMA_ATOMIC)
+			goto inval;
+
+	switch (new_state) {
+	case IB_QPS_RESET:
+		if (qp->state != IB_QPS_RESET) {
+			qp->state = IB_QPS_RESET;
+			spin_lock(&dev->pending_lock);
+			if (!list_empty(&qp->iowait))
+				list_del_init(&qp->iowait);
+			spin_unlock(&dev->pending_lock);
+			qp->s_flags &= ~(QIB_S_TIMER | QIB_S_ANY_WAIT);
+			spin_unlock(&qp->s_lock);
+			spin_unlock_irq(&qp->r_lock);
+			/* Stop the sending work queue and retry timer */
+			cancel_work_sync(&qp->s_work);
+			del_timer_sync(&qp->s_timer);
+			wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+			if (qp->s_tx) {
+				qib_put_txreq(qp->s_tx);
+				qp->s_tx = NULL;
+			}
+			remove_qp(dev, qp);
+			wait_event(qp->wait, !atomic_read(&qp->refcount));
+			spin_lock_irq(&qp->r_lock);
+			spin_lock(&qp->s_lock);
+			clear_mr_refs(qp, 1);
+			qib_reset_qp(qp, ibqp->qp_type);
+		}
+		break;
+
+	case IB_QPS_RTR:
+		/* Allow event to retrigger if QP set to RTR more than once */
+		qp->r_flags &= ~QIB_R_COMM_EST;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_SQD:
+		qp->s_draining = qp->s_last != qp->s_cur;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_SQE:
+		if (qp->ibqp.qp_type == IB_QPT_RC)
+			goto inval;
+		qp->state = new_state;
+		break;
+
+	case IB_QPS_ERR:
+		lastwqe = qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+		break;
+
+	default:
+		qp->state = new_state;
+		break;
+	}
+
+	if (attr_mask & IB_QP_PKEY_INDEX)
+		qp->s_pkey_index = attr->pkey_index;
+
+	if (attr_mask & IB_QP_PORT)
+		qp->port_num = attr->port_num;
+
+	if (attr_mask & IB_QP_DEST_QPN)
+		qp->remote_qpn = attr->dest_qp_num;
+
+	if (attr_mask & IB_QP_SQ_PSN) {
+		qp->s_next_psn = attr->sq_psn & QIB_PSN_MASK;
+		qp->s_psn = qp->s_next_psn;
+		qp->s_sending_psn = qp->s_next_psn;
+		qp->s_last_psn = qp->s_next_psn - 1;
+		qp->s_sending_hpsn = qp->s_last_psn;
+	}
+
+	if (attr_mask & IB_QP_RQ_PSN)
+		qp->r_psn = attr->rq_psn & QIB_PSN_MASK;
+
+	if (attr_mask & IB_QP_ACCESS_FLAGS)
+		qp->qp_access_flags = attr->qp_access_flags;
+
+	if (attr_mask & IB_QP_AV) {
+		qp->remote_ah_attr = attr->ah_attr;
+		qp->s_srate = attr->ah_attr.static_rate;
+	}
+
+	if (attr_mask & IB_QP_ALT_PATH) {
+		qp->alt_ah_attr = attr->alt_ah_attr;
+		qp->s_alt_pkey_index = attr->alt_pkey_index;
+	}
+
+	if (attr_mask & IB_QP_PATH_MIG_STATE) {
+		qp->s_mig_state = attr->path_mig_state;
+		if (mig) {
+			qp->remote_ah_attr = qp->alt_ah_attr;
+			qp->port_num = qp->alt_ah_attr.port_num;
+			qp->s_pkey_index = qp->s_alt_pkey_index;
+		}
+	}
+
+	if (attr_mask & IB_QP_PATH_MTU) {
+		qp->path_mtu = pmtu;
+		qp->pmtu = ib_mtu_enum_to_int(pmtu);
+	}
+
+	if (attr_mask & IB_QP_RETRY_CNT) {
+		qp->s_retry_cnt = attr->retry_cnt;
+		qp->s_retry = attr->retry_cnt;
+	}
+
+	if (attr_mask & IB_QP_RNR_RETRY) {
+		qp->s_rnr_retry_cnt = attr->rnr_retry;
+		qp->s_rnr_retry = attr->rnr_retry;
+	}
+
+	if (attr_mask & IB_QP_MIN_RNR_TIMER)
+		qp->r_min_rnr_timer = attr->min_rnr_timer;
+
+	if (attr_mask & IB_QP_TIMEOUT) {
+		qp->timeout = attr->timeout;
+		qp->timeout_jiffies =
+			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+				1000UL);
+	}
+
+	if (attr_mask & IB_QP_QKEY)
+		qp->qkey = attr->qkey;
+
+	if (attr_mask & IB_QP_MAX_DEST_RD_ATOMIC)
+		qp->r_max_rd_atomic = attr->max_dest_rd_atomic;
+
+	if (attr_mask & IB_QP_MAX_QP_RD_ATOMIC)
+		qp->s_max_rd_atomic = attr->max_rd_atomic;
+
+	spin_unlock(&qp->s_lock);
+	spin_unlock_irq(&qp->r_lock);
+
+	if (cur_state == IB_QPS_RESET && new_state == IB_QPS_INIT)
+		insert_qp(dev, qp);
+
+	if (lastwqe) {
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+	if (mig) {
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_PATH_MIG;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+	ret = 0;
+	goto bail;
+
+inval:
+	spin_unlock(&qp->s_lock);
+	spin_unlock_irq(&qp->r_lock);
+	ret = -EINVAL;
+
+bail:
+	return ret;
+}
+
+int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		 int attr_mask, struct ib_qp_init_attr *init_attr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+
+	attr->qp_state = qp->state;
+	attr->cur_qp_state = attr->qp_state;
+	attr->path_mtu = qp->path_mtu;
+	attr->path_mig_state = qp->s_mig_state;
+	attr->qkey = qp->qkey;
+	attr->rq_psn = qp->r_psn & QIB_PSN_MASK;
+	attr->sq_psn = qp->s_next_psn & QIB_PSN_MASK;
+	attr->dest_qp_num = qp->remote_qpn;
+	attr->qp_access_flags = qp->qp_access_flags;
+	attr->cap.max_send_wr = qp->s_size - 1;
+	attr->cap.max_recv_wr = qp->ibqp.srq ? 0 : qp->r_rq.size - 1;
+	attr->cap.max_send_sge = qp->s_max_sge;
+	attr->cap.max_recv_sge = qp->r_rq.max_sge;
+	attr->cap.max_inline_data = 0;
+	attr->ah_attr = qp->remote_ah_attr;
+	attr->alt_ah_attr = qp->alt_ah_attr;
+	attr->pkey_index = qp->s_pkey_index;
+	attr->alt_pkey_index = qp->s_alt_pkey_index;
+	attr->en_sqd_async_notify = 0;
+	attr->sq_draining = qp->s_draining;
+	attr->max_rd_atomic = qp->s_max_rd_atomic;
+	attr->max_dest_rd_atomic = qp->r_max_rd_atomic;
+	attr->min_rnr_timer = qp->r_min_rnr_timer;
+	attr->port_num = qp->port_num;
+	attr->timeout = qp->timeout;
+	attr->retry_cnt = qp->s_retry_cnt;
+	attr->rnr_retry = qp->s_rnr_retry_cnt;
+	attr->alt_port_num = qp->alt_ah_attr.port_num;
+	attr->alt_timeout = qp->alt_timeout;
+
+	init_attr->event_handler = qp->ibqp.event_handler;
+	init_attr->qp_context = qp->ibqp.qp_context;
+	init_attr->send_cq = qp->ibqp.send_cq;
+	init_attr->recv_cq = qp->ibqp.recv_cq;
+	init_attr->srq = qp->ibqp.srq;
+	init_attr->cap = attr->cap;
+	if (qp->s_flags & QIB_S_SIGNAL_REQ_WR)
+		init_attr->sq_sig_type = IB_SIGNAL_REQ_WR;
+	else
+		init_attr->sq_sig_type = IB_SIGNAL_ALL_WR;
+	init_attr->qp_type = qp->ibqp.qp_type;
+	init_attr->port_num = qp->port_num;
+	return 0;
+}
+
+/**
+ * qib_compute_aeth - compute the AETH (syndrome + MSN)
+ * @qp: the queue pair to compute the AETH for
+ *
+ * Returns the AETH.
+ */
+__be32 qib_compute_aeth(struct qib_qp *qp)
+{
+	u32 aeth = qp->r_msn & QIB_MSN_MASK;
+
+	if (qp->ibqp.srq) {
+		/*
+		 * Shared receive queues don't generate credits.
+		 * Set the credit field to the invalid value.
+		 */
+		aeth |= QIB_AETH_CREDIT_INVAL << QIB_AETH_CREDIT_SHIFT;
+	} else {
+		u32 min, max, x;
+		u32 credits;
+		struct qib_rwq *wq = qp->r_rq.wq;
+		u32 head;
+		u32 tail;
+
+		/* sanity check pointers before trusting them */
+		head = wq->head;
+		if (head >= qp->r_rq.size)
+			head = 0;
+		tail = wq->tail;
+		if (tail >= qp->r_rq.size)
+			tail = 0;
+		/*
+		 * Compute the number of credits available (RWQEs).
+		 * XXX Not holding the r_rq.lock here so there is a small
+		 * chance that the pair of reads are not atomic.
+		 */
+		credits = head - tail;
+		if ((int)credits < 0)
+			credits += qp->r_rq.size;
+		/*
+		 * Binary search the credit table to find the code to
+		 * use.
+		 */
+		min = 0;
+		max = 31;
+		for (;;) {
+			x = (min + max) / 2;
+			if (credit_table[x] == credits)
+				break;
+			if (credit_table[x] > credits)
+				max = x;
+			else if (min == x)
+				break;
+			else
+				min = x;
+		}
+		aeth |= x << QIB_AETH_CREDIT_SHIFT;
+	}
+	return cpu_to_be32(aeth);
+}
+
+/**
+ * qib_create_qp - create a queue pair for a device
+ * @ibpd: the protection domain who's device we create the queue pair for
+ * @init_attr: the attributes of the queue pair
+ * @udata: user data for libibverbs.so
+ *
+ * Returns the queue pair on success, otherwise returns an errno.
+ *
+ * Called by the ib_create_qp() core verbs function.
+ */
+struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata)
+{
+	struct qib_qp *qp;
+	int err;
+	struct qib_swqe *swq = NULL;
+	struct qib_ibdev *dev;
+	struct qib_devdata *dd;
+	size_t sz;
+	size_t sg_list_sz;
+	struct ib_qp *ret;
+
+	if (init_attr->cap.max_send_sge > ib_qib_max_sges ||
+	    init_attr->cap.max_send_wr > ib_qib_max_qp_wrs ||
+	    init_attr->create_flags) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	/* Check receive queue parameters if no SRQ is specified. */
+	if (!init_attr->srq) {
+		if (init_attr->cap.max_recv_sge > ib_qib_max_sges ||
+		    init_attr->cap.max_recv_wr > ib_qib_max_qp_wrs) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+		if (init_attr->cap.max_send_sge +
+		    init_attr->cap.max_send_wr +
+		    init_attr->cap.max_recv_sge +
+		    init_attr->cap.max_recv_wr == 0) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+	}
+
+	switch (init_attr->qp_type) {
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		if (init_attr->port_num == 0 ||
+		    init_attr->port_num > ibpd->device->phys_port_cnt) {
+			ret = ERR_PTR(-EINVAL);
+			goto bail;
+		}
+	case IB_QPT_UC:
+	case IB_QPT_RC:
+	case IB_QPT_UD:
+		sz = sizeof(struct qib_sge) *
+			init_attr->cap.max_send_sge +
+			sizeof(struct qib_swqe);
+		swq = vmalloc((init_attr->cap.max_send_wr + 1) * sz);
+		if (swq == NULL) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail;
+		}
+		sz = sizeof(*qp);
+		sg_list_sz = 0;
+		if (init_attr->srq) {
+			struct qib_srq *srq = to_isrq(init_attr->srq);
+
+			if (srq->rq.max_sge > 1)
+				sg_list_sz = sizeof(*qp->r_sg_list) *
+					(srq->rq.max_sge - 1);
+		} else if (init_attr->cap.max_recv_sge > 1)
+			sg_list_sz = sizeof(*qp->r_sg_list) *
+				(init_attr->cap.max_recv_sge - 1);
+		qp = kzalloc(sz + sg_list_sz, GFP_KERNEL);
+		if (!qp) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_swq;
+		}
+		RCU_INIT_POINTER(qp->next, NULL);
+		qp->s_hdr = kzalloc(sizeof(*qp->s_hdr), GFP_KERNEL);
+		if (!qp->s_hdr) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_qp;
+		}
+		qp->timeout_jiffies =
+			usecs_to_jiffies((4096UL * (1UL << qp->timeout)) /
+				1000UL);
+		if (init_attr->srq)
+			sz = 0;
+		else {
+			qp->r_rq.size = init_attr->cap.max_recv_wr + 1;
+			qp->r_rq.max_sge = init_attr->cap.max_recv_sge;
+			sz = (sizeof(struct ib_sge) * qp->r_rq.max_sge) +
+				sizeof(struct qib_rwqe);
+			qp->r_rq.wq = vmalloc_user(sizeof(struct qib_rwq) +
+						   qp->r_rq.size * sz);
+			if (!qp->r_rq.wq) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_qp;
+			}
+		}
+
+		/*
+		 * ib_create_qp() will initialize qp->ibqp
+		 * except for qp->ibqp.qp_num.
+		 */
+		spin_lock_init(&qp->r_lock);
+		spin_lock_init(&qp->s_lock);
+		spin_lock_init(&qp->r_rq.lock);
+		atomic_set(&qp->refcount, 0);
+		init_waitqueue_head(&qp->wait);
+		init_waitqueue_head(&qp->wait_dma);
+		init_timer(&qp->s_timer);
+		qp->s_timer.data = (unsigned long)qp;
+		INIT_WORK(&qp->s_work, qib_do_send);
+		INIT_LIST_HEAD(&qp->iowait);
+		INIT_LIST_HEAD(&qp->rspwait);
+		qp->state = IB_QPS_RESET;
+		qp->s_wq = swq;
+		qp->s_size = init_attr->cap.max_send_wr + 1;
+		qp->s_max_sge = init_attr->cap.max_send_sge;
+		if (init_attr->sq_sig_type == IB_SIGNAL_REQ_WR)
+			qp->s_flags = QIB_S_SIGNAL_REQ_WR;
+		dev = to_idev(ibpd->device);
+		dd = dd_from_dev(dev);
+		err = alloc_qpn(dd, &dev->qpn_table, init_attr->qp_type,
+				init_attr->port_num);
+		if (err < 0) {
+			ret = ERR_PTR(err);
+			vfree(qp->r_rq.wq);
+			goto bail_qp;
+		}
+		qp->ibqp.qp_num = err;
+		qp->port_num = init_attr->port_num;
+		qib_reset_qp(qp, init_attr->qp_type);
+		break;
+
+	default:
+		/* Don't support raw QPs */
+		ret = ERR_PTR(-ENOSYS);
+		goto bail;
+	}
+
+	init_attr->cap.max_inline_data = 0;
+
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		if (!qp->r_rq.wq) {
+			__u64 offset = 0;
+
+			err = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_ip;
+			}
+		} else {
+			u32 s = sizeof(struct qib_rwq) + qp->r_rq.size * sz;
+
+			qp->ip = qib_create_mmap_info(dev, s,
+						      ibpd->uobject->context,
+						      qp->r_rq.wq);
+			if (!qp->ip) {
+				ret = ERR_PTR(-ENOMEM);
+				goto bail_ip;
+			}
+
+			err = ib_copy_to_udata(udata, &(qp->ip->offset),
+					       sizeof(qp->ip->offset));
+			if (err) {
+				ret = ERR_PTR(err);
+				goto bail_ip;
+			}
+		}
+	}
+
+	spin_lock(&dev->n_qps_lock);
+	if (dev->n_qps_allocated == ib_qib_max_qps) {
+		spin_unlock(&dev->n_qps_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_qps_allocated++;
+	spin_unlock(&dev->n_qps_lock);
+
+	if (qp->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&qp->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = &qp->ibqp;
+	goto bail;
+
+bail_ip:
+	if (qp->ip)
+		kref_put(&qp->ip->ref, qib_release_mmap_info);
+	else
+		vfree(qp->r_rq.wq);
+	free_qpn(&dev->qpn_table, qp->ibqp.qp_num);
+bail_qp:
+	kfree(qp->s_hdr);
+	kfree(qp);
+bail_swq:
+	vfree(swq);
+bail:
+	return ret;
+}
+
+/**
+ * qib_destroy_qp - destroy a queue pair
+ * @ibqp: the queue pair to destroy
+ *
+ * Returns 0 on success.
+ *
+ * Note that this can be called while the QP is actively sending or
+ * receiving!
+ */
+int qib_destroy_qp(struct ib_qp *ibqp)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+
+	/* Make sure HW and driver activity is stopped. */
+	spin_lock_irq(&qp->s_lock);
+	if (qp->state != IB_QPS_RESET) {
+		qp->state = IB_QPS_RESET;
+		spin_lock(&dev->pending_lock);
+		if (!list_empty(&qp->iowait))
+			list_del_init(&qp->iowait);
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_ANY_WAIT);
+		spin_unlock_irq(&qp->s_lock);
+		cancel_work_sync(&qp->s_work);
+		del_timer_sync(&qp->s_timer);
+		wait_event(qp->wait_dma, !atomic_read(&qp->s_dma_busy));
+		if (qp->s_tx) {
+			qib_put_txreq(qp->s_tx);
+			qp->s_tx = NULL;
+		}
+		remove_qp(dev, qp);
+		wait_event(qp->wait, !atomic_read(&qp->refcount));
+		clear_mr_refs(qp, 1);
+	} else
+		spin_unlock_irq(&qp->s_lock);
+
+	/* all user's cleaned up, mark it available */
+	free_qpn(&dev->qpn_table, qp->ibqp.qp_num);
+	spin_lock(&dev->n_qps_lock);
+	dev->n_qps_allocated--;
+	spin_unlock(&dev->n_qps_lock);
+
+	if (qp->ip)
+		kref_put(&qp->ip->ref, qib_release_mmap_info);
+	else
+		vfree(qp->r_rq.wq);
+	vfree(qp->s_wq);
+	kfree(qp->s_hdr);
+	kfree(qp);
+	return 0;
+}
+
+/**
+ * qib_init_qpn_table - initialize the QP number table for a device
+ * @qpt: the QPN table
+ */
+void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt)
+{
+	spin_lock_init(&qpt->lock);
+	qpt->last = 1;          /* start with QPN 2 */
+	qpt->nmaps = 1;
+	qpt->mask = dd->qpn_mask;
+}
+
+/**
+ * qib_free_qpn_table - free the QP number table for a device
+ * @qpt: the QPN table
+ */
+void qib_free_qpn_table(struct qib_qpn_table *qpt)
+{
+	int i;
+
+	for (i = 0; i < ARRAY_SIZE(qpt->map); i++)
+		if (qpt->map[i].page)
+			free_page((unsigned long) qpt->map[i].page);
+}
+
+/**
+ * qib_get_credit - flush the send work queue of a QP
+ * @qp: the qp who's send work queue to flush
+ * @aeth: the Acknowledge Extended Transport Header
+ *
+ * The QP s_lock should be held.
+ */
+void qib_get_credit(struct qib_qp *qp, u32 aeth)
+{
+	u32 credit = (aeth >> QIB_AETH_CREDIT_SHIFT) & QIB_AETH_CREDIT_MASK;
+
+	/*
+	 * If the credit is invalid, we can send
+	 * as many packets as we like.  Otherwise, we have to
+	 * honor the credit field.
+	 */
+	if (credit == QIB_AETH_CREDIT_INVAL) {
+		if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT)) {
+			qp->s_flags |= QIB_S_UNLIMITED_CREDIT;
+			if (qp->s_flags & QIB_S_WAIT_SSN_CREDIT) {
+				qp->s_flags &= ~QIB_S_WAIT_SSN_CREDIT;
+				qib_schedule_send(qp);
+			}
+		}
+	} else if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT)) {
+		/* Compute new LSN (i.e., MSN + credit) */
+		credit = (aeth + credit_table[credit]) & QIB_MSN_MASK;
+		if (qib_cmp24(credit, qp->s_lsn) > 0) {
+			qp->s_lsn = credit;
+			if (qp->s_flags & QIB_S_WAIT_SSN_CREDIT) {
+				qp->s_flags &= ~QIB_S_WAIT_SSN_CREDIT;
+				qib_schedule_send(qp);
+			}
+		}
+	}
+}
+
+#ifdef CONFIG_DEBUG_FS
+
+struct qib_qp_iter {
+	struct qib_ibdev *dev;
+	struct qib_qp *qp;
+	int n;
+};
+
+struct qib_qp_iter *qib_qp_iter_init(struct qib_ibdev *dev)
+{
+	struct qib_qp_iter *iter;
+
+	iter = kzalloc(sizeof(*iter), GFP_KERNEL);
+	if (!iter)
+		return NULL;
+
+	iter->dev = dev;
+	if (qib_qp_iter_next(iter)) {
+		kfree(iter);
+		return NULL;
+	}
+
+	return iter;
+}
+
+int qib_qp_iter_next(struct qib_qp_iter *iter)
+{
+	struct qib_ibdev *dev = iter->dev;
+	int n = iter->n;
+	int ret = 1;
+	struct qib_qp *pqp = iter->qp;
+	struct qib_qp *qp;
+
+	rcu_read_lock();
+	for (; n < dev->qp_table_size; n++) {
+		if (pqp)
+			qp = rcu_dereference(pqp->next);
+		else
+			qp = rcu_dereference(dev->qp_table[n]);
+		pqp = qp;
+		if (qp) {
+			if (iter->qp)
+				atomic_dec(&iter->qp->refcount);
+			atomic_inc(&qp->refcount);
+			rcu_read_unlock();
+			iter->qp = qp;
+			iter->n = n;
+			return 0;
+		}
+	}
+	rcu_read_unlock();
+	if (iter->qp)
+		atomic_dec(&iter->qp->refcount);
+	return ret;
+}
+
+static const char * const qp_type_str[] = {
+	"SMI", "GSI", "RC", "UC", "UD",
+};
+
+void qib_qp_iter_print(struct seq_file *s, struct qib_qp_iter *iter)
+{
+	struct qib_swqe *wqe;
+	struct qib_qp *qp = iter->qp;
+
+	wqe = get_swqe_ptr(qp, qp->s_last);
+	seq_printf(s,
+		   "N %d QP%u %s %u %u %u f=%x %u %u %u %u %u PSN %x %x %x %x %x (%u %u %u %u %u %u) QP%u LID %x\n",
+		   iter->n,
+		   qp->ibqp.qp_num,
+		   qp_type_str[qp->ibqp.qp_type],
+		   qp->state,
+		   wqe->wr.opcode,
+		   qp->s_hdrwords,
+		   qp->s_flags,
+		   atomic_read(&qp->s_dma_busy),
+		   !list_empty(&qp->iowait),
+		   qp->timeout,
+		   wqe->ssn,
+		   qp->s_lsn,
+		   qp->s_last_psn,
+		   qp->s_psn, qp->s_next_psn,
+		   qp->s_sending_psn, qp->s_sending_hpsn,
+		   qp->s_last, qp->s_acked, qp->s_cur,
+		   qp->s_tail, qp->s_head, qp->s_size,
+		   qp->remote_qpn,
+		   qp->remote_ah_attr.dlid);
+}
+
+#endif
diff --git a/drivers/infiniband/hw/qib/qib_qsfp.c b/drivers/infiniband/hw/qib/qib_qsfp.c
new file mode 100644
index 00000000000..fa71b1e666c
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qsfp.c
@@ -0,0 +1,556 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+#include "qib_qsfp.h"
+
+/*
+ * QSFP support for ib_qib driver, using "Two Wire Serial Interface" driver
+ * in qib_twsi.c
+ */
+#define QSFP_MAX_RETRY 4
+
+static int qsfp_read(struct qib_pportdata *ppd, int addr, void *bp, int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt, pass = 0;
+	int stuck = 0;
+	u8 *buff = bp;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL, and there
+	 * is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port,
+				"QSFP interface Reset for read failed\n");
+		ret = -EIO;
+		stuck = 1;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_rd(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		/* Some QSFP's fail first try. Retry as experiment */
+		if (ret && cnt == 0 && ++pass < QSFP_MAX_RETRY)
+			continue;
+		if (ret) {
+			/* qib_twsi_blk_rd() 1 for error, else 0 */
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST. LP all high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	if (stuck)
+		qib_dev_err(dd, "QSFP interface bus stuck non-idle\n");
+
+	if (pass >= QSFP_MAX_RETRY && ret)
+		qib_dev_porterr(dd, ppd->port, "QSFP failed even retrying\n");
+	else if (pass)
+		qib_dev_porterr(dd, ppd->port, "QSFP retries: %d\n", pass);
+
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * qsfp_write
+ * We do not ordinarily write the QSFP, but this is needed to select
+ * the page on non-flat QSFPs, and possibly later unusual cases
+ */
+static int qib_qsfp_write(struct qib_pportdata *ppd, int addr, void *bp,
+			  int len)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u32 out, mask;
+	int ret, cnt;
+	u8 *buff = bp;
+
+	ret = mutex_lock_interruptible(&dd->eep_lock);
+	if (ret)
+		goto no_unlock;
+
+	if (dd->twsi_eeprom_dev == QIB_TWSI_NO_DEV) {
+		ret = -ENXIO;
+		goto bail;
+	}
+
+	/*
+	 * We presume, if we are called at all, that this board has
+	 * QSFP. This is on the same i2c chain as the legacy parts,
+	 * but only responds if the module is selected via GPIO pins.
+	 * Further, there are very long setup and hold requirements
+	 * on MODSEL.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	out = QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	if (ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		out <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, out, mask, mask);
+
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL,
+	 * and there is no way to tell if it is ready, so we must wait.
+	 */
+	msleep(2);
+
+	/* Make sure TWSI bus is in sane state. */
+	ret = qib_twsi_reset(dd);
+	if (ret) {
+		qib_dev_porterr(dd, ppd->port,
+				"QSFP interface Reset for write failed\n");
+		ret = -EIO;
+		goto deselect;
+	}
+
+	/* All QSFP modules are at A0 */
+
+	cnt = 0;
+	while (cnt < len) {
+		unsigned in_page;
+		int wlen = len - cnt;
+		in_page = addr % QSFP_PAGESIZE;
+		if ((in_page + wlen) > QSFP_PAGESIZE)
+			wlen = QSFP_PAGESIZE - in_page;
+		ret = qib_twsi_blk_wr(dd, QSFP_DEV, addr, buff + cnt, wlen);
+		if (ret) {
+			/* qib_twsi_blk_wr() 1 for error, else 0 */
+			ret = -EIO;
+			goto deselect;
+		}
+		addr += wlen;
+		cnt += wlen;
+	}
+	ret = cnt;
+
+deselect:
+	/*
+	 * Module could take up to 10 uSec after transfer before
+	 * ready to respond to MOD_SEL negation, and there is no way
+	 * to tell if it is ready, so we must wait.
+	 */
+	udelay(10);
+	/* set QSFP MODSEL, RST, LP high */
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	/*
+	 * Module could take up to 2 Msec to respond to MOD_SEL
+	 * going away, and there is no way to tell if it is ready.
+	 * so we must wait.
+	 */
+	msleep(2);
+
+bail:
+	mutex_unlock(&dd->eep_lock);
+
+no_unlock:
+	return ret;
+}
+
+/*
+ * For validation, we want to check the checksums, even of the
+ * fields we do not otherwise use. This function reads the bytes from
+ * <first> to <next-1> and returns the 8lsbs of the sum, or <0 for errors
+ */
+static int qsfp_cks(struct qib_pportdata *ppd, int first, int next)
+{
+	int ret;
+	u16 cks;
+	u8 bval;
+
+	cks = 0;
+	while (first < next) {
+		ret = qsfp_read(ppd, first, &bval, 1);
+		if (ret < 0)
+			goto bail;
+		cks += bval;
+		++first;
+	}
+	ret = cks & 0xFF;
+bail:
+	return ret;
+
+}
+
+int qib_refresh_qsfp_cache(struct qib_pportdata *ppd, struct qib_qsfp_cache *cp)
+{
+	int ret;
+	int idx;
+	u16 cks;
+	u8 peek[4];
+
+	/* ensure sane contents on invalid reads, for cable swaps */
+	memset(cp, 0, sizeof(*cp));
+
+	if (!qib_qsfp_mod_present(ppd)) {
+		ret = -ENODEV;
+		goto bail;
+	}
+
+	ret = qsfp_read(ppd, 0, peek, 3);
+	if (ret < 0)
+		goto bail;
+	if ((peek[0] & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP byte0 is 0x%02X, S/B 0x0C/D\n", peek[0]);
+
+	if ((peek[2] & 2) == 0) {
+		/*
+		 * If cable is paged, rather than "flat memory", we need to
+		 * set the page to zero, Even if it already appears to be zero.
+		 */
+		u8 poke = 0;
+		ret = qib_qsfp_write(ppd, 127, &poke, 1);
+		udelay(50);
+		if (ret != 1) {
+			qib_dev_porterr(ppd->dd, ppd->port,
+					"Failed QSFP Page set\n");
+			goto bail;
+		}
+	}
+
+	ret = qsfp_read(ppd, QSFP_MOD_ID_OFFS, &cp->id, 1);
+	if (ret < 0)
+		goto bail;
+	if ((cp->id & 0xFE) != 0x0C)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP ID byte is 0x%02X, S/B 0x0C/D\n", cp->id);
+	cks = cp->id;
+
+	ret = qsfp_read(ppd, QSFP_MOD_PWR_OFFS, &cp->pwr, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->pwr;
+
+	ret = qsfp_cks(ppd, QSFP_MOD_PWR_OFFS + 1, QSFP_MOD_LEN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_MOD_LEN_OFFS, &cp->len, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->len;
+
+	ret = qsfp_read(ppd, QSFP_MOD_TECH_OFFS, &cp->tech, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->tech;
+
+	ret = qsfp_read(ppd, QSFP_VEND_OFFS, &cp->vendor, QSFP_VEND_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VEND_LEN; ++idx)
+		cks += cp->vendor[idx];
+
+	ret = qsfp_read(ppd, QSFP_IBXCV_OFFS, &cp->xt_xcv, 1);
+	if (ret < 0)
+		goto bail;
+	cks += cp->xt_xcv;
+
+	ret = qsfp_read(ppd, QSFP_VOUI_OFFS, &cp->oui, QSFP_VOUI_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_VOUI_LEN; ++idx)
+		cks += cp->oui[idx];
+
+	ret = qsfp_read(ppd, QSFP_PN_OFFS, &cp->partnum, QSFP_PN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_PN_LEN; ++idx)
+		cks += cp->partnum[idx];
+
+	ret = qsfp_read(ppd, QSFP_REV_OFFS, &cp->rev, QSFP_REV_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_REV_LEN; ++idx)
+		cks += cp->rev[idx];
+
+	ret = qsfp_read(ppd, QSFP_ATTEN_OFFS, &cp->atten, QSFP_ATTEN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_ATTEN_LEN; ++idx)
+		cks += cp->atten[idx];
+
+	ret = qsfp_cks(ppd, QSFP_ATTEN_OFFS + QSFP_ATTEN_LEN, QSFP_CC_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	cks &= 0xFF;
+	ret = qsfp_read(ppd, QSFP_CC_OFFS, &cp->cks1, 1);
+	if (ret < 0)
+		goto bail;
+	if (cks != cp->cks1)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks1 is %02X, computed %02X\n", cp->cks1,
+				cks);
+
+	/* Second checksum covers 192 to (serial, date, lot) */
+	ret = qsfp_cks(ppd, QSFP_CC_OFFS + 1, QSFP_SN_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks = ret;
+
+	ret = qsfp_read(ppd, QSFP_SN_OFFS, &cp->serial, QSFP_SN_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_SN_LEN; ++idx)
+		cks += cp->serial[idx];
+
+	ret = qsfp_read(ppd, QSFP_DATE_OFFS, &cp->date, QSFP_DATE_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_DATE_LEN; ++idx)
+		cks += cp->date[idx];
+
+	ret = qsfp_read(ppd, QSFP_LOT_OFFS, &cp->lot, QSFP_LOT_LEN);
+	if (ret < 0)
+		goto bail;
+	for (idx = 0; idx < QSFP_LOT_LEN; ++idx)
+		cks += cp->lot[idx];
+
+	ret = qsfp_cks(ppd, QSFP_LOT_OFFS + QSFP_LOT_LEN, QSFP_CC_EXT_OFFS);
+	if (ret < 0)
+		goto bail;
+	cks += ret;
+
+	ret = qsfp_read(ppd, QSFP_CC_EXT_OFFS, &cp->cks2, 1);
+	if (ret < 0)
+		goto bail;
+	cks &= 0xFF;
+	if (cks != cp->cks2)
+		qib_dev_porterr(ppd->dd, ppd->port,
+				"QSFP cks2 is %02X, computed %02X\n", cp->cks2,
+				cks);
+	return 0;
+
+bail:
+	cp->id = 0;
+	return ret;
+}
+
+const char * const qib_qsfp_devtech[16] = {
+	"850nm VCSEL", "1310nm VCSEL", "1550nm VCSEL", "1310nm FP",
+	"1310nm DFB", "1550nm DFB", "1310nm EML", "1550nm EML",
+	"Cu Misc", "1490nm DFB", "Cu NoEq", "Cu Eq",
+	"Undef", "Cu Active BothEq", "Cu FarEq", "Cu NearEq"
+};
+
+#define QSFP_DUMP_CHUNK 16 /* Holds longest string */
+#define QSFP_DEFAULT_HDR_CNT 224
+
+static const char *pwr_codes = "1.5W2.0W2.5W3.5W";
+
+int qib_qsfp_mod_present(struct qib_pportdata *ppd)
+{
+	u32 mask;
+	int ret;
+
+	mask = QSFP_GPIO_MOD_PRS_N <<
+		(ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT);
+	ret = ppd->dd->f_gpio_mod(ppd->dd, 0, 0, 0);
+
+	return !((ret & mask) >>
+		 ((ppd->hw_pidx * QSFP_GPIO_PORT2_SHIFT) + 3));
+}
+
+/*
+ * Initialize structures that control access to QSFP. Called once per port
+ * on cards that support QSFP.
+ */
+void qib_qsfp_init(struct qib_qsfp_data *qd,
+		   void (*fevent)(struct work_struct *))
+{
+	u32 mask, highs;
+
+	struct qib_devdata *dd = qd->ppd->dd;
+
+	/* Initialize work struct for later QSFP events */
+	INIT_WORK(&qd->work, fevent);
+
+	/*
+	 * Later, we may want more validation. For now, just set up pins and
+	 * blip reset. If module is present, call qib_refresh_qsfp_cache(),
+	 * to do further init.
+	 */
+	mask = QSFP_GPIO_MOD_SEL_N | QSFP_GPIO_MOD_RST_N | QSFP_GPIO_LP_MODE;
+	highs = mask - QSFP_GPIO_MOD_RST_N;
+	if (qd->ppd->hw_pidx) {
+		mask <<= QSFP_GPIO_PORT2_SHIFT;
+		highs <<= QSFP_GPIO_PORT2_SHIFT;
+	}
+	dd->f_gpio_mod(dd, highs, mask, mask);
+	udelay(20); /* Generous RST dwell */
+
+	dd->f_gpio_mod(dd, mask, mask, mask);
+	return;
+}
+
+void qib_qsfp_deinit(struct qib_qsfp_data *qd)
+{
+	/*
+	 * There is nothing to do here for now.  our work is scheduled
+	 * with queue_work(), and flush_workqueue() from remove_one
+	 * will block until all work setup with queue_work()
+	 * completes.
+	 */
+}
+
+int qib_qsfp_dump(struct qib_pportdata *ppd, char *buf, int len)
+{
+	struct qib_qsfp_cache cd;
+	u8 bin_buff[QSFP_DUMP_CHUNK];
+	char lenstr[6];
+	int sofar, ret;
+	int bidx = 0;
+
+	sofar = 0;
+	ret = qib_refresh_qsfp_cache(ppd, &cd);
+	if (ret < 0)
+		goto bail;
+
+	lenstr[0] = ' ';
+	lenstr[1] = '\0';
+	if (QSFP_IS_CU(cd.tech))
+		sprintf(lenstr, "%dM ", cd.len);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "PWR:%.3sW\n", pwr_codes +
+			   (QSFP_PWR(cd.pwr) * 4));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "TECH:%s%s\n", lenstr,
+			   qib_qsfp_devtech[cd.tech >> 4]);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Vendor:%.*s\n",
+			   QSFP_VEND_LEN, cd.vendor);
+
+	sofar += scnprintf(buf + sofar, len - sofar, "OUI:%06X\n",
+			   QSFP_OUI(cd.oui));
+
+	sofar += scnprintf(buf + sofar, len - sofar, "Part#:%.*s\n",
+			   QSFP_PN_LEN, cd.partnum);
+	sofar += scnprintf(buf + sofar, len - sofar, "Rev:%.*s\n",
+			   QSFP_REV_LEN, cd.rev);
+	if (QSFP_IS_CU(cd.tech))
+		sofar += scnprintf(buf + sofar, len - sofar, "Atten:%d, %d\n",
+				   QSFP_ATTEN_SDR(cd.atten),
+				   QSFP_ATTEN_DDR(cd.atten));
+	sofar += scnprintf(buf + sofar, len - sofar, "Serial:%.*s\n",
+			   QSFP_SN_LEN, cd.serial);
+	sofar += scnprintf(buf + sofar, len - sofar, "Date:%.*s\n",
+			   QSFP_DATE_LEN, cd.date);
+	sofar += scnprintf(buf + sofar, len - sofar, "Lot:%.*s\n",
+			   QSFP_LOT_LEN, cd.date);
+
+	while (bidx < QSFP_DEFAULT_HDR_CNT) {
+		int iidx;
+		ret = qsfp_read(ppd, bidx, bin_buff, QSFP_DUMP_CHUNK);
+		if (ret < 0)
+			goto bail;
+		for (iidx = 0; iidx < ret; ++iidx) {
+			sofar += scnprintf(buf + sofar, len-sofar, " %02X",
+				bin_buff[iidx]);
+		}
+		sofar += scnprintf(buf + sofar, len - sofar, "\n");
+		bidx += QSFP_DUMP_CHUNK;
+	}
+	ret = sofar;
+bail:
+	return ret;
+}
diff --git a/drivers/infiniband/hw/qib/qib_qsfp.h b/drivers/infiniband/hw/qib/qib_qsfp.h
new file mode 100644
index 00000000000..91908f533a2
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_qsfp.h
@@ -0,0 +1,189 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+/* QSFP support common definitions, for ib_qib driver */
+
+#define QSFP_DEV 0xA0
+#define QSFP_PWR_LAG_MSEC 2000
+#define QSFP_MODPRS_LAG_MSEC 20
+
+/*
+ * Below are masks for various QSFP signals, for Port 1.
+ * Port2 equivalents are shifted by QSFP_GPIO_PORT2_SHIFT.
+ * _N means asserted low
+ */
+#define QSFP_GPIO_MOD_SEL_N (4)
+#define QSFP_GPIO_MOD_PRS_N (8)
+#define QSFP_GPIO_INT_N (0x10)
+#define QSFP_GPIO_MOD_RST_N (0x20)
+#define QSFP_GPIO_LP_MODE (0x40)
+#define QSFP_GPIO_PORT2_SHIFT 5
+
+#define QSFP_PAGESIZE 128
+/* Defined fields that QLogic requires of qualified cables */
+/* Byte 0 is Identifier, not checked */
+/* Byte 1 is reserved "status MSB" */
+/* Byte 2 is "status LSB" We only care that D2 "Flat Mem" is set. */
+/*
+ * Rest of first 128 not used, although 127 is reserved for page select
+ * if module is not "Flat memory".
+ */
+/* Byte 128 is Identifier: must be 0x0c for QSFP, or 0x0d for QSFP+ */
+#define QSFP_MOD_ID_OFFS 128
+/*
+ * Byte 129 is "Extended Identifier". We only care about D7,D6: Power class
+ *  0:1.5W, 1:2.0W, 2:2.5W, 3:3.5W
+ */
+#define QSFP_MOD_PWR_OFFS 129
+/* Byte 130 is Connector type. Not QLogic req'd */
+/* Bytes 131..138 are Transceiver types, bit maps for various tech, none IB */
+/* Byte 139 is encoding. code 0x01 is 8b10b. Not QLogic req'd */
+/* byte 140 is nominal bit-rate, in units of 100Mbits/sec Not QLogic req'd */
+/* Byte 141 is Extended Rate Select. Not QLogic req'd */
+/* Bytes 142..145 are lengths for various fiber types. Not QLogic req'd */
+/* Byte 146 is length for Copper. Units of 1 meter */
+#define QSFP_MOD_LEN_OFFS 146
+/*
+ * Byte 147 is Device technology. D0..3 not Qlogc req'd
+ * D4..7 select from 15 choices, translated by table:
+ */
+#define QSFP_MOD_TECH_OFFS 147
+extern const char *const qib_qsfp_devtech[16];
+/* Active Equalization includes fiber, copper full EQ, and copper near Eq */
+#define QSFP_IS_ACTIVE(tech) ((0xA2FF >> ((tech) >> 4)) & 1)
+/* Active Equalization includes fiber, copper full EQ, and copper far Eq */
+#define QSFP_IS_ACTIVE_FAR(tech) ((0x32FF >> ((tech) >> 4)) & 1)
+/* Attenuation should be valid for copper other than full/near Eq */
+#define QSFP_HAS_ATTEN(tech) ((0x4D00 >> ((tech) >> 4)) & 1)
+/* Length is only valid if technology is "copper" */
+#define QSFP_IS_CU(tech) ((0xED00 >> ((tech) >> 4)) & 1)
+#define QSFP_TECH_1490 9
+
+#define QSFP_OUI(oui) (((unsigned)oui[0] << 16) | ((unsigned)oui[1] << 8) | \
+			oui[2])
+#define QSFP_OUI_AMPHENOL 0x415048
+#define QSFP_OUI_FINISAR  0x009065
+#define QSFP_OUI_GORE     0x002177
+
+/* Bytes 148..163 are Vendor Name, Left-justified Blank-filled */
+#define QSFP_VEND_OFFS 148
+#define QSFP_VEND_LEN 16
+/* Byte 164 is IB Extended tranceiver codes Bits D0..3 are SDR,DDR,QDR,EDR */
+#define QSFP_IBXCV_OFFS 164
+/* Bytes 165..167 are Vendor OUI number */
+#define QSFP_VOUI_OFFS 165
+#define QSFP_VOUI_LEN 3
+/* Bytes 168..183 are Vendor Part Number, string */
+#define QSFP_PN_OFFS 168
+#define QSFP_PN_LEN 16
+/* Bytes 184,185 are Vendor Rev. Left Justified, Blank-filled */
+#define QSFP_REV_OFFS 184
+#define QSFP_REV_LEN 2
+/*
+ * Bytes 186,187 are Wavelength, if Optical. Not Qlogic req'd
+ *  If copper, they are attenuation in dB:
+ * Byte 186 is at 2.5Gb/sec (SDR), Byte 187 at 5.0Gb/sec (DDR)
+ */
+#define QSFP_ATTEN_OFFS 186
+#define QSFP_ATTEN_LEN 2
+/* Bytes 188,189 are Wavelength tolerance, not QLogic req'd */
+/* Byte 190 is Max Case Temp. Not QLogic req'd */
+/* Byte 191 is LSB of sum of bytes 128..190. Not QLogic req'd */
+#define QSFP_CC_OFFS 191
+/* Bytes 192..195 are Options implemented in qsfp. Not Qlogic req'd */
+/* Bytes 196..211 are Serial Number, String */
+#define QSFP_SN_OFFS 196
+#define QSFP_SN_LEN 16
+/* Bytes 212..219 are date-code YYMMDD (MM==1 for Jan) */
+#define QSFP_DATE_OFFS 212
+#define QSFP_DATE_LEN 6
+/* Bytes 218,219 are optional lot-code, string */
+#define QSFP_LOT_OFFS 218
+#define QSFP_LOT_LEN 2
+/* Bytes 220, 221 indicate monitoring options, Not QLogic req'd */
+/* Byte 223 is LSB of sum of bytes 192..222 */
+#define QSFP_CC_EXT_OFFS 223
+
+/*
+ * struct qib_qsfp_data encapsulates state of QSFP device for one port.
+ * it will be part of port-chip-specific data if a board supports QSFP.
+ *
+ * Since multiple board-types use QSFP, and their pport_data structs
+ * differ (in the chip-specific section), we need a pointer to its head.
+ *
+ * Avoiding premature optimization, we will have one work_struct per port,
+ * and let the (increasingly inaccurately named) eep_lock arbitrate
+ * access to common resources.
+ *
+ */
+
+/*
+ * Hold the parts of the onboard EEPROM that we care about, so we aren't
+ * coonstantly bit-boffing
+ */
+struct qib_qsfp_cache {
+	u8 id;	/* must be 0x0C or 0x0D; 0 indicates invalid EEPROM read */
+	u8 pwr; /* in D6,7 */
+	u8 len;	/* in meters, Cu only */
+	u8 tech;
+	char vendor[QSFP_VEND_LEN];
+	u8 xt_xcv; /* Ext. tranceiver codes, 4 lsbs are IB speed supported */
+	u8 oui[QSFP_VOUI_LEN];
+	u8 partnum[QSFP_PN_LEN];
+	u8 rev[QSFP_REV_LEN];
+	u8 atten[QSFP_ATTEN_LEN];
+	u8 cks1;	/* Checksum of bytes 128..190 */
+	u8 serial[QSFP_SN_LEN];
+	u8 date[QSFP_DATE_LEN];
+	u8 lot[QSFP_LOT_LEN];
+	u8 cks2;	/* Checsum of bytes 192..222 */
+};
+
+#define QSFP_PWR(pbyte) (((pbyte) >> 6) & 3)
+#define QSFP_ATTEN_SDR(attenarray) (attenarray[0])
+#define QSFP_ATTEN_DDR(attenarray) (attenarray[1])
+
+struct qib_qsfp_data {
+	/* Helps to find our way */
+	struct qib_pportdata *ppd;
+	struct work_struct work;
+	struct qib_qsfp_cache cache;
+	unsigned long t_insert;
+	u8 modpresent;
+};
+
+extern int qib_refresh_qsfp_cache(struct qib_pportdata *ppd,
+				  struct qib_qsfp_cache *cp);
+extern int qib_qsfp_mod_present(struct qib_pportdata *ppd);
+extern void qib_qsfp_init(struct qib_qsfp_data *qd,
+			  void (*fevent)(struct work_struct *));
+extern void qib_qsfp_deinit(struct qib_qsfp_data *qd);
diff --git a/drivers/infiniband/hw/qib/qib_rc.c b/drivers/infiniband/hw/qib/qib_rc.c
new file mode 100644
index 00000000000..2f2501890c4
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_rc.c
@@ -0,0 +1,2290 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/io.h>
+
+#include "qib.h"
+
+/* cut down ridiculously long IB macro names */
+#define OP(x) IB_OPCODE_RC_##x
+
+static void rc_timeout(unsigned long arg);
+
+static u32 restart_sge(struct qib_sge_state *ss, struct qib_swqe *wqe,
+		       u32 psn, u32 pmtu)
+{
+	u32 len;
+
+	len = ((psn - wqe->psn) & QIB_PSN_MASK) * pmtu;
+	ss->sge = wqe->sg_list[0];
+	ss->sg_list = wqe->sg_list + 1;
+	ss->num_sge = wqe->wr.num_sge;
+	ss->total_len = wqe->length;
+	qib_skip_sge(ss, len, 0);
+	return wqe->length - len;
+}
+
+static void start_timer(struct qib_qp *qp)
+{
+	qp->s_flags |= QIB_S_TIMER;
+	qp->s_timer.function = rc_timeout;
+	/* 4.096 usec. * (1 << qp->timeout) */
+	qp->s_timer.expires = jiffies + qp->timeout_jiffies;
+	add_timer(&qp->s_timer);
+}
+
+/**
+ * qib_make_rc_ack - construct a response packet (ACK, NAK, or RDMA read)
+ * @dev: the device for this QP
+ * @qp: a pointer to the QP
+ * @ohdr: a pointer to the IB header being constructed
+ * @pmtu: the path MTU
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ * Note that we are in the responder's side of the QP context.
+ * Note the QP s_lock must be held.
+ */
+static int qib_make_rc_ack(struct qib_ibdev *dev, struct qib_qp *qp,
+			   struct qib_other_headers *ohdr, u32 pmtu)
+{
+	struct qib_ack_entry *e;
+	u32 hwords;
+	u32 len;
+	u32 bth0;
+	u32 bth2;
+
+	/* Don't send an ACK if we aren't supposed to. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto bail;
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+
+	switch (qp->s_ack_state) {
+	case OP(RDMA_READ_RESPONSE_LAST):
+	case OP(RDMA_READ_RESPONSE_ONLY):
+		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+		if (e->rdma_sge.mr) {
+			qib_put_mr(e->rdma_sge.mr);
+			e->rdma_sge.mr = NULL;
+		}
+		/* FALLTHROUGH */
+	case OP(ATOMIC_ACKNOWLEDGE):
+		/*
+		 * We can increment the tail pointer now that the last
+		 * response has been sent instead of only being
+		 * constructed.
+		 */
+		if (++qp->s_tail_ack_queue > QIB_MAX_RDMA_ATOMIC)
+			qp->s_tail_ack_queue = 0;
+		/* FALLTHROUGH */
+	case OP(SEND_ONLY):
+	case OP(ACKNOWLEDGE):
+		/* Check for no next entry in the queue. */
+		if (qp->r_head_ack_queue == qp->s_tail_ack_queue) {
+			if (qp->s_flags & QIB_S_ACK_PENDING)
+				goto normal;
+			goto bail;
+		}
+
+		e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST)) {
+			/*
+			 * If a RDMA read response is being resent and
+			 * we haven't seen the duplicate request yet,
+			 * then stop sending the remaining responses the
+			 * responder has seen until the requester resends it.
+			 */
+			len = e->rdma_sge.sge_length;
+			if (len && !e->rdma_sge.mr) {
+				qp->s_tail_ack_queue = qp->r_head_ack_queue;
+				goto bail;
+			}
+			/* Copy SGE state in case we need to resend */
+			qp->s_rdma_mr = e->rdma_sge.mr;
+			if (qp->s_rdma_mr)
+				qib_get_mr(qp->s_rdma_mr);
+			qp->s_ack_rdma_sge.sge = e->rdma_sge;
+			qp->s_ack_rdma_sge.num_sge = 1;
+			qp->s_cur_sge = &qp->s_ack_rdma_sge;
+			if (len > pmtu) {
+				len = pmtu;
+				qp->s_ack_state = OP(RDMA_READ_RESPONSE_FIRST);
+			} else {
+				qp->s_ack_state = OP(RDMA_READ_RESPONSE_ONLY);
+				e->sent = 1;
+			}
+			ohdr->u.aeth = qib_compute_aeth(qp);
+			hwords++;
+			qp->s_ack_rdma_psn = e->psn;
+			bth2 = qp->s_ack_rdma_psn++ & QIB_PSN_MASK;
+		} else {
+			/* COMPARE_SWAP or FETCH_ADD */
+			qp->s_cur_sge = NULL;
+			len = 0;
+			qp->s_ack_state = OP(ATOMIC_ACKNOWLEDGE);
+			ohdr->u.at.aeth = qib_compute_aeth(qp);
+			ohdr->u.at.atomic_ack_eth[0] =
+				cpu_to_be32(e->atomic_data >> 32);
+			ohdr->u.at.atomic_ack_eth[1] =
+				cpu_to_be32(e->atomic_data);
+			hwords += sizeof(ohdr->u.at) / sizeof(u32);
+			bth2 = e->psn & QIB_PSN_MASK;
+			e->sent = 1;
+		}
+		bth0 = qp->s_ack_state << 24;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		qp->s_ack_state = OP(RDMA_READ_RESPONSE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		qp->s_cur_sge = &qp->s_ack_rdma_sge;
+		qp->s_rdma_mr = qp->s_ack_rdma_sge.sge.mr;
+		if (qp->s_rdma_mr)
+			qib_get_mr(qp->s_rdma_mr);
+		len = qp->s_ack_rdma_sge.sge.sge_length;
+		if (len > pmtu)
+			len = pmtu;
+		else {
+			ohdr->u.aeth = qib_compute_aeth(qp);
+			hwords++;
+			qp->s_ack_state = OP(RDMA_READ_RESPONSE_LAST);
+			e = &qp->s_ack_queue[qp->s_tail_ack_queue];
+			e->sent = 1;
+		}
+		bth0 = qp->s_ack_state << 24;
+		bth2 = qp->s_ack_rdma_psn++ & QIB_PSN_MASK;
+		break;
+
+	default:
+normal:
+		/*
+		 * Send a regular ACK.
+		 * Set the s_ack_state so we wait until after sending
+		 * the ACK before setting s_ack_state to ACKNOWLEDGE
+		 * (see above).
+		 */
+		qp->s_ack_state = OP(SEND_ONLY);
+		qp->s_flags &= ~QIB_S_ACK_PENDING;
+		qp->s_cur_sge = NULL;
+		if (qp->s_nak_state)
+			ohdr->u.aeth =
+				cpu_to_be32((qp->r_msn & QIB_MSN_MASK) |
+					    (qp->s_nak_state <<
+					     QIB_AETH_CREDIT_SHIFT));
+		else
+			ohdr->u.aeth = qib_compute_aeth(qp);
+		hwords++;
+		len = 0;
+		bth0 = OP(ACKNOWLEDGE) << 24;
+		bth2 = qp->s_ack_psn & QIB_PSN_MASK;
+	}
+	qp->s_rdma_ack_cnt++;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0, bth2);
+	return 1;
+
+bail:
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+	qp->s_flags &= ~(QIB_S_RESP_PENDING | QIB_S_ACK_PENDING);
+	return 0;
+}
+
+/**
+ * qib_make_rc_req - construct a request packet (SEND, RDMA r/w, ATOMIC)
+ * @qp: a pointer to the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_rc_req(struct qib_qp *qp)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_other_headers *ohdr;
+	struct qib_sge_state *ss;
+	struct qib_swqe *wqe;
+	u32 hwords;
+	u32 len;
+	u32 bth0;
+	u32 bth2;
+	u32 pmtu = qp->pmtu;
+	char newreq;
+	unsigned long flags;
+	int ret = 0;
+	int delta;
+
+	ohdr = &qp->s_hdr->u.oth;
+	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+		ohdr = &qp->s_hdr->u.l.oth;
+
+	/*
+	 * The lock is needed to synchronize between the sending tasklet,
+	 * the receive interrupt handler, and timeout resends.
+	 */
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Sending responses has higher priority over sending requests. */
+	if ((qp->s_flags & QIB_S_RESP_PENDING) &&
+	    qib_make_rc_ack(dev, qp, ohdr, pmtu))
+		goto done;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, qp->s_last != qp->s_acked ?
+			IB_WC_SUCCESS : IB_WC_WR_FLUSH_ERR);
+		/* will get called again */
+		goto done;
+	}
+
+	if (qp->s_flags & (QIB_S_WAIT_RNR | QIB_S_WAIT_ACK))
+		goto bail;
+
+	if (qib_cmp24(qp->s_psn, qp->s_sending_hpsn) <= 0) {
+		if (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0) {
+			qp->s_flags |= QIB_S_WAIT_PSN;
+			goto bail;
+		}
+		qp->s_sending_psn = qp->s_psn;
+		qp->s_sending_hpsn = qp->s_psn - 1;
+	}
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+	bth0 = 0;
+
+	/* Send a request. */
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	switch (qp->s_state) {
+	default:
+		if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK))
+			goto bail;
+		/*
+		 * Resend an old request or start a new one.
+		 *
+		 * We keep track of the current SWQE so that
+		 * we don't reset the "furthest progress" state
+		 * if we need to back up.
+		 */
+		newreq = 0;
+		if (qp->s_cur == qp->s_tail) {
+			/* Check if send work queue is empty. */
+			if (qp->s_tail == qp->s_head)
+				goto bail;
+			/*
+			 * If a fence is requested, wait for previous
+			 * RDMA read and atomic operations to finish.
+			 */
+			if ((wqe->wr.send_flags & IB_SEND_FENCE) &&
+			    qp->s_num_rd_atomic) {
+				qp->s_flags |= QIB_S_WAIT_FENCE;
+				goto bail;
+			}
+			wqe->psn = qp->s_next_psn;
+			newreq = 1;
+		}
+		/*
+		 * Note that we have to be careful not to modify the
+		 * original work request since we may need to resend
+		 * it.
+		 */
+		len = wqe->length;
+		ss = &qp->s_sge;
+		bth2 = qp->s_psn & QIB_PSN_MASK;
+		switch (wqe->wr.opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_IMM:
+			/* If no credit, return. */
+			if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT) &&
+			    qib_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= QIB_S_WAIT_SSN_CREDIT;
+				goto bail;
+			}
+			wqe->lpsn = wqe->psn;
+			if (len > pmtu) {
+				wqe->lpsn += (len - 1) / pmtu;
+				qp->s_state = OP(SEND_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_SEND)
+				qp->s_state = OP(SEND_ONLY);
+			else {
+				qp->s_state = OP(SEND_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the BTH */
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+			}
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_WRITE:
+			if (newreq && !(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+				qp->s_lsn++;
+			/* FALLTHROUGH */
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			/* If no credit, return. */
+			if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT) &&
+			    qib_cmp24(wqe->ssn, qp->s_lsn + 1) > 0) {
+				qp->s_flags |= QIB_S_WAIT_SSN_CREDIT;
+				goto bail;
+			}
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			hwords += sizeof(struct ib_reth) / sizeof(u32);
+			wqe->lpsn = wqe->psn;
+			if (len > pmtu) {
+				wqe->lpsn += (len - 1) / pmtu;
+				qp->s_state = OP(RDMA_WRITE_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+				qp->s_state = OP(RDMA_WRITE_ONLY);
+			else {
+				qp->s_state =
+					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after RETH */
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+					bth0 |= IB_BTH_SOLICITED;
+			}
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_READ:
+			/*
+			 * Don't allow more operations to be started
+			 * than the QP limits allow.
+			 */
+			if (newreq) {
+				if (qp->s_num_rd_atomic >=
+				    qp->s_max_rd_atomic) {
+					qp->s_flags |= QIB_S_WAIT_RDMAR;
+					goto bail;
+				}
+				qp->s_num_rd_atomic++;
+				if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+					qp->s_lsn++;
+				/*
+				 * Adjust s_next_psn to count the
+				 * expected number of responses.
+				 */
+				if (len > pmtu)
+					qp->s_next_psn += (len - 1) / pmtu;
+				wqe->lpsn = qp->s_next_psn++;
+			}
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			qp->s_state = OP(RDMA_READ_REQUEST);
+			hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
+			ss = NULL;
+			len = 0;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_ATOMIC_CMP_AND_SWP:
+		case IB_WR_ATOMIC_FETCH_AND_ADD:
+			/*
+			 * Don't allow more operations to be started
+			 * than the QP limits allow.
+			 */
+			if (newreq) {
+				if (qp->s_num_rd_atomic >=
+				    qp->s_max_rd_atomic) {
+					qp->s_flags |= QIB_S_WAIT_RDMAR;
+					goto bail;
+				}
+				qp->s_num_rd_atomic++;
+				if (!(qp->s_flags & QIB_S_UNLIMITED_CREDIT))
+					qp->s_lsn++;
+				wqe->lpsn = wqe->psn;
+			}
+			if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP) {
+				qp->s_state = OP(COMPARE_SWAP);
+				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
+					wqe->wr.wr.atomic.swap);
+				ohdr->u.atomic_eth.compare_data = cpu_to_be64(
+					wqe->wr.wr.atomic.compare_add);
+			} else {
+				qp->s_state = OP(FETCH_ADD);
+				ohdr->u.atomic_eth.swap_data = cpu_to_be64(
+					wqe->wr.wr.atomic.compare_add);
+				ohdr->u.atomic_eth.compare_data = 0;
+			}
+			ohdr->u.atomic_eth.vaddr[0] = cpu_to_be32(
+				wqe->wr.wr.atomic.remote_addr >> 32);
+			ohdr->u.atomic_eth.vaddr[1] = cpu_to_be32(
+				wqe->wr.wr.atomic.remote_addr);
+			ohdr->u.atomic_eth.rkey = cpu_to_be32(
+				wqe->wr.wr.atomic.rkey);
+			hwords += sizeof(struct ib_atomic_eth) / sizeof(u32);
+			ss = NULL;
+			len = 0;
+			bth2 |= IB_BTH_REQ_ACK;
+			if (++qp->s_cur == qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		default:
+			goto bail;
+		}
+		qp->s_sge.sge = wqe->sg_list[0];
+		qp->s_sge.sg_list = wqe->sg_list + 1;
+		qp->s_sge.num_sge = wqe->wr.num_sge;
+		qp->s_sge.total_len = wqe->length;
+		qp->s_len = wqe->length;
+		if (newreq) {
+			qp->s_tail++;
+			if (qp->s_tail >= qp->s_size)
+				qp->s_tail = 0;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_READ)
+			qp->s_psn = wqe->lpsn + 1;
+		else {
+			qp->s_psn++;
+			if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+				qp->s_next_psn = qp->s_psn;
+		}
+		break;
+
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_FIRST is used by the ACK processing
+		 * thread to indicate a SEND needs to be restarted from an
+		 * earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
+		/* FALLTHROUGH */
+	case OP(SEND_FIRST):
+		qp->s_state = OP(SEND_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		bth2 = qp->s_psn++ & QIB_PSN_MASK;
+		if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+			qp->s_next_psn = qp->s_psn;
+		ss = &qp->s_sge;
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_SEND)
+			qp->s_state = OP(SEND_LAST);
+		else {
+			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+		}
+		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+			bth0 |= IB_BTH_SOLICITED;
+		bth2 |= IB_BTH_REQ_ACK;
+		qp->s_cur++;
+		if (qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_LAST):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_LAST is used by the ACK processing
+		 * thread to indicate a RDMA write needs to be restarted from
+		 * an earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		qp->s_len = restart_sge(&qp->s_sge, wqe, qp->s_psn, pmtu);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_FIRST):
+		qp->s_state = OP(RDMA_WRITE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		bth2 = qp->s_psn++ & QIB_PSN_MASK;
+		if (qib_cmp24(qp->s_psn, qp->s_next_psn) > 0)
+			qp->s_next_psn = qp->s_psn;
+		ss = &qp->s_sge;
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			qp->s_state = OP(RDMA_WRITE_LAST);
+		else {
+			qp->s_state = OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+		}
+		bth2 |= IB_BTH_REQ_ACK;
+		qp->s_cur++;
+		if (qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		/*
+		 * qp->s_state is normally set to the opcode of the
+		 * last packet constructed for new requests and therefore
+		 * is never set to RDMA read response.
+		 * RDMA_READ_RESPONSE_MIDDLE is used by the ACK processing
+		 * thread to indicate a RDMA read needs to be restarted from
+		 * an earlier PSN without interferring with the sending thread.
+		 * See qib_restart_rc().
+		 */
+		len = ((qp->s_psn - wqe->psn) & QIB_PSN_MASK) * pmtu;
+		ohdr->u.rc.reth.vaddr =
+			cpu_to_be64(wqe->wr.wr.rdma.remote_addr + len);
+		ohdr->u.rc.reth.rkey =
+			cpu_to_be32(wqe->wr.wr.rdma.rkey);
+		ohdr->u.rc.reth.length = cpu_to_be32(wqe->length - len);
+		qp->s_state = OP(RDMA_READ_REQUEST);
+		hwords += sizeof(ohdr->u.rc.reth) / sizeof(u32);
+		bth2 = (qp->s_psn & QIB_PSN_MASK) | IB_BTH_REQ_ACK;
+		qp->s_psn = wqe->lpsn + 1;
+		ss = NULL;
+		len = 0;
+		qp->s_cur++;
+		if (qp->s_cur == qp->s_size)
+			qp->s_cur = 0;
+		break;
+	}
+	qp->s_sending_hpsn = bth2;
+	delta = (((int) bth2 - (int) wqe->psn) << 8) >> 8;
+	if (delta && delta % QIB_PSN_CREDIT == 0)
+		bth2 |= IB_BTH_REQ_ACK;
+	if (qp->s_flags & QIB_S_SEND_ONE) {
+		qp->s_flags &= ~QIB_S_SEND_ONE;
+		qp->s_flags |= QIB_S_WAIT_ACK;
+		bth2 |= IB_BTH_REQ_ACK;
+	}
+	qp->s_len -= len;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_sge = ss;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24), bth2);
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_send_rc_ack - Construct an ACK packet and send it
+ * @qp: a pointer to the QP
+ *
+ * This is called from qib_rc_rcv() and qib_kreceive().
+ * Note that RDMA reads and atomics are handled in the
+ * send side QP state and tasklet.
+ */
+void qib_send_rc_ack(struct qib_qp *qp)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	u64 pbc;
+	u16 lrh0;
+	u32 bth0;
+	u32 hwords;
+	u32 pbufn;
+	u32 __iomem *piobuf;
+	struct qib_ib_header hdr;
+	struct qib_other_headers *ohdr;
+	u32 control;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto unlock;
+
+	/* Don't send ACK or NAK if a RDMA read or atomic is pending. */
+	if ((qp->s_flags & QIB_S_RESP_PENDING) || qp->s_rdma_ack_cnt)
+		goto queue_ack;
+
+	/* Construct the header with s_lock held so APM doesn't change it. */
+	ohdr = &hdr.u.oth;
+	lrh0 = QIB_LRH_BTH;
+	/* header size in 32-bit words LRH+BTH+AETH = (8+12+4)/4. */
+	hwords = 6;
+	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+		hwords += qib_make_grh(ibp, &hdr.u.l.grh,
+				       &qp->remote_ah_attr.grh, hwords, 0);
+		ohdr = &hdr.u.l.oth;
+		lrh0 = QIB_LRH_GRH;
+	}
+	/* read pkey_index w/o lock (its atomic) */
+	bth0 = qib_get_pkey(ibp, qp->s_pkey_index) | (OP(ACKNOWLEDGE) << 24);
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	if (qp->r_nak_state)
+		ohdr->u.aeth = cpu_to_be32((qp->r_msn & QIB_MSN_MASK) |
+					    (qp->r_nak_state <<
+					     QIB_AETH_CREDIT_SHIFT));
+	else
+		ohdr->u.aeth = qib_compute_aeth(qp);
+	lrh0 |= ibp->sl_to_vl[qp->remote_ah_attr.sl] << 12 |
+		qp->remote_ah_attr.sl << 4;
+	hdr.lrh[0] = cpu_to_be16(lrh0);
+	hdr.lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+	hdr.lrh[2] = cpu_to_be16(hwords + SIZE_OF_CRC);
+	hdr.lrh[3] = cpu_to_be16(ppd->lid | qp->remote_ah_attr.src_path_bits);
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(qp->r_ack_psn & QIB_PSN_MASK);
+
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	/* Don't try to send ACKs if the link isn't ACTIVE */
+	if (!(ppd->lflags & QIBL_LINKACTIVE))
+		goto done;
+
+	control = dd->f_setpbc_control(ppd, hwords + SIZE_OF_CRC,
+				       qp->s_srate, lrh0 >> 12);
+	/* length is + 1 for the control dword */
+	pbc = ((u64) control << 32) | (hwords + 1);
+
+	piobuf = dd->f_getsendbuf(ppd, pbc, &pbufn);
+	if (!piobuf) {
+		/*
+		 * We are out of PIO buffers at the moment.
+		 * Pass responsibility for sending the ACK to the
+		 * send tasklet so that when a PIO buffer becomes
+		 * available, the ACK is sent ahead of other outgoing
+		 * packets.
+		 */
+		spin_lock_irqsave(&qp->s_lock, flags);
+		goto queue_ack;
+	}
+
+	/*
+	 * Write the pbc.
+	 * We have to flush after the PBC for correctness
+	 * on some cpus or WC buffer can be written out of order.
+	 */
+	writeq(pbc, piobuf);
+
+	if (dd->flags & QIB_PIO_FLUSH_WC) {
+		u32 *hdrp = (u32 *) &hdr;
+
+		qib_flush_wc();
+		qib_pio_copy(piobuf + 2, hdrp, hwords - 1);
+		qib_flush_wc();
+		__raw_writel(hdrp[hwords - 1], piobuf + hwords + 1);
+	} else
+		qib_pio_copy(piobuf + 2, (u32 *) &hdr, hwords);
+
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf + spcl_off);
+	}
+
+	qib_flush_wc();
+	qib_sendbuf_done(dd, pbufn);
+
+	this_cpu_inc(ibp->pmastats->n_unicast_xmit);
+	goto done;
+
+queue_ack:
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		ibp->n_rc_qacks++;
+		qp->s_flags |= QIB_S_ACK_PENDING | QIB_S_RESP_PENDING;
+		qp->s_nak_state = qp->r_nak_state;
+		qp->s_ack_psn = qp->r_ack_psn;
+
+		/* Schedule the send tasklet. */
+		qib_schedule_send(qp);
+	}
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+done:
+	return;
+}
+
+/**
+ * reset_psn - reset the QP state to send starting from PSN
+ * @qp: the QP
+ * @psn: the packet sequence number to restart at
+ *
+ * This is called from qib_rc_rcv() to process an incoming RC ACK
+ * for the given QP.
+ * Called at interrupt level with the QP s_lock held.
+ */
+static void reset_psn(struct qib_qp *qp, u32 psn)
+{
+	u32 n = qp->s_acked;
+	struct qib_swqe *wqe = get_swqe_ptr(qp, n);
+	u32 opcode;
+
+	qp->s_cur = n;
+
+	/*
+	 * If we are starting the request from the beginning,
+	 * let the normal send code handle initialization.
+	 */
+	if (qib_cmp24(psn, wqe->psn) <= 0) {
+		qp->s_state = OP(SEND_LAST);
+		goto done;
+	}
+
+	/* Find the work request opcode corresponding to the given PSN. */
+	opcode = wqe->wr.opcode;
+	for (;;) {
+		int diff;
+
+		if (++n == qp->s_size)
+			n = 0;
+		if (n == qp->s_tail)
+			break;
+		wqe = get_swqe_ptr(qp, n);
+		diff = qib_cmp24(psn, wqe->psn);
+		if (diff < 0)
+			break;
+		qp->s_cur = n;
+		/*
+		 * If we are starting the request from the beginning,
+		 * let the normal send code handle initialization.
+		 */
+		if (diff == 0) {
+			qp->s_state = OP(SEND_LAST);
+			goto done;
+		}
+		opcode = wqe->wr.opcode;
+	}
+
+	/*
+	 * Set the state to restart in the middle of a request.
+	 * Don't change the s_sge, s_cur_sge, or s_cur_size.
+	 * See qib_make_rc_req().
+	 */
+	switch (opcode) {
+	case IB_WR_SEND:
+	case IB_WR_SEND_WITH_IMM:
+		qp->s_state = OP(RDMA_READ_RESPONSE_FIRST);
+		break;
+
+	case IB_WR_RDMA_WRITE:
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		qp->s_state = OP(RDMA_READ_RESPONSE_LAST);
+		break;
+
+	case IB_WR_RDMA_READ:
+		qp->s_state = OP(RDMA_READ_RESPONSE_MIDDLE);
+		break;
+
+	default:
+		/*
+		 * This case shouldn't happen since its only
+		 * one PSN per req.
+		 */
+		qp->s_state = OP(SEND_LAST);
+	}
+done:
+	qp->s_psn = psn;
+	/*
+	 * Set QIB_S_WAIT_PSN as qib_rc_complete() may start the timer
+	 * asynchronously before the send tasklet can get scheduled.
+	 * Doing it in qib_make_rc_req() is too late.
+	 */
+	if ((qib_cmp24(qp->s_psn, qp->s_sending_hpsn) <= 0) &&
+	    (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0))
+		qp->s_flags |= QIB_S_WAIT_PSN;
+}
+
+/*
+ * Back up requester to resend the last un-ACKed request.
+ * The QP r_lock and s_lock should be held and interrupts disabled.
+ */
+static void qib_restart_rc(struct qib_qp *qp, u32 psn, int wait)
+{
+	struct qib_swqe *wqe = get_swqe_ptr(qp, qp->s_acked);
+	struct qib_ibport *ibp;
+
+	if (qp->s_retry == 0) {
+		if (qp->s_mig_state == IB_MIG_ARMED) {
+			qib_migrate_qp(qp);
+			qp->s_retry = qp->s_retry_cnt;
+		} else if (qp->s_last == qp->s_acked) {
+			qib_send_complete(qp, wqe, IB_WC_RETRY_EXC_ERR);
+			qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+			return;
+		} else /* XXX need to handle delayed completion */
+			return;
+	} else
+		qp->s_retry--;
+
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+	if (wqe->wr.opcode == IB_WR_RDMA_READ)
+		ibp->n_rc_resends++;
+	else
+		ibp->n_rc_resends += (qp->s_psn - psn) & QIB_PSN_MASK;
+
+	qp->s_flags &= ~(QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR |
+			 QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_PSN |
+			 QIB_S_WAIT_ACK);
+	if (wait)
+		qp->s_flags |= QIB_S_SEND_ONE;
+	reset_psn(qp, psn);
+}
+
+/*
+ * This is called from s_timer for missing responses.
+ */
+static void rc_timeout(unsigned long arg)
+{
+	struct qib_qp *qp = (struct qib_qp *)arg;
+	struct qib_ibport *ibp;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->r_lock, flags);
+	spin_lock(&qp->s_lock);
+	if (qp->s_flags & QIB_S_TIMER) {
+		ibp = to_iport(qp->ibqp.device, qp->port_num);
+		ibp->n_rc_timeouts++;
+		qp->s_flags &= ~QIB_S_TIMER;
+		del_timer(&qp->s_timer);
+		qib_restart_rc(qp, qp->s_last_psn + 1, 1);
+		qib_schedule_send(qp);
+	}
+	spin_unlock(&qp->s_lock);
+	spin_unlock_irqrestore(&qp->r_lock, flags);
+}
+
+/*
+ * This is called from s_timer for RNR timeouts.
+ */
+void qib_rc_rnr_retry(unsigned long arg)
+{
+	struct qib_qp *qp = (struct qib_qp *)arg;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (qp->s_flags & QIB_S_WAIT_RNR) {
+		qp->s_flags &= ~QIB_S_WAIT_RNR;
+		del_timer(&qp->s_timer);
+		qib_schedule_send(qp);
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
+
+/*
+ * Set qp->s_sending_psn to the next PSN after the given one.
+ * This would be psn+1 except when RDMA reads are present.
+ */
+static void reset_sending_psn(struct qib_qp *qp, u32 psn)
+{
+	struct qib_swqe *wqe;
+	u32 n = qp->s_last;
+
+	/* Find the work request corresponding to the given PSN. */
+	for (;;) {
+		wqe = get_swqe_ptr(qp, n);
+		if (qib_cmp24(psn, wqe->lpsn) <= 0) {
+			if (wqe->wr.opcode == IB_WR_RDMA_READ)
+				qp->s_sending_psn = wqe->lpsn + 1;
+			else
+				qp->s_sending_psn = psn + 1;
+			break;
+		}
+		if (++n == qp->s_size)
+			n = 0;
+		if (n == qp->s_tail)
+			break;
+	}
+}
+
+/*
+ * This should be called with the QP s_lock held and interrupts disabled.
+ */
+void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr)
+{
+	struct qib_other_headers *ohdr;
+	struct qib_swqe *wqe;
+	struct ib_wc wc;
+	unsigned i;
+	u32 opcode;
+	u32 psn;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		return;
+
+	/* Find out where the BTH is */
+	if ((be16_to_cpu(hdr->lrh[0]) & 3) == QIB_LRH_BTH)
+		ohdr = &hdr->u.oth;
+	else
+		ohdr = &hdr->u.l.oth;
+
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
+	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
+		WARN_ON(!qp->s_rdma_ack_cnt);
+		qp->s_rdma_ack_cnt--;
+		return;
+	}
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	reset_sending_psn(qp, psn);
+
+	/*
+	 * Start timer after a packet requesting an ACK has been sent and
+	 * there are still requests that haven't been acked.
+	 */
+	if ((psn & IB_BTH_REQ_ACK) && qp->s_acked != qp->s_tail &&
+	    !(qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR | QIB_S_WAIT_PSN)) &&
+	    (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		start_timer(qp);
+
+	while (qp->s_last != qp->s_acked) {
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) >= 0 &&
+		    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)
+			break;
+		for (i = 0; i < wqe->wr.num_sge; i++) {
+			struct qib_sge *sge = &wqe->sg_list[i];
+
+			qib_put_mr(sge->mr);
+		}
+		/* Post a send completion queue entry if requested. */
+		if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
+			memset(&wc, 0, sizeof wc);
+			wc.wr_id = wqe->wr.wr_id;
+			wc.status = IB_WC_SUCCESS;
+			wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+			wc.byte_len = wqe->length;
+			wc.qp = &qp->ibqp;
+			qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
+		}
+		if (++qp->s_last >= qp->s_size)
+			qp->s_last = 0;
+	}
+	/*
+	 * If we were waiting for sends to complete before resending,
+	 * and they are now complete, restart sending.
+	 */
+	if (qp->s_flags & QIB_S_WAIT_PSN &&
+	    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
+		qp->s_flags &= ~QIB_S_WAIT_PSN;
+		qp->s_sending_psn = qp->s_psn;
+		qp->s_sending_hpsn = qp->s_psn - 1;
+		qib_schedule_send(qp);
+	}
+}
+
+static inline void update_last_psn(struct qib_qp *qp, u32 psn)
+{
+	qp->s_last_psn = psn;
+}
+
+/*
+ * Generate a SWQE completion.
+ * This is similar to qib_send_complete but has to check to be sure
+ * that the SGEs are not being referenced if the SWQE is being resent.
+ */
+static struct qib_swqe *do_rc_completion(struct qib_qp *qp,
+					 struct qib_swqe *wqe,
+					 struct qib_ibport *ibp)
+{
+	struct ib_wc wc;
+	unsigned i;
+
+	/*
+	 * Don't decrement refcount and don't generate a
+	 * completion if the SWQE is being resent until the send
+	 * is finished.
+	 */
+	if (qib_cmp24(wqe->lpsn, qp->s_sending_psn) < 0 ||
+	    qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) > 0) {
+		for (i = 0; i < wqe->wr.num_sge; i++) {
+			struct qib_sge *sge = &wqe->sg_list[i];
+
+			qib_put_mr(sge->mr);
+		}
+		/* Post a send completion queue entry if requested. */
+		if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+		    (wqe->wr.send_flags & IB_SEND_SIGNALED)) {
+			memset(&wc, 0, sizeof wc);
+			wc.wr_id = wqe->wr.wr_id;
+			wc.status = IB_WC_SUCCESS;
+			wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+			wc.byte_len = wqe->length;
+			wc.qp = &qp->ibqp;
+			qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc, 0);
+		}
+		if (++qp->s_last >= qp->s_size)
+			qp->s_last = 0;
+	} else
+		ibp->n_rc_delayed_comp++;
+
+	qp->s_retry = qp->s_retry_cnt;
+	update_last_psn(qp, wqe->lpsn);
+
+	/*
+	 * If we are completing a request which is in the process of
+	 * being resent, we can stop resending it since we know the
+	 * responder has already seen it.
+	 */
+	if (qp->s_acked == qp->s_cur) {
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		qp->s_acked = qp->s_cur;
+		wqe = get_swqe_ptr(qp, qp->s_cur);
+		if (qp->s_acked != qp->s_tail) {
+			qp->s_state = OP(SEND_LAST);
+			qp->s_psn = wqe->psn;
+		}
+	} else {
+		if (++qp->s_acked >= qp->s_size)
+			qp->s_acked = 0;
+		if (qp->state == IB_QPS_SQD && qp->s_acked == qp->s_cur)
+			qp->s_draining = 0;
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+	}
+	return wqe;
+}
+
+/**
+ * do_rc_ack - process an incoming RC ACK
+ * @qp: the QP the ACK came in on
+ * @psn: the packet sequence number of the ACK
+ * @opcode: the opcode of the request that resulted in the ACK
+ *
+ * This is called from qib_rc_rcv_resp() to process an incoming RC ACK
+ * for the given QP.
+ * Called at interrupt level with the QP s_lock held.
+ * Returns 1 if OK, 0 if current operation should be aborted (NAK).
+ */
+static int do_rc_ack(struct qib_qp *qp, u32 aeth, u32 psn, int opcode,
+		     u64 val, struct qib_ctxtdata *rcd)
+{
+	struct qib_ibport *ibp;
+	enum ib_wc_status status;
+	struct qib_swqe *wqe;
+	int ret = 0;
+	u32 ack_psn;
+	int diff;
+
+	/* Remove QP from retry timer */
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+
+	/*
+	 * Note that NAKs implicitly ACK outstanding SEND and RDMA write
+	 * requests and implicitly NAK RDMA read and atomic requests issued
+	 * before the NAK'ed request.  The MSN won't include the NAK'ed
+	 * request but will include an ACK'ed request(s).
+	 */
+	ack_psn = psn;
+	if (aeth >> 29)
+		ack_psn--;
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+
+	/*
+	 * The MSN might be for a later WQE than the PSN indicates so
+	 * only complete WQEs that the PSN finishes.
+	 */
+	while ((diff = qib_cmp24(ack_psn, wqe->lpsn)) >= 0) {
+		/*
+		 * RDMA_READ_RESPONSE_ONLY is a special case since
+		 * we want to generate completion events for everything
+		 * before the RDMA read, copy the data, then generate
+		 * the completion for the read.
+		 */
+		if (wqe->wr.opcode == IB_WR_RDMA_READ &&
+		    opcode == OP(RDMA_READ_RESPONSE_ONLY) &&
+		    diff == 0) {
+			ret = 1;
+			goto bail;
+		}
+		/*
+		 * If this request is a RDMA read or atomic, and the ACK is
+		 * for a later operation, this ACK NAKs the RDMA read or
+		 * atomic.  In other words, only a RDMA_READ_LAST or ONLY
+		 * can ACK a RDMA read and likewise for atomic ops.  Note
+		 * that the NAK case can only happen if relaxed ordering is
+		 * used and requests are sent after an RDMA read or atomic
+		 * is sent but before the response is received.
+		 */
+		if ((wqe->wr.opcode == IB_WR_RDMA_READ &&
+		     (opcode != OP(RDMA_READ_RESPONSE_LAST) || diff != 0)) ||
+		    ((wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		      wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) &&
+		     (opcode != OP(ATOMIC_ACKNOWLEDGE) || diff != 0))) {
+			/* Retry this request. */
+			if (!(qp->r_flags & QIB_R_RDMAR_SEQ)) {
+				qp->r_flags |= QIB_R_RDMAR_SEQ;
+				qib_restart_rc(qp, qp->s_last_psn + 1, 0);
+				if (list_empty(&qp->rspwait)) {
+					qp->r_flags |= QIB_R_RSP_SEND;
+					atomic_inc(&qp->refcount);
+					list_add_tail(&qp->rspwait,
+						      &rcd->qp_wait_list);
+				}
+			}
+			/*
+			 * No need to process the ACK/NAK since we are
+			 * restarting an earlier request.
+			 */
+			goto bail;
+		}
+		if (wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) {
+			u64 *vaddr = wqe->sg_list[0].vaddr;
+			*vaddr = val;
+		}
+		if (qp->s_num_rd_atomic &&
+		    (wqe->wr.opcode == IB_WR_RDMA_READ ||
+		     wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		     wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)) {
+			qp->s_num_rd_atomic--;
+			/* Restart sending task if fence is complete */
+			if ((qp->s_flags & QIB_S_WAIT_FENCE) &&
+			    !qp->s_num_rd_atomic) {
+				qp->s_flags &= ~(QIB_S_WAIT_FENCE |
+						 QIB_S_WAIT_ACK);
+				qib_schedule_send(qp);
+			} else if (qp->s_flags & QIB_S_WAIT_RDMAR) {
+				qp->s_flags &= ~(QIB_S_WAIT_RDMAR |
+						 QIB_S_WAIT_ACK);
+				qib_schedule_send(qp);
+			}
+		}
+		wqe = do_rc_completion(qp, wqe, ibp);
+		if (qp->s_acked == qp->s_tail)
+			break;
+	}
+
+	switch (aeth >> 29) {
+	case 0:         /* ACK */
+		ibp->n_rc_acks++;
+		if (qp->s_acked != qp->s_tail) {
+			/*
+			 * We are expecting more ACKs so
+			 * reset the retransmit timer.
+			 */
+			start_timer(qp);
+			/*
+			 * We can stop resending the earlier packets and
+			 * continue with the next packet the receiver wants.
+			 */
+			if (qib_cmp24(qp->s_psn, psn) <= 0)
+				reset_psn(qp, psn + 1);
+		} else if (qib_cmp24(qp->s_psn, psn) <= 0) {
+			qp->s_state = OP(SEND_LAST);
+			qp->s_psn = psn + 1;
+		}
+		if (qp->s_flags & QIB_S_WAIT_ACK) {
+			qp->s_flags &= ~QIB_S_WAIT_ACK;
+			qib_schedule_send(qp);
+		}
+		qib_get_credit(qp, aeth);
+		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
+		qp->s_retry = qp->s_retry_cnt;
+		update_last_psn(qp, psn);
+		ret = 1;
+		goto bail;
+
+	case 1:         /* RNR NAK */
+		ibp->n_rnr_naks++;
+		if (qp->s_acked == qp->s_tail)
+			goto bail;
+		if (qp->s_flags & QIB_S_WAIT_RNR)
+			goto bail;
+		if (qp->s_rnr_retry == 0) {
+			status = IB_WC_RNR_RETRY_EXC_ERR;
+			goto class_b;
+		}
+		if (qp->s_rnr_retry_cnt < 7)
+			qp->s_rnr_retry--;
+
+		/* The last valid PSN is the previous PSN. */
+		update_last_psn(qp, psn - 1);
+
+		ibp->n_rc_resends += (qp->s_psn - psn) & QIB_PSN_MASK;
+
+		reset_psn(qp, psn);
+
+		qp->s_flags &= ~(QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_ACK);
+		qp->s_flags |= QIB_S_WAIT_RNR;
+		qp->s_timer.function = qib_rc_rnr_retry;
+		qp->s_timer.expires = jiffies + usecs_to_jiffies(
+			ib_qib_rnr_table[(aeth >> QIB_AETH_CREDIT_SHIFT) &
+					   QIB_AETH_CREDIT_MASK]);
+		add_timer(&qp->s_timer);
+		goto bail;
+
+	case 3:         /* NAK */
+		if (qp->s_acked == qp->s_tail)
+			goto bail;
+		/* The last valid PSN is the previous PSN. */
+		update_last_psn(qp, psn - 1);
+		switch ((aeth >> QIB_AETH_CREDIT_SHIFT) &
+			QIB_AETH_CREDIT_MASK) {
+		case 0: /* PSN sequence error */
+			ibp->n_seq_naks++;
+			/*
+			 * Back up to the responder's expected PSN.
+			 * Note that we might get a NAK in the middle of an
+			 * RDMA READ response which terminates the RDMA
+			 * READ.
+			 */
+			qib_restart_rc(qp, psn, 0);
+			qib_schedule_send(qp);
+			break;
+
+		case 1: /* Invalid Request */
+			status = IB_WC_REM_INV_REQ_ERR;
+			ibp->n_other_naks++;
+			goto class_b;
+
+		case 2: /* Remote Access Error */
+			status = IB_WC_REM_ACCESS_ERR;
+			ibp->n_other_naks++;
+			goto class_b;
+
+		case 3: /* Remote Operation Error */
+			status = IB_WC_REM_OP_ERR;
+			ibp->n_other_naks++;
+class_b:
+			if (qp->s_last == qp->s_acked) {
+				qib_send_complete(qp, wqe, status);
+				qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+			}
+			break;
+
+		default:
+			/* Ignore other reserved NAK error codes */
+			goto reserved;
+		}
+		qp->s_retry = qp->s_retry_cnt;
+		qp->s_rnr_retry = qp->s_rnr_retry_cnt;
+		goto bail;
+
+	default:                /* 2: reserved */
+reserved:
+		/* Ignore reserved NAK codes. */
+		goto bail;
+	}
+
+bail:
+	return ret;
+}
+
+/*
+ * We have seen an out of sequence RDMA read middle or last packet.
+ * This ACKs SENDs and RDMA writes up to the first RDMA read or atomic SWQE.
+ */
+static void rdma_seq_err(struct qib_qp *qp, struct qib_ibport *ibp, u32 psn,
+			 struct qib_ctxtdata *rcd)
+{
+	struct qib_swqe *wqe;
+
+	/* Remove QP from retry timer */
+	if (qp->s_flags & (QIB_S_TIMER | QIB_S_WAIT_RNR)) {
+		qp->s_flags &= ~(QIB_S_TIMER | QIB_S_WAIT_RNR);
+		del_timer(&qp->s_timer);
+	}
+
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+
+	while (qib_cmp24(psn, wqe->lpsn) > 0) {
+		if (wqe->wr.opcode == IB_WR_RDMA_READ ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_CMP_AND_SWP ||
+		    wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD)
+			break;
+		wqe = do_rc_completion(qp, wqe, ibp);
+	}
+
+	ibp->n_rdma_seq++;
+	qp->r_flags |= QIB_R_RDMAR_SEQ;
+	qib_restart_rc(qp, qp->s_last_psn + 1, 0);
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_SEND;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+}
+
+/**
+ * qib_rc_rcv_resp - process an incoming RC response packet
+ * @ibp: the port this packet came in on
+ * @ohdr: the other headers for this packet
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP for this packet
+ * @opcode: the opcode for this packet
+ * @psn: the packet sequence number for this packet
+ * @hdrsize: the header length
+ * @pmtu: the path MTU
+ *
+ * This is called from qib_rc_rcv() to process an incoming RC response
+ * packet for the given QP.
+ * Called at interrupt level.
+ */
+static void qib_rc_rcv_resp(struct qib_ibport *ibp,
+			    struct qib_other_headers *ohdr,
+			    void *data, u32 tlen,
+			    struct qib_qp *qp,
+			    u32 opcode,
+			    u32 psn, u32 hdrsize, u32 pmtu,
+			    struct qib_ctxtdata *rcd)
+{
+	struct qib_swqe *wqe;
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	enum ib_wc_status status;
+	unsigned long flags;
+	int diff;
+	u32 pad;
+	u32 aeth;
+	u64 val;
+
+	if (opcode != OP(RDMA_READ_RESPONSE_MIDDLE)) {
+		/*
+		 * If ACK'd PSN on SDMA busy list try to make progress to
+		 * reclaim SDMA credits.
+		 */
+		if ((qib_cmp24(psn, qp->s_sending_psn) >= 0) &&
+		    (qib_cmp24(qp->s_sending_psn, qp->s_sending_hpsn) <= 0)) {
+
+			/*
+			 * If send tasklet not running attempt to progress
+			 * SDMA queue.
+			 */
+			if (!(qp->s_flags & QIB_S_BUSY)) {
+				/* Acquire SDMA Lock */
+				spin_lock_irqsave(&ppd->sdma_lock, flags);
+				/* Invoke sdma make progress */
+				qib_sdma_make_progress(ppd);
+				/* Release SDMA Lock */
+				spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+			}
+		}
+	}
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK))
+		goto ack_done;
+
+	/* Ignore invalid responses. */
+	if (qib_cmp24(psn, qp->s_next_psn) >= 0)
+		goto ack_done;
+
+	/* Ignore duplicate responses. */
+	diff = qib_cmp24(psn, qp->s_last_psn);
+	if (unlikely(diff <= 0)) {
+		/* Update credits for "ghost" ACKs */
+		if (diff == 0 && opcode == OP(ACKNOWLEDGE)) {
+			aeth = be32_to_cpu(ohdr->u.aeth);
+			if ((aeth >> 29) == 0)
+				qib_get_credit(qp, aeth);
+		}
+		goto ack_done;
+	}
+
+	/*
+	 * Skip everything other than the PSN we expect, if we are waiting
+	 * for a reply to a restarted RDMA read or atomic op.
+	 */
+	if (qp->r_flags & QIB_R_RDMAR_SEQ) {
+		if (qib_cmp24(psn, qp->s_last_psn + 1) != 0)
+			goto ack_done;
+		qp->r_flags &= ~QIB_R_RDMAR_SEQ;
+	}
+
+	if (unlikely(qp->s_acked == qp->s_tail))
+		goto ack_done;
+	wqe = get_swqe_ptr(qp, qp->s_acked);
+	status = IB_WC_SUCCESS;
+
+	switch (opcode) {
+	case OP(ACKNOWLEDGE):
+	case OP(ATOMIC_ACKNOWLEDGE):
+	case OP(RDMA_READ_RESPONSE_FIRST):
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		if (opcode == OP(ATOMIC_ACKNOWLEDGE)) {
+			__be32 *p = ohdr->u.at.atomic_ack_eth;
+
+			val = ((u64) be32_to_cpu(p[0]) << 32) |
+				be32_to_cpu(p[1]);
+		} else
+			val = 0;
+		if (!do_rc_ack(qp, aeth, psn, opcode, val, rcd) ||
+		    opcode != OP(RDMA_READ_RESPONSE_FIRST))
+			goto ack_done;
+		hdrsize += 4;
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+		/*
+		 * If this is a response to a resent RDMA read, we
+		 * have to be careful to copy the data to the right
+		 * location.
+		 */
+		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
+						  wqe, psn, pmtu);
+		goto read_middle;
+
+	case OP(RDMA_READ_RESPONSE_MIDDLE):
+		/* no AETH, no ACK */
+		if (unlikely(qib_cmp24(psn, qp->s_last_psn + 1)))
+			goto ack_seq_err;
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+read_middle:
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto ack_len_err;
+		if (unlikely(pmtu >= qp->s_rdma_read_len))
+			goto ack_len_err;
+
+		/*
+		 * We got a response so update the timeout.
+		 * 4.096 usec. * (1 << qp->timeout)
+		 */
+		qp->s_flags |= QIB_S_TIMER;
+		mod_timer(&qp->s_timer, jiffies + qp->timeout_jiffies);
+		if (qp->s_flags & QIB_S_WAIT_ACK) {
+			qp->s_flags &= ~QIB_S_WAIT_ACK;
+			qib_schedule_send(qp);
+		}
+
+		if (opcode == OP(RDMA_READ_RESPONSE_MIDDLE))
+			qp->s_retry = qp->s_retry_cnt;
+
+		/*
+		 * Update the RDMA receive state but do the copy w/o
+		 * holding the locks and blocking interrupts.
+		 */
+		qp->s_rdma_read_len -= pmtu;
+		update_last_psn(qp, psn);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		qib_copy_sge(&qp->s_rdma_read_sge, data, pmtu, 0);
+		goto bail;
+
+	case OP(RDMA_READ_RESPONSE_ONLY):
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		if (!do_rc_ack(qp, aeth, psn, opcode, 0, rcd))
+			goto ack_done;
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/*
+		 * Check that the data size is >= 0 && <= pmtu.
+		 * Remember to account for the AETH header (4) and
+		 * ICRC (4).
+		 */
+		if (unlikely(tlen < (hdrsize + pad + 8)))
+			goto ack_len_err;
+		/*
+		 * If this is a response to a resent RDMA read, we
+		 * have to be careful to copy the data to the right
+		 * location.
+		 */
+		wqe = get_swqe_ptr(qp, qp->s_acked);
+		qp->s_rdma_read_len = restart_sge(&qp->s_rdma_read_sge,
+						  wqe, psn, pmtu);
+		goto read_last;
+
+	case OP(RDMA_READ_RESPONSE_LAST):
+		/* ACKs READ req. */
+		if (unlikely(qib_cmp24(psn, qp->s_last_psn + 1)))
+			goto ack_seq_err;
+		if (unlikely(wqe->wr.opcode != IB_WR_RDMA_READ))
+			goto ack_op_err;
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/*
+		 * Check that the data size is >= 1 && <= pmtu.
+		 * Remember to account for the AETH header (4) and
+		 * ICRC (4).
+		 */
+		if (unlikely(tlen <= (hdrsize + pad + 8)))
+			goto ack_len_err;
+read_last:
+		tlen -= hdrsize + pad + 8;
+		if (unlikely(tlen != qp->s_rdma_read_len))
+			goto ack_len_err;
+		aeth = be32_to_cpu(ohdr->u.aeth);
+		qib_copy_sge(&qp->s_rdma_read_sge, data, tlen, 0);
+		WARN_ON(qp->s_rdma_read_sge.num_sge);
+		(void) do_rc_ack(qp, aeth, psn,
+				 OP(RDMA_READ_RESPONSE_LAST), 0, rcd);
+		goto ack_done;
+	}
+
+ack_op_err:
+	status = IB_WC_LOC_QP_OP_ERR;
+	goto ack_err;
+
+ack_seq_err:
+	rdma_seq_err(qp, ibp, psn, rcd);
+	goto ack_done;
+
+ack_len_err:
+	status = IB_WC_LOC_LEN_ERR;
+ack_err:
+	if (qp->s_last == qp->s_acked) {
+		qib_send_complete(qp, wqe, status);
+		qib_error_qp(qp, IB_WC_WR_FLUSH_ERR);
+	}
+ack_done:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+bail:
+	return;
+}
+
+/**
+ * qib_rc_rcv_error - process an incoming duplicate or error RC packet
+ * @ohdr: the other headers for this packet
+ * @data: the packet data
+ * @qp: the QP for this packet
+ * @opcode: the opcode for this packet
+ * @psn: the packet sequence number for this packet
+ * @diff: the difference between the PSN and the expected PSN
+ *
+ * This is called from qib_rc_rcv() to process an unexpected
+ * incoming RC packet for the given QP.
+ * Called at interrupt level.
+ * Return 1 if no more processing is needed; otherwise return 0 to
+ * schedule a response to be sent.
+ */
+static int qib_rc_rcv_error(struct qib_other_headers *ohdr,
+			    void *data,
+			    struct qib_qp *qp,
+			    u32 opcode,
+			    u32 psn,
+			    int diff,
+			    struct qib_ctxtdata *rcd)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_ack_entry *e;
+	unsigned long flags;
+	u8 i, prev;
+	int old_req;
+
+	if (diff > 0) {
+		/*
+		 * Packet sequence error.
+		 * A NAK will ACK earlier sends and RDMA writes.
+		 * Don't queue the NAK if we already sent one.
+		 */
+		if (!qp->r_nak_state) {
+			ibp->n_rc_seqnak++;
+			qp->r_nak_state = IB_NAK_PSN_ERROR;
+			/* Use the expected PSN. */
+			qp->r_ack_psn = qp->r_psn;
+			/*
+			 * Wait to send the sequence NAK until all packets
+			 * in the receive queue have been processed.
+			 * Otherwise, we end up propagating congestion.
+			 */
+			if (list_empty(&qp->rspwait)) {
+				qp->r_flags |= QIB_R_RSP_NAK;
+				atomic_inc(&qp->refcount);
+				list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+			}
+		}
+		goto done;
+	}
+
+	/*
+	 * Handle a duplicate request.  Don't re-execute SEND, RDMA
+	 * write or atomic op.  Don't NAK errors, just silently drop
+	 * the duplicate request.  Note that r_sge, r_len, and
+	 * r_rcv_len may be in use so don't modify them.
+	 *
+	 * We are supposed to ACK the earliest duplicate PSN but we
+	 * can coalesce an outstanding duplicate ACK.  We have to
+	 * send the earliest so that RDMA reads can be restarted at
+	 * the requester's expected PSN.
+	 *
+	 * First, find where this duplicate PSN falls within the
+	 * ACKs previously sent.
+	 * old_req is true if there is an older response that is scheduled
+	 * to be sent before sending this one.
+	 */
+	e = NULL;
+	old_req = 1;
+	ibp->n_rc_dupreq++;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	for (i = qp->r_head_ack_queue; ; i = prev) {
+		if (i == qp->s_tail_ack_queue)
+			old_req = 0;
+		if (i)
+			prev = i - 1;
+		else
+			prev = QIB_MAX_RDMA_ATOMIC;
+		if (prev == qp->r_head_ack_queue) {
+			e = NULL;
+			break;
+		}
+		e = &qp->s_ack_queue[prev];
+		if (!e->opcode) {
+			e = NULL;
+			break;
+		}
+		if (qib_cmp24(psn, e->psn) >= 0) {
+			if (prev == qp->s_tail_ack_queue &&
+			    qib_cmp24(psn, e->lpsn) <= 0)
+				old_req = 0;
+			break;
+		}
+	}
+	switch (opcode) {
+	case OP(RDMA_READ_REQUEST): {
+		struct ib_reth *reth;
+		u32 offset;
+		u32 len;
+
+		/*
+		 * If we didn't find the RDMA read request in the ack queue,
+		 * we can ignore this request.
+		 */
+		if (!e || e->opcode != OP(RDMA_READ_REQUEST))
+			goto unlock_done;
+		/* RETH comes after BTH */
+		reth = &ohdr->u.rc.reth;
+		/*
+		 * Address range must be a subset of the original
+		 * request and start on pmtu boundaries.
+		 * We reuse the old ack_queue slot since the requester
+		 * should not back up and request an earlier PSN for the
+		 * same request.
+		 */
+		offset = ((psn - e->psn) & QIB_PSN_MASK) *
+			qp->pmtu;
+		len = be32_to_cpu(reth->length);
+		if (unlikely(offset + len != e->rdma_sge.sge_length))
+			goto unlock_done;
+		if (e->rdma_sge.mr) {
+			qib_put_mr(e->rdma_sge.mr);
+			e->rdma_sge.mr = NULL;
+		}
+		if (len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			ok = qib_rkey_ok(qp, &e->rdma_sge, len, vaddr, rkey,
+					 IB_ACCESS_REMOTE_READ);
+			if (unlikely(!ok))
+				goto unlock_done;
+		} else {
+			e->rdma_sge.vaddr = NULL;
+			e->rdma_sge.length = 0;
+			e->rdma_sge.sge_length = 0;
+		}
+		e->psn = psn;
+		if (old_req)
+			goto unlock_done;
+		qp->s_tail_ack_queue = prev;
+		break;
+	}
+
+	case OP(COMPARE_SWAP):
+	case OP(FETCH_ADD): {
+		/*
+		 * If we didn't find the atomic request in the ack queue
+		 * or the send tasklet is already backed up to send an
+		 * earlier entry, we can ignore this request.
+		 */
+		if (!e || e->opcode != (u8) opcode || old_req)
+			goto unlock_done;
+		qp->s_tail_ack_queue = prev;
+		break;
+	}
+
+	default:
+		/*
+		 * Ignore this operation if it doesn't request an ACK
+		 * or an earlier RDMA read or atomic is going to be resent.
+		 */
+		if (!(psn & IB_BTH_REQ_ACK) || old_req)
+			goto unlock_done;
+		/*
+		 * Resend the most recent ACK if this request is
+		 * after all the previous RDMA reads and atomics.
+		 */
+		if (i == qp->r_head_ack_queue) {
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qp->r_nak_state = 0;
+			qp->r_ack_psn = qp->r_psn - 1;
+			goto send_ack;
+		}
+		/*
+		 * Try to send a simple ACK to work around a Mellanox bug
+		 * which doesn't accept a RDMA read response or atomic
+		 * response as an ACK for earlier SENDs or RDMA writes.
+		 */
+		if (!(qp->s_flags & QIB_S_RESP_PENDING)) {
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qp->r_nak_state = 0;
+			qp->r_ack_psn = qp->s_ack_queue[i].psn - 1;
+			goto send_ack;
+		}
+		/*
+		 * Resend the RDMA read or atomic op which
+		 * ACKs this duplicate request.
+		 */
+		qp->s_tail_ack_queue = i;
+		break;
+	}
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+	qp->s_flags |= QIB_S_RESP_PENDING;
+	qp->r_nak_state = 0;
+	qib_schedule_send(qp);
+
+unlock_done:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+done:
+	return 1;
+
+send_ack:
+	return 0;
+}
+
+void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err)
+{
+	unsigned long flags;
+	int lastwqe;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	lastwqe = qib_error_qp(qp, err);
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	if (lastwqe) {
+		struct ib_event ev;
+
+		ev.device = qp->ibqp.device;
+		ev.element.qp = &qp->ibqp;
+		ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+		qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+	}
+}
+
+static inline void qib_update_ack_queue(struct qib_qp *qp, unsigned n)
+{
+	unsigned next;
+
+	next = n + 1;
+	if (next > QIB_MAX_RDMA_ATOMIC)
+		next = 0;
+	qp->s_tail_ack_queue = next;
+	qp->s_ack_state = OP(ACKNOWLEDGE);
+}
+
+/**
+ * qib_rc_rcv - process an incoming RC packet
+ * @rcd: the context pointer
+ * @hdr: the header of this packet
+ * @has_grh: true if the header has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP for this packet
+ *
+ * This is called from qib_qp_rcv() to process an incoming RC packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = &rcd->ppd->ibport_data;
+	struct qib_other_headers *ohdr;
+	u32 opcode;
+	u32 hdrsize;
+	u32 psn;
+	u32 pad;
+	struct ib_wc wc;
+	u32 pmtu = qp->pmtu;
+	int diff;
+	struct ib_reth *reth;
+	unsigned long flags;
+	int ret;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12;       /* LRH + BTH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
+	}
+
+	opcode = be32_to_cpu(ohdr->bth[0]);
+	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
+		return;
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	opcode >>= 24;
+
+	/*
+	 * Process responses (ACKs) before anything else.  Note that the
+	 * packet sequence number will be for something in the send work
+	 * queue rather than the expected receive packet sequence number.
+	 * In other words, this QP is the requester.
+	 */
+	if (opcode >= OP(RDMA_READ_RESPONSE_FIRST) &&
+	    opcode <= OP(ATOMIC_ACKNOWLEDGE)) {
+		qib_rc_rcv_resp(ibp, ohdr, data, tlen, qp, opcode, psn,
+				hdrsize, pmtu, rcd);
+		return;
+	}
+
+	/* Compute 24 bits worth of difference. */
+	diff = qib_cmp24(psn, qp->r_psn);
+	if (unlikely(diff)) {
+		if (qib_rc_rcv_error(ohdr, data, qp, opcode, psn, diff, rcd))
+			return;
+		goto send_ack;
+	}
+
+	/* Check for opcode sequence errors. */
+	switch (qp->r_state) {
+	case OP(SEND_FIRST):
+	case OP(SEND_MIDDLE):
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE))
+			break;
+		goto nack_inv;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_MIDDLE):
+		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			break;
+		goto nack_inv;
+
+	default:
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			goto nack_inv;
+		/*
+		 * Note that it is up to the requester to not send a new
+		 * RDMA read or atomic operation before receiving an ACK
+		 * for the previous operation.
+		 */
+		break;
+	}
+
+	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
+		qp->r_flags |= QIB_R_COMM_EST;
+		if (qp->ibqp.event_handler) {
+			struct ib_event ev;
+
+			ev.device = qp->ibqp.device;
+			ev.element.qp = &qp->ibqp;
+			ev.event = IB_EVENT_COMM_EST;
+			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+		}
+	}
+
+	/* OK, process the packet. */
+	switch (opcode) {
+	case OP(SEND_FIRST):
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		qp->r_rcv_len = 0;
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+	case OP(RDMA_WRITE_MIDDLE):
+send_middle:
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto nack_inv;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto nack_inv;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
+		/* consume RWQE */
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		goto send_last_imm;
+
+	case OP(SEND_ONLY):
+	case OP(SEND_ONLY_WITH_IMMEDIATE):
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		qp->r_rcv_len = 0;
+		if (opcode == OP(SEND_ONLY))
+			goto no_immediate_data;
+		/* FALLTHROUGH for SEND_ONLY_WITH_IMMEDIATE */
+	case OP(SEND_LAST_WITH_IMMEDIATE):
+send_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		goto send_last;
+	case OP(SEND_LAST):
+	case OP(RDMA_WRITE_LAST):
+no_immediate_data:
+		wc.wc_flags = 0;
+		wc.ex.imm_data = 0;
+send_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto nack_inv;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		wc.byte_len = tlen + qp->r_rcv_len;
+		if (unlikely(wc.byte_len > qp->r_len))
+			goto nack_inv;
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		qib_put_ss(&qp->r_sge);
+		qp->r_msn++;
+		if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+			break;
+		wc.wr_id = qp->r_wr_id;
+		wc.status = IB_WC_SUCCESS;
+		if (opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		else
+			wc.opcode = IB_WC_RECV;
+		wc.qp = &qp->ibqp;
+		wc.src_qp = qp->remote_qpn;
+		wc.slid = qp->remote_ah_attr.dlid;
+		wc.sl = qp->remote_ah_attr.sl;
+		/* zero fields that are N/A */
+		wc.vendor_err = 0;
+		wc.pkey_index = 0;
+		wc.dlid_path_bits = 0;
+		wc.port_num = 0;
+		/* Signal completion event if the solicited bit is set. */
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+			     (ohdr->bth[0] &
+			      cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_ONLY):
+	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto nack_inv;
+		/* consume RWQE */
+		reth = &ohdr->u.rc.reth;
+		hdrsize += sizeof(*reth);
+		qp->r_len = be32_to_cpu(reth->length);
+		qp->r_rcv_len = 0;
+		qp->r_sge.sg_list = NULL;
+		if (qp->r_len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey & NAK */
+			ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len, vaddr,
+					 rkey, IB_ACCESS_REMOTE_WRITE);
+			if (unlikely(!ok))
+				goto nack_acc;
+			qp->r_sge.num_sge = 1;
+		} else {
+			qp->r_sge.num_sge = 0;
+			qp->r_sge.sge.mr = NULL;
+			qp->r_sge.sge.vaddr = NULL;
+			qp->r_sge.sge.length = 0;
+			qp->r_sge.sge.sge_length = 0;
+		}
+		if (opcode == OP(RDMA_WRITE_FIRST))
+			goto send_middle;
+		else if (opcode == OP(RDMA_WRITE_ONLY))
+			goto no_immediate_data;
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto nack_op_err;
+		if (!ret)
+			goto rnr_nak;
+		wc.ex.imm_data = ohdr->u.rc.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		goto send_last;
+
+	case OP(RDMA_READ_REQUEST): {
+		struct qib_ack_entry *e;
+		u32 len;
+		u8 next;
+
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+			goto nack_inv;
+		next = qp->r_head_ack_queue + 1;
+		/* s_ack_queue is size QIB_MAX_RDMA_ATOMIC+1 so use > not >= */
+		if (next > QIB_MAX_RDMA_ATOMIC)
+			next = 0;
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv_unlck;
+			qib_update_ack_queue(qp, next);
+		}
+		e = &qp->s_ack_queue[qp->r_head_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
+			qib_put_mr(e->rdma_sge.mr);
+			e->rdma_sge.mr = NULL;
+		}
+		reth = &ohdr->u.rc.reth;
+		len = be32_to_cpu(reth->length);
+		if (len) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey & NAK */
+			ok = qib_rkey_ok(qp, &e->rdma_sge, len, vaddr,
+					 rkey, IB_ACCESS_REMOTE_READ);
+			if (unlikely(!ok))
+				goto nack_acc_unlck;
+			/*
+			 * Update the next expected PSN.  We add 1 later
+			 * below, so only add the remainder here.
+			 */
+			if (len > pmtu)
+				qp->r_psn += (len - 1) / pmtu;
+		} else {
+			e->rdma_sge.mr = NULL;
+			e->rdma_sge.vaddr = NULL;
+			e->rdma_sge.length = 0;
+			e->rdma_sge.sge_length = 0;
+		}
+		e->opcode = opcode;
+		e->sent = 0;
+		e->psn = psn;
+		e->lpsn = qp->r_psn;
+		/*
+		 * We need to increment the MSN here instead of when we
+		 * finish sending the result since a duplicate request would
+		 * increment it more than once.
+		 */
+		qp->r_msn++;
+		qp->r_psn++;
+		qp->r_state = opcode;
+		qp->r_nak_state = 0;
+		qp->r_head_ack_queue = next;
+
+		/* Schedule the send tasklet. */
+		qp->s_flags |= QIB_S_RESP_PENDING;
+		qib_schedule_send(qp);
+
+		goto sunlock;
+	}
+
+	case OP(COMPARE_SWAP):
+	case OP(FETCH_ADD): {
+		struct ib_atomic_eth *ateth;
+		struct qib_ack_entry *e;
+		u64 vaddr;
+		atomic64_t *maddr;
+		u64 sdata;
+		u32 rkey;
+		u8 next;
+
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+			goto nack_inv;
+		next = qp->r_head_ack_queue + 1;
+		if (next > QIB_MAX_RDMA_ATOMIC)
+			next = 0;
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (unlikely(next == qp->s_tail_ack_queue)) {
+			if (!qp->s_ack_queue[next].sent)
+				goto nack_inv_unlck;
+			qib_update_ack_queue(qp, next);
+		}
+		e = &qp->s_ack_queue[qp->r_head_ack_queue];
+		if (e->opcode == OP(RDMA_READ_REQUEST) && e->rdma_sge.mr) {
+			qib_put_mr(e->rdma_sge.mr);
+			e->rdma_sge.mr = NULL;
+		}
+		ateth = &ohdr->u.atomic_eth;
+		vaddr = ((u64) be32_to_cpu(ateth->vaddr[0]) << 32) |
+			be32_to_cpu(ateth->vaddr[1]);
+		if (unlikely(vaddr & (sizeof(u64) - 1)))
+			goto nack_inv_unlck;
+		rkey = be32_to_cpu(ateth->rkey);
+		/* Check rkey & NAK */
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+					  vaddr, rkey,
+					  IB_ACCESS_REMOTE_ATOMIC)))
+			goto nack_acc_unlck;
+		/* Perform atomic OP and save result. */
+		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
+		sdata = be64_to_cpu(ateth->swap_data);
+		e->atomic_data = (opcode == OP(FETCH_ADD)) ?
+			(u64) atomic64_add_return(sdata, maddr) - sdata :
+			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
+				      be64_to_cpu(ateth->compare_data),
+				      sdata);
+		qib_put_mr(qp->r_sge.sge.mr);
+		qp->r_sge.num_sge = 0;
+		e->opcode = opcode;
+		e->sent = 0;
+		e->psn = psn;
+		e->lpsn = psn;
+		qp->r_msn++;
+		qp->r_psn++;
+		qp->r_state = opcode;
+		qp->r_nak_state = 0;
+		qp->r_head_ack_queue = next;
+
+		/* Schedule the send tasklet. */
+		qp->s_flags |= QIB_S_RESP_PENDING;
+		qib_schedule_send(qp);
+
+		goto sunlock;
+	}
+
+	default:
+		/* NAK unknown opcodes. */
+		goto nack_inv;
+	}
+	qp->r_psn++;
+	qp->r_state = opcode;
+	qp->r_ack_psn = psn;
+	qp->r_nak_state = 0;
+	/* Send an ACK if requested or required. */
+	if (psn & (1 << 31))
+		goto send_ack;
+	return;
+
+rnr_nak:
+	qp->r_nak_state = IB_RNR_NAK | qp->r_min_rnr_timer;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue RNR NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	return;
+
+nack_op_err:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	qp->r_nak_state = IB_NAK_REMOTE_OPERATIONAL_ERROR;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	return;
+
+nack_inv_unlck:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+nack_inv:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	qp->r_nak_state = IB_NAK_INVALID_REQUEST;
+	qp->r_ack_psn = qp->r_psn;
+	/* Queue NAK for later */
+	if (list_empty(&qp->rspwait)) {
+		qp->r_flags |= QIB_R_RSP_NAK;
+		atomic_inc(&qp->refcount);
+		list_add_tail(&qp->rspwait, &rcd->qp_wait_list);
+	}
+	return;
+
+nack_acc_unlck:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+nack_acc:
+	qib_rc_error(qp, IB_WC_LOC_PROT_ERR);
+	qp->r_nak_state = IB_NAK_REMOTE_ACCESS_ERROR;
+	qp->r_ack_psn = qp->r_psn;
+send_ack:
+	qib_send_rc_ack(qp);
+	return;
+
+sunlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+}
diff --git a/drivers/infiniband/hw/qib/qib_ruc.c b/drivers/infiniband/hw/qib/qib_ruc.c
new file mode 100644
index 00000000000..4c07a8b34ff
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_ruc.c
@@ -0,0 +1,819 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/*
+ * Convert the AETH RNR timeout code into the number of microseconds.
+ */
+const u32 ib_qib_rnr_table[32] = {
+	655360,	/* 00: 655.36 */
+	10,	/* 01:    .01 */
+	20,	/* 02     .02 */
+	30,	/* 03:    .03 */
+	40,	/* 04:    .04 */
+	60,	/* 05:    .06 */
+	80,	/* 06:    .08 */
+	120,	/* 07:    .12 */
+	160,	/* 08:    .16 */
+	240,	/* 09:    .24 */
+	320,	/* 0A:    .32 */
+	480,	/* 0B:    .48 */
+	640,	/* 0C:    .64 */
+	960,	/* 0D:    .96 */
+	1280,	/* 0E:   1.28 */
+	1920,	/* 0F:   1.92 */
+	2560,	/* 10:   2.56 */
+	3840,	/* 11:   3.84 */
+	5120,	/* 12:   5.12 */
+	7680,	/* 13:   7.68 */
+	10240,	/* 14:  10.24 */
+	15360,	/* 15:  15.36 */
+	20480,	/* 16:  20.48 */
+	30720,	/* 17:  30.72 */
+	40960,	/* 18:  40.96 */
+	61440,	/* 19:  61.44 */
+	81920,	/* 1A:  81.92 */
+	122880,	/* 1B: 122.88 */
+	163840,	/* 1C: 163.84 */
+	245760,	/* 1D: 245.76 */
+	327680,	/* 1E: 327.68 */
+	491520	/* 1F: 491.52 */
+};
+
+/*
+ * Validate a RWQE and fill in the SGE state.
+ * Return 1 if OK.
+ */
+static int qib_init_sge(struct qib_qp *qp, struct qib_rwqe *wqe)
+{
+	int i, j, ret;
+	struct ib_wc wc;
+	struct qib_lkey_table *rkt;
+	struct qib_pd *pd;
+	struct qib_sge_state *ss;
+
+	rkt = &to_idev(qp->ibqp.device)->lk_table;
+	pd = to_ipd(qp->ibqp.srq ? qp->ibqp.srq->pd : qp->ibqp.pd);
+	ss = &qp->r_sge;
+	ss->sg_list = qp->r_sg_list;
+	qp->r_len = 0;
+	for (i = j = 0; i < wqe->num_sge; i++) {
+		if (wqe->sg_list[i].length == 0)
+			continue;
+		/* Check LKEY */
+		if (!qib_lkey_ok(rkt, pd, j ? &ss->sg_list[j - 1] : &ss->sge,
+				 &wqe->sg_list[i], IB_ACCESS_LOCAL_WRITE))
+			goto bad_lkey;
+		qp->r_len += wqe->sg_list[i].length;
+		j++;
+	}
+	ss->num_sge = j;
+	ss->total_len = qp->r_len;
+	ret = 1;
+	goto bail;
+
+bad_lkey:
+	while (j) {
+		struct qib_sge *sge = --j ? &ss->sg_list[j - 1] : &ss->sge;
+
+		qib_put_mr(sge->mr);
+	}
+	ss->num_sge = 0;
+	memset(&wc, 0, sizeof(wc));
+	wc.wr_id = wqe->wr_id;
+	wc.status = IB_WC_LOC_PROT_ERR;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	/* Signal solicited completion event. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc, 1);
+	ret = 0;
+bail:
+	return ret;
+}
+
+/**
+ * qib_get_rwqe - copy the next RWQE into the QP's RWQE
+ * @qp: the QP
+ * @wr_id_only: update qp->r_wr_id only, not qp->r_sge
+ *
+ * Return -1 if there is a local error, 0 if no RWQE is available,
+ * otherwise return 1.
+ *
+ * Can be called from interrupt level.
+ */
+int qib_get_rwqe(struct qib_qp *qp, int wr_id_only)
+{
+	unsigned long flags;
+	struct qib_rq *rq;
+	struct qib_rwq *wq;
+	struct qib_srq *srq;
+	struct qib_rwqe *wqe;
+	void (*handler)(struct ib_event *, void *);
+	u32 tail;
+	int ret;
+
+	if (qp->ibqp.srq) {
+		srq = to_isrq(qp->ibqp.srq);
+		handler = srq->ibsrq.event_handler;
+		rq = &srq->rq;
+	} else {
+		srq = NULL;
+		handler = NULL;
+		rq = &qp->r_rq;
+	}
+
+	spin_lock_irqsave(&rq->lock, flags);
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ret = 0;
+		goto unlock;
+	}
+
+	wq = rq->wq;
+	tail = wq->tail;
+	/* Validate tail before using it since it is user writable. */
+	if (tail >= rq->size)
+		tail = 0;
+	if (unlikely(tail == wq->head)) {
+		ret = 0;
+		goto unlock;
+	}
+	/* Make sure entry is read after head index is read. */
+	smp_rmb();
+	wqe = get_rwqe_ptr(rq, tail);
+	/*
+	 * Even though we update the tail index in memory, the verbs
+	 * consumer is not supposed to post more entries until a
+	 * completion is generated.
+	 */
+	if (++tail >= rq->size)
+		tail = 0;
+	wq->tail = tail;
+	if (!wr_id_only && !qib_init_sge(qp, wqe)) {
+		ret = -1;
+		goto unlock;
+	}
+	qp->r_wr_id = wqe->wr_id;
+
+	ret = 1;
+	set_bit(QIB_R_WRID_VALID, &qp->r_aflags);
+	if (handler) {
+		u32 n;
+
+		/*
+		 * Validate head pointer value and compute
+		 * the number of remaining WQEs.
+		 */
+		n = wq->head;
+		if (n >= rq->size)
+			n = 0;
+		if (n < tail)
+			n += rq->size - tail;
+		else
+			n -= tail;
+		if (n < srq->limit) {
+			struct ib_event ev;
+
+			srq->limit = 0;
+			spin_unlock_irqrestore(&rq->lock, flags);
+			ev.device = qp->ibqp.device;
+			ev.element.srq = qp->ibqp.srq;
+			ev.event = IB_EVENT_SRQ_LIMIT_REACHED;
+			handler(&ev, srq->ibsrq.srq_context);
+			goto bail;
+		}
+	}
+unlock:
+	spin_unlock_irqrestore(&rq->lock, flags);
+bail:
+	return ret;
+}
+
+/*
+ * Switch to alternate path.
+ * The QP s_lock should be held and interrupts disabled.
+ */
+void qib_migrate_qp(struct qib_qp *qp)
+{
+	struct ib_event ev;
+
+	qp->s_mig_state = IB_MIG_MIGRATED;
+	qp->remote_ah_attr = qp->alt_ah_attr;
+	qp->port_num = qp->alt_ah_attr.port_num;
+	qp->s_pkey_index = qp->s_alt_pkey_index;
+
+	ev.device = qp->ibqp.device;
+	ev.element.qp = &qp->ibqp;
+	ev.event = IB_EVENT_PATH_MIG;
+	qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+}
+
+static __be64 get_sguid(struct qib_ibport *ibp, unsigned index)
+{
+	if (!index) {
+		struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+		return ppd->guid;
+	} else
+		return ibp->guids[index - 1];
+}
+
+static int gid_ok(union ib_gid *gid, __be64 gid_prefix, __be64 id)
+{
+	return (gid->global.interface_id == id &&
+		(gid->global.subnet_prefix == gid_prefix ||
+		 gid->global.subnet_prefix == IB_DEFAULT_GID_PREFIX));
+}
+
+/*
+ *
+ * This should be called with the QP r_lock held.
+ *
+ * The s_lock will be acquired around the qib_migrate_qp() call.
+ */
+int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		      int has_grh, struct qib_qp *qp, u32 bth0)
+{
+	__be64 guid;
+	unsigned long flags;
+
+	if (qp->s_mig_state == IB_MIG_ARMED && (bth0 & IB_BTH_MIG_REQ)) {
+		if (!has_grh) {
+			if (qp->alt_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->alt_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp, qp->alt_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->alt_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->alt_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_alt_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 and 17.2.8 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->alt_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->alt_ah_attr.port_num)
+			goto err;
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qib_migrate_qp(qp);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	} else {
+		if (!has_grh) {
+			if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+				goto err;
+		} else {
+			if (!(qp->remote_ah_attr.ah_flags & IB_AH_GRH))
+				goto err;
+			guid = get_sguid(ibp,
+					 qp->remote_ah_attr.grh.sgid_index);
+			if (!gid_ok(&hdr->u.l.grh.dgid, ibp->gid_prefix, guid))
+				goto err;
+			if (!gid_ok(&hdr->u.l.grh.sgid,
+			    qp->remote_ah_attr.grh.dgid.global.subnet_prefix,
+			    qp->remote_ah_attr.grh.dgid.global.interface_id))
+				goto err;
+		}
+		if (!qib_pkey_ok((u16)bth0,
+				 qib_get_pkey(ibp, qp->s_pkey_index))) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+				      (u16)bth0,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      0, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			goto err;
+		}
+		/* Validate the SLID. See Ch. 9.6.1.5 */
+		if (be16_to_cpu(hdr->lrh[3]) != qp->remote_ah_attr.dlid ||
+		    ppd_from_ibp(ibp)->port != qp->port_num)
+			goto err;
+		if (qp->s_mig_state == IB_MIG_REARM &&
+		    !(bth0 & IB_BTH_MIG_REQ))
+			qp->s_mig_state = IB_MIG_ARMED;
+	}
+
+	return 0;
+
+err:
+	return 1;
+}
+
+/**
+ * qib_ruc_loopback - handle UC and RC lookback requests
+ * @sqp: the sending QP
+ *
+ * This is called from qib_do_send() to
+ * forward a WQE addressed to the same HCA.
+ * Note that although we are single threaded due to the tasklet, we still
+ * have to protect against post_send().  We don't have to worry about
+ * receive interrupts since this is a connected protocol and all packets
+ * will pass through here.
+ */
+static void qib_ruc_loopback(struct qib_qp *sqp)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_qp *qp;
+	struct qib_swqe *wqe;
+	struct qib_sge *sge;
+	unsigned long flags;
+	struct ib_wc wc;
+	u64 sdata;
+	atomic64_t *maddr;
+	enum ib_wc_status send_status;
+	int release;
+	int ret;
+
+	/*
+	 * Note that we check the responder QP state after
+	 * checking the requester's state.
+	 */
+	qp = qib_lookup_qpn(ibp, sqp->remote_qpn);
+
+	spin_lock_irqsave(&sqp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if ((sqp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT)) ||
+	    !(ib_qib_state_ops[sqp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		goto unlock;
+
+	sqp->s_flags |= QIB_S_BUSY;
+
+again:
+	if (sqp->s_last == sqp->s_head)
+		goto clr_busy;
+	wqe = get_swqe_ptr(sqp, sqp->s_last);
+
+	/* Return if it is not OK to start a new work reqeust. */
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[sqp->state] & QIB_FLUSH_SEND))
+			goto clr_busy;
+		/* We are in the error state, flush the work request. */
+		send_status = IB_WC_WR_FLUSH_ERR;
+		goto flush_send;
+	}
+
+	/*
+	 * We can rely on the entry not changing without the s_lock
+	 * being held until we update s_last.
+	 * We increment s_cur to indicate s_last is in progress.
+	 */
+	if (sqp->s_last == sqp->s_cur) {
+		if (++sqp->s_cur >= sqp->s_size)
+			sqp->s_cur = 0;
+	}
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+
+	if (!qp || !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) ||
+	    qp->ibqp.qp_type != sqp->ibqp.qp_type) {
+		ibp->n_pkt_drops++;
+		/*
+		 * For RC, the requester would timeout and retry so
+		 * shortcut the timeouts and just signal too many retries.
+		 */
+		if (sqp->ibqp.qp_type == IB_QPT_RC)
+			send_status = IB_WC_RETRY_EXC_ERR;
+		else
+			send_status = IB_WC_SUCCESS;
+		goto serr;
+	}
+
+	memset(&wc, 0, sizeof wc);
+	send_status = IB_WC_SUCCESS;
+
+	release = 1;
+	sqp->s_sge.sge = wqe->sg_list[0];
+	sqp->s_sge.sg_list = wqe->sg_list + 1;
+	sqp->s_sge.num_sge = wqe->wr.num_sge;
+	sqp->s_len = wqe->length;
+	switch (wqe->wr.opcode) {
+	case IB_WR_SEND_WITH_IMM:
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		/* FALLTHROUGH */
+	case IB_WR_SEND:
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		break;
+
+	case IB_WR_RDMA_WRITE_WITH_IMM:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = wqe->wr.ex.imm_data;
+		ret = qib_get_rwqe(qp, 1);
+		if (ret < 0)
+			goto op_err;
+		if (!ret)
+			goto rnr_nak;
+		/* FALLTHROUGH */
+	case IB_WR_RDMA_WRITE:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_WRITE)))
+			goto inv_err;
+		if (wqe->length == 0)
+			break;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_WRITE)))
+			goto acc_err;
+		qp->r_sge.sg_list = NULL;
+		qp->r_sge.num_sge = 1;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_RDMA_READ:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_READ)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &sqp->s_sge.sge, wqe->length,
+					  wqe->wr.wr.rdma.remote_addr,
+					  wqe->wr.wr.rdma.rkey,
+					  IB_ACCESS_REMOTE_READ)))
+			goto acc_err;
+		release = 0;
+		sqp->s_sge.sg_list = NULL;
+		sqp->s_sge.num_sge = 1;
+		qp->r_sge.sge = wqe->sg_list[0];
+		qp->r_sge.sg_list = wqe->sg_list + 1;
+		qp->r_sge.num_sge = wqe->wr.num_sge;
+		qp->r_sge.total_len = wqe->length;
+		break;
+
+	case IB_WR_ATOMIC_CMP_AND_SWP:
+	case IB_WR_ATOMIC_FETCH_AND_ADD:
+		if (unlikely(!(qp->qp_access_flags & IB_ACCESS_REMOTE_ATOMIC)))
+			goto inv_err;
+		if (unlikely(!qib_rkey_ok(qp, &qp->r_sge.sge, sizeof(u64),
+					  wqe->wr.wr.atomic.remote_addr,
+					  wqe->wr.wr.atomic.rkey,
+					  IB_ACCESS_REMOTE_ATOMIC)))
+			goto acc_err;
+		/* Perform atomic OP and save result. */
+		maddr = (atomic64_t *) qp->r_sge.sge.vaddr;
+		sdata = wqe->wr.wr.atomic.compare_add;
+		*(u64 *) sqp->s_sge.sge.vaddr =
+			(wqe->wr.opcode == IB_WR_ATOMIC_FETCH_AND_ADD) ?
+			(u64) atomic64_add_return(sdata, maddr) - sdata :
+			(u64) cmpxchg((u64 *) qp->r_sge.sge.vaddr,
+				      sdata, wqe->wr.wr.atomic.swap);
+		qib_put_mr(qp->r_sge.sge.mr);
+		qp->r_sge.num_sge = 0;
+		goto send_comp;
+
+	default:
+		send_status = IB_WC_LOC_QP_OP_ERR;
+		goto serr;
+	}
+
+	sge = &sqp->s_sge.sge;
+	while (sqp->s_len) {
+		u32 len = sqp->s_len;
+
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, release);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (!release)
+				qib_put_mr(sge->mr);
+			if (--sqp->s_sge.num_sge)
+				*sge = *sqp->s_sge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		sqp->s_len -= len;
+	}
+	if (release)
+		qib_put_ss(&qp->r_sge);
+
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto send_comp;
+
+	if (wqe->wr.opcode == IB_WR_RDMA_WRITE_WITH_IMM)
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+	else
+		wc.opcode = IB_WC_RECV;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.byte_len = wqe->length;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = qp->remote_qpn;
+	wc.slid = qp->remote_ah_attr.dlid;
+	wc.sl = qp->remote_ah_attr.sl;
+	wc.port_num = 1;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		       wqe->wr.send_flags & IB_SEND_SOLICITED);
+
+send_comp:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	ibp->n_loop_pkts++;
+flush_send:
+	sqp->s_rnr_retry = sqp->s_rnr_retry_cnt;
+	qib_send_complete(sqp, wqe, send_status);
+	goto again;
+
+rnr_nak:
+	/* Handle RNR NAK */
+	if (qp->ibqp.qp_type == IB_QPT_UC)
+		goto send_comp;
+	ibp->n_rnr_naks++;
+	/*
+	 * Note: we don't need the s_lock held since the BUSY flag
+	 * makes this single threaded.
+	 */
+	if (sqp->s_rnr_retry == 0) {
+		send_status = IB_WC_RNR_RETRY_EXC_ERR;
+		goto serr;
+	}
+	if (sqp->s_rnr_retry_cnt < 7)
+		sqp->s_rnr_retry--;
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	if (!(ib_qib_state_ops[sqp->state] & QIB_PROCESS_RECV_OK))
+		goto clr_busy;
+	sqp->s_flags |= QIB_S_WAIT_RNR;
+	sqp->s_timer.function = qib_rc_rnr_retry;
+	sqp->s_timer.expires = jiffies +
+		usecs_to_jiffies(ib_qib_rnr_table[qp->r_min_rnr_timer]);
+	add_timer(&sqp->s_timer);
+	goto clr_busy;
+
+op_err:
+	send_status = IB_WC_REM_OP_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+inv_err:
+	send_status = IB_WC_REM_INV_REQ_ERR;
+	wc.status = IB_WC_LOC_QP_OP_ERR;
+	goto err;
+
+acc_err:
+	send_status = IB_WC_REM_ACCESS_ERR;
+	wc.status = IB_WC_LOC_PROT_ERR;
+err:
+	/* responder goes to error state */
+	qib_rc_error(qp, wc.status);
+
+serr:
+	spin_lock_irqsave(&sqp->s_lock, flags);
+	qib_send_complete(sqp, wqe, send_status);
+	if (sqp->ibqp.qp_type == IB_QPT_RC) {
+		int lastwqe = qib_error_qp(sqp, IB_WC_WR_FLUSH_ERR);
+
+		sqp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock_irqrestore(&sqp->s_lock, flags);
+		if (lastwqe) {
+			struct ib_event ev;
+
+			ev.device = sqp->ibqp.device;
+			ev.element.qp = &sqp->ibqp;
+			ev.event = IB_EVENT_QP_LAST_WQE_REACHED;
+			sqp->ibqp.event_handler(&ev, sqp->ibqp.qp_context);
+		}
+		goto done;
+	}
+clr_busy:
+	sqp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&sqp->s_lock, flags);
+done:
+	if (qp && atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_grh - construct a GRH header
+ * @ibp: a pointer to the IB port
+ * @hdr: a pointer to the GRH header being constructed
+ * @grh: the global route address to send to
+ * @hwords: the number of 32 bit words of header being sent
+ * @nwords: the number of 32 bit words of data being sent
+ *
+ * Return the size of the header in 32 bit words.
+ */
+u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
+		 struct ib_global_route *grh, u32 hwords, u32 nwords)
+{
+	hdr->version_tclass_flow =
+		cpu_to_be32((IB_GRH_VERSION << IB_GRH_VERSION_SHIFT) |
+			    (grh->traffic_class << IB_GRH_TCLASS_SHIFT) |
+			    (grh->flow_label << IB_GRH_FLOW_SHIFT));
+	hdr->paylen = cpu_to_be16((hwords - 2 + nwords + SIZE_OF_CRC) << 2);
+	/* next_hdr is defined by C8-7 in ch. 8.4.1 */
+	hdr->next_hdr = IB_GRH_NEXT_HDR;
+	hdr->hop_limit = grh->hop_limit;
+	/* The SGID is 32-bit aligned. */
+	hdr->sgid.global.subnet_prefix = ibp->gid_prefix;
+	hdr->sgid.global.interface_id = grh->sgid_index ?
+		ibp->guids[grh->sgid_index - 1] : ppd_from_ibp(ibp)->guid;
+	hdr->dgid = grh->dgid;
+
+	/* GRH header size in 32-bit words. */
+	return sizeof(struct ib_grh) / sizeof(u32);
+}
+
+void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
+			 u32 bth0, u32 bth2)
+{
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	u16 lrh0;
+	u32 nwords;
+	u32 extra_bytes;
+
+	/* Construct the header. */
+	extra_bytes = -qp->s_cur_size & 3;
+	nwords = (qp->s_cur_size + extra_bytes) >> 2;
+	lrh0 = QIB_LRH_BTH;
+	if (unlikely(qp->remote_ah_attr.ah_flags & IB_AH_GRH)) {
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr->u.l.grh,
+					       &qp->remote_ah_attr.grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+	}
+	lrh0 |= ibp->sl_to_vl[qp->remote_ah_attr.sl] << 12 |
+		qp->remote_ah_attr.sl << 4;
+	qp->s_hdr->lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr->lrh[1] = cpu_to_be16(qp->remote_ah_attr.dlid);
+	qp->s_hdr->lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	qp->s_hdr->lrh[3] = cpu_to_be16(ppd_from_ibp(ibp)->lid |
+				       qp->remote_ah_attr.src_path_bits);
+	bth0 |= qib_get_pkey(ibp, qp->s_pkey_index);
+	bth0 |= extra_bytes << 20;
+	if (qp->s_mig_state == IB_MIG_MIGRATED)
+		bth0 |= IB_BTH_MIG_REQ;
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	ohdr->bth[1] = cpu_to_be32(qp->remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(bth2);
+	this_cpu_inc(ibp->pmastats->n_unicast_xmit);
+}
+
+/**
+ * qib_do_send - perform a send on a QP
+ * @work: contains a pointer to the QP
+ *
+ * Process entries in the send work queue until credit or queue is
+ * exhausted.  Only allow one CPU to send a packet per QP (tasklet).
+ * Otherwise, two threads could send packets out of order.
+ */
+void qib_do_send(struct work_struct *work)
+{
+	struct qib_qp *qp = container_of(work, struct qib_qp, s_work);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	int (*make_req)(struct qib_qp *qp);
+	unsigned long flags;
+
+	if ((qp->ibqp.qp_type == IB_QPT_RC ||
+	     qp->ibqp.qp_type == IB_QPT_UC) &&
+	    (qp->remote_ah_attr.dlid & ~((1 << ppd->lmc) - 1)) == ppd->lid) {
+		qib_ruc_loopback(qp);
+		return;
+	}
+
+	if (qp->ibqp.qp_type == IB_QPT_RC)
+		make_req = qib_make_rc_req;
+	else if (qp->ibqp.qp_type == IB_QPT_UC)
+		make_req = qib_make_uc_req;
+	else
+		make_req = qib_make_ud_req;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Return if we are already busy processing a work request. */
+	if (!qib_send_ok(qp)) {
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		return;
+	}
+
+	qp->s_flags |= QIB_S_BUSY;
+
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	do {
+		/* Check for a constructed packet to be sent. */
+		if (qp->s_hdrwords != 0) {
+			/*
+			 * If the packet cannot be sent now, return and
+			 * the send tasklet will be woken up later.
+			 */
+			if (qib_verbs_send(qp, qp->s_hdr, qp->s_hdrwords,
+					   qp->s_cur_sge, qp->s_cur_size))
+				break;
+			/* Record that s_hdr is empty. */
+			qp->s_hdrwords = 0;
+		}
+	} while (make_req(qp));
+}
+
+/*
+ * This should be called with s_lock held.
+ */
+void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
+		       enum ib_wc_status status)
+{
+	u32 old_last, last;
+	unsigned i;
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_OR_FLUSH_SEND))
+		return;
+
+	for (i = 0; i < wqe->wr.num_sge; i++) {
+		struct qib_sge *sge = &wqe->sg_list[i];
+
+		qib_put_mr(sge->mr);
+	}
+	if (qp->ibqp.qp_type == IB_QPT_UD ||
+	    qp->ibqp.qp_type == IB_QPT_SMI ||
+	    qp->ibqp.qp_type == IB_QPT_GSI)
+		atomic_dec(&to_iah(wqe->wr.wr.ud.ah)->refcount);
+
+	/* See ch. 11.2.4.1 and 10.7.3.1 */
+	if (!(qp->s_flags & QIB_S_SIGNAL_REQ_WR) ||
+	    (wqe->wr.send_flags & IB_SEND_SIGNALED) ||
+	    status != IB_WC_SUCCESS) {
+		struct ib_wc wc;
+
+		memset(&wc, 0, sizeof wc);
+		wc.wr_id = wqe->wr.wr_id;
+		wc.status = status;
+		wc.opcode = ib_qib_wc_opcode[wqe->wr.opcode];
+		wc.qp = &qp->ibqp;
+		if (status == IB_WC_SUCCESS)
+			wc.byte_len = wqe->length;
+		qib_cq_enter(to_icq(qp->ibqp.send_cq), &wc,
+			     status != IB_WC_SUCCESS);
+	}
+
+	last = qp->s_last;
+	old_last = last;
+	if (++last >= qp->s_size)
+		last = 0;
+	qp->s_last = last;
+	if (qp->s_acked == old_last)
+		qp->s_acked = last;
+	if (qp->s_cur == old_last)
+		qp->s_cur = last;
+	if (qp->s_tail == old_last)
+		qp->s_tail = last;
+	if (qp->state == IB_QPS_SQD && last == qp->s_cur)
+		qp->s_draining = 0;
+}
diff --git a/drivers/infiniband/hw/ipath/ipath_sd7220.c b/drivers/infiniband/hw/qib/qib_sd7220.c
index aa47eb54952..911205d3d5a 100644
--- a/drivers/infiniband/hw/ipath/ipath_sd7220.c
+++ b/drivers/infiniband/hw/qib/qib_sd7220.c
@@ -1,5 +1,6 @@
 /*
- * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2013 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
  * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
  *
  * This software is available to you under a choice of one of two
@@ -32,22 +33,45 @@
  */
 /*
  * This file contains all of the code that is specific to the SerDes
- * on the InfiniPath 7220 chip.
+ * on the QLogic_IB 7220 chip.
  */
 
 #include <linux/pci.h>
 #include <linux/delay.h>
+#include <linux/module.h>
+#include <linux/firmware.h>
 
-#include "ipath_kernel.h"
-#include "ipath_registers.h"
-#include "ipath_7220.h"
+#include "qib.h"
+#include "qib_7220.h"
+
+#define SD7220_FW_NAME "qlogic/sd7220.fw"
+MODULE_FIRMWARE(SD7220_FW_NAME);
+
+/*
+ * Same as in qib_iba7220.c, but just the registers needed here.
+ * Could move whole set to qib_7220.h, but decided better to keep
+ * local.
+ */
+#define KREG_IDX(regname) (QIB_7220_##regname##_OFFS / sizeof(u64))
+#define kr_hwerrclear KREG_IDX(HwErrClear)
+#define kr_hwerrmask KREG_IDX(HwErrMask)
+#define kr_hwerrstatus KREG_IDX(HwErrStatus)
+#define kr_ibcstatus KREG_IDX(IBCStatus)
+#define kr_ibserdesctrl KREG_IDX(IBSerDesCtrl)
+#define kr_scratch KREG_IDX(Scratch)
+#define kr_xgxs_cfg KREG_IDX(XGXSCfg)
+/* these are used only here, not in qib_iba7220.c */
+#define kr_ibsd_epb_access_ctrl KREG_IDX(ibsd_epb_access_ctrl)
+#define kr_ibsd_epb_transaction_reg KREG_IDX(ibsd_epb_transaction_reg)
+#define kr_pciesd_epb_transaction_reg KREG_IDX(pciesd_epb_transaction_reg)
+#define kr_pciesd_epb_access_ctrl KREG_IDX(pciesd_epb_access_ctrl)
+#define kr_serdes_ddsrxeq0 KREG_IDX(SerDes_DDSRXEQ0)
 
 /*
  * The IBSerDesMappTable is a memory that holds values to be stored in
- * various SerDes registers by IBC. It is not part of the normal kregs
- * map and is used in exactly one place, hence the #define below.
+ * various SerDes registers by IBC.
  */
-#define KR_IBSerDesMappTable (0x94000 / (sizeof(uint64_t)))
+#define kr_serdes_maptable KREG_IDX(IBSerDesMappTable)
 
 /*
  * Below used for sdnum parameter, selecting one of the two sections
@@ -71,42 +95,44 @@
 #define EPB_GLOBAL_WR (1U << (EPB_ADDR_SHF + 8))
 
 /* Forward declarations. */
-static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
-				u32 data, u32 mask);
-static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val,
+static int qib_sd7220_reg_mod(struct qib_devdata *dd, int sdnum, u32 loc,
+			      u32 data, u32 mask);
+static int ibsd_mod_allchnls(struct qib_devdata *dd, int loc, int val,
 			     int mask);
-static int ipath_sd_trimdone_poll(struct ipath_devdata *dd);
-static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd,
-				      const char *where);
-static int ipath_sd_setvals(struct ipath_devdata *dd);
-static int ipath_sd_early(struct ipath_devdata *dd);
-static int ipath_sd_dactrim(struct ipath_devdata *dd);
-/* Set the registers that IBC may muck with to their default "preset" values */
-int ipath_sd7220_presets(struct ipath_devdata *dd);
-static int ipath_internal_presets(struct ipath_devdata *dd);
+static int qib_sd_trimdone_poll(struct qib_devdata *dd);
+static void qib_sd_trimdone_monitor(struct qib_devdata *dd, const char *where);
+static int qib_sd_setvals(struct qib_devdata *dd);
+static int qib_sd_early(struct qib_devdata *dd);
+static int qib_sd_dactrim(struct qib_devdata *dd);
+static int qib_internal_presets(struct qib_devdata *dd);
 /* Tweak the register (CMUCTRL5) that contains the TRIMSELF controls */
-static int ipath_sd_trimself(struct ipath_devdata *dd, int val);
-static int epb_access(struct ipath_devdata *dd, int sdnum, int claim);
-
-void ipath_set_relock_poll(struct ipath_devdata *dd, int ibup);
+static int qib_sd_trimself(struct qib_devdata *dd, int val);
+static int epb_access(struct qib_devdata *dd, int sdnum, int claim);
+static int qib_sd7220_ib_load(struct qib_devdata *dd,
+			      const struct firmware *fw);
+static int qib_sd7220_ib_vfy(struct qib_devdata *dd,
+			     const struct firmware *fw);
 
 /*
  * Below keeps track of whether the "once per power-on" initialization has
  * been done, because uC code Version 1.32.17 or higher allows the uC to
  * be reset at will, and Automatic Equalization may require it. So the
- * state of the reset "pin", as reflected in was_reset parameter to
- * ipath_sd7220_init() is no longer valid. Instead, we check for the
+ * state of the reset "pin", is no longer valid. Instead, we check for the
  * actual uC code having been loaded.
  */
-static int ipath_ibsd_ucode_loaded(struct ipath_devdata *dd)
+static int qib_ibsd_ucode_loaded(struct qib_pportdata *ppd,
+				 const struct firmware *fw)
 {
-	if (!dd->serdes_first_init_done && (ipath_sd7220_ib_vfy(dd) > 0))
-		dd->serdes_first_init_done = 1;
-	return dd->serdes_first_init_done;
+	struct qib_devdata *dd = ppd->dd;
+
+	if (!dd->cspec->serdes_first_init_done &&
+	    qib_sd7220_ib_vfy(dd, fw) > 0)
+		dd->cspec->serdes_first_init_done = 1;
+	return dd->cspec->serdes_first_init_done;
 }
 
-/* repeat #define for local use. "Real" #define is in ipath_iba7220.c */
-#define INFINIPATH_HWE_IB_UC_MEMORYPARITYERR      0x0000004000000000ULL
+/* repeat #define for local use. "Real" #define is in qib_iba7220.c */
+#define QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR      0x0000004000000000ULL
 #define IB_MPREG5 (EPB_LOC(6, 0, 0xE) | (1L << EPB_IB_UC_CS_SHF))
 #define IB_MPREG6 (EPB_LOC(6, 0, 0xF) | (1U << EPB_IB_UC_CS_SHF))
 #define UC_PAR_CLR_D 8
@@ -114,25 +140,25 @@ static int ipath_ibsd_ucode_loaded(struct ipath_devdata *dd)
 #define IB_CTRL2(chn) (EPB_LOC(chn, 7, 3) | EPB_IB_QUAD0_CS)
 #define START_EQ1(chan) EPB_LOC(chan, 7, 0x27)
 
-void ipath_sd7220_clr_ibpar(struct ipath_devdata *dd)
+void qib_sd7220_clr_ibpar(struct qib_devdata *dd)
 {
 	int ret;
 
 	/* clear, then re-enable parity errs */
-	ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6,
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6,
 		UC_PAR_CLR_D, UC_PAR_CLR_M);
 	if (ret < 0) {
-		ipath_dev_err(dd, "Failed clearing IBSerDes Parity err\n");
+		qib_dev_err(dd, "Failed clearing IBSerDes Parity err\n");
 		goto bail;
 	}
-	ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0,
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0,
 		UC_PAR_CLR_M);
 
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+	qib_read_kreg32(dd, kr_scratch);
 	udelay(4);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrclear,
-		INFINIPATH_HWE_IB_UC_MEMORYPARITYERR);
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+	qib_write_kreg(dd, kr_hwerrclear,
+		QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
+	qib_read_kreg32(dd, kr_scratch);
 bail:
 	return;
 }
@@ -146,7 +172,7 @@ bail:
 #define IB_PGUDP(chn) (EPB_LOC((chn), 2, 1) | EPB_IB_QUAD0_CS)
 #define IB_CMUDONE(chn) (EPB_LOC((chn), 7, 0xF) | EPB_IB_QUAD0_CS)
 
-static int ipath_resync_ibepb(struct ipath_devdata *dd)
+static int qib_resync_ibepb(struct qib_devdata *dd)
 {
 	int ret, pat, tries, chn;
 	u32 loc;
@@ -155,43 +181,42 @@ static int ipath_resync_ibepb(struct ipath_devdata *dd)
 	chn = 0;
 	for (tries = 0; tries < (4 * IBSD_RESYNC_TRIES); ++tries) {
 		loc = IB_PGUDP(chn);
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed read in resync\n");
+			qib_dev_err(dd, "Failed read in resync\n");
 			continue;
 		}
 		if (ret != 0xF0 && ret != 0x55 && tries == 0)
-			ipath_dev_err(dd, "unexpected pattern in resync\n");
+			qib_dev_err(dd, "unexpected pattern in resync\n");
 		pat = ret ^ 0xA5; /* alternate F0 and 55 */
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, pat, 0xFF);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, pat, 0xFF);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed write in resync\n");
+			qib_dev_err(dd, "Failed write in resync\n");
 			continue;
 		}
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed re-read in resync\n");
+			qib_dev_err(dd, "Failed re-read in resync\n");
 			continue;
 		}
 		if (ret != pat) {
-			ipath_dev_err(dd, "Failed compare1 in resync\n");
+			qib_dev_err(dd, "Failed compare1 in resync\n");
 			continue;
 		}
 		loc = IB_CMUDONE(chn);
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, 0, 0);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed CMUDONE rd in resync\n");
+			qib_dev_err(dd, "Failed CMUDONE rd in resync\n");
 			continue;
 		}
 		if ((ret & 0x70) != ((chn << 4) | 0x40)) {
-			ipath_dev_err(dd, "Bad CMUDONE value %02X, chn %d\n",
-				ret, chn);
+			qib_dev_err(dd, "Bad CMUDONE value %02X, chn %d\n",
+				    ret, chn);
 			continue;
 		}
 		if (++chn == 4)
 			break;  /* Success */
 	}
-	ipath_cdbg(VERBOSE, "Resync in %d tries\n", tries);
 	return (ret > 0) ? 0 : ret;
 }
 
@@ -199,32 +224,32 @@ static int ipath_resync_ibepb(struct ipath_devdata *dd)
  * Localize the stuff that should be done to change IB uC reset
  * returns <0 for errors.
  */
-static int ipath_ibsd_reset(struct ipath_devdata *dd, int assert_rst)
+static int qib_ibsd_reset(struct qib_devdata *dd, int assert_rst)
 {
 	u64 rst_val;
 	int ret = 0;
 	unsigned long flags;
 
-	rst_val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl);
+	rst_val = qib_read_kreg64(dd, kr_ibserdesctrl);
 	if (assert_rst) {
 		/*
 		 * Vendor recommends "interrupting" uC before reset, to
 		 * minimize possible glitches.
 		 */
-		spin_lock_irqsave(&dd->ipath_sdepb_lock, flags);
+		spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
 		epb_access(dd, IB_7220_SERDES, 1);
 		rst_val |= 1ULL;
 		/* Squelch possible parity error from _asserting_ reset */
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-			dd->ipath_hwerrmask &
-			~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR);
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, rst_val);
+		qib_write_kreg(dd, kr_hwerrmask,
+			       dd->cspec->hwerrmask &
+			       ~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
+		qib_write_kreg(dd, kr_ibserdesctrl, rst_val);
 		/* flush write, delay to ensure it took effect */
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+		qib_read_kreg32(dd, kr_scratch);
 		udelay(2);
 		/* once it's reset, can remove interrupt */
 		epb_access(dd, IB_7220_SERDES, -1);
-		spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags);
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
 	} else {
 		/*
 		 * Before we de-assert reset, we need to deal with
@@ -235,47 +260,47 @@ static int ipath_ibsd_reset(struct ipath_devdata *dd, int assert_rst)
 		 */
 		u64 val;
 		rst_val &= ~(1ULL);
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-			dd->ipath_hwerrmask &
-			~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR);
+		qib_write_kreg(dd, kr_hwerrmask,
+			       dd->cspec->hwerrmask &
+			       ~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR);
 
-		ret = ipath_resync_ibepb(dd);
+		ret = qib_resync_ibepb(dd);
 		if (ret < 0)
-			ipath_dev_err(dd, "unable to re-sync IB EPB\n");
+			qib_dev_err(dd, "unable to re-sync IB EPB\n");
 
 		/* set uC control regs to suppress parity errs */
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG5, 1, 1);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG5, 1, 1);
 		if (ret < 0)
 			goto bail;
 		/* IB uC code past Version 1.32.17 allow suppression of wdog */
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80,
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80,
 			0x80);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed to set WDOG disable\n");
+			qib_dev_err(dd, "Failed to set WDOG disable\n");
 			goto bail;
 		}
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, rst_val);
+		qib_write_kreg(dd, kr_ibserdesctrl, rst_val);
 		/* flush write, delay for startup */
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+		qib_read_kreg32(dd, kr_scratch);
 		udelay(1);
 		/* clear, then re-enable parity errs */
-		ipath_sd7220_clr_ibpar(dd);
-		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_hwerrstatus);
-		if (val & INFINIPATH_HWE_IB_UC_MEMORYPARITYERR) {
-			ipath_dev_err(dd, "IBUC Parity still set after RST\n");
-			dd->ipath_hwerrmask &=
-				~INFINIPATH_HWE_IB_UC_MEMORYPARITYERR;
+		qib_sd7220_clr_ibpar(dd);
+		val = qib_read_kreg64(dd, kr_hwerrstatus);
+		if (val & QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR) {
+			qib_dev_err(dd, "IBUC Parity still set after RST\n");
+			dd->cspec->hwerrmask &=
+				~QLOGIC_IB_HWE_IB_UC_MEMORYPARITYERR;
 		}
-		ipath_write_kreg(dd, dd->ipath_kregs->kr_hwerrmask,
-			dd->ipath_hwerrmask);
+		qib_write_kreg(dd, kr_hwerrmask,
+			dd->cspec->hwerrmask);
 	}
 
 bail:
 	return ret;
 }
 
-static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd,
-       const char *where)
+static void qib_sd_trimdone_monitor(struct qib_devdata *dd,
+	const char *where)
 {
 	int ret, chn, baduns;
 	u64 val;
@@ -286,70 +311,74 @@ static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd,
 	/* give time for reset to settle out in EPB */
 	udelay(2);
 
-	ret = ipath_resync_ibepb(dd);
+	ret = qib_resync_ibepb(dd);
 	if (ret < 0)
-		ipath_dev_err(dd, "not able to re-sync IB EPB (%s)\n", where);
+		qib_dev_err(dd, "not able to re-sync IB EPB (%s)\n", where);
 
 	/* Do "sacrificial read" to get EPB in sane state after reset */
-	ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_CTRL2(0), 0, 0);
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_CTRL2(0), 0, 0);
 	if (ret < 0)
-		ipath_dev_err(dd, "Failed TRIMDONE 1st read, (%s)\n", where);
+		qib_dev_err(dd, "Failed TRIMDONE 1st read, (%s)\n", where);
 
 	/* Check/show "summary" Trim-done bit in IBCStatus */
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
-	if (val & (1ULL << 11))
-		ipath_cdbg(VERBOSE, "IBCS TRIMDONE set (%s)\n", where);
-	else
-		ipath_dev_err(dd, "IBCS TRIMDONE clear (%s)\n", where);
-
+	val = qib_read_kreg64(dd, kr_ibcstatus);
+	if (!(val & (1ULL << 11)))
+		qib_dev_err(dd, "IBCS TRIMDONE clear (%s)\n", where);
+	/*
+	 * Do "dummy read/mod/wr" to get EPB in sane state after reset
+	 * The default value for MPREG6 is 0.
+	 */
 	udelay(2);
 
-	ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80, 0x80);
+	ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, IB_MPREG6, 0x80, 0x80);
 	if (ret < 0)
-		ipath_dev_err(dd, "Failed Dummy RMW, (%s)\n", where);
+		qib_dev_err(dd, "Failed Dummy RMW, (%s)\n", where);
 	udelay(10);
 
 	baduns = 0;
 
 	for (chn = 3; chn >= 0; --chn) {
 		/* Read CTRL reg for each channel to check TRIMDONE */
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
 			IB_CTRL2(chn), 0, 0);
 		if (ret < 0)
-			ipath_dev_err(dd, "Failed checking TRIMDONE, chn %d"
-				" (%s)\n", chn, where);
+			qib_dev_err(dd,
+				"Failed checking TRIMDONE, chn %d (%s)\n",
+				chn, where);
 
 		if (!(ret & 0x10)) {
 			int probe;
+
 			baduns |= (1 << chn);
-			ipath_dev_err(dd, "TRIMDONE cleared on chn %d (%02X)."
-				" (%s)\n", chn, ret, where);
-			probe = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
+			qib_dev_err(dd,
+				"TRIMDONE cleared on chn %d (%02X). (%s)\n",
+				chn, ret, where);
+			probe = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
 				IB_PGUDP(0), 0, 0);
-			ipath_dev_err(dd, "probe is %d (%02X)\n",
+			qib_dev_err(dd, "probe is %d (%02X)\n",
 				probe, probe);
-			probe = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
+			probe = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
 				IB_CTRL2(chn), 0, 0);
-			ipath_dev_err(dd, "re-read: %d (%02X)\n",
+			qib_dev_err(dd, "re-read: %d (%02X)\n",
 				probe, probe);
-			ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
 				IB_CTRL2(chn), 0x10, 0x10);
 			if (ret < 0)
-				ipath_dev_err(dd,
+				qib_dev_err(dd,
 					"Err on TRIMDONE rewrite1\n");
 		}
 	}
 	for (chn = 3; chn >= 0; --chn) {
 		/* Read CTRL reg for each channel to check TRIMDONE */
 		if (baduns & (1 << chn)) {
-			ipath_dev_err(dd,
-				"Reseting TRIMDONE on chn %d (%s)\n",
+			qib_dev_err(dd,
+				"Resetting TRIMDONE on chn %d (%s)\n",
 				chn, where);
-			ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
 				IB_CTRL2(chn), 0x10, 0x10);
 			if (ret < 0)
-				ipath_dev_err(dd, "Failed re-setting "
-					"TRIMDONE, chn %d (%s)\n",
+				qib_dev_err(dd,
+					"Failed re-setting TRIMDONE, chn %d (%s)\n",
 					chn, where);
 		}
 	}
@@ -361,96 +390,94 @@ static void ipath_sd_trimdone_monitor(struct ipath_devdata *dd,
  * Post IB uC code version 1.32.17, was_reset being 1 is not really
  * informative, so we double-check.
  */
-int ipath_sd7220_init(struct ipath_devdata *dd, int was_reset)
+int qib_sd7220_init(struct qib_devdata *dd)
 {
+	const struct firmware *fw;
 	int ret = 1; /* default to failure */
-	int first_reset;
-	int val_stat;
+	int first_reset, was_reset;
 
+	/* SERDES MPU reset recorded in D0 */
+	was_reset = (qib_read_kreg64(dd, kr_ibserdesctrl) & 1);
 	if (!was_reset) {
 		/* entered with reset not asserted, we need to do it */
-		ipath_ibsd_reset(dd, 1);
-		ipath_sd_trimdone_monitor(dd, "Driver-reload");
+		qib_ibsd_reset(dd, 1);
+		qib_sd_trimdone_monitor(dd, "Driver-reload");
 	}
 
-	/* Substitute our deduced value for was_reset */
-	ret = ipath_ibsd_ucode_loaded(dd);
-	if (ret < 0) {
-		ret = 1;
+	ret = request_firmware(&fw, SD7220_FW_NAME, &dd->pcidev->dev);
+	if (ret) {
+		qib_dev_err(dd, "Failed to load IB SERDES image\n");
 		goto done;
 	}
-	first_reset = !ret; /* First reset if IBSD uCode not yet loaded */
 
+	/* Substitute our deduced value for was_reset */
+	ret = qib_ibsd_ucode_loaded(dd->pport, fw);
+	if (ret < 0)
+		goto bail;
+
+	first_reset = !ret; /* First reset if IBSD uCode not yet loaded */
 	/*
 	 * Alter some regs per vendor latest doc, reset-defaults
 	 * are not right for IB.
 	 */
-	ret = ipath_sd_early(dd);
+	ret = qib_sd_early(dd);
 	if (ret < 0) {
-		ipath_dev_err(dd, "Failed to set IB SERDES early defaults\n");
-		ret = 1;
-		goto done;
+		qib_dev_err(dd, "Failed to set IB SERDES early defaults\n");
+		goto bail;
 	}
-
 	/*
 	 * Set DAC manual trim IB.
 	 * We only do this once after chip has been reset (usually
 	 * same as once per system boot).
 	 */
 	if (first_reset) {
-		ret = ipath_sd_dactrim(dd);
+		ret = qib_sd_dactrim(dd);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed IB SERDES DAC trim\n");
-			ret = 1;
-			goto done;
+			qib_dev_err(dd, "Failed IB SERDES DAC trim\n");
+			goto bail;
 		}
 	}
-
 	/*
 	 * Set various registers (DDS and RXEQ) that will be
 	 * controlled by IBC (in 1.2 mode) to reasonable preset values
 	 * Calling the "internal" version avoids the "check for needed"
 	 * and "trimdone monitor" that might be counter-productive.
 	 */
-	ret = ipath_internal_presets(dd);
+	ret = qib_internal_presets(dd);
 	if (ret < 0) {
-		ipath_dev_err(dd, "Failed to set IB SERDES presets\n");
-		ret = 1;
-		goto done;
+		qib_dev_err(dd, "Failed to set IB SERDES presets\n");
+		goto bail;
 	}
-	ret = ipath_sd_trimself(dd, 0x80);
+	ret = qib_sd_trimself(dd, 0x80);
 	if (ret < 0) {
-		ipath_dev_err(dd, "Failed to set IB SERDES TRIMSELF\n");
-		ret = 1;
-		goto done;
+		qib_dev_err(dd, "Failed to set IB SERDES TRIMSELF\n");
+		goto bail;
 	}
 
 	/* Load image, then try to verify */
-	ret = 0;	/* Assume success */
+	ret = 0;        /* Assume success */
 	if (first_reset) {
 		int vfy;
 		int trim_done;
-		ipath_dbg("SerDes uC was reset, reloading PRAM\n");
-		ret = ipath_sd7220_ib_load(dd);
+
+		ret = qib_sd7220_ib_load(dd, fw);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed to load IB SERDES image\n");
-			ret = 1;
-			goto done;
-		}
+			qib_dev_err(dd, "Failed to load IB SERDES image\n");
+			goto bail;
+		} else {
+			/* Loaded image, try to verify */
+			vfy = qib_sd7220_ib_vfy(dd, fw);
+			if (vfy != ret) {
+				qib_dev_err(dd, "SERDES PRAM VFY failed\n");
+				goto bail;
+			} /* end if verified */
+		} /* end if loaded */
 
-		/* Loaded image, try to verify */
-		vfy = ipath_sd7220_ib_vfy(dd);
-		if (vfy != ret) {
-			ipath_dev_err(dd, "SERDES PRAM VFY failed\n");
-			ret = 1;
-			goto done;
-		}
 		/*
 		 * Loaded and verified. Almost good...
 		 * hold "success" in ret
 		 */
 		ret = 0;
-
 		/*
 		 * Prev steps all worked, continue bringup
 		 * De-assert RESET to uC, only in first reset, to allow
@@ -461,45 +488,49 @@ int ipath_sd7220_init(struct ipath_devdata *dd, int was_reset)
 		 */
 		ret = ibsd_mod_allchnls(dd, START_EQ1(0), 0, 0x38);
 		if (ret < 0) {
-			ipath_dev_err(dd, "Failed clearing START_EQ1\n");
-			ret = 1;
-			goto done;
+			qib_dev_err(dd, "Failed clearing START_EQ1\n");
+			goto bail;
 		}
 
-		ipath_ibsd_reset(dd, 0);
+		qib_ibsd_reset(dd, 0);
 		/*
 		 * If this is not the first reset, trimdone should be set
-		 * already.
+		 * already. We may need to check about this.
 		 */
-		trim_done = ipath_sd_trimdone_poll(dd);
+		trim_done = qib_sd_trimdone_poll(dd);
 		/*
 		 * Whether or not trimdone succeeded, we need to put the
 		 * uC back into reset to avoid a possible fight with the
 		 * IBC state-machine.
 		 */
-		ipath_ibsd_reset(dd, 1);
+		qib_ibsd_reset(dd, 1);
 
 		if (!trim_done) {
-			ipath_dev_err(dd, "No TRIMDONE seen\n");
-			ret = 1;
-			goto done;
+			qib_dev_err(dd, "No TRIMDONE seen\n");
+			goto bail;
 		}
-
-		ipath_sd_trimdone_monitor(dd, "First-reset");
+		/*
+		 * DEBUG: check each time we reset if trimdone bits have
+		 * gotten cleared, and re-set them.
+		 */
+		qib_sd_trimdone_monitor(dd, "First-reset");
 		/* Remember so we do not re-do the load, dactrim, etc. */
-		dd->serdes_first_init_done = 1;
+		dd->cspec->serdes_first_init_done = 1;
 	}
 	/*
-	 * Setup for channel training and load values for
+	 * setup for channel training and load values for
 	 * RxEq and DDS in tables used by IBC in IB1.2 mode
 	 */
-
-	val_stat = ipath_sd_setvals(dd);
-	if (val_stat < 0)
-		ret = 1;
+	ret = 0;
+	if (qib_sd_setvals(dd) >= 0)
+		goto done;
+bail:
+	ret = 1;
 done:
 	/* start relock timer regardless, but start at 1 second */
-	ipath_set_relock_poll(dd, -1);
+	set_7220_relock_poll(dd, -1);
+
+	release_firmware(fw);
 	return ret;
 }
 
@@ -517,7 +548,7 @@ done:
  * the "claim" parameter is >0 to claim, <0 to release, 0 to query.
  * Returns <0 for errors, >0 if we had ownership, else 0.
  */
-static int epb_access(struct ipath_devdata *dd, int sdnum, int claim)
+static int epb_access(struct qib_devdata *dd, int sdnum, int claim)
 {
 	u16 acc;
 	u64 accval;
@@ -525,28 +556,30 @@ static int epb_access(struct ipath_devdata *dd, int sdnum, int claim)
 	u64 oct_sel = 0;
 
 	switch (sdnum) {
-	case IB_7220_SERDES :
+	case IB_7220_SERDES:
 		/*
 		 * The IB SERDES "ownership" is fairly simple. A single each
 		 * request/grant.
 		 */
-		acc = dd->ipath_kregs->kr_ib_epbacc;
+		acc = kr_ibsd_epb_access_ctrl;
 		break;
-	case PCIE_SERDES0 :
-	case PCIE_SERDES1 :
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
 		/* PCIe SERDES has two "octants", need to select which */
-		acc = dd->ipath_kregs->kr_pcie_epbacc;
+		acc = kr_pciesd_epb_access_ctrl;
 		oct_sel = (2 << (sdnum - PCIE_SERDES0));
 		break;
-	default :
+
+	default:
 		return 0;
 	}
 
 	/* Make sure any outstanding transaction was seen */
-	ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+	qib_read_kreg32(dd, kr_scratch);
 	udelay(15);
 
-	accval = ipath_read_kreg32(dd, acc);
+	accval = qib_read_kreg32(dd, acc);
 
 	owned = !!(accval & EPB_ACC_GNT);
 	if (claim < 0) {
@@ -557,22 +590,22 @@ static int epb_access(struct ipath_devdata *dd, int sdnum, int claim)
 		 * Both should be clear
 		 */
 		u64 newval = 0;
-		ipath_write_kreg(dd, acc, newval);
+		qib_write_kreg(dd, acc, newval);
 		/* First read after write is not trustworthy */
-		pollval = ipath_read_kreg32(dd, acc);
+		pollval = qib_read_kreg32(dd, acc);
 		udelay(5);
-		pollval = ipath_read_kreg32(dd, acc);
+		pollval = qib_read_kreg32(dd, acc);
 		if (pollval & EPB_ACC_GNT)
 			owned = -1;
 	} else if (claim > 0) {
 		/* Need to claim */
 		u64 pollval;
 		u64 newval = EPB_ACC_REQ | oct_sel;
-		ipath_write_kreg(dd, acc, newval);
+		qib_write_kreg(dd, acc, newval);
 		/* First read after write is not trustworthy */
-		pollval = ipath_read_kreg32(dd, acc);
+		pollval = qib_read_kreg32(dd, acc);
 		udelay(5);
-		pollval = ipath_read_kreg32(dd, acc);
+		pollval = qib_read_kreg32(dd, acc);
 		if (!(pollval & EPB_ACC_GNT))
 			owned = -1;
 	}
@@ -582,18 +615,17 @@ static int epb_access(struct ipath_devdata *dd, int sdnum, int claim)
 /*
  * Lemma to deal with race condition of write..read to epb regs
  */
-static int epb_trans(struct ipath_devdata *dd, u16 reg, u64 i_val, u64 *o_vp)
+static int epb_trans(struct qib_devdata *dd, u16 reg, u64 i_val, u64 *o_vp)
 {
 	int tries;
 	u64 transval;
 
-
-	ipath_write_kreg(dd, reg, i_val);
+	qib_write_kreg(dd, reg, i_val);
 	/* Throw away first read, as RDY bit may be stale */
-	transval = ipath_read_kreg64(dd, reg);
+	transval = qib_read_kreg64(dd, reg);
 
 	for (tries = EPB_TRANS_TRIES; tries; --tries) {
-		transval = ipath_read_kreg32(dd, reg);
+		transval = qib_read_kreg32(dd, reg);
 		if (transval & EPB_TRANS_RDY)
 			break;
 		udelay(5);
@@ -606,9 +638,8 @@ static int epb_trans(struct ipath_devdata *dd, u16 reg, u64 i_val, u64 *o_vp)
 }
 
 /**
- *
- * ipath_sd7220_reg_mod - modify SERDES register
- * @dd: the infinipath device
+ * qib_sd7220_reg_mod - modify SERDES register
+ * @dd: the qlogic_ib device
  * @sdnum: which SERDES to access
  * @loc: location - channel, element, register, as packed by EPB_LOC() macro.
  * @wd: Write Data - value to set in register
@@ -619,8 +650,8 @@ static int epb_trans(struct ipath_devdata *dd, u16 reg, u64 i_val, u64 *o_vp)
  * returns current (presumed, if a write was done) contents of selected
  * register, or <0 if errors.
  */
-static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
-				u32 wd, u32 mask)
+static int qib_sd7220_reg_mod(struct qib_devdata *dd, int sdnum, u32 loc,
+			      u32 wd, u32 mask)
 {
 	u16 trans;
 	u64 transval;
@@ -629,14 +660,16 @@ static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
 	unsigned long flags;
 
 	switch (sdnum) {
-	case IB_7220_SERDES :
-		trans = dd->ipath_kregs->kr_ib_epbtrans;
+	case IB_7220_SERDES:
+		trans = kr_ibsd_epb_transaction_reg;
 		break;
-	case PCIE_SERDES0 :
-	case PCIE_SERDES1 :
-		trans = dd->ipath_kregs->kr_pcie_epbtrans;
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
+		trans = kr_pciesd_epb_transaction_reg;
 		break;
-	default :
+
+	default:
 		return -1;
 	}
 
@@ -644,23 +677,23 @@ static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
 	 * All access is locked in software (vs other host threads) and
 	 * hardware (vs uC access).
 	 */
-	spin_lock_irqsave(&dd->ipath_sdepb_lock, flags);
+	spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
 
 	owned = epb_access(dd, sdnum, 1);
 	if (owned < 0) {
-		spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags);
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
 		return -1;
 	}
 	ret = 0;
 	for (tries = EPB_TRANS_TRIES; tries; --tries) {
-		transval = ipath_read_kreg32(dd, trans);
+		transval = qib_read_kreg32(dd, trans);
 		if (transval & EPB_TRANS_RDY)
 			break;
 		udelay(5);
 	}
 
 	if (tries > 0) {
-		tries = 1;	/* to make read-skip work */
+		tries = 1;      /* to make read-skip work */
 		if (mask != 0xFF) {
 			/*
 			 * Not a pure write, so need to read.
@@ -688,7 +721,7 @@ static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
 	else
 		ret = transval & EPB_DATA_MASK;
 
-	spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags);
+	spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
 	if (tries <= 0)
 		ret = -1;
 	return ret;
@@ -707,7 +740,7 @@ static int ipath_sd7220_reg_mod(struct ipath_devdata *dd, int sdnum, u32 loc,
 #define EPB_RAMDATA EPB_LOC(6, 0, 5)
 
 /* Transfer date to/from uC Program RAM of IB or PCIe SerDes */
-static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
+static int qib_sd7220_ram_xfer(struct qib_devdata *dd, int sdnum, u32 loc,
 			       u8 *buf, int cnt, int rd_notwr)
 {
 	u16 trans;
@@ -723,29 +756,28 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 
 	/* Pick appropriate transaction reg and "Chip select" for this serdes */
 	switch (sdnum) {
-	case IB_7220_SERDES :
+	case IB_7220_SERDES:
 		csbit = 1ULL << EPB_IB_UC_CS_SHF;
-		trans = dd->ipath_kregs->kr_ib_epbtrans;
+		trans = kr_ibsd_epb_transaction_reg;
 		break;
-	case PCIE_SERDES0 :
-	case PCIE_SERDES1 :
+
+	case PCIE_SERDES0:
+	case PCIE_SERDES1:
 		/* PCIe SERDES has uC "chip select" in different bit, too */
 		csbit = 1ULL << EPB_PCIE_UC_CS_SHF;
-		trans = dd->ipath_kregs->kr_pcie_epbtrans;
+		trans = kr_pciesd_epb_transaction_reg;
 		break;
-	default :
+
+	default:
 		return -1;
 	}
 
 	op = rd_notwr ? "Rd" : "Wr";
-	spin_lock_irqsave(&dd->ipath_sdepb_lock, flags);
+	spin_lock_irqsave(&dd->cspec->sdepb_lock, flags);
 
 	owned = epb_access(dd, sdnum, 1);
 	if (owned < 0) {
-		spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags);
-		ipath_dbg("Could not get %s access to %s EPB: %X, loc %X\n",
-			op, (sdnum == IB_7220_SERDES) ? "IB" : "PCIe",
-			owned, loc);
+		spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
 		return -1;
 	}
 
@@ -758,16 +790,14 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 	 */
 	addr = loc & 0x1FFF;
 	for (tries = EPB_TRANS_TRIES; tries; --tries) {
-		transval = ipath_read_kreg32(dd, trans);
+		transval = qib_read_kreg32(dd, trans);
 		if (transval & EPB_TRANS_RDY)
 			break;
 		udelay(5);
 	}
 
 	sofar = 0;
-	if (tries <= 0)
-		ipath_dbg("No initial RDY on EPB access request\n");
-	else {
+	if (tries > 0) {
 		/*
 		 * Every "memory" access is doubly-indirect.
 		 * We set two bytes of address, then read/write
@@ -778,8 +808,6 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 		transval = csbit | EPB_UC_CTL |
 			(rd_notwr ? EPB_ROM_R : EPB_ROM_W);
 		tries = epb_trans(dd, trans, transval, &transval);
-		if (tries <= 0)
-			ipath_dbg("No EPB response to uC %s cmd\n", op);
 		while (tries > 0 && sofar < cnt) {
 			if (!sofar) {
 				/* Only set address at start of chunk */
@@ -787,18 +815,14 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 				transval = csbit | EPB_MADDRH | addrbyte;
 				tries = epb_trans(dd, trans, transval,
 						  &transval);
-				if (tries <= 0) {
-					ipath_dbg("No EPB response ADDRH\n");
+				if (tries <= 0)
 					break;
-				}
 				addrbyte = (addr + sofar) & 0xFF;
 				transval = csbit | EPB_MADDRL | addrbyte;
 				tries = epb_trans(dd, trans, transval,
 						 &transval);
-				if (tries <= 0) {
-					ipath_dbg("No EPB response ADDRL\n");
+				if (tries <= 0)
 					break;
-				}
 			}
 
 			if (rd_notwr)
@@ -806,10 +830,8 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 			else
 				transval = csbit | EPB_ROMDATA | buf[sofar];
 			tries = epb_trans(dd, trans, transval, &transval);
-			if (tries <= 0) {
-				ipath_dbg("No EPB response DATA\n");
+			if (tries <= 0)
 				break;
-			}
 			if (rd_notwr)
 				buf[sofar] = transval & EPB_DATA_MASK;
 			++sofar;
@@ -817,8 +839,6 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 		/* Finally, clear control-bit for Read or Write */
 		transval = csbit | EPB_UC_CTL;
 		tries = epb_trans(dd, trans, transval, &transval);
-		if (tries <= 0)
-			ipath_dbg("No EPB response to drop of uC %s cmd\n", op);
 	}
 
 	ret = sofar;
@@ -826,18 +846,16 @@ static int ipath_sd7220_ram_xfer(struct ipath_devdata *dd, int sdnum, u32 loc,
 	if (epb_access(dd, sdnum, -1) < 0)
 		ret = -1;
 
-	spin_unlock_irqrestore(&dd->ipath_sdepb_lock, flags);
-	if (tries <= 0) {
-		ipath_dbg("SERDES PRAM %s failed after %d bytes\n", op, sofar);
+	spin_unlock_irqrestore(&dd->cspec->sdepb_lock, flags);
+	if (tries <= 0)
 		ret = -1;
-	}
 	return ret;
 }
 
 #define PROG_CHUNK 64
 
-int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum,
-	u8 *img, int len, int offset)
+static int qib_sd7220_prog_ld(struct qib_devdata *dd, int sdnum,
+			      const u8 *img, int len, int offset)
 {
 	int cnt, sofar, req;
 
@@ -846,8 +864,8 @@ int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum,
 		req = len - sofar;
 		if (req > PROG_CHUNK)
 			req = PROG_CHUNK;
-		cnt = ipath_sd7220_ram_xfer(dd, sdnum, offset + sofar,
-					  img + sofar, req, 0);
+		cnt = qib_sd7220_ram_xfer(dd, sdnum, offset + sofar,
+					  (u8 *)img + sofar, req, 0);
 		if (cnt < req) {
 			sofar = -1;
 			break;
@@ -860,8 +878,8 @@ int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum,
 #define VFY_CHUNK 64
 #define SD_PRAM_ERROR_LIMIT 42
 
-int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum,
-	const u8 *img, int len, int offset)
+static int qib_sd7220_prog_vfy(struct qib_devdata *dd, int sdnum,
+			       const u8 *img, int len, int offset)
 {
 	int cnt, sofar, req, idx, errors;
 	unsigned char readback[VFY_CHUNK];
@@ -872,7 +890,7 @@ int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum,
 		req = len - sofar;
 		if (req > VFY_CHUNK)
 			req = VFY_CHUNK;
-		cnt = ipath_sd7220_ram_xfer(dd, sdnum, sofar + offset,
+		cnt = qib_sd7220_ram_xfer(dd, sdnum, sofar + offset,
 					  readback, req, 1);
 		if (cnt < req) {
 			/* failed in read itself */
@@ -888,11 +906,25 @@ int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum,
 	return errors ? -errors : sofar;
 }
 
-/* IRQ not set up at this point in init, so we poll. */
+static int
+qib_sd7220_ib_load(struct qib_devdata *dd, const struct firmware *fw)
+{
+	return qib_sd7220_prog_ld(dd, IB_7220_SERDES, fw->data, fw->size, 0);
+}
+
+static int
+qib_sd7220_ib_vfy(struct qib_devdata *dd, const struct firmware *fw)
+{
+	return qib_sd7220_prog_vfy(dd, IB_7220_SERDES, fw->data, fw->size, 0);
+}
+
+/*
+ * IRQ not set up at this point in init, so we poll.
+ */
 #define IB_SERDES_TRIM_DONE (1ULL << 11)
 #define TRIM_TMO (30)
 
-static int ipath_sd_trimdone_poll(struct ipath_devdata *dd)
+static int qib_sd_trimdone_poll(struct qib_devdata *dd)
 {
 	int trim_tmo, ret;
 	uint64_t val;
@@ -903,16 +935,15 @@ static int ipath_sd_trimdone_poll(struct ipath_devdata *dd)
 	 */
 	ret = 0;
 	for (trim_tmo = 0; trim_tmo < TRIM_TMO; ++trim_tmo) {
-		val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
+		val = qib_read_kreg64(dd, kr_ibcstatus);
 		if (val & IB_SERDES_TRIM_DONE) {
-			ipath_cdbg(VERBOSE, "TRIMDONE after %d\n", trim_tmo);
 			ret = 1;
 			break;
 		}
 		msleep(10);
 	}
 	if (trim_tmo >= TRIM_TMO) {
-		ipath_dev_err(dd, "No TRIMDONE in %d tries\n", trim_tmo);
+		qib_dev_err(dd, "No TRIMDONE in %d tries\n", trim_tmo);
 		ret = 0;
 	}
 	return ret;
@@ -964,8 +995,7 @@ static struct dds_init {
 };
 
 /*
- * Next, values related to Receive Equalization.
- * In comments, FDR (Full) is IB DDR, HDR (Half) is IB SDR
+ * Now the RXEQ section of the table.
  */
 /* Hardware packs an element number and register address thus: */
 #define RXEQ_INIT_RDESC(elt, addr) (((elt) & 0xF) | ((addr) << 4))
@@ -981,23 +1011,23 @@ static struct dds_init {
 #define RXEQ_SDR_ZCNT 23
 
 static struct rxeq_init {
-	u16 rdesc;	/* in form used in SerDesDDSRXEQ */
+	u16 rdesc;      /* in form used in SerDesDDSRXEQ */
 	u8  rdata[4];
 } rxeq_init_vals[] = {
 	/* Set Rcv Eq. to Preset node */
 	RXEQ_VAL_ALL(7, 0x27, 0x10),
 	/* Set DFELTHFDR/HDR thresholds */
-	RXEQ_VAL(7, 8,    0, 0, 0, 0), /* FDR */
+	RXEQ_VAL(7, 8,    0, 0, 0, 0), /* FDR, was 0, 1, 2, 3 */
 	RXEQ_VAL(7, 0x21, 0, 0, 0, 0), /* HDR */
 	/* Set TLTHFDR/HDR theshold */
-	RXEQ_VAL(7, 9,    2, 2, 2, 2), /* FDR */
-	RXEQ_VAL(7, 0x23, 2, 2, 2, 2), /* HDR */
+	RXEQ_VAL(7, 9,    2, 2, 2, 2), /* FDR, was 0, 2, 4, 6 */
+	RXEQ_VAL(7, 0x23, 2, 2, 2, 2), /* HDR, was  0, 1, 2, 3 */
 	/* Set Preamp setting 2 (ZFR/ZCNT) */
-	RXEQ_VAL(7, 0x1B, 12, 12, 12, 12), /* FDR */
-	RXEQ_VAL(7, 0x1C, 12, 12, 12, 12), /* HDR */
+	RXEQ_VAL(7, 0x1B, 12, 12, 12, 12), /* FDR, was 12, 16, 20, 24 */
+	RXEQ_VAL(7, 0x1C, 12, 12, 12, 12), /* HDR, was 12, 16, 20, 24 */
 	/* Set Preamp DC gain and Setting 1 (GFR/GHR) */
-	RXEQ_VAL(7, 0x1E, 0x10, 0x10, 0x10, 0x10), /* FDR */
-	RXEQ_VAL(7, 0x1F, 0x10, 0x10, 0x10, 0x10), /* HDR */
+	RXEQ_VAL(7, 0x1E, 16, 16, 16, 16), /* FDR, was 16, 17, 18, 20 */
+	RXEQ_VAL(7, 0x1F, 16, 16, 16, 16), /* HDR, was 16, 17, 18, 20 */
 	/* Toggle RELOCK (in VCDL_CTRL0) to lock to data */
 	RXEQ_VAL_ALL(6, 6, 0x20), /* Set D5 High */
 	RXEQ_VAL_ALL(6, 6, 0), /* Set D5 Low */
@@ -1007,27 +1037,27 @@ static struct rxeq_init {
 #define DDS_ROWS (16)
 #define RXEQ_ROWS ARRAY_SIZE(rxeq_init_vals)
 
-static int ipath_sd_setvals(struct ipath_devdata *dd)
+static int qib_sd_setvals(struct qib_devdata *dd)
 {
 	int idx, midx;
-	int min_idx;	 /* Minimum index for this portion of table */
+	int min_idx;     /* Minimum index for this portion of table */
 	uint32_t dds_reg_map;
 	u64 __iomem *taddr, *iaddr;
 	uint64_t data;
 	uint64_t sdctl;
 
-	taddr = dd->ipath_kregbase + KR_IBSerDesMappTable;
-	iaddr = dd->ipath_kregbase + dd->ipath_kregs->kr_ib_ddsrxeq;
+	taddr = dd->kregbase + kr_serdes_maptable;
+	iaddr = dd->kregbase + kr_serdes_ddsrxeq0;
 
 	/*
 	 * Init the DDS section of the table.
 	 * Each "row" of the table provokes NUM_DDS_REG writes, to the
 	 * registers indicated in DDS_REG_MAP.
 	 */
-	sdctl = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibserdesctrl);
+	sdctl = qib_read_kreg64(dd, kr_ibserdesctrl);
 	sdctl = (sdctl & ~(0x1f << 8)) | (NUM_DDS_REGS << 8);
 	sdctl = (sdctl & ~(0x1f << 13)) | (RXEQ_ROWS << 13);
-	ipath_write_kreg(dd, dd->ipath_kregs->kr_ibserdesctrl, sdctl);
+	qib_write_kreg(dd, kr_ibserdesctrl, sdctl);
 
 	/*
 	 * Iterate down table within loop for each register to store.
@@ -1037,21 +1067,21 @@ static int ipath_sd_setvals(struct ipath_devdata *dd)
 		data = ((dds_reg_map & 0xF) << 4) | TX_FAST_ELT;
 		writeq(data, iaddr + idx);
 		mmiowb();
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+		qib_read_kreg32(dd, kr_scratch);
 		dds_reg_map >>= 4;
 		for (midx = 0; midx < DDS_ROWS; ++midx) {
 			u64 __iomem *daddr = taddr + ((midx << 4) + idx);
 			data = dds_init_vals[midx].reg_vals[idx];
 			writeq(data, daddr);
 			mmiowb();
-			ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+			qib_read_kreg32(dd, kr_scratch);
 		} /* End inner for (vals for this reg, each row) */
 	} /* end outer for (regs to be stored) */
 
 	/*
-	 * Init the RXEQ section of the table. As explained above the table
-	 * rxeq_init_vals[], this runs in a different order, as the pattern
-	 * of register references is more complex, but there are only
+	 * Init the RXEQ section of the table.
+	 * This runs in a different order, as the pattern of
+	 * register references is more complex, but there are only
 	 * four "data" values per register.
 	 */
 	min_idx = idx; /* RXEQ indices pick up where DDS left off */
@@ -1066,13 +1096,13 @@ static int ipath_sd_setvals(struct ipath_devdata *dd)
 		/* Store the next RXEQ register address */
 		writeq(rxeq_init_vals[idx].rdesc, iaddr + didx);
 		mmiowb();
-		ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+		qib_read_kreg32(dd, kr_scratch);
 		/* Iterate through RXEQ values */
 		for (vidx = 0; vidx < 4; vidx++) {
 			data = rxeq_init_vals[idx].rdata[vidx];
 			writeq(data, taddr + (vidx << 6) + idx);
 			mmiowb();
-			ipath_read_kreg32(dd, dd->ipath_kregs->kr_scratch);
+			qib_read_kreg32(dd, kr_scratch);
 		}
 	} /* end outer for (Reg-writes for RXEQ) */
 	return 0;
@@ -1085,33 +1115,18 @@ static int ipath_sd_setvals(struct ipath_devdata *dd)
 #define VCDL_CTRL2(chan) EPB_LOC(chan, 6, 8)
 #define START_EQ2(chan) EPB_LOC(chan, 7, 0x28)
 
-static int ibsd_sto_noisy(struct ipath_devdata *dd, int loc, int val, int mask)
-{
-	int ret = -1;
-	int sloc; /* shifted loc, for messages */
-
-	loc |= (1U << EPB_IB_QUAD0_CS_SHF);
-	sloc = loc >> EPB_ADDR_SHF;
-
-	ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, mask);
-	if (ret < 0)
-		ipath_dev_err(dd, "Write failed: elt %d,"
-			" addr 0x%X, chnl %d, val 0x%02X, mask 0x%02X\n",
-			(sloc & 0xF), (sloc >> 9) & 0x3f, (sloc >> 4) & 7,
-			val & 0xFF, mask & 0xFF);
-	return ret;
-}
-
 /*
  * Repeat a "store" across all channels of the IB SerDes.
  * Although nominally it inherits the "read value" of the last
  * channel it modified, the only really useful return is <0 for
  * failure, >= 0 for success. The parameter 'loc' is assumed to
- * be the location for the channel-0 copy of the register to
- * be modified.
+ * be the location in some channel of the register to be modified
+ * The caller can specify use of the "gang write" option of EPB,
+ * in which case we use the specified channel data for any fields
+ * not explicitely written.
  */
-static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val,
-	int mask)
+static int ibsd_mod_allchnls(struct qib_devdata *dd, int loc, int val,
+			     int mask)
 {
 	int ret = -1;
 	int chnl;
@@ -1126,23 +1141,26 @@ static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val,
 		loc |= (1U << EPB_IB_QUAD0_CS_SHF);
 		chnl = (loc >> (4 + EPB_ADDR_SHF)) & 7;
 		if (mask != 0xFF) {
-			ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES,
-				loc & ~EPB_GLOBAL_WR, 0, 0);
+			ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES,
+						 loc & ~EPB_GLOBAL_WR, 0, 0);
 			if (ret < 0) {
 				int sloc = loc >> EPB_ADDR_SHF;
-				ipath_dev_err(dd, "pre-read failed: elt %d,"
-					" addr 0x%X, chnl %d\n", (sloc & 0xF),
+
+				qib_dev_err(dd,
+					"pre-read failed: elt %d, addr 0x%X, chnl %d\n",
+					(sloc & 0xF),
 					(sloc >> 9) & 0x3f, chnl);
 				return ret;
 			}
 			val = (ret & ~mask) | (val & mask);
 		}
 		loc &=  ~(7 << (4+EPB_ADDR_SHF));
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF);
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF);
 		if (ret < 0) {
 			int sloc = loc >> EPB_ADDR_SHF;
-			ipath_dev_err(dd, "Global WR failed: elt %d,"
-				" addr 0x%X, val %02X\n",
+
+			qib_dev_err(dd,
+				"Global WR failed: elt %d, addr 0x%X, val %02X\n",
 				(sloc & 0xF), (sloc >> 9) & 0x3f, val);
 		}
 		return ret;
@@ -1151,14 +1169,14 @@ static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val,
 	loc &=  ~(7 << (4+EPB_ADDR_SHF));
 	loc |= (1U << EPB_IB_QUAD0_CS_SHF);
 	for (chnl = 0; chnl < 4; ++chnl) {
-		int cloc;
-		cloc = loc | (chnl << (4+EPB_ADDR_SHF));
-		ret = ipath_sd7220_reg_mod(dd, IB_7220_SERDES, cloc, val, mask);
+		int cloc = loc | (chnl << (4+EPB_ADDR_SHF));
+
+		ret = qib_sd7220_reg_mod(dd, IB_7220_SERDES, cloc, val, mask);
 		if (ret < 0) {
 			int sloc = loc >> EPB_ADDR_SHF;
-			ipath_dev_err(dd, "Write failed: elt %d,"
-				" addr 0x%X, chnl %d, val 0x%02X,"
-				" mask 0x%02X\n",
+
+			qib_dev_err(dd,
+				"Write failed: elt %d, addr 0x%X, chnl %d, val 0x%02X, mask 0x%02X\n",
 				(sloc & 0xF), (sloc >> 9) & 0x3f, chnl,
 				val & 0xFF, mask & 0xFF);
 			break;
@@ -1171,7 +1189,7 @@ static int ibsd_mod_allchnls(struct ipath_devdata *dd, int loc, int val,
  * Set the Tx values normally modified by IBC in IB1.2 mode to default
  * values, as gotten from first row of init table.
  */
-static int set_dds_vals(struct ipath_devdata *dd, struct dds_init *ddi)
+static int set_dds_vals(struct qib_devdata *dd, struct dds_init *ddi)
 {
 	int ret;
 	int idx, reg, data;
@@ -1194,7 +1212,7 @@ static int set_dds_vals(struct ipath_devdata *dd, struct dds_init *ddi)
  * Set the Rx values normally modified by IBC in IB1.2 mode to default
  * values, as gotten from selected column of init table.
  */
-static int set_rxeq_vals(struct ipath_devdata *dd, int vsel)
+static int set_rxeq_vals(struct qib_devdata *dd, int vsel)
 {
 	int ret;
 	int ridx;
@@ -1202,6 +1220,7 @@ static int set_rxeq_vals(struct ipath_devdata *dd, int vsel)
 
 	for (ridx = 0; ridx < cnt; ++ridx) {
 		int elt, reg, val, loc;
+
 		elt = rxeq_init_vals[ridx].rdesc & 0xF;
 		reg = rxeq_init_vals[ridx].rdesc >> 4;
 		loc = EPB_LOC(0, elt, reg);
@@ -1217,83 +1236,66 @@ static int set_rxeq_vals(struct ipath_devdata *dd, int vsel)
 /*
  * Set the default values (row 0) for DDR Driver Demphasis.
  * we do this initially and whenever we turn off IB-1.2
+ *
  * The "default" values for Rx equalization are also stored to
  * SerDes registers. Formerly (and still default), we used set 2.
  * For experimenting with cables and link-partners, we allow changing
  * that via a module parameter.
  */
-static unsigned ipath_rxeq_set = 2;
-module_param_named(rxeq_default_set, ipath_rxeq_set, uint,
-	S_IWUSR | S_IRUGO);
+static unsigned qib_rxeq_set = 2;
+module_param_named(rxeq_default_set, qib_rxeq_set, uint,
+		   S_IWUSR | S_IRUGO);
 MODULE_PARM_DESC(rxeq_default_set,
-	"Which set [0..3] of Rx Equalization values is default");
+		 "Which set [0..3] of Rx Equalization values is default");
 
-static int ipath_internal_presets(struct ipath_devdata *dd)
+static int qib_internal_presets(struct qib_devdata *dd)
 {
 	int ret = 0;
 
 	ret = set_dds_vals(dd, dds_init_vals + DDS_3M);
 
 	if (ret < 0)
-		ipath_dev_err(dd, "Failed to set default DDS values\n");
-	ret = set_rxeq_vals(dd, ipath_rxeq_set & 3);
+		qib_dev_err(dd, "Failed to set default DDS values\n");
+	ret = set_rxeq_vals(dd, qib_rxeq_set & 3);
 	if (ret < 0)
-		ipath_dev_err(dd, "Failed to set default RXEQ values\n");
+		qib_dev_err(dd, "Failed to set default RXEQ values\n");
 	return ret;
 }
 
-int ipath_sd7220_presets(struct ipath_devdata *dd)
+int qib_sd7220_presets(struct qib_devdata *dd)
 {
 	int ret = 0;
 
-	if (!dd->ipath_presets_needed)
+	if (!dd->cspec->presets_needed)
 		return ret;
-	dd->ipath_presets_needed = 0;
+	dd->cspec->presets_needed = 0;
 	/* Assert uC reset, so we don't clash with it. */
-	ipath_ibsd_reset(dd, 1);
+	qib_ibsd_reset(dd, 1);
 	udelay(2);
-	ipath_sd_trimdone_monitor(dd, "link-down");
+	qib_sd_trimdone_monitor(dd, "link-down");
 
-	ret = ipath_internal_presets(dd);
-return ret;
+	ret = qib_internal_presets(dd);
+	return ret;
 }
 
-static int ipath_sd_trimself(struct ipath_devdata *dd, int val)
+static int qib_sd_trimself(struct qib_devdata *dd, int val)
 {
-	return ibsd_sto_noisy(dd, CMUCTRL5, val, 0xFF);
+	int loc = CMUCTRL5 | (1U << EPB_IB_QUAD0_CS_SHF);
+
+	return qib_sd7220_reg_mod(dd, IB_7220_SERDES, loc, val, 0xFF);
 }
 
-static int ipath_sd_early(struct ipath_devdata *dd)
+static int qib_sd_early(struct qib_devdata *dd)
 {
-	int ret = -1; /* Default failed */
-	int chnl;
+	int ret;
 
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, RXHSCTRL0(chnl), 0xD4, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, VCDL_DAC2(chnl), 0x2D, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
-	/* more fine-tuning of what will be default */
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, VCDL_CTRL2(chnl), 3, 0xF);
-		if (ret < 0)
-			goto bail;
-	}
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, START_EQ1(chnl), 0x10, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, START_EQ2(chnl), 0x30, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
+	ret = ibsd_mod_allchnls(dd, RXHSCTRL0(0) | EPB_GLOBAL_WR, 0xD4, 0xFF);
+	if (ret < 0)
+		goto bail;
+	ret = ibsd_mod_allchnls(dd, START_EQ1(0) | EPB_GLOBAL_WR, 0x10, 0xFF);
+	if (ret < 0)
+		goto bail;
+	ret = ibsd_mod_allchnls(dd, START_EQ2(0) | EPB_GLOBAL_WR, 0x30, 0xFF);
 bail:
 	return ret;
 }
@@ -1302,50 +1304,53 @@ bail:
 #define LDOUTCTRL1(chnl) EPB_LOC(chnl, 7, 6)
 #define RXHSSTATUS(chnl) EPB_LOC(chnl, 6, 0xF)
 
-static int ipath_sd_dactrim(struct ipath_devdata *dd)
+static int qib_sd_dactrim(struct qib_devdata *dd)
 {
-	int ret = -1; /* Default failed */
-	int chnl;
+	int ret;
+
+	ret = ibsd_mod_allchnls(dd, VCDL_DAC2(0) | EPB_GLOBAL_WR, 0x2D, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	/* more fine-tuning of what will be default */
+	ret = ibsd_mod_allchnls(dd, VCDL_CTRL2(0), 3, 0xF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, BACTRL(0) | EPB_GLOBAL_WR, 0x40, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, LDOUTCTRL1(0) | EPB_GLOBAL_WR, 0x04, 0xFF);
+	if (ret < 0)
+		goto bail;
+
+	ret = ibsd_mod_allchnls(dd, RXHSSTATUS(0) | EPB_GLOBAL_WR, 0x04, 0xFF);
+	if (ret < 0)
+		goto bail;
 
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, BACTRL(chnl), 0x40, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, LDOUTCTRL1(chnl), 0x04, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, RXHSSTATUS(chnl), 0x04, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
 	/*
-	 * delay for max possible number of steps, with slop.
+	 * Delay for max possible number of steps, with slop.
 	 * Each step is about 4usec.
 	 */
 	udelay(415);
-	for (chnl = 0; chnl < 4; ++chnl) {
-		ret = ibsd_sto_noisy(dd, LDOUTCTRL1(chnl), 0x00, 0xFF);
-		if (ret < 0)
-			goto bail;
-	}
+
+	ret = ibsd_mod_allchnls(dd, LDOUTCTRL1(0) | EPB_GLOBAL_WR, 0x00, 0xFF);
+
 bail:
 	return ret;
 }
 
 #define RELOCK_FIRST_MS 3
 #define RXLSPPM(chan) EPB_LOC(chan, 0, 2)
-void ipath_toggle_rclkrls(struct ipath_devdata *dd)
+void toggle_7220_rclkrls(struct qib_devdata *dd)
 {
 	int loc = RXLSPPM(0) | EPB_GLOBAL_WR;
 	int ret;
 
 	ret = ibsd_mod_allchnls(dd, loc, 0, 0x80);
 	if (ret < 0)
-		ipath_dev_err(dd, "RCLKRLS failed to clear D7\n");
+		qib_dev_err(dd, "RCLKRLS failed to clear D7\n");
 	else {
 		udelay(1);
 		ibsd_mod_allchnls(dd, loc, 0x80, 0x80);
@@ -1354,109 +1359,91 @@ void ipath_toggle_rclkrls(struct ipath_devdata *dd)
 	udelay(1);
 	ret = ibsd_mod_allchnls(dd, loc, 0, 0x80);
 	if (ret < 0)
-		ipath_dev_err(dd, "RCLKRLS failed to clear D7\n");
+		qib_dev_err(dd, "RCLKRLS failed to clear D7\n");
 	else {
 		udelay(1);
 		ibsd_mod_allchnls(dd, loc, 0x80, 0x80);
 	}
 	/* Now reset xgxs and IBC to complete the recovery */
-	dd->ipath_f_xgxs_reset(dd);
+	dd->f_xgxs_reset(dd->pport);
 }
 
 /*
  * Shut down the timer that polls for relock occasions, if needed
- * this is "hooked" from ipath_7220_quiet_serdes(), which is called
- * just before ipath_shutdown_device() in ipath_driver.c shuts down all
+ * this is "hooked" from qib_7220_quiet_serdes(), which is called
+ * just before qib_shutdown_device() in qib_driver.c shuts down all
  * the other timers
  */
-void ipath_shutdown_relock_poll(struct ipath_devdata *dd)
+void shutdown_7220_relock_poll(struct qib_devdata *dd)
 {
-	struct ipath_relock *irp = &dd->ipath_relock_singleton;
-	if (atomic_read(&irp->ipath_relock_timer_active)) {
-		del_timer_sync(&irp->ipath_relock_timer);
-		atomic_set(&irp->ipath_relock_timer_active, 0);
-	}
+	if (dd->cspec->relock_timer_active)
+		del_timer_sync(&dd->cspec->relock_timer);
 }
 
-static unsigned ipath_relock_by_timer = 1;
-module_param_named(relock_by_timer, ipath_relock_by_timer, uint,
-	S_IWUSR | S_IRUGO);
+static unsigned qib_relock_by_timer = 1;
+module_param_named(relock_by_timer, qib_relock_by_timer, uint,
+		   S_IWUSR | S_IRUGO);
 MODULE_PARM_DESC(relock_by_timer, "Allow relock attempt if link not up");
 
-static void ipath_run_relock(unsigned long opaque)
+static void qib_run_relock(unsigned long opaque)
 {
-	struct ipath_devdata *dd = (struct ipath_devdata *)opaque;
-	struct ipath_relock *irp = &dd->ipath_relock_singleton;
-	u64 val, ltstate;
-
-	if (!(dd->ipath_flags & IPATH_INITTED)) {
-		/* Not yet up, just reenable the timer for later */
-		irp->ipath_relock_interval = HZ;
-		mod_timer(&irp->ipath_relock_timer, jiffies + HZ);
-		return;
-	}
+	struct qib_devdata *dd = (struct qib_devdata *)opaque;
+	struct qib_pportdata *ppd = dd->pport;
+	struct qib_chip_specific *cs = dd->cspec;
+	int timeoff;
 
 	/*
-	 * Check link-training state for "stuck" state.
+	 * Check link-training state for "stuck" state, when down.
 	 * if found, try relock and schedule another try at
 	 * exponentially growing delay, maxed at one second.
 	 * if not stuck, our work is done.
 	 */
-	val = ipath_read_kreg64(dd, dd->ipath_kregs->kr_ibcstatus);
-	ltstate = ipath_ib_linktrstate(dd, val);
-
-	if (ltstate <= INFINIPATH_IBCS_LT_STATE_CFGWAITRMT
-		&& ltstate != INFINIPATH_IBCS_LT_STATE_LINKUP) {
-		int timeoff;
-		/* Not up yet. Try again, if allowed by module-param */
-		if (ipath_relock_by_timer) {
-			if (dd->ipath_flags & IPATH_IB_AUTONEG_INPROG)
-				ipath_cdbg(VERBOSE, "Skip RELOCK in AUTONEG\n");
-			else if (!(dd->ipath_flags & IPATH_IB_LINK_DISABLED)) {
-				ipath_cdbg(VERBOSE, "RELOCK\n");
-				ipath_toggle_rclkrls(dd);
-			}
+	if ((dd->flags & QIB_INITTED) && !(ppd->lflags &
+	    (QIBL_IB_AUTONEG_INPROG | QIBL_LINKINIT | QIBL_LINKARMED |
+	     QIBL_LINKACTIVE))) {
+		if (qib_relock_by_timer) {
+			if (!(ppd->lflags & QIBL_IB_LINK_DISABLED))
+				toggle_7220_rclkrls(dd);
 		}
 		/* re-set timer for next check */
-		timeoff = irp->ipath_relock_interval << 1;
+		timeoff = cs->relock_interval << 1;
 		if (timeoff > HZ)
 			timeoff = HZ;
-		irp->ipath_relock_interval = timeoff;
-
-		mod_timer(&irp->ipath_relock_timer, jiffies + timeoff);
-	} else {
-		/* Up, so no more need to check so often */
-		mod_timer(&irp->ipath_relock_timer, jiffies + HZ);
-	}
+		cs->relock_interval = timeoff;
+	} else
+		timeoff = HZ;
+	mod_timer(&cs->relock_timer, jiffies + timeoff);
 }
 
-void ipath_set_relock_poll(struct ipath_devdata *dd, int ibup)
+void set_7220_relock_poll(struct qib_devdata *dd, int ibup)
 {
-	struct ipath_relock *irp = &dd->ipath_relock_singleton;
+	struct qib_chip_specific *cs = dd->cspec;
 
-	if (ibup > 0) {
-		/* we are now up, so relax timer to 1 second interval */
-		if (atomic_read(&irp->ipath_relock_timer_active))
-			mod_timer(&irp->ipath_relock_timer, jiffies + HZ);
+	if (ibup) {
+		/* We are now up, relax timer to 1 second interval */
+		if (cs->relock_timer_active) {
+			cs->relock_interval = HZ;
+			mod_timer(&cs->relock_timer, jiffies + HZ);
+		}
 	} else {
 		/* Transition to down, (re-)set timer to short interval. */
-		int timeout;
-		timeout = (HZ * ((ibup == -1) ? 1000 : RELOCK_FIRST_MS))/1000;
+		unsigned int timeout;
+
+		timeout = msecs_to_jiffies(RELOCK_FIRST_MS);
 		if (timeout == 0)
 			timeout = 1;
 		/* If timer has not yet been started, do so. */
-		if (atomic_inc_return(&irp->ipath_relock_timer_active) == 1) {
-			init_timer(&irp->ipath_relock_timer);
-			irp->ipath_relock_timer.function = ipath_run_relock;
-			irp->ipath_relock_timer.data = (unsigned long) dd;
-			irp->ipath_relock_interval = timeout;
-			irp->ipath_relock_timer.expires = jiffies + timeout;
-			add_timer(&irp->ipath_relock_timer);
+		if (!cs->relock_timer_active) {
+			cs->relock_timer_active = 1;
+			init_timer(&cs->relock_timer);
+			cs->relock_timer.function = qib_run_relock;
+			cs->relock_timer.data = (unsigned long) dd;
+			cs->relock_interval = timeout;
+			cs->relock_timer.expires = jiffies + timeout;
+			add_timer(&cs->relock_timer);
 		} else {
-			irp->ipath_relock_interval = timeout;
-			mod_timer(&irp->ipath_relock_timer, jiffies + timeout);
-			atomic_dec(&irp->ipath_relock_timer_active);
+			cs->relock_interval = timeout;
+			mod_timer(&cs->relock_timer, jiffies + timeout);
 		}
 	}
 }
-
diff --git a/drivers/infiniband/hw/qib/qib_sdma.c b/drivers/infiniband/hw/qib/qib_sdma.c
new file mode 100644
index 00000000000..c6d6a54d2e1
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sdma.c
@@ -0,0 +1,1039 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2007 - 2012 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/netdevice.h>
+#include <linux/moduleparam.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+/* default pio off, sdma on */
+static ushort sdma_descq_cnt = 256;
+module_param_named(sdma_descq_cnt, sdma_descq_cnt, ushort, S_IRUGO);
+MODULE_PARM_DESC(sdma_descq_cnt, "Number of SDMA descq entries");
+
+/*
+ * Bits defined in the send DMA descriptor.
+ */
+#define SDMA_DESC_LAST          (1ULL << 11)
+#define SDMA_DESC_FIRST         (1ULL << 12)
+#define SDMA_DESC_DMA_HEAD      (1ULL << 13)
+#define SDMA_DESC_USE_LARGE_BUF (1ULL << 14)
+#define SDMA_DESC_INTR          (1ULL << 15)
+#define SDMA_DESC_COUNT_LSB     16
+#define SDMA_DESC_GEN_LSB       30
+
+char *qib_sdma_state_names[] = {
+	[qib_sdma_state_s00_hw_down]          = "s00_HwDown",
+	[qib_sdma_state_s10_hw_start_up_wait] = "s10_HwStartUpWait",
+	[qib_sdma_state_s20_idle]             = "s20_Idle",
+	[qib_sdma_state_s30_sw_clean_up_wait] = "s30_SwCleanUpWait",
+	[qib_sdma_state_s40_hw_clean_up_wait] = "s40_HwCleanUpWait",
+	[qib_sdma_state_s50_hw_halt_wait]     = "s50_HwHaltWait",
+	[qib_sdma_state_s99_running]          = "s99_Running",
+};
+
+char *qib_sdma_event_names[] = {
+	[qib_sdma_event_e00_go_hw_down]   = "e00_GoHwDown",
+	[qib_sdma_event_e10_go_hw_start]  = "e10_GoHwStart",
+	[qib_sdma_event_e20_hw_started]   = "e20_HwStarted",
+	[qib_sdma_event_e30_go_running]   = "e30_GoRunning",
+	[qib_sdma_event_e40_sw_cleaned]   = "e40_SwCleaned",
+	[qib_sdma_event_e50_hw_cleaned]   = "e50_HwCleaned",
+	[qib_sdma_event_e60_hw_halted]    = "e60_HwHalted",
+	[qib_sdma_event_e70_go_idle]      = "e70_GoIdle",
+	[qib_sdma_event_e7220_err_halted] = "e7220_ErrHalted",
+	[qib_sdma_event_e7322_err_halted] = "e7322_ErrHalted",
+	[qib_sdma_event_e90_timer_tick]   = "e90_TimerTick",
+};
+
+/* declare all statics here rather than keep sorting */
+static int alloc_sdma(struct qib_pportdata *);
+static void sdma_complete(struct kref *);
+static void sdma_finalput(struct qib_sdma_state *);
+static void sdma_get(struct qib_sdma_state *);
+static void sdma_put(struct qib_sdma_state *);
+static void sdma_set_state(struct qib_pportdata *, enum qib_sdma_states);
+static void sdma_start_sw_clean_up(struct qib_pportdata *);
+static void sdma_sw_clean_up_task(unsigned long);
+static void unmap_desc(struct qib_pportdata *, unsigned);
+
+static void sdma_get(struct qib_sdma_state *ss)
+{
+	kref_get(&ss->kref);
+}
+
+static void sdma_complete(struct kref *kref)
+{
+	struct qib_sdma_state *ss =
+		container_of(kref, struct qib_sdma_state, kref);
+
+	complete(&ss->comp);
+}
+
+static void sdma_put(struct qib_sdma_state *ss)
+{
+	kref_put(&ss->kref, sdma_complete);
+}
+
+static void sdma_finalput(struct qib_sdma_state *ss)
+{
+	sdma_put(ss);
+	wait_for_completion(&ss->comp);
+}
+
+/*
+ * Complete all the sdma requests on the active list, in the correct
+ * order, and with appropriate processing.   Called when cleaning up
+ * after sdma shutdown, and when new sdma requests are submitted for
+ * a link that is down.   This matches what is done for requests
+ * that complete normally, it's just the full list.
+ *
+ * Must be called with sdma_lock held
+ */
+static void clear_sdma_activelist(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_txreq *txp, *txp_next;
+
+	list_for_each_entry_safe(txp, txp_next, &ppd->sdma_activelist, list) {
+		list_del_init(&txp->list);
+		if (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) {
+			unsigned idx;
+
+			idx = txp->start_idx;
+			while (idx != txp->next_descq_idx) {
+				unmap_desc(ppd, idx);
+				if (++idx == ppd->sdma_descq_cnt)
+					idx = 0;
+			}
+		}
+		if (txp->callback)
+			(*txp->callback)(txp, QIB_SDMA_TXREQ_S_ABORTED);
+	}
+}
+
+static void sdma_sw_clean_up_task(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *) opaque;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	/*
+	 * At this point, the following should always be true:
+	 * - We are halted, so no more descriptors are getting retired.
+	 * - We are not running, so no one is submitting new work.
+	 * - Only we can send the e40_sw_cleaned, so we can't start
+	 *   running again until we say so.  So, the active list and
+	 *   descq are ours to play with.
+	 */
+
+	/* Process all retired requests. */
+	qib_sdma_make_progress(ppd);
+
+	clear_sdma_activelist(ppd);
+
+	/*
+	 * Resync count of added and removed.  It is VERY important that
+	 * sdma_descq_removed NEVER decrement - user_sdma depends on it.
+	 */
+	ppd->sdma_descq_removed = ppd->sdma_descq_added;
+
+	/*
+	 * Reset our notion of head and tail.
+	 * Note that the HW registers will be reset when switching states
+	 * due to calling __qib_sdma_process_event() below.
+	 */
+	ppd->sdma_descq_tail = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_head_dma[0] = 0;
+	ppd->sdma_generation = 0;
+
+	__qib_sdma_process_event(ppd, qib_sdma_event_e40_sw_cleaned);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+/*
+ * This is called when changing to state qib_sdma_state_s10_hw_start_up_wait
+ * as a result of send buffer errors or send DMA descriptor errors.
+ * We want to disarm the buffers in these cases.
+ */
+static void sdma_hw_start_up(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	unsigned bufno;
+
+	for (bufno = ss->first_sendbuf; bufno < ss->last_sendbuf; ++bufno)
+		ppd->dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_BUF(bufno));
+
+	ppd->dd->f_sdma_hw_start_up(ppd);
+}
+
+static void sdma_sw_tear_down(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	/* Releasing this reference means the state machine has stopped. */
+	sdma_put(ss);
+}
+
+static void sdma_start_sw_clean_up(struct qib_pportdata *ppd)
+{
+	tasklet_hi_schedule(&ppd->sdma_sw_clean_up_task);
+}
+
+static void sdma_set_state(struct qib_pportdata *ppd,
+	enum qib_sdma_states next_state)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+	struct sdma_set_state_action *action = ss->set_state_action;
+	unsigned op = 0;
+
+	/* debugging bookkeeping */
+	ss->previous_state = ss->current_state;
+	ss->previous_op = ss->current_op;
+
+	ss->current_state = next_state;
+
+	if (action[next_state].op_enable)
+		op |= QIB_SDMA_SENDCTRL_OP_ENABLE;
+
+	if (action[next_state].op_intenable)
+		op |= QIB_SDMA_SENDCTRL_OP_INTENABLE;
+
+	if (action[next_state].op_halt)
+		op |= QIB_SDMA_SENDCTRL_OP_HALT;
+
+	if (action[next_state].op_drain)
+		op |= QIB_SDMA_SENDCTRL_OP_DRAIN;
+
+	if (action[next_state].go_s99_running_tofalse)
+		ss->go_s99_running = 0;
+
+	if (action[next_state].go_s99_running_totrue)
+		ss->go_s99_running = 1;
+
+	ss->current_op = op;
+
+	ppd->dd->f_sdma_sendctrl(ppd, ss->current_op);
+}
+
+static void unmap_desc(struct qib_pportdata *ppd, unsigned head)
+{
+	__le64 *descqp = &ppd->sdma_descq[head].qw[0];
+	u64 desc[2];
+	dma_addr_t addr;
+	size_t len;
+
+	desc[0] = le64_to_cpu(descqp[0]);
+	desc[1] = le64_to_cpu(descqp[1]);
+
+	addr = (desc[1] << 32) | (desc[0] >> 32);
+	len = (desc[0] >> 14) & (0x7ffULL << 2);
+	dma_unmap_single(&ppd->dd->pcidev->dev, addr, len, DMA_TO_DEVICE);
+}
+
+static int alloc_sdma(struct qib_pportdata *ppd)
+{
+	ppd->sdma_descq_cnt = sdma_descq_cnt;
+	if (!ppd->sdma_descq_cnt)
+		ppd->sdma_descq_cnt = 256;
+
+	/* Allocate memory for SendDMA descriptor FIFO */
+	ppd->sdma_descq = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), &ppd->sdma_descq_phys,
+		GFP_KERNEL);
+
+	if (!ppd->sdma_descq) {
+		qib_dev_err(ppd->dd,
+			"failed to allocate SendDMA descriptor FIFO memory\n");
+		goto bail;
+	}
+
+	/* Allocate memory for DMA of head register to memory */
+	ppd->sdma_head_dma = dma_alloc_coherent(&ppd->dd->pcidev->dev,
+		PAGE_SIZE, &ppd->sdma_head_phys, GFP_KERNEL);
+	if (!ppd->sdma_head_dma) {
+		qib_dev_err(ppd->dd,
+			"failed to allocate SendDMA head memory\n");
+		goto cleanup_descq;
+	}
+	ppd->sdma_head_dma[0] = 0;
+	return 0;
+
+cleanup_descq:
+	dma_free_coherent(&ppd->dd->pcidev->dev,
+		ppd->sdma_descq_cnt * sizeof(u64[2]), (void *)ppd->sdma_descq,
+		ppd->sdma_descq_phys);
+	ppd->sdma_descq = NULL;
+	ppd->sdma_descq_phys = 0;
+bail:
+	ppd->sdma_descq_cnt = 0;
+	return -ENOMEM;
+}
+
+static void free_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+
+	if (ppd->sdma_head_dma) {
+		dma_free_coherent(&dd->pcidev->dev, PAGE_SIZE,
+				  (void *)ppd->sdma_head_dma,
+				  ppd->sdma_head_phys);
+		ppd->sdma_head_dma = NULL;
+		ppd->sdma_head_phys = 0;
+	}
+
+	if (ppd->sdma_descq) {
+		dma_free_coherent(&dd->pcidev->dev,
+				  ppd->sdma_descq_cnt * sizeof(u64[2]),
+				  ppd->sdma_descq, ppd->sdma_descq_phys);
+		ppd->sdma_descq = NULL;
+		ppd->sdma_descq_phys = 0;
+	}
+}
+
+static inline void make_sdma_desc(struct qib_pportdata *ppd,
+				  u64 *sdmadesc, u64 addr, u64 dwlen,
+				  u64 dwoffset)
+{
+
+	WARN_ON(addr & 3);
+	/* SDmaPhyAddr[47:32] */
+	sdmadesc[1] = addr >> 32;
+	/* SDmaPhyAddr[31:0] */
+	sdmadesc[0] = (addr & 0xfffffffcULL) << 32;
+	/* SDmaGeneration[1:0] */
+	sdmadesc[0] |= (ppd->sdma_generation & 3ULL) <<
+		SDMA_DESC_GEN_LSB;
+	/* SDmaDwordCount[10:0] */
+	sdmadesc[0] |= (dwlen & 0x7ffULL) << SDMA_DESC_COUNT_LSB;
+	/* SDmaBufOffset[12:2] */
+	sdmadesc[0] |= dwoffset & 0x7ffULL;
+}
+
+/* sdma_lock must be held */
+int qib_sdma_make_progress(struct qib_pportdata *ppd)
+{
+	struct list_head *lp = NULL;
+	struct qib_sdma_txreq *txp = NULL;
+	struct qib_devdata *dd = ppd->dd;
+	int progress = 0;
+	u16 hwhead;
+	u16 idx = 0;
+
+	hwhead = dd->f_sdma_gethead(ppd);
+
+	/* The reason for some of the complexity of this code is that
+	 * not all descriptors have corresponding txps.  So, we have to
+	 * be able to skip over descs until we wander into the range of
+	 * the next txp on the list.
+	 */
+
+	if (!list_empty(&ppd->sdma_activelist)) {
+		lp = ppd->sdma_activelist.next;
+		txp = list_entry(lp, struct qib_sdma_txreq, list);
+		idx = txp->start_idx;
+	}
+
+	while (ppd->sdma_descq_head != hwhead) {
+		/* if desc is part of this txp, unmap if needed */
+		if (txp && (txp->flags & QIB_SDMA_TXREQ_F_FREEDESC) &&
+		    (idx == ppd->sdma_descq_head)) {
+			unmap_desc(ppd, ppd->sdma_descq_head);
+			if (++idx == ppd->sdma_descq_cnt)
+				idx = 0;
+		}
+
+		/* increment dequed desc count */
+		ppd->sdma_descq_removed++;
+
+		/* advance head, wrap if needed */
+		if (++ppd->sdma_descq_head == ppd->sdma_descq_cnt)
+			ppd->sdma_descq_head = 0;
+
+		/* if now past this txp's descs, do the callback */
+		if (txp && txp->next_descq_idx == ppd->sdma_descq_head) {
+			/* remove from active list */
+			list_del_init(&txp->list);
+			if (txp->callback)
+				(*txp->callback)(txp, QIB_SDMA_TXREQ_S_OK);
+			/* see if there is another txp */
+			if (list_empty(&ppd->sdma_activelist))
+				txp = NULL;
+			else {
+				lp = ppd->sdma_activelist.next;
+				txp = list_entry(lp, struct qib_sdma_txreq,
+					list);
+				idx = txp->start_idx;
+			}
+		}
+		progress = 1;
+	}
+	if (progress)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+	return progress;
+}
+
+/*
+ * This is called from interrupt context.
+ */
+void qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_intr(ppd);
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_intr(struct qib_pportdata *ppd)
+{
+	if (__qib_sdma_running(ppd)) {
+		qib_sdma_make_progress(ppd);
+		if (!list_empty(&ppd->sdma_userpending))
+			qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);
+	}
+}
+
+int qib_setup_sdma(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	int ret = 0;
+
+	ret = alloc_sdma(ppd);
+	if (ret)
+		goto bail;
+
+	/* set consistent sdma state */
+	ppd->dd->f_sdma_init_early(ppd);
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	/* set up reference counting */
+	kref_init(&ppd->sdma_state.kref);
+	init_completion(&ppd->sdma_state.comp);
+
+	ppd->sdma_generation = 0;
+	ppd->sdma_descq_head = 0;
+	ppd->sdma_descq_removed = 0;
+	ppd->sdma_descq_added = 0;
+
+	ppd->sdma_intrequest = 0;
+	INIT_LIST_HEAD(&ppd->sdma_userpending);
+
+	INIT_LIST_HEAD(&ppd->sdma_activelist);
+
+	tasklet_init(&ppd->sdma_sw_clean_up_task, sdma_sw_clean_up_task,
+		(unsigned long)ppd);
+
+	ret = dd->f_init_sdma_regs(ppd);
+	if (ret)
+		goto bail_alloc;
+
+	qib_sdma_process_event(ppd, qib_sdma_event_e10_go_hw_start);
+
+	return 0;
+
+bail_alloc:
+	qib_teardown_sdma(ppd);
+bail:
+	return ret;
+}
+
+void qib_teardown_sdma(struct qib_pportdata *ppd)
+{
+	qib_sdma_process_event(ppd, qib_sdma_event_e00_go_hw_down);
+
+	/*
+	 * This waits for the state machine to exit so it is not
+	 * necessary to kill the sdma_sw_clean_up_task to make sure
+	 * it is not running.
+	 */
+	sdma_finalput(&ppd->sdma_state);
+
+	free_sdma(ppd);
+}
+
+int qib_sdma_running(struct qib_pportdata *ppd)
+{
+	unsigned long flags;
+	int ret;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = __qib_sdma_running(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/*
+ * Complete a request when sdma not running; likely only request
+ * but to simplify the code, always queue it, then process the full
+ * activelist.  We process the entire list to ensure that this particular
+ * request does get it's callback, but in the correct order.
+ * Must be called with sdma_lock held
+ */
+static void complete_sdma_err_req(struct qib_pportdata *ppd,
+				  struct qib_verbs_txreq *tx)
+{
+	atomic_inc(&tx->qp->s_dma_busy);
+	/* no sdma descriptors, so no unmap_desc */
+	tx->txreq.start_idx = 0;
+	tx->txreq.next_descq_idx = 0;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	clear_sdma_activelist(ppd);
+}
+
+/*
+ * This function queues one IB packet onto the send DMA queue per call.
+ * The caller is responsible for checking:
+ * 1) The number of send DMA descriptor entries is less than the size of
+ *    the descriptor queue.
+ * 2) The IB SGE addresses and lengths are 32-bit aligned
+ *    (except possibly the last SGE's length)
+ * 3) The SGE addresses are suitable for passing to dma_map_single().
+ */
+int qib_sdma_verbs_send(struct qib_pportdata *ppd,
+			struct qib_sge_state *ss, u32 dwords,
+			struct qib_verbs_txreq *tx)
+{
+	unsigned long flags;
+	struct qib_sge *sge;
+	struct qib_qp *qp;
+	int ret = 0;
+	u16 tail;
+	__le64 *descqp;
+	u64 sdmadesc[2];
+	u32 dwoffset;
+	dma_addr_t addr;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+retry:
+	if (unlikely(!__qib_sdma_running(ppd))) {
+		complete_sdma_err_req(ppd, tx);
+		goto unlock;
+	}
+
+	if (tx->txreq.sg_count > qib_sdma_descq_freecnt(ppd)) {
+		if (qib_sdma_make_progress(ppd))
+			goto retry;
+		if (ppd->dd->flags & QIB_HAS_SDMA_TIMEOUT)
+			ppd->dd->f_sdma_set_desc_cnt(ppd,
+					ppd->sdma_descq_cnt / 2);
+		goto busy;
+	}
+
+	dwoffset = tx->hdr_dwords;
+	make_sdma_desc(ppd, sdmadesc, (u64) tx->txreq.addr, dwoffset, 0);
+
+	sdmadesc[0] |= SDMA_DESC_FIRST;
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+		sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+
+	/* write to the descq */
+	tail = ppd->sdma_descq_tail;
+	descqp = &ppd->sdma_descq[tail].qw[0];
+	*descqp++ = cpu_to_le64(sdmadesc[0]);
+	*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+	/* increment the tail */
+	if (++tail == ppd->sdma_descq_cnt) {
+		tail = 0;
+		descqp = &ppd->sdma_descq[0].qw[0];
+		++ppd->sdma_generation;
+	}
+
+	tx->txreq.start_idx = tail;
+
+	sge = &ss->sge;
+	while (dwords) {
+		u32 dw;
+		u32 len;
+
+		len = dwords << 2;
+		if (len > sge->length)
+			len = sge->length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		dw = (len + 3) >> 2;
+		addr = dma_map_single(&ppd->dd->pcidev->dev, sge->vaddr,
+				      dw << 2, DMA_TO_DEVICE);
+		if (dma_mapping_error(&ppd->dd->pcidev->dev, addr))
+			goto unmap;
+		sdmadesc[0] = 0;
+		make_sdma_desc(ppd, sdmadesc, (u64) addr, dw, dwoffset);
+		/* SDmaUseLargeBuf has to be set in every descriptor */
+		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_USELARGEBUF)
+			sdmadesc[0] |= SDMA_DESC_USE_LARGE_BUF;
+		/* write to the descq */
+		*descqp++ = cpu_to_le64(sdmadesc[0]);
+		*descqp++ = cpu_to_le64(sdmadesc[1]);
+
+		/* increment the tail */
+		if (++tail == ppd->sdma_descq_cnt) {
+			tail = 0;
+			descqp = &ppd->sdma_descq[0].qw[0];
+			++ppd->sdma_generation;
+		}
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+
+		dwoffset += dw;
+		dwords -= dw;
+	}
+
+	if (!tail)
+		descqp = &ppd->sdma_descq[ppd->sdma_descq_cnt].qw[0];
+	descqp -= 2;
+	descqp[0] |= cpu_to_le64(SDMA_DESC_LAST);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_HEADTOHOST)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_DMA_HEAD);
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_INTREQ)
+		descqp[0] |= cpu_to_le64(SDMA_DESC_INTR);
+
+	atomic_inc(&tx->qp->s_dma_busy);
+	tx->txreq.next_descq_idx = tail;
+	ppd->dd->f_sdma_update_tail(ppd, tail);
+	ppd->sdma_descq_added += tx->txreq.sg_count;
+	list_add_tail(&tx->txreq.list, &ppd->sdma_activelist);
+	goto unlock;
+
+unmap:
+	for (;;) {
+		if (!tail)
+			tail = ppd->sdma_descq_cnt - 1;
+		else
+			tail--;
+		if (tail == ppd->sdma_descq_tail)
+			break;
+		unmap_desc(ppd, tail);
+	}
+	qp = tx->qp;
+	qib_put_txreq(tx);
+	spin_lock(&qp->r_lock);
+	spin_lock(&qp->s_lock);
+	if (qp->ibqp.qp_type == IB_QPT_RC) {
+		/* XXX what about error sending RDMA read responses? */
+		if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)
+			qib_error_qp(qp, IB_WC_GENERAL_ERR);
+	} else if (qp->s_wqe)
+		qib_send_complete(qp, qp->s_wqe, IB_WC_GENERAL_ERR);
+	spin_unlock(&qp->s_lock);
+	spin_unlock(&qp->r_lock);
+	/* return zero to process the next send work request */
+	goto unlock;
+
+busy:
+	qp = tx->qp;
+	spin_lock(&qp->s_lock);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		struct qib_ibdev *dev;
+
+		/*
+		 * If we couldn't queue the DMA request, save the info
+		 * and try again later rather than destroying the
+		 * buffer and undoing the side effects of the copy.
+		 */
+		tx->ss = ss;
+		tx->dwords = dwords;
+		qp->s_tx = tx;
+		dev = &ppd->dd->verbs_dev;
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			struct qib_ibport *ibp;
+
+			ibp = &ppd->ibport_data;
+			ibp->n_dmawait++;
+			qp->s_flags |= QIB_S_WAIT_DMA_DESC;
+			list_add_tail(&qp->iowait, &dev->dmawait);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock(&qp->s_lock);
+		ret = -EBUSY;
+	} else {
+		spin_unlock(&qp->s_lock);
+		qib_put_txreq(tx);
+	}
+unlock:
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+	return ret;
+}
+
+/*
+ * sdma_lock should be acquired before calling this routine
+ */
+void dump_sdma_state(struct qib_pportdata *ppd)
+{
+	struct qib_sdma_desc *descq;
+	struct qib_sdma_txreq *txp, *txpnext;
+	__le64 *descqp;
+	u64 desc[2];
+	u64 addr;
+	u16 gen, dwlen, dwoffset;
+	u16 head, tail, cnt;
+
+	head = ppd->sdma_descq_head;
+	tail = ppd->sdma_descq_tail;
+	cnt = qib_sdma_descq_freecnt(ppd);
+	descq = ppd->sdma_descq;
+
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA ppd->sdma_descq_head: %u\n", head);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA ppd->sdma_descq_tail: %u\n", tail);
+	qib_dev_porterr(ppd->dd, ppd->port,
+		"SDMA sdma_descq_freecnt: %u\n", cnt);
+
+	/* print info for each entry in the descriptor queue */
+	while (head != tail) {
+		char flags[6] = { 'x', 'x', 'x', 'x', 'x', 0 };
+
+		descqp = &descq[head].qw[0];
+		desc[0] = le64_to_cpu(descqp[0]);
+		desc[1] = le64_to_cpu(descqp[1]);
+		flags[0] = (desc[0] & 1<<15) ? 'I' : '-';
+		flags[1] = (desc[0] & 1<<14) ? 'L' : 'S';
+		flags[2] = (desc[0] & 1<<13) ? 'H' : '-';
+		flags[3] = (desc[0] & 1<<12) ? 'F' : '-';
+		flags[4] = (desc[0] & 1<<11) ? 'L' : '-';
+		addr = (desc[1] << 32) | ((desc[0] >> 32) & 0xfffffffcULL);
+		gen = (desc[0] >> 30) & 3ULL;
+		dwlen = (desc[0] >> 14) & (0x7ffULL << 2);
+		dwoffset = (desc[0] & 0x7ffULL) << 2;
+		qib_dev_porterr(ppd->dd, ppd->port,
+			"SDMA sdmadesc[%u]: flags:%s addr:0x%016llx gen:%u len:%u bytes offset:%u bytes\n",
+			 head, flags, addr, gen, dwlen, dwoffset);
+		if (++head == ppd->sdma_descq_cnt)
+			head = 0;
+	}
+
+	/* print dma descriptor indices from the TX requests */
+	list_for_each_entry_safe(txp, txpnext, &ppd->sdma_activelist,
+				 list)
+		qib_dev_porterr(ppd->dd, ppd->port,
+			"SDMA txp->start_idx: %u txp->next_descq_idx: %u\n",
+			txp->start_idx, txp->next_descq_idx);
+}
+
+void qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+
+	__qib_sdma_process_event(ppd, event);
+
+	if (ppd->sdma_state.current_state == qib_sdma_state_s99_running)
+		qib_verbs_sdma_desc_avail(ppd, qib_sdma_descq_freecnt(ppd));
+
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+}
+
+void __qib_sdma_process_event(struct qib_pportdata *ppd,
+	enum qib_sdma_events event)
+{
+	struct qib_sdma_state *ss = &ppd->sdma_state;
+
+	switch (ss->current_state) {
+	case qib_sdma_state_s00_hw_down:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			break;
+		case qib_sdma_event_e30_go_running:
+			/*
+			 * If down, but running requested (usually result
+			 * of link up, then we need to start up.
+			 * This can happen when hw down is requested while
+			 * bringing the link up with traffic active on
+			 * 7220, e.g. */
+			ss->go_s99_running = 1;
+			/* fall through and start dma engine */
+		case qib_sdma_event_e10_go_hw_start:
+			/* This reference means the state machine is started */
+			sdma_get(&ppd->sdma_state);
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s10_hw_start_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			sdma_set_state(ppd, ss->go_s99_running ?
+				       qib_sdma_state_s99_running :
+				       qib_sdma_state_s20_idle);
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s20_idle:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_sw_tear_down(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			sdma_set_state(ppd, qib_sdma_state_s99_running);
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s30_sw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s10_hw_start_up_wait);
+			sdma_hw_start_up(ppd);
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s40_hw_clean_up_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s50_hw_halt_wait:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			ss->go_s99_running = 1;
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s40_hw_clean_up_wait);
+			ppd->dd->f_sdma_hw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+
+	case qib_sdma_state_s99_running:
+		switch (event) {
+		case qib_sdma_event_e00_go_hw_down:
+			sdma_set_state(ppd, qib_sdma_state_s00_hw_down);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e10_go_hw_start:
+			break;
+		case qib_sdma_event_e20_hw_started:
+			break;
+		case qib_sdma_event_e30_go_running:
+			break;
+		case qib_sdma_event_e40_sw_cleaned:
+			break;
+		case qib_sdma_event_e50_hw_cleaned:
+			break;
+		case qib_sdma_event_e60_hw_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e70_go_idle:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			ss->go_s99_running = 0;
+			break;
+		case qib_sdma_event_e7220_err_halted:
+			sdma_set_state(ppd,
+				       qib_sdma_state_s30_sw_clean_up_wait);
+			sdma_start_sw_clean_up(ppd);
+			break;
+		case qib_sdma_event_e7322_err_halted:
+			sdma_set_state(ppd, qib_sdma_state_s50_hw_halt_wait);
+			break;
+		case qib_sdma_event_e90_timer_tick:
+			break;
+		}
+		break;
+	}
+
+	ss->last_event = event;
+}
diff --git a/drivers/infiniband/hw/qib/qib_srq.c b/drivers/infiniband/hw/qib/qib_srq.c
new file mode 100644
index 00000000000..d6235931a1b
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_srq.c
@@ -0,0 +1,380 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/err.h>
+#include <linux/slab.h>
+#include <linux/vmalloc.h>
+
+#include "qib_verbs.h"
+
+/**
+ * qib_post_srq_receive - post a receive on a shared receive queue
+ * @ibsrq: the SRQ to post the receive on
+ * @wr: the list of work requests to post
+ * @bad_wr: A pointer to the first WR to cause a problem is put here
+ *
+ * This may be called from interrupt context.
+ */
+int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	unsigned long flags;
+	int ret;
+
+	for (; wr; wr = wr->next) {
+		struct qib_rwqe *wqe;
+		u32 next;
+		int i;
+
+		if ((unsigned) wr->num_sge > srq->rq.max_sge) {
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		spin_lock_irqsave(&srq->rq.lock, flags);
+		wq = srq->rq.wq;
+		next = wq->head + 1;
+		if (next >= srq->rq.size)
+			next = 0;
+		if (next == wq->tail) {
+			spin_unlock_irqrestore(&srq->rq.lock, flags);
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		wqe = get_rwqe_ptr(&srq->rq, wq->head);
+		wqe->wr_id = wr->wr_id;
+		wqe->num_sge = wr->num_sge;
+		for (i = 0; i < wr->num_sge; i++)
+			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
+		wq->head = next;
+		spin_unlock_irqrestore(&srq->rq.lock, flags);
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_create_srq - create a shared receive queue
+ * @ibpd: the protection domain of the SRQ to create
+ * @srq_init_attr: the attributes of the SRQ
+ * @udata: data from libibverbs when creating a user SRQ
+ */
+struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
+			      struct ib_srq_init_attr *srq_init_attr,
+			      struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibpd->device);
+	struct qib_srq *srq;
+	u32 sz;
+	struct ib_srq *ret;
+
+	if (srq_init_attr->srq_type != IB_SRQT_BASIC) {
+		ret = ERR_PTR(-ENOSYS);
+		goto done;
+	}
+
+	if (srq_init_attr->attr.max_sge == 0 ||
+	    srq_init_attr->attr.max_sge > ib_qib_max_srq_sges ||
+	    srq_init_attr->attr.max_wr == 0 ||
+	    srq_init_attr->attr.max_wr > ib_qib_max_srq_wrs) {
+		ret = ERR_PTR(-EINVAL);
+		goto done;
+	}
+
+	srq = kmalloc(sizeof(*srq), GFP_KERNEL);
+	if (!srq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto done;
+	}
+
+	/*
+	 * Need to use vmalloc() if we want to support large #s of entries.
+	 */
+	srq->rq.size = srq_init_attr->attr.max_wr + 1;
+	srq->rq.max_sge = srq_init_attr->attr.max_sge;
+	sz = sizeof(struct ib_sge) * srq->rq.max_sge +
+		sizeof(struct qib_rwqe);
+	srq->rq.wq = vmalloc_user(sizeof(struct qib_rwq) + srq->rq.size * sz);
+	if (!srq->rq.wq) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_srq;
+	}
+
+	/*
+	 * Return the address of the RWQ as the offset to mmap.
+	 * See qib_mmap() for details.
+	 */
+	if (udata && udata->outlen >= sizeof(__u64)) {
+		int err;
+		u32 s = sizeof(struct qib_rwq) + srq->rq.size * sz;
+
+		srq->ip =
+		    qib_create_mmap_info(dev, s, ibpd->uobject->context,
+					 srq->rq.wq);
+		if (!srq->ip) {
+			ret = ERR_PTR(-ENOMEM);
+			goto bail_wq;
+		}
+
+		err = ib_copy_to_udata(udata, &srq->ip->offset,
+				       sizeof(srq->ip->offset));
+		if (err) {
+			ret = ERR_PTR(err);
+			goto bail_ip;
+		}
+	} else
+		srq->ip = NULL;
+
+	/*
+	 * ib_create_srq() will initialize srq->ibsrq.
+	 */
+	spin_lock_init(&srq->rq.lock);
+	srq->rq.wq->head = 0;
+	srq->rq.wq->tail = 0;
+	srq->limit = srq_init_attr->attr.srq_limit;
+
+	spin_lock(&dev->n_srqs_lock);
+	if (dev->n_srqs_allocated == ib_qib_max_srqs) {
+		spin_unlock(&dev->n_srqs_lock);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail_ip;
+	}
+
+	dev->n_srqs_allocated++;
+	spin_unlock(&dev->n_srqs_lock);
+
+	if (srq->ip) {
+		spin_lock_irq(&dev->pending_lock);
+		list_add(&srq->ip->pending_mmaps, &dev->pending_mmaps);
+		spin_unlock_irq(&dev->pending_lock);
+	}
+
+	ret = &srq->ibsrq;
+	goto done;
+
+bail_ip:
+	kfree(srq->ip);
+bail_wq:
+	vfree(srq->rq.wq);
+bail_srq:
+	kfree(srq);
+done:
+	return ret;
+}
+
+/**
+ * qib_modify_srq - modify a shared receive queue
+ * @ibsrq: the SRQ to modify
+ * @attr: the new attributes of the SRQ
+ * @attr_mask: indicates which attributes to modify
+ * @udata: user data for libibverbs.so
+ */
+int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		   enum ib_srq_attr_mask attr_mask,
+		   struct ib_udata *udata)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_rwq *wq;
+	int ret = 0;
+
+	if (attr_mask & IB_SRQ_MAX_WR) {
+		struct qib_rwq *owq;
+		struct qib_rwqe *p;
+		u32 sz, size, n, head, tail;
+
+		/* Check that the requested sizes are below the limits. */
+		if ((attr->max_wr > ib_qib_max_srq_wrs) ||
+		    ((attr_mask & IB_SRQ_LIMIT) ?
+		     attr->srq_limit : srq->limit) > attr->max_wr) {
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		sz = sizeof(struct qib_rwqe) +
+			srq->rq.max_sge * sizeof(struct ib_sge);
+		size = attr->max_wr + 1;
+		wq = vmalloc_user(sizeof(struct qib_rwq) + size * sz);
+		if (!wq) {
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		/* Check that we can write the offset to mmap. */
+		if (udata && udata->inlen >= sizeof(__u64)) {
+			__u64 offset_addr;
+			__u64 offset = 0;
+
+			ret = ib_copy_from_udata(&offset_addr, udata,
+						 sizeof(offset_addr));
+			if (ret)
+				goto bail_free;
+			udata->outbuf =
+				(void __user *) (unsigned long) offset_addr;
+			ret = ib_copy_to_udata(udata, &offset,
+					       sizeof(offset));
+			if (ret)
+				goto bail_free;
+		}
+
+		spin_lock_irq(&srq->rq.lock);
+		/*
+		 * validate head and tail pointer values and compute
+		 * the number of remaining WQEs.
+		 */
+		owq = srq->rq.wq;
+		head = owq->head;
+		tail = owq->tail;
+		if (head >= srq->rq.size || tail >= srq->rq.size) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = head;
+		if (n < tail)
+			n += srq->rq.size - tail;
+		else
+			n -= tail;
+		if (size <= n) {
+			ret = -EINVAL;
+			goto bail_unlock;
+		}
+		n = 0;
+		p = wq->wq;
+		while (tail != head) {
+			struct qib_rwqe *wqe;
+			int i;
+
+			wqe = get_rwqe_ptr(&srq->rq, tail);
+			p->wr_id = wqe->wr_id;
+			p->num_sge = wqe->num_sge;
+			for (i = 0; i < wqe->num_sge; i++)
+				p->sg_list[i] = wqe->sg_list[i];
+			n++;
+			p = (struct qib_rwqe *)((char *) p + sz);
+			if (++tail >= srq->rq.size)
+				tail = 0;
+		}
+		srq->rq.wq = wq;
+		srq->rq.size = size;
+		wq->head = n;
+		wq->tail = 0;
+		if (attr_mask & IB_SRQ_LIMIT)
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+
+		vfree(owq);
+
+		if (srq->ip) {
+			struct qib_mmap_info *ip = srq->ip;
+			struct qib_ibdev *dev = to_idev(srq->ibsrq.device);
+			u32 s = sizeof(struct qib_rwq) + size * sz;
+
+			qib_update_mmap_info(dev, ip, s, wq);
+
+			/*
+			 * Return the offset to mmap.
+			 * See qib_mmap() for details.
+			 */
+			if (udata && udata->inlen >= sizeof(__u64)) {
+				ret = ib_copy_to_udata(udata, &ip->offset,
+						       sizeof(ip->offset));
+				if (ret)
+					goto bail;
+			}
+
+			/*
+			 * Put user mapping info onto the pending list
+			 * unless it already is on the list.
+			 */
+			spin_lock_irq(&dev->pending_lock);
+			if (list_empty(&ip->pending_mmaps))
+				list_add(&ip->pending_mmaps,
+					 &dev->pending_mmaps);
+			spin_unlock_irq(&dev->pending_lock);
+		}
+	} else if (attr_mask & IB_SRQ_LIMIT) {
+		spin_lock_irq(&srq->rq.lock);
+		if (attr->srq_limit >= srq->rq.size)
+			ret = -EINVAL;
+		else
+			srq->limit = attr->srq_limit;
+		spin_unlock_irq(&srq->rq.lock);
+	}
+	goto bail;
+
+bail_unlock:
+	spin_unlock_irq(&srq->rq.lock);
+bail_free:
+	vfree(wq);
+bail:
+	return ret;
+}
+
+int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+
+	attr->max_wr = srq->rq.size - 1;
+	attr->max_sge = srq->rq.max_sge;
+	attr->srq_limit = srq->limit;
+	return 0;
+}
+
+/**
+ * qib_destroy_srq - destroy a shared receive queue
+ * @ibsrq: the SRQ to destroy
+ */
+int qib_destroy_srq(struct ib_srq *ibsrq)
+{
+	struct qib_srq *srq = to_isrq(ibsrq);
+	struct qib_ibdev *dev = to_idev(ibsrq->device);
+
+	spin_lock(&dev->n_srqs_lock);
+	dev->n_srqs_allocated--;
+	spin_unlock(&dev->n_srqs_lock);
+	if (srq->ip)
+		kref_put(&srq->ip->ref, qib_release_mmap_info);
+	else
+		vfree(srq->rq.wq);
+	kfree(srq);
+
+	return 0;
+}
diff --git a/drivers/infiniband/hw/qib/qib_sysfs.c b/drivers/infiniband/hw/qib/qib_sysfs.c
new file mode 100644
index 00000000000..3c8e4e3caca
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_sysfs.c
@@ -0,0 +1,842 @@
+/*
+ * Copyright (c) 2012 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/ctype.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/* start of per-port functions */
+/*
+ * Get/Set heartbeat enable. OR of 1=enabled, 2=auto
+ */
+static ssize_t show_hrtbt_enb(struct qib_pportdata *ppd, char *buf)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+
+	ret = dd->f_get_ib_cfg(ppd, QIB_IB_CFG_HRTBT);
+	ret = scnprintf(buf, PAGE_SIZE, "%d\n", ret);
+	return ret;
+}
+
+static ssize_t store_hrtbt_enb(struct qib_pportdata *ppd, const char *buf,
+			       size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = kstrtou16(buf, 0, &val);
+	if (ret) {
+		qib_dev_err(dd, "attempt to set invalid Heartbeat enable\n");
+		return ret;
+	}
+
+	/*
+	 * Set the "intentional" heartbeat enable per either of
+	 * "Enable" and "Auto", as these are normally set together.
+	 * This bit is consulted when leaving loopback mode,
+	 * because entering loopback mode overrides it and automatically
+	 * disables heartbeat.
+	 */
+	ret = dd->f_set_ib_cfg(ppd, QIB_IB_CFG_HRTBT, val);
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t store_loopback(struct qib_pportdata *ppd, const char *buf,
+			      size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret = count, r;
+
+	r = dd->f_set_ib_loopback(ppd, buf);
+	if (r < 0)
+		ret = r;
+
+	return ret;
+}
+
+static ssize_t store_led_override(struct qib_pportdata *ppd, const char *buf,
+				  size_t count)
+{
+	struct qib_devdata *dd = ppd->dd;
+	int ret;
+	u16 val;
+
+	ret = kstrtou16(buf, 0, &val);
+	if (ret) {
+		qib_dev_err(dd, "attempt to set invalid LED override\n");
+		return ret;
+	}
+
+	qib_set_led_override(ppd, val);
+	return count;
+}
+
+static ssize_t show_status(struct qib_pportdata *ppd, char *buf)
+{
+	ssize_t ret;
+
+	if (!ppd->statusp)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "0x%llx\n",
+				(unsigned long long) *(ppd->statusp));
+	return ret;
+}
+
+/*
+ * For userland compatibility, these offsets must remain fixed.
+ * They are strings for QIB_STATUS_*
+ */
+static const char * const qib_status_str[] = {
+	"Initted",
+	"",
+	"",
+	"",
+	"",
+	"Present",
+	"IB_link_up",
+	"IB_configured",
+	"",
+	"Fatal_Hardware_Error",
+	NULL,
+};
+
+static ssize_t show_status_str(struct qib_pportdata *ppd, char *buf)
+{
+	int i, any;
+	u64 s;
+	ssize_t ret;
+
+	if (!ppd->statusp) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	s = *(ppd->statusp);
+	*buf = '\0';
+	for (any = i = 0; s && qib_status_str[i]; i++) {
+		if (s & 1) {
+			/* if overflow */
+			if (any && strlcat(buf, " ", PAGE_SIZE) >= PAGE_SIZE)
+				break;
+			if (strlcat(buf, qib_status_str[i], PAGE_SIZE) >=
+					PAGE_SIZE)
+				break;
+			any = 1;
+		}
+		s >>= 1;
+	}
+	if (any)
+		strlcat(buf, "\n", PAGE_SIZE);
+
+	ret = strlen(buf);
+
+bail:
+	return ret;
+}
+
+/* end of per-port functions */
+
+/*
+ * Start of per-port file structures and support code
+ * Because we are fitting into other infrastructure, we have to supply the
+ * full set of kobject/sysfs_ops structures and routines.
+ */
+#define QIB_PORT_ATTR(name, mode, show, store) \
+	static struct qib_port_attr qib_port_attr_##name = \
+		__ATTR(name, mode, show, store)
+
+struct qib_port_attr {
+	struct attribute attr;
+	ssize_t (*show)(struct qib_pportdata *, char *);
+	ssize_t (*store)(struct qib_pportdata *, const char *, size_t);
+};
+
+QIB_PORT_ATTR(loopback, S_IWUSR, NULL, store_loopback);
+QIB_PORT_ATTR(led_override, S_IWUSR, NULL, store_led_override);
+QIB_PORT_ATTR(hrtbt_enable, S_IWUSR | S_IRUGO, show_hrtbt_enb,
+	      store_hrtbt_enb);
+QIB_PORT_ATTR(status, S_IRUGO, show_status, NULL);
+QIB_PORT_ATTR(status_str, S_IRUGO, show_status_str, NULL);
+
+static struct attribute *port_default_attributes[] = {
+	&qib_port_attr_loopback.attr,
+	&qib_port_attr_led_override.attr,
+	&qib_port_attr_hrtbt_enable.attr,
+	&qib_port_attr_status.attr,
+	&qib_port_attr_status_str.attr,
+	NULL
+};
+
+/*
+ * Start of per-port congestion control structures and support code
+ */
+
+/*
+ * Congestion control table size followed by table entries
+ */
+static ssize_t read_cc_table_bin(struct file *filp, struct kobject *kobj,
+		struct bin_attribute *bin_attr,
+		char *buf, loff_t pos, size_t count)
+{
+	int ret;
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_cc_kobj);
+
+	if (!qib_cc_table_size || !ppd->ccti_entries_shadow)
+		return -EINVAL;
+
+	ret = ppd->total_cct_entry * sizeof(struct ib_cc_table_entry_shadow)
+		 + sizeof(__be16);
+
+	if (pos > ret)
+		return -EINVAL;
+
+	if (count > ret - pos)
+		count = ret - pos;
+
+	if (!count)
+		return count;
+
+	spin_lock(&ppd->cc_shadow_lock);
+	memcpy(buf, ppd->ccti_entries_shadow, count);
+	spin_unlock(&ppd->cc_shadow_lock);
+
+	return count;
+}
+
+static void qib_port_release(struct kobject *kobj)
+{
+	/* nothing to do since memory is freed by qib_free_devdata() */
+}
+
+static struct kobj_type qib_port_cc_ktype = {
+	.release = qib_port_release,
+};
+
+static struct bin_attribute cc_table_bin_attr = {
+	.attr = {.name = "cc_table_bin", .mode = 0444},
+	.read = read_cc_table_bin,
+	.size = PAGE_SIZE,
+};
+
+/*
+ * Congestion settings: port control, control map and an array of 16
+ * entries for the congestion entries - increase, timer, event log
+ * trigger threshold and the minimum injection rate delay.
+ */
+static ssize_t read_cc_setting_bin(struct file *filp, struct kobject *kobj,
+		struct bin_attribute *bin_attr,
+		char *buf, loff_t pos, size_t count)
+{
+	int ret;
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_cc_kobj);
+
+	if (!qib_cc_table_size || !ppd->congestion_entries_shadow)
+		return -EINVAL;
+
+	ret = sizeof(struct ib_cc_congestion_setting_attr_shadow);
+
+	if (pos > ret)
+		return -EINVAL;
+	if (count > ret - pos)
+		count = ret - pos;
+
+	if (!count)
+		return count;
+
+	spin_lock(&ppd->cc_shadow_lock);
+	memcpy(buf, ppd->congestion_entries_shadow, count);
+	spin_unlock(&ppd->cc_shadow_lock);
+
+	return count;
+}
+
+static struct bin_attribute cc_setting_bin_attr = {
+	.attr = {.name = "cc_settings_bin", .mode = 0444},
+	.read = read_cc_setting_bin,
+	.size = PAGE_SIZE,
+};
+
+
+static ssize_t qib_portattr_show(struct kobject *kobj,
+	struct attribute *attr, char *buf)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->show(ppd, buf);
+}
+
+static ssize_t qib_portattr_store(struct kobject *kobj,
+	struct attribute *attr, const char *buf, size_t len)
+{
+	struct qib_port_attr *pattr =
+		container_of(attr, struct qib_port_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, pport_kobj);
+
+	return pattr->store(ppd, buf, len);
+}
+
+
+static const struct sysfs_ops qib_port_ops = {
+	.show = qib_portattr_show,
+	.store = qib_portattr_store,
+};
+
+static struct kobj_type qib_port_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_port_ops,
+	.default_attrs = port_default_attributes
+};
+
+/* Start sl2vl */
+
+#define QIB_SL2VL_ATTR(N) \
+	static struct qib_sl2vl_attr qib_sl2vl_attr_##N = { \
+		.attr = { .name = __stringify(N), .mode = 0444 }, \
+		.sl = N \
+	}
+
+struct qib_sl2vl_attr {
+	struct attribute attr;
+	int sl;
+};
+
+QIB_SL2VL_ATTR(0);
+QIB_SL2VL_ATTR(1);
+QIB_SL2VL_ATTR(2);
+QIB_SL2VL_ATTR(3);
+QIB_SL2VL_ATTR(4);
+QIB_SL2VL_ATTR(5);
+QIB_SL2VL_ATTR(6);
+QIB_SL2VL_ATTR(7);
+QIB_SL2VL_ATTR(8);
+QIB_SL2VL_ATTR(9);
+QIB_SL2VL_ATTR(10);
+QIB_SL2VL_ATTR(11);
+QIB_SL2VL_ATTR(12);
+QIB_SL2VL_ATTR(13);
+QIB_SL2VL_ATTR(14);
+QIB_SL2VL_ATTR(15);
+
+static struct attribute *sl2vl_default_attributes[] = {
+	&qib_sl2vl_attr_0.attr,
+	&qib_sl2vl_attr_1.attr,
+	&qib_sl2vl_attr_2.attr,
+	&qib_sl2vl_attr_3.attr,
+	&qib_sl2vl_attr_4.attr,
+	&qib_sl2vl_attr_5.attr,
+	&qib_sl2vl_attr_6.attr,
+	&qib_sl2vl_attr_7.attr,
+	&qib_sl2vl_attr_8.attr,
+	&qib_sl2vl_attr_9.attr,
+	&qib_sl2vl_attr_10.attr,
+	&qib_sl2vl_attr_11.attr,
+	&qib_sl2vl_attr_12.attr,
+	&qib_sl2vl_attr_13.attr,
+	&qib_sl2vl_attr_14.attr,
+	&qib_sl2vl_attr_15.attr,
+	NULL
+};
+
+static ssize_t sl2vl_attr_show(struct kobject *kobj, struct attribute *attr,
+			       char *buf)
+{
+	struct qib_sl2vl_attr *sattr =
+		container_of(attr, struct qib_sl2vl_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, sl2vl_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+
+	return sprintf(buf, "%u\n", qibp->sl_to_vl[sattr->sl]);
+}
+
+static const struct sysfs_ops qib_sl2vl_ops = {
+	.show = sl2vl_attr_show,
+};
+
+static struct kobj_type qib_sl2vl_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_sl2vl_ops,
+	.default_attrs = sl2vl_default_attributes
+};
+
+/* End sl2vl */
+
+/* Start diag_counters */
+
+#define QIB_DIAGC_ATTR(N) \
+	static struct qib_diagc_attr qib_diagc_attr_##N = { \
+		.attr = { .name = __stringify(N), .mode = 0664 }, \
+		.counter = offsetof(struct qib_ibport, n_##N) \
+	}
+
+struct qib_diagc_attr {
+	struct attribute attr;
+	size_t counter;
+};
+
+QIB_DIAGC_ATTR(rc_resends);
+QIB_DIAGC_ATTR(rc_acks);
+QIB_DIAGC_ATTR(rc_qacks);
+QIB_DIAGC_ATTR(rc_delayed_comp);
+QIB_DIAGC_ATTR(seq_naks);
+QIB_DIAGC_ATTR(rdma_seq);
+QIB_DIAGC_ATTR(rnr_naks);
+QIB_DIAGC_ATTR(other_naks);
+QIB_DIAGC_ATTR(rc_timeouts);
+QIB_DIAGC_ATTR(loop_pkts);
+QIB_DIAGC_ATTR(pkt_drops);
+QIB_DIAGC_ATTR(dmawait);
+QIB_DIAGC_ATTR(unaligned);
+QIB_DIAGC_ATTR(rc_dupreq);
+QIB_DIAGC_ATTR(rc_seqnak);
+
+static struct attribute *diagc_default_attributes[] = {
+	&qib_diagc_attr_rc_resends.attr,
+	&qib_diagc_attr_rc_acks.attr,
+	&qib_diagc_attr_rc_qacks.attr,
+	&qib_diagc_attr_rc_delayed_comp.attr,
+	&qib_diagc_attr_seq_naks.attr,
+	&qib_diagc_attr_rdma_seq.attr,
+	&qib_diagc_attr_rnr_naks.attr,
+	&qib_diagc_attr_other_naks.attr,
+	&qib_diagc_attr_rc_timeouts.attr,
+	&qib_diagc_attr_loop_pkts.attr,
+	&qib_diagc_attr_pkt_drops.attr,
+	&qib_diagc_attr_dmawait.attr,
+	&qib_diagc_attr_unaligned.attr,
+	&qib_diagc_attr_rc_dupreq.attr,
+	&qib_diagc_attr_rc_seqnak.attr,
+	NULL
+};
+
+static ssize_t diagc_attr_show(struct kobject *kobj, struct attribute *attr,
+			       char *buf)
+{
+	struct qib_diagc_attr *dattr =
+		container_of(attr, struct qib_diagc_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, diagc_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+
+	return sprintf(buf, "%u\n", *(u32 *)((char *)qibp + dattr->counter));
+}
+
+static ssize_t diagc_attr_store(struct kobject *kobj, struct attribute *attr,
+				const char *buf, size_t size)
+{
+	struct qib_diagc_attr *dattr =
+		container_of(attr, struct qib_diagc_attr, attr);
+	struct qib_pportdata *ppd =
+		container_of(kobj, struct qib_pportdata, diagc_kobj);
+	struct qib_ibport *qibp = &ppd->ibport_data;
+	u32 val;
+	int ret;
+
+	ret = kstrtou32(buf, 0, &val);
+	if (ret)
+		return ret;
+	*(u32 *)((char *) qibp + dattr->counter) = val;
+	return size;
+}
+
+static const struct sysfs_ops qib_diagc_ops = {
+	.show = diagc_attr_show,
+	.store = diagc_attr_store,
+};
+
+static struct kobj_type qib_diagc_ktype = {
+	.release = qib_port_release,
+	.sysfs_ops = &qib_diagc_ops,
+	.default_attrs = diagc_default_attributes
+};
+
+/* End diag_counters */
+
+/* end of per-port file structures and support code */
+
+/*
+ * Start of per-unit (or driver, in some cases, but replicated
+ * per unit) functions (these get a device *)
+ */
+static ssize_t show_rev(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+
+	return sprintf(buf, "%x\n", dd_from_dev(dev)->minrev);
+}
+
+static ssize_t show_hca(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (!dd->boardname)
+		ret = -EINVAL;
+	else
+		ret = scnprintf(buf, PAGE_SIZE, "%s\n", dd->boardname);
+	return ret;
+}
+
+static ssize_t show_version(struct device *device,
+			    struct device_attribute *attr, char *buf)
+{
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", (char *)ib_qib_version);
+}
+
+static ssize_t show_boardversion(struct device *device,
+				 struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->boardversion);
+}
+
+
+static ssize_t show_localbus_info(struct device *device,
+				  struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* The string printed here is already newline-terminated. */
+	return scnprintf(buf, PAGE_SIZE, "%s", dd->lbus_info);
+}
+
+
+static ssize_t show_nctxts(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* Return the number of user ports (contexts) available. */
+	/* The calculation below deals with a special case where
+	 * cfgctxts is set to 1 on a single-port board. */
+	return scnprintf(buf, PAGE_SIZE, "%u\n",
+			(dd->first_user_ctxt > dd->cfgctxts) ? 0 :
+			(dd->cfgctxts - dd->first_user_ctxt));
+}
+
+static ssize_t show_nfreectxts(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	/* Return the number of free user ports (contexts) available. */
+	return scnprintf(buf, PAGE_SIZE, "%u\n", dd->freectxts);
+}
+
+static ssize_t show_serial(struct device *device,
+			   struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+
+	buf[sizeof dd->serial] = '\0';
+	memcpy(buf, dd->serial, sizeof dd->serial);
+	strcat(buf, "\n");
+	return strlen(buf);
+}
+
+static ssize_t store_chip_reset(struct device *device,
+				struct device_attribute *attr, const char *buf,
+				size_t count)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+
+	if (count < 5 || memcmp(buf, "reset", 5) || !dd->diag_client) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	ret = qib_reset_device(dd->unit);
+bail:
+	return ret < 0 ? ret : count;
+}
+
+static ssize_t show_logged_errs(struct device *device,
+				struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int idx, count;
+
+	/* force consistency with actual EEPROM */
+	if (qib_update_eeprom_log(dd) != 0)
+		return -ENXIO;
+
+	count = 0;
+	for (idx = 0; idx < QIB_EEP_LOG_CNT; ++idx) {
+		count += scnprintf(buf + count, PAGE_SIZE - count, "%d%c",
+				   dd->eep_st_errs[idx],
+				   idx == (QIB_EEP_LOG_CNT - 1) ? '\n' : ' ');
+	}
+
+	return count;
+}
+
+/*
+ * Dump tempsense regs. in decimal, to ease shell-scripts.
+ */
+static ssize_t show_tempsense(struct device *device,
+			      struct device_attribute *attr, char *buf)
+{
+	struct qib_ibdev *dev =
+		container_of(device, struct qib_ibdev, ibdev.dev);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	int ret;
+	int idx;
+	u8 regvals[8];
+
+	ret = -ENXIO;
+	for (idx = 0; idx < 8; ++idx) {
+		if (idx == 6)
+			continue;
+		ret = dd->f_tempsense_rd(dd, idx);
+		if (ret < 0)
+			break;
+		regvals[idx] = ret;
+	}
+	if (idx == 8)
+		ret = scnprintf(buf, PAGE_SIZE, "%d %d %02X %02X %d %d\n",
+				*(signed char *)(regvals),
+				*(signed char *)(regvals + 1),
+				regvals[2], regvals[3],
+				*(signed char *)(regvals + 5),
+				*(signed char *)(regvals + 7));
+	return ret;
+}
+
+/*
+ * end of per-unit (or driver, in some cases, but replicated
+ * per unit) functions
+ */
+
+/* start of per-unit file structures and support code */
+static DEVICE_ATTR(hw_rev, S_IRUGO, show_rev, NULL);
+static DEVICE_ATTR(hca_type, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, show_hca, NULL);
+static DEVICE_ATTR(version, S_IRUGO, show_version, NULL);
+static DEVICE_ATTR(nctxts, S_IRUGO, show_nctxts, NULL);
+static DEVICE_ATTR(nfreectxts, S_IRUGO, show_nfreectxts, NULL);
+static DEVICE_ATTR(serial, S_IRUGO, show_serial, NULL);
+static DEVICE_ATTR(boardversion, S_IRUGO, show_boardversion, NULL);
+static DEVICE_ATTR(logged_errors, S_IRUGO, show_logged_errs, NULL);
+static DEVICE_ATTR(tempsense, S_IRUGO, show_tempsense, NULL);
+static DEVICE_ATTR(localbus_info, S_IRUGO, show_localbus_info, NULL);
+static DEVICE_ATTR(chip_reset, S_IWUSR, NULL, store_chip_reset);
+
+static struct device_attribute *qib_attributes[] = {
+	&dev_attr_hw_rev,
+	&dev_attr_hca_type,
+	&dev_attr_board_id,
+	&dev_attr_version,
+	&dev_attr_nctxts,
+	&dev_attr_nfreectxts,
+	&dev_attr_serial,
+	&dev_attr_boardversion,
+	&dev_attr_logged_errors,
+	&dev_attr_tempsense,
+	&dev_attr_localbus_info,
+	&dev_attr_chip_reset,
+};
+
+int qib_create_port_files(struct ib_device *ibdev, u8 port_num,
+			  struct kobject *kobj)
+{
+	struct qib_pportdata *ppd;
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret;
+
+	if (!port_num || port_num > dd->num_pports) {
+		qib_dev_err(dd,
+			"Skipping infiniband class with invalid port %u\n",
+			port_num);
+		ret = -ENODEV;
+		goto bail;
+	}
+	ppd = &dd->pport[port_num - 1];
+
+	ret = kobject_init_and_add(&ppd->pport_kobj, &qib_port_ktype, kobj,
+				   "linkcontrol");
+	if (ret) {
+		qib_dev_err(dd,
+			"Skipping linkcontrol sysfs info, (err %d) port %u\n",
+			ret, port_num);
+		goto bail;
+	}
+	kobject_uevent(&ppd->pport_kobj, KOBJ_ADD);
+
+	ret = kobject_init_and_add(&ppd->sl2vl_kobj, &qib_sl2vl_ktype, kobj,
+				   "sl2vl");
+	if (ret) {
+		qib_dev_err(dd,
+			"Skipping sl2vl sysfs info, (err %d) port %u\n",
+			ret, port_num);
+		goto bail_link;
+	}
+	kobject_uevent(&ppd->sl2vl_kobj, KOBJ_ADD);
+
+	ret = kobject_init_and_add(&ppd->diagc_kobj, &qib_diagc_ktype, kobj,
+				   "diag_counters");
+	if (ret) {
+		qib_dev_err(dd,
+			"Skipping diag_counters sysfs info, (err %d) port %u\n",
+			ret, port_num);
+		goto bail_sl;
+	}
+	kobject_uevent(&ppd->diagc_kobj, KOBJ_ADD);
+
+	if (!qib_cc_table_size || !ppd->congestion_entries_shadow)
+		return 0;
+
+	ret = kobject_init_and_add(&ppd->pport_cc_kobj, &qib_port_cc_ktype,
+				kobj, "CCMgtA");
+	if (ret) {
+		qib_dev_err(dd,
+		 "Skipping Congestion Control sysfs info, (err %d) port %u\n",
+		 ret, port_num);
+		goto bail_diagc;
+	}
+
+	kobject_uevent(&ppd->pport_cc_kobj, KOBJ_ADD);
+
+	ret = sysfs_create_bin_file(&ppd->pport_cc_kobj,
+				&cc_setting_bin_attr);
+	if (ret) {
+		qib_dev_err(dd,
+		 "Skipping Congestion Control setting sysfs info, (err %d) port %u\n",
+		 ret, port_num);
+		goto bail_cc;
+	}
+
+	ret = sysfs_create_bin_file(&ppd->pport_cc_kobj,
+				&cc_table_bin_attr);
+	if (ret) {
+		qib_dev_err(dd,
+		 "Skipping Congestion Control table sysfs info, (err %d) port %u\n",
+		 ret, port_num);
+		goto bail_cc_entry_bin;
+	}
+
+	qib_devinfo(dd->pcidev,
+		"IB%u: Congestion Control Agent enabled for port %d\n",
+		dd->unit, port_num);
+
+	return 0;
+
+bail_cc_entry_bin:
+	sysfs_remove_bin_file(&ppd->pport_cc_kobj, &cc_setting_bin_attr);
+bail_cc:
+	kobject_put(&ppd->pport_cc_kobj);
+bail_diagc:
+	kobject_put(&ppd->diagc_kobj);
+bail_sl:
+	kobject_put(&ppd->sl2vl_kobj);
+bail_link:
+	kobject_put(&ppd->pport_kobj);
+bail:
+	return ret;
+}
+
+/*
+ * Register and create our files in /sys/class/infiniband.
+ */
+int qib_verbs_register_sysfs(struct qib_devdata *dd)
+{
+	struct ib_device *dev = &dd->verbs_dev.ibdev;
+	int i, ret;
+
+	for (i = 0; i < ARRAY_SIZE(qib_attributes); ++i) {
+		ret = device_create_file(&dev->dev, qib_attributes[i]);
+		if (ret)
+			goto bail;
+	}
+
+	return 0;
+bail:
+	for (i = 0; i < ARRAY_SIZE(qib_attributes); ++i)
+		device_remove_file(&dev->dev, qib_attributes[i]);
+	return ret;
+}
+
+/*
+ * Unregister and remove our files in /sys/class/infiniband.
+ */
+void qib_verbs_unregister_sysfs(struct qib_devdata *dd)
+{
+	struct qib_pportdata *ppd;
+	int i;
+
+	for (i = 0; i < dd->num_pports; i++) {
+		ppd = &dd->pport[i];
+		if (qib_cc_table_size &&
+			ppd->congestion_entries_shadow) {
+			sysfs_remove_bin_file(&ppd->pport_cc_kobj,
+				&cc_setting_bin_attr);
+			sysfs_remove_bin_file(&ppd->pport_cc_kobj,
+				&cc_table_bin_attr);
+			kobject_put(&ppd->pport_cc_kobj);
+		}
+		kobject_put(&ppd->sl2vl_kobj);
+		kobject_put(&ppd->pport_kobj);
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_twsi.c b/drivers/infiniband/hw/qib/qib_twsi.c
new file mode 100644
index 00000000000..647f7beb1b0
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_twsi.c
@@ -0,0 +1,500 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/delay.h>
+#include <linux/pci.h>
+#include <linux/vmalloc.h>
+
+#include "qib.h"
+
+/*
+ * QLogic_IB "Two Wire Serial Interface" driver.
+ * Originally written for a not-quite-i2c serial eeprom, which is
+ * still used on some supported boards. Later boards have added a
+ * variety of other uses, most board-specific, so the bit-boffing
+ * part has been split off to this file, while the other parts
+ * have been moved to chip-specific files.
+ *
+ * We have also dropped all pretense of fully generic (e.g. pretend
+ * we don't know whether '1' is the higher voltage) interface, as
+ * the restrictions of the generic i2c interface (e.g. no access from
+ * driver itself) make it unsuitable for this use.
+ */
+
+#define READ_CMD 1
+#define WRITE_CMD 0
+
+/**
+ * i2c_wait_for_writes - wait for a write
+ * @dd: the qlogic_ib device
+ *
+ * We use this instead of udelay directly, so we can make sure
+ * that previous register writes have been flushed all the way
+ * to the chip.  Since we are delaying anyway, the cost doesn't
+ * hurt, and makes the bit twiddling more regular
+ */
+static void i2c_wait_for_writes(struct qib_devdata *dd)
+{
+	/*
+	 * implicit read of EXTStatus is as good as explicit
+	 * read of scratch, if all we want to do is flush
+	 * writes.
+	 */
+	dd->f_gpio_mod(dd, 0, 0, 0);
+	rmb(); /* inlined, so prevent compiler reordering */
+}
+
+/*
+ * QSFP modules are allowed to hold SCL low for 500uSec. Allow twice that
+ * for "almost compliant" modules
+ */
+#define SCL_WAIT_USEC 1000
+
+/* BUF_WAIT is time bus must be free between STOP or ACK and to next START.
+ * Should be 20, but some chips need more.
+ */
+#define TWSI_BUF_WAIT_USEC 60
+
+static void scl_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	udelay(1);
+
+	mask = 1UL << dd->gpio_scl_num;
+
+	/* SCL is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	/*
+	 * Allow for slow slaves by simple
+	 * delay for falling edge, sampling on rise.
+	 */
+	if (!bit)
+		udelay(2);
+	else {
+		int rise_usec;
+		for (rise_usec = SCL_WAIT_USEC; rise_usec > 0; rise_usec -= 2) {
+			if (mask & dd->f_gpio_mod(dd, 0, 0, 0))
+				break;
+			udelay(2);
+		}
+		if (rise_usec <= 0)
+			qib_dev_err(dd, "SCL interface stuck low > %d uSec\n",
+				    SCL_WAIT_USEC);
+	}
+	i2c_wait_for_writes(dd);
+}
+
+static void sda_out(struct qib_devdata *dd, u8 bit)
+{
+	u32 mask;
+
+	mask = 1UL << dd->gpio_sda_num;
+
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, bit ? 0 : mask, mask);
+
+	i2c_wait_for_writes(dd);
+	udelay(2);
+}
+
+static u8 sda_in(struct qib_devdata *dd, int wait)
+{
+	int bnum;
+	u32 read_val, mask;
+
+	bnum = dd->gpio_sda_num;
+	mask = (1UL << bnum);
+	/* SDA is meant to be bare-drain, so never set "OUT", just DIR */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+	read_val = dd->f_gpio_mod(dd, 0, 0, 0);
+	if (wait)
+		i2c_wait_for_writes(dd);
+	return (read_val & mask) >> bnum;
+}
+
+/**
+ * i2c_ackrcv - see if ack following write is true
+ * @dd: the qlogic_ib device
+ */
+static int i2c_ackrcv(struct qib_devdata *dd)
+{
+	u8 ack_received;
+
+	/* AT ENTRY SCL = LOW */
+	/* change direction, ignore data */
+	ack_received = sda_in(dd, 1);
+	scl_out(dd, 1);
+	ack_received = sda_in(dd, 1) == 0;
+	scl_out(dd, 0);
+	return ack_received;
+}
+
+static void stop_cmd(struct qib_devdata *dd);
+
+/**
+ * rd_byte - read a byte, sending STOP on last, else ACK
+ * @dd: the qlogic_ib device
+ *
+ * Returns byte shifted out of device
+ */
+static int rd_byte(struct qib_devdata *dd, int last)
+{
+	int bit_cntr, data;
+
+	data = 0;
+
+	for (bit_cntr = 7; bit_cntr >= 0; --bit_cntr) {
+		data <<= 1;
+		scl_out(dd, 1);
+		data |= sda_in(dd, 0);
+		scl_out(dd, 0);
+	}
+	if (last) {
+		scl_out(dd, 1);
+		stop_cmd(dd);
+	} else {
+		sda_out(dd, 0);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+		sda_out(dd, 1);
+	}
+	return data;
+}
+
+/**
+ * wr_byte - write a byte, one bit at a time
+ * @dd: the qlogic_ib device
+ * @data: the byte to write
+ *
+ * Returns 0 if we got the following ack, otherwise 1
+ */
+static int wr_byte(struct qib_devdata *dd, u8 data)
+{
+	int bit_cntr;
+	u8 bit;
+
+	for (bit_cntr = 7; bit_cntr >= 0; bit_cntr--) {
+		bit = (data >> bit_cntr) & 1;
+		sda_out(dd, bit);
+		scl_out(dd, 1);
+		scl_out(dd, 0);
+	}
+	return (!i2c_ackrcv(dd)) ? 1 : 0;
+}
+
+/*
+ * issue TWSI start sequence:
+ * (both clock/data high, clock high, data low while clock is high)
+ */
+static void start_seq(struct qib_devdata *dd)
+{
+	sda_out(dd, 1);
+	scl_out(dd, 1);
+	sda_out(dd, 0);
+	udelay(1);
+	scl_out(dd, 0);
+}
+
+/**
+ * stop_seq - transmit the stop sequence
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_seq(struct qib_devdata *dd)
+{
+	scl_out(dd, 0);
+	sda_out(dd, 0);
+	scl_out(dd, 1);
+	sda_out(dd, 1);
+}
+
+/**
+ * stop_cmd - transmit the stop condition
+ * @dd: the qlogic_ib device
+ *
+ * (both clock/data low, clock high, data high while clock is high)
+ */
+static void stop_cmd(struct qib_devdata *dd)
+{
+	stop_seq(dd);
+	udelay(TWSI_BUF_WAIT_USEC);
+}
+
+/**
+ * qib_twsi_reset - reset I2C communication
+ * @dd: the qlogic_ib device
+ */
+
+int qib_twsi_reset(struct qib_devdata *dd)
+{
+	int clock_cycles_left = 9;
+	int was_high = 0;
+	u32 pins, mask;
+
+	/* Both SCL and SDA should be high. If not, there
+	 * is something wrong.
+	 */
+	mask = (1UL << dd->gpio_scl_num) | (1UL << dd->gpio_sda_num);
+
+	/*
+	 * Force pins to desired innocuous state.
+	 * This is the default power-on state with out=0 and dir=0,
+	 * So tri-stated and should be floating high (barring HW problems)
+	 */
+	dd->f_gpio_mod(dd, 0, 0, mask);
+
+	/*
+	 * Clock nine times to get all listeners into a sane state.
+	 * If SDA does not go high at any point, we are wedged.
+	 * One vendor recommends then issuing START followed by STOP.
+	 * we cannot use our "normal" functions to do that, because
+	 * if SCL drops between them, another vendor's part will
+	 * wedge, dropping SDA and keeping it low forever, at the end of
+	 * the next transaction (even if it was not the device addressed).
+	 * So our START and STOP take place with SCL held high.
+	 */
+	while (clock_cycles_left--) {
+		scl_out(dd, 0);
+		scl_out(dd, 1);
+		/* Note if SDA is high, but keep clocking to sync slave */
+		was_high |= sda_in(dd, 0);
+	}
+
+	if (was_high) {
+		/*
+		 * We saw a high, which we hope means the slave is sync'd.
+		 * Issue START, STOP, pause for T_BUF.
+		 */
+
+		pins = dd->f_gpio_mod(dd, 0, 0, 0);
+		if ((pins & mask) != mask)
+			qib_dev_err(dd, "GPIO pins not at rest: %d\n",
+				    pins & mask);
+		/* Drop SDA to issue START */
+		udelay(1); /* Guarantee .6 uSec setup */
+		sda_out(dd, 0);
+		udelay(1); /* Guarantee .6 uSec hold */
+		/* At this point, SCL is high, SDA low. Raise SDA for STOP */
+		sda_out(dd, 1);
+		udelay(TWSI_BUF_WAIT_USEC);
+	}
+
+	return !was_high;
+}
+
+#define QIB_TWSI_START 0x100
+#define QIB_TWSI_STOP 0x200
+
+/* Write byte to TWSI, optionally prefixed with START or suffixed with
+ * STOP.
+ * returns 0 if OK (ACK received), else != 0
+ */
+static int qib_twsi_wr(struct qib_devdata *dd, int data, int flags)
+{
+	int ret = 1;
+	if (flags & QIB_TWSI_START)
+		start_seq(dd);
+
+	ret = wr_byte(dd, data); /* Leaves SCL low (from i2c_ackrcv()) */
+
+	if (flags & QIB_TWSI_STOP)
+		stop_cmd(dd);
+	return ret;
+}
+
+/* Added functionality for IBA7220-based cards */
+#define QIB_TEMP_DEV 0x98
+
+/*
+ * qib_twsi_blk_rd
+ * Formerly called qib_eeprom_internal_read, and only used for eeprom,
+ * but now the general interface for data transfer from twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device from which data should
+ * be read.
+ */
+int qib_twsi_blk_rd(struct qib_devdata *dd, int dev, int addr,
+		    void *buffer, int len)
+{
+	int ret;
+	u8 *bp = buffer;
+
+	ret = 1;
+
+	if (dev == QIB_TWSI_NO_DEV) {
+		/* legacy not-really-I2C */
+		addr = (addr << 1) | READ_CMD;
+		ret = qib_twsi_wr(dd, addr, QIB_TWSI_START);
+	} else {
+		/* Actual I2C */
+		ret = qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START);
+		if (ret) {
+			stop_cmd(dd);
+			ret = 1;
+			goto bail;
+		}
+		/*
+		 * SFF spec claims we do _not_ stop after the addr
+		 * but simply issue a start with the "read" dev-addr.
+		 * Since we are implicitely waiting for ACK here,
+		 * we need t_buf (nominally 20uSec) before that start,
+		 * and cannot rely on the delay built in to the STOP
+		 */
+		ret = qib_twsi_wr(dd, addr, 0);
+		udelay(TWSI_BUF_WAIT_USEC);
+
+		if (ret) {
+			qib_dev_err(dd,
+				"Failed to write interface read addr %02X\n",
+				addr);
+			ret = 1;
+			goto bail;
+		}
+		ret = qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START);
+	}
+	if (ret) {
+		stop_cmd(dd);
+		ret = 1;
+		goto bail;
+	}
+
+	/*
+	 * block devices keeps clocking data out as long as we ack,
+	 * automatically incrementing the address. Some have "pages"
+	 * whose boundaries will not be crossed, but the handling
+	 * of these is left to the caller, who is in a better
+	 * position to know.
+	 */
+	while (len-- > 0) {
+		/*
+		 * Get and store data, sending ACK if length remaining,
+		 * else STOP
+		 */
+		*bp++ = rd_byte(dd, !len);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/*
+ * qib_twsi_blk_wr
+ * Formerly called qib_eeprom_internal_write, and only used for eeprom,
+ * but now the general interface for data transfer to twsi devices.
+ * One vestige of its former role is that it recognizes a device
+ * QIB_TWSI_NO_DEV and does the correct operation for the legacy part,
+ * which responded to all TWSI device codes, interpreting them as
+ * address within device. On all other devices found on board handled by
+ * this driver, the device is followed by a one-byte "address" which selects
+ * the "register" or "offset" within the device to which data should
+ * be written.
+ */
+int qib_twsi_blk_wr(struct qib_devdata *dd, int dev, int addr,
+		    const void *buffer, int len)
+{
+	int sub_len;
+	const u8 *bp = buffer;
+	int max_wait_time, i;
+	int ret;
+	ret = 1;
+
+	while (len > 0) {
+		if (dev == QIB_TWSI_NO_DEV) {
+			if (qib_twsi_wr(dd, (addr << 1) | WRITE_CMD,
+					QIB_TWSI_START)) {
+				goto failed_write;
+			}
+		} else {
+			/* Real I2C */
+			if (qib_twsi_wr(dd, dev | WRITE_CMD, QIB_TWSI_START))
+				goto failed_write;
+			ret = qib_twsi_wr(dd, addr, 0);
+			if (ret) {
+				qib_dev_err(dd,
+					"Failed to write interface write addr %02X\n",
+					addr);
+				goto failed_write;
+			}
+		}
+
+		sub_len = min(len, 4);
+		addr += sub_len;
+		len -= sub_len;
+
+		for (i = 0; i < sub_len; i++)
+			if (qib_twsi_wr(dd, *bp++, 0))
+				goto failed_write;
+
+		stop_cmd(dd);
+
+		/*
+		 * Wait for write complete by waiting for a successful
+		 * read (the chip replies with a zero after the write
+		 * cmd completes, and before it writes to the eeprom.
+		 * The startcmd for the read will fail the ack until
+		 * the writes have completed.   We do this inline to avoid
+		 * the debug prints that are in the real read routine
+		 * if the startcmd fails.
+		 * We also use the proper device address, so it doesn't matter
+		 * whether we have real eeprom_dev. Legacy likes any address.
+		 */
+		max_wait_time = 100;
+		while (qib_twsi_wr(dd, dev | READ_CMD, QIB_TWSI_START)) {
+			stop_cmd(dd);
+			if (!--max_wait_time)
+				goto failed_write;
+		}
+		/* now read (and ignore) the resulting byte */
+		rd_byte(dd, 1);
+	}
+
+	ret = 0;
+	goto bail;
+
+failed_write:
+	stop_cmd(dd);
+	ret = 1;
+
+bail:
+	return ret;
+}
diff --git a/drivers/infiniband/hw/qib/qib_tx.c b/drivers/infiniband/hw/qib/qib_tx.c
new file mode 100644
index 00000000000..31d3561400a
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_tx.c
@@ -0,0 +1,571 @@
+/*
+ * Copyright (c) 2008, 2009, 2010 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/spinlock.h>
+#include <linux/pci.h>
+#include <linux/io.h>
+#include <linux/delay.h>
+#include <linux/netdevice.h>
+#include <linux/vmalloc.h>
+#include <linux/moduleparam.h>
+
+#include "qib.h"
+
+static unsigned qib_hol_timeout_ms = 3000;
+module_param_named(hol_timeout_ms, qib_hol_timeout_ms, uint, S_IRUGO);
+MODULE_PARM_DESC(hol_timeout_ms,
+		 "duration of user app suspension after link failure");
+
+unsigned qib_sdma_fetch_arb = 1;
+module_param_named(fetch_arb, qib_sdma_fetch_arb, uint, S_IRUGO);
+MODULE_PARM_DESC(fetch_arb, "IBA7220: change SDMA descriptor arbitration");
+
+/**
+ * qib_disarm_piobufs - cancel a range of PIO buffers
+ * @dd: the qlogic_ib device
+ * @first: the first PIO buffer to cancel
+ * @cnt: the number of PIO buffers to cancel
+ *
+ * Cancel a range of PIO buffers. Used at user process close,
+ * in case it died while writing to a PIO buffer.
+ */
+void qib_disarm_piobufs(struct qib_devdata *dd, unsigned first, unsigned cnt)
+{
+	unsigned long flags;
+	unsigned i;
+	unsigned last;
+
+	last = first + cnt;
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = first; i < last; i++) {
+		__clear_bit(i, dd->pio_need_disarm);
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * This is called by a user process when it sees the DISARM_BUFS event
+ * bit is set.
+ */
+int qib_disarm_piobufs_ifneeded(struct qib_ctxtdata *rcd)
+{
+	struct qib_devdata *dd = rcd->dd;
+	unsigned i;
+	unsigned last;
+	unsigned n = 0;
+
+	last = rcd->pio_base + rcd->piocnt;
+	/*
+	 * Don't need uctxt_lock here, since user has called in to us.
+	 * Clear at start in case more interrupts set bits while we
+	 * are disarming
+	 */
+	if (rcd->user_event_mask) {
+		/*
+		 * subctxt_cnt is 0 if not shared, so do base
+		 * separately, first, then remaining subctxt, if any
+		 */
+		clear_bit(_QIB_EVENT_DISARM_BUFS_BIT, &rcd->user_event_mask[0]);
+		for (i = 1; i < rcd->subctxt_cnt; i++)
+			clear_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				  &rcd->user_event_mask[i]);
+	}
+	spin_lock_irq(&dd->pioavail_lock);
+	for (i = rcd->pio_base; i < last; i++) {
+		if (__test_and_clear_bit(i, dd->pio_need_disarm)) {
+			n++;
+			dd->f_sendctrl(rcd->ppd, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+	}
+	spin_unlock_irq(&dd->pioavail_lock);
+	return 0;
+}
+
+static struct qib_pportdata *is_sdma_buf(struct qib_devdata *dd, unsigned i)
+{
+	struct qib_pportdata *ppd;
+	unsigned pidx;
+
+	for (pidx = 0; pidx < dd->num_pports; pidx++) {
+		ppd = dd->pport + pidx;
+		if (i >= ppd->sdma_state.first_sendbuf &&
+		    i < ppd->sdma_state.last_sendbuf)
+			return ppd;
+	}
+	return NULL;
+}
+
+/*
+ * Return true if send buffer is being used by a user context.
+ * Sets  _QIB_EVENT_DISARM_BUFS_BIT in user_event_mask as a side effect
+ */
+static int find_ctxt(struct qib_devdata *dd, unsigned bufn)
+{
+	struct qib_ctxtdata *rcd;
+	unsigned ctxt;
+	int ret = 0;
+
+	spin_lock(&dd->uctxt_lock);
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		rcd = dd->rcd[ctxt];
+		if (!rcd || bufn < rcd->pio_base ||
+		    bufn >= rcd->pio_base + rcd->piocnt)
+			continue;
+		if (rcd->user_event_mask) {
+			int i;
+			/*
+			 * subctxt_cnt is 0 if not shared, so do base
+			 * separately, first, then remaining subctxt, if any
+			 */
+			set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+				&rcd->user_event_mask[0]);
+			for (i = 1; i < rcd->subctxt_cnt; i++)
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[i]);
+		}
+		ret = 1;
+		break;
+	}
+	spin_unlock(&dd->uctxt_lock);
+
+	return ret;
+}
+
+/*
+ * Disarm a set of send buffers.  If the buffer might be actively being
+ * written to, mark the buffer to be disarmed later when it is not being
+ * written to.
+ *
+ * This should only be called from the IRQ error handler.
+ */
+void qib_disarm_piobufs_set(struct qib_devdata *dd, unsigned long *mask,
+			    unsigned cnt)
+{
+	struct qib_pportdata *ppd, *pppd[QIB_MAX_IB_PORTS];
+	unsigned i;
+	unsigned long flags;
+
+	for (i = 0; i < dd->num_pports; i++)
+		pppd[i] = NULL;
+
+	for (i = 0; i < cnt; i++) {
+		int which;
+		if (!test_bit(i, mask))
+			continue;
+		/*
+		 * If the buffer is owned by the DMA hardware,
+		 * reset the DMA engine.
+		 */
+		ppd = is_sdma_buf(dd, i);
+		if (ppd) {
+			pppd[ppd->port] = ppd;
+			continue;
+		}
+		/*
+		 * If the kernel is writing the buffer or the buffer is
+		 * owned by a user process, we can't clear it yet.
+		 */
+		spin_lock_irqsave(&dd->pioavail_lock, flags);
+		if (test_bit(i, dd->pio_writing) ||
+		    (!test_bit(i << 1, dd->pioavailkernel) &&
+		     find_ctxt(dd, i))) {
+			__set_bit(i, dd->pio_need_disarm);
+			which = 0;
+		} else {
+			which = 1;
+			dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(i));
+		}
+		spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+	}
+
+	/* do cancel_sends once per port that had sdma piobufs in error */
+	for (i = 0; i < dd->num_pports; i++)
+		if (pppd[i])
+			qib_cancel_sends(pppd[i]);
+}
+
+/**
+ * update_send_bufs - update shadow copy of the PIO availability map
+ * @dd: the qlogic_ib device
+ *
+ * called whenever our local copy indicates we have run out of send buffers
+ */
+static void update_send_bufs(struct qib_devdata *dd)
+{
+	unsigned long flags;
+	unsigned i;
+	const unsigned piobregs = dd->pioavregs;
+
+	/*
+	 * If the generation (check) bits have changed, then we update the
+	 * busy bit for the corresponding PIO buffer.  This algorithm will
+	 * modify positions to the value they already have in some cases
+	 * (i.e., no change), but it's faster than changing only the bits
+	 * that have changed.
+	 *
+	 * We would like to do this atomicly, to avoid spinlocks in the
+	 * critical send path, but that's not really possible, given the
+	 * type of changes, and that this routine could be called on
+	 * multiple cpu's simultaneously, so we lock in this routine only,
+	 * to avoid conflicting updates; all we change is the shadow, and
+	 * it's a single 64 bit memory location, so by definition the update
+	 * is atomic in terms of what other cpu's can see in testing the
+	 * bits.  The spin_lock overhead isn't too bad, since it only
+	 * happens when all buffers are in use, so only cpu overhead, not
+	 * latency or bandwidth is affected.
+	 */
+	if (!dd->pioavailregs_dma)
+		return;
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	for (i = 0; i < piobregs; i++) {
+		u64 pchbusy, pchg, piov, pnew;
+
+		piov = le64_to_cpu(dd->pioavailregs_dma[i]);
+		pchg = dd->pioavailkernel[i] &
+			~(dd->pioavailshadow[i] ^ piov);
+		pchbusy = pchg << QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT;
+		if (pchg && (pchbusy & dd->pioavailshadow[i])) {
+			pnew = dd->pioavailshadow[i] & ~pchbusy;
+			pnew |= piov & pchbusy;
+			dd->pioavailshadow[i] = pnew;
+		}
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/*
+ * Debugging code and stats updates if no pio buffers available.
+ */
+static noinline void no_send_bufs(struct qib_devdata *dd)
+{
+	dd->upd_pio_shadow = 1;
+
+	/* not atomic, but if we lose a stat count in a while, that's OK */
+	qib_stats.sps_nopiobufs++;
+}
+
+/*
+ * Common code for normal driver send buffer allocation, and reserved
+ * allocation.
+ *
+ * Do appropriate marking as busy, etc.
+ * Returns buffer pointer if one is found, otherwise NULL.
+ */
+u32 __iomem *qib_getsendbuf_range(struct qib_devdata *dd, u32 *pbufnum,
+				  u32 first, u32 last)
+{
+	unsigned i, j, updated = 0;
+	unsigned nbufs;
+	unsigned long flags;
+	unsigned long *shadow = dd->pioavailshadow;
+	u32 __iomem *buf;
+
+	if (!(dd->flags & QIB_PRESENT))
+		return NULL;
+
+	nbufs = last - first + 1; /* number in range to check */
+	if (dd->upd_pio_shadow) {
+update_shadow:
+		/*
+		 * Minor optimization.  If we had no buffers on last call,
+		 * start out by doing the update; continue and do scan even
+		 * if no buffers were updated, to be paranoid.
+		 */
+		update_send_bufs(dd);
+		updated++;
+	}
+	i = first;
+	/*
+	 * While test_and_set_bit() is atomic, we do that and then the
+	 * change_bit(), and the pair is not.  See if this is the cause
+	 * of the remaining armlaunch errors.
+	 */
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	if (dd->last_pio >= first && dd->last_pio <= last)
+		i = dd->last_pio + 1;
+	if (!first)
+		/* adjust to min possible  */
+		nbufs = last - dd->min_kernel_pio + 1;
+	for (j = 0; j < nbufs; j++, i++) {
+		if (i > last)
+			i = !first ? dd->min_kernel_pio : first;
+		if (__test_and_set_bit((2 * i) + 1, shadow))
+			continue;
+		/* flip generation bit */
+		__change_bit(2 * i, shadow);
+		/* remember that the buffer can be written to now */
+		__set_bit(i, dd->pio_writing);
+		if (!first && first != last) /* first == last on VL15, avoid */
+			dd->last_pio = i;
+		break;
+	}
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	if (j == nbufs) {
+		if (!updated)
+			/*
+			 * First time through; shadow exhausted, but may be
+			 * buffers available, try an update and then rescan.
+			 */
+			goto update_shadow;
+		no_send_bufs(dd);
+		buf = NULL;
+	} else {
+		if (i < dd->piobcnt2k)
+			buf = (u32 __iomem *)(dd->pio2kbase +
+				i * dd->palign);
+		else if (i < dd->piobcnt2k + dd->piobcnt4k || !dd->piovl15base)
+			buf = (u32 __iomem *)(dd->pio4kbase +
+				(i - dd->piobcnt2k) * dd->align4k);
+		else
+			buf = (u32 __iomem *)(dd->piovl15base +
+				(i - (dd->piobcnt2k + dd->piobcnt4k)) *
+				dd->align4k);
+		if (pbufnum)
+			*pbufnum = i;
+		dd->upd_pio_shadow = 0;
+	}
+
+	return buf;
+}
+
+/*
+ * Record that the caller is finished writing to the buffer so we don't
+ * disarm it while it is being written and disarm it now if needed.
+ */
+void qib_sendbuf_done(struct qib_devdata *dd, unsigned n)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	__clear_bit(n, dd->pio_writing);
+	if (__test_and_clear_bit(n, dd->pio_need_disarm))
+		dd->f_sendctrl(dd->pport, QIB_SENDCTRL_DISARM_BUF(n));
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+}
+
+/**
+ * qib_chg_pioavailkernel - change which send buffers are available for kernel
+ * @dd: the qlogic_ib device
+ * @start: the starting send buffer number
+ * @len: the number of send buffers
+ * @avail: true if the buffers are available for kernel use, false otherwise
+ */
+void qib_chg_pioavailkernel(struct qib_devdata *dd, unsigned start,
+	unsigned len, u32 avail, struct qib_ctxtdata *rcd)
+{
+	unsigned long flags;
+	unsigned end;
+	unsigned ostart = start;
+
+	/* There are two bits per send buffer (busy and generation) */
+	start *= 2;
+	end = start + len * 2;
+
+	spin_lock_irqsave(&dd->pioavail_lock, flags);
+	/* Set or clear the busy bit in the shadow. */
+	while (start < end) {
+		if (avail) {
+			unsigned long dma;
+			int i;
+
+			/*
+			 * The BUSY bit will never be set, because we disarm
+			 * the user buffers before we hand them back to the
+			 * kernel.  We do have to make sure the generation
+			 * bit is set correctly in shadow, since it could
+			 * have changed many times while allocated to user.
+			 * We can't use the bitmap functions on the full
+			 * dma array because it is always little-endian, so
+			 * we have to flip to host-order first.
+			 * BITS_PER_LONG is slightly wrong, since it's
+			 * always 64 bits per register in chip...
+			 * We only work on 64 bit kernels, so that's OK.
+			 */
+			i = start / BITS_PER_LONG;
+			__clear_bit(QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT + start,
+				    dd->pioavailshadow);
+			dma = (unsigned long)
+				le64_to_cpu(dd->pioavailregs_dma[i]);
+			if (test_bit((QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+				      start) % BITS_PER_LONG, &dma))
+				__set_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT +
+					  start, dd->pioavailshadow);
+			else
+				__clear_bit(QLOGIC_IB_SENDPIOAVAIL_CHECK_SHIFT
+					    + start, dd->pioavailshadow);
+			__set_bit(start, dd->pioavailkernel);
+			if ((start >> 1) < dd->min_kernel_pio)
+				dd->min_kernel_pio = start >> 1;
+		} else {
+			__set_bit(start + QLOGIC_IB_SENDPIOAVAIL_BUSY_SHIFT,
+				  dd->pioavailshadow);
+			__clear_bit(start, dd->pioavailkernel);
+			if ((start >> 1) > dd->min_kernel_pio)
+				dd->min_kernel_pio = start >> 1;
+		}
+		start += 2;
+	}
+
+	if (dd->min_kernel_pio > 0 && dd->last_pio < dd->min_kernel_pio - 1)
+		dd->last_pio = dd->min_kernel_pio - 1;
+	spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+
+	dd->f_txchk_change(dd, ostart, len, avail, rcd);
+}
+
+/*
+ * Flush all sends that might be in the ready to send state, as well as any
+ * that are in the process of being sent.  Used whenever we need to be
+ * sure the send side is idle.  Cleans up all buffer state by canceling
+ * all pio buffers, and issuing an abort, which cleans up anything in the
+ * launch fifo.  The cancel is superfluous on some chip versions, but
+ * it's safer to always do it.
+ * PIOAvail bits are updated by the chip as if a normal send had happened.
+ */
+void qib_cancel_sends(struct qib_pportdata *ppd)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct qib_ctxtdata *rcd;
+	unsigned long flags;
+	unsigned ctxt;
+	unsigned i;
+	unsigned last;
+
+	/*
+	 * Tell PSM to disarm buffers again before trying to reuse them.
+	 * We need to be sure the rcd doesn't change out from under us
+	 * while we do so.  We hold the two locks sequentially.  We might
+	 * needlessly set some need_disarm bits as a result, if the
+	 * context is closed after we release the uctxt_lock, but that's
+	 * fairly benign, and safer than nesting the locks.
+	 */
+	for (ctxt = dd->first_user_ctxt; ctxt < dd->cfgctxts; ctxt++) {
+		spin_lock_irqsave(&dd->uctxt_lock, flags);
+		rcd = dd->rcd[ctxt];
+		if (rcd && rcd->ppd == ppd) {
+			last = rcd->pio_base + rcd->piocnt;
+			if (rcd->user_event_mask) {
+				/*
+				 * subctxt_cnt is 0 if not shared, so do base
+				 * separately, first, then remaining subctxt,
+				 * if any
+				 */
+				set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+					&rcd->user_event_mask[0]);
+				for (i = 1; i < rcd->subctxt_cnt; i++)
+					set_bit(_QIB_EVENT_DISARM_BUFS_BIT,
+						&rcd->user_event_mask[i]);
+			}
+			i = rcd->pio_base;
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+			spin_lock_irqsave(&dd->pioavail_lock, flags);
+			for (; i < last; i++)
+				__set_bit(i, dd->pio_need_disarm);
+			spin_unlock_irqrestore(&dd->pioavail_lock, flags);
+		} else
+			spin_unlock_irqrestore(&dd->uctxt_lock, flags);
+	}
+
+	if (!(dd->flags & QIB_HAS_SEND_DMA))
+		dd->f_sendctrl(ppd, QIB_SENDCTRL_DISARM_ALL |
+				    QIB_SENDCTRL_FLUSH);
+}
+
+/*
+ * Force an update of in-memory copy of the pioavail registers, when
+ * needed for any of a variety of reasons.
+ * If already off, this routine is a nop, on the assumption that the
+ * caller (or set of callers) will "do the right thing".
+ * This is a per-device operation, so just the first port.
+ */
+void qib_force_pio_avail_update(struct qib_devdata *dd)
+{
+	dd->f_sendctrl(dd->pport, QIB_SENDCTRL_AVAIL_BLIP);
+}
+
+void qib_hol_down(struct qib_pportdata *ppd)
+{
+	/*
+	 * Cancel sends when the link goes DOWN so that we aren't doing it
+	 * at INIT when we might be trying to send SMI packets.
+	 */
+	if (!(ppd->lflags & QIBL_IB_AUTONEG_INPROG))
+		qib_cancel_sends(ppd);
+}
+
+/*
+ * Link is at INIT.
+ * We start the HoL timer so we can detect stuck packets blocking SMP replies.
+ * Timer may already be running, so use mod_timer, not add_timer.
+ */
+void qib_hol_init(struct qib_pportdata *ppd)
+{
+	if (ppd->hol_state != QIB_HOL_INIT) {
+		ppd->hol_state = QIB_HOL_INIT;
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}
+
+/*
+ * Link is up, continue any user processes, and ensure timer
+ * is a nop, if running.  Let timer keep running, if set; it
+ * will nop when it sees the link is up.
+ */
+void qib_hol_up(struct qib_pportdata *ppd)
+{
+	ppd->hol_state = QIB_HOL_UP;
+}
+
+/*
+ * This is only called via the timer.
+ */
+void qib_hol_event(unsigned long opaque)
+{
+	struct qib_pportdata *ppd = (struct qib_pportdata *)opaque;
+
+	/* If hardware error, etc, skip. */
+	if (!(ppd->dd->flags & QIB_INITTED))
+		return;
+
+	if (ppd->hol_state != QIB_HOL_UP) {
+		/*
+		 * Try to flush sends in case a stuck packet is blocking
+		 * SMP replies.
+		 */
+		qib_hol_down(ppd);
+		mod_timer(&ppd->hol_timer,
+			  jiffies + msecs_to_jiffies(qib_hol_timeout_ms));
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_uc.c b/drivers/infiniband/hw/qib/qib_uc.c
new file mode 100644
index 00000000000..aa3a8035bb6
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_uc.c
@@ -0,0 +1,536 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009, 2010 QLogic Corporation.
+ * All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include "qib.h"
+
+/* cut down ridiculously long IB macro names */
+#define OP(x) IB_OPCODE_UC_##x
+
+/**
+ * qib_make_uc_req - construct a request packet (SEND, RDMA write)
+ * @qp: a pointer to the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_uc_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 hwords;
+	u32 bth0;
+	u32 len;
+	u32 pmtu = qp->pmtu;
+	int ret = 0;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	ohdr = &qp->s_hdr->u.oth;
+	if (qp->remote_ah_attr.ah_flags & IB_AH_GRH)
+		ohdr = &qp->s_hdr->u.l.oth;
+
+	/* header size in 32-bit words LRH+BTH = (8+12)/4. */
+	hwords = 5;
+	bth0 = 0;
+
+	/* Get the next send request. */
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	qp->s_wqe = NULL;
+	switch (qp->s_state) {
+	default:
+		if (!(ib_qib_state_ops[qp->state] &
+		    QIB_PROCESS_NEXT_SEND_OK))
+			goto bail;
+		/* Check if send work queue is empty. */
+		if (qp->s_cur == qp->s_head)
+			goto bail;
+		/*
+		 * Start a new request.
+		 */
+		wqe->psn = qp->s_next_psn;
+		qp->s_psn = qp->s_next_psn;
+		qp->s_sge.sge = wqe->sg_list[0];
+		qp->s_sge.sg_list = wqe->sg_list + 1;
+		qp->s_sge.num_sge = wqe->wr.num_sge;
+		qp->s_sge.total_len = wqe->length;
+		len = wqe->length;
+		qp->s_len = len;
+		switch (wqe->wr.opcode) {
+		case IB_WR_SEND:
+		case IB_WR_SEND_WITH_IMM:
+			if (len > pmtu) {
+				qp->s_state = OP(SEND_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_SEND)
+				qp->s_state = OP(SEND_ONLY);
+			else {
+				qp->s_state =
+					OP(SEND_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the BTH */
+				ohdr->u.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+			}
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		case IB_WR_RDMA_WRITE:
+		case IB_WR_RDMA_WRITE_WITH_IMM:
+			ohdr->u.rc.reth.vaddr =
+				cpu_to_be64(wqe->wr.wr.rdma.remote_addr);
+			ohdr->u.rc.reth.rkey =
+				cpu_to_be32(wqe->wr.wr.rdma.rkey);
+			ohdr->u.rc.reth.length = cpu_to_be32(len);
+			hwords += sizeof(struct ib_reth) / 4;
+			if (len > pmtu) {
+				qp->s_state = OP(RDMA_WRITE_FIRST);
+				len = pmtu;
+				break;
+			}
+			if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+				qp->s_state = OP(RDMA_WRITE_ONLY);
+			else {
+				qp->s_state =
+					OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE);
+				/* Immediate data comes after the RETH */
+				ohdr->u.rc.imm_data = wqe->wr.ex.imm_data;
+				hwords += 1;
+				if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+					bth0 |= IB_BTH_SOLICITED;
+			}
+			qp->s_wqe = wqe;
+			if (++qp->s_cur >= qp->s_size)
+				qp->s_cur = 0;
+			break;
+
+		default:
+			goto bail;
+		}
+		break;
+
+	case OP(SEND_FIRST):
+		qp->s_state = OP(SEND_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_SEND)
+			qp->s_state = OP(SEND_LAST);
+		else {
+			qp->s_state = OP(SEND_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+		}
+		if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+			bth0 |= IB_BTH_SOLICITED;
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+		qp->s_state = OP(RDMA_WRITE_MIDDLE);
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		len = qp->s_len;
+		if (len > pmtu) {
+			len = pmtu;
+			break;
+		}
+		if (wqe->wr.opcode == IB_WR_RDMA_WRITE)
+			qp->s_state = OP(RDMA_WRITE_LAST);
+		else {
+			qp->s_state =
+				OP(RDMA_WRITE_LAST_WITH_IMMEDIATE);
+			/* Immediate data comes after the BTH */
+			ohdr->u.imm_data = wqe->wr.ex.imm_data;
+			hwords += 1;
+			if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+				bth0 |= IB_BTH_SOLICITED;
+		}
+		qp->s_wqe = wqe;
+		if (++qp->s_cur >= qp->s_size)
+			qp->s_cur = 0;
+		break;
+	}
+	qp->s_len -= len;
+	qp->s_hdrwords = hwords;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_cur_size = len;
+	qib_make_ruc_header(qp, ohdr, bth0 | (qp->s_state << 24),
+			    qp->s_next_psn++ & QIB_PSN_MASK);
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_uc_rcv - handle an incoming UC packet
+ * @ibp: the port the packet came in on
+ * @hdr: the header of the packet
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the length of the packet
+ * @qp: the QP for this packet.
+ *
+ * This is called from qib_qp_rcv() to process an incoming UC packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	u32 opcode;
+	u32 hdrsize;
+	u32 psn;
+	u32 pad;
+	struct ib_wc wc;
+	u32 pmtu = qp->pmtu;
+	struct ib_reth *reth;
+	int ret;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12;       /* LRH + BTH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12;  /* LRH + GRH + BTH */
+	}
+
+	opcode = be32_to_cpu(ohdr->bth[0]);
+	if (qib_ruc_check_hdr(ibp, hdr, has_grh, qp, opcode))
+		return;
+
+	psn = be32_to_cpu(ohdr->bth[2]);
+	opcode >>= 24;
+
+	/* Compare the PSN verses the expected PSN. */
+	if (unlikely(qib_cmp24(psn, qp->r_psn) != 0)) {
+		/*
+		 * Handle a sequence error.
+		 * Silently drop any current message.
+		 */
+		qp->r_psn = psn;
+inv:
+		if (qp->r_state == OP(SEND_FIRST) ||
+		    qp->r_state == OP(SEND_MIDDLE)) {
+			set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
+			qp->r_sge.num_sge = 0;
+		} else
+			qib_put_ss(&qp->r_sge);
+		qp->r_state = OP(SEND_LAST);
+		switch (opcode) {
+		case OP(SEND_FIRST):
+		case OP(SEND_ONLY):
+		case OP(SEND_ONLY_WITH_IMMEDIATE):
+			goto send_first;
+
+		case OP(RDMA_WRITE_FIRST):
+		case OP(RDMA_WRITE_ONLY):
+		case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE):
+			goto rdma_first;
+
+		default:
+			goto drop;
+		}
+	}
+
+	/* Check for opcode sequence errors. */
+	switch (qp->r_state) {
+	case OP(SEND_FIRST):
+	case OP(SEND_MIDDLE):
+		if (opcode == OP(SEND_MIDDLE) ||
+		    opcode == OP(SEND_LAST) ||
+		    opcode == OP(SEND_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_MIDDLE):
+		if (opcode == OP(RDMA_WRITE_MIDDLE) ||
+		    opcode == OP(RDMA_WRITE_LAST) ||
+		    opcode == OP(RDMA_WRITE_LAST_WITH_IMMEDIATE))
+			break;
+		goto inv;
+
+	default:
+		if (opcode == OP(SEND_FIRST) ||
+		    opcode == OP(SEND_ONLY) ||
+		    opcode == OP(SEND_ONLY_WITH_IMMEDIATE) ||
+		    opcode == OP(RDMA_WRITE_FIRST) ||
+		    opcode == OP(RDMA_WRITE_ONLY) ||
+		    opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE))
+			break;
+		goto inv;
+	}
+
+	if (qp->state == IB_QPS_RTR && !(qp->r_flags & QIB_R_COMM_EST)) {
+		qp->r_flags |= QIB_R_COMM_EST;
+		if (qp->ibqp.event_handler) {
+			struct ib_event ev;
+
+			ev.device = qp->ibqp.device;
+			ev.element.qp = &qp->ibqp;
+			ev.event = IB_EVENT_COMM_EST;
+			qp->ibqp.event_handler(&ev, qp->ibqp.qp_context);
+		}
+	}
+
+	/* OK, process the packet. */
+	switch (opcode) {
+	case OP(SEND_FIRST):
+	case OP(SEND_ONLY):
+	case OP(SEND_ONLY_WITH_IMMEDIATE):
+send_first:
+		if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
+			qp->r_sge = qp->s_rdma_read_sge;
+		else {
+			ret = qib_get_rwqe(qp, 0);
+			if (ret < 0)
+				goto op_err;
+			if (!ret)
+				goto drop;
+			/*
+			 * qp->s_rdma_read_sge will be the owner
+			 * of the mr references.
+			 */
+			qp->s_rdma_read_sge = qp->r_sge;
+		}
+		qp->r_rcv_len = 0;
+		if (opcode == OP(SEND_ONLY))
+			goto no_immediate_data;
+		else if (opcode == OP(SEND_ONLY_WITH_IMMEDIATE))
+			goto send_last_imm;
+		/* FALLTHROUGH */
+	case OP(SEND_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto rewind;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto rewind;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 0);
+		break;
+
+	case OP(SEND_LAST_WITH_IMMEDIATE):
+send_last_imm:
+		wc.ex.imm_data = ohdr->u.imm_data;
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		goto send_last;
+	case OP(SEND_LAST):
+no_immediate_data:
+		wc.ex.imm_data = 0;
+		wc.wc_flags = 0;
+send_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto rewind;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		wc.byte_len = tlen + qp->r_rcv_len;
+		if (unlikely(wc.byte_len > qp->r_len))
+			goto rewind;
+		wc.opcode = IB_WC_RECV;
+		qib_copy_sge(&qp->r_sge, data, tlen, 0);
+		qib_put_ss(&qp->s_rdma_read_sge);
+last_imm:
+		wc.wr_id = qp->r_wr_id;
+		wc.status = IB_WC_SUCCESS;
+		wc.qp = &qp->ibqp;
+		wc.src_qp = qp->remote_qpn;
+		wc.slid = qp->remote_ah_attr.dlid;
+		wc.sl = qp->remote_ah_attr.sl;
+		/* zero fields that are N/A */
+		wc.vendor_err = 0;
+		wc.pkey_index = 0;
+		wc.dlid_path_bits = 0;
+		wc.port_num = 0;
+		/* Signal completion event if the solicited bit is set. */
+		qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+			     (ohdr->bth[0] &
+				cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+		break;
+
+	case OP(RDMA_WRITE_FIRST):
+	case OP(RDMA_WRITE_ONLY):
+	case OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE): /* consume RWQE */
+rdma_first:
+		if (unlikely(!(qp->qp_access_flags &
+			       IB_ACCESS_REMOTE_WRITE))) {
+			goto drop;
+		}
+		reth = &ohdr->u.rc.reth;
+		hdrsize += sizeof(*reth);
+		qp->r_len = be32_to_cpu(reth->length);
+		qp->r_rcv_len = 0;
+		qp->r_sge.sg_list = NULL;
+		if (qp->r_len != 0) {
+			u32 rkey = be32_to_cpu(reth->rkey);
+			u64 vaddr = be64_to_cpu(reth->vaddr);
+			int ok;
+
+			/* Check rkey */
+			ok = qib_rkey_ok(qp, &qp->r_sge.sge, qp->r_len,
+					 vaddr, rkey, IB_ACCESS_REMOTE_WRITE);
+			if (unlikely(!ok))
+				goto drop;
+			qp->r_sge.num_sge = 1;
+		} else {
+			qp->r_sge.num_sge = 0;
+			qp->r_sge.sge.mr = NULL;
+			qp->r_sge.sge.vaddr = NULL;
+			qp->r_sge.sge.length = 0;
+			qp->r_sge.sge.sge_length = 0;
+		}
+		if (opcode == OP(RDMA_WRITE_ONLY))
+			goto rdma_last;
+		else if (opcode == OP(RDMA_WRITE_ONLY_WITH_IMMEDIATE)) {
+			wc.ex.imm_data = ohdr->u.rc.imm_data;
+			goto rdma_last_imm;
+		}
+		/* FALLTHROUGH */
+	case OP(RDMA_WRITE_MIDDLE):
+		/* Check for invalid length PMTU or posted rwqe len. */
+		if (unlikely(tlen != (hdrsize + pmtu + 4)))
+			goto drop;
+		qp->r_rcv_len += pmtu;
+		if (unlikely(qp->r_rcv_len > qp->r_len))
+			goto drop;
+		qib_copy_sge(&qp->r_sge, data, pmtu, 1);
+		break;
+
+	case OP(RDMA_WRITE_LAST_WITH_IMMEDIATE):
+		wc.ex.imm_data = ohdr->u.imm_data;
+rdma_last_imm:
+		hdrsize += 4;
+		wc.wc_flags = IB_WC_WITH_IMM;
+
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto drop;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
+			goto drop;
+		if (test_and_clear_bit(QIB_R_REWIND_SGE, &qp->r_aflags))
+			qib_put_ss(&qp->s_rdma_read_sge);
+		else {
+			ret = qib_get_rwqe(qp, 1);
+			if (ret < 0)
+				goto op_err;
+			if (!ret)
+				goto drop;
+		}
+		wc.byte_len = qp->r_len;
+		wc.opcode = IB_WC_RECV_RDMA_WITH_IMM;
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		qib_put_ss(&qp->r_sge);
+		goto last_imm;
+
+	case OP(RDMA_WRITE_LAST):
+rdma_last:
+		/* Get the number of bytes the message was padded by. */
+		pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+		/* Check for invalid length. */
+		/* XXX LAST len should be >= 1 */
+		if (unlikely(tlen < (hdrsize + pad + 4)))
+			goto drop;
+		/* Don't count the CRC. */
+		tlen -= (hdrsize + pad + 4);
+		if (unlikely(tlen + qp->r_rcv_len != qp->r_len))
+			goto drop;
+		qib_copy_sge(&qp->r_sge, data, tlen, 1);
+		qib_put_ss(&qp->r_sge);
+		break;
+
+	default:
+		/* Drop packet for unknown opcodes. */
+		goto drop;
+	}
+	qp->r_psn++;
+	qp->r_state = opcode;
+	return;
+
+rewind:
+	set_bit(QIB_R_REWIND_SGE, &qp->r_aflags);
+	qp->r_sge.num_sge = 0;
+drop:
+	ibp->n_pkt_drops++;
+	return;
+
+op_err:
+	qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+	return;
+
+}
diff --git a/drivers/infiniband/hw/qib/qib_ud.c b/drivers/infiniband/hw/qib/qib_ud.c
new file mode 100644
index 00000000000..aaf7039f8ed
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_ud.c
@@ -0,0 +1,590 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_smi.h>
+
+#include "qib.h"
+#include "qib_mad.h"
+
+/**
+ * qib_ud_loopback - handle send on loopback QPs
+ * @sqp: the sending QP
+ * @swqe: the send work request
+ *
+ * This is called from qib_make_ud_req() to forward a WQE addressed
+ * to the same HCA.
+ * Note that the receive interrupt handler may be calling qib_ud_rcv()
+ * while this is being called.
+ */
+static void qib_ud_loopback(struct qib_qp *sqp, struct qib_swqe *swqe)
+{
+	struct qib_ibport *ibp = to_iport(sqp->ibqp.device, sqp->port_num);
+	struct qib_pportdata *ppd;
+	struct qib_qp *qp;
+	struct ib_ah_attr *ah_attr;
+	unsigned long flags;
+	struct qib_sge_state ssge;
+	struct qib_sge *sge;
+	struct ib_wc wc;
+	u32 length;
+	enum ib_qp_type sqptype, dqptype;
+
+	qp = qib_lookup_qpn(ibp, swqe->wr.wr.ud.remote_qpn);
+	if (!qp) {
+		ibp->n_pkt_drops++;
+		return;
+	}
+
+	sqptype = sqp->ibqp.qp_type == IB_QPT_GSI ?
+			IB_QPT_UD : sqp->ibqp.qp_type;
+	dqptype = qp->ibqp.qp_type == IB_QPT_GSI ?
+			IB_QPT_UD : qp->ibqp.qp_type;
+
+	if (dqptype != sqptype ||
+	    !(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ibp->n_pkt_drops++;
+		goto drop;
+	}
+
+	ah_attr = &to_iah(swqe->wr.wr.ud.ah)->attr;
+	ppd = ppd_from_ibp(ibp);
+
+	if (qp->ibqp.qp_num > 1) {
+		u16 pkey1;
+		u16 pkey2;
+		u16 lid;
+
+		pkey1 = qib_get_pkey(ibp, sqp->s_pkey_index);
+		pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+		if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY, pkey1,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * Check that the qkey matches (except for QP0, see 9.6.1.4.1).
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	if (qp->ibqp.qp_num) {
+		u32 qkey;
+
+		qkey = (int)swqe->wr.wr.ud.remote_qkey < 0 ?
+			sqp->qkey : swqe->wr.wr.ud.remote_qkey;
+		if (unlikely(qkey != qp->qkey)) {
+			u16 lid;
+
+			lid = ppd->lid | (ah_attr->src_path_bits &
+					  ((1 << ppd->lmc) - 1));
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      ah_attr->sl,
+				      sqp->ibqp.qp_num, qp->ibqp.qp_num,
+				      cpu_to_be16(lid),
+				      cpu_to_be16(ah_attr->dlid));
+			goto drop;
+		}
+	}
+
+	/*
+	 * A GRH is expected to precede the data even if not
+	 * present on the wire.
+	 */
+	length = swqe->length;
+	memset(&wc, 0, sizeof wc);
+	wc.byte_len = length + sizeof(struct ib_grh);
+
+	if (swqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		wc.wc_flags = IB_WC_WITH_IMM;
+		wc.ex.imm_data = swqe->wr.ex.imm_data;
+	}
+
+	spin_lock_irqsave(&qp->r_lock, flags);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			goto bail_unlock;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			goto bail_unlock;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		ibp->n_pkt_drops++;
+		goto bail_unlock;
+	}
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		qib_copy_sge(&qp->r_sge, &ah_attr->grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	ssge.sg_list = swqe->sg_list + 1;
+	ssge.sge = *swqe->sg_list;
+	ssge.num_sge = swqe->wr.num_sge;
+	sge = &ssge.sge;
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		qib_copy_sge(&qp->r_sge, sge->vaddr, len, 1);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ssge.num_sge)
+				*sge = *ssge.sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		length -= len;
+	}
+	qib_put_ss(&qp->r_sge);
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		goto bail_unlock;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = sqp->ibqp.qp_num;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		swqe->wr.wr.ud.pkey_index : 0;
+	wc.slid = ppd->lid | (ah_attr->src_path_bits & ((1 << ppd->lmc) - 1));
+	wc.sl = ah_attr->sl;
+	wc.dlid_path_bits = ah_attr->dlid & ((1 << ppd->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     swqe->wr.send_flags & IB_SEND_SOLICITED);
+	ibp->n_loop_pkts++;
+bail_unlock:
+	spin_unlock_irqrestore(&qp->r_lock, flags);
+drop:
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+}
+
+/**
+ * qib_make_ud_req - construct a UD request packet
+ * @qp: the QP
+ *
+ * Return 1 if constructed; otherwise, return 0.
+ */
+int qib_make_ud_req(struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	struct ib_ah_attr *ah_attr;
+	struct qib_pportdata *ppd;
+	struct qib_ibport *ibp;
+	struct qib_swqe *wqe;
+	unsigned long flags;
+	u32 nwords;
+	u32 extra_bytes;
+	u32 bth0;
+	u16 lrh0;
+	u16 lid;
+	int ret = 0;
+	int next_cur;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_NEXT_SEND_OK)) {
+		if (!(ib_qib_state_ops[qp->state] & QIB_FLUSH_SEND))
+			goto bail;
+		/* We are in the error state, flush the work request. */
+		if (qp->s_last == qp->s_head)
+			goto bail;
+		/* If DMAs are in progress, we can't flush immediately. */
+		if (atomic_read(&qp->s_dma_busy)) {
+			qp->s_flags |= QIB_S_WAIT_DMA;
+			goto bail;
+		}
+		wqe = get_swqe_ptr(qp, qp->s_last);
+		qib_send_complete(qp, wqe, IB_WC_WR_FLUSH_ERR);
+		goto done;
+	}
+
+	if (qp->s_cur == qp->s_head)
+		goto bail;
+
+	wqe = get_swqe_ptr(qp, qp->s_cur);
+	next_cur = qp->s_cur + 1;
+	if (next_cur >= qp->s_size)
+		next_cur = 0;
+
+	/* Construct the header. */
+	ibp = to_iport(qp->ibqp.device, qp->port_num);
+	ppd = ppd_from_ibp(ibp);
+	ah_attr = &to_iah(wqe->wr.wr.ud.ah)->attr;
+	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE) {
+		if (ah_attr->dlid != QIB_PERMISSIVE_LID)
+			this_cpu_inc(ibp->pmastats->n_multicast_xmit);
+		else
+			this_cpu_inc(ibp->pmastats->n_unicast_xmit);
+	} else {
+		this_cpu_inc(ibp->pmastats->n_unicast_xmit);
+		lid = ah_attr->dlid & ~((1 << ppd->lmc) - 1);
+		if (unlikely(lid == ppd->lid)) {
+			/*
+			 * If DMAs are in progress, we can't generate
+			 * a completion for the loopback packet since
+			 * it would be out of order.
+			 * XXX Instead of waiting, we could queue a
+			 * zero length descriptor so we get a callback.
+			 */
+			if (atomic_read(&qp->s_dma_busy)) {
+				qp->s_flags |= QIB_S_WAIT_DMA;
+				goto bail;
+			}
+			qp->s_cur = next_cur;
+			spin_unlock_irqrestore(&qp->s_lock, flags);
+			qib_ud_loopback(qp, wqe);
+			spin_lock_irqsave(&qp->s_lock, flags);
+			qib_send_complete(qp, wqe, IB_WC_SUCCESS);
+			goto done;
+		}
+	}
+
+	qp->s_cur = next_cur;
+	extra_bytes = -wqe->length & 3;
+	nwords = (wqe->length + extra_bytes) >> 2;
+
+	/* header size in 32-bit words LRH+BTH+DETH = (8+12+8)/4. */
+	qp->s_hdrwords = 7;
+	qp->s_cur_size = wqe->length;
+	qp->s_cur_sge = &qp->s_sge;
+	qp->s_srate = ah_attr->static_rate;
+	qp->s_wqe = wqe;
+	qp->s_sge.sge = wqe->sg_list[0];
+	qp->s_sge.sg_list = wqe->sg_list + 1;
+	qp->s_sge.num_sge = wqe->wr.num_sge;
+	qp->s_sge.total_len = wqe->length;
+
+	if (ah_attr->ah_flags & IB_AH_GRH) {
+		/* Header size in 32-bit words. */
+		qp->s_hdrwords += qib_make_grh(ibp, &qp->s_hdr->u.l.grh,
+					       &ah_attr->grh,
+					       qp->s_hdrwords, nwords);
+		lrh0 = QIB_LRH_GRH;
+		ohdr = &qp->s_hdr->u.l.oth;
+		/*
+		 * Don't worry about sending to locally attached multicast
+		 * QPs.  It is unspecified by the spec. what happens.
+		 */
+	} else {
+		/* Header size in 32-bit words. */
+		lrh0 = QIB_LRH_BTH;
+		ohdr = &qp->s_hdr->u.oth;
+	}
+	if (wqe->wr.opcode == IB_WR_SEND_WITH_IMM) {
+		qp->s_hdrwords++;
+		ohdr->u.ud.imm_data = wqe->wr.ex.imm_data;
+		bth0 = IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE << 24;
+	} else
+		bth0 = IB_OPCODE_UD_SEND_ONLY << 24;
+	lrh0 |= ah_attr->sl << 4;
+	if (qp->ibqp.qp_type == IB_QPT_SMI)
+		lrh0 |= 0xF000; /* Set VL (see ch. 13.5.3.1) */
+	else
+		lrh0 |= ibp->sl_to_vl[ah_attr->sl] << 12;
+	qp->s_hdr->lrh[0] = cpu_to_be16(lrh0);
+	qp->s_hdr->lrh[1] = cpu_to_be16(ah_attr->dlid);  /* DEST LID */
+	qp->s_hdr->lrh[2] = cpu_to_be16(qp->s_hdrwords + nwords + SIZE_OF_CRC);
+	lid = ppd->lid;
+	if (lid) {
+		lid |= ah_attr->src_path_bits & ((1 << ppd->lmc) - 1);
+		qp->s_hdr->lrh[3] = cpu_to_be16(lid);
+	} else
+		qp->s_hdr->lrh[3] = IB_LID_PERMISSIVE;
+	if (wqe->wr.send_flags & IB_SEND_SOLICITED)
+		bth0 |= IB_BTH_SOLICITED;
+	bth0 |= extra_bytes << 20;
+	bth0 |= qp->ibqp.qp_type == IB_QPT_SMI ? QIB_DEFAULT_P_KEY :
+		qib_get_pkey(ibp, qp->ibqp.qp_type == IB_QPT_GSI ?
+			     wqe->wr.wr.ud.pkey_index : qp->s_pkey_index);
+	ohdr->bth[0] = cpu_to_be32(bth0);
+	/*
+	 * Use the multicast QP if the destination LID is a multicast LID.
+	 */
+	ohdr->bth[1] = ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
+		ah_attr->dlid != QIB_PERMISSIVE_LID ?
+		cpu_to_be32(QIB_MULTICAST_QPN) :
+		cpu_to_be32(wqe->wr.wr.ud.remote_qpn);
+	ohdr->bth[2] = cpu_to_be32(qp->s_next_psn++ & QIB_PSN_MASK);
+	/*
+	 * Qkeys with the high order bit set mean use the
+	 * qkey from the QP context instead of the WR (see 10.2.5).
+	 */
+	ohdr->u.ud.deth[0] = cpu_to_be32((int)wqe->wr.wr.ud.remote_qkey < 0 ?
+					 qp->qkey : wqe->wr.wr.ud.remote_qkey);
+	ohdr->u.ud.deth[1] = cpu_to_be32(qp->ibqp.qp_num);
+
+done:
+	ret = 1;
+	goto unlock;
+
+bail:
+	qp->s_flags &= ~QIB_S_BUSY;
+unlock:
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static unsigned qib_lookup_pkey(struct qib_ibport *ibp, u16 pkey)
+{
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = ppd->dd;
+	unsigned ctxt = ppd->hw_pidx;
+	unsigned i;
+
+	pkey &= 0x7fff;	/* remove limited/full membership bit */
+
+	for (i = 0; i < ARRAY_SIZE(dd->rcd[ctxt]->pkeys); ++i)
+		if ((dd->rcd[ctxt]->pkeys[i] & 0x7fff) == pkey)
+			return i;
+
+	/*
+	 * Should not get here, this means hardware failed to validate pkeys.
+	 * Punt and return index 0.
+	 */
+	return 0;
+}
+
+/**
+ * qib_ud_rcv - receive an incoming UD packet
+ * @ibp: the port the packet came in on
+ * @hdr: the packet header
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP the packet came on
+ *
+ * This is called from qib_qp_rcv() to process an incoming UD packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_other_headers *ohdr;
+	int opcode;
+	u32 hdrsize;
+	u32 pad;
+	struct ib_wc wc;
+	u32 qkey;
+	u32 src_qp;
+	u16 dlid;
+
+	/* Check for GRH */
+	if (!has_grh) {
+		ohdr = &hdr->u.oth;
+		hdrsize = 8 + 12 + 8;   /* LRH + BTH + DETH */
+	} else {
+		ohdr = &hdr->u.l.oth;
+		hdrsize = 8 + 40 + 12 + 8; /* LRH + GRH + BTH + DETH */
+	}
+	qkey = be32_to_cpu(ohdr->u.ud.deth[0]);
+	src_qp = be32_to_cpu(ohdr->u.ud.deth[1]) & QIB_QPN_MASK;
+
+	/*
+	 * Get the number of bytes the message was padded by
+	 * and drop incomplete packets.
+	 */
+	pad = (be32_to_cpu(ohdr->bth[0]) >> 20) & 3;
+	if (unlikely(tlen < (hdrsize + pad + 4)))
+		goto drop;
+
+	tlen -= hdrsize + pad + 4;
+
+	/*
+	 * Check that the permissive LID is only used on QP0
+	 * and the QKEY matches (see 9.6.1.4.1 and 9.6.1.5.1).
+	 */
+	if (qp->ibqp.qp_num) {
+		if (unlikely(hdr->lrh[1] == IB_LID_PERMISSIVE ||
+			     hdr->lrh[3] == IB_LID_PERMISSIVE))
+			goto drop;
+		if (qp->ibqp.qp_num > 1) {
+			u16 pkey1, pkey2;
+
+			pkey1 = be32_to_cpu(ohdr->bth[0]);
+			pkey2 = qib_get_pkey(ibp, qp->s_pkey_index);
+			if (unlikely(!qib_pkey_ok(pkey1, pkey2))) {
+				qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_PKEY,
+					      pkey1,
+					      (be16_to_cpu(hdr->lrh[0]) >> 4) &
+						0xF,
+					      src_qp, qp->ibqp.qp_num,
+					      hdr->lrh[3], hdr->lrh[1]);
+				return;
+			}
+		}
+		if (unlikely(qkey != qp->qkey)) {
+			qib_bad_pqkey(ibp, IB_NOTICE_TRAP_BAD_QKEY, qkey,
+				      (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF,
+				      src_qp, qp->ibqp.qp_num,
+				      hdr->lrh[3], hdr->lrh[1]);
+			return;
+		}
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (unlikely(qp->ibqp.qp_num == 1 &&
+			     (tlen != 256 ||
+			      (be16_to_cpu(hdr->lrh[0]) >> 12) == 15)))
+			goto drop;
+	} else {
+		struct ib_smp *smp;
+
+		/* Drop invalid MAD packets (see 13.5.3.1). */
+		if (tlen != 256 || (be16_to_cpu(hdr->lrh[0]) >> 12) != 15)
+			goto drop;
+		smp = (struct ib_smp *) data;
+		if ((hdr->lrh[1] == IB_LID_PERMISSIVE ||
+		     hdr->lrh[3] == IB_LID_PERMISSIVE) &&
+		    smp->mgmt_class != IB_MGMT_CLASS_SUBN_DIRECTED_ROUTE)
+			goto drop;
+	}
+
+	/*
+	 * The opcode is in the low byte when its in network order
+	 * (top byte when in host order).
+	 */
+	opcode = be32_to_cpu(ohdr->bth[0]) >> 24;
+	if (qp->ibqp.qp_num > 1 &&
+	    opcode == IB_OPCODE_UD_SEND_ONLY_WITH_IMMEDIATE) {
+		wc.ex.imm_data = ohdr->u.ud.imm_data;
+		wc.wc_flags = IB_WC_WITH_IMM;
+		tlen -= sizeof(u32);
+	} else if (opcode == IB_OPCODE_UD_SEND_ONLY) {
+		wc.ex.imm_data = 0;
+		wc.wc_flags = 0;
+	} else
+		goto drop;
+
+	/*
+	 * A GRH is expected to precede the data even if not
+	 * present on the wire.
+	 */
+	wc.byte_len = tlen + sizeof(struct ib_grh);
+
+	/*
+	 * Get the next work request entry to find where to put the data.
+	 */
+	if (qp->r_flags & QIB_R_REUSE_SGE)
+		qp->r_flags &= ~QIB_R_REUSE_SGE;
+	else {
+		int ret;
+
+		ret = qib_get_rwqe(qp, 0);
+		if (ret < 0) {
+			qib_rc_error(qp, IB_WC_LOC_QP_OP_ERR);
+			return;
+		}
+		if (!ret) {
+			if (qp->ibqp.qp_num == 0)
+				ibp->n_vl15_dropped++;
+			return;
+		}
+	}
+	/* Silently drop packets which are too big. */
+	if (unlikely(wc.byte_len > qp->r_len)) {
+		qp->r_flags |= QIB_R_REUSE_SGE;
+		goto drop;
+	}
+	if (has_grh) {
+		qib_copy_sge(&qp->r_sge, &hdr->u.l.grh,
+			     sizeof(struct ib_grh), 1);
+		wc.wc_flags |= IB_WC_GRH;
+	} else
+		qib_skip_sge(&qp->r_sge, sizeof(struct ib_grh), 1);
+	qib_copy_sge(&qp->r_sge, data, wc.byte_len - sizeof(struct ib_grh), 1);
+	qib_put_ss(&qp->r_sge);
+	if (!test_and_clear_bit(QIB_R_WRID_VALID, &qp->r_aflags))
+		return;
+	wc.wr_id = qp->r_wr_id;
+	wc.status = IB_WC_SUCCESS;
+	wc.opcode = IB_WC_RECV;
+	wc.vendor_err = 0;
+	wc.qp = &qp->ibqp;
+	wc.src_qp = src_qp;
+	wc.pkey_index = qp->ibqp.qp_type == IB_QPT_GSI ?
+		qib_lookup_pkey(ibp, be32_to_cpu(ohdr->bth[0])) : 0;
+	wc.slid = be16_to_cpu(hdr->lrh[3]);
+	wc.sl = (be16_to_cpu(hdr->lrh[0]) >> 4) & 0xF;
+	dlid = be16_to_cpu(hdr->lrh[1]);
+	/*
+	 * Save the LMC lower bits if the destination LID is a unicast LID.
+	 */
+	wc.dlid_path_bits = dlid >= QIB_MULTICAST_LID_BASE ? 0 :
+		dlid & ((1 << ppd_from_ibp(ibp)->lmc) - 1);
+	wc.port_num = qp->port_num;
+	/* Signal completion event if the solicited bit is set. */
+	qib_cq_enter(to_icq(qp->ibqp.recv_cq), &wc,
+		     (ohdr->bth[0] &
+			cpu_to_be32(IB_BTH_SOLICITED)) != 0);
+	return;
+
+drop:
+	ibp->n_pkt_drops++;
+}
diff --git a/drivers/infiniband/hw/qib/qib_user_pages.c b/drivers/infiniband/hw/qib/qib_user_pages.c
new file mode 100644
index 00000000000..2bc1d2b9629
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_pages.c
@@ -0,0 +1,157 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mm.h>
+#include <linux/device.h>
+
+#include "qib.h"
+
+static void __qib_release_user_pages(struct page **p, size_t num_pages,
+				     int dirty)
+{
+	size_t i;
+
+	for (i = 0; i < num_pages; i++) {
+		if (dirty)
+			set_page_dirty_lock(p[i]);
+		put_page(p[i]);
+	}
+}
+
+/*
+ * Call with current->mm->mmap_sem held.
+ */
+static int __qib_get_user_pages(unsigned long start_page, size_t num_pages,
+				struct page **p, struct vm_area_struct **vma)
+{
+	unsigned long lock_limit;
+	size_t got;
+	int ret;
+
+	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+
+	if (num_pages > lock_limit && !capable(CAP_IPC_LOCK)) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	for (got = 0; got < num_pages; got += ret) {
+		ret = get_user_pages(current, current->mm,
+				     start_page + got * PAGE_SIZE,
+				     num_pages - got, 1, 1,
+				     p + got, vma);
+		if (ret < 0)
+			goto bail_release;
+	}
+
+	current->mm->pinned_vm += num_pages;
+
+	ret = 0;
+	goto bail;
+
+bail_release:
+	__qib_release_user_pages(p, got, 0);
+bail:
+	return ret;
+}
+
+/**
+ * qib_map_page - a safety wrapper around pci_map_page()
+ *
+ * A dma_addr of all 0's is interpreted by the chip as "disabled".
+ * Unfortunately, it can also be a valid dma_addr returned on some
+ * architectures.
+ *
+ * The powerpc iommu assigns dma_addrs in ascending order, so we don't
+ * have to bother with retries or mapping a dummy page to insure we
+ * don't just get the same mapping again.
+ *
+ * I'm sure we won't be so lucky with other iommu's, so FIXME.
+ */
+dma_addr_t qib_map_page(struct pci_dev *hwdev, struct page *page,
+			unsigned long offset, size_t size, int direction)
+{
+	dma_addr_t phys;
+
+	phys = pci_map_page(hwdev, page, offset, size, direction);
+
+	if (phys == 0) {
+		pci_unmap_page(hwdev, phys, size, direction);
+		phys = pci_map_page(hwdev, page, offset, size, direction);
+		/*
+		 * FIXME: If we get 0 again, we should keep this page,
+		 * map another, then free the 0 page.
+		 */
+	}
+
+	return phys;
+}
+
+/**
+ * qib_get_user_pages - lock user pages into memory
+ * @start_page: the start page
+ * @num_pages: the number of pages
+ * @p: the output page structures
+ *
+ * This function takes a given start page (page aligned user virtual
+ * address) and pins it and the following specified number of pages.  For
+ * now, num_pages is always 1, but that will probably change at some point
+ * (because caller is doing expected sends on a single virtually contiguous
+ * buffer, so we can do all pages at once).
+ */
+int qib_get_user_pages(unsigned long start_page, size_t num_pages,
+		       struct page **p)
+{
+	int ret;
+
+	down_write(&current->mm->mmap_sem);
+
+	ret = __qib_get_user_pages(start_page, num_pages, p, NULL);
+
+	up_write(&current->mm->mmap_sem);
+
+	return ret;
+}
+
+void qib_release_user_pages(struct page **p, size_t num_pages)
+{
+	if (current->mm) /* during close after signal, mm can be NULL */
+		down_write(&current->mm->mmap_sem);
+
+	__qib_release_user_pages(p, num_pages, 1);
+
+	if (current->mm) {
+		current->mm->pinned_vm -= num_pages;
+		up_write(&current->mm->mmap_sem);
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.c b/drivers/infiniband/hw/qib/qib_user_sdma.c
new file mode 100644
index 00000000000..d2806cae234
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.c
@@ -0,0 +1,1465 @@
+/*
+ * Copyright (c) 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/mm.h>
+#include <linux/types.h>
+#include <linux/device.h>
+#include <linux/dmapool.h>
+#include <linux/slab.h>
+#include <linux/list.h>
+#include <linux/highmem.h>
+#include <linux/io.h>
+#include <linux/uio.h>
+#include <linux/rbtree.h>
+#include <linux/spinlock.h>
+#include <linux/delay.h>
+
+#include "qib.h"
+#include "qib_user_sdma.h"
+
+/* minimum size of header */
+#define QIB_USER_SDMA_MIN_HEADER_LENGTH 64
+/* expected size of headers (for dma_pool) */
+#define QIB_USER_SDMA_EXP_HEADER_LENGTH 64
+/* attempt to drain the queue for 5secs */
+#define QIB_USER_SDMA_DRAIN_TIMEOUT 500
+
+/*
+ * track how many times a process open this driver.
+ */
+static struct rb_root qib_user_sdma_rb_root = RB_ROOT;
+
+struct qib_user_sdma_rb_node {
+	struct rb_node node;
+	int refcount;
+	pid_t pid;
+};
+
+struct qib_user_sdma_pkt {
+	struct list_head list;  /* list element */
+
+	u8  tiddma;		/* if this is NEW tid-sdma */
+	u8  largepkt;		/* this is large pkt from kmalloc */
+	u16 frag_size;		/* frag size used by PSM */
+	u16 index;              /* last header index or push index */
+	u16 naddr;              /* dimension of addr (1..3) ... */
+	u16 addrlimit;		/* addr array size */
+	u16 tidsmidx;		/* current tidsm index */
+	u16 tidsmcount;		/* tidsm array item count */
+	u16 payload_size;	/* payload size so far for header */
+	u32 bytes_togo;		/* bytes for processing */
+	u32 counter;            /* sdma pkts queued counter for this entry */
+	struct qib_tid_session_member *tidsm;	/* tid session member array */
+	struct qib_user_sdma_queue *pq;	/* which pq this pkt belongs to */
+	u64 added;              /* global descq number of entries */
+
+	struct {
+		u16 offset;                     /* offset for kvaddr, addr */
+		u16 length;                     /* length in page */
+		u16 first_desc;			/* first desc */
+		u16 last_desc;			/* last desc */
+		u16 put_page;                   /* should we put_page? */
+		u16 dma_mapped;                 /* is page dma_mapped? */
+		u16 dma_length;			/* for dma_unmap_page() */
+		u16 padding;
+		struct page *page;              /* may be NULL (coherent mem) */
+		void *kvaddr;                   /* FIXME: only for pio hack */
+		dma_addr_t addr;
+	} addr[4];   /* max pages, any more and we coalesce */
+};
+
+struct qib_user_sdma_queue {
+	/*
+	 * pkts sent to dma engine are queued on this
+	 * list head.  the type of the elements of this
+	 * list are struct qib_user_sdma_pkt...
+	 */
+	struct list_head sent;
+
+	/*
+	 * Because above list will be accessed by both process and
+	 * signal handler, we need a spinlock for it.
+	 */
+	spinlock_t sent_lock ____cacheline_aligned_in_smp;
+
+	/* headers with expected length are allocated from here... */
+	char header_cache_name[64];
+	struct dma_pool *header_cache;
+
+	/* packets are allocated from the slab cache... */
+	char pkt_slab_name[64];
+	struct kmem_cache *pkt_slab;
+
+	/* as packets go on the queued queue, they are counted... */
+	u32 counter;
+	u32 sent_counter;
+	/* pending packets, not sending yet */
+	u32 num_pending;
+	/* sending packets, not complete yet */
+	u32 num_sending;
+	/* global descq number of entry of last sending packet */
+	u64 added;
+
+	/* dma page table */
+	struct rb_root dma_pages_root;
+
+	struct qib_user_sdma_rb_node *sdma_rb_node;
+
+	/* protect everything above... */
+	struct mutex lock;
+};
+
+static struct qib_user_sdma_rb_node *
+qib_user_sdma_rb_search(struct rb_root *root, pid_t pid)
+{
+	struct qib_user_sdma_rb_node *sdma_rb_node;
+	struct rb_node *node = root->rb_node;
+
+	while (node) {
+		sdma_rb_node = container_of(node,
+			struct qib_user_sdma_rb_node, node);
+		if (pid < sdma_rb_node->pid)
+			node = node->rb_left;
+		else if (pid > sdma_rb_node->pid)
+			node = node->rb_right;
+		else
+			return sdma_rb_node;
+	}
+	return NULL;
+}
+
+static int
+qib_user_sdma_rb_insert(struct rb_root *root, struct qib_user_sdma_rb_node *new)
+{
+	struct rb_node **node = &(root->rb_node);
+	struct rb_node *parent = NULL;
+	struct qib_user_sdma_rb_node *got;
+
+	while (*node) {
+		got = container_of(*node, struct qib_user_sdma_rb_node, node);
+		parent = *node;
+		if (new->pid < got->pid)
+			node = &((*node)->rb_left);
+		else if (new->pid > got->pid)
+			node = &((*node)->rb_right);
+		else
+			return 0;
+	}
+
+	rb_link_node(&new->node, parent, node);
+	rb_insert_color(&new->node, root);
+	return 1;
+}
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int ctxt, int sctxt)
+{
+	struct qib_user_sdma_queue *pq =
+		kmalloc(sizeof(struct qib_user_sdma_queue), GFP_KERNEL);
+	struct qib_user_sdma_rb_node *sdma_rb_node;
+
+	if (!pq)
+		goto done;
+
+	pq->counter = 0;
+	pq->sent_counter = 0;
+	pq->num_pending = 0;
+	pq->num_sending = 0;
+	pq->added = 0;
+	pq->sdma_rb_node = NULL;
+
+	INIT_LIST_HEAD(&pq->sent);
+	spin_lock_init(&pq->sent_lock);
+	mutex_init(&pq->lock);
+
+	snprintf(pq->pkt_slab_name, sizeof(pq->pkt_slab_name),
+		 "qib-user-sdma-pkts-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->pkt_slab = kmem_cache_create(pq->pkt_slab_name,
+					 sizeof(struct qib_user_sdma_pkt),
+					 0, 0, NULL);
+
+	if (!pq->pkt_slab)
+		goto err_kfree;
+
+	snprintf(pq->header_cache_name, sizeof(pq->header_cache_name),
+		 "qib-user-sdma-headers-%u-%02u.%02u", unit, ctxt, sctxt);
+	pq->header_cache = dma_pool_create(pq->header_cache_name,
+					   dev,
+					   QIB_USER_SDMA_EXP_HEADER_LENGTH,
+					   4, 0);
+	if (!pq->header_cache)
+		goto err_slab;
+
+	pq->dma_pages_root = RB_ROOT;
+
+	sdma_rb_node = qib_user_sdma_rb_search(&qib_user_sdma_rb_root,
+					current->pid);
+	if (sdma_rb_node) {
+		sdma_rb_node->refcount++;
+	} else {
+		int ret;
+		sdma_rb_node = kmalloc(sizeof(
+			struct qib_user_sdma_rb_node), GFP_KERNEL);
+		if (!sdma_rb_node)
+			goto err_rb;
+
+		sdma_rb_node->refcount = 1;
+		sdma_rb_node->pid = current->pid;
+
+		ret = qib_user_sdma_rb_insert(&qib_user_sdma_rb_root,
+					sdma_rb_node);
+		BUG_ON(ret == 0);
+	}
+	pq->sdma_rb_node = sdma_rb_node;
+
+	goto done;
+
+err_rb:
+	dma_pool_destroy(pq->header_cache);
+err_slab:
+	kmem_cache_destroy(pq->pkt_slab);
+err_kfree:
+	kfree(pq);
+	pq = NULL;
+
+done:
+	return pq;
+}
+
+static void qib_user_sdma_init_frag(struct qib_user_sdma_pkt *pkt,
+				    int i, u16 offset, u16 len,
+				    u16 first_desc, u16 last_desc,
+				    u16 put_page, u16 dma_mapped,
+				    struct page *page, void *kvaddr,
+				    dma_addr_t dma_addr, u16 dma_length)
+{
+	pkt->addr[i].offset = offset;
+	pkt->addr[i].length = len;
+	pkt->addr[i].first_desc = first_desc;
+	pkt->addr[i].last_desc = last_desc;
+	pkt->addr[i].put_page = put_page;
+	pkt->addr[i].dma_mapped = dma_mapped;
+	pkt->addr[i].page = page;
+	pkt->addr[i].kvaddr = kvaddr;
+	pkt->addr[i].addr = dma_addr;
+	pkt->addr[i].dma_length = dma_length;
+}
+
+static void *qib_user_sdma_alloc_header(struct qib_user_sdma_queue *pq,
+				size_t len, dma_addr_t *dma_addr)
+{
+	void *hdr;
+
+	if (len == QIB_USER_SDMA_EXP_HEADER_LENGTH)
+		hdr = dma_pool_alloc(pq->header_cache, GFP_KERNEL,
+					     dma_addr);
+	else
+		hdr = NULL;
+
+	if (!hdr) {
+		hdr = kmalloc(len, GFP_KERNEL);
+		if (!hdr)
+			return NULL;
+
+		*dma_addr = 0;
+	}
+
+	return hdr;
+}
+
+static int qib_user_sdma_page_to_frags(const struct qib_devdata *dd,
+				       struct qib_user_sdma_queue *pq,
+				       struct qib_user_sdma_pkt *pkt,
+				       struct page *page, u16 put,
+				       u16 offset, u16 len, void *kvaddr)
+{
+	__le16 *pbc16;
+	void *pbcvaddr;
+	struct qib_message_header *hdr;
+	u16 newlen, pbclen, lastdesc, dma_mapped;
+	u32 vcto;
+	union qib_seqnum seqnum;
+	dma_addr_t pbcdaddr;
+	dma_addr_t dma_addr =
+		dma_map_page(&dd->pcidev->dev,
+			page, offset, len, DMA_TO_DEVICE);
+	int ret = 0;
+
+	if (dma_mapping_error(&dd->pcidev->dev, dma_addr)) {
+		/*
+		 * dma mapping error, pkt has not managed
+		 * this page yet, return the page here so
+		 * the caller can ignore this page.
+		 */
+		if (put) {
+			put_page(page);
+		} else {
+			/* coalesce case */
+			kunmap(page);
+			__free_page(page);
+		}
+		ret = -ENOMEM;
+		goto done;
+	}
+	offset = 0;
+	dma_mapped = 1;
+
+
+next_fragment:
+
+	/*
+	 * In tid-sdma, the transfer length is restricted by
+	 * receiver side current tid page length.
+	 */
+	if (pkt->tiddma && len > pkt->tidsm[pkt->tidsmidx].length)
+		newlen = pkt->tidsm[pkt->tidsmidx].length;
+	else
+		newlen = len;
+
+	/*
+	 * Then the transfer length is restricted by MTU.
+	 * the last descriptor flag is determined by:
+	 * 1. the current packet is at frag size length.
+	 * 2. the current tid page is done if tid-sdma.
+	 * 3. there is no more byte togo if sdma.
+	 */
+	lastdesc = 0;
+	if ((pkt->payload_size + newlen) >= pkt->frag_size) {
+		newlen = pkt->frag_size - pkt->payload_size;
+		lastdesc = 1;
+	} else if (pkt->tiddma) {
+		if (newlen == pkt->tidsm[pkt->tidsmidx].length)
+			lastdesc = 1;
+	} else {
+		if (newlen == pkt->bytes_togo)
+			lastdesc = 1;
+	}
+
+	/* fill the next fragment in this page */
+	qib_user_sdma_init_frag(pkt, pkt->naddr, /* index */
+		offset, newlen,		/* offset, len */
+		0, lastdesc,		/* first last desc */
+		put, dma_mapped,	/* put page, dma mapped */
+		page, kvaddr,		/* struct page, virt addr */
+		dma_addr, len);		/* dma addr, dma length */
+	pkt->bytes_togo -= newlen;
+	pkt->payload_size += newlen;
+	pkt->naddr++;
+	if (pkt->naddr == pkt->addrlimit) {
+		ret = -EFAULT;
+		goto done;
+	}
+
+	/* If there is no more byte togo. (lastdesc==1) */
+	if (pkt->bytes_togo == 0) {
+		/* The packet is done, header is not dma mapped yet.
+		 * it should be from kmalloc */
+		if (!pkt->addr[pkt->index].addr) {
+			pkt->addr[pkt->index].addr =
+				dma_map_single(&dd->pcidev->dev,
+					pkt->addr[pkt->index].kvaddr,
+					pkt->addr[pkt->index].dma_length,
+					DMA_TO_DEVICE);
+			if (dma_mapping_error(&dd->pcidev->dev,
+					pkt->addr[pkt->index].addr)) {
+				ret = -ENOMEM;
+				goto done;
+			}
+			pkt->addr[pkt->index].dma_mapped = 1;
+		}
+
+		goto done;
+	}
+
+	/* If tid-sdma, advance tid info. */
+	if (pkt->tiddma) {
+		pkt->tidsm[pkt->tidsmidx].length -= newlen;
+		if (pkt->tidsm[pkt->tidsmidx].length) {
+			pkt->tidsm[pkt->tidsmidx].offset += newlen;
+		} else {
+			pkt->tidsmidx++;
+			if (pkt->tidsmidx == pkt->tidsmcount) {
+				ret = -EFAULT;
+				goto done;
+			}
+		}
+	}
+
+	/*
+	 * If this is NOT the last descriptor. (newlen==len)
+	 * the current packet is not done yet, but the current
+	 * send side page is done.
+	 */
+	if (lastdesc == 0)
+		goto done;
+
+	/*
+	 * If running this driver under PSM with message size
+	 * fitting into one transfer unit, it is not possible
+	 * to pass this line. otherwise, it is a buggggg.
+	 */
+
+	/*
+	 * Since the current packet is done, and there are more
+	 * bytes togo, we need to create a new sdma header, copying
+	 * from previous sdma header and modify both.
+	 */
+	pbclen = pkt->addr[pkt->index].length;
+	pbcvaddr = qib_user_sdma_alloc_header(pq, pbclen, &pbcdaddr);
+	if (!pbcvaddr) {
+		ret = -ENOMEM;
+		goto done;
+	}
+	/* Copy the previous sdma header to new sdma header */
+	pbc16 = (__le16 *)pkt->addr[pkt->index].kvaddr;
+	memcpy(pbcvaddr, pbc16, pbclen);
+
+	/* Modify the previous sdma header */
+	hdr = (struct qib_message_header *)&pbc16[4];
+
+	/* New pbc length */
+	pbc16[0] = cpu_to_le16(le16_to_cpu(pbc16[0])-(pkt->bytes_togo>>2));
+
+	/* New packet length */
+	hdr->lrh[2] = cpu_to_be16(le16_to_cpu(pbc16[0]));
+
+	if (pkt->tiddma) {
+		/* turn on the header suppression */
+		hdr->iph.pkt_flags =
+			cpu_to_le16(le16_to_cpu(hdr->iph.pkt_flags)|0x2);
+		/* turn off ACK_REQ: 0x04 and EXPECTED_DONE: 0x20 */
+		hdr->flags &= ~(0x04|0x20);
+	} else {
+		/* turn off extra bytes: 20-21 bits */
+		hdr->bth[0] = cpu_to_be32(be32_to_cpu(hdr->bth[0])&0xFFCFFFFF);
+		/* turn off ACK_REQ: 0x04 */
+		hdr->flags &= ~(0x04);
+	}
+
+	/* New kdeth checksum */
+	vcto = le32_to_cpu(hdr->iph.ver_ctxt_tid_offset);
+	hdr->iph.chksum = cpu_to_le16(QIB_LRH_BTH +
+		be16_to_cpu(hdr->lrh[2]) -
+		((vcto>>16)&0xFFFF) - (vcto&0xFFFF) -
+		le16_to_cpu(hdr->iph.pkt_flags));
+
+	/* The packet is done, header is not dma mapped yet.
+	 * it should be from kmalloc */
+	if (!pkt->addr[pkt->index].addr) {
+		pkt->addr[pkt->index].addr =
+			dma_map_single(&dd->pcidev->dev,
+				pkt->addr[pkt->index].kvaddr,
+				pkt->addr[pkt->index].dma_length,
+				DMA_TO_DEVICE);
+		if (dma_mapping_error(&dd->pcidev->dev,
+				pkt->addr[pkt->index].addr)) {
+			ret = -ENOMEM;
+			goto done;
+		}
+		pkt->addr[pkt->index].dma_mapped = 1;
+	}
+
+	/* Modify the new sdma header */
+	pbc16 = (__le16 *)pbcvaddr;
+	hdr = (struct qib_message_header *)&pbc16[4];
+
+	/* New pbc length */
+	pbc16[0] = cpu_to_le16(le16_to_cpu(pbc16[0])-(pkt->payload_size>>2));
+
+	/* New packet length */
+	hdr->lrh[2] = cpu_to_be16(le16_to_cpu(pbc16[0]));
+
+	if (pkt->tiddma) {
+		/* Set new tid and offset for new sdma header */
+		hdr->iph.ver_ctxt_tid_offset = cpu_to_le32(
+			(le32_to_cpu(hdr->iph.ver_ctxt_tid_offset)&0xFF000000) +
+			(pkt->tidsm[pkt->tidsmidx].tid<<QLOGIC_IB_I_TID_SHIFT) +
+			(pkt->tidsm[pkt->tidsmidx].offset>>2));
+	} else {
+		/* Middle protocol new packet offset */
+		hdr->uwords[2] += pkt->payload_size;
+	}
+
+	/* New kdeth checksum */
+	vcto = le32_to_cpu(hdr->iph.ver_ctxt_tid_offset);
+	hdr->iph.chksum = cpu_to_le16(QIB_LRH_BTH +
+		be16_to_cpu(hdr->lrh[2]) -
+		((vcto>>16)&0xFFFF) - (vcto&0xFFFF) -
+		le16_to_cpu(hdr->iph.pkt_flags));
+
+	/* Next sequence number in new sdma header */
+	seqnum.val = be32_to_cpu(hdr->bth[2]);
+	if (pkt->tiddma)
+		seqnum.seq++;
+	else
+		seqnum.pkt++;
+	hdr->bth[2] = cpu_to_be32(seqnum.val);
+
+	/* Init new sdma header. */
+	qib_user_sdma_init_frag(pkt, pkt->naddr, /* index */
+		0, pbclen,		/* offset, len */
+		1, 0,			/* first last desc */
+		0, 0,			/* put page, dma mapped */
+		NULL, pbcvaddr,		/* struct page, virt addr */
+		pbcdaddr, pbclen);	/* dma addr, dma length */
+	pkt->index = pkt->naddr;
+	pkt->payload_size = 0;
+	pkt->naddr++;
+	if (pkt->naddr == pkt->addrlimit) {
+		ret = -EFAULT;
+		goto done;
+	}
+
+	/* Prepare for next fragment in this page */
+	if (newlen != len) {
+		if (dma_mapped) {
+			put = 0;
+			dma_mapped = 0;
+			page = NULL;
+			kvaddr = NULL;
+		}
+		len -= newlen;
+		offset += newlen;
+
+		goto next_fragment;
+	}
+
+done:
+	return ret;
+}
+
+/* we've too many pages in the iovec, coalesce to a single page */
+static int qib_user_sdma_coalesce(const struct qib_devdata *dd,
+				  struct qib_user_sdma_queue *pq,
+				  struct qib_user_sdma_pkt *pkt,
+				  const struct iovec *iov,
+				  unsigned long niov)
+{
+	int ret = 0;
+	struct page *page = alloc_page(GFP_KERNEL);
+	void *mpage_save;
+	char *mpage;
+	int i;
+	int len = 0;
+
+	if (!page) {
+		ret = -ENOMEM;
+		goto done;
+	}
+
+	mpage = kmap(page);
+	mpage_save = mpage;
+	for (i = 0; i < niov; i++) {
+		int cfur;
+
+		cfur = copy_from_user(mpage,
+				      iov[i].iov_base, iov[i].iov_len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_unmap;
+		}
+
+		mpage += iov[i].iov_len;
+		len += iov[i].iov_len;
+	}
+
+	ret = qib_user_sdma_page_to_frags(dd, pq, pkt,
+			page, 0, 0, len, mpage_save);
+	goto done;
+
+free_unmap:
+	kunmap(page);
+	__free_page(page);
+done:
+	return ret;
+}
+
+/*
+ * How many pages in this iovec element?
+ */
+static int qib_user_sdma_num_pages(const struct iovec *iov)
+{
+	const unsigned long addr  = (unsigned long) iov->iov_base;
+	const unsigned long  len  = iov->iov_len;
+	const unsigned long spage = addr & PAGE_MASK;
+	const unsigned long epage = (addr + len - 1) & PAGE_MASK;
+
+	return 1 + ((epage - spage) >> PAGE_SHIFT);
+}
+
+static void qib_user_sdma_free_pkt_frag(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct qib_user_sdma_pkt *pkt,
+					int frag)
+{
+	const int i = frag;
+
+	if (pkt->addr[i].page) {
+		/* only user data has page */
+		if (pkt->addr[i].dma_mapped)
+			dma_unmap_page(dev,
+				       pkt->addr[i].addr,
+				       pkt->addr[i].dma_length,
+				       DMA_TO_DEVICE);
+
+		if (pkt->addr[i].kvaddr)
+			kunmap(pkt->addr[i].page);
+
+		if (pkt->addr[i].put_page)
+			put_page(pkt->addr[i].page);
+		else
+			__free_page(pkt->addr[i].page);
+	} else if (pkt->addr[i].kvaddr) {
+		/* for headers */
+		if (pkt->addr[i].dma_mapped) {
+			/* from kmalloc & dma mapped */
+			dma_unmap_single(dev,
+				       pkt->addr[i].addr,
+				       pkt->addr[i].dma_length,
+				       DMA_TO_DEVICE);
+			kfree(pkt->addr[i].kvaddr);
+		} else if (pkt->addr[i].addr) {
+			/* free coherent mem from cache... */
+			dma_pool_free(pq->header_cache,
+			      pkt->addr[i].kvaddr, pkt->addr[i].addr);
+		} else {
+			/* from kmalloc but not dma mapped */
+			kfree(pkt->addr[i].kvaddr);
+		}
+	}
+}
+
+/* return number of pages pinned... */
+static int qib_user_sdma_pin_pages(const struct qib_devdata *dd,
+				   struct qib_user_sdma_queue *pq,
+				   struct qib_user_sdma_pkt *pkt,
+				   unsigned long addr, int tlen, int npages)
+{
+	struct page *pages[8];
+	int i, j;
+	int ret = 0;
+
+	while (npages) {
+		if (npages > 8)
+			j = 8;
+		else
+			j = npages;
+
+		ret = get_user_pages_fast(addr, j, 0, pages);
+		if (ret != j) {
+			i = 0;
+			j = ret;
+			ret = -ENOMEM;
+			goto free_pages;
+		}
+
+		for (i = 0; i < j; i++) {
+			/* map the pages... */
+			unsigned long fofs = addr & ~PAGE_MASK;
+			int flen = ((fofs + tlen) > PAGE_SIZE) ?
+				(PAGE_SIZE - fofs) : tlen;
+
+			ret = qib_user_sdma_page_to_frags(dd, pq, pkt,
+				pages[i], 1, fofs, flen, NULL);
+			if (ret < 0) {
+				/* current page has beed taken
+				 * care of inside above call.
+				 */
+				i++;
+				goto free_pages;
+			}
+
+			addr += flen;
+			tlen -= flen;
+		}
+
+		npages -= j;
+	}
+
+	goto done;
+
+	/* if error, return all pages not managed by pkt */
+free_pages:
+	while (i < j)
+		put_page(pages[i++]);
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_pin_pkt(const struct qib_devdata *dd,
+				 struct qib_user_sdma_queue *pq,
+				 struct qib_user_sdma_pkt *pkt,
+				 const struct iovec *iov,
+				 unsigned long niov)
+{
+	int ret = 0;
+	unsigned long idx;
+
+	for (idx = 0; idx < niov; idx++) {
+		const int npages = qib_user_sdma_num_pages(iov + idx);
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+
+		ret = qib_user_sdma_pin_pages(dd, pq, pkt, addr,
+					      iov[idx].iov_len, npages);
+		if (ret < 0)
+			goto free_pkt;
+	}
+
+	goto done;
+
+free_pkt:
+	/* we need to ignore the first entry here */
+	for (idx = 1; idx < pkt->naddr; idx++)
+		qib_user_sdma_free_pkt_frag(&dd->pcidev->dev, pq, pkt, idx);
+
+	/* need to dma unmap the first entry, this is to restore to
+	 * the original state so that caller can free the memory in
+	 * error condition. Caller does not know if dma mapped or not*/
+	if (pkt->addr[0].dma_mapped) {
+		dma_unmap_single(&dd->pcidev->dev,
+		       pkt->addr[0].addr,
+		       pkt->addr[0].dma_length,
+		       DMA_TO_DEVICE);
+		pkt->addr[0].addr = 0;
+		pkt->addr[0].dma_mapped = 0;
+	}
+
+done:
+	return ret;
+}
+
+static int qib_user_sdma_init_payload(const struct qib_devdata *dd,
+				      struct qib_user_sdma_queue *pq,
+				      struct qib_user_sdma_pkt *pkt,
+				      const struct iovec *iov,
+				      unsigned long niov, int npages)
+{
+	int ret = 0;
+
+	if (pkt->frag_size == pkt->bytes_togo &&
+			npages >= ARRAY_SIZE(pkt->addr))
+		ret = qib_user_sdma_coalesce(dd, pq, pkt, iov, niov);
+	else
+		ret = qib_user_sdma_pin_pkt(dd, pq, pkt, iov, niov);
+
+	return ret;
+}
+
+/* free a packet list -- return counter value of last packet */
+static void qib_user_sdma_free_pkt_list(struct device *dev,
+					struct qib_user_sdma_queue *pq,
+					struct list_head *list)
+{
+	struct qib_user_sdma_pkt *pkt, *pkt_next;
+
+	list_for_each_entry_safe(pkt, pkt_next, list, list) {
+		int i;
+
+		for (i = 0; i < pkt->naddr; i++)
+			qib_user_sdma_free_pkt_frag(dev, pq, pkt, i);
+
+		if (pkt->largepkt)
+			kfree(pkt);
+		else
+			kmem_cache_free(pq->pkt_slab, pkt);
+	}
+	INIT_LIST_HEAD(list);
+}
+
+/*
+ * copy headers, coalesce etc -- pq->lock must be held
+ *
+ * we queue all the packets to list, returning the
+ * number of bytes total.  list must be empty initially,
+ * as, if there is an error we clean it...
+ */
+static int qib_user_sdma_queue_pkts(const struct qib_devdata *dd,
+				    struct qib_pportdata *ppd,
+				    struct qib_user_sdma_queue *pq,
+				    const struct iovec *iov,
+				    unsigned long niov,
+				    struct list_head *list,
+				    int *maxpkts, int *ndesc)
+{
+	unsigned long idx = 0;
+	int ret = 0;
+	int npkts = 0;
+	__le32 *pbc;
+	dma_addr_t dma_addr;
+	struct qib_user_sdma_pkt *pkt = NULL;
+	size_t len;
+	size_t nw;
+	u32 counter = pq->counter;
+	u16 frag_size;
+
+	while (idx < niov && npkts < *maxpkts) {
+		const unsigned long addr = (unsigned long) iov[idx].iov_base;
+		const unsigned long idx_save = idx;
+		unsigned pktnw;
+		unsigned pktnwc;
+		int nfrags = 0;
+		int npages = 0;
+		int bytes_togo = 0;
+		int tiddma = 0;
+		int cfur;
+
+		len = iov[idx].iov_len;
+		nw = len >> 2;
+
+		if (len < QIB_USER_SDMA_MIN_HEADER_LENGTH ||
+		    len > PAGE_SIZE || len & 3 || addr & 3) {
+			ret = -EINVAL;
+			goto free_list;
+		}
+
+		pbc = qib_user_sdma_alloc_header(pq, len, &dma_addr);
+		if (!pbc) {
+			ret = -ENOMEM;
+			goto free_list;
+		}
+
+		cfur = copy_from_user(pbc, iov[idx].iov_base, len);
+		if (cfur) {
+			ret = -EFAULT;
+			goto free_pbc;
+		}
+
+		/*
+		 * This assignment is a bit strange.  it's because the
+		 * the pbc counts the number of 32 bit words in the full
+		 * packet _except_ the first word of the pbc itself...
+		 */
+		pktnwc = nw - 1;
+
+		/*
+		 * pktnw computation yields the number of 32 bit words
+		 * that the caller has indicated in the PBC.  note that
+		 * this is one less than the total number of words that
+		 * goes to the send DMA engine as the first 32 bit word
+		 * of the PBC itself is not counted.  Armed with this count,
+		 * we can verify that the packet is consistent with the
+		 * iovec lengths.
+		 */
+		pktnw = le32_to_cpu(*pbc) & 0xFFFF;
+		if (pktnw < pktnwc) {
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		idx++;
+		while (pktnwc < pktnw && idx < niov) {
+			const size_t slen = iov[idx].iov_len;
+			const unsigned long faddr =
+				(unsigned long) iov[idx].iov_base;
+
+			if (slen & 3 || faddr & 3 || !slen) {
+				ret = -EINVAL;
+				goto free_pbc;
+			}
+
+			npages += qib_user_sdma_num_pages(&iov[idx]);
+
+			bytes_togo += slen;
+			pktnwc += slen >> 2;
+			idx++;
+			nfrags++;
+		}
+
+		if (pktnwc != pktnw) {
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		frag_size = ((le32_to_cpu(*pbc))>>16) & 0xFFFF;
+		if (((frag_size ? frag_size : bytes_togo) + len) >
+						ppd->ibmaxlen) {
+			ret = -EINVAL;
+			goto free_pbc;
+		}
+
+		if (frag_size) {
+			int pktsize, tidsmsize, n;
+
+			n = npages*((2*PAGE_SIZE/frag_size)+1);
+			pktsize = sizeof(*pkt) + sizeof(pkt->addr[0])*n;
+
+			/*
+			 * Determine if this is tid-sdma or just sdma.
+			 */
+			tiddma = (((le32_to_cpu(pbc[7])>>
+				QLOGIC_IB_I_TID_SHIFT)&
+				QLOGIC_IB_I_TID_MASK) !=
+				QLOGIC_IB_I_TID_MASK);
+
+			if (tiddma)
+				tidsmsize = iov[idx].iov_len;
+			else
+				tidsmsize = 0;
+
+			pkt = kmalloc(pktsize+tidsmsize, GFP_KERNEL);
+			if (!pkt) {
+				ret = -ENOMEM;
+				goto free_pbc;
+			}
+			pkt->largepkt = 1;
+			pkt->frag_size = frag_size;
+			pkt->addrlimit = n + ARRAY_SIZE(pkt->addr);
+
+			if (tiddma) {
+				char *tidsm = (char *)pkt + pktsize;
+				cfur = copy_from_user(tidsm,
+					iov[idx].iov_base, tidsmsize);
+				if (cfur) {
+					ret = -EFAULT;
+					goto free_pkt;
+				}
+				pkt->tidsm =
+					(struct qib_tid_session_member *)tidsm;
+				pkt->tidsmcount = tidsmsize/
+					sizeof(struct qib_tid_session_member);
+				pkt->tidsmidx = 0;
+				idx++;
+			}
+
+			/*
+			 * pbc 'fill1' field is borrowed to pass frag size,
+			 * we need to clear it after picking frag size, the
+			 * hardware requires this field to be zero.
+			 */
+			*pbc = cpu_to_le32(le32_to_cpu(*pbc) & 0x0000FFFF);
+		} else {
+			pkt = kmem_cache_alloc(pq->pkt_slab, GFP_KERNEL);
+			if (!pkt) {
+				ret = -ENOMEM;
+				goto free_pbc;
+			}
+			pkt->largepkt = 0;
+			pkt->frag_size = bytes_togo;
+			pkt->addrlimit = ARRAY_SIZE(pkt->addr);
+		}
+		pkt->bytes_togo = bytes_togo;
+		pkt->payload_size = 0;
+		pkt->counter = counter;
+		pkt->tiddma = tiddma;
+
+		/* setup the first header */
+		qib_user_sdma_init_frag(pkt, 0, /* index */
+			0, len,		/* offset, len */
+			1, 0,		/* first last desc */
+			0, 0,		/* put page, dma mapped */
+			NULL, pbc,	/* struct page, virt addr */
+			dma_addr, len);	/* dma addr, dma length */
+		pkt->index = 0;
+		pkt->naddr = 1;
+
+		if (nfrags) {
+			ret = qib_user_sdma_init_payload(dd, pq, pkt,
+							 iov + idx_save + 1,
+							 nfrags, npages);
+			if (ret < 0)
+				goto free_pkt;
+		} else {
+			/* since there is no payload, mark the
+			 * header as the last desc. */
+			pkt->addr[0].last_desc = 1;
+
+			if (dma_addr == 0) {
+				/*
+				 * the header is not dma mapped yet.
+				 * it should be from kmalloc.
+				 */
+				dma_addr = dma_map_single(&dd->pcidev->dev,
+					pbc, len, DMA_TO_DEVICE);
+				if (dma_mapping_error(&dd->pcidev->dev,
+								dma_addr)) {
+					ret = -ENOMEM;
+					goto free_pkt;
+				}
+				pkt->addr[0].addr = dma_addr;
+				pkt->addr[0].dma_mapped = 1;
+			}
+		}
+
+		counter++;
+		npkts++;
+		pkt->pq = pq;
+		pkt->index = 0; /* reset index for push on hw */
+		*ndesc += pkt->naddr;
+
+		list_add_tail(&pkt->list, list);
+	}
+
+	*maxpkts = npkts;
+	ret = idx;
+	goto done;
+
+free_pkt:
+	if (pkt->largepkt)
+		kfree(pkt);
+	else
+		kmem_cache_free(pq->pkt_slab, pkt);
+free_pbc:
+	if (dma_addr)
+		dma_pool_free(pq->header_cache, pbc, dma_addr);
+	else
+		kfree(pbc);
+free_list:
+	qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, list);
+done:
+	return ret;
+}
+
+static void qib_user_sdma_set_complete_counter(struct qib_user_sdma_queue *pq,
+					       u32 c)
+{
+	pq->sent_counter = c;
+}
+
+/* try to clean out queue -- needs pq->lock */
+static int qib_user_sdma_queue_clean(struct qib_pportdata *ppd,
+				     struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	struct list_head free_list;
+	struct qib_user_sdma_pkt *pkt;
+	struct qib_user_sdma_pkt *pkt_prev;
+	unsigned long flags;
+	int ret = 0;
+
+	if (!pq->num_sending)
+		return 0;
+
+	INIT_LIST_HEAD(&free_list);
+
+	/*
+	 * We need this spin lock here because interrupt handler
+	 * might modify this list in qib_user_sdma_send_desc(), also
+	 * we can not get interrupted, otherwise it is a deadlock.
+	 */
+	spin_lock_irqsave(&pq->sent_lock, flags);
+	list_for_each_entry_safe(pkt, pkt_prev, &pq->sent, list) {
+		s64 descd = ppd->sdma_descq_removed - pkt->added;
+
+		if (descd < 0)
+			break;
+
+		list_move_tail(&pkt->list, &free_list);
+
+		/* one more packet cleaned */
+		ret++;
+		pq->num_sending--;
+	}
+	spin_unlock_irqrestore(&pq->sent_lock, flags);
+
+	if (!list_empty(&free_list)) {
+		u32 counter;
+
+		pkt = list_entry(free_list.prev,
+				 struct qib_user_sdma_pkt, list);
+		counter = pkt->counter;
+
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		qib_user_sdma_set_complete_counter(pq, counter);
+	}
+
+	return ret;
+}
+
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq)
+{
+	if (!pq)
+		return;
+
+	pq->sdma_rb_node->refcount--;
+	if (pq->sdma_rb_node->refcount == 0) {
+		rb_erase(&pq->sdma_rb_node->node, &qib_user_sdma_rb_root);
+		kfree(pq->sdma_rb_node);
+	}
+	dma_pool_destroy(pq->header_cache);
+	kmem_cache_destroy(pq->pkt_slab);
+	kfree(pq);
+}
+
+/* clean descriptor queue, returns > 0 if some elements cleaned */
+static int qib_user_sdma_hwqueue_clean(struct qib_pportdata *ppd)
+{
+	int ret;
+	unsigned long flags;
+
+	spin_lock_irqsave(&ppd->sdma_lock, flags);
+	ret = qib_sdma_make_progress(ppd);
+	spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+	return ret;
+}
+
+/* we're in close, drain packets so that we can cleanup successfully... */
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq)
+{
+	struct qib_devdata *dd = ppd->dd;
+	unsigned long flags;
+	int i;
+
+	if (!pq)
+		return;
+
+	for (i = 0; i < QIB_USER_SDMA_DRAIN_TIMEOUT; i++) {
+		mutex_lock(&pq->lock);
+		if (!pq->num_pending && !pq->num_sending) {
+			mutex_unlock(&pq->lock);
+			break;
+		}
+		qib_user_sdma_hwqueue_clean(ppd);
+		qib_user_sdma_queue_clean(ppd, pq);
+		mutex_unlock(&pq->lock);
+		msleep(10);
+	}
+
+	if (pq->num_pending || pq->num_sending) {
+		struct qib_user_sdma_pkt *pkt;
+		struct qib_user_sdma_pkt *pkt_prev;
+		struct list_head free_list;
+
+		mutex_lock(&pq->lock);
+		spin_lock_irqsave(&ppd->sdma_lock, flags);
+		/*
+		 * Since we hold sdma_lock, it is safe without sent_lock.
+		 */
+		if (pq->num_pending) {
+			list_for_each_entry_safe(pkt, pkt_prev,
+					&ppd->sdma_userpending, list) {
+				if (pkt->pq == pq) {
+					list_move_tail(&pkt->list, &pq->sent);
+					pq->num_pending--;
+					pq->num_sending++;
+				}
+			}
+		}
+		spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+
+		qib_dev_err(dd, "user sdma lists not empty: forcing!\n");
+		INIT_LIST_HEAD(&free_list);
+		list_splice_init(&pq->sent, &free_list);
+		pq->num_sending = 0;
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &free_list);
+		mutex_unlock(&pq->lock);
+	}
+}
+
+static inline __le64 qib_sdma_make_desc0(u8 gen,
+					 u64 addr, u64 dwlen, u64 dwoffset)
+{
+	return cpu_to_le64(/* SDmaPhyAddr[31:0] */
+			   ((addr & 0xfffffffcULL) << 32) |
+			   /* SDmaGeneration[1:0] */
+			   ((gen & 3ULL) << 30) |
+			   /* SDmaDwordCount[10:0] */
+			   ((dwlen & 0x7ffULL) << 16) |
+			   /* SDmaBufOffset[12:2] */
+			   (dwoffset & 0x7ffULL));
+}
+
+static inline __le64 qib_sdma_make_first_desc0(__le64 descq)
+{
+	return descq | cpu_to_le64(1ULL << 12);
+}
+
+static inline __le64 qib_sdma_make_last_desc0(__le64 descq)
+{
+					      /* last */  /* dma head */
+	return descq | cpu_to_le64(1ULL << 11 | 1ULL << 13);
+}
+
+static inline __le64 qib_sdma_make_desc1(u64 addr)
+{
+	/* SDmaPhyAddr[47:32] */
+	return cpu_to_le64(addr >> 32);
+}
+
+static void qib_user_sdma_send_frag(struct qib_pportdata *ppd,
+				    struct qib_user_sdma_pkt *pkt, int idx,
+				    unsigned ofs, u16 tail, u8 gen)
+{
+	const u64 addr = (u64) pkt->addr[idx].addr +
+		(u64) pkt->addr[idx].offset;
+	const u64 dwlen = (u64) pkt->addr[idx].length / 4;
+	__le64 *descqp;
+	__le64 descq0;
+
+	descqp = &ppd->sdma_descq[tail].qw[0];
+
+	descq0 = qib_sdma_make_desc0(gen, addr, dwlen, ofs);
+	if (pkt->addr[idx].first_desc)
+		descq0 = qib_sdma_make_first_desc0(descq0);
+	if (pkt->addr[idx].last_desc) {
+		descq0 = qib_sdma_make_last_desc0(descq0);
+		if (ppd->sdma_intrequest) {
+			descq0 |= cpu_to_le64(1ULL << 15);
+			ppd->sdma_intrequest = 0;
+		}
+	}
+
+	descqp[0] = descq0;
+	descqp[1] = qib_sdma_make_desc1(addr);
+}
+
+void qib_user_sdma_send_desc(struct qib_pportdata *ppd,
+				struct list_head *pktlist)
+{
+	struct qib_devdata *dd = ppd->dd;
+	u16 nfree, nsent;
+	u16 tail, tail_c;
+	u8 gen, gen_c;
+
+	nfree = qib_sdma_descq_freecnt(ppd);
+	if (!nfree)
+		return;
+
+retry:
+	nsent = 0;
+	tail_c = tail = ppd->sdma_descq_tail;
+	gen_c = gen = ppd->sdma_generation;
+	while (!list_empty(pktlist)) {
+		struct qib_user_sdma_pkt *pkt =
+			list_entry(pktlist->next, struct qib_user_sdma_pkt,
+				   list);
+		int i, j, c = 0;
+		unsigned ofs = 0;
+		u16 dtail = tail;
+
+		for (i = pkt->index; i < pkt->naddr && nfree; i++) {
+			qib_user_sdma_send_frag(ppd, pkt, i, ofs, tail, gen);
+			ofs += pkt->addr[i].length >> 2;
+
+			if (++tail == ppd->sdma_descq_cnt) {
+				tail = 0;
+				++gen;
+				ppd->sdma_intrequest = 1;
+			} else if (tail == (ppd->sdma_descq_cnt>>1)) {
+				ppd->sdma_intrequest = 1;
+			}
+			nfree--;
+			if (pkt->addr[i].last_desc == 0)
+				continue;
+
+			/*
+			 * If the packet is >= 2KB mtu equivalent, we
+			 * have to use the large buffers, and have to
+			 * mark each descriptor as part of a large
+			 * buffer packet.
+			 */
+			if (ofs > dd->piosize2kmax_dwords) {
+				for (j = pkt->index; j <= i; j++) {
+					ppd->sdma_descq[dtail].qw[0] |=
+						cpu_to_le64(1ULL << 14);
+					if (++dtail == ppd->sdma_descq_cnt)
+						dtail = 0;
+				}
+			}
+			c += i + 1 - pkt->index;
+			pkt->index = i + 1; /* index for next first */
+			tail_c = dtail = tail;
+			gen_c = gen;
+			ofs = 0;  /* reset for next packet */
+		}
+
+		ppd->sdma_descq_added += c;
+		nsent += c;
+		if (pkt->index == pkt->naddr) {
+			pkt->added = ppd->sdma_descq_added;
+			pkt->pq->added = pkt->added;
+			pkt->pq->num_pending--;
+			spin_lock(&pkt->pq->sent_lock);
+			pkt->pq->num_sending++;
+			list_move_tail(&pkt->list, &pkt->pq->sent);
+			spin_unlock(&pkt->pq->sent_lock);
+		}
+		if (!nfree || (nsent<<2) > ppd->sdma_descq_cnt)
+			break;
+	}
+
+	/* advance the tail on the chip if necessary */
+	if (ppd->sdma_descq_tail != tail_c) {
+		ppd->sdma_generation = gen_c;
+		dd->f_sdma_update_tail(ppd, tail_c);
+	}
+
+	if (nfree && !list_empty(pktlist))
+		goto retry;
+
+	return;
+}
+
+/* pq->lock must be held, get packets on the wire... */
+static int qib_user_sdma_push_pkts(struct qib_pportdata *ppd,
+				 struct qib_user_sdma_queue *pq,
+				 struct list_head *pktlist, int count)
+{
+	unsigned long flags;
+
+	if (unlikely(!(ppd->lflags & QIBL_LINKACTIVE)))
+		return -ECOMM;
+
+	/* non-blocking mode */
+	if (pq->sdma_rb_node->refcount > 1) {
+		spin_lock_irqsave(&ppd->sdma_lock, flags);
+		if (unlikely(!__qib_sdma_running(ppd))) {
+			spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+			return -ECOMM;
+		}
+		pq->num_pending += count;
+		list_splice_tail_init(pktlist, &ppd->sdma_userpending);
+		qib_user_sdma_send_desc(ppd, &ppd->sdma_userpending);
+		spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+		return 0;
+	}
+
+	/* In this case, descriptors from this process are not
+	 * linked to ppd pending queue, interrupt handler
+	 * won't update this process, it is OK to directly
+	 * modify without sdma lock.
+	 */
+
+
+	pq->num_pending += count;
+	/*
+	 * Blocking mode for single rail process, we must
+	 * release/regain sdma_lock to give other process
+	 * chance to make progress. This is important for
+	 * performance.
+	 */
+	do {
+		spin_lock_irqsave(&ppd->sdma_lock, flags);
+		if (unlikely(!__qib_sdma_running(ppd))) {
+			spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+			return -ECOMM;
+		}
+		qib_user_sdma_send_desc(ppd, pktlist);
+		if (!list_empty(pktlist))
+			qib_sdma_make_progress(ppd);
+		spin_unlock_irqrestore(&ppd->sdma_lock, flags);
+	} while (!list_empty(pktlist));
+
+	return 0;
+}
+
+int qib_user_sdma_writev(struct qib_ctxtdata *rcd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim)
+{
+	struct qib_devdata *dd = rcd->dd;
+	struct qib_pportdata *ppd = rcd->ppd;
+	int ret = 0;
+	struct list_head list;
+	int npkts = 0;
+
+	INIT_LIST_HEAD(&list);
+
+	mutex_lock(&pq->lock);
+
+	/* why not -ECOMM like qib_user_sdma_push_pkts() below? */
+	if (!qib_sdma_running(ppd))
+		goto done_unlock;
+
+	/* if I have packets not complete yet */
+	if (pq->added > ppd->sdma_descq_removed)
+		qib_user_sdma_hwqueue_clean(ppd);
+	/* if I have complete packets to be freed */
+	if (pq->num_sending)
+		qib_user_sdma_queue_clean(ppd, pq);
+
+	while (dim) {
+		int mxp = 1;
+		int ndesc = 0;
+
+		ret = qib_user_sdma_queue_pkts(dd, ppd, pq,
+				iov, dim, &list, &mxp, &ndesc);
+		if (ret < 0)
+			goto done_unlock;
+		else {
+			dim -= ret;
+			iov += ret;
+		}
+
+		/* force packets onto the sdma hw queue... */
+		if (!list_empty(&list)) {
+			/*
+			 * Lazily clean hw queue.
+			 */
+			if (qib_sdma_descq_freecnt(ppd) < ndesc) {
+				qib_user_sdma_hwqueue_clean(ppd);
+				if (pq->num_sending)
+					qib_user_sdma_queue_clean(ppd, pq);
+			}
+
+			ret = qib_user_sdma_push_pkts(ppd, pq, &list, mxp);
+			if (ret < 0)
+				goto done_unlock;
+			else {
+				npkts += mxp;
+				pq->counter += mxp;
+			}
+		}
+	}
+
+done_unlock:
+	if (!list_empty(&list))
+		qib_user_sdma_free_pkt_list(&dd->pcidev->dev, pq, &list);
+	mutex_unlock(&pq->lock);
+
+	return (ret < 0) ? ret : npkts;
+}
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq)
+{
+	int ret = 0;
+
+	mutex_lock(&pq->lock);
+	qib_user_sdma_hwqueue_clean(ppd);
+	ret = qib_user_sdma_queue_clean(ppd, pq);
+	mutex_unlock(&pq->lock);
+
+	return ret;
+}
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->sent_counter : 0;
+}
+
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq)
+{
+	return pq ? pq->counter : 0;
+}
diff --git a/drivers/infiniband/hw/qib/qib_user_sdma.h b/drivers/infiniband/hw/qib/qib_user_sdma.h
new file mode 100644
index 00000000000..ce8cbaf6a5c
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_user_sdma.h
@@ -0,0 +1,52 @@
+/*
+ * Copyright (c) 2007, 2008 QLogic Corporation. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+#include <linux/device.h>
+
+struct qib_user_sdma_queue;
+
+struct qib_user_sdma_queue *
+qib_user_sdma_queue_create(struct device *dev, int unit, int port, int sport);
+void qib_user_sdma_queue_destroy(struct qib_user_sdma_queue *pq);
+
+int qib_user_sdma_writev(struct qib_ctxtdata *pd,
+			 struct qib_user_sdma_queue *pq,
+			 const struct iovec *iov,
+			 unsigned long dim);
+
+int qib_user_sdma_make_progress(struct qib_pportdata *ppd,
+				struct qib_user_sdma_queue *pq);
+
+void qib_user_sdma_queue_drain(struct qib_pportdata *ppd,
+			       struct qib_user_sdma_queue *pq);
+
+u32 qib_user_sdma_complete_counter(const struct qib_user_sdma_queue *pq);
+u32 qib_user_sdma_inflight_counter(struct qib_user_sdma_queue *pq);
diff --git a/drivers/infiniband/hw/qib/qib_verbs.c b/drivers/infiniband/hw/qib/qib_verbs.c
new file mode 100644
index 00000000000..9bcfbd84298
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs.c
@@ -0,0 +1,2336 @@
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <rdma/ib_mad.h>
+#include <rdma/ib_user_verbs.h>
+#include <linux/io.h>
+#include <linux/module.h>
+#include <linux/utsname.h>
+#include <linux/rculist.h>
+#include <linux/mm.h>
+#include <linux/random.h>
+
+#include "qib.h"
+#include "qib_common.h"
+
+static unsigned int ib_qib_qp_table_size = 256;
+module_param_named(qp_table_size, ib_qib_qp_table_size, uint, S_IRUGO);
+MODULE_PARM_DESC(qp_table_size, "QP table size");
+
+unsigned int ib_qib_lkey_table_size = 16;
+module_param_named(lkey_table_size, ib_qib_lkey_table_size, uint,
+		   S_IRUGO);
+MODULE_PARM_DESC(lkey_table_size,
+		 "LKEY table size in bits (2^n, 1 <= n <= 23)");
+
+static unsigned int ib_qib_max_pds = 0xFFFF;
+module_param_named(max_pds, ib_qib_max_pds, uint, S_IRUGO);
+MODULE_PARM_DESC(max_pds,
+		 "Maximum number of protection domains to support");
+
+static unsigned int ib_qib_max_ahs = 0xFFFF;
+module_param_named(max_ahs, ib_qib_max_ahs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_ahs, "Maximum number of address handles to support");
+
+unsigned int ib_qib_max_cqes = 0x2FFFF;
+module_param_named(max_cqes, ib_qib_max_cqes, uint, S_IRUGO);
+MODULE_PARM_DESC(max_cqes,
+		 "Maximum number of completion queue entries to support");
+
+unsigned int ib_qib_max_cqs = 0x1FFFF;
+module_param_named(max_cqs, ib_qib_max_cqs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_cqs, "Maximum number of completion queues to support");
+
+unsigned int ib_qib_max_qp_wrs = 0x3FFF;
+module_param_named(max_qp_wrs, ib_qib_max_qp_wrs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_qp_wrs, "Maximum number of QP WRs to support");
+
+unsigned int ib_qib_max_qps = 16384;
+module_param_named(max_qps, ib_qib_max_qps, uint, S_IRUGO);
+MODULE_PARM_DESC(max_qps, "Maximum number of QPs to support");
+
+unsigned int ib_qib_max_sges = 0x60;
+module_param_named(max_sges, ib_qib_max_sges, uint, S_IRUGO);
+MODULE_PARM_DESC(max_sges, "Maximum number of SGEs to support");
+
+unsigned int ib_qib_max_mcast_grps = 16384;
+module_param_named(max_mcast_grps, ib_qib_max_mcast_grps, uint, S_IRUGO);
+MODULE_PARM_DESC(max_mcast_grps,
+		 "Maximum number of multicast groups to support");
+
+unsigned int ib_qib_max_mcast_qp_attached = 16;
+module_param_named(max_mcast_qp_attached, ib_qib_max_mcast_qp_attached,
+		   uint, S_IRUGO);
+MODULE_PARM_DESC(max_mcast_qp_attached,
+		 "Maximum number of attached QPs to support");
+
+unsigned int ib_qib_max_srqs = 1024;
+module_param_named(max_srqs, ib_qib_max_srqs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srqs, "Maximum number of SRQs to support");
+
+unsigned int ib_qib_max_srq_sges = 128;
+module_param_named(max_srq_sges, ib_qib_max_srq_sges, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srq_sges, "Maximum number of SRQ SGEs to support");
+
+unsigned int ib_qib_max_srq_wrs = 0x1FFFF;
+module_param_named(max_srq_wrs, ib_qib_max_srq_wrs, uint, S_IRUGO);
+MODULE_PARM_DESC(max_srq_wrs, "Maximum number of SRQ WRs support");
+
+static unsigned int ib_qib_disable_sma;
+module_param_named(disable_sma, ib_qib_disable_sma, uint, S_IWUSR | S_IRUGO);
+MODULE_PARM_DESC(disable_sma, "Disable the SMA");
+
+/*
+ * Note that it is OK to post send work requests in the SQE and ERR
+ * states; qib_do_send() will process them and generate error
+ * completions as per IB 1.2 C10-96.
+ */
+const int ib_qib_state_ops[IB_QPS_ERR + 1] = {
+	[IB_QPS_RESET] = 0,
+	[IB_QPS_INIT] = QIB_POST_RECV_OK,
+	[IB_QPS_RTR] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK,
+	[IB_QPS_RTS] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_PROCESS_SEND_OK |
+	    QIB_PROCESS_NEXT_SEND_OK,
+	[IB_QPS_SQD] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_PROCESS_SEND_OK,
+	[IB_QPS_SQE] = QIB_POST_RECV_OK | QIB_PROCESS_RECV_OK |
+	    QIB_POST_SEND_OK | QIB_FLUSH_SEND,
+	[IB_QPS_ERR] = QIB_POST_RECV_OK | QIB_FLUSH_RECV |
+	    QIB_POST_SEND_OK | QIB_FLUSH_SEND,
+};
+
+struct qib_ucontext {
+	struct ib_ucontext ibucontext;
+};
+
+static inline struct qib_ucontext *to_iucontext(struct ib_ucontext
+						  *ibucontext)
+{
+	return container_of(ibucontext, struct qib_ucontext, ibucontext);
+}
+
+/*
+ * Translate ib_wr_opcode into ib_wc_opcode.
+ */
+const enum ib_wc_opcode ib_qib_wc_opcode[] = {
+	[IB_WR_RDMA_WRITE] = IB_WC_RDMA_WRITE,
+	[IB_WR_RDMA_WRITE_WITH_IMM] = IB_WC_RDMA_WRITE,
+	[IB_WR_SEND] = IB_WC_SEND,
+	[IB_WR_SEND_WITH_IMM] = IB_WC_SEND,
+	[IB_WR_RDMA_READ] = IB_WC_RDMA_READ,
+	[IB_WR_ATOMIC_CMP_AND_SWP] = IB_WC_COMP_SWAP,
+	[IB_WR_ATOMIC_FETCH_AND_ADD] = IB_WC_FETCH_ADD
+};
+
+/*
+ * System image GUID.
+ */
+__be64 ib_qib_sys_image_guid;
+
+/**
+ * qib_copy_sge - copy data to SGE memory
+ * @ss: the SGE state
+ * @data: the data to copy
+ * @length: the length of the data
+ */
+void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length, int release)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		memcpy(sge->vaddr, data, len);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (release)
+				qib_put_mr(sge->mr);
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		data += len;
+		length -= len;
+	}
+}
+
+/**
+ * qib_skip_sge - skip over SGE memory - XXX almost dup of prev func
+ * @ss: the SGE state
+ * @length: the number of bytes to skip
+ */
+void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (release)
+				qib_put_mr(sge->mr);
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		length -= len;
+	}
+}
+
+/*
+ * Count the number of DMA descriptors needed to send length bytes of data.
+ * Don't modify the qib_sge_state to get the count.
+ * Return zero if any of the segments is not aligned.
+ */
+static u32 qib_count_sge(struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sg_list = ss->sg_list;
+	struct qib_sge sge = ss->sge;
+	u8 num_sge = ss->num_sge;
+	u32 ndesc = 1;  /* count the header */
+
+	while (length) {
+		u32 len = sge.length;
+
+		if (len > length)
+			len = length;
+		if (len > sge.sge_length)
+			len = sge.sge_length;
+		BUG_ON(len == 0);
+		if (((long) sge.vaddr & (sizeof(u32) - 1)) ||
+		    (len != length && (len & (sizeof(u32) - 1)))) {
+			ndesc = 0;
+			break;
+		}
+		ndesc++;
+		sge.vaddr += len;
+		sge.length -= len;
+		sge.sge_length -= len;
+		if (sge.sge_length == 0) {
+			if (--num_sge)
+				sge = *sg_list++;
+		} else if (sge.length == 0 && sge.mr->lkey) {
+			if (++sge.n >= QIB_SEGSZ) {
+				if (++sge.m >= sge.mr->mapsz)
+					break;
+				sge.n = 0;
+			}
+			sge.vaddr =
+				sge.mr->map[sge.m]->segs[sge.n].vaddr;
+			sge.length =
+				sge.mr->map[sge.m]->segs[sge.n].length;
+		}
+		length -= len;
+	}
+	return ndesc;
+}
+
+/*
+ * Copy from the SGEs to the data buffer.
+ */
+static void qib_copy_from_sge(void *data, struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	while (length) {
+		u32 len = sge->length;
+
+		if (len > length)
+			len = length;
+		if (len > sge->sge_length)
+			len = sge->sge_length;
+		BUG_ON(len == 0);
+		memcpy(data, sge->vaddr, len);
+		sge->vaddr += len;
+		sge->length -= len;
+		sge->sge_length -= len;
+		if (sge->sge_length == 0) {
+			if (--ss->num_sge)
+				*sge = *ss->sg_list++;
+		} else if (sge->length == 0 && sge->mr->lkey) {
+			if (++sge->n >= QIB_SEGSZ) {
+				if (++sge->m >= sge->mr->mapsz)
+					break;
+				sge->n = 0;
+			}
+			sge->vaddr =
+				sge->mr->map[sge->m]->segs[sge->n].vaddr;
+			sge->length =
+				sge->mr->map[sge->m]->segs[sge->n].length;
+		}
+		data += len;
+		length -= len;
+	}
+}
+
+/**
+ * qib_post_one_send - post one RC, UC, or UD send work request
+ * @qp: the QP to post on
+ * @wr: the work request to send
+ */
+static int qib_post_one_send(struct qib_qp *qp, struct ib_send_wr *wr,
+	int *scheduled)
+{
+	struct qib_swqe *wqe;
+	u32 next;
+	int i;
+	int j;
+	int acc;
+	int ret;
+	unsigned long flags;
+	struct qib_lkey_table *rkt;
+	struct qib_pd *pd;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+
+	/* Check that state is OK to post send. */
+	if (unlikely(!(ib_qib_state_ops[qp->state] & QIB_POST_SEND_OK)))
+		goto bail_inval;
+
+	/* IB spec says that num_sge == 0 is OK. */
+	if (wr->num_sge > qp->s_max_sge)
+		goto bail_inval;
+
+	/*
+	 * Don't allow RDMA reads or atomic operations on UC or
+	 * undefined operations.
+	 * Make sure buffer is large enough to hold the result for atomics.
+	 */
+	if (wr->opcode == IB_WR_FAST_REG_MR) {
+		if (qib_fast_reg_mr(qp, wr))
+			goto bail_inval;
+	} else if (qp->ibqp.qp_type == IB_QPT_UC) {
+		if ((unsigned) wr->opcode >= IB_WR_RDMA_READ)
+			goto bail_inval;
+	} else if (qp->ibqp.qp_type != IB_QPT_RC) {
+		/* Check IB_QPT_SMI, IB_QPT_GSI, IB_QPT_UD opcode */
+		if (wr->opcode != IB_WR_SEND &&
+		    wr->opcode != IB_WR_SEND_WITH_IMM)
+			goto bail_inval;
+		/* Check UD destination address PD */
+		if (qp->ibqp.pd != wr->wr.ud.ah->pd)
+			goto bail_inval;
+	} else if ((unsigned) wr->opcode > IB_WR_ATOMIC_FETCH_AND_ADD)
+		goto bail_inval;
+	else if (wr->opcode >= IB_WR_ATOMIC_CMP_AND_SWP &&
+		   (wr->num_sge == 0 ||
+		    wr->sg_list[0].length < sizeof(u64) ||
+		    wr->sg_list[0].addr & (sizeof(u64) - 1)))
+		goto bail_inval;
+	else if (wr->opcode >= IB_WR_RDMA_READ && !qp->s_max_rd_atomic)
+		goto bail_inval;
+
+	next = qp->s_head + 1;
+	if (next >= qp->s_size)
+		next = 0;
+	if (next == qp->s_last) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+
+	rkt = &to_idev(qp->ibqp.device)->lk_table;
+	pd = to_ipd(qp->ibqp.pd);
+	wqe = get_swqe_ptr(qp, qp->s_head);
+	wqe->wr = *wr;
+	wqe->length = 0;
+	j = 0;
+	if (wr->num_sge) {
+		acc = wr->opcode >= IB_WR_RDMA_READ ?
+			IB_ACCESS_LOCAL_WRITE : 0;
+		for (i = 0; i < wr->num_sge; i++) {
+			u32 length = wr->sg_list[i].length;
+			int ok;
+
+			if (length == 0)
+				continue;
+			ok = qib_lkey_ok(rkt, pd, &wqe->sg_list[j],
+					 &wr->sg_list[i], acc);
+			if (!ok)
+				goto bail_inval_free;
+			wqe->length += length;
+			j++;
+		}
+		wqe->wr.num_sge = j;
+	}
+	if (qp->ibqp.qp_type == IB_QPT_UC ||
+	    qp->ibqp.qp_type == IB_QPT_RC) {
+		if (wqe->length > 0x80000000U)
+			goto bail_inval_free;
+	} else if (wqe->length > (dd_from_ibdev(qp->ibqp.device)->pport +
+				  qp->port_num - 1)->ibmtu)
+		goto bail_inval_free;
+	else
+		atomic_inc(&to_iah(wr->wr.ud.ah)->refcount);
+	wqe->ssn = qp->s_ssn++;
+	qp->s_head = next;
+
+	ret = 0;
+	goto bail;
+
+bail_inval_free:
+	while (j) {
+		struct qib_sge *sge = &wqe->sg_list[--j];
+
+		qib_put_mr(sge->mr);
+	}
+bail_inval:
+	ret = -EINVAL;
+bail:
+	if (!ret && !wr->next &&
+	 !qib_sdma_empty(
+	   dd_from_ibdev(qp->ibqp.device)->pport + qp->port_num - 1)) {
+		qib_schedule_send(qp);
+		*scheduled = 1;
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+/**
+ * qib_post_send - post a send on a QP
+ * @ibqp: the QP to post the send on
+ * @wr: the list of work requests to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+			 struct ib_send_wr **bad_wr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	int err = 0;
+	int scheduled = 0;
+
+	for (; wr; wr = wr->next) {
+		err = qib_post_one_send(qp, wr, &scheduled);
+		if (err) {
+			*bad_wr = wr;
+			goto bail;
+		}
+	}
+
+	/* Try to do the send work in the caller's context. */
+	if (!scheduled)
+		qib_do_send(&qp->s_work);
+
+bail:
+	return err;
+}
+
+/**
+ * qib_post_receive - post a receive on a QP
+ * @ibqp: the QP to post the receive on
+ * @wr: the WR to post
+ * @bad_wr: the first bad WR is put here
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_post_receive(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+			    struct ib_recv_wr **bad_wr)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_rwq *wq = qp->r_rq.wq;
+	unsigned long flags;
+	int ret;
+
+	/* Check that state is OK to post receive. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_POST_RECV_OK) || !wq) {
+		*bad_wr = wr;
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	for (; wr; wr = wr->next) {
+		struct qib_rwqe *wqe;
+		u32 next;
+		int i;
+
+		if ((unsigned) wr->num_sge > qp->r_rq.max_sge) {
+			*bad_wr = wr;
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		spin_lock_irqsave(&qp->r_rq.lock, flags);
+		next = wq->head + 1;
+		if (next >= qp->r_rq.size)
+			next = 0;
+		if (next == wq->tail) {
+			spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+			*bad_wr = wr;
+			ret = -ENOMEM;
+			goto bail;
+		}
+
+		wqe = get_rwqe_ptr(&qp->r_rq, wq->head);
+		wqe->wr_id = wr->wr_id;
+		wqe->num_sge = wr->num_sge;
+		for (i = 0; i < wr->num_sge; i++)
+			wqe->sg_list[i] = wr->sg_list[i];
+		/* Make sure queue entry is written before the head index. */
+		smp_wmb();
+		wq->head = next;
+		spin_unlock_irqrestore(&qp->r_rq.lock, flags);
+	}
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_qp_rcv - processing an incoming packet on a QP
+ * @rcd: the context pointer
+ * @hdr: the packet header
+ * @has_grh: true if the packet has a GRH
+ * @data: the packet data
+ * @tlen: the packet length
+ * @qp: the QP the packet came on
+ *
+ * This is called from qib_ib_rcv() to process an incoming packet
+ * for the given QP.
+ * Called at interrupt level.
+ */
+static void qib_qp_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		       int has_grh, void *data, u32 tlen, struct qib_qp *qp)
+{
+	struct qib_ibport *ibp = &rcd->ppd->ibport_data;
+
+	spin_lock(&qp->r_lock);
+
+	/* Check for valid receive state. */
+	if (!(ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK)) {
+		ibp->n_pkt_drops++;
+		goto unlock;
+	}
+
+	switch (qp->ibqp.qp_type) {
+	case IB_QPT_SMI:
+	case IB_QPT_GSI:
+		if (ib_qib_disable_sma)
+			break;
+		/* FALLTHROUGH */
+	case IB_QPT_UD:
+		qib_ud_rcv(ibp, hdr, has_grh, data, tlen, qp);
+		break;
+
+	case IB_QPT_RC:
+		qib_rc_rcv(rcd, hdr, has_grh, data, tlen, qp);
+		break;
+
+	case IB_QPT_UC:
+		qib_uc_rcv(ibp, hdr, has_grh, data, tlen, qp);
+		break;
+
+	default:
+		break;
+	}
+
+unlock:
+	spin_unlock(&qp->r_lock);
+}
+
+/**
+ * qib_ib_rcv - process an incoming packet
+ * @rcd: the context pointer
+ * @rhdr: the header of the packet
+ * @data: the packet payload
+ * @tlen: the packet length
+ *
+ * This is called from qib_kreceive() to process an incoming packet at
+ * interrupt level. Tlen is the length of the header + data + CRC in bytes.
+ */
+void qib_ib_rcv(struct qib_ctxtdata *rcd, void *rhdr, void *data, u32 tlen)
+{
+	struct qib_pportdata *ppd = rcd->ppd;
+	struct qib_ibport *ibp = &ppd->ibport_data;
+	struct qib_ib_header *hdr = rhdr;
+	struct qib_other_headers *ohdr;
+	struct qib_qp *qp;
+	u32 qp_num;
+	int lnh;
+	u8 opcode;
+	u16 lid;
+
+	/* 24 == LRH+BTH+CRC */
+	if (unlikely(tlen < 24))
+		goto drop;
+
+	/* Check for a valid destination LID (see ch. 7.11.1). */
+	lid = be16_to_cpu(hdr->lrh[1]);
+	if (lid < QIB_MULTICAST_LID_BASE) {
+		lid &= ~((1 << ppd->lmc) - 1);
+		if (unlikely(lid != ppd->lid))
+			goto drop;
+	}
+
+	/* Check for GRH */
+	lnh = be16_to_cpu(hdr->lrh[0]) & 3;
+	if (lnh == QIB_LRH_BTH)
+		ohdr = &hdr->u.oth;
+	else if (lnh == QIB_LRH_GRH) {
+		u32 vtf;
+
+		ohdr = &hdr->u.l.oth;
+		if (hdr->u.l.grh.next_hdr != IB_GRH_NEXT_HDR)
+			goto drop;
+		vtf = be32_to_cpu(hdr->u.l.grh.version_tclass_flow);
+		if ((vtf >> IB_GRH_VERSION_SHIFT) != IB_GRH_VERSION)
+			goto drop;
+	} else
+		goto drop;
+
+	opcode = (be32_to_cpu(ohdr->bth[0]) >> 24) & 0x7f;
+#ifdef CONFIG_DEBUG_FS
+	rcd->opstats->stats[opcode].n_bytes += tlen;
+	rcd->opstats->stats[opcode].n_packets++;
+#endif
+
+	/* Get the destination QP number. */
+	qp_num = be32_to_cpu(ohdr->bth[1]) & QIB_QPN_MASK;
+	if (qp_num == QIB_MULTICAST_QPN) {
+		struct qib_mcast *mcast;
+		struct qib_mcast_qp *p;
+
+		if (lnh != QIB_LRH_GRH)
+			goto drop;
+		mcast = qib_mcast_find(ibp, &hdr->u.l.grh.dgid);
+		if (mcast == NULL)
+			goto drop;
+		this_cpu_inc(ibp->pmastats->n_multicast_rcv);
+		list_for_each_entry_rcu(p, &mcast->qp_list, list)
+			qib_qp_rcv(rcd, hdr, 1, data, tlen, p->qp);
+		/*
+		 * Notify qib_multicast_detach() if it is waiting for us
+		 * to finish.
+		 */
+		if (atomic_dec_return(&mcast->refcount) <= 1)
+			wake_up(&mcast->wait);
+	} else {
+		if (rcd->lookaside_qp) {
+			if (rcd->lookaside_qpn != qp_num) {
+				if (atomic_dec_and_test(
+					&rcd->lookaside_qp->refcount))
+					wake_up(
+					 &rcd->lookaside_qp->wait);
+				rcd->lookaside_qp = NULL;
+			}
+		}
+		if (!rcd->lookaside_qp) {
+			qp = qib_lookup_qpn(ibp, qp_num);
+			if (!qp)
+				goto drop;
+			rcd->lookaside_qp = qp;
+			rcd->lookaside_qpn = qp_num;
+		} else
+			qp = rcd->lookaside_qp;
+		this_cpu_inc(ibp->pmastats->n_unicast_rcv);
+		qib_qp_rcv(rcd, hdr, lnh == QIB_LRH_GRH, data, tlen, qp);
+	}
+	return;
+
+drop:
+	ibp->n_pkt_drops++;
+}
+
+/*
+ * This is called from a timer to check for QPs
+ * which need kernel memory in order to send a packet.
+ */
+static void mem_timer(unsigned long data)
+{
+	struct qib_ibdev *dev = (struct qib_ibdev *) data;
+	struct list_head *list = &dev->memwait;
+	struct qib_qp *qp = NULL;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->pending_lock, flags);
+	if (!list_empty(list)) {
+		qp = list_entry(list->next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		if (!list_empty(list))
+			mod_timer(&dev->mem_timer, jiffies + 1);
+	}
+	spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+	if (qp) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_KMEM) {
+			qp->s_flags &= ~QIB_S_WAIT_KMEM;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+static void update_sge(struct qib_sge_state *ss, u32 length)
+{
+	struct qib_sge *sge = &ss->sge;
+
+	sge->vaddr += length;
+	sge->length -= length;
+	sge->sge_length -= length;
+	if (sge->sge_length == 0) {
+		if (--ss->num_sge)
+			*sge = *ss->sg_list++;
+	} else if (sge->length == 0 && sge->mr->lkey) {
+		if (++sge->n >= QIB_SEGSZ) {
+			if (++sge->m >= sge->mr->mapsz)
+				return;
+			sge->n = 0;
+		}
+		sge->vaddr = sge->mr->map[sge->m]->segs[sge->n].vaddr;
+		sge->length = sge->mr->map[sge->m]->segs[sge->n].length;
+	}
+}
+
+#ifdef __LITTLE_ENDIAN
+static inline u32 get_upper_bits(u32 data, u32 shift)
+{
+	return data >> shift;
+}
+
+static inline u32 set_upper_bits(u32 data, u32 shift)
+{
+	return data << shift;
+}
+
+static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
+{
+	data <<= ((sizeof(u32) - n) * BITS_PER_BYTE);
+	data >>= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
+	return data;
+}
+#else
+static inline u32 get_upper_bits(u32 data, u32 shift)
+{
+	return data << shift;
+}
+
+static inline u32 set_upper_bits(u32 data, u32 shift)
+{
+	return data >> shift;
+}
+
+static inline u32 clear_upper_bytes(u32 data, u32 n, u32 off)
+{
+	data >>= ((sizeof(u32) - n) * BITS_PER_BYTE);
+	data <<= ((sizeof(u32) - n - off) * BITS_PER_BYTE);
+	return data;
+}
+#endif
+
+static void copy_io(u32 __iomem *piobuf, struct qib_sge_state *ss,
+		    u32 length, unsigned flush_wc)
+{
+	u32 extra = 0;
+	u32 data = 0;
+	u32 last;
+
+	while (1) {
+		u32 len = ss->sge.length;
+		u32 off;
+
+		if (len > length)
+			len = length;
+		if (len > ss->sge.sge_length)
+			len = ss->sge.sge_length;
+		BUG_ON(len == 0);
+		/* If the source address is not aligned, try to align it. */
+		off = (unsigned long)ss->sge.vaddr & (sizeof(u32) - 1);
+		if (off) {
+			u32 *addr = (u32 *)((unsigned long)ss->sge.vaddr &
+					    ~(sizeof(u32) - 1));
+			u32 v = get_upper_bits(*addr, off * BITS_PER_BYTE);
+			u32 y;
+
+			y = sizeof(u32) - off;
+			if (len > y)
+				len = y;
+			if (len + extra >= sizeof(u32)) {
+				data |= set_upper_bits(v, extra *
+						       BITS_PER_BYTE);
+				len = sizeof(u32) - extra;
+				if (len == length) {
+					last = data;
+					break;
+				}
+				__raw_writel(data, piobuf);
+				piobuf++;
+				extra = 0;
+				data = 0;
+			} else {
+				/* Clear unused upper bytes */
+				data |= clear_upper_bytes(v, len, extra);
+				if (len == length) {
+					last = data;
+					break;
+				}
+				extra += len;
+			}
+		} else if (extra) {
+			/* Source address is aligned. */
+			u32 *addr = (u32 *) ss->sge.vaddr;
+			int shift = extra * BITS_PER_BYTE;
+			int ushift = 32 - shift;
+			u32 l = len;
+
+			while (l >= sizeof(u32)) {
+				u32 v = *addr;
+
+				data |= set_upper_bits(v, shift);
+				__raw_writel(data, piobuf);
+				data = get_upper_bits(v, ushift);
+				piobuf++;
+				addr++;
+				l -= sizeof(u32);
+			}
+			/*
+			 * We still have 'extra' number of bytes leftover.
+			 */
+			if (l) {
+				u32 v = *addr;
+
+				if (l + extra >= sizeof(u32)) {
+					data |= set_upper_bits(v, shift);
+					len -= l + extra - sizeof(u32);
+					if (len == length) {
+						last = data;
+						break;
+					}
+					__raw_writel(data, piobuf);
+					piobuf++;
+					extra = 0;
+					data = 0;
+				} else {
+					/* Clear unused upper bytes */
+					data |= clear_upper_bytes(v, l, extra);
+					if (len == length) {
+						last = data;
+						break;
+					}
+					extra += l;
+				}
+			} else if (len == length) {
+				last = data;
+				break;
+			}
+		} else if (len == length) {
+			u32 w;
+
+			/*
+			 * Need to round up for the last dword in the
+			 * packet.
+			 */
+			w = (len + 3) >> 2;
+			qib_pio_copy(piobuf, ss->sge.vaddr, w - 1);
+			piobuf += w - 1;
+			last = ((u32 *) ss->sge.vaddr)[w - 1];
+			break;
+		} else {
+			u32 w = len >> 2;
+
+			qib_pio_copy(piobuf, ss->sge.vaddr, w);
+			piobuf += w;
+
+			extra = len & (sizeof(u32) - 1);
+			if (extra) {
+				u32 v = ((u32 *) ss->sge.vaddr)[w];
+
+				/* Clear unused upper bytes */
+				data = clear_upper_bytes(v, extra, 0);
+			}
+		}
+		update_sge(ss, len);
+		length -= len;
+	}
+	/* Update address before sending packet. */
+	update_sge(ss, length);
+	if (flush_wc) {
+		/* must flush early everything before trigger word */
+		qib_flush_wc();
+		__raw_writel(last, piobuf);
+		/* be sure trigger word is written */
+		qib_flush_wc();
+	} else
+		__raw_writel(last, piobuf);
+}
+
+static noinline struct qib_verbs_txreq *__get_txreq(struct qib_ibdev *dev,
+					   struct qib_qp *qp)
+{
+	struct qib_verbs_txreq *tx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	spin_lock(&dev->pending_lock);
+
+	if (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+
+		list_del(l);
+		spin_unlock(&dev->pending_lock);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+	} else {
+		if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK &&
+		    list_empty(&qp->iowait)) {
+			dev->n_txwait++;
+			qp->s_flags |= QIB_S_WAIT_TX;
+			list_add_tail(&qp->iowait, &dev->txwait);
+		}
+		qp->s_flags &= ~QIB_S_BUSY;
+		spin_unlock(&dev->pending_lock);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+		tx = ERR_PTR(-EBUSY);
+	}
+	return tx;
+}
+
+static inline struct qib_verbs_txreq *get_txreq(struct qib_ibdev *dev,
+					 struct qib_qp *qp)
+{
+	struct qib_verbs_txreq *tx;
+	unsigned long flags;
+
+	spin_lock_irqsave(&dev->pending_lock, flags);
+	/* assume the list non empty */
+	if (likely(!list_empty(&dev->txreq_free))) {
+		struct list_head *l = dev->txreq_free.next;
+
+		list_del(l);
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+	} else {
+		/* call slow path to get the extra lock */
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+		tx =  __get_txreq(dev, qp);
+	}
+	return tx;
+}
+
+void qib_put_txreq(struct qib_verbs_txreq *tx)
+{
+	struct qib_ibdev *dev;
+	struct qib_qp *qp;
+	unsigned long flags;
+
+	qp = tx->qp;
+	dev = to_idev(qp->ibqp.device);
+
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+	if (tx->mr) {
+		qib_put_mr(tx->mr);
+		tx->mr = NULL;
+	}
+	if (tx->txreq.flags & QIB_SDMA_TXREQ_F_FREEBUF) {
+		tx->txreq.flags &= ~QIB_SDMA_TXREQ_F_FREEBUF;
+		dma_unmap_single(&dd_from_dev(dev)->pcidev->dev,
+				 tx->txreq.addr, tx->hdr_dwords << 2,
+				 DMA_TO_DEVICE);
+		kfree(tx->align_buf);
+	}
+
+	spin_lock_irqsave(&dev->pending_lock, flags);
+
+	/* Put struct back on free list */
+	list_add(&tx->txreq.list, &dev->txreq_free);
+
+	if (!list_empty(&dev->txwait)) {
+		/* Wake up first QP wanting a free struct */
+		qp = list_entry(dev->txwait.next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_TX) {
+			qp->s_flags &= ~QIB_S_WAIT_TX;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	} else
+		spin_unlock_irqrestore(&dev->pending_lock, flags);
+}
+
+/*
+ * This is called when there are send DMA descriptors that might be
+ * available.
+ *
+ * This is called with ppd->sdma_lock held.
+ */
+void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail)
+{
+	struct qib_qp *qp, *nqp;
+	struct qib_qp *qps[20];
+	struct qib_ibdev *dev;
+	unsigned i, n;
+
+	n = 0;
+	dev = &ppd->dd->verbs_dev;
+	spin_lock(&dev->pending_lock);
+
+	/* Search wait list for first QP wanting DMA descriptors. */
+	list_for_each_entry_safe(qp, nqp, &dev->dmawait, iowait) {
+		if (qp->port_num != ppd->port)
+			continue;
+		if (n == ARRAY_SIZE(qps))
+			break;
+		if (qp->s_tx->txreq.sg_count > avail)
+			break;
+		avail -= qp->s_tx->txreq.sg_count;
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		qps[n++] = qp;
+	}
+
+	spin_unlock(&dev->pending_lock);
+
+	for (i = 0; i < n; i++) {
+		qp = qps[i];
+		spin_lock(&qp->s_lock);
+		if (qp->s_flags & QIB_S_WAIT_DMA_DESC) {
+			qp->s_flags &= ~QIB_S_WAIT_DMA_DESC;
+			qib_schedule_send(qp);
+		}
+		spin_unlock(&qp->s_lock);
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+/*
+ * This is called with ppd->sdma_lock held.
+ */
+static void sdma_complete(struct qib_sdma_txreq *cookie, int status)
+{
+	struct qib_verbs_txreq *tx =
+		container_of(cookie, struct qib_verbs_txreq, txreq);
+	struct qib_qp *qp = tx->qp;
+
+	spin_lock(&qp->s_lock);
+	if (tx->wqe)
+		qib_send_complete(qp, tx->wqe, IB_WC_SUCCESS);
+	else if (qp->ibqp.qp_type == IB_QPT_RC) {
+		struct qib_ib_header *hdr;
+
+		if (tx->txreq.flags & QIB_SDMA_TXREQ_F_FREEBUF)
+			hdr = &tx->align_buf->hdr;
+		else {
+			struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+
+			hdr = &dev->pio_hdrs[tx->hdr_inx].hdr;
+		}
+		qib_rc_send_complete(qp, hdr);
+	}
+	if (atomic_dec_and_test(&qp->s_dma_busy)) {
+		if (qp->state == IB_QPS_RESET)
+			wake_up(&qp->wait_dma);
+		else if (qp->s_flags & QIB_S_WAIT_DMA) {
+			qp->s_flags &= ~QIB_S_WAIT_DMA;
+			qib_schedule_send(qp);
+		}
+	}
+	spin_unlock(&qp->s_lock);
+
+	qib_put_txreq(tx);
+}
+
+static int wait_kmem(struct qib_ibdev *dev, struct qib_qp *qp)
+{
+	unsigned long flags;
+	int ret = 0;
+
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			if (list_empty(&dev->memwait))
+				mod_timer(&dev->mem_timer, jiffies + 1);
+			qp->s_flags |= QIB_S_WAIT_KMEM;
+			list_add_tail(&qp->iowait, &dev->memwait);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		ret = -EBUSY;
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+
+	return ret;
+}
+
+static int qib_verbs_send_dma(struct qib_qp *qp, struct qib_ib_header *hdr,
+			      u32 hdrwords, struct qib_sge_state *ss, u32 len,
+			      u32 plen, u32 dwords)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_devdata *dd = dd_from_dev(dev);
+	struct qib_ibport *ibp = to_iport(qp->ibqp.device, qp->port_num);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_verbs_txreq *tx;
+	struct qib_pio_header *phdr;
+	u32 control;
+	u32 ndesc;
+	int ret;
+
+	tx = qp->s_tx;
+	if (tx) {
+		qp->s_tx = NULL;
+		/* resend previously constructed packet */
+		ret = qib_sdma_verbs_send(ppd, tx->ss, tx->dwords, tx);
+		goto bail;
+	}
+
+	tx = get_txreq(dev, qp);
+	if (IS_ERR(tx))
+		goto bail_tx;
+
+	control = dd->f_setpbc_control(ppd, plen, qp->s_srate,
+				       be16_to_cpu(hdr->lrh[0]) >> 12);
+	tx->qp = qp;
+	atomic_inc(&qp->refcount);
+	tx->wqe = qp->s_wqe;
+	tx->mr = qp->s_rdma_mr;
+	if (qp->s_rdma_mr)
+		qp->s_rdma_mr = NULL;
+	tx->txreq.callback = sdma_complete;
+	if (dd->flags & QIB_HAS_SDMA_TIMEOUT)
+		tx->txreq.flags = QIB_SDMA_TXREQ_F_HEADTOHOST;
+	else
+		tx->txreq.flags = QIB_SDMA_TXREQ_F_INTREQ;
+	if (plen + 1 > dd->piosize2kmax_dwords)
+		tx->txreq.flags |= QIB_SDMA_TXREQ_F_USELARGEBUF;
+
+	if (len) {
+		/*
+		 * Don't try to DMA if it takes more descriptors than
+		 * the queue holds.
+		 */
+		ndesc = qib_count_sge(ss, len);
+		if (ndesc >= ppd->sdma_descq_cnt)
+			ndesc = 0;
+	} else
+		ndesc = 1;
+	if (ndesc) {
+		phdr = &dev->pio_hdrs[tx->hdr_inx];
+		phdr->pbc[0] = cpu_to_le32(plen);
+		phdr->pbc[1] = cpu_to_le32(control);
+		memcpy(&phdr->hdr, hdr, hdrwords << 2);
+		tx->txreq.flags |= QIB_SDMA_TXREQ_F_FREEDESC;
+		tx->txreq.sg_count = ndesc;
+		tx->txreq.addr = dev->pio_hdrs_phys +
+			tx->hdr_inx * sizeof(struct qib_pio_header);
+		tx->hdr_dwords = hdrwords + 2; /* add PBC length */
+		ret = qib_sdma_verbs_send(ppd, ss, dwords, tx);
+		goto bail;
+	}
+
+	/* Allocate a buffer and copy the header and payload to it. */
+	tx->hdr_dwords = plen + 1;
+	phdr = kmalloc(tx->hdr_dwords << 2, GFP_ATOMIC);
+	if (!phdr)
+		goto err_tx;
+	phdr->pbc[0] = cpu_to_le32(plen);
+	phdr->pbc[1] = cpu_to_le32(control);
+	memcpy(&phdr->hdr, hdr, hdrwords << 2);
+	qib_copy_from_sge((u32 *) &phdr->hdr + hdrwords, ss, len);
+
+	tx->txreq.addr = dma_map_single(&dd->pcidev->dev, phdr,
+					tx->hdr_dwords << 2, DMA_TO_DEVICE);
+	if (dma_mapping_error(&dd->pcidev->dev, tx->txreq.addr))
+		goto map_err;
+	tx->align_buf = phdr;
+	tx->txreq.flags |= QIB_SDMA_TXREQ_F_FREEBUF;
+	tx->txreq.sg_count = 1;
+	ret = qib_sdma_verbs_send(ppd, NULL, 0, tx);
+	goto unaligned;
+
+map_err:
+	kfree(phdr);
+err_tx:
+	qib_put_txreq(tx);
+	ret = wait_kmem(dev, qp);
+unaligned:
+	ibp->n_unaligned++;
+bail:
+	return ret;
+bail_tx:
+	ret = PTR_ERR(tx);
+	goto bail;
+}
+
+/*
+ * If we are now in the error state, return zero to flush the
+ * send work request.
+ */
+static int no_bufs_available(struct qib_qp *qp)
+{
+	struct qib_ibdev *dev = to_idev(qp->ibqp.device);
+	struct qib_devdata *dd;
+	unsigned long flags;
+	int ret = 0;
+
+	/*
+	 * Note that as soon as want_buffer() is called and
+	 * possibly before it returns, qib_ib_piobufavail()
+	 * could be called. Therefore, put QP on the I/O wait list before
+	 * enabling the PIO avail interrupt.
+	 */
+	spin_lock_irqsave(&qp->s_lock, flags);
+	if (ib_qib_state_ops[qp->state] & QIB_PROCESS_RECV_OK) {
+		spin_lock(&dev->pending_lock);
+		if (list_empty(&qp->iowait)) {
+			dev->n_piowait++;
+			qp->s_flags |= QIB_S_WAIT_PIO;
+			list_add_tail(&qp->iowait, &dev->piowait);
+			dd = dd_from_dev(dev);
+			dd->f_wantpiobuf_intr(dd, 1);
+		}
+		spin_unlock(&dev->pending_lock);
+		qp->s_flags &= ~QIB_S_BUSY;
+		ret = -EBUSY;
+	}
+	spin_unlock_irqrestore(&qp->s_lock, flags);
+	return ret;
+}
+
+static int qib_verbs_send_pio(struct qib_qp *qp, struct qib_ib_header *ibhdr,
+			      u32 hdrwords, struct qib_sge_state *ss, u32 len,
+			      u32 plen, u32 dwords)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	struct qib_pportdata *ppd = dd->pport + qp->port_num - 1;
+	u32 *hdr = (u32 *) ibhdr;
+	u32 __iomem *piobuf_orig;
+	u32 __iomem *piobuf;
+	u64 pbc;
+	unsigned long flags;
+	unsigned flush_wc;
+	u32 control;
+	u32 pbufn;
+
+	control = dd->f_setpbc_control(ppd, plen, qp->s_srate,
+		be16_to_cpu(ibhdr->lrh[0]) >> 12);
+	pbc = ((u64) control << 32) | plen;
+	piobuf = dd->f_getsendbuf(ppd, pbc, &pbufn);
+	if (unlikely(piobuf == NULL))
+		return no_bufs_available(qp);
+
+	/*
+	 * Write the pbc.
+	 * We have to flush after the PBC for correctness on some cpus
+	 * or WC buffer can be written out of order.
+	 */
+	writeq(pbc, piobuf);
+	piobuf_orig = piobuf;
+	piobuf += 2;
+
+	flush_wc = dd->flags & QIB_PIO_FLUSH_WC;
+	if (len == 0) {
+		/*
+		 * If there is just the header portion, must flush before
+		 * writing last word of header for correctness, and after
+		 * the last header word (trigger word).
+		 */
+		if (flush_wc) {
+			qib_flush_wc();
+			qib_pio_copy(piobuf, hdr, hdrwords - 1);
+			qib_flush_wc();
+			__raw_writel(hdr[hdrwords - 1], piobuf + hdrwords - 1);
+			qib_flush_wc();
+		} else
+			qib_pio_copy(piobuf, hdr, hdrwords);
+		goto done;
+	}
+
+	if (flush_wc)
+		qib_flush_wc();
+	qib_pio_copy(piobuf, hdr, hdrwords);
+	piobuf += hdrwords;
+
+	/* The common case is aligned and contained in one segment. */
+	if (likely(ss->num_sge == 1 && len <= ss->sge.length &&
+		   !((unsigned long)ss->sge.vaddr & (sizeof(u32) - 1)))) {
+		u32 *addr = (u32 *) ss->sge.vaddr;
+
+		/* Update address before sending packet. */
+		update_sge(ss, len);
+		if (flush_wc) {
+			qib_pio_copy(piobuf, addr, dwords - 1);
+			/* must flush early everything before trigger word */
+			qib_flush_wc();
+			__raw_writel(addr[dwords - 1], piobuf + dwords - 1);
+			/* be sure trigger word is written */
+			qib_flush_wc();
+		} else
+			qib_pio_copy(piobuf, addr, dwords);
+		goto done;
+	}
+	copy_io(piobuf, ss, len, flush_wc);
+done:
+	if (dd->flags & QIB_USE_SPCL_TRIG) {
+		u32 spcl_off = (pbufn >= dd->piobcnt2k) ? 2047 : 1023;
+		qib_flush_wc();
+		__raw_writel(0xaebecede, piobuf_orig + spcl_off);
+	}
+	qib_sendbuf_done(dd, pbufn);
+	if (qp->s_rdma_mr) {
+		qib_put_mr(qp->s_rdma_mr);
+		qp->s_rdma_mr = NULL;
+	}
+	if (qp->s_wqe) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qib_send_complete(qp, qp->s_wqe, IB_WC_SUCCESS);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	} else if (qp->ibqp.qp_type == IB_QPT_RC) {
+		spin_lock_irqsave(&qp->s_lock, flags);
+		qib_rc_send_complete(qp, ibhdr);
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+	}
+	return 0;
+}
+
+/**
+ * qib_verbs_send - send a packet
+ * @qp: the QP to send on
+ * @hdr: the packet header
+ * @hdrwords: the number of 32-bit words in the header
+ * @ss: the SGE to send
+ * @len: the length of the packet in bytes
+ *
+ * Return zero if packet is sent or queued OK.
+ * Return non-zero and clear qp->s_flags QIB_S_BUSY otherwise.
+ */
+int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
+		   u32 hdrwords, struct qib_sge_state *ss, u32 len)
+{
+	struct qib_devdata *dd = dd_from_ibdev(qp->ibqp.device);
+	u32 plen;
+	int ret;
+	u32 dwords = (len + 3) >> 2;
+
+	/*
+	 * Calculate the send buffer trigger address.
+	 * The +1 counts for the pbc control dword following the pbc length.
+	 */
+	plen = hdrwords + dwords + 1;
+
+	/*
+	 * VL15 packets (IB_QPT_SMI) will always use PIO, so we
+	 * can defer SDMA restart until link goes ACTIVE without
+	 * worrying about just how we got there.
+	 */
+	if (qp->ibqp.qp_type == IB_QPT_SMI ||
+	    !(dd->flags & QIB_HAS_SEND_DMA))
+		ret = qib_verbs_send_pio(qp, hdr, hdrwords, ss, len,
+					 plen, dwords);
+	else
+		ret = qib_verbs_send_dma(qp, hdr, hdrwords, ss, len,
+					 plen, dwords);
+
+	return ret;
+}
+
+int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
+			  u64 *rwords, u64 *spkts, u64 *rpkts,
+			  u64 *xmit_wait)
+{
+	int ret;
+	struct qib_devdata *dd = ppd->dd;
+
+	if (!(dd->flags & QIB_PRESENT)) {
+		/* no hardware, freeze, etc. */
+		ret = -EINVAL;
+		goto bail;
+	}
+	*swords = dd->f_portcntr(ppd, QIBPORTCNTR_WORDSEND);
+	*rwords = dd->f_portcntr(ppd, QIBPORTCNTR_WORDRCV);
+	*spkts = dd->f_portcntr(ppd, QIBPORTCNTR_PKTSEND);
+	*rpkts = dd->f_portcntr(ppd, QIBPORTCNTR_PKTRCV);
+	*xmit_wait = dd->f_portcntr(ppd, QIBPORTCNTR_SENDSTALL);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_get_counters - get various chip counters
+ * @dd: the qlogic_ib device
+ * @cntrs: counters are placed here
+ *
+ * Return the counters needed by recv_pma_get_portcounters().
+ */
+int qib_get_counters(struct qib_pportdata *ppd,
+		     struct qib_verbs_counters *cntrs)
+{
+	int ret;
+
+	if (!(ppd->dd->flags & QIB_PRESENT)) {
+		/* no hardware, freeze, etc. */
+		ret = -EINVAL;
+		goto bail;
+	}
+	cntrs->symbol_error_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBSYMBOLERR);
+	cntrs->link_error_recovery_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBLINKERRRECOV);
+	/*
+	 * The link downed counter counts when the other side downs the
+	 * connection.  We add in the number of times we downed the link
+	 * due to local link integrity errors to compensate.
+	 */
+	cntrs->link_downed_counter =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_IBLINKDOWN);
+	cntrs->port_rcv_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXDROPPKT) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RCVOVFL) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERR_RLEN) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_INVALIDRLEN) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRLINK) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRICRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRVCRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_ERRLPCRC) +
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_BADFORMAT);
+	cntrs->port_rcv_errors +=
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXLOCALPHYERR);
+	cntrs->port_rcv_errors +=
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RXVLERR);
+	cntrs->port_rcv_remphys_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_RCVEBP);
+	cntrs->port_xmit_discards =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_UNSUPVL);
+	cntrs->port_xmit_data = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_WORDSEND);
+	cntrs->port_rcv_data = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_WORDRCV);
+	cntrs->port_xmit_packets = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_PKTSEND);
+	cntrs->port_rcv_packets = ppd->dd->f_portcntr(ppd,
+			QIBPORTCNTR_PKTRCV);
+	cntrs->local_link_integrity_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_LLI);
+	cntrs->excessive_buffer_overrun_errors =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_EXCESSBUFOVFL);
+	cntrs->vl15_dropped =
+		ppd->dd->f_portcntr(ppd, QIBPORTCNTR_VL15PKTDROP);
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_ib_piobufavail - callback when a PIO buffer is available
+ * @dd: the device pointer
+ *
+ * This is called from qib_intr() at interrupt level when a PIO buffer is
+ * available after qib_verbs_send() returned an error that no buffers were
+ * available. Disable the interrupt if there are no more QPs waiting.
+ */
+void qib_ib_piobufavail(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct list_head *list;
+	struct qib_qp *qps[5];
+	struct qib_qp *qp;
+	unsigned long flags;
+	unsigned i, n;
+
+	list = &dev->piowait;
+	n = 0;
+
+	/*
+	 * Note: checking that the piowait list is empty and clearing
+	 * the buffer available interrupt needs to be atomic or we
+	 * could end up with QPs on the wait list with the interrupt
+	 * disabled.
+	 */
+	spin_lock_irqsave(&dev->pending_lock, flags);
+	while (!list_empty(list)) {
+		if (n == ARRAY_SIZE(qps))
+			goto full;
+		qp = list_entry(list->next, struct qib_qp, iowait);
+		list_del_init(&qp->iowait);
+		atomic_inc(&qp->refcount);
+		qps[n++] = qp;
+	}
+	dd->f_wantpiobuf_intr(dd, 0);
+full:
+	spin_unlock_irqrestore(&dev->pending_lock, flags);
+
+	for (i = 0; i < n; i++) {
+		qp = qps[i];
+
+		spin_lock_irqsave(&qp->s_lock, flags);
+		if (qp->s_flags & QIB_S_WAIT_PIO) {
+			qp->s_flags &= ~QIB_S_WAIT_PIO;
+			qib_schedule_send(qp);
+		}
+		spin_unlock_irqrestore(&qp->s_lock, flags);
+
+		/* Notify qib_destroy_qp() if it is waiting. */
+		if (atomic_dec_and_test(&qp->refcount))
+			wake_up(&qp->wait);
+	}
+}
+
+static int qib_query_device(struct ib_device *ibdev,
+			    struct ib_device_attr *props)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	struct qib_ibdev *dev = to_idev(ibdev);
+
+	memset(props, 0, sizeof(*props));
+
+	props->device_cap_flags = IB_DEVICE_BAD_PKEY_CNTR |
+		IB_DEVICE_BAD_QKEY_CNTR | IB_DEVICE_SHUTDOWN_PORT |
+		IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_RC_RNR_NAK_GEN |
+		IB_DEVICE_PORT_ACTIVE_EVENT | IB_DEVICE_SRQ_RESIZE;
+	props->page_size_cap = PAGE_SIZE;
+	props->vendor_id =
+		QIB_SRC_OUI_1 << 16 | QIB_SRC_OUI_2 << 8 | QIB_SRC_OUI_3;
+	props->vendor_part_id = dd->deviceid;
+	props->hw_ver = dd->minrev;
+	props->sys_image_guid = ib_qib_sys_image_guid;
+	props->max_mr_size = ~0ULL;
+	props->max_qp = ib_qib_max_qps;
+	props->max_qp_wr = ib_qib_max_qp_wrs;
+	props->max_sge = ib_qib_max_sges;
+	props->max_cq = ib_qib_max_cqs;
+	props->max_ah = ib_qib_max_ahs;
+	props->max_cqe = ib_qib_max_cqes;
+	props->max_mr = dev->lk_table.max;
+	props->max_fmr = dev->lk_table.max;
+	props->max_map_per_fmr = 32767;
+	props->max_pd = ib_qib_max_pds;
+	props->max_qp_rd_atom = QIB_MAX_RDMA_ATOMIC;
+	props->max_qp_init_rd_atom = 255;
+	/* props->max_res_rd_atom */
+	props->max_srq = ib_qib_max_srqs;
+	props->max_srq_wr = ib_qib_max_srq_wrs;
+	props->max_srq_sge = ib_qib_max_srq_sges;
+	/* props->local_ca_ack_delay */
+	props->atomic_cap = IB_ATOMIC_GLOB;
+	props->max_pkeys = qib_get_npkeys(dd);
+	props->max_mcast_grp = ib_qib_max_mcast_grps;
+	props->max_mcast_qp_attach = ib_qib_max_mcast_qp_attached;
+	props->max_total_mcast_qp_attach = props->max_mcast_qp_attach *
+		props->max_mcast_grp;
+
+	return 0;
+}
+
+static int qib_query_port(struct ib_device *ibdev, u8 port,
+			  struct ib_port_attr *props)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	enum ib_mtu mtu;
+	u16 lid = ppd->lid;
+
+	memset(props, 0, sizeof(*props));
+	props->lid = lid ? lid : be16_to_cpu(IB_LID_PERMISSIVE);
+	props->lmc = ppd->lmc;
+	props->sm_lid = ibp->sm_lid;
+	props->sm_sl = ibp->sm_sl;
+	props->state = dd->f_iblink_state(ppd->lastibcstat);
+	props->phys_state = dd->f_ibphys_portstate(ppd->lastibcstat);
+	props->port_cap_flags = ibp->port_cap_flags;
+	props->gid_tbl_len = QIB_GUIDS_PER_PORT;
+	props->max_msg_sz = 0x80000000;
+	props->pkey_tbl_len = qib_get_npkeys(dd);
+	props->bad_pkey_cntr = ibp->pkey_violations;
+	props->qkey_viol_cntr = ibp->qkey_violations;
+	props->active_width = ppd->link_width_active;
+	/* See rate_show() */
+	props->active_speed = ppd->link_speed_active;
+	props->max_vl_num = qib_num_vls(ppd->vls_supported);
+	props->init_type_reply = 0;
+
+	props->max_mtu = qib_ibmtu ? qib_ibmtu : IB_MTU_4096;
+	switch (ppd->ibmtu) {
+	case 4096:
+		mtu = IB_MTU_4096;
+		break;
+	case 2048:
+		mtu = IB_MTU_2048;
+		break;
+	case 1024:
+		mtu = IB_MTU_1024;
+		break;
+	case 512:
+		mtu = IB_MTU_512;
+		break;
+	case 256:
+		mtu = IB_MTU_256;
+		break;
+	default:
+		mtu = IB_MTU_2048;
+	}
+	props->active_mtu = mtu;
+	props->subnet_timeout = ibp->subnet_timeout;
+
+	return 0;
+}
+
+static int qib_modify_device(struct ib_device *device,
+			     int device_modify_mask,
+			     struct ib_device_modify *device_modify)
+{
+	struct qib_devdata *dd = dd_from_ibdev(device);
+	unsigned i;
+	int ret;
+
+	if (device_modify_mask & ~(IB_DEVICE_MODIFY_SYS_IMAGE_GUID |
+				   IB_DEVICE_MODIFY_NODE_DESC)) {
+		ret = -EOPNOTSUPP;
+		goto bail;
+	}
+
+	if (device_modify_mask & IB_DEVICE_MODIFY_NODE_DESC) {
+		memcpy(device->node_desc, device_modify->node_desc, 64);
+		for (i = 0; i < dd->num_pports; i++) {
+			struct qib_ibport *ibp = &dd->pport[i].ibport_data;
+
+			qib_node_desc_chg(ibp);
+		}
+	}
+
+	if (device_modify_mask & IB_DEVICE_MODIFY_SYS_IMAGE_GUID) {
+		ib_qib_sys_image_guid =
+			cpu_to_be64(device_modify->sys_image_guid);
+		for (i = 0; i < dd->num_pports; i++) {
+			struct qib_ibport *ibp = &dd->pport[i].ibport_data;
+
+			qib_sys_guid_chg(ibp);
+		}
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+static int qib_modify_port(struct ib_device *ibdev, u8 port,
+			   int port_modify_mask, struct ib_port_modify *props)
+{
+	struct qib_ibport *ibp = to_iport(ibdev, port);
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+	ibp->port_cap_flags |= props->set_port_cap_mask;
+	ibp->port_cap_flags &= ~props->clr_port_cap_mask;
+	if (props->set_port_cap_mask || props->clr_port_cap_mask)
+		qib_cap_mask_chg(ibp);
+	if (port_modify_mask & IB_PORT_SHUTDOWN)
+		qib_set_linkstate(ppd, QIB_IB_LINKDOWN);
+	if (port_modify_mask & IB_PORT_RESET_QKEY_CNTR)
+		ibp->qkey_violations = 0;
+	return 0;
+}
+
+static int qib_query_gid(struct ib_device *ibdev, u8 port,
+			 int index, union ib_gid *gid)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret = 0;
+
+	if (!port || port > dd->num_pports)
+		ret = -EINVAL;
+	else {
+		struct qib_ibport *ibp = to_iport(ibdev, port);
+		struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+		gid->global.subnet_prefix = ibp->gid_prefix;
+		if (index == 0)
+			gid->global.interface_id = ppd->guid;
+		else if (index < QIB_GUIDS_PER_PORT)
+			gid->global.interface_id = ibp->guids[index - 1];
+		else
+			ret = -EINVAL;
+	}
+
+	return ret;
+}
+
+static struct ib_pd *qib_alloc_pd(struct ib_device *ibdev,
+				  struct ib_ucontext *context,
+				  struct ib_udata *udata)
+{
+	struct qib_ibdev *dev = to_idev(ibdev);
+	struct qib_pd *pd;
+	struct ib_pd *ret;
+
+	/*
+	 * This is actually totally arbitrary.  Some correctness tests
+	 * assume there's a maximum number of PDs that can be allocated.
+	 * We don't actually have this limit, but we fail the test if
+	 * we allow allocations of more than we report for this value.
+	 */
+
+	pd = kmalloc(sizeof *pd, GFP_KERNEL);
+	if (!pd) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	spin_lock(&dev->n_pds_lock);
+	if (dev->n_pds_allocated == ib_qib_max_pds) {
+		spin_unlock(&dev->n_pds_lock);
+		kfree(pd);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	dev->n_pds_allocated++;
+	spin_unlock(&dev->n_pds_lock);
+
+	/* ib_alloc_pd() will initialize pd->ibpd. */
+	pd->user = udata != NULL;
+
+	ret = &pd->ibpd;
+
+bail:
+	return ret;
+}
+
+static int qib_dealloc_pd(struct ib_pd *ibpd)
+{
+	struct qib_pd *pd = to_ipd(ibpd);
+	struct qib_ibdev *dev = to_idev(ibpd->device);
+
+	spin_lock(&dev->n_pds_lock);
+	dev->n_pds_allocated--;
+	spin_unlock(&dev->n_pds_lock);
+
+	kfree(pd);
+
+	return 0;
+}
+
+int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr)
+{
+	/* A multicast address requires a GRH (see ch. 8.4.1). */
+	if (ah_attr->dlid >= QIB_MULTICAST_LID_BASE &&
+	    ah_attr->dlid != QIB_PERMISSIVE_LID &&
+	    !(ah_attr->ah_flags & IB_AH_GRH))
+		goto bail;
+	if ((ah_attr->ah_flags & IB_AH_GRH) &&
+	    ah_attr->grh.sgid_index >= QIB_GUIDS_PER_PORT)
+		goto bail;
+	if (ah_attr->dlid == 0)
+		goto bail;
+	if (ah_attr->port_num < 1 ||
+	    ah_attr->port_num > ibdev->phys_port_cnt)
+		goto bail;
+	if (ah_attr->static_rate != IB_RATE_PORT_CURRENT &&
+	    ib_rate_to_mult(ah_attr->static_rate) < 0)
+		goto bail;
+	if (ah_attr->sl > 15)
+		goto bail;
+	return 0;
+bail:
+	return -EINVAL;
+}
+
+/**
+ * qib_create_ah - create an address handle
+ * @pd: the protection domain
+ * @ah_attr: the attributes of the AH
+ *
+ * This may be called from interrupt context.
+ */
+static struct ib_ah *qib_create_ah(struct ib_pd *pd,
+				   struct ib_ah_attr *ah_attr)
+{
+	struct qib_ah *ah;
+	struct ib_ah *ret;
+	struct qib_ibdev *dev = to_idev(pd->device);
+	unsigned long flags;
+
+	if (qib_check_ah(pd->device, ah_attr)) {
+		ret = ERR_PTR(-EINVAL);
+		goto bail;
+	}
+
+	ah = kmalloc(sizeof *ah, GFP_ATOMIC);
+	if (!ah) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	spin_lock_irqsave(&dev->n_ahs_lock, flags);
+	if (dev->n_ahs_allocated == ib_qib_max_ahs) {
+		spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+		kfree(ah);
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	dev->n_ahs_allocated++;
+	spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+
+	/* ib_create_ah() will initialize ah->ibah. */
+	ah->attr = *ah_attr;
+	atomic_set(&ah->refcount, 0);
+
+	ret = &ah->ibah;
+
+bail:
+	return ret;
+}
+
+struct ib_ah *qib_create_qp0_ah(struct qib_ibport *ibp, u16 dlid)
+{
+	struct ib_ah_attr attr;
+	struct ib_ah *ah = ERR_PTR(-EINVAL);
+	struct qib_qp *qp0;
+
+	memset(&attr, 0, sizeof attr);
+	attr.dlid = dlid;
+	attr.port_num = ppd_from_ibp(ibp)->port;
+	rcu_read_lock();
+	qp0 = rcu_dereference(ibp->qp0);
+	if (qp0)
+		ah = ib_create_ah(qp0->ibqp.pd, &attr);
+	rcu_read_unlock();
+	return ah;
+}
+
+/**
+ * qib_destroy_ah - destroy an address handle
+ * @ibah: the AH to destroy
+ *
+ * This may be called from interrupt context.
+ */
+static int qib_destroy_ah(struct ib_ah *ibah)
+{
+	struct qib_ibdev *dev = to_idev(ibah->device);
+	struct qib_ah *ah = to_iah(ibah);
+	unsigned long flags;
+
+	if (atomic_read(&ah->refcount) != 0)
+		return -EBUSY;
+
+	spin_lock_irqsave(&dev->n_ahs_lock, flags);
+	dev->n_ahs_allocated--;
+	spin_unlock_irqrestore(&dev->n_ahs_lock, flags);
+
+	kfree(ah);
+
+	return 0;
+}
+
+static int qib_modify_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
+{
+	struct qib_ah *ah = to_iah(ibah);
+
+	if (qib_check_ah(ibah->device, ah_attr))
+		return -EINVAL;
+
+	ah->attr = *ah_attr;
+
+	return 0;
+}
+
+static int qib_query_ah(struct ib_ah *ibah, struct ib_ah_attr *ah_attr)
+{
+	struct qib_ah *ah = to_iah(ibah);
+
+	*ah_attr = ah->attr;
+
+	return 0;
+}
+
+/**
+ * qib_get_npkeys - return the size of the PKEY table for context 0
+ * @dd: the qlogic_ib device
+ */
+unsigned qib_get_npkeys(struct qib_devdata *dd)
+{
+	return ARRAY_SIZE(dd->rcd[0]->pkeys);
+}
+
+/*
+ * Return the indexed PKEY from the port PKEY table.
+ * No need to validate rcd[ctxt]; the port is setup if we are here.
+ */
+unsigned qib_get_pkey(struct qib_ibport *ibp, unsigned index)
+{
+	struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+	struct qib_devdata *dd = ppd->dd;
+	unsigned ctxt = ppd->hw_pidx;
+	unsigned ret;
+
+	/* dd->rcd null if mini_init or some init failures */
+	if (!dd->rcd || index >= ARRAY_SIZE(dd->rcd[ctxt]->pkeys))
+		ret = 0;
+	else
+		ret = dd->rcd[ctxt]->pkeys[index];
+
+	return ret;
+}
+
+static int qib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+			  u16 *pkey)
+{
+	struct qib_devdata *dd = dd_from_ibdev(ibdev);
+	int ret;
+
+	if (index >= qib_get_npkeys(dd)) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	*pkey = qib_get_pkey(to_iport(ibdev, port), index);
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+/**
+ * qib_alloc_ucontext - allocate a ucontest
+ * @ibdev: the infiniband device
+ * @udata: not used by the QLogic_IB driver
+ */
+
+static struct ib_ucontext *qib_alloc_ucontext(struct ib_device *ibdev,
+					      struct ib_udata *udata)
+{
+	struct qib_ucontext *context;
+	struct ib_ucontext *ret;
+
+	context = kmalloc(sizeof *context, GFP_KERNEL);
+	if (!context) {
+		ret = ERR_PTR(-ENOMEM);
+		goto bail;
+	}
+
+	ret = &context->ibucontext;
+
+bail:
+	return ret;
+}
+
+static int qib_dealloc_ucontext(struct ib_ucontext *context)
+{
+	kfree(to_iucontext(context));
+	return 0;
+}
+
+static void init_ibport(struct qib_pportdata *ppd)
+{
+	struct qib_verbs_counters cntrs;
+	struct qib_ibport *ibp = &ppd->ibport_data;
+
+	spin_lock_init(&ibp->lock);
+	/* Set the prefix to the default value (see ch. 4.1.1) */
+	ibp->gid_prefix = IB_DEFAULT_GID_PREFIX;
+	ibp->sm_lid = be16_to_cpu(IB_LID_PERMISSIVE);
+	ibp->port_cap_flags = IB_PORT_SYS_IMAGE_GUID_SUP |
+		IB_PORT_CLIENT_REG_SUP | IB_PORT_SL_MAP_SUP |
+		IB_PORT_TRAP_SUP | IB_PORT_AUTO_MIGR_SUP |
+		IB_PORT_DR_NOTICE_SUP | IB_PORT_CAP_MASK_NOTICE_SUP |
+		IB_PORT_OTHER_LOCAL_CHANGES_SUP;
+	if (ppd->dd->flags & QIB_HAS_LINK_LATENCY)
+		ibp->port_cap_flags |= IB_PORT_LINK_LATENCY_SUP;
+	ibp->pma_counter_select[0] = IB_PMA_PORT_XMIT_DATA;
+	ibp->pma_counter_select[1] = IB_PMA_PORT_RCV_DATA;
+	ibp->pma_counter_select[2] = IB_PMA_PORT_XMIT_PKTS;
+	ibp->pma_counter_select[3] = IB_PMA_PORT_RCV_PKTS;
+	ibp->pma_counter_select[4] = IB_PMA_PORT_XMIT_WAIT;
+
+	/* Snapshot current HW counters to "clear" them. */
+	qib_get_counters(ppd, &cntrs);
+	ibp->z_symbol_error_counter = cntrs.symbol_error_counter;
+	ibp->z_link_error_recovery_counter =
+		cntrs.link_error_recovery_counter;
+	ibp->z_link_downed_counter = cntrs.link_downed_counter;
+	ibp->z_port_rcv_errors = cntrs.port_rcv_errors;
+	ibp->z_port_rcv_remphys_errors = cntrs.port_rcv_remphys_errors;
+	ibp->z_port_xmit_discards = cntrs.port_xmit_discards;
+	ibp->z_port_xmit_data = cntrs.port_xmit_data;
+	ibp->z_port_rcv_data = cntrs.port_rcv_data;
+	ibp->z_port_xmit_packets = cntrs.port_xmit_packets;
+	ibp->z_port_rcv_packets = cntrs.port_rcv_packets;
+	ibp->z_local_link_integrity_errors =
+		cntrs.local_link_integrity_errors;
+	ibp->z_excessive_buffer_overrun_errors =
+		cntrs.excessive_buffer_overrun_errors;
+	ibp->z_vl15_dropped = cntrs.vl15_dropped;
+	RCU_INIT_POINTER(ibp->qp0, NULL);
+	RCU_INIT_POINTER(ibp->qp1, NULL);
+}
+
+/**
+ * qib_register_ib_device - register our device with the infiniband core
+ * @dd: the device data structure
+ * Return the allocated qib_ibdev pointer or NULL on error.
+ */
+int qib_register_ib_device(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct ib_device *ibdev = &dev->ibdev;
+	struct qib_pportdata *ppd = dd->pport;
+	unsigned i, lk_tab_size;
+	int ret;
+
+	dev->qp_table_size = ib_qib_qp_table_size;
+	get_random_bytes(&dev->qp_rnd, sizeof(dev->qp_rnd));
+	dev->qp_table = kmalloc(dev->qp_table_size * sizeof *dev->qp_table,
+				GFP_KERNEL);
+	if (!dev->qp_table) {
+		ret = -ENOMEM;
+		goto err_qpt;
+	}
+	for (i = 0; i < dev->qp_table_size; i++)
+		RCU_INIT_POINTER(dev->qp_table[i], NULL);
+
+	for (i = 0; i < dd->num_pports; i++)
+		init_ibport(ppd + i);
+
+	/* Only need to initialize non-zero fields. */
+	spin_lock_init(&dev->qpt_lock);
+	spin_lock_init(&dev->n_pds_lock);
+	spin_lock_init(&dev->n_ahs_lock);
+	spin_lock_init(&dev->n_cqs_lock);
+	spin_lock_init(&dev->n_qps_lock);
+	spin_lock_init(&dev->n_srqs_lock);
+	spin_lock_init(&dev->n_mcast_grps_lock);
+	init_timer(&dev->mem_timer);
+	dev->mem_timer.function = mem_timer;
+	dev->mem_timer.data = (unsigned long) dev;
+
+	qib_init_qpn_table(dd, &dev->qpn_table);
+
+	/*
+	 * The top ib_qib_lkey_table_size bits are used to index the
+	 * table.  The lower 8 bits can be owned by the user (copied from
+	 * the LKEY).  The remaining bits act as a generation number or tag.
+	 */
+	spin_lock_init(&dev->lk_table.lock);
+	dev->lk_table.max = 1 << ib_qib_lkey_table_size;
+	lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table);
+	dev->lk_table.table = (struct qib_mregion __rcu **)
+		__get_free_pages(GFP_KERNEL, get_order(lk_tab_size));
+	if (dev->lk_table.table == NULL) {
+		ret = -ENOMEM;
+		goto err_lk;
+	}
+	RCU_INIT_POINTER(dev->dma_mr, NULL);
+	for (i = 0; i < dev->lk_table.max; i++)
+		RCU_INIT_POINTER(dev->lk_table.table[i], NULL);
+	INIT_LIST_HEAD(&dev->pending_mmaps);
+	spin_lock_init(&dev->pending_lock);
+	dev->mmap_offset = PAGE_SIZE;
+	spin_lock_init(&dev->mmap_offset_lock);
+	INIT_LIST_HEAD(&dev->piowait);
+	INIT_LIST_HEAD(&dev->dmawait);
+	INIT_LIST_HEAD(&dev->txwait);
+	INIT_LIST_HEAD(&dev->memwait);
+	INIT_LIST_HEAD(&dev->txreq_free);
+
+	if (ppd->sdma_descq_cnt) {
+		dev->pio_hdrs = dma_alloc_coherent(&dd->pcidev->dev,
+						ppd->sdma_descq_cnt *
+						sizeof(struct qib_pio_header),
+						&dev->pio_hdrs_phys,
+						GFP_KERNEL);
+		if (!dev->pio_hdrs) {
+			ret = -ENOMEM;
+			goto err_hdrs;
+		}
+	}
+
+	for (i = 0; i < ppd->sdma_descq_cnt; i++) {
+		struct qib_verbs_txreq *tx;
+
+		tx = kzalloc(sizeof *tx, GFP_KERNEL);
+		if (!tx) {
+			ret = -ENOMEM;
+			goto err_tx;
+		}
+		tx->hdr_inx = i;
+		list_add(&tx->txreq.list, &dev->txreq_free);
+	}
+
+	/*
+	 * The system image GUID is supposed to be the same for all
+	 * IB HCAs in a single system but since there can be other
+	 * device types in the system, we can't be sure this is unique.
+	 */
+	if (!ib_qib_sys_image_guid)
+		ib_qib_sys_image_guid = ppd->guid;
+
+	strlcpy(ibdev->name, "qib%d", IB_DEVICE_NAME_MAX);
+	ibdev->owner = THIS_MODULE;
+	ibdev->node_guid = ppd->guid;
+	ibdev->uverbs_abi_ver = QIB_UVERBS_ABI_VERSION;
+	ibdev->uverbs_cmd_mask =
+		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT)         |
+		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE)        |
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT)          |
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD)            |
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD)          |
+		(1ull << IB_USER_VERBS_CMD_CREATE_AH)           |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_AH)           |
+		(1ull << IB_USER_VERBS_CMD_QUERY_AH)            |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_AH)          |
+		(1ull << IB_USER_VERBS_CMD_REG_MR)              |
+		(1ull << IB_USER_VERBS_CMD_DEREG_MR)            |
+		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ)           |
+		(1ull << IB_USER_VERBS_CMD_RESIZE_CQ)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ)          |
+		(1ull << IB_USER_VERBS_CMD_POLL_CQ)             |
+		(1ull << IB_USER_VERBS_CMD_REQ_NOTIFY_CQ)       |
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP)           |
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP)            |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_QP)          |
+		(1ull << IB_USER_VERBS_CMD_POST_SEND)           |
+		(1ull << IB_USER_VERBS_CMD_POST_RECV)           |
+		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST)        |
+		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST)        |
+		(1ull << IB_USER_VERBS_CMD_CREATE_SRQ)          |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_SRQ)          |
+		(1ull << IB_USER_VERBS_CMD_QUERY_SRQ)           |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_SRQ)         |
+		(1ull << IB_USER_VERBS_CMD_POST_SRQ_RECV);
+	ibdev->node_type = RDMA_NODE_IB_CA;
+	ibdev->phys_port_cnt = dd->num_pports;
+	ibdev->num_comp_vectors = 1;
+	ibdev->dma_device = &dd->pcidev->dev;
+	ibdev->query_device = qib_query_device;
+	ibdev->modify_device = qib_modify_device;
+	ibdev->query_port = qib_query_port;
+	ibdev->modify_port = qib_modify_port;
+	ibdev->query_pkey = qib_query_pkey;
+	ibdev->query_gid = qib_query_gid;
+	ibdev->alloc_ucontext = qib_alloc_ucontext;
+	ibdev->dealloc_ucontext = qib_dealloc_ucontext;
+	ibdev->alloc_pd = qib_alloc_pd;
+	ibdev->dealloc_pd = qib_dealloc_pd;
+	ibdev->create_ah = qib_create_ah;
+	ibdev->destroy_ah = qib_destroy_ah;
+	ibdev->modify_ah = qib_modify_ah;
+	ibdev->query_ah = qib_query_ah;
+	ibdev->create_srq = qib_create_srq;
+	ibdev->modify_srq = qib_modify_srq;
+	ibdev->query_srq = qib_query_srq;
+	ibdev->destroy_srq = qib_destroy_srq;
+	ibdev->create_qp = qib_create_qp;
+	ibdev->modify_qp = qib_modify_qp;
+	ibdev->query_qp = qib_query_qp;
+	ibdev->destroy_qp = qib_destroy_qp;
+	ibdev->post_send = qib_post_send;
+	ibdev->post_recv = qib_post_receive;
+	ibdev->post_srq_recv = qib_post_srq_receive;
+	ibdev->create_cq = qib_create_cq;
+	ibdev->destroy_cq = qib_destroy_cq;
+	ibdev->resize_cq = qib_resize_cq;
+	ibdev->poll_cq = qib_poll_cq;
+	ibdev->req_notify_cq = qib_req_notify_cq;
+	ibdev->get_dma_mr = qib_get_dma_mr;
+	ibdev->reg_phys_mr = qib_reg_phys_mr;
+	ibdev->reg_user_mr = qib_reg_user_mr;
+	ibdev->dereg_mr = qib_dereg_mr;
+	ibdev->alloc_fast_reg_mr = qib_alloc_fast_reg_mr;
+	ibdev->alloc_fast_reg_page_list = qib_alloc_fast_reg_page_list;
+	ibdev->free_fast_reg_page_list = qib_free_fast_reg_page_list;
+	ibdev->alloc_fmr = qib_alloc_fmr;
+	ibdev->map_phys_fmr = qib_map_phys_fmr;
+	ibdev->unmap_fmr = qib_unmap_fmr;
+	ibdev->dealloc_fmr = qib_dealloc_fmr;
+	ibdev->attach_mcast = qib_multicast_attach;
+	ibdev->detach_mcast = qib_multicast_detach;
+	ibdev->process_mad = qib_process_mad;
+	ibdev->mmap = qib_mmap;
+	ibdev->dma_ops = &qib_dma_mapping_ops;
+
+	snprintf(ibdev->node_desc, sizeof(ibdev->node_desc),
+		 "Intel Infiniband HCA %s", init_utsname()->nodename);
+
+	ret = ib_register_device(ibdev, qib_create_port_files);
+	if (ret)
+		goto err_reg;
+
+	ret = qib_create_agents(dev);
+	if (ret)
+		goto err_agents;
+
+	ret = qib_verbs_register_sysfs(dd);
+	if (ret)
+		goto err_class;
+
+	goto bail;
+
+err_class:
+	qib_free_agents(dev);
+err_agents:
+	ib_unregister_device(ibdev);
+err_reg:
+err_tx:
+	while (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+		struct qib_verbs_txreq *tx;
+
+		list_del(l);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+		kfree(tx);
+	}
+	if (ppd->sdma_descq_cnt)
+		dma_free_coherent(&dd->pcidev->dev,
+				  ppd->sdma_descq_cnt *
+					sizeof(struct qib_pio_header),
+				  dev->pio_hdrs, dev->pio_hdrs_phys);
+err_hdrs:
+	free_pages((unsigned long) dev->lk_table.table, get_order(lk_tab_size));
+err_lk:
+	kfree(dev->qp_table);
+err_qpt:
+	qib_dev_err(dd, "cannot register verbs: %d!\n", -ret);
+bail:
+	return ret;
+}
+
+void qib_unregister_ib_device(struct qib_devdata *dd)
+{
+	struct qib_ibdev *dev = &dd->verbs_dev;
+	struct ib_device *ibdev = &dev->ibdev;
+	u32 qps_inuse;
+	unsigned lk_tab_size;
+
+	qib_verbs_unregister_sysfs(dd);
+
+	qib_free_agents(dev);
+
+	ib_unregister_device(ibdev);
+
+	if (!list_empty(&dev->piowait))
+		qib_dev_err(dd, "piowait list not empty!\n");
+	if (!list_empty(&dev->dmawait))
+		qib_dev_err(dd, "dmawait list not empty!\n");
+	if (!list_empty(&dev->txwait))
+		qib_dev_err(dd, "txwait list not empty!\n");
+	if (!list_empty(&dev->memwait))
+		qib_dev_err(dd, "memwait list not empty!\n");
+	if (dev->dma_mr)
+		qib_dev_err(dd, "DMA MR not NULL!\n");
+
+	qps_inuse = qib_free_all_qps(dd);
+	if (qps_inuse)
+		qib_dev_err(dd, "QP memory leak! %u still in use\n",
+			    qps_inuse);
+
+	del_timer_sync(&dev->mem_timer);
+	qib_free_qpn_table(&dev->qpn_table);
+	while (!list_empty(&dev->txreq_free)) {
+		struct list_head *l = dev->txreq_free.next;
+		struct qib_verbs_txreq *tx;
+
+		list_del(l);
+		tx = list_entry(l, struct qib_verbs_txreq, txreq.list);
+		kfree(tx);
+	}
+	if (dd->pport->sdma_descq_cnt)
+		dma_free_coherent(&dd->pcidev->dev,
+				  dd->pport->sdma_descq_cnt *
+					sizeof(struct qib_pio_header),
+				  dev->pio_hdrs, dev->pio_hdrs_phys);
+	lk_tab_size = dev->lk_table.max * sizeof(*dev->lk_table.table);
+	free_pages((unsigned long) dev->lk_table.table,
+		   get_order(lk_tab_size));
+	kfree(dev->qp_table);
+}
+
+/*
+ * This must be called with s_lock held.
+ */
+void qib_schedule_send(struct qib_qp *qp)
+{
+	if (qib_send_ok(qp)) {
+		struct qib_ibport *ibp =
+			to_iport(qp->ibqp.device, qp->port_num);
+		struct qib_pportdata *ppd = ppd_from_ibp(ibp);
+
+		queue_work(ppd->qib_wq, &qp->s_work);
+	}
+}
diff --git a/drivers/infiniband/hw/qib/qib_verbs.h b/drivers/infiniband/hw/qib/qib_verbs.h
new file mode 100644
index 00000000000..bfc8948fdd3
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs.h
@@ -0,0 +1,1173 @@
+/*
+ * Copyright (c) 2012, 2013 Intel Corporation.  All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#ifndef QIB_VERBS_H
+#define QIB_VERBS_H
+
+#include <linux/types.h>
+#include <linux/spinlock.h>
+#include <linux/kernel.h>
+#include <linux/interrupt.h>
+#include <linux/kref.h>
+#include <linux/workqueue.h>
+#include <linux/kthread.h>
+#include <linux/completion.h>
+#include <rdma/ib_pack.h>
+#include <rdma/ib_user_verbs.h>
+
+struct qib_ctxtdata;
+struct qib_pportdata;
+struct qib_devdata;
+struct qib_verbs_txreq;
+
+#define QIB_MAX_RDMA_ATOMIC     16
+#define QIB_GUIDS_PER_PORT	5
+
+#define QPN_MAX                 (1 << 24)
+#define QPNMAP_ENTRIES          (QPN_MAX / PAGE_SIZE / BITS_PER_BYTE)
+
+/*
+ * Increment this value if any changes that break userspace ABI
+ * compatibility are made.
+ */
+#define QIB_UVERBS_ABI_VERSION       2
+
+/*
+ * Define an ib_cq_notify value that is not valid so we know when CQ
+ * notifications are armed.
+ */
+#define IB_CQ_NONE      (IB_CQ_NEXT_COMP + 1)
+
+#define IB_SEQ_NAK	(3 << 29)
+
+/* AETH NAK opcode values */
+#define IB_RNR_NAK                      0x20
+#define IB_NAK_PSN_ERROR                0x60
+#define IB_NAK_INVALID_REQUEST          0x61
+#define IB_NAK_REMOTE_ACCESS_ERROR      0x62
+#define IB_NAK_REMOTE_OPERATIONAL_ERROR 0x63
+#define IB_NAK_INVALID_RD_REQUEST       0x64
+
+/* Flags for checking QP state (see ib_qib_state_ops[]) */
+#define QIB_POST_SEND_OK                0x01
+#define QIB_POST_RECV_OK                0x02
+#define QIB_PROCESS_RECV_OK             0x04
+#define QIB_PROCESS_SEND_OK             0x08
+#define QIB_PROCESS_NEXT_SEND_OK        0x10
+#define QIB_FLUSH_SEND			0x20
+#define QIB_FLUSH_RECV			0x40
+#define QIB_PROCESS_OR_FLUSH_SEND \
+	(QIB_PROCESS_SEND_OK | QIB_FLUSH_SEND)
+
+/* IB Performance Manager status values */
+#define IB_PMA_SAMPLE_STATUS_DONE       0x00
+#define IB_PMA_SAMPLE_STATUS_STARTED    0x01
+#define IB_PMA_SAMPLE_STATUS_RUNNING    0x02
+
+/* Mandatory IB performance counter select values. */
+#define IB_PMA_PORT_XMIT_DATA   cpu_to_be16(0x0001)
+#define IB_PMA_PORT_RCV_DATA    cpu_to_be16(0x0002)
+#define IB_PMA_PORT_XMIT_PKTS   cpu_to_be16(0x0003)
+#define IB_PMA_PORT_RCV_PKTS    cpu_to_be16(0x0004)
+#define IB_PMA_PORT_XMIT_WAIT   cpu_to_be16(0x0005)
+
+#define QIB_VENDOR_IPG		cpu_to_be16(0xFFA0)
+
+#define IB_BTH_REQ_ACK		(1 << 31)
+#define IB_BTH_SOLICITED	(1 << 23)
+#define IB_BTH_MIG_REQ		(1 << 22)
+
+/* XXX Should be defined in ib_verbs.h enum ib_port_cap_flags */
+#define IB_PORT_OTHER_LOCAL_CHANGES_SUP (1 << 26)
+
+#define IB_GRH_VERSION		6
+#define IB_GRH_VERSION_MASK	0xF
+#define IB_GRH_VERSION_SHIFT	28
+#define IB_GRH_TCLASS_MASK	0xFF
+#define IB_GRH_TCLASS_SHIFT	20
+#define IB_GRH_FLOW_MASK	0xFFFFF
+#define IB_GRH_FLOW_SHIFT	0
+#define IB_GRH_NEXT_HDR		0x1B
+
+#define IB_DEFAULT_GID_PREFIX	cpu_to_be64(0xfe80000000000000ULL)
+
+/* Values for set/get portinfo VLCap OperationalVLs */
+#define IB_VL_VL0       1
+#define IB_VL_VL0_1     2
+#define IB_VL_VL0_3     3
+#define IB_VL_VL0_7     4
+#define IB_VL_VL0_14    5
+
+static inline int qib_num_vls(int vls)
+{
+	switch (vls) {
+	default:
+	case IB_VL_VL0:
+		return 1;
+	case IB_VL_VL0_1:
+		return 2;
+	case IB_VL_VL0_3:
+		return 4;
+	case IB_VL_VL0_7:
+		return 8;
+	case IB_VL_VL0_14:
+		return 15;
+	}
+}
+
+struct ib_reth {
+	__be64 vaddr;
+	__be32 rkey;
+	__be32 length;
+} __packed;
+
+struct ib_atomic_eth {
+	__be32 vaddr[2];        /* unaligned so access as 2 32-bit words */
+	__be32 rkey;
+	__be64 swap_data;
+	__be64 compare_data;
+} __packed;
+
+struct qib_other_headers {
+	__be32 bth[3];
+	union {
+		struct {
+			__be32 deth[2];
+			__be32 imm_data;
+		} ud;
+		struct {
+			struct ib_reth reth;
+			__be32 imm_data;
+		} rc;
+		struct {
+			__be32 aeth;
+			__be32 atomic_ack_eth[2];
+		} at;
+		__be32 imm_data;
+		__be32 aeth;
+		struct ib_atomic_eth atomic_eth;
+	} u;
+} __packed;
+
+/*
+ * Note that UD packets with a GRH header are 8+40+12+8 = 68 bytes
+ * long (72 w/ imm_data).  Only the first 56 bytes of the IB header
+ * will be in the eager header buffer.  The remaining 12 or 16 bytes
+ * are in the data buffer.
+ */
+struct qib_ib_header {
+	__be16 lrh[4];
+	union {
+		struct {
+			struct ib_grh grh;
+			struct qib_other_headers oth;
+		} l;
+		struct qib_other_headers oth;
+	} u;
+} __packed;
+
+struct qib_pio_header {
+	__le32 pbc[2];
+	struct qib_ib_header hdr;
+} __packed;
+
+/*
+ * There is one struct qib_mcast for each multicast GID.
+ * All attached QPs are then stored as a list of
+ * struct qib_mcast_qp.
+ */
+struct qib_mcast_qp {
+	struct list_head list;
+	struct qib_qp *qp;
+};
+
+struct qib_mcast {
+	struct rb_node rb_node;
+	union ib_gid mgid;
+	struct list_head qp_list;
+	wait_queue_head_t wait;
+	atomic_t refcount;
+	int n_attached;
+};
+
+/* Protection domain */
+struct qib_pd {
+	struct ib_pd ibpd;
+	int user;               /* non-zero if created from user space */
+};
+
+/* Address Handle */
+struct qib_ah {
+	struct ib_ah ibah;
+	struct ib_ah_attr attr;
+	atomic_t refcount;
+};
+
+/*
+ * This structure is used by qib_mmap() to validate an offset
+ * when an mmap() request is made.  The vm_area_struct then uses
+ * this as its vm_private_data.
+ */
+struct qib_mmap_info {
+	struct list_head pending_mmaps;
+	struct ib_ucontext *context;
+	void *obj;
+	__u64 offset;
+	struct kref ref;
+	unsigned size;
+};
+
+/*
+ * This structure is used to contain the head pointer, tail pointer,
+ * and completion queue entries as a single memory allocation so
+ * it can be mmap'ed into user space.
+ */
+struct qib_cq_wc {
+	u32 head;               /* index of next entry to fill */
+	u32 tail;               /* index of next ib_poll_cq() entry */
+	union {
+		/* these are actually size ibcq.cqe + 1 */
+		struct ib_uverbs_wc uqueue[0];
+		struct ib_wc kqueue[0];
+	};
+};
+
+/*
+ * The completion queue structure.
+ */
+struct qib_cq {
+	struct ib_cq ibcq;
+	struct kthread_work comptask;
+	struct qib_devdata *dd;
+	spinlock_t lock; /* protect changes in this struct */
+	u8 notify;
+	u8 triggered;
+	struct qib_cq_wc *queue;
+	struct qib_mmap_info *ip;
+};
+
+/*
+ * A segment is a linear region of low physical memory.
+ * XXX Maybe we should use phys addr here and kmap()/kunmap().
+ * Used by the verbs layer.
+ */
+struct qib_seg {
+	void *vaddr;
+	size_t length;
+};
+
+/* The number of qib_segs that fit in a page. */
+#define QIB_SEGSZ     (PAGE_SIZE / sizeof(struct qib_seg))
+
+struct qib_segarray {
+	struct qib_seg segs[QIB_SEGSZ];
+};
+
+struct qib_mregion {
+	struct ib_pd *pd;       /* shares refcnt of ibmr.pd */
+	u64 user_base;          /* User's address for this region */
+	u64 iova;               /* IB start address of this region */
+	size_t length;
+	u32 lkey;
+	u32 offset;             /* offset (bytes) to start of region */
+	int access_flags;
+	u32 max_segs;           /* number of qib_segs in all the arrays */
+	u32 mapsz;              /* size of the map array */
+	u8  page_shift;         /* 0 - non unform/non powerof2 sizes */
+	u8  lkey_published;     /* in global table */
+	struct completion comp; /* complete when refcount goes to zero */
+	struct rcu_head list;
+	atomic_t refcount;
+	struct qib_segarray *map[0];    /* the segments */
+};
+
+/*
+ * These keep track of the copy progress within a memory region.
+ * Used by the verbs layer.
+ */
+struct qib_sge {
+	struct qib_mregion *mr;
+	void *vaddr;            /* kernel virtual address of segment */
+	u32 sge_length;         /* length of the SGE */
+	u32 length;             /* remaining length of the segment */
+	u16 m;                  /* current index: mr->map[m] */
+	u16 n;                  /* current index: mr->map[m]->segs[n] */
+};
+
+/* Memory region */
+struct qib_mr {
+	struct ib_mr ibmr;
+	struct ib_umem *umem;
+	struct qib_mregion mr;  /* must be last */
+};
+
+/*
+ * Send work request queue entry.
+ * The size of the sg_list is determined when the QP is created and stored
+ * in qp->s_max_sge.
+ */
+struct qib_swqe {
+	struct ib_send_wr wr;   /* don't use wr.sg_list */
+	u32 psn;                /* first packet sequence number */
+	u32 lpsn;               /* last packet sequence number */
+	u32 ssn;                /* send sequence number */
+	u32 length;             /* total length of data in sg_list */
+	struct qib_sge sg_list[0];
+};
+
+/*
+ * Receive work request queue entry.
+ * The size of the sg_list is determined when the QP (or SRQ) is created
+ * and stored in qp->r_rq.max_sge (or srq->rq.max_sge).
+ */
+struct qib_rwqe {
+	u64 wr_id;
+	u8 num_sge;
+	struct ib_sge sg_list[0];
+};
+
+/*
+ * This structure is used to contain the head pointer, tail pointer,
+ * and receive work queue entries as a single memory allocation so
+ * it can be mmap'ed into user space.
+ * Note that the wq array elements are variable size so you can't
+ * just index into the array to get the N'th element;
+ * use get_rwqe_ptr() instead.
+ */
+struct qib_rwq {
+	u32 head;               /* new work requests posted to the head */
+	u32 tail;               /* receives pull requests from here. */
+	struct qib_rwqe wq[0];
+};
+
+struct qib_rq {
+	struct qib_rwq *wq;
+	u32 size;               /* size of RWQE array */
+	u8 max_sge;
+	spinlock_t lock /* protect changes in this struct */
+		____cacheline_aligned_in_smp;
+};
+
+struct qib_srq {
+	struct ib_srq ibsrq;
+	struct qib_rq rq;
+	struct qib_mmap_info *ip;
+	/* send signal when number of RWQEs < limit */
+	u32 limit;
+};
+
+struct qib_sge_state {
+	struct qib_sge *sg_list;      /* next SGE to be used if any */
+	struct qib_sge sge;   /* progress state for the current SGE */
+	u32 total_len;
+	u8 num_sge;
+};
+
+/*
+ * This structure holds the information that the send tasklet needs
+ * to send a RDMA read response or atomic operation.
+ */
+struct qib_ack_entry {
+	u8 opcode;
+	u8 sent;
+	u32 psn;
+	u32 lpsn;
+	union {
+		struct qib_sge rdma_sge;
+		u64 atomic_data;
+	};
+};
+
+/*
+ * Variables prefixed with s_ are for the requester (sender).
+ * Variables prefixed with r_ are for the responder (receiver).
+ * Variables prefixed with ack_ are for responder replies.
+ *
+ * Common variables are protected by both r_rq.lock and s_lock in that order
+ * which only happens in modify_qp() or changing the QP 'state'.
+ */
+struct qib_qp {
+	struct ib_qp ibqp;
+	/* read mostly fields above and below */
+	struct ib_ah_attr remote_ah_attr;
+	struct ib_ah_attr alt_ah_attr;
+	struct qib_qp __rcu *next;            /* link list for QPN hash table */
+	struct qib_swqe *s_wq;  /* send work queue */
+	struct qib_mmap_info *ip;
+	struct qib_ib_header *s_hdr;     /* next packet header to send */
+	unsigned long timeout_jiffies;  /* computed from timeout */
+
+	enum ib_mtu path_mtu;
+	u32 remote_qpn;
+	u32 pmtu;		/* decoded from path_mtu */
+	u32 qkey;               /* QKEY for this QP (for UD or RD) */
+	u32 s_size;             /* send work queue size */
+	u32 s_rnr_timeout;      /* number of milliseconds for RNR timeout */
+
+	u8 state;               /* QP state */
+	u8 qp_access_flags;
+	u8 alt_timeout;         /* Alternate path timeout for this QP */
+	u8 timeout;             /* Timeout for this QP */
+	u8 s_srate;
+	u8 s_mig_state;
+	u8 port_num;
+	u8 s_pkey_index;        /* PKEY index to use */
+	u8 s_alt_pkey_index;    /* Alternate path PKEY index to use */
+	u8 r_max_rd_atomic;     /* max number of RDMA read/atomic to receive */
+	u8 s_max_rd_atomic;     /* max number of RDMA read/atomic to send */
+	u8 s_retry_cnt;         /* number of times to retry */
+	u8 s_rnr_retry_cnt;
+	u8 r_min_rnr_timer;     /* retry timeout value for RNR NAKs */
+	u8 s_max_sge;           /* size of s_wq->sg_list */
+	u8 s_draining;
+
+	/* start of read/write fields */
+
+	atomic_t refcount ____cacheline_aligned_in_smp;
+	wait_queue_head_t wait;
+
+
+	struct qib_ack_entry s_ack_queue[QIB_MAX_RDMA_ATOMIC + 1]
+		____cacheline_aligned_in_smp;
+	struct qib_sge_state s_rdma_read_sge;
+
+	spinlock_t r_lock ____cacheline_aligned_in_smp;      /* used for APM */
+	unsigned long r_aflags;
+	u64 r_wr_id;            /* ID for current receive WQE */
+	u32 r_ack_psn;          /* PSN for next ACK or atomic ACK */
+	u32 r_len;              /* total length of r_sge */
+	u32 r_rcv_len;          /* receive data len processed */
+	u32 r_psn;              /* expected rcv packet sequence number */
+	u32 r_msn;              /* message sequence number */
+
+	u8 r_state;             /* opcode of last packet received */
+	u8 r_flags;
+	u8 r_head_ack_queue;    /* index into s_ack_queue[] */
+
+	struct list_head rspwait;       /* link for waititing to respond */
+
+	struct qib_sge_state r_sge;     /* current receive data */
+	struct qib_rq r_rq;             /* receive work queue */
+
+	spinlock_t s_lock ____cacheline_aligned_in_smp;
+	struct qib_sge_state *s_cur_sge;
+	u32 s_flags;
+	struct qib_verbs_txreq *s_tx;
+	struct qib_swqe *s_wqe;
+	struct qib_sge_state s_sge;     /* current send request data */
+	struct qib_mregion *s_rdma_mr;
+	atomic_t s_dma_busy;
+	u32 s_cur_size;         /* size of send packet in bytes */
+	u32 s_len;              /* total length of s_sge */
+	u32 s_rdma_read_len;    /* total length of s_rdma_read_sge */
+	u32 s_next_psn;         /* PSN for next request */
+	u32 s_last_psn;         /* last response PSN processed */
+	u32 s_sending_psn;      /* lowest PSN that is being sent */
+	u32 s_sending_hpsn;     /* highest PSN that is being sent */
+	u32 s_psn;              /* current packet sequence number */
+	u32 s_ack_rdma_psn;     /* PSN for sending RDMA read responses */
+	u32 s_ack_psn;          /* PSN for acking sends and RDMA writes */
+	u32 s_head;             /* new entries added here */
+	u32 s_tail;             /* next entry to process */
+	u32 s_cur;              /* current work queue entry */
+	u32 s_acked;            /* last un-ACK'ed entry */
+	u32 s_last;             /* last completed entry */
+	u32 s_ssn;              /* SSN of tail entry */
+	u32 s_lsn;              /* limit sequence number (credit) */
+	u16 s_hdrwords;         /* size of s_hdr in 32 bit words */
+	u16 s_rdma_ack_cnt;
+	u8 s_state;             /* opcode of last packet sent */
+	u8 s_ack_state;         /* opcode of packet to ACK */
+	u8 s_nak_state;         /* non-zero if NAK is pending */
+	u8 r_nak_state;         /* non-zero if NAK is pending */
+	u8 s_retry;             /* requester retry counter */
+	u8 s_rnr_retry;         /* requester RNR retry counter */
+	u8 s_num_rd_atomic;     /* number of RDMA read/atomic pending */
+	u8 s_tail_ack_queue;    /* index into s_ack_queue[] */
+
+	struct qib_sge_state s_ack_rdma_sge;
+	struct timer_list s_timer;
+	struct list_head iowait;        /* link for wait PIO buf */
+
+	struct work_struct s_work;
+
+	wait_queue_head_t wait_dma;
+
+	struct qib_sge r_sg_list[0] /* verified SGEs */
+		____cacheline_aligned_in_smp;
+};
+
+/*
+ * Atomic bit definitions for r_aflags.
+ */
+#define QIB_R_WRID_VALID        0
+#define QIB_R_REWIND_SGE        1
+
+/*
+ * Bit definitions for r_flags.
+ */
+#define QIB_R_REUSE_SGE 0x01
+#define QIB_R_RDMAR_SEQ 0x02
+#define QIB_R_RSP_NAK   0x04
+#define QIB_R_RSP_SEND  0x08
+#define QIB_R_COMM_EST  0x10
+
+/*
+ * Bit definitions for s_flags.
+ *
+ * QIB_S_SIGNAL_REQ_WR - set if QP send WRs contain completion signaled
+ * QIB_S_BUSY - send tasklet is processing the QP
+ * QIB_S_TIMER - the RC retry timer is active
+ * QIB_S_ACK_PENDING - an ACK is waiting to be sent after RDMA read/atomics
+ * QIB_S_WAIT_FENCE - waiting for all prior RDMA read or atomic SWQEs
+ *                         before processing the next SWQE
+ * QIB_S_WAIT_RDMAR - waiting for a RDMA read or atomic SWQE to complete
+ *                         before processing the next SWQE
+ * QIB_S_WAIT_RNR - waiting for RNR timeout
+ * QIB_S_WAIT_SSN_CREDIT - waiting for RC credits to process next SWQE
+ * QIB_S_WAIT_DMA - waiting for send DMA queue to drain before generating
+ *                  next send completion entry not via send DMA
+ * QIB_S_WAIT_PIO - waiting for a send buffer to be available
+ * QIB_S_WAIT_TX - waiting for a struct qib_verbs_txreq to be available
+ * QIB_S_WAIT_DMA_DESC - waiting for DMA descriptors to be available
+ * QIB_S_WAIT_KMEM - waiting for kernel memory to be available
+ * QIB_S_WAIT_PSN - waiting for a packet to exit the send DMA queue
+ * QIB_S_WAIT_ACK - waiting for an ACK packet before sending more requests
+ * QIB_S_SEND_ONE - send one packet, request ACK, then wait for ACK
+ */
+#define QIB_S_SIGNAL_REQ_WR	0x0001
+#define QIB_S_BUSY		0x0002
+#define QIB_S_TIMER		0x0004
+#define QIB_S_RESP_PENDING	0x0008
+#define QIB_S_ACK_PENDING	0x0010
+#define QIB_S_WAIT_FENCE	0x0020
+#define QIB_S_WAIT_RDMAR	0x0040
+#define QIB_S_WAIT_RNR		0x0080
+#define QIB_S_WAIT_SSN_CREDIT	0x0100
+#define QIB_S_WAIT_DMA		0x0200
+#define QIB_S_WAIT_PIO		0x0400
+#define QIB_S_WAIT_TX		0x0800
+#define QIB_S_WAIT_DMA_DESC	0x1000
+#define QIB_S_WAIT_KMEM		0x2000
+#define QIB_S_WAIT_PSN		0x4000
+#define QIB_S_WAIT_ACK		0x8000
+#define QIB_S_SEND_ONE		0x10000
+#define QIB_S_UNLIMITED_CREDIT	0x20000
+
+/*
+ * Wait flags that would prevent any packet type from being sent.
+ */
+#define QIB_S_ANY_WAIT_IO (QIB_S_WAIT_PIO | QIB_S_WAIT_TX | \
+	QIB_S_WAIT_DMA_DESC | QIB_S_WAIT_KMEM)
+
+/*
+ * Wait flags that would prevent send work requests from making progress.
+ */
+#define QIB_S_ANY_WAIT_SEND (QIB_S_WAIT_FENCE | QIB_S_WAIT_RDMAR | \
+	QIB_S_WAIT_RNR | QIB_S_WAIT_SSN_CREDIT | QIB_S_WAIT_DMA | \
+	QIB_S_WAIT_PSN | QIB_S_WAIT_ACK)
+
+#define QIB_S_ANY_WAIT (QIB_S_ANY_WAIT_IO | QIB_S_ANY_WAIT_SEND)
+
+#define QIB_PSN_CREDIT  16
+
+/*
+ * Since struct qib_swqe is not a fixed size, we can't simply index into
+ * struct qib_qp.s_wq.  This function does the array index computation.
+ */
+static inline struct qib_swqe *get_swqe_ptr(struct qib_qp *qp,
+					      unsigned n)
+{
+	return (struct qib_swqe *)((char *)qp->s_wq +
+				     (sizeof(struct qib_swqe) +
+				      qp->s_max_sge *
+				      sizeof(struct qib_sge)) * n);
+}
+
+/*
+ * Since struct qib_rwqe is not a fixed size, we can't simply index into
+ * struct qib_rwq.wq.  This function does the array index computation.
+ */
+static inline struct qib_rwqe *get_rwqe_ptr(struct qib_rq *rq, unsigned n)
+{
+	return (struct qib_rwqe *)
+		((char *) rq->wq->wq +
+		 (sizeof(struct qib_rwqe) +
+		  rq->max_sge * sizeof(struct ib_sge)) * n);
+}
+
+/*
+ * QPN-map pages start out as NULL, they get allocated upon
+ * first use and are never deallocated. This way,
+ * large bitmaps are not allocated unless large numbers of QPs are used.
+ */
+struct qpn_map {
+	void *page;
+};
+
+struct qib_qpn_table {
+	spinlock_t lock; /* protect changes in this struct */
+	unsigned flags;         /* flags for QP0/1 allocated for each port */
+	u32 last;               /* last QP number allocated */
+	u32 nmaps;              /* size of the map table */
+	u16 limit;
+	u16 mask;
+	/* bit map of free QP numbers other than 0/1 */
+	struct qpn_map map[QPNMAP_ENTRIES];
+};
+
+struct qib_lkey_table {
+	spinlock_t lock; /* protect changes in this struct */
+	u32 next;               /* next unused index (speeds search) */
+	u32 gen;                /* generation count */
+	u32 max;                /* size of the table */
+	struct qib_mregion __rcu **table;
+};
+
+struct qib_opcode_stats {
+	u64 n_packets;          /* number of packets */
+	u64 n_bytes;            /* total number of bytes */
+};
+
+struct qib_opcode_stats_perctx {
+	struct qib_opcode_stats stats[128];
+};
+
+struct qib_pma_counters {
+	u64 n_unicast_xmit;     /* total unicast packets sent */
+	u64 n_unicast_rcv;      /* total unicast packets received */
+	u64 n_multicast_xmit;   /* total multicast packets sent */
+	u64 n_multicast_rcv;    /* total multicast packets received */
+};
+
+struct qib_ibport {
+	struct qib_qp __rcu *qp0;
+	struct qib_qp __rcu *qp1;
+	struct ib_mad_agent *send_agent;	/* agent for SMI (traps) */
+	struct qib_ah *sm_ah;
+	struct qib_ah *smi_ah;
+	struct rb_root mcast_tree;
+	spinlock_t lock;		/* protect changes in this struct */
+
+	/* non-zero when timer is set */
+	unsigned long mkey_lease_timeout;
+	unsigned long trap_timeout;
+	__be64 gid_prefix;      /* in network order */
+	__be64 mkey;
+	__be64 guids[QIB_GUIDS_PER_PORT	- 1];	/* writable GUIDs */
+	u64 tid;		/* TID for traps */
+	struct qib_pma_counters __percpu *pmastats;
+	u64 z_unicast_xmit;     /* starting count for PMA */
+	u64 z_unicast_rcv;      /* starting count for PMA */
+	u64 z_multicast_xmit;   /* starting count for PMA */
+	u64 z_multicast_rcv;    /* starting count for PMA */
+	u64 z_symbol_error_counter;             /* starting count for PMA */
+	u64 z_link_error_recovery_counter;      /* starting count for PMA */
+	u64 z_link_downed_counter;              /* starting count for PMA */
+	u64 z_port_rcv_errors;                  /* starting count for PMA */
+	u64 z_port_rcv_remphys_errors;          /* starting count for PMA */
+	u64 z_port_xmit_discards;               /* starting count for PMA */
+	u64 z_port_xmit_data;                   /* starting count for PMA */
+	u64 z_port_rcv_data;                    /* starting count for PMA */
+	u64 z_port_xmit_packets;                /* starting count for PMA */
+	u64 z_port_rcv_packets;                 /* starting count for PMA */
+	u32 z_local_link_integrity_errors;      /* starting count for PMA */
+	u32 z_excessive_buffer_overrun_errors;  /* starting count for PMA */
+	u32 z_vl15_dropped;                     /* starting count for PMA */
+	u32 n_rc_resends;
+	u32 n_rc_acks;
+	u32 n_rc_qacks;
+	u32 n_rc_delayed_comp;
+	u32 n_seq_naks;
+	u32 n_rdma_seq;
+	u32 n_rnr_naks;
+	u32 n_other_naks;
+	u32 n_loop_pkts;
+	u32 n_pkt_drops;
+	u32 n_vl15_dropped;
+	u32 n_rc_timeouts;
+	u32 n_dmawait;
+	u32 n_unaligned;
+	u32 n_rc_dupreq;
+	u32 n_rc_seqnak;
+	u32 port_cap_flags;
+	u32 pma_sample_start;
+	u32 pma_sample_interval;
+	__be16 pma_counter_select[5];
+	u16 pma_tag;
+	u16 pkey_violations;
+	u16 qkey_violations;
+	u16 mkey_violations;
+	u16 mkey_lease_period;
+	u16 sm_lid;
+	u16 repress_traps;
+	u8 sm_sl;
+	u8 mkeyprot;
+	u8 subnet_timeout;
+	u8 vl_high_limit;
+	u8 sl_to_vl[16];
+
+};
+
+
+struct qib_ibdev {
+	struct ib_device ibdev;
+	struct list_head pending_mmaps;
+	spinlock_t mmap_offset_lock; /* protect mmap_offset */
+	u32 mmap_offset;
+	struct qib_mregion __rcu *dma_mr;
+
+	/* QP numbers are shared by all IB ports */
+	struct qib_qpn_table qpn_table;
+	struct qib_lkey_table lk_table;
+	struct list_head piowait;       /* list for wait PIO buf */
+	struct list_head dmawait;	/* list for wait DMA */
+	struct list_head txwait;        /* list for wait qib_verbs_txreq */
+	struct list_head memwait;       /* list for wait kernel memory */
+	struct list_head txreq_free;
+	struct timer_list mem_timer;
+	struct qib_qp __rcu **qp_table;
+	struct qib_pio_header *pio_hdrs;
+	dma_addr_t pio_hdrs_phys;
+	/* list of QPs waiting for RNR timer */
+	spinlock_t pending_lock; /* protect wait lists, PMA counters, etc. */
+	u32 qp_table_size; /* size of the hash table */
+	u32 qp_rnd; /* random bytes for hash */
+	spinlock_t qpt_lock;
+
+	u32 n_piowait;
+	u32 n_txwait;
+
+	u32 n_pds_allocated;    /* number of PDs allocated for device */
+	spinlock_t n_pds_lock;
+	u32 n_ahs_allocated;    /* number of AHs allocated for device */
+	spinlock_t n_ahs_lock;
+	u32 n_cqs_allocated;    /* number of CQs allocated for device */
+	spinlock_t n_cqs_lock;
+	u32 n_qps_allocated;    /* number of QPs allocated for device */
+	spinlock_t n_qps_lock;
+	u32 n_srqs_allocated;   /* number of SRQs allocated for device */
+	spinlock_t n_srqs_lock;
+	u32 n_mcast_grps_allocated; /* number of mcast groups allocated */
+	spinlock_t n_mcast_grps_lock;
+#ifdef CONFIG_DEBUG_FS
+	/* per HCA debugfs */
+	struct dentry *qib_ibdev_dbg;
+#endif
+};
+
+struct qib_verbs_counters {
+	u64 symbol_error_counter;
+	u64 link_error_recovery_counter;
+	u64 link_downed_counter;
+	u64 port_rcv_errors;
+	u64 port_rcv_remphys_errors;
+	u64 port_xmit_discards;
+	u64 port_xmit_data;
+	u64 port_rcv_data;
+	u64 port_xmit_packets;
+	u64 port_rcv_packets;
+	u32 local_link_integrity_errors;
+	u32 excessive_buffer_overrun_errors;
+	u32 vl15_dropped;
+};
+
+static inline struct qib_mr *to_imr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct qib_mr, ibmr);
+}
+
+static inline struct qib_pd *to_ipd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct qib_pd, ibpd);
+}
+
+static inline struct qib_ah *to_iah(struct ib_ah *ibah)
+{
+	return container_of(ibah, struct qib_ah, ibah);
+}
+
+static inline struct qib_cq *to_icq(struct ib_cq *ibcq)
+{
+	return container_of(ibcq, struct qib_cq, ibcq);
+}
+
+static inline struct qib_srq *to_isrq(struct ib_srq *ibsrq)
+{
+	return container_of(ibsrq, struct qib_srq, ibsrq);
+}
+
+static inline struct qib_qp *to_iqp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct qib_qp, ibqp);
+}
+
+static inline struct qib_ibdev *to_idev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct qib_ibdev, ibdev);
+}
+
+/*
+ * Send if not busy or waiting for I/O and either
+ * a RC response is pending or we can process send work requests.
+ */
+static inline int qib_send_ok(struct qib_qp *qp)
+{
+	return !(qp->s_flags & (QIB_S_BUSY | QIB_S_ANY_WAIT_IO)) &&
+		(qp->s_hdrwords || (qp->s_flags & QIB_S_RESP_PENDING) ||
+		 !(qp->s_flags & QIB_S_ANY_WAIT_SEND));
+}
+
+/*
+ * This must be called with s_lock held.
+ */
+void qib_schedule_send(struct qib_qp *qp);
+
+static inline int qib_pkey_ok(u16 pkey1, u16 pkey2)
+{
+	u16 p1 = pkey1 & 0x7FFF;
+	u16 p2 = pkey2 & 0x7FFF;
+
+	/*
+	 * Low 15 bits must be non-zero and match, and
+	 * one of the two must be a full member.
+	 */
+	return p1 && p1 == p2 && ((__s16)pkey1 < 0 || (__s16)pkey2 < 0);
+}
+
+void qib_bad_pqkey(struct qib_ibport *ibp, __be16 trap_num, u32 key, u32 sl,
+		   u32 qp1, u32 qp2, __be16 lid1, __be16 lid2);
+void qib_cap_mask_chg(struct qib_ibport *ibp);
+void qib_sys_guid_chg(struct qib_ibport *ibp);
+void qib_node_desc_chg(struct qib_ibport *ibp);
+int qib_process_mad(struct ib_device *ibdev, int mad_flags, u8 port_num,
+		    struct ib_wc *in_wc, struct ib_grh *in_grh,
+		    struct ib_mad *in_mad, struct ib_mad *out_mad);
+int qib_create_agents(struct qib_ibdev *dev);
+void qib_free_agents(struct qib_ibdev *dev);
+
+/*
+ * Compare the lower 24 bits of the two values.
+ * Returns an integer <, ==, or > than zero.
+ */
+static inline int qib_cmp24(u32 a, u32 b)
+{
+	return (((int) a) - ((int) b)) << 8;
+}
+
+struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid);
+
+int qib_snapshot_counters(struct qib_pportdata *ppd, u64 *swords,
+			  u64 *rwords, u64 *spkts, u64 *rpkts,
+			  u64 *xmit_wait);
+
+int qib_get_counters(struct qib_pportdata *ppd,
+		     struct qib_verbs_counters *cntrs);
+
+int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+
+int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid);
+
+int qib_mcast_tree_empty(struct qib_ibport *ibp);
+
+__be32 qib_compute_aeth(struct qib_qp *qp);
+
+struct qib_qp *qib_lookup_qpn(struct qib_ibport *ibp, u32 qpn);
+
+struct ib_qp *qib_create_qp(struct ib_pd *ibpd,
+			    struct ib_qp_init_attr *init_attr,
+			    struct ib_udata *udata);
+
+int qib_destroy_qp(struct ib_qp *ibqp);
+
+int qib_error_qp(struct qib_qp *qp, enum ib_wc_status err);
+
+int qib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		  int attr_mask, struct ib_udata *udata);
+
+int qib_query_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+		 int attr_mask, struct ib_qp_init_attr *init_attr);
+
+unsigned qib_free_all_qps(struct qib_devdata *dd);
+
+void qib_init_qpn_table(struct qib_devdata *dd, struct qib_qpn_table *qpt);
+
+void qib_free_qpn_table(struct qib_qpn_table *qpt);
+
+#ifdef CONFIG_DEBUG_FS
+
+struct qib_qp_iter;
+
+struct qib_qp_iter *qib_qp_iter_init(struct qib_ibdev *dev);
+
+int qib_qp_iter_next(struct qib_qp_iter *iter);
+
+void qib_qp_iter_print(struct seq_file *s, struct qib_qp_iter *iter);
+
+#endif
+
+void qib_get_credit(struct qib_qp *qp, u32 aeth);
+
+unsigned qib_pkt_delay(u32 plen, u8 snd_mult, u8 rcv_mult);
+
+void qib_verbs_sdma_desc_avail(struct qib_pportdata *ppd, unsigned avail);
+
+void qib_put_txreq(struct qib_verbs_txreq *tx);
+
+int qib_verbs_send(struct qib_qp *qp, struct qib_ib_header *hdr,
+		   u32 hdrwords, struct qib_sge_state *ss, u32 len);
+
+void qib_copy_sge(struct qib_sge_state *ss, void *data, u32 length,
+		  int release);
+
+void qib_skip_sge(struct qib_sge_state *ss, u32 length, int release);
+
+void qib_uc_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+void qib_rc_rcv(struct qib_ctxtdata *rcd, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+int qib_check_ah(struct ib_device *ibdev, struct ib_ah_attr *ah_attr);
+
+struct ib_ah *qib_create_qp0_ah(struct qib_ibport *ibp, u16 dlid);
+
+void qib_rc_rnr_retry(unsigned long arg);
+
+void qib_rc_send_complete(struct qib_qp *qp, struct qib_ib_header *hdr);
+
+void qib_rc_error(struct qib_qp *qp, enum ib_wc_status err);
+
+int qib_post_ud_send(struct qib_qp *qp, struct ib_send_wr *wr);
+
+void qib_ud_rcv(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		int has_grh, void *data, u32 tlen, struct qib_qp *qp);
+
+int qib_alloc_lkey(struct qib_mregion *mr, int dma_region);
+
+void qib_free_lkey(struct qib_mregion *mr);
+
+int qib_lkey_ok(struct qib_lkey_table *rkt, struct qib_pd *pd,
+		struct qib_sge *isge, struct ib_sge *sge, int acc);
+
+int qib_rkey_ok(struct qib_qp *qp, struct qib_sge *sge,
+		u32 len, u64 vaddr, u32 rkey, int acc);
+
+int qib_post_srq_receive(struct ib_srq *ibsrq, struct ib_recv_wr *wr,
+			 struct ib_recv_wr **bad_wr);
+
+struct ib_srq *qib_create_srq(struct ib_pd *ibpd,
+			      struct ib_srq_init_attr *srq_init_attr,
+			      struct ib_udata *udata);
+
+int qib_modify_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr,
+		   enum ib_srq_attr_mask attr_mask,
+		   struct ib_udata *udata);
+
+int qib_query_srq(struct ib_srq *ibsrq, struct ib_srq_attr *attr);
+
+int qib_destroy_srq(struct ib_srq *ibsrq);
+
+int qib_cq_init(struct qib_devdata *dd);
+
+void qib_cq_exit(struct qib_devdata *dd);
+
+void qib_cq_enter(struct qib_cq *cq, struct ib_wc *entry, int sig);
+
+int qib_poll_cq(struct ib_cq *ibcq, int num_entries, struct ib_wc *entry);
+
+struct ib_cq *qib_create_cq(struct ib_device *ibdev, int entries,
+			    int comp_vector, struct ib_ucontext *context,
+			    struct ib_udata *udata);
+
+int qib_destroy_cq(struct ib_cq *ibcq);
+
+int qib_req_notify_cq(struct ib_cq *ibcq, enum ib_cq_notify_flags notify_flags);
+
+int qib_resize_cq(struct ib_cq *ibcq, int cqe, struct ib_udata *udata);
+
+struct ib_mr *qib_get_dma_mr(struct ib_pd *pd, int acc);
+
+struct ib_mr *qib_reg_phys_mr(struct ib_pd *pd,
+			      struct ib_phys_buf *buffer_list,
+			      int num_phys_buf, int acc, u64 *iova_start);
+
+struct ib_mr *qib_reg_user_mr(struct ib_pd *pd, u64 start, u64 length,
+			      u64 virt_addr, int mr_access_flags,
+			      struct ib_udata *udata);
+
+int qib_dereg_mr(struct ib_mr *ibmr);
+
+struct ib_mr *qib_alloc_fast_reg_mr(struct ib_pd *pd, int max_page_list_len);
+
+struct ib_fast_reg_page_list *qib_alloc_fast_reg_page_list(
+				struct ib_device *ibdev, int page_list_len);
+
+void qib_free_fast_reg_page_list(struct ib_fast_reg_page_list *pl);
+
+int qib_fast_reg_mr(struct qib_qp *qp, struct ib_send_wr *wr);
+
+struct ib_fmr *qib_alloc_fmr(struct ib_pd *pd, int mr_access_flags,
+			     struct ib_fmr_attr *fmr_attr);
+
+int qib_map_phys_fmr(struct ib_fmr *ibfmr, u64 *page_list,
+		     int list_len, u64 iova);
+
+int qib_unmap_fmr(struct list_head *fmr_list);
+
+int qib_dealloc_fmr(struct ib_fmr *ibfmr);
+
+static inline void qib_get_mr(struct qib_mregion *mr)
+{
+	atomic_inc(&mr->refcount);
+}
+
+void mr_rcu_callback(struct rcu_head *list);
+
+static inline void qib_put_mr(struct qib_mregion *mr)
+{
+	if (unlikely(atomic_dec_and_test(&mr->refcount)))
+		call_rcu(&mr->list, mr_rcu_callback);
+}
+
+static inline void qib_put_ss(struct qib_sge_state *ss)
+{
+	while (ss->num_sge) {
+		qib_put_mr(ss->sge.mr);
+		if (--ss->num_sge)
+			ss->sge = *ss->sg_list++;
+	}
+}
+
+
+void qib_release_mmap_info(struct kref *ref);
+
+struct qib_mmap_info *qib_create_mmap_info(struct qib_ibdev *dev, u32 size,
+					   struct ib_ucontext *context,
+					   void *obj);
+
+void qib_update_mmap_info(struct qib_ibdev *dev, struct qib_mmap_info *ip,
+			  u32 size, void *obj);
+
+int qib_mmap(struct ib_ucontext *context, struct vm_area_struct *vma);
+
+int qib_get_rwqe(struct qib_qp *qp, int wr_id_only);
+
+void qib_migrate_qp(struct qib_qp *qp);
+
+int qib_ruc_check_hdr(struct qib_ibport *ibp, struct qib_ib_header *hdr,
+		      int has_grh, struct qib_qp *qp, u32 bth0);
+
+u32 qib_make_grh(struct qib_ibport *ibp, struct ib_grh *hdr,
+		 struct ib_global_route *grh, u32 hwords, u32 nwords);
+
+void qib_make_ruc_header(struct qib_qp *qp, struct qib_other_headers *ohdr,
+			 u32 bth0, u32 bth2);
+
+void qib_do_send(struct work_struct *work);
+
+void qib_send_complete(struct qib_qp *qp, struct qib_swqe *wqe,
+		       enum ib_wc_status status);
+
+void qib_send_rc_ack(struct qib_qp *qp);
+
+int qib_make_rc_req(struct qib_qp *qp);
+
+int qib_make_uc_req(struct qib_qp *qp);
+
+int qib_make_ud_req(struct qib_qp *qp);
+
+int qib_register_ib_device(struct qib_devdata *);
+
+void qib_unregister_ib_device(struct qib_devdata *);
+
+void qib_ib_rcv(struct qib_ctxtdata *, void *, void *, u32);
+
+void qib_ib_piobufavail(struct qib_devdata *);
+
+unsigned qib_get_npkeys(struct qib_devdata *);
+
+unsigned qib_get_pkey(struct qib_ibport *, unsigned);
+
+extern const enum ib_wc_opcode ib_qib_wc_opcode[];
+
+/*
+ * Below  HCA-independent IB PhysPortState values, returned
+ * by the f_ibphys_portstate() routine.
+ */
+#define IB_PHYSPORTSTATE_SLEEP 1
+#define IB_PHYSPORTSTATE_POLL 2
+#define IB_PHYSPORTSTATE_DISABLED 3
+#define IB_PHYSPORTSTATE_CFG_TRAIN 4
+#define IB_PHYSPORTSTATE_LINKUP 5
+#define IB_PHYSPORTSTATE_LINK_ERR_RECOVER 6
+#define IB_PHYSPORTSTATE_CFG_DEBOUNCE 8
+#define IB_PHYSPORTSTATE_CFG_IDLE 0xB
+#define IB_PHYSPORTSTATE_RECOVERY_RETRAIN 0xC
+#define IB_PHYSPORTSTATE_RECOVERY_WAITRMT 0xE
+#define IB_PHYSPORTSTATE_RECOVERY_IDLE 0xF
+#define IB_PHYSPORTSTATE_CFG_ENH 0x10
+#define IB_PHYSPORTSTATE_CFG_WAIT_ENH 0x13
+
+extern const int ib_qib_state_ops[];
+
+extern __be64 ib_qib_sys_image_guid;    /* in network order */
+
+extern unsigned int ib_qib_lkey_table_size;
+
+extern unsigned int ib_qib_max_cqes;
+
+extern unsigned int ib_qib_max_cqs;
+
+extern unsigned int ib_qib_max_qp_wrs;
+
+extern unsigned int ib_qib_max_qps;
+
+extern unsigned int ib_qib_max_sges;
+
+extern unsigned int ib_qib_max_mcast_grps;
+
+extern unsigned int ib_qib_max_mcast_qp_attached;
+
+extern unsigned int ib_qib_max_srqs;
+
+extern unsigned int ib_qib_max_srq_sges;
+
+extern unsigned int ib_qib_max_srq_wrs;
+
+extern const u32 ib_qib_rnr_table[];
+
+extern struct ib_dma_mapping_ops qib_dma_mapping_ops;
+
+#endif                          /* QIB_VERBS_H */
diff --git a/drivers/infiniband/hw/qib/qib_verbs_mcast.c b/drivers/infiniband/hw/qib/qib_verbs_mcast.c
new file mode 100644
index 00000000000..dabb697b1c2
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_verbs_mcast.c
@@ -0,0 +1,368 @@
+/*
+ * Copyright (c) 2006, 2007, 2008, 2009 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/rculist.h>
+
+#include "qib.h"
+
+/**
+ * qib_mcast_qp_alloc - alloc a struct to link a QP to mcast GID struct
+ * @qp: the QP to link
+ */
+static struct qib_mcast_qp *qib_mcast_qp_alloc(struct qib_qp *qp)
+{
+	struct qib_mcast_qp *mqp;
+
+	mqp = kmalloc(sizeof *mqp, GFP_KERNEL);
+	if (!mqp)
+		goto bail;
+
+	mqp->qp = qp;
+	atomic_inc(&qp->refcount);
+
+bail:
+	return mqp;
+}
+
+static void qib_mcast_qp_free(struct qib_mcast_qp *mqp)
+{
+	struct qib_qp *qp = mqp->qp;
+
+	/* Notify qib_destroy_qp() if it is waiting. */
+	if (atomic_dec_and_test(&qp->refcount))
+		wake_up(&qp->wait);
+
+	kfree(mqp);
+}
+
+/**
+ * qib_mcast_alloc - allocate the multicast GID structure
+ * @mgid: the multicast GID
+ *
+ * A list of QPs will be attached to this structure.
+ */
+static struct qib_mcast *qib_mcast_alloc(union ib_gid *mgid)
+{
+	struct qib_mcast *mcast;
+
+	mcast = kmalloc(sizeof *mcast, GFP_KERNEL);
+	if (!mcast)
+		goto bail;
+
+	mcast->mgid = *mgid;
+	INIT_LIST_HEAD(&mcast->qp_list);
+	init_waitqueue_head(&mcast->wait);
+	atomic_set(&mcast->refcount, 0);
+	mcast->n_attached = 0;
+
+bail:
+	return mcast;
+}
+
+static void qib_mcast_free(struct qib_mcast *mcast)
+{
+	struct qib_mcast_qp *p, *tmp;
+
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list)
+		qib_mcast_qp_free(p);
+
+	kfree(mcast);
+}
+
+/**
+ * qib_mcast_find - search the global table for the given multicast GID
+ * @ibp: the IB port structure
+ * @mgid: the multicast GID to search for
+ *
+ * Returns NULL if not found.
+ *
+ * The caller is responsible for decrementing the reference count if found.
+ */
+struct qib_mcast *qib_mcast_find(struct qib_ibport *ibp, union ib_gid *mgid)
+{
+	struct rb_node *n;
+	unsigned long flags;
+	struct qib_mcast *mcast;
+
+	spin_lock_irqsave(&ibp->lock, flags);
+	n = ibp->mcast_tree.rb_node;
+	while (n) {
+		int ret;
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+
+		ret = memcmp(mgid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else {
+			atomic_inc(&mcast->refcount);
+			spin_unlock_irqrestore(&ibp->lock, flags);
+			goto bail;
+		}
+	}
+	spin_unlock_irqrestore(&ibp->lock, flags);
+
+	mcast = NULL;
+
+bail:
+	return mcast;
+}
+
+/**
+ * qib_mcast_add - insert mcast GID into table and attach QP struct
+ * @mcast: the mcast GID table
+ * @mqp: the QP to attach
+ *
+ * Return zero if both were added.  Return EEXIST if the GID was already in
+ * the table but the QP was added.  Return ESRCH if the QP was already
+ * attached and neither structure was added.
+ */
+static int qib_mcast_add(struct qib_ibdev *dev, struct qib_ibport *ibp,
+			 struct qib_mcast *mcast, struct qib_mcast_qp *mqp)
+{
+	struct rb_node **n = &ibp->mcast_tree.rb_node;
+	struct rb_node *pn = NULL;
+	int ret;
+
+	spin_lock_irq(&ibp->lock);
+
+	while (*n) {
+		struct qib_mcast *tmcast;
+		struct qib_mcast_qp *p;
+
+		pn = *n;
+		tmcast = rb_entry(pn, struct qib_mcast, rb_node);
+
+		ret = memcmp(mcast->mgid.raw, tmcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0) {
+			n = &pn->rb_left;
+			continue;
+		}
+		if (ret > 0) {
+			n = &pn->rb_right;
+			continue;
+		}
+
+		/* Search the QP list to see if this is already there. */
+		list_for_each_entry_rcu(p, &tmcast->qp_list, list) {
+			if (p->qp == mqp->qp) {
+				ret = ESRCH;
+				goto bail;
+			}
+		}
+		if (tmcast->n_attached == ib_qib_max_mcast_qp_attached) {
+			ret = ENOMEM;
+			goto bail;
+		}
+
+		tmcast->n_attached++;
+
+		list_add_tail_rcu(&mqp->list, &tmcast->qp_list);
+		ret = EEXIST;
+		goto bail;
+	}
+
+	spin_lock(&dev->n_mcast_grps_lock);
+	if (dev->n_mcast_grps_allocated == ib_qib_max_mcast_grps) {
+		spin_unlock(&dev->n_mcast_grps_lock);
+		ret = ENOMEM;
+		goto bail;
+	}
+
+	dev->n_mcast_grps_allocated++;
+	spin_unlock(&dev->n_mcast_grps_lock);
+
+	mcast->n_attached++;
+
+	list_add_tail_rcu(&mqp->list, &mcast->qp_list);
+
+	atomic_inc(&mcast->refcount);
+	rb_link_node(&mcast->rb_node, pn, n);
+	rb_insert_color(&mcast->rb_node, &ibp->mcast_tree);
+
+	ret = 0;
+
+bail:
+	spin_unlock_irq(&ibp->lock);
+
+	return ret;
+}
+
+int qib_multicast_attach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp;
+	struct qib_mcast *mcast;
+	struct qib_mcast_qp *mqp;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	/*
+	 * Allocate data structures since its better to do this outside of
+	 * spin locks and it will most likely be needed.
+	 */
+	mcast = qib_mcast_alloc(gid);
+	if (mcast == NULL) {
+		ret = -ENOMEM;
+		goto bail;
+	}
+	mqp = qib_mcast_qp_alloc(qp);
+	if (mqp == NULL) {
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+	}
+	ibp = to_iport(ibqp->device, qp->port_num);
+	switch (qib_mcast_add(dev, ibp, mcast, mqp)) {
+	case ESRCH:
+		/* Neither was used: OK to attach the same QP twice. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		break;
+
+	case EEXIST:            /* The mcast wasn't used */
+		qib_mcast_free(mcast);
+		break;
+
+	case ENOMEM:
+		/* Exceeded the maximum number of mcast groups. */
+		qib_mcast_qp_free(mqp);
+		qib_mcast_free(mcast);
+		ret = -ENOMEM;
+		goto bail;
+
+	default:
+		break;
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_multicast_detach(struct ib_qp *ibqp, union ib_gid *gid, u16 lid)
+{
+	struct qib_qp *qp = to_iqp(ibqp);
+	struct qib_ibdev *dev = to_idev(ibqp->device);
+	struct qib_ibport *ibp = to_iport(ibqp->device, qp->port_num);
+	struct qib_mcast *mcast = NULL;
+	struct qib_mcast_qp *p, *tmp;
+	struct rb_node *n;
+	int last = 0;
+	int ret;
+
+	if (ibqp->qp_num <= 1 || qp->state == IB_QPS_RESET) {
+		ret = -EINVAL;
+		goto bail;
+	}
+
+	spin_lock_irq(&ibp->lock);
+
+	/* Find the GID in the mcast table. */
+	n = ibp->mcast_tree.rb_node;
+	while (1) {
+		if (n == NULL) {
+			spin_unlock_irq(&ibp->lock);
+			ret = -EINVAL;
+			goto bail;
+		}
+
+		mcast = rb_entry(n, struct qib_mcast, rb_node);
+		ret = memcmp(gid->raw, mcast->mgid.raw,
+			     sizeof(union ib_gid));
+		if (ret < 0)
+			n = n->rb_left;
+		else if (ret > 0)
+			n = n->rb_right;
+		else
+			break;
+	}
+
+	/* Search the QP list. */
+	list_for_each_entry_safe(p, tmp, &mcast->qp_list, list) {
+		if (p->qp != qp)
+			continue;
+		/*
+		 * We found it, so remove it, but don't poison the forward
+		 * link until we are sure there are no list walkers.
+		 */
+		list_del_rcu(&p->list);
+		mcast->n_attached--;
+
+		/* If this was the last attached QP, remove the GID too. */
+		if (list_empty(&mcast->qp_list)) {
+			rb_erase(&mcast->rb_node, &ibp->mcast_tree);
+			last = 1;
+		}
+		break;
+	}
+
+	spin_unlock_irq(&ibp->lock);
+
+	if (p) {
+		/*
+		 * Wait for any list walkers to finish before freeing the
+		 * list element.
+		 */
+		wait_event(mcast->wait, atomic_read(&mcast->refcount) <= 1);
+		qib_mcast_qp_free(p);
+	}
+	if (last) {
+		atomic_dec(&mcast->refcount);
+		wait_event(mcast->wait, !atomic_read(&mcast->refcount));
+		qib_mcast_free(mcast);
+		spin_lock_irq(&dev->n_mcast_grps_lock);
+		dev->n_mcast_grps_allocated--;
+		spin_unlock_irq(&dev->n_mcast_grps_lock);
+	}
+
+	ret = 0;
+
+bail:
+	return ret;
+}
+
+int qib_mcast_tree_empty(struct qib_ibport *ibp)
+{
+	return ibp->mcast_tree.rb_node == NULL;
+}
diff --git a/drivers/infiniband/hw/ipath/ipath_7220.h b/drivers/infiniband/hw/qib/qib_wc_ppc64.c
index 74fa5cc5131..673cf4c22eb 100644
--- a/drivers/infiniband/hw/ipath/ipath_7220.h
+++ b/drivers/infiniband/hw/qib/qib_wc_ppc64.c
@@ -1,7 +1,5 @@
-#ifndef _IPATH_7220_H
-#define _IPATH_7220_H
 /*
- * Copyright (c) 2007 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2006, 2007, 2008 QLogic Corporation. All rights reserved.
  *
  * This software is available to you under a choice of one of two
  * licenses.  You may choose to be licensed under the terms of the GNU
@@ -33,25 +31,32 @@
  */
 
 /*
- * This header file provides the declarations and common definitions
- * for (mostly) manipulation of the SerDes blocks within the IBA7220.
- * the functions declared should only be called from within other
- * 7220-related files such as ipath_iba7220.c or ipath_sd7220.c.
+ * This file is conditionally built on PowerPC only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
  */
-int ipath_sd7220_presets(struct ipath_devdata *dd);
-int ipath_sd7220_init(struct ipath_devdata *dd, int was_reset);
-int ipath_sd7220_prog_ld(struct ipath_devdata *dd, int sdnum, u8 *img,
-	int len, int offset);
-int ipath_sd7220_prog_vfy(struct ipath_devdata *dd, int sdnum, const u8 *img,
-	int len, int offset);
-/*
- * Below used for sdnum parameter, selecting one of the two sections
- * used for PCIe, or the single SerDes used for IB, which is the
- * only one currently used
- */
-#define IB_7220_SERDES 2
 
-int ipath_sd7220_ib_load(struct ipath_devdata *dd);
-int ipath_sd7220_ib_vfy(struct ipath_devdata *dd);
+#include "qib.h"
 
-#endif /* _IPATH_7220_H */
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * Nothing to do on PowerPC, so just return without error.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	return 0;
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is unordered
+ *
+ * Because our performance depends on our ability to do write
+ * combining mmio writes in the most efficient way, we need to
+ * know if we are on a processor that may reorder stores when
+ * write combining.
+ */
+int qib_unordered_wc(void)
+{
+	return 1;
+}
diff --git a/drivers/infiniband/hw/qib/qib_wc_x86_64.c b/drivers/infiniband/hw/qib/qib_wc_x86_64.c
new file mode 100644
index 00000000000..1d7281c5a02
--- /dev/null
+++ b/drivers/infiniband/hw/qib/qib_wc_x86_64.c
@@ -0,0 +1,171 @@
+/*
+ * Copyright (c) 2012 Intel Corporation. All rights reserved.
+ * Copyright (c) 2006 - 2012 QLogic Corporation. All rights reserved.
+ * Copyright (c) 2003, 2004, 2005, 2006 PathScale, Inc. All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+/*
+ * This file is conditionally built on x86_64 only.  Otherwise weak symbol
+ * versions of the functions exported from here are used.
+ */
+
+#include <linux/pci.h>
+#include <asm/mtrr.h>
+#include <asm/processor.h>
+
+#include "qib.h"
+
+/**
+ * qib_enable_wc - enable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ *
+ * This routine is x86_64-specific; it twiddles the CPU's MTRRs to enable
+ * write combining.
+ */
+int qib_enable_wc(struct qib_devdata *dd)
+{
+	int ret = 0;
+	u64 pioaddr, piolen;
+	unsigned bits;
+	const unsigned long addr = pci_resource_start(dd->pcidev, 0);
+	const size_t len = pci_resource_len(dd->pcidev, 0);
+
+	/*
+	 * Set the PIO buffers to be WCCOMB, so we get HT bursts to the
+	 * chip.  Linux (possibly the hardware) requires it to be on a power
+	 * of 2 address matching the length (which has to be a power of 2).
+	 * For rev1, that means the base address, for rev2, it will be just
+	 * the PIO buffers themselves.
+	 * For chips with two sets of buffers, the calculations are
+	 * somewhat more complicated; we need to sum, and the piobufbase
+	 * register has both offsets, 2K in low 32 bits, 4K in high 32 bits.
+	 * The buffers are still packed, so a single range covers both.
+	 */
+	if (dd->piobcnt2k && dd->piobcnt4k) {
+		/* 2 sizes for chip */
+		unsigned long pio2kbase, pio4kbase;
+		pio2kbase = dd->piobufbase & 0xffffffffUL;
+		pio4kbase = (dd->piobufbase >> 32) & 0xffffffffUL;
+		if (pio2kbase < pio4kbase) {
+			/* all current chips */
+			pioaddr = addr + pio2kbase;
+			piolen = pio4kbase - pio2kbase +
+				dd->piobcnt4k * dd->align4k;
+		} else {
+			pioaddr = addr + pio4kbase;
+			piolen = pio2kbase - pio4kbase +
+				dd->piobcnt2k * dd->palign;
+		}
+	} else {  /* single buffer size (2K, currently) */
+		pioaddr = addr + dd->piobufbase;
+		piolen = dd->piobcnt2k * dd->palign +
+			dd->piobcnt4k * dd->align4k;
+	}
+
+	for (bits = 0; !(piolen & (1ULL << bits)); bits++)
+		/* do nothing */ ;
+
+	if (piolen != (1ULL << bits)) {
+		piolen >>= bits;
+		while (piolen >>= 1)
+			bits++;
+		piolen = 1ULL << (bits + 1);
+	}
+	if (pioaddr & (piolen - 1)) {
+		u64 atmp;
+		atmp = pioaddr & ~(piolen - 1);
+		if (atmp < addr || (atmp + piolen) > (addr + len)) {
+			qib_dev_err(dd,
+				"No way to align address/size (%llx/%llx), no WC mtrr\n",
+				(unsigned long long) atmp,
+				(unsigned long long) piolen << 1);
+			ret = -ENODEV;
+		} else {
+			pioaddr = atmp;
+			piolen <<= 1;
+		}
+	}
+
+	if (!ret) {
+		int cookie;
+
+		cookie = mtrr_add(pioaddr, piolen, MTRR_TYPE_WRCOMB, 0);
+		if (cookie < 0) {
+			{
+				qib_devinfo(dd->pcidev,
+					 "mtrr_add()  WC for PIO bufs failed (%d)\n",
+					 cookie);
+				ret = -EINVAL;
+			}
+		} else {
+			dd->wc_cookie = cookie;
+			dd->wc_base = (unsigned long) pioaddr;
+			dd->wc_len = (unsigned long) piolen;
+		}
+	}
+
+	return ret;
+}
+
+/**
+ * qib_disable_wc - disable write combining for MMIO writes to the device
+ * @dd: qlogic_ib device
+ */
+void qib_disable_wc(struct qib_devdata *dd)
+{
+	if (dd->wc_cookie) {
+		int r;
+
+		r = mtrr_del(dd->wc_cookie, dd->wc_base,
+			     dd->wc_len);
+		if (r < 0)
+			qib_devinfo(dd->pcidev,
+				 "mtrr_del(%lx, %lx, %lx) failed: %d\n",
+				 dd->wc_cookie, dd->wc_base,
+				 dd->wc_len, r);
+		dd->wc_cookie = 0; /* even on failure */
+	}
+}
+
+/**
+ * qib_unordered_wc - indicate whether write combining is ordered
+ *
+ * Because our performance depends on our ability to do write combining mmio
+ * writes in the most efficient way, we need to know if we are on an Intel
+ * or AMD x86_64 processor.  AMD x86_64 processors flush WC buffers out in
+ * the order completed, and so no special flushing is required to get
+ * correct ordering.  Intel processors, however, will flush write buffers
+ * out in "random" orders, and so explicit ordering is needed at times.
+ */
+int qib_unordered_wc(void)
+{
+	return boot_cpu_data.x86_vendor != X86_VENDOR_AMD;
+}
diff --git a/drivers/infiniband/hw/usnic/Kconfig b/drivers/infiniband/hw/usnic/Kconfig
new file mode 100644
index 00000000000..29ab11c34f3
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/Kconfig
@@ -0,0 +1,10 @@
+config INFINIBAND_USNIC
+	tristate "Verbs support for Cisco VIC"
+	depends on NETDEVICES && ETHERNET && INET && PCI && INTEL_IOMMU
+	select ENIC
+	select NET_VENDOR_CISCO
+	select PCI_IOV
+	select INFINIBAND_USER_ACCESS
+	---help---
+	  This is a low-level driver for Cisco's Virtual Interface
+	  Cards (VICs), including the VIC 1240 and 1280 cards.
diff --git a/drivers/infiniband/hw/usnic/Makefile b/drivers/infiniband/hw/usnic/Makefile
new file mode 100644
index 00000000000..99fb2db47cd
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/Makefile
@@ -0,0 +1,15 @@
+ccflags-y := -Idrivers/net/ethernet/cisco/enic
+
+obj-$(CONFIG_INFINIBAND_USNIC)+= usnic_verbs.o
+
+usnic_verbs-y=\
+usnic_fwd.o \
+usnic_transport.o \
+usnic_uiom.o \
+usnic_uiom_interval_tree.o \
+usnic_vnic.o \
+usnic_ib_main.o \
+usnic_ib_qp_grp.o \
+usnic_ib_sysfs.o \
+usnic_ib_verbs.o \
+usnic_debugfs.o \
diff --git a/drivers/infiniband/hw/usnic/usnic.h b/drivers/infiniband/hw/usnic/usnic.h
new file mode 100644
index 00000000000..5be13d8991b
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_H_
+#define USNIC_H_
+
+#define DRV_NAME	"usnic_verbs"
+
+#define PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC	0x00cf	/* User space NIC */
+
+#define DRV_VERSION    "1.0.3"
+#define DRV_RELDATE    "December 19, 2013"
+
+#endif /* USNIC_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_abi.h b/drivers/infiniband/hw/usnic/usnic_abi.h
new file mode 100644
index 00000000000..04a66229584
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_abi.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+
+#ifndef USNIC_ABI_H
+#define USNIC_ABI_H
+
+/* ABI between userspace and kernel */
+#define USNIC_UVERBS_ABI_VERSION	4
+
+#define USNIC_QP_GRP_MAX_WQS		8
+#define USNIC_QP_GRP_MAX_RQS		8
+#define USNIC_QP_GRP_MAX_CQS		16
+
+enum usnic_transport_type {
+	USNIC_TRANSPORT_UNKNOWN		= 0,
+	USNIC_TRANSPORT_ROCE_CUSTOM	= 1,
+	USNIC_TRANSPORT_IPV4_UDP	= 2,
+	USNIC_TRANSPORT_MAX		= 3,
+};
+
+struct usnic_transport_spec {
+	enum usnic_transport_type	trans_type;
+	union {
+		struct {
+			uint16_t	port_num;
+		} usnic_roce;
+		struct {
+			uint32_t	sock_fd;
+		} udp;
+	};
+};
+
+struct usnic_ib_create_qp_cmd {
+	struct usnic_transport_spec	spec;
+};
+
+/*TODO: Future - usnic_modify_qp needs to pass in generic filters */
+struct usnic_ib_create_qp_resp {
+	u32				vfid;
+	u32				qp_grp_id;
+	u64				bar_bus_addr;
+	u32				bar_len;
+/*
+ * WQ, RQ, CQ are explicity specified bc exposing a generic resources inteface
+ * expands the scope of ABI to many files.
+ */
+	u32				wq_cnt;
+	u32				rq_cnt;
+	u32				cq_cnt;
+	u32				wq_idx[USNIC_QP_GRP_MAX_WQS];
+	u32				rq_idx[USNIC_QP_GRP_MAX_RQS];
+	u32				cq_idx[USNIC_QP_GRP_MAX_CQS];
+	u32				transport;
+	u32				reserved[9];
+};
+
+#endif /* USNIC_ABI_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h b/drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h
new file mode 100644
index 00000000000..39356726614
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_common_pkt_hdr.h
@@ -0,0 +1,27 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_CMN_PKT_HDR_H
+#define USNIC_CMN_PKT_HDR_H
+
+#define USNIC_ROCE_ETHERTYPE		(0x8915)
+#define USNIC_ROCE_GRH_VER              (8)
+#define USNIC_PROTO_VER                 (1)
+#define USNIC_ROCE_GRH_VER_SHIFT        (4)
+
+#endif /* USNIC_COMMON_PKT_HDR_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_common_util.h b/drivers/infiniband/hw/usnic/usnic_common_util.h
new file mode 100644
index 00000000000..9d737ed5e55
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_common_util.h
@@ -0,0 +1,68 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_CMN_UTIL_H
+#define USNIC_CMN_UTIL_H
+
+static inline void
+usnic_mac_to_gid(const char *const mac, char *raw_gid)
+{
+	raw_gid[0] = 0xfe;
+	raw_gid[1] = 0x80;
+	memset(&raw_gid[2], 0, 6);
+	raw_gid[8] = mac[0]^2;
+	raw_gid[9] = mac[1];
+	raw_gid[10] = mac[2];
+	raw_gid[11] = 0xff;
+	raw_gid[12] = 0xfe;
+	raw_gid[13] = mac[3];
+	raw_gid[14] = mac[4];
+	raw_gid[15] = mac[5];
+}
+
+static inline void
+usnic_mac_ip_to_gid(const char *const mac, const __be32 inaddr, char *raw_gid)
+{
+	raw_gid[0] = 0xfe;
+	raw_gid[1] = 0x80;
+	memset(&raw_gid[2], 0, 2);
+	memcpy(&raw_gid[4], &inaddr, 4);
+	raw_gid[8] = mac[0]^2;
+	raw_gid[9] = mac[1];
+	raw_gid[10] = mac[2];
+	raw_gid[11] = 0xff;
+	raw_gid[12] = 0xfe;
+	raw_gid[13] = mac[3];
+	raw_gid[14] = mac[4];
+	raw_gid[15] = mac[5];
+}
+
+static inline void
+usnic_write_gid_if_id_from_mac(char *mac, char *raw_gid)
+{
+	raw_gid[8] = mac[0]^2;
+	raw_gid[9] = mac[1];
+	raw_gid[10] = mac[2];
+	raw_gid[11] = 0xff;
+	raw_gid[12] = 0xfe;
+	raw_gid[13] = mac[3];
+	raw_gid[14] = mac[4];
+	raw_gid[15] = mac[5];
+}
+
+#endif /* USNIC_COMMON_UTIL_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_debugfs.c b/drivers/infiniband/hw/usnic/usnic_debugfs.c
new file mode 100644
index 00000000000..5d13860161a
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_debugfs.c
@@ -0,0 +1,154 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/debugfs.h>
+#include <linux/module.h>
+
+#include "usnic.h"
+#include "usnic_log.h"
+#include "usnic_debugfs.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_transport.h"
+
+static struct dentry *debugfs_root;
+static struct dentry *flows_dentry;
+
+static ssize_t usnic_debugfs_buildinfo_read(struct file *f, char __user *data,
+						size_t count, loff_t *ppos)
+{
+	char buf[500];
+	int res;
+
+	if (*ppos > 0)
+		return 0;
+
+	res = scnprintf(buf, sizeof(buf),
+			"version:       %s\n"
+			"build date:    %s\n",
+			DRV_VERSION, DRV_RELDATE);
+
+	return simple_read_from_buffer(data, count, ppos, buf, res);
+}
+
+static const struct file_operations usnic_debugfs_buildinfo_ops = {
+	.owner = THIS_MODULE,
+	.open = simple_open,
+	.read = usnic_debugfs_buildinfo_read
+};
+
+static ssize_t flowinfo_read(struct file *f, char __user *data,
+				size_t count, loff_t *ppos)
+{
+	struct usnic_ib_qp_grp_flow *qp_flow;
+	int n;
+	int left;
+	char *ptr;
+	char buf[512];
+
+	qp_flow = f->private_data;
+	ptr = buf;
+	left = count;
+
+	if (*ppos > 0)
+		return 0;
+
+	spin_lock(&qp_flow->qp_grp->lock);
+	n = scnprintf(ptr, left,
+			"QP Grp ID: %d Transport: %s ",
+			qp_flow->qp_grp->grp_id,
+			usnic_transport_to_str(qp_flow->trans_type));
+	UPDATE_PTR_LEFT(n, ptr, left);
+	if (qp_flow->trans_type == USNIC_TRANSPORT_ROCE_CUSTOM) {
+		n = scnprintf(ptr, left, "Port_Num:%hu\n",
+					qp_flow->usnic_roce.port_num);
+		UPDATE_PTR_LEFT(n, ptr, left);
+	} else if (qp_flow->trans_type == USNIC_TRANSPORT_IPV4_UDP) {
+		n = usnic_transport_sock_to_str(ptr, left,
+				qp_flow->udp.sock);
+		UPDATE_PTR_LEFT(n, ptr, left);
+		n = scnprintf(ptr, left, "\n");
+		UPDATE_PTR_LEFT(n, ptr, left);
+	}
+	spin_unlock(&qp_flow->qp_grp->lock);
+
+	return simple_read_from_buffer(data, count, ppos, buf, ptr - buf);
+}
+
+static const struct file_operations flowinfo_ops = {
+	.owner = THIS_MODULE,
+	.open = simple_open,
+	.read = flowinfo_read,
+};
+
+void usnic_debugfs_init(void)
+{
+	debugfs_root = debugfs_create_dir(DRV_NAME, NULL);
+	if (IS_ERR(debugfs_root)) {
+		usnic_err("Failed to create debugfs root dir, check if debugfs is enabled in kernel configuration\n");
+		goto out_clear_root;
+	}
+
+	flows_dentry = debugfs_create_dir("flows", debugfs_root);
+	if (IS_ERR_OR_NULL(flows_dentry)) {
+		usnic_err("Failed to create debugfs flow dir with err %ld\n",
+				PTR_ERR(flows_dentry));
+		goto out_free_root;
+	}
+
+	debugfs_create_file("build-info", S_IRUGO, debugfs_root,
+				NULL, &usnic_debugfs_buildinfo_ops);
+	return;
+
+out_free_root:
+	debugfs_remove_recursive(debugfs_root);
+out_clear_root:
+	debugfs_root = NULL;
+}
+
+void usnic_debugfs_exit(void)
+{
+	if (!debugfs_root)
+		return;
+
+	debugfs_remove_recursive(debugfs_root);
+	debugfs_root = NULL;
+}
+
+void usnic_debugfs_flow_add(struct usnic_ib_qp_grp_flow *qp_flow)
+{
+	if (IS_ERR_OR_NULL(flows_dentry))
+		return;
+
+	scnprintf(qp_flow->dentry_name, sizeof(qp_flow->dentry_name),
+			"%u", qp_flow->flow->flow_id);
+	qp_flow->dbgfs_dentry = debugfs_create_file(qp_flow->dentry_name,
+							S_IRUGO,
+							flows_dentry,
+							qp_flow,
+							&flowinfo_ops);
+	if (IS_ERR_OR_NULL(qp_flow->dbgfs_dentry)) {
+		usnic_err("Failed to create dbg fs entry for flow %u\n",
+				qp_flow->flow->flow_id);
+	}
+}
+
+void usnic_debugfs_flow_remove(struct usnic_ib_qp_grp_flow *qp_flow)
+{
+	if (!IS_ERR_OR_NULL(qp_flow->dbgfs_dentry))
+		debugfs_remove(qp_flow->dbgfs_dentry);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_debugfs.h b/drivers/infiniband/hw/usnic/usnic_debugfs.h
new file mode 100644
index 00000000000..4087d24a88f
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_debugfs.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#ifndef USNIC_DEBUGFS_H_
+#define USNIC_DEBUGFS_H_
+
+#include "usnic_ib_qp_grp.h"
+
+void usnic_debugfs_init(void);
+
+void usnic_debugfs_exit(void);
+void usnic_debugfs_flow_add(struct usnic_ib_qp_grp_flow *qp_flow);
+void usnic_debugfs_flow_remove(struct usnic_ib_qp_grp_flow *qp_flow);
+
+#endif /*!USNIC_DEBUGFS_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_fwd.c b/drivers/infiniband/hw/usnic/usnic_fwd.c
new file mode 100644
index 00000000000..e3c9bd9d3ba
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_fwd.c
@@ -0,0 +1,350 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+
+#include "enic_api.h"
+#include "usnic_common_pkt_hdr.h"
+#include "usnic_fwd.h"
+#include "usnic_log.h"
+
+static int usnic_fwd_devcmd_locked(struct usnic_fwd_dev *ufdev, int vnic_idx,
+					enum vnic_devcmd_cmd cmd, u64 *a0,
+					u64 *a1)
+{
+	int status;
+	struct net_device *netdev = ufdev->netdev;
+
+	lockdep_assert_held(&ufdev->lock);
+
+	status = enic_api_devcmd_proxy_by_index(netdev,
+			vnic_idx,
+			cmd,
+			a0, a1,
+			1000);
+	if (status) {
+		if (status == ERR_EINVAL && cmd == CMD_DEL_FILTER) {
+			usnic_dbg("Dev %s vnic idx %u cmd %u already deleted",
+					ufdev->name, vnic_idx, cmd);
+		} else {
+			usnic_err("Dev %s vnic idx %u cmd %u failed with status %d\n",
+					ufdev->name, vnic_idx, cmd,
+					status);
+		}
+	} else {
+		usnic_dbg("Dev %s vnic idx %u cmd %u success",
+				ufdev->name, vnic_idx, cmd);
+	}
+
+	return status;
+}
+
+static int usnic_fwd_devcmd(struct usnic_fwd_dev *ufdev, int vnic_idx,
+				enum vnic_devcmd_cmd cmd, u64 *a0, u64 *a1)
+{
+	int status;
+
+	spin_lock(&ufdev->lock);
+	status = usnic_fwd_devcmd_locked(ufdev, vnic_idx, cmd, a0, a1);
+	spin_unlock(&ufdev->lock);
+
+	return status;
+}
+
+struct usnic_fwd_dev *usnic_fwd_dev_alloc(struct pci_dev *pdev)
+{
+	struct usnic_fwd_dev *ufdev;
+
+	ufdev = kzalloc(sizeof(*ufdev), GFP_KERNEL);
+	if (!ufdev)
+		return NULL;
+
+	ufdev->pdev = pdev;
+	ufdev->netdev = pci_get_drvdata(pdev);
+	spin_lock_init(&ufdev->lock);
+	strncpy(ufdev->name, netdev_name(ufdev->netdev),
+			sizeof(ufdev->name) - 1);
+
+	return ufdev;
+}
+
+void usnic_fwd_dev_free(struct usnic_fwd_dev *ufdev)
+{
+	kfree(ufdev);
+}
+
+void usnic_fwd_set_mac(struct usnic_fwd_dev *ufdev, char mac[ETH_ALEN])
+{
+	spin_lock(&ufdev->lock);
+	memcpy(&ufdev->mac, mac, sizeof(ufdev->mac));
+	spin_unlock(&ufdev->lock);
+}
+
+int usnic_fwd_add_ipaddr(struct usnic_fwd_dev *ufdev, __be32 inaddr)
+{
+	int status;
+
+	spin_lock(&ufdev->lock);
+	if (ufdev->inaddr == 0) {
+		ufdev->inaddr = inaddr;
+		status = 0;
+	} else {
+		status = -EFAULT;
+	}
+	spin_unlock(&ufdev->lock);
+
+	return status;
+}
+
+void usnic_fwd_del_ipaddr(struct usnic_fwd_dev *ufdev)
+{
+	spin_lock(&ufdev->lock);
+	ufdev->inaddr = 0;
+	spin_unlock(&ufdev->lock);
+}
+
+void usnic_fwd_carrier_up(struct usnic_fwd_dev *ufdev)
+{
+	spin_lock(&ufdev->lock);
+	ufdev->link_up = 1;
+	spin_unlock(&ufdev->lock);
+}
+
+void usnic_fwd_carrier_down(struct usnic_fwd_dev *ufdev)
+{
+	spin_lock(&ufdev->lock);
+	ufdev->link_up = 0;
+	spin_unlock(&ufdev->lock);
+}
+
+void usnic_fwd_set_mtu(struct usnic_fwd_dev *ufdev, unsigned int mtu)
+{
+	spin_lock(&ufdev->lock);
+	ufdev->mtu = mtu;
+	spin_unlock(&ufdev->lock);
+}
+
+static int usnic_fwd_dev_ready_locked(struct usnic_fwd_dev *ufdev)
+{
+	lockdep_assert_held(&ufdev->lock);
+
+	if (!ufdev->link_up)
+		return -EPERM;
+
+	return 0;
+}
+
+static int validate_filter_locked(struct usnic_fwd_dev *ufdev,
+					struct filter *filter)
+{
+
+	lockdep_assert_held(&ufdev->lock);
+
+	if (filter->type == FILTER_IPV4_5TUPLE) {
+		if (!(filter->u.ipv4.flags & FILTER_FIELD_5TUP_DST_AD))
+			return -EACCES;
+		if (!(filter->u.ipv4.flags & FILTER_FIELD_5TUP_DST_PT))
+			return -EBUSY;
+		else if (ufdev->inaddr == 0)
+			return -EINVAL;
+		else if (filter->u.ipv4.dst_port == 0)
+			return -ERANGE;
+		else if (ntohl(ufdev->inaddr) != filter->u.ipv4.dst_addr)
+			return -EFAULT;
+		else
+			return 0;
+	}
+
+	return 0;
+}
+
+static void fill_tlv(struct filter_tlv *tlv, struct filter *filter,
+		struct filter_action *action)
+{
+	tlv->type = CLSF_TLV_FILTER;
+	tlv->length = sizeof(struct filter);
+	*((struct filter *)&tlv->val) = *filter;
+
+	tlv = (struct filter_tlv *)((char *)tlv + sizeof(struct filter_tlv) +
+			sizeof(struct filter));
+	tlv->type = CLSF_TLV_ACTION;
+	tlv->length = sizeof(struct filter_action);
+	*((struct filter_action *)&tlv->val) = *action;
+}
+
+struct usnic_fwd_flow*
+usnic_fwd_alloc_flow(struct usnic_fwd_dev *ufdev, struct filter *filter,
+				struct usnic_filter_action *uaction)
+{
+	struct filter_tlv *tlv;
+	struct pci_dev *pdev;
+	struct usnic_fwd_flow *flow;
+	uint64_t a0, a1;
+	uint64_t tlv_size;
+	dma_addr_t tlv_pa;
+	int status;
+
+	pdev = ufdev->pdev;
+	tlv_size = (2*sizeof(struct filter_tlv) + sizeof(struct filter) +
+			sizeof(struct filter_action));
+
+	flow = kzalloc(sizeof(*flow), GFP_ATOMIC);
+	if (!flow)
+		return ERR_PTR(-ENOMEM);
+
+	tlv = pci_alloc_consistent(pdev, tlv_size, &tlv_pa);
+	if (!tlv) {
+		usnic_err("Failed to allocate memory\n");
+		status = -ENOMEM;
+		goto out_free_flow;
+	}
+
+	fill_tlv(tlv, filter, &uaction->action);
+
+	spin_lock(&ufdev->lock);
+	status = usnic_fwd_dev_ready_locked(ufdev);
+	if (status) {
+		usnic_err("Forwarding dev %s not ready with status %d\n",
+				ufdev->name, status);
+		goto out_free_tlv;
+	}
+
+	status = validate_filter_locked(ufdev, filter);
+	if (status) {
+		usnic_err("Failed to validate filter with status %d\n",
+				status);
+		goto out_free_tlv;
+	}
+
+	/* Issue Devcmd */
+	a0 = tlv_pa;
+	a1 = tlv_size;
+	status = usnic_fwd_devcmd_locked(ufdev, uaction->vnic_idx,
+						CMD_ADD_FILTER, &a0, &a1);
+	if (status) {
+		usnic_err("VF %s Filter add failed with status:%d",
+				ufdev->name, status);
+		status = -EFAULT;
+		goto out_free_tlv;
+	} else {
+		usnic_dbg("VF %s FILTER ID:%llu", ufdev->name, a0);
+	}
+
+	flow->flow_id = (uint32_t) a0;
+	flow->vnic_idx = uaction->vnic_idx;
+	flow->ufdev = ufdev;
+
+out_free_tlv:
+	spin_unlock(&ufdev->lock);
+	pci_free_consistent(pdev, tlv_size, tlv, tlv_pa);
+	if (!status)
+		return flow;
+out_free_flow:
+	kfree(flow);
+	return ERR_PTR(status);
+}
+
+int usnic_fwd_dealloc_flow(struct usnic_fwd_flow *flow)
+{
+	int status;
+	u64 a0, a1;
+
+	a0 = flow->flow_id;
+
+	status = usnic_fwd_devcmd(flow->ufdev, flow->vnic_idx,
+					CMD_DEL_FILTER, &a0, &a1);
+	if (status) {
+		if (status == ERR_EINVAL) {
+			usnic_dbg("Filter %u already deleted for VF Idx %u pf: %s status: %d",
+					flow->flow_id, flow->vnic_idx,
+					flow->ufdev->name, status);
+		} else {
+			usnic_err("PF %s VF Idx %u Filter: %u FILTER DELETE failed with status %d",
+					flow->ufdev->name, flow->vnic_idx,
+					flow->flow_id, status);
+		}
+		status = 0;
+		/*
+		 * Log the error and fake success to the caller because if
+		 * a flow fails to be deleted in the firmware, it is an
+		 * unrecoverable error.
+		 */
+	} else {
+		usnic_dbg("PF %s VF Idx %u Filter: %u FILTER DELETED",
+				flow->ufdev->name, flow->vnic_idx,
+				flow->flow_id);
+	}
+
+	kfree(flow);
+	return status;
+}
+
+int usnic_fwd_enable_qp(struct usnic_fwd_dev *ufdev, int vnic_idx, int qp_idx)
+{
+	int status;
+	struct net_device *pf_netdev;
+	u64 a0, a1;
+
+	pf_netdev = ufdev->netdev;
+	a0 = qp_idx;
+	a1 = CMD_QP_RQWQ;
+
+	status = usnic_fwd_devcmd(ufdev, vnic_idx, CMD_QP_ENABLE,
+						&a0, &a1);
+	if (status) {
+		usnic_err("PF %s VNIC Index %u RQ Index: %u ENABLE Failed with status %d",
+				netdev_name(pf_netdev),
+				vnic_idx,
+				qp_idx,
+				status);
+	} else {
+		usnic_dbg("PF %s VNIC Index %u RQ Index: %u ENABLED",
+				netdev_name(pf_netdev),
+				vnic_idx, qp_idx);
+	}
+
+	return status;
+}
+
+int usnic_fwd_disable_qp(struct usnic_fwd_dev *ufdev, int vnic_idx, int qp_idx)
+{
+	int status;
+	u64 a0, a1;
+	struct net_device *pf_netdev;
+
+	pf_netdev = ufdev->netdev;
+	a0 = qp_idx;
+	a1 = CMD_QP_RQWQ;
+
+	status = usnic_fwd_devcmd(ufdev, vnic_idx, CMD_QP_DISABLE,
+			&a0, &a1);
+	if (status) {
+		usnic_err("PF %s VNIC Index %u RQ Index: %u DISABLE Failed with status %d",
+				netdev_name(pf_netdev),
+				vnic_idx,
+				qp_idx,
+				status);
+	} else {
+		usnic_dbg("PF %s VNIC Index %u RQ Index: %u DISABLED",
+				netdev_name(pf_netdev),
+				vnic_idx,
+				qp_idx);
+	}
+
+	return status;
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_fwd.h b/drivers/infiniband/hw/usnic/usnic_fwd.h
new file mode 100644
index 00000000000..93713a2230b
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_fwd.h
@@ -0,0 +1,113 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_FWD_H_
+#define USNIC_FWD_H_
+
+#include <linux/if.h>
+#include <linux/netdevice.h>
+#include <linux/pci.h>
+#include <linux/in.h>
+
+#include "usnic_abi.h"
+#include "usnic_common_pkt_hdr.h"
+#include "vnic_devcmd.h"
+
+struct usnic_fwd_dev {
+	struct pci_dev			*pdev;
+	struct net_device		*netdev;
+	spinlock_t			lock;
+	/*
+	 * The following fields can be read directly off the device.
+	 * However, they should be set by a accessor function, except name,
+	 * which cannot be changed.
+	 */
+	bool				link_up;
+	char				mac[ETH_ALEN];
+	unsigned int			mtu;
+	__be32				inaddr;
+	char				name[IFNAMSIZ+1];
+};
+
+struct usnic_fwd_flow {
+	uint32_t			flow_id;
+	struct usnic_fwd_dev		*ufdev;
+	unsigned int			vnic_idx;
+};
+
+struct usnic_filter_action {
+	int				vnic_idx;
+	struct filter_action		action;
+};
+
+struct usnic_fwd_dev *usnic_fwd_dev_alloc(struct pci_dev *pdev);
+void usnic_fwd_dev_free(struct usnic_fwd_dev *ufdev);
+
+void usnic_fwd_set_mac(struct usnic_fwd_dev *ufdev, char mac[ETH_ALEN]);
+int usnic_fwd_add_ipaddr(struct usnic_fwd_dev *ufdev, __be32 inaddr);
+void usnic_fwd_del_ipaddr(struct usnic_fwd_dev *ufdev);
+void usnic_fwd_carrier_up(struct usnic_fwd_dev *ufdev);
+void usnic_fwd_carrier_down(struct usnic_fwd_dev *ufdev);
+void usnic_fwd_set_mtu(struct usnic_fwd_dev *ufdev, unsigned int mtu);
+
+/*
+ * Allocate a flow on this forwarding device. Whoever calls this function,
+ * must monitor netdev events on ufdev's netdevice. If NETDEV_REBOOT or
+ * NETDEV_DOWN is seen, flow will no longer function and must be
+ * immediately freed by calling usnic_dealloc_flow.
+ */
+struct usnic_fwd_flow*
+usnic_fwd_alloc_flow(struct usnic_fwd_dev *ufdev, struct filter *filter,
+				struct usnic_filter_action *action);
+int usnic_fwd_dealloc_flow(struct usnic_fwd_flow *flow);
+int usnic_fwd_enable_qp(struct usnic_fwd_dev *ufdev, int vnic_idx, int qp_idx);
+int usnic_fwd_disable_qp(struct usnic_fwd_dev *ufdev, int vnic_idx, int qp_idx);
+
+static inline void usnic_fwd_init_usnic_filter(struct filter *filter,
+						uint32_t usnic_id)
+{
+	filter->type = FILTER_USNIC_ID;
+	filter->u.usnic.ethtype = USNIC_ROCE_ETHERTYPE;
+	filter->u.usnic.flags = FILTER_FIELD_USNIC_ETHTYPE |
+				FILTER_FIELD_USNIC_ID |
+				FILTER_FIELD_USNIC_PROTO;
+	filter->u.usnic.proto_version = (USNIC_ROCE_GRH_VER <<
+					 USNIC_ROCE_GRH_VER_SHIFT) |
+					 USNIC_PROTO_VER;
+	filter->u.usnic.usnic_id = usnic_id;
+}
+
+static inline void usnic_fwd_init_udp_filter(struct filter *filter,
+						uint32_t daddr, uint16_t dport)
+{
+	filter->type = FILTER_IPV4_5TUPLE;
+	filter->u.ipv4.flags = FILTER_FIELD_5TUP_PROTO;
+	filter->u.ipv4.protocol = PROTO_UDP;
+
+	if (daddr) {
+		filter->u.ipv4.flags |= FILTER_FIELD_5TUP_DST_AD;
+		filter->u.ipv4.dst_addr = daddr;
+	}
+
+	if (dport) {
+		filter->u.ipv4.flags |= FILTER_FIELD_5TUP_DST_PT;
+		filter->u.ipv4.dst_port = dport;
+	}
+}
+
+#endif /* !USNIC_FWD_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib.h b/drivers/infiniband/hw/usnic/usnic_ib.h
new file mode 100644
index 00000000000..e5a9297dd1b
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib.h
@@ -0,0 +1,118 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_H_
+#define USNIC_IB_H_
+
+#include <linux/iommu.h>
+#include <linux/netdevice.h>
+
+#include <rdma/ib_verbs.h>
+
+
+#include "usnic.h"
+#include "usnic_abi.h"
+#include "usnic_vnic.h"
+
+#define USNIC_IB_PORT_CNT		1
+#define USNIC_IB_NUM_COMP_VECTORS	1
+
+extern unsigned int usnic_ib_share_vf;
+
+struct usnic_ib_ucontext {
+	struct ib_ucontext		ibucontext;
+	/* Protected by usnic_ib_dev->usdev_lock */
+	struct list_head		qp_grp_list;
+	struct list_head		link;
+};
+
+struct usnic_ib_pd {
+	struct ib_pd			ibpd;
+	struct usnic_uiom_pd		*umem_pd;
+};
+
+struct usnic_ib_mr {
+	struct ib_mr			ibmr;
+	struct usnic_uiom_reg		*umem;
+};
+
+struct usnic_ib_dev {
+	struct ib_device		ib_dev;
+	struct pci_dev			*pdev;
+	struct net_device		*netdev;
+	struct usnic_fwd_dev		*ufdev;
+	struct list_head		ib_dev_link;
+	struct list_head		vf_dev_list;
+	struct list_head		ctx_list;
+	struct mutex			usdev_lock;
+
+	/* provisioning information */
+	struct kref			vf_cnt;
+	unsigned int			vf_res_cnt[USNIC_VNIC_RES_TYPE_MAX];
+
+	/* sysfs vars for QPN reporting */
+	struct kobject *qpn_kobj;
+};
+
+struct usnic_ib_vf {
+	struct usnic_ib_dev		*pf;
+	spinlock_t			lock;
+	struct usnic_vnic		*vnic;
+	unsigned int			qp_grp_ref_cnt;
+	struct usnic_ib_pd		*pd;
+	struct list_head		link;
+};
+
+static inline
+struct usnic_ib_dev *to_usdev(struct ib_device *ibdev)
+{
+	return container_of(ibdev, struct usnic_ib_dev, ib_dev);
+}
+
+static inline
+struct usnic_ib_ucontext *to_ucontext(struct ib_ucontext *ibucontext)
+{
+	return container_of(ibucontext, struct usnic_ib_ucontext, ibucontext);
+}
+
+static inline
+struct usnic_ib_pd *to_upd(struct ib_pd *ibpd)
+{
+	return container_of(ibpd, struct usnic_ib_pd, ibpd);
+}
+
+static inline
+struct usnic_ib_ucontext *to_uucontext(struct ib_ucontext *ibucontext)
+{
+	return container_of(ibucontext, struct usnic_ib_ucontext, ibucontext);
+}
+
+static inline
+struct usnic_ib_mr *to_umr(struct ib_mr *ibmr)
+{
+	return container_of(ibmr, struct usnic_ib_mr, ibmr);
+}
+void usnic_ib_log_vf(struct usnic_ib_vf *vf);
+
+#define UPDATE_PTR_LEFT(N, P, L)			\
+do {							\
+	L -= (N);					\
+	P += (N);					\
+} while (0)
+
+#endif /* USNIC_IB_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_main.c b/drivers/infiniband/hw/usnic/usnic_ib_main.c
new file mode 100644
index 00000000000..fb6d026f92c
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_main.c
@@ -0,0 +1,682 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ * Author: Upinder Malhi <umalhi@cisco.com>
+ * Author: Anant Deepak <anadeepa@cisco.com>
+ * Author: Cesare Cantu' <cantuc@cisco.com>
+ * Author: Jeff Squyres <jsquyres@cisco.com>
+ * Author: Kiran Thirumalai <kithirum@cisco.com>
+ * Author: Xuyang Wang <xuywang@cisco.com>
+ * Author: Reese Faucette <rfaucett@cisco.com>
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/inetdevice.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+#include <linux/pci.h>
+#include <linux/netdevice.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_abi.h"
+#include "usnic_common_util.h"
+#include "usnic_ib.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_log.h"
+#include "usnic_fwd.h"
+#include "usnic_debugfs.h"
+#include "usnic_ib_verbs.h"
+#include "usnic_transport.h"
+#include "usnic_uiom.h"
+#include "usnic_ib_sysfs.h"
+
+unsigned int usnic_log_lvl = USNIC_LOG_LVL_ERR;
+unsigned int usnic_ib_share_vf = 1;
+
+static const char usnic_version[] =
+	DRV_NAME ": Cisco VIC (USNIC) Verbs Driver v"
+	DRV_VERSION " (" DRV_RELDATE ")\n";
+
+static DEFINE_MUTEX(usnic_ib_ibdev_list_lock);
+static LIST_HEAD(usnic_ib_ibdev_list);
+
+/* Callback dump funcs */
+static int usnic_ib_dump_vf_hdr(void *obj, char *buf, int buf_sz)
+{
+	struct usnic_ib_vf *vf = obj;
+	return scnprintf(buf, buf_sz, "PF: %s ", vf->pf->ib_dev.name);
+}
+/* End callback dump funcs */
+
+static void usnic_ib_dump_vf(struct usnic_ib_vf *vf, char *buf, int buf_sz)
+{
+	usnic_vnic_dump(vf->vnic, buf, buf_sz, vf,
+			usnic_ib_dump_vf_hdr,
+			usnic_ib_qp_grp_dump_hdr, usnic_ib_qp_grp_dump_rows);
+}
+
+void usnic_ib_log_vf(struct usnic_ib_vf *vf)
+{
+	char buf[1000];
+	usnic_ib_dump_vf(vf, buf, sizeof(buf));
+	usnic_dbg("%s\n", buf);
+}
+
+/* Start of netdev section */
+static inline const char *usnic_ib_netdev_event_to_string(unsigned long event)
+{
+	const char *event2str[] = {"NETDEV_NONE", "NETDEV_UP", "NETDEV_DOWN",
+		"NETDEV_REBOOT", "NETDEV_CHANGE",
+		"NETDEV_REGISTER", "NETDEV_UNREGISTER", "NETDEV_CHANGEMTU",
+		"NETDEV_CHANGEADDR", "NETDEV_GOING_DOWN", "NETDEV_FEAT_CHANGE",
+		"NETDEV_BONDING_FAILOVER", "NETDEV_PRE_UP",
+		"NETDEV_PRE_TYPE_CHANGE", "NETDEV_POST_TYPE_CHANGE",
+		"NETDEV_POST_INT", "NETDEV_UNREGISTER_FINAL", "NETDEV_RELEASE",
+		"NETDEV_NOTIFY_PEERS", "NETDEV_JOIN"
+	};
+
+	if (event >= ARRAY_SIZE(event2str))
+		return "UNKNOWN_NETDEV_EVENT";
+	else
+		return event2str[event];
+}
+
+static void usnic_ib_qp_grp_modify_active_to_err(struct usnic_ib_dev *us_ibdev)
+{
+	struct usnic_ib_ucontext *ctx;
+	struct usnic_ib_qp_grp *qp_grp;
+	enum ib_qp_state cur_state;
+	int status;
+
+	BUG_ON(!mutex_is_locked(&us_ibdev->usdev_lock));
+
+	list_for_each_entry(ctx, &us_ibdev->ctx_list, link) {
+		list_for_each_entry(qp_grp, &ctx->qp_grp_list, link) {
+			cur_state = qp_grp->state;
+			if (cur_state == IB_QPS_INIT ||
+				cur_state == IB_QPS_RTR ||
+				cur_state == IB_QPS_RTS) {
+				status = usnic_ib_qp_grp_modify(qp_grp,
+								IB_QPS_ERR,
+								NULL);
+				if (status) {
+					usnic_err("Failed to transistion qp grp %u from %s to %s\n",
+						qp_grp->grp_id,
+						usnic_ib_qp_grp_state_to_string
+						(cur_state),
+						usnic_ib_qp_grp_state_to_string
+						(IB_QPS_ERR));
+				}
+			}
+		}
+	}
+}
+
+static void usnic_ib_handle_usdev_event(struct usnic_ib_dev *us_ibdev,
+					unsigned long event)
+{
+	struct net_device *netdev;
+	struct ib_event ib_event;
+
+	memset(&ib_event, 0, sizeof(ib_event));
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	netdev = us_ibdev->netdev;
+	switch (event) {
+	case NETDEV_REBOOT:
+		usnic_info("PF Reset on %s\n", us_ibdev->ib_dev.name);
+		usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+		ib_event.event = IB_EVENT_PORT_ERR;
+		ib_event.device = &us_ibdev->ib_dev;
+		ib_event.element.port_num = 1;
+		ib_dispatch_event(&ib_event);
+		break;
+	case NETDEV_UP:
+	case NETDEV_DOWN:
+	case NETDEV_CHANGE:
+		if (!us_ibdev->ufdev->link_up &&
+				netif_carrier_ok(netdev)) {
+			usnic_fwd_carrier_up(us_ibdev->ufdev);
+			usnic_info("Link UP on %s\n", us_ibdev->ib_dev.name);
+			ib_event.event = IB_EVENT_PORT_ACTIVE;
+			ib_event.device = &us_ibdev->ib_dev;
+			ib_event.element.port_num = 1;
+			ib_dispatch_event(&ib_event);
+		} else if (us_ibdev->ufdev->link_up &&
+				!netif_carrier_ok(netdev)) {
+			usnic_fwd_carrier_down(us_ibdev->ufdev);
+			usnic_info("Link DOWN on %s\n", us_ibdev->ib_dev.name);
+			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+			ib_event.event = IB_EVENT_PORT_ERR;
+			ib_event.device = &us_ibdev->ib_dev;
+			ib_event.element.port_num = 1;
+			ib_dispatch_event(&ib_event);
+		} else {
+			usnic_dbg("Ignoring %s on %s\n",
+					usnic_ib_netdev_event_to_string(event),
+					us_ibdev->ib_dev.name);
+		}
+		break;
+	case NETDEV_CHANGEADDR:
+		if (!memcmp(us_ibdev->ufdev->mac, netdev->dev_addr,
+				sizeof(us_ibdev->ufdev->mac))) {
+			usnic_dbg("Ignoring addr change on %s\n",
+					us_ibdev->ib_dev.name);
+		} else {
+			usnic_info(" %s old mac: %pM new mac: %pM\n",
+					us_ibdev->ib_dev.name,
+					us_ibdev->ufdev->mac,
+					netdev->dev_addr);
+			usnic_fwd_set_mac(us_ibdev->ufdev, netdev->dev_addr);
+			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+			ib_event.event = IB_EVENT_GID_CHANGE;
+			ib_event.device = &us_ibdev->ib_dev;
+			ib_event.element.port_num = 1;
+			ib_dispatch_event(&ib_event);
+		}
+
+		break;
+	case NETDEV_CHANGEMTU:
+		if (us_ibdev->ufdev->mtu != netdev->mtu) {
+			usnic_info("MTU Change on %s old: %u new: %u\n",
+					us_ibdev->ib_dev.name,
+					us_ibdev->ufdev->mtu, netdev->mtu);
+			usnic_fwd_set_mtu(us_ibdev->ufdev, netdev->mtu);
+			usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+		} else {
+			usnic_dbg("Ignoring MTU change on %s\n",
+					us_ibdev->ib_dev.name);
+		}
+		break;
+	default:
+		usnic_dbg("Ignoring event %s on %s",
+				usnic_ib_netdev_event_to_string(event),
+				us_ibdev->ib_dev.name);
+	}
+	mutex_unlock(&us_ibdev->usdev_lock);
+}
+
+static int usnic_ib_netdevice_event(struct notifier_block *notifier,
+					unsigned long event, void *ptr)
+{
+	struct usnic_ib_dev *us_ibdev;
+
+	struct net_device *netdev = netdev_notifier_info_to_dev(ptr);
+
+	mutex_lock(&usnic_ib_ibdev_list_lock);
+	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
+		if (us_ibdev->netdev == netdev) {
+			usnic_ib_handle_usdev_event(us_ibdev, event);
+			break;
+		}
+	}
+	mutex_unlock(&usnic_ib_ibdev_list_lock);
+
+	return NOTIFY_DONE;
+}
+
+static struct notifier_block usnic_ib_netdevice_notifier = {
+	.notifier_call = usnic_ib_netdevice_event
+};
+/* End of netdev section */
+
+/* Start of inet section */
+static int usnic_ib_handle_inet_event(struct usnic_ib_dev *us_ibdev,
+					unsigned long event, void *ptr)
+{
+	struct in_ifaddr *ifa = ptr;
+	struct ib_event ib_event;
+
+	mutex_lock(&us_ibdev->usdev_lock);
+
+	switch (event) {
+	case NETDEV_DOWN:
+		usnic_info("%s via ip notifiers",
+				usnic_ib_netdev_event_to_string(event));
+		usnic_fwd_del_ipaddr(us_ibdev->ufdev);
+		usnic_ib_qp_grp_modify_active_to_err(us_ibdev);
+		ib_event.event = IB_EVENT_GID_CHANGE;
+		ib_event.device = &us_ibdev->ib_dev;
+		ib_event.element.port_num = 1;
+		ib_dispatch_event(&ib_event);
+		break;
+	case NETDEV_UP:
+		usnic_fwd_add_ipaddr(us_ibdev->ufdev, ifa->ifa_address);
+		usnic_info("%s via ip notifiers: ip %pI4",
+				usnic_ib_netdev_event_to_string(event),
+				&us_ibdev->ufdev->inaddr);
+		ib_event.event = IB_EVENT_GID_CHANGE;
+		ib_event.device = &us_ibdev->ib_dev;
+		ib_event.element.port_num = 1;
+		ib_dispatch_event(&ib_event);
+		break;
+	default:
+		usnic_info("Ignoring event %s on %s",
+				usnic_ib_netdev_event_to_string(event),
+				us_ibdev->ib_dev.name);
+	}
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return NOTIFY_DONE;
+}
+
+static int usnic_ib_inetaddr_event(struct notifier_block *notifier,
+					unsigned long event, void *ptr)
+{
+	struct usnic_ib_dev *us_ibdev;
+	struct in_ifaddr *ifa = ptr;
+	struct net_device *netdev = ifa->ifa_dev->dev;
+
+	mutex_lock(&usnic_ib_ibdev_list_lock);
+	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
+		if (us_ibdev->netdev == netdev) {
+			usnic_ib_handle_inet_event(us_ibdev, event, ptr);
+			break;
+		}
+	}
+	mutex_unlock(&usnic_ib_ibdev_list_lock);
+
+	return NOTIFY_DONE;
+}
+static struct notifier_block usnic_ib_inetaddr_notifier = {
+	.notifier_call = usnic_ib_inetaddr_event
+};
+/* End of inet section*/
+
+/* Start of PF discovery section */
+static void *usnic_ib_device_add(struct pci_dev *dev)
+{
+	struct usnic_ib_dev *us_ibdev;
+	union ib_gid gid;
+	struct in_ifaddr *in;
+	struct net_device *netdev;
+
+	usnic_dbg("\n");
+	netdev = pci_get_drvdata(dev);
+
+	us_ibdev = (struct usnic_ib_dev *)ib_alloc_device(sizeof(*us_ibdev));
+	if (IS_ERR_OR_NULL(us_ibdev)) {
+		usnic_err("Device %s context alloc failed\n",
+				netdev_name(pci_get_drvdata(dev)));
+		return ERR_PTR(us_ibdev ? PTR_ERR(us_ibdev) : -EFAULT);
+	}
+
+	us_ibdev->ufdev = usnic_fwd_dev_alloc(dev);
+	if (IS_ERR_OR_NULL(us_ibdev->ufdev)) {
+		usnic_err("Failed to alloc ufdev for %s with err %ld\n",
+				pci_name(dev), PTR_ERR(us_ibdev->ufdev));
+		goto err_dealloc;
+	}
+
+	mutex_init(&us_ibdev->usdev_lock);
+	INIT_LIST_HEAD(&us_ibdev->vf_dev_list);
+	INIT_LIST_HEAD(&us_ibdev->ctx_list);
+
+	us_ibdev->pdev = dev;
+	us_ibdev->netdev = pci_get_drvdata(dev);
+	us_ibdev->ib_dev.owner = THIS_MODULE;
+	us_ibdev->ib_dev.node_type = RDMA_NODE_USNIC_UDP;
+	us_ibdev->ib_dev.phys_port_cnt = USNIC_IB_PORT_CNT;
+	us_ibdev->ib_dev.num_comp_vectors = USNIC_IB_NUM_COMP_VECTORS;
+	us_ibdev->ib_dev.dma_device = &dev->dev;
+	us_ibdev->ib_dev.uverbs_abi_ver = USNIC_UVERBS_ABI_VERSION;
+	strlcpy(us_ibdev->ib_dev.name, "usnic_%d", IB_DEVICE_NAME_MAX);
+
+	us_ibdev->ib_dev.uverbs_cmd_mask =
+		(1ull << IB_USER_VERBS_CMD_GET_CONTEXT) |
+		(1ull << IB_USER_VERBS_CMD_QUERY_DEVICE) |
+		(1ull << IB_USER_VERBS_CMD_QUERY_PORT) |
+		(1ull << IB_USER_VERBS_CMD_ALLOC_PD) |
+		(1ull << IB_USER_VERBS_CMD_DEALLOC_PD) |
+		(1ull << IB_USER_VERBS_CMD_REG_MR) |
+		(1ull << IB_USER_VERBS_CMD_DEREG_MR) |
+		(1ull << IB_USER_VERBS_CMD_CREATE_COMP_CHANNEL) |
+		(1ull << IB_USER_VERBS_CMD_CREATE_CQ) |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_CQ) |
+		(1ull << IB_USER_VERBS_CMD_CREATE_QP) |
+		(1ull << IB_USER_VERBS_CMD_MODIFY_QP) |
+		(1ull << IB_USER_VERBS_CMD_QUERY_QP) |
+		(1ull << IB_USER_VERBS_CMD_DESTROY_QP) |
+		(1ull << IB_USER_VERBS_CMD_ATTACH_MCAST) |
+		(1ull << IB_USER_VERBS_CMD_DETACH_MCAST) |
+		(1ull << IB_USER_VERBS_CMD_OPEN_QP);
+
+	us_ibdev->ib_dev.query_device = usnic_ib_query_device;
+	us_ibdev->ib_dev.query_port = usnic_ib_query_port;
+	us_ibdev->ib_dev.query_pkey = usnic_ib_query_pkey;
+	us_ibdev->ib_dev.query_gid = usnic_ib_query_gid;
+	us_ibdev->ib_dev.get_link_layer = usnic_ib_port_link_layer;
+	us_ibdev->ib_dev.alloc_pd = usnic_ib_alloc_pd;
+	us_ibdev->ib_dev.dealloc_pd = usnic_ib_dealloc_pd;
+	us_ibdev->ib_dev.create_qp = usnic_ib_create_qp;
+	us_ibdev->ib_dev.modify_qp = usnic_ib_modify_qp;
+	us_ibdev->ib_dev.query_qp = usnic_ib_query_qp;
+	us_ibdev->ib_dev.destroy_qp = usnic_ib_destroy_qp;
+	us_ibdev->ib_dev.create_cq = usnic_ib_create_cq;
+	us_ibdev->ib_dev.destroy_cq = usnic_ib_destroy_cq;
+	us_ibdev->ib_dev.reg_user_mr = usnic_ib_reg_mr;
+	us_ibdev->ib_dev.dereg_mr = usnic_ib_dereg_mr;
+	us_ibdev->ib_dev.alloc_ucontext = usnic_ib_alloc_ucontext;
+	us_ibdev->ib_dev.dealloc_ucontext = usnic_ib_dealloc_ucontext;
+	us_ibdev->ib_dev.mmap = usnic_ib_mmap;
+	us_ibdev->ib_dev.create_ah = usnic_ib_create_ah;
+	us_ibdev->ib_dev.destroy_ah = usnic_ib_destroy_ah;
+	us_ibdev->ib_dev.post_send = usnic_ib_post_send;
+	us_ibdev->ib_dev.post_recv = usnic_ib_post_recv;
+	us_ibdev->ib_dev.poll_cq = usnic_ib_poll_cq;
+	us_ibdev->ib_dev.req_notify_cq = usnic_ib_req_notify_cq;
+	us_ibdev->ib_dev.get_dma_mr = usnic_ib_get_dma_mr;
+
+
+	if (ib_register_device(&us_ibdev->ib_dev, NULL))
+		goto err_fwd_dealloc;
+
+	usnic_fwd_set_mtu(us_ibdev->ufdev, us_ibdev->netdev->mtu);
+	usnic_fwd_set_mac(us_ibdev->ufdev, us_ibdev->netdev->dev_addr);
+	if (netif_carrier_ok(us_ibdev->netdev))
+		usnic_fwd_carrier_up(us_ibdev->ufdev);
+
+	in = ((struct in_device *)(netdev->ip_ptr))->ifa_list;
+	if (in != NULL)
+		usnic_fwd_add_ipaddr(us_ibdev->ufdev, in->ifa_address);
+
+	usnic_mac_ip_to_gid(us_ibdev->netdev->perm_addr,
+				us_ibdev->ufdev->inaddr, &gid.raw[0]);
+	memcpy(&us_ibdev->ib_dev.node_guid, &gid.global.interface_id,
+		sizeof(gid.global.interface_id));
+	kref_init(&us_ibdev->vf_cnt);
+
+	usnic_info("Added ibdev: %s netdev: %s with mac %pM Link: %u MTU: %u\n",
+			us_ibdev->ib_dev.name, netdev_name(us_ibdev->netdev),
+			us_ibdev->ufdev->mac, us_ibdev->ufdev->link_up,
+			us_ibdev->ufdev->mtu);
+	return us_ibdev;
+
+err_fwd_dealloc:
+	usnic_fwd_dev_free(us_ibdev->ufdev);
+err_dealloc:
+	usnic_err("failed -- deallocing device\n");
+	ib_dealloc_device(&us_ibdev->ib_dev);
+	return NULL;
+}
+
+static void usnic_ib_device_remove(struct usnic_ib_dev *us_ibdev)
+{
+	usnic_info("Unregistering %s\n", us_ibdev->ib_dev.name);
+	usnic_ib_sysfs_unregister_usdev(us_ibdev);
+	usnic_fwd_dev_free(us_ibdev->ufdev);
+	ib_unregister_device(&us_ibdev->ib_dev);
+	ib_dealloc_device(&us_ibdev->ib_dev);
+}
+
+static void usnic_ib_undiscover_pf(struct kref *kref)
+{
+	struct usnic_ib_dev *us_ibdev, *tmp;
+	struct pci_dev *dev;
+	bool found = false;
+
+	dev = container_of(kref, struct usnic_ib_dev, vf_cnt)->pdev;
+	mutex_lock(&usnic_ib_ibdev_list_lock);
+	list_for_each_entry_safe(us_ibdev, tmp,
+				&usnic_ib_ibdev_list, ib_dev_link) {
+		if (us_ibdev->pdev == dev) {
+			list_del(&us_ibdev->ib_dev_link);
+			usnic_ib_device_remove(us_ibdev);
+			found = true;
+			break;
+		}
+	}
+
+	WARN(!found, "Failed to remove PF %s\n", pci_name(dev));
+
+	mutex_unlock(&usnic_ib_ibdev_list_lock);
+}
+
+static struct usnic_ib_dev *usnic_ib_discover_pf(struct usnic_vnic *vnic)
+{
+	struct usnic_ib_dev *us_ibdev;
+	struct pci_dev *parent_pci, *vf_pci;
+	int err;
+
+	vf_pci = usnic_vnic_get_pdev(vnic);
+	parent_pci = pci_physfn(vf_pci);
+
+	BUG_ON(!parent_pci);
+
+	mutex_lock(&usnic_ib_ibdev_list_lock);
+	list_for_each_entry(us_ibdev, &usnic_ib_ibdev_list, ib_dev_link) {
+		if (us_ibdev->pdev == parent_pci) {
+			kref_get(&us_ibdev->vf_cnt);
+			goto out;
+		}
+	}
+
+	us_ibdev = usnic_ib_device_add(parent_pci);
+	if (IS_ERR_OR_NULL(us_ibdev)) {
+		us_ibdev = us_ibdev ? us_ibdev : ERR_PTR(-EFAULT);
+		goto out;
+	}
+
+	err = usnic_ib_sysfs_register_usdev(us_ibdev);
+	if (err) {
+		usnic_ib_device_remove(us_ibdev);
+		us_ibdev = ERR_PTR(err);
+		goto out;
+	}
+
+	list_add(&us_ibdev->ib_dev_link, &usnic_ib_ibdev_list);
+out:
+	mutex_unlock(&usnic_ib_ibdev_list_lock);
+	return us_ibdev;
+}
+/* End of PF discovery section */
+
+/* Start of PCI section */
+
+static DEFINE_PCI_DEVICE_TABLE(usnic_ib_pci_ids) = {
+	{PCI_DEVICE(PCI_VENDOR_ID_CISCO, PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC)},
+	{0,}
+};
+
+static int usnic_ib_pci_probe(struct pci_dev *pdev,
+				const struct pci_device_id *id)
+{
+	int err;
+	struct usnic_ib_dev *pf;
+	struct usnic_ib_vf *vf;
+	enum usnic_vnic_res_type res_type;
+
+	vf = kzalloc(sizeof(*vf), GFP_KERNEL);
+	if (!vf)
+		return -ENOMEM;
+
+	err = pci_enable_device(pdev);
+	if (err) {
+		usnic_err("Failed to enable %s with err %d\n",
+				pci_name(pdev), err);
+		goto out_clean_vf;
+	}
+
+	err = pci_request_regions(pdev, DRV_NAME);
+	if (err) {
+		usnic_err("Failed to request region for %s with err %d\n",
+				pci_name(pdev), err);
+		goto out_disable_device;
+	}
+
+	pci_set_master(pdev);
+	pci_set_drvdata(pdev, vf);
+
+	vf->vnic = usnic_vnic_alloc(pdev);
+	if (IS_ERR_OR_NULL(vf->vnic)) {
+		err = vf->vnic ? PTR_ERR(vf->vnic) : -ENOMEM;
+		usnic_err("Failed to alloc vnic for %s with err %d\n",
+				pci_name(pdev), err);
+		goto out_release_regions;
+	}
+
+	pf = usnic_ib_discover_pf(vf->vnic);
+	if (IS_ERR_OR_NULL(pf)) {
+		usnic_err("Failed to discover pf of vnic %s with err%ld\n",
+				pci_name(pdev), PTR_ERR(pf));
+		err = pf ? PTR_ERR(pf) : -EFAULT;
+		goto out_clean_vnic;
+	}
+
+	vf->pf = pf;
+	spin_lock_init(&vf->lock);
+	mutex_lock(&pf->usdev_lock);
+	list_add_tail(&vf->link, &pf->vf_dev_list);
+	/*
+	 * Save max settings (will be same for each VF, easier to re-write than
+	 * to say "if (!set) { set_values(); set=1; }
+	 */
+	for (res_type = USNIC_VNIC_RES_TYPE_EOL+1;
+			res_type < USNIC_VNIC_RES_TYPE_MAX;
+			res_type++) {
+		pf->vf_res_cnt[res_type] = usnic_vnic_res_cnt(vf->vnic,
+								res_type);
+	}
+
+	mutex_unlock(&pf->usdev_lock);
+
+	usnic_info("Registering usnic VF %s into PF %s\n", pci_name(pdev),
+			pf->ib_dev.name);
+	usnic_ib_log_vf(vf);
+	return 0;
+
+out_clean_vnic:
+	usnic_vnic_free(vf->vnic);
+out_release_regions:
+	pci_set_drvdata(pdev, NULL);
+	pci_clear_master(pdev);
+	pci_release_regions(pdev);
+out_disable_device:
+	pci_disable_device(pdev);
+out_clean_vf:
+	kfree(vf);
+	return err;
+}
+
+static void usnic_ib_pci_remove(struct pci_dev *pdev)
+{
+	struct usnic_ib_vf *vf = pci_get_drvdata(pdev);
+	struct usnic_ib_dev *pf = vf->pf;
+
+	mutex_lock(&pf->usdev_lock);
+	list_del(&vf->link);
+	mutex_unlock(&pf->usdev_lock);
+
+	kref_put(&pf->vf_cnt, usnic_ib_undiscover_pf);
+	usnic_vnic_free(vf->vnic);
+	pci_set_drvdata(pdev, NULL);
+	pci_clear_master(pdev);
+	pci_release_regions(pdev);
+	pci_disable_device(pdev);
+	kfree(vf);
+
+	usnic_info("Removed VF %s\n", pci_name(pdev));
+}
+
+/* PCI driver entry points */
+static struct pci_driver usnic_ib_pci_driver = {
+	.name = DRV_NAME,
+	.id_table = usnic_ib_pci_ids,
+	.probe = usnic_ib_pci_probe,
+	.remove = usnic_ib_pci_remove,
+};
+/* End of PCI section */
+
+/* Start of module section */
+static int __init usnic_ib_init(void)
+{
+	int err;
+
+	printk_once(KERN_INFO "%s", usnic_version);
+
+	err = usnic_uiom_init(DRV_NAME);
+	if (err) {
+		usnic_err("Unable to initalize umem with err %d\n", err);
+		return err;
+	}
+
+	if (pci_register_driver(&usnic_ib_pci_driver)) {
+		usnic_err("Unable to register with PCI\n");
+		goto out_umem_fini;
+	}
+
+	err = register_netdevice_notifier(&usnic_ib_netdevice_notifier);
+	if (err) {
+		usnic_err("Failed to register netdev notifier\n");
+		goto out_pci_unreg;
+	}
+
+	err = register_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
+	if (err) {
+		usnic_err("Failed to register inet addr notifier\n");
+		goto out_unreg_netdev_notifier;
+	}
+
+	err = usnic_transport_init();
+	if (err) {
+		usnic_err("Failed to initialize transport\n");
+		goto out_unreg_inetaddr_notifier;
+	}
+
+	usnic_debugfs_init();
+
+	return 0;
+
+out_unreg_inetaddr_notifier:
+	unregister_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
+out_unreg_netdev_notifier:
+	unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
+out_pci_unreg:
+	pci_unregister_driver(&usnic_ib_pci_driver);
+out_umem_fini:
+	usnic_uiom_fini();
+
+	return err;
+}
+
+static void __exit usnic_ib_destroy(void)
+{
+	usnic_dbg("\n");
+	usnic_debugfs_exit();
+	usnic_transport_fini();
+	unregister_inetaddr_notifier(&usnic_ib_inetaddr_notifier);
+	unregister_netdevice_notifier(&usnic_ib_netdevice_notifier);
+	pci_unregister_driver(&usnic_ib_pci_driver);
+	usnic_uiom_fini();
+}
+
+MODULE_DESCRIPTION("Cisco VIC (usNIC) Verbs Driver");
+MODULE_AUTHOR("Upinder Malhi <umalhi@cisco.com>");
+MODULE_LICENSE("Dual BSD/GPL");
+MODULE_VERSION(DRV_VERSION);
+module_param(usnic_log_lvl, uint, S_IRUGO | S_IWUSR);
+module_param(usnic_ib_share_vf, uint, S_IRUGO | S_IWUSR);
+MODULE_PARM_DESC(usnic_log_lvl, " Off=0, Err=1, Info=2, Debug=3");
+MODULE_PARM_DESC(usnic_ib_share_vf, "Off=0, On=1 VF sharing amongst QPs");
+MODULE_DEVICE_TABLE(pci, usnic_ib_pci_ids);
+
+module_init(usnic_ib_init);
+module_exit(usnic_ib_destroy);
+/* End of module section */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c
new file mode 100644
index 00000000000..f8dfd76be89
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.c
@@ -0,0 +1,761 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/bug.h>
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/spinlock.h>
+
+#include "usnic_log.h"
+#include "usnic_vnic.h"
+#include "usnic_fwd.h"
+#include "usnic_uiom.h"
+#include "usnic_debugfs.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_ib_sysfs.h"
+#include "usnic_transport.h"
+
+#define DFLT_RQ_IDX	0
+
+const char *usnic_ib_qp_grp_state_to_string(enum ib_qp_state state)
+{
+	switch (state) {
+	case IB_QPS_RESET:
+		return "Rst";
+	case IB_QPS_INIT:
+		return "Init";
+	case IB_QPS_RTR:
+		return "RTR";
+	case IB_QPS_RTS:
+		return "RTS";
+	case IB_QPS_SQD:
+		return "SQD";
+	case IB_QPS_SQE:
+		return "SQE";
+	case IB_QPS_ERR:
+		return "ERR";
+	default:
+		return "UNKOWN STATE";
+
+	}
+}
+
+int usnic_ib_qp_grp_dump_hdr(char *buf, int buf_sz)
+{
+	return scnprintf(buf, buf_sz, "|QPN\t|State\t|PID\t|VF Idx\t|Fil ID");
+}
+
+int usnic_ib_qp_grp_dump_rows(void *obj, char *buf, int buf_sz)
+{
+	struct usnic_ib_qp_grp *qp_grp = obj;
+	struct usnic_ib_qp_grp_flow *default_flow;
+	if (obj) {
+		default_flow = list_first_entry(&qp_grp->flows_lst,
+					struct usnic_ib_qp_grp_flow, link);
+		return scnprintf(buf, buf_sz, "|%d\t|%s\t|%d\t|%hu\t|%d",
+					qp_grp->ibqp.qp_num,
+					usnic_ib_qp_grp_state_to_string(
+							qp_grp->state),
+					qp_grp->owner_pid,
+					usnic_vnic_get_index(qp_grp->vf->vnic),
+					default_flow->flow->flow_id);
+	} else {
+		return scnprintf(buf, buf_sz, "|N/A\t|N/A\t|N/A\t|N/A\t|N/A");
+	}
+}
+
+static struct usnic_vnic_res_chunk *
+get_qp_res_chunk(struct usnic_ib_qp_grp *qp_grp)
+{
+	lockdep_assert_held(&qp_grp->lock);
+	/*
+	 * The QP res chunk, used to derive qp indices,
+	 * are just indices of the RQs
+	 */
+	return usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+}
+
+static int enable_qp_grp(struct usnic_ib_qp_grp *qp_grp)
+{
+
+	int status;
+	int i, vnic_idx;
+	struct usnic_vnic_res_chunk *res_chunk;
+	struct usnic_vnic_res *res;
+
+	lockdep_assert_held(&qp_grp->lock);
+
+	vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+
+	res_chunk = get_qp_res_chunk(qp_grp);
+	if (IS_ERR_OR_NULL(res_chunk)) {
+		usnic_err("Unable to get qp res with err %ld\n",
+				PTR_ERR(res_chunk));
+		return res_chunk ? PTR_ERR(res_chunk) : -ENOMEM;
+	}
+
+	for (i = 0; i < res_chunk->cnt; i++) {
+		res = res_chunk->res[i];
+		status = usnic_fwd_enable_qp(qp_grp->ufdev, vnic_idx,
+						res->vnic_idx);
+		if (status) {
+			usnic_err("Failed to enable qp %d of %s:%d\n with err %d\n",
+					res->vnic_idx, qp_grp->ufdev->name,
+					vnic_idx, status);
+			goto out_err;
+		}
+	}
+
+	return 0;
+
+out_err:
+	for (i--; i >= 0; i--) {
+		res = res_chunk->res[i];
+		usnic_fwd_disable_qp(qp_grp->ufdev, vnic_idx,
+					res->vnic_idx);
+	}
+
+	return status;
+}
+
+static int disable_qp_grp(struct usnic_ib_qp_grp *qp_grp)
+{
+	int i, vnic_idx;
+	struct usnic_vnic_res_chunk *res_chunk;
+	struct usnic_vnic_res *res;
+	int status = 0;
+
+	lockdep_assert_held(&qp_grp->lock);
+	vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+
+	res_chunk = get_qp_res_chunk(qp_grp);
+	if (IS_ERR_OR_NULL(res_chunk)) {
+		usnic_err("Unable to get qp res with err %ld\n",
+			PTR_ERR(res_chunk));
+		return res_chunk ? PTR_ERR(res_chunk) : -ENOMEM;
+	}
+
+	for (i = 0; i < res_chunk->cnt; i++) {
+		res = res_chunk->res[i];
+		status = usnic_fwd_disable_qp(qp_grp->ufdev, vnic_idx,
+						res->vnic_idx);
+		if (status) {
+			usnic_err("Failed to disable rq %d of %s:%d\n with err %d\n",
+					res->vnic_idx,
+					qp_grp->ufdev->name,
+					vnic_idx, status);
+		}
+	}
+
+	return status;
+
+}
+
+static int init_filter_action(struct usnic_ib_qp_grp *qp_grp,
+				struct usnic_filter_action *uaction)
+{
+	struct usnic_vnic_res_chunk *res_chunk;
+
+	res_chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+	if (IS_ERR_OR_NULL(res_chunk)) {
+		usnic_err("Unable to get %s with err %ld\n",
+			usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_RQ),
+			PTR_ERR(res_chunk));
+		return res_chunk ? PTR_ERR(res_chunk) : -ENOMEM;
+	}
+
+	uaction->vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+	uaction->action.type = FILTER_ACTION_RQ_STEERING;
+	uaction->action.u.rq_idx = res_chunk->res[DFLT_RQ_IDX]->vnic_idx;
+
+	return 0;
+}
+
+static struct usnic_ib_qp_grp_flow*
+create_roce_custom_flow(struct usnic_ib_qp_grp *qp_grp,
+			struct usnic_transport_spec *trans_spec)
+{
+	uint16_t port_num;
+	int err;
+	struct filter filter;
+	struct usnic_filter_action uaction;
+	struct usnic_ib_qp_grp_flow *qp_flow;
+	struct usnic_fwd_flow *flow;
+	enum usnic_transport_type trans_type;
+
+	trans_type = trans_spec->trans_type;
+	port_num = trans_spec->usnic_roce.port_num;
+
+	/* Reserve Port */
+	port_num = usnic_transport_rsrv_port(trans_type, port_num);
+	if (port_num == 0)
+		return ERR_PTR(-EINVAL);
+
+	/* Create Flow */
+	usnic_fwd_init_usnic_filter(&filter, port_num);
+	err = init_filter_action(qp_grp, &uaction);
+	if (err)
+		goto out_unreserve_port;
+
+	flow = usnic_fwd_alloc_flow(qp_grp->ufdev, &filter, &uaction);
+	if (IS_ERR_OR_NULL(flow)) {
+		usnic_err("Unable to alloc flow failed with err %ld\n",
+				PTR_ERR(flow));
+		err = flow ? PTR_ERR(flow) : -EFAULT;
+		goto out_unreserve_port;
+	}
+
+	/* Create Flow Handle */
+	qp_flow = kzalloc(sizeof(*qp_flow), GFP_ATOMIC);
+	if (IS_ERR_OR_NULL(qp_flow)) {
+		err = qp_flow ? PTR_ERR(qp_flow) : -ENOMEM;
+		goto out_dealloc_flow;
+	}
+	qp_flow->flow = flow;
+	qp_flow->trans_type = trans_type;
+	qp_flow->usnic_roce.port_num = port_num;
+	qp_flow->qp_grp = qp_grp;
+	return qp_flow;
+
+out_dealloc_flow:
+	usnic_fwd_dealloc_flow(flow);
+out_unreserve_port:
+	usnic_transport_unrsrv_port(trans_type, port_num);
+	return ERR_PTR(err);
+}
+
+static void release_roce_custom_flow(struct usnic_ib_qp_grp_flow *qp_flow)
+{
+	usnic_fwd_dealloc_flow(qp_flow->flow);
+	usnic_transport_unrsrv_port(qp_flow->trans_type,
+					qp_flow->usnic_roce.port_num);
+	kfree(qp_flow);
+}
+
+static struct usnic_ib_qp_grp_flow*
+create_udp_flow(struct usnic_ib_qp_grp *qp_grp,
+		struct usnic_transport_spec *trans_spec)
+{
+	struct socket *sock;
+	int sock_fd;
+	int err;
+	struct filter filter;
+	struct usnic_filter_action uaction;
+	struct usnic_ib_qp_grp_flow *qp_flow;
+	struct usnic_fwd_flow *flow;
+	enum usnic_transport_type trans_type;
+	uint32_t addr;
+	uint16_t port_num;
+	int proto;
+
+	trans_type = trans_spec->trans_type;
+	sock_fd = trans_spec->udp.sock_fd;
+
+	/* Get and check socket */
+	sock = usnic_transport_get_socket(sock_fd);
+	if (IS_ERR_OR_NULL(sock))
+		return ERR_CAST(sock);
+
+	err = usnic_transport_sock_get_addr(sock, &proto, &addr, &port_num);
+	if (err)
+		goto out_put_sock;
+
+	if (proto != IPPROTO_UDP) {
+		usnic_err("Protocol for fd %d is not UDP", sock_fd);
+		err = -EPERM;
+		goto out_put_sock;
+	}
+
+	/* Create flow */
+	usnic_fwd_init_udp_filter(&filter, addr, port_num);
+	err = init_filter_action(qp_grp, &uaction);
+	if (err)
+		goto out_put_sock;
+
+	flow = usnic_fwd_alloc_flow(qp_grp->ufdev, &filter, &uaction);
+	if (IS_ERR_OR_NULL(flow)) {
+		usnic_err("Unable to alloc flow failed with err %ld\n",
+				PTR_ERR(flow));
+		err = flow ? PTR_ERR(flow) : -EFAULT;
+		goto out_put_sock;
+	}
+
+	/* Create qp_flow */
+	qp_flow = kzalloc(sizeof(*qp_flow), GFP_ATOMIC);
+	if (IS_ERR_OR_NULL(qp_flow)) {
+		err = qp_flow ? PTR_ERR(qp_flow) : -ENOMEM;
+		goto out_dealloc_flow;
+	}
+	qp_flow->flow = flow;
+	qp_flow->trans_type = trans_type;
+	qp_flow->udp.sock = sock;
+	qp_flow->qp_grp = qp_grp;
+	return qp_flow;
+
+out_dealloc_flow:
+	usnic_fwd_dealloc_flow(flow);
+out_put_sock:
+	usnic_transport_put_socket(sock);
+	return ERR_PTR(err);
+}
+
+static void release_udp_flow(struct usnic_ib_qp_grp_flow *qp_flow)
+{
+	usnic_fwd_dealloc_flow(qp_flow->flow);
+	usnic_transport_put_socket(qp_flow->udp.sock);
+	kfree(qp_flow);
+}
+
+static struct usnic_ib_qp_grp_flow*
+create_and_add_flow(struct usnic_ib_qp_grp *qp_grp,
+			struct usnic_transport_spec *trans_spec)
+{
+	struct usnic_ib_qp_grp_flow *qp_flow;
+	enum usnic_transport_type trans_type;
+
+	trans_type = trans_spec->trans_type;
+	switch (trans_type) {
+	case USNIC_TRANSPORT_ROCE_CUSTOM:
+		qp_flow = create_roce_custom_flow(qp_grp, trans_spec);
+		break;
+	case USNIC_TRANSPORT_IPV4_UDP:
+		qp_flow = create_udp_flow(qp_grp, trans_spec);
+		break;
+	default:
+		usnic_err("Unsupported transport %u\n",
+				trans_spec->trans_type);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (!IS_ERR_OR_NULL(qp_flow)) {
+		list_add_tail(&qp_flow->link, &qp_grp->flows_lst);
+		usnic_debugfs_flow_add(qp_flow);
+	}
+
+
+	return qp_flow;
+}
+
+static void release_and_remove_flow(struct usnic_ib_qp_grp_flow *qp_flow)
+{
+	usnic_debugfs_flow_remove(qp_flow);
+	list_del(&qp_flow->link);
+
+	switch (qp_flow->trans_type) {
+	case USNIC_TRANSPORT_ROCE_CUSTOM:
+		release_roce_custom_flow(qp_flow);
+		break;
+	case USNIC_TRANSPORT_IPV4_UDP:
+		release_udp_flow(qp_flow);
+		break;
+	default:
+		WARN(1, "Unsupported transport %u\n",
+				qp_flow->trans_type);
+		break;
+	}
+}
+
+static void release_and_remove_all_flows(struct usnic_ib_qp_grp *qp_grp)
+{
+	struct usnic_ib_qp_grp_flow *qp_flow, *tmp;
+	list_for_each_entry_safe(qp_flow, tmp, &qp_grp->flows_lst, link)
+		release_and_remove_flow(qp_flow);
+}
+
+int usnic_ib_qp_grp_modify(struct usnic_ib_qp_grp *qp_grp,
+				enum ib_qp_state new_state,
+				void *data)
+{
+	int status = 0;
+	int vnic_idx;
+	struct ib_event ib_event;
+	enum ib_qp_state old_state;
+	struct usnic_transport_spec *trans_spec;
+	struct usnic_ib_qp_grp_flow *qp_flow;
+
+	old_state = qp_grp->state;
+	vnic_idx = usnic_vnic_get_index(qp_grp->vf->vnic);
+	trans_spec = (struct usnic_transport_spec *) data;
+
+	spin_lock(&qp_grp->lock);
+	switch (new_state) {
+	case IB_QPS_RESET:
+		switch (old_state) {
+		case IB_QPS_RESET:
+			/* NO-OP */
+			break;
+		case IB_QPS_INIT:
+			release_and_remove_all_flows(qp_grp);
+			status = 0;
+			break;
+		case IB_QPS_RTR:
+		case IB_QPS_RTS:
+		case IB_QPS_ERR:
+			status = disable_qp_grp(qp_grp);
+			release_and_remove_all_flows(qp_grp);
+			break;
+		default:
+			status = -EINVAL;
+		}
+		break;
+	case IB_QPS_INIT:
+		switch (old_state) {
+		case IB_QPS_RESET:
+			if (trans_spec) {
+				qp_flow = create_and_add_flow(qp_grp,
+								trans_spec);
+				if (IS_ERR_OR_NULL(qp_flow)) {
+					status = qp_flow ? PTR_ERR(qp_flow) : -EFAULT;
+					break;
+				}
+			} else {
+				/*
+				 * Optional to specify filters.
+				 */
+				status = 0;
+			}
+			break;
+		case IB_QPS_INIT:
+			if (trans_spec) {
+				qp_flow = create_and_add_flow(qp_grp,
+								trans_spec);
+				if (IS_ERR_OR_NULL(qp_flow)) {
+					status = qp_flow ? PTR_ERR(qp_flow) : -EFAULT;
+					break;
+				}
+			} else {
+				/*
+				 * Doesn't make sense to go into INIT state
+				 * from INIT state w/o adding filters.
+				 */
+				status = -EINVAL;
+			}
+			break;
+		case IB_QPS_RTR:
+			status = disable_qp_grp(qp_grp);
+			break;
+		case IB_QPS_RTS:
+			status = disable_qp_grp(qp_grp);
+			break;
+		default:
+			status = -EINVAL;
+		}
+		break;
+	case IB_QPS_RTR:
+		switch (old_state) {
+		case IB_QPS_INIT:
+			status = enable_qp_grp(qp_grp);
+			break;
+		default:
+			status = -EINVAL;
+		}
+		break;
+	case IB_QPS_RTS:
+		switch (old_state) {
+		case IB_QPS_RTR:
+			/* NO-OP FOR NOW */
+			break;
+		default:
+			status = -EINVAL;
+		}
+		break;
+	case IB_QPS_ERR:
+		ib_event.device = &qp_grp->vf->pf->ib_dev;
+		ib_event.element.qp = &qp_grp->ibqp;
+		ib_event.event = IB_EVENT_QP_FATAL;
+
+		switch (old_state) {
+		case IB_QPS_RESET:
+			qp_grp->ibqp.event_handler(&ib_event,
+					qp_grp->ibqp.qp_context);
+			break;
+		case IB_QPS_INIT:
+			release_and_remove_all_flows(qp_grp);
+			qp_grp->ibqp.event_handler(&ib_event,
+					qp_grp->ibqp.qp_context);
+			break;
+		case IB_QPS_RTR:
+		case IB_QPS_RTS:
+			status = disable_qp_grp(qp_grp);
+			release_and_remove_all_flows(qp_grp);
+			qp_grp->ibqp.event_handler(&ib_event,
+					qp_grp->ibqp.qp_context);
+			break;
+		default:
+			status = -EINVAL;
+		}
+		break;
+	default:
+		status = -EINVAL;
+	}
+	spin_unlock(&qp_grp->lock);
+
+	if (!status) {
+		qp_grp->state = new_state;
+		usnic_info("Transistioned %u from %s to %s",
+		qp_grp->grp_id,
+		usnic_ib_qp_grp_state_to_string(old_state),
+		usnic_ib_qp_grp_state_to_string(new_state));
+	} else {
+		usnic_err("Failed to transistion %u from %s to %s",
+		qp_grp->grp_id,
+		usnic_ib_qp_grp_state_to_string(old_state),
+		usnic_ib_qp_grp_state_to_string(new_state));
+	}
+
+	return status;
+}
+
+static struct usnic_vnic_res_chunk**
+alloc_res_chunk_list(struct usnic_vnic *vnic,
+			struct usnic_vnic_res_spec *res_spec, void *owner_obj)
+{
+	enum usnic_vnic_res_type res_type;
+	struct usnic_vnic_res_chunk **res_chunk_list;
+	int err, i, res_cnt, res_lst_sz;
+
+	for (res_lst_sz = 0;
+		res_spec->resources[res_lst_sz].type != USNIC_VNIC_RES_TYPE_EOL;
+		res_lst_sz++) {
+		/* Do Nothing */
+	}
+
+	res_chunk_list = kzalloc(sizeof(*res_chunk_list)*(res_lst_sz+1),
+					GFP_ATOMIC);
+	if (!res_chunk_list)
+		return ERR_PTR(-ENOMEM);
+
+	for (i = 0; res_spec->resources[i].type != USNIC_VNIC_RES_TYPE_EOL;
+		i++) {
+		res_type = res_spec->resources[i].type;
+		res_cnt = res_spec->resources[i].cnt;
+
+		res_chunk_list[i] = usnic_vnic_get_resources(vnic, res_type,
+					res_cnt, owner_obj);
+		if (IS_ERR_OR_NULL(res_chunk_list[i])) {
+			err = res_chunk_list[i] ?
+					PTR_ERR(res_chunk_list[i]) : -ENOMEM;
+			usnic_err("Failed to get %s from %s with err %d\n",
+				usnic_vnic_res_type_to_str(res_type),
+				usnic_vnic_pci_name(vnic),
+				err);
+			goto out_free_res;
+		}
+	}
+
+	return res_chunk_list;
+
+out_free_res:
+	for (i--; i > 0; i--)
+		usnic_vnic_put_resources(res_chunk_list[i]);
+	kfree(res_chunk_list);
+	return ERR_PTR(err);
+}
+
+static void free_qp_grp_res(struct usnic_vnic_res_chunk **res_chunk_list)
+{
+	int i;
+	for (i = 0; res_chunk_list[i]; i++)
+		usnic_vnic_put_resources(res_chunk_list[i]);
+	kfree(res_chunk_list);
+}
+
+static int qp_grp_and_vf_bind(struct usnic_ib_vf *vf,
+				struct usnic_ib_pd *pd,
+				struct usnic_ib_qp_grp *qp_grp)
+{
+	int err;
+	struct pci_dev *pdev;
+
+	lockdep_assert_held(&vf->lock);
+
+	pdev = usnic_vnic_get_pdev(vf->vnic);
+	if (vf->qp_grp_ref_cnt == 0) {
+		err = usnic_uiom_attach_dev_to_pd(pd->umem_pd, &pdev->dev);
+		if (err) {
+			usnic_err("Failed to attach %s to domain\n",
+					pci_name(pdev));
+			return err;
+		}
+		vf->pd = pd;
+	}
+	vf->qp_grp_ref_cnt++;
+
+	WARN_ON(vf->pd != pd);
+	qp_grp->vf = vf;
+
+	return 0;
+}
+
+static void qp_grp_and_vf_unbind(struct usnic_ib_qp_grp *qp_grp)
+{
+	struct pci_dev *pdev;
+	struct usnic_ib_pd *pd;
+
+	lockdep_assert_held(&qp_grp->vf->lock);
+
+	pd = qp_grp->vf->pd;
+	pdev = usnic_vnic_get_pdev(qp_grp->vf->vnic);
+	if (--qp_grp->vf->qp_grp_ref_cnt == 0) {
+		qp_grp->vf->pd = NULL;
+		usnic_uiom_detach_dev_from_pd(pd->umem_pd, &pdev->dev);
+	}
+	qp_grp->vf = NULL;
+}
+
+static void log_spec(struct usnic_vnic_res_spec *res_spec)
+{
+	char buf[512];
+	usnic_vnic_spec_dump(buf, sizeof(buf), res_spec);
+	usnic_dbg("%s\n", buf);
+}
+
+static int qp_grp_id_from_flow(struct usnic_ib_qp_grp_flow *qp_flow,
+				uint32_t *id)
+{
+	enum usnic_transport_type trans_type = qp_flow->trans_type;
+	int err;
+	uint16_t port_num = 0;
+
+	switch (trans_type) {
+	case USNIC_TRANSPORT_ROCE_CUSTOM:
+		*id = qp_flow->usnic_roce.port_num;
+		break;
+	case USNIC_TRANSPORT_IPV4_UDP:
+		err = usnic_transport_sock_get_addr(qp_flow->udp.sock,
+							NULL, NULL,
+							&port_num);
+		if (err)
+			return err;
+		/*
+		 * Copy port_num to stack first and then to *id,
+		 * so that the short to int cast works for little
+		 * and big endian systems.
+		 */
+		*id = port_num;
+		break;
+	default:
+		usnic_err("Unsupported transport %u\n", trans_type);
+		return -EINVAL;
+	}
+
+	return 0;
+}
+
+struct usnic_ib_qp_grp *
+usnic_ib_qp_grp_create(struct usnic_fwd_dev *ufdev, struct usnic_ib_vf *vf,
+			struct usnic_ib_pd *pd,
+			struct usnic_vnic_res_spec *res_spec,
+			struct usnic_transport_spec *transport_spec)
+{
+	struct usnic_ib_qp_grp *qp_grp;
+	int err;
+	enum usnic_transport_type transport = transport_spec->trans_type;
+	struct usnic_ib_qp_grp_flow *qp_flow;
+
+	lockdep_assert_held(&vf->lock);
+
+	err = usnic_vnic_res_spec_satisfied(&min_transport_spec[transport],
+						res_spec);
+	if (err) {
+		usnic_err("Spec does not meet miniumum req for transport %d\n",
+				transport);
+		log_spec(res_spec);
+		return ERR_PTR(err);
+	}
+
+	qp_grp = kzalloc(sizeof(*qp_grp), GFP_ATOMIC);
+	if (!qp_grp) {
+		usnic_err("Unable to alloc qp_grp - Out of memory\n");
+		return NULL;
+	}
+
+	qp_grp->res_chunk_list = alloc_res_chunk_list(vf->vnic, res_spec,
+							qp_grp);
+	if (IS_ERR_OR_NULL(qp_grp->res_chunk_list)) {
+		err = qp_grp->res_chunk_list ?
+				PTR_ERR(qp_grp->res_chunk_list) : -ENOMEM;
+		usnic_err("Unable to alloc res for %d with err %d\n",
+				qp_grp->grp_id, err);
+		goto out_free_qp_grp;
+	}
+
+	err = qp_grp_and_vf_bind(vf, pd, qp_grp);
+	if (err)
+		goto out_free_res;
+
+	INIT_LIST_HEAD(&qp_grp->flows_lst);
+	spin_lock_init(&qp_grp->lock);
+	qp_grp->ufdev = ufdev;
+	qp_grp->state = IB_QPS_RESET;
+	qp_grp->owner_pid = current->pid;
+
+	qp_flow = create_and_add_flow(qp_grp, transport_spec);
+	if (IS_ERR_OR_NULL(qp_flow)) {
+		usnic_err("Unable to create and add flow with err %ld\n",
+				PTR_ERR(qp_flow));
+		err = qp_flow ? PTR_ERR(qp_flow) : -EFAULT;
+		goto out_qp_grp_vf_unbind;
+	}
+
+	err = qp_grp_id_from_flow(qp_flow, &qp_grp->grp_id);
+	if (err)
+		goto out_release_flow;
+	qp_grp->ibqp.qp_num = qp_grp->grp_id;
+
+	usnic_ib_sysfs_qpn_add(qp_grp);
+
+	return qp_grp;
+
+out_release_flow:
+	release_and_remove_flow(qp_flow);
+out_qp_grp_vf_unbind:
+	qp_grp_and_vf_unbind(qp_grp);
+out_free_res:
+	free_qp_grp_res(qp_grp->res_chunk_list);
+out_free_qp_grp:
+	kfree(qp_grp);
+
+	return ERR_PTR(err);
+}
+
+void usnic_ib_qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp)
+{
+
+	WARN_ON(qp_grp->state != IB_QPS_RESET);
+	lockdep_assert_held(&qp_grp->vf->lock);
+
+	release_and_remove_all_flows(qp_grp);
+	usnic_ib_sysfs_qpn_remove(qp_grp);
+	qp_grp_and_vf_unbind(qp_grp);
+	free_qp_grp_res(qp_grp->res_chunk_list);
+	kfree(qp_grp);
+}
+
+struct usnic_vnic_res_chunk*
+usnic_ib_qp_grp_get_chunk(struct usnic_ib_qp_grp *qp_grp,
+				enum usnic_vnic_res_type res_type)
+{
+	int i;
+
+	for (i = 0; qp_grp->res_chunk_list[i]; i++) {
+		if (qp_grp->res_chunk_list[i]->type == res_type)
+			return qp_grp->res_chunk_list[i];
+	}
+
+	return ERR_PTR(-EINVAL);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h
new file mode 100644
index 00000000000..b0aafe8db0c
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_qp_grp.h
@@ -0,0 +1,117 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_QP_GRP_H_
+#define USNIC_IB_QP_GRP_H_
+
+#include <linux/debugfs.h>
+#include <rdma/ib_verbs.h>
+
+#include "usnic_ib.h"
+#include "usnic_abi.h"
+#include "usnic_fwd.h"
+#include "usnic_vnic.h"
+
+/*
+ * The qp group struct represents all the hw resources needed to present a ib_qp
+ */
+struct usnic_ib_qp_grp {
+	struct ib_qp				ibqp;
+	enum ib_qp_state			state;
+	int					grp_id;
+
+	struct usnic_fwd_dev			*ufdev;
+	struct usnic_ib_ucontext		*ctx;
+	struct list_head			flows_lst;
+
+	struct usnic_vnic_res_chunk		**res_chunk_list;
+
+	pid_t					owner_pid;
+	struct usnic_ib_vf			*vf;
+	struct list_head			link;
+
+	spinlock_t				lock;
+
+	struct kobject				kobj;
+};
+
+struct usnic_ib_qp_grp_flow {
+	struct usnic_fwd_flow		*flow;
+	enum usnic_transport_type	trans_type;
+	union {
+		struct {
+			uint16_t	port_num;
+		} usnic_roce;
+		struct {
+			struct socket	*sock;
+		} udp;
+	};
+	struct usnic_ib_qp_grp		*qp_grp;
+	struct list_head		link;
+
+	/* Debug FS */
+	struct dentry			*dbgfs_dentry;
+	char				dentry_name[32];
+};
+
+static const struct
+usnic_vnic_res_spec min_transport_spec[USNIC_TRANSPORT_MAX] = {
+	{ /*USNIC_TRANSPORT_UNKNOWN*/
+		.resources = {
+			{.type = USNIC_VNIC_RES_TYPE_EOL,	.cnt = 0,},
+		},
+	},
+	{ /*USNIC_TRANSPORT_ROCE_CUSTOM*/
+		.resources = {
+			{.type = USNIC_VNIC_RES_TYPE_WQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_RQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_CQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_EOL,	.cnt = 0,},
+		},
+	},
+	{ /*USNIC_TRANSPORT_IPV4_UDP*/
+		.resources = {
+			{.type = USNIC_VNIC_RES_TYPE_WQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_RQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_CQ,	.cnt = 1,},
+			{.type = USNIC_VNIC_RES_TYPE_EOL,	.cnt = 0,},
+		},
+	},
+};
+
+const char *usnic_ib_qp_grp_state_to_string(enum ib_qp_state state);
+int usnic_ib_qp_grp_dump_hdr(char *buf, int buf_sz);
+int usnic_ib_qp_grp_dump_rows(void *obj, char *buf, int buf_sz);
+struct usnic_ib_qp_grp *
+usnic_ib_qp_grp_create(struct usnic_fwd_dev *ufdev, struct usnic_ib_vf *vf,
+			struct usnic_ib_pd *pd,
+			struct usnic_vnic_res_spec *res_spec,
+			struct usnic_transport_spec *trans_spec);
+void usnic_ib_qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp);
+int usnic_ib_qp_grp_modify(struct usnic_ib_qp_grp *qp_grp,
+				enum ib_qp_state new_state,
+				void *data);
+struct usnic_vnic_res_chunk
+*usnic_ib_qp_grp_get_chunk(struct usnic_ib_qp_grp *qp_grp,
+				enum usnic_vnic_res_type type);
+static inline
+struct usnic_ib_qp_grp *to_uqp_grp(struct ib_qp *ibqp)
+{
+	return container_of(ibqp, struct usnic_ib_qp_grp, ibqp);
+}
+#endif /* USNIC_IB_QP_GRP_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c
new file mode 100644
index 00000000000..27dc67c1689
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.c
@@ -0,0 +1,341 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/errno.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_common_util.h"
+#include "usnic_ib.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_vnic.h"
+#include "usnic_ib_verbs.h"
+#include "usnic_log.h"
+
+static ssize_t usnic_ib_show_fw_ver(struct device *device,
+					struct device_attribute *attr,
+					char *buf)
+{
+	struct usnic_ib_dev *us_ibdev =
+		container_of(device, struct usnic_ib_dev, ib_dev.dev);
+	struct ethtool_drvinfo info;
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	us_ibdev->netdev->ethtool_ops->get_drvinfo(us_ibdev->netdev, &info);
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return scnprintf(buf, PAGE_SIZE, "%s\n", info.fw_version);
+}
+
+static ssize_t usnic_ib_show_board(struct device *device,
+					struct device_attribute *attr,
+					char *buf)
+{
+	struct usnic_ib_dev *us_ibdev =
+		container_of(device, struct usnic_ib_dev, ib_dev.dev);
+	unsigned short subsystem_device_id;
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	subsystem_device_id = us_ibdev->pdev->subsystem_device;
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return scnprintf(buf, PAGE_SIZE, "%hu\n", subsystem_device_id);
+}
+
+/*
+ * Report the configuration for this PF
+ */
+static ssize_t
+usnic_ib_show_config(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct usnic_ib_dev *us_ibdev;
+	char *ptr;
+	unsigned left;
+	unsigned n;
+	enum usnic_vnic_res_type res_type;
+
+	us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+	/* Buffer space limit is 1 page */
+	ptr = buf;
+	left = PAGE_SIZE;
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	if (atomic_read(&us_ibdev->vf_cnt.refcount) > 0) {
+		char *busname;
+
+		/*
+		 * bus name seems to come with annoying prefix.
+		 * Remove it if it is predictable
+		 */
+		busname = us_ibdev->pdev->bus->name;
+		if (strncmp(busname, "PCI Bus ", 8) == 0)
+			busname += 8;
+
+		n = scnprintf(ptr, left,
+			"%s: %s:%d.%d, %s, %pM, %u VFs\n Per VF:",
+			us_ibdev->ib_dev.name,
+			busname,
+			PCI_SLOT(us_ibdev->pdev->devfn),
+			PCI_FUNC(us_ibdev->pdev->devfn),
+			netdev_name(us_ibdev->netdev),
+			us_ibdev->ufdev->mac,
+			atomic_read(&us_ibdev->vf_cnt.refcount));
+		UPDATE_PTR_LEFT(n, ptr, left);
+
+		for (res_type = USNIC_VNIC_RES_TYPE_EOL;
+				res_type < USNIC_VNIC_RES_TYPE_MAX;
+				res_type++) {
+			if (us_ibdev->vf_res_cnt[res_type] == 0)
+				continue;
+			n = scnprintf(ptr, left, " %d %s%s",
+				us_ibdev->vf_res_cnt[res_type],
+				usnic_vnic_res_type_to_str(res_type),
+				(res_type < (USNIC_VNIC_RES_TYPE_MAX - 1)) ?
+				 "," : "");
+			UPDATE_PTR_LEFT(n, ptr, left);
+		}
+		n = scnprintf(ptr, left, "\n");
+		UPDATE_PTR_LEFT(n, ptr, left);
+	} else {
+		n = scnprintf(ptr, left, "%s: no VFs\n",
+				us_ibdev->ib_dev.name);
+		UPDATE_PTR_LEFT(n, ptr, left);
+	}
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return ptr - buf;
+}
+
+static ssize_t
+usnic_ib_show_iface(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct usnic_ib_dev *us_ibdev;
+
+	us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+	return scnprintf(buf, PAGE_SIZE, "%s\n",
+			netdev_name(us_ibdev->netdev));
+}
+
+static ssize_t
+usnic_ib_show_max_vf(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct usnic_ib_dev *us_ibdev;
+
+	us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+	return scnprintf(buf, PAGE_SIZE, "%u\n",
+			atomic_read(&us_ibdev->vf_cnt.refcount));
+}
+
+static ssize_t
+usnic_ib_show_qp_per_vf(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct usnic_ib_dev *us_ibdev;
+	int qp_per_vf;
+
+	us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+	qp_per_vf = max(us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_WQ],
+			us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_RQ]);
+
+	return scnprintf(buf, PAGE_SIZE,
+				"%d\n", qp_per_vf);
+}
+
+static ssize_t
+usnic_ib_show_cq_per_vf(struct device *device, struct device_attribute *attr,
+			char *buf)
+{
+	struct usnic_ib_dev *us_ibdev;
+
+	us_ibdev = container_of(device, struct usnic_ib_dev, ib_dev.dev);
+
+	return scnprintf(buf, PAGE_SIZE, "%d\n",
+			us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_CQ]);
+}
+
+static DEVICE_ATTR(fw_ver, S_IRUGO, usnic_ib_show_fw_ver, NULL);
+static DEVICE_ATTR(board_id, S_IRUGO, usnic_ib_show_board, NULL);
+static DEVICE_ATTR(config, S_IRUGO, usnic_ib_show_config, NULL);
+static DEVICE_ATTR(iface, S_IRUGO, usnic_ib_show_iface, NULL);
+static DEVICE_ATTR(max_vf, S_IRUGO, usnic_ib_show_max_vf, NULL);
+static DEVICE_ATTR(qp_per_vf, S_IRUGO, usnic_ib_show_qp_per_vf, NULL);
+static DEVICE_ATTR(cq_per_vf, S_IRUGO, usnic_ib_show_cq_per_vf, NULL);
+
+static struct device_attribute *usnic_class_attributes[] = {
+	&dev_attr_fw_ver,
+	&dev_attr_board_id,
+	&dev_attr_config,
+	&dev_attr_iface,
+	&dev_attr_max_vf,
+	&dev_attr_qp_per_vf,
+	&dev_attr_cq_per_vf,
+};
+
+struct qpn_attribute {
+	struct attribute attr;
+	ssize_t (*show)(struct usnic_ib_qp_grp *, char *buf);
+};
+
+/*
+ * Definitions for supporting QPN entries in sysfs
+ */
+static ssize_t
+usnic_ib_qpn_attr_show(struct kobject *kobj, struct attribute *attr, char *buf)
+{
+	struct usnic_ib_qp_grp *qp_grp;
+	struct qpn_attribute *qpn_attr;
+
+	qp_grp = container_of(kobj, struct usnic_ib_qp_grp, kobj);
+	qpn_attr = container_of(attr, struct qpn_attribute, attr);
+
+	return qpn_attr->show(qp_grp, buf);
+}
+
+static const struct sysfs_ops usnic_ib_qpn_sysfs_ops = {
+	.show = usnic_ib_qpn_attr_show
+};
+
+#define QPN_ATTR_RO(NAME) \
+struct qpn_attribute qpn_attr_##NAME = __ATTR_RO(NAME)
+
+static ssize_t context_show(struct usnic_ib_qp_grp *qp_grp, char *buf)
+{
+	return scnprintf(buf, PAGE_SIZE, "0x%p\n", qp_grp->ctx);
+}
+
+static ssize_t summary_show(struct usnic_ib_qp_grp *qp_grp, char *buf)
+{
+	int i, j, n;
+	int left;
+	char *ptr;
+	struct usnic_vnic_res_chunk *res_chunk;
+	struct usnic_vnic_res *vnic_res;
+
+	left = PAGE_SIZE;
+	ptr = buf;
+
+	n = scnprintf(ptr, left,
+			"QPN: %d State: (%s) PID: %u VF Idx: %hu ",
+			qp_grp->ibqp.qp_num,
+			usnic_ib_qp_grp_state_to_string(qp_grp->state),
+			qp_grp->owner_pid,
+			usnic_vnic_get_index(qp_grp->vf->vnic));
+	UPDATE_PTR_LEFT(n, ptr, left);
+
+	for (i = 0; qp_grp->res_chunk_list[i]; i++) {
+		res_chunk = qp_grp->res_chunk_list[i];
+		for (j = 0; j < res_chunk->cnt; j++) {
+			vnic_res = res_chunk->res[j];
+			n = scnprintf(ptr, left, "%s[%d] ",
+				usnic_vnic_res_type_to_str(vnic_res->type),
+				vnic_res->vnic_idx);
+			UPDATE_PTR_LEFT(n, ptr, left);
+		}
+	}
+
+	n = scnprintf(ptr, left, "\n");
+	UPDATE_PTR_LEFT(n, ptr, left);
+
+	return ptr - buf;
+}
+
+static QPN_ATTR_RO(context);
+static QPN_ATTR_RO(summary);
+
+static struct attribute *usnic_ib_qpn_default_attrs[] = {
+	&qpn_attr_context.attr,
+	&qpn_attr_summary.attr,
+	NULL
+};
+
+static struct kobj_type usnic_ib_qpn_type = {
+	.sysfs_ops = &usnic_ib_qpn_sysfs_ops,
+	.default_attrs = usnic_ib_qpn_default_attrs
+};
+
+int usnic_ib_sysfs_register_usdev(struct usnic_ib_dev *us_ibdev)
+{
+	int i;
+	int err;
+	for (i = 0; i < ARRAY_SIZE(usnic_class_attributes); ++i) {
+		err = device_create_file(&us_ibdev->ib_dev.dev,
+						usnic_class_attributes[i]);
+		if (err) {
+			usnic_err("Failed to create device file %d for %s eith err %d",
+				i, us_ibdev->ib_dev.name, err);
+			return -EINVAL;
+		}
+	}
+
+	/* create kernel object for looking at individual QPs */
+	kobject_get(&us_ibdev->ib_dev.dev.kobj);
+	us_ibdev->qpn_kobj = kobject_create_and_add("qpn",
+			&us_ibdev->ib_dev.dev.kobj);
+	if (us_ibdev->qpn_kobj == NULL) {
+		kobject_put(&us_ibdev->ib_dev.dev.kobj);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+void usnic_ib_sysfs_unregister_usdev(struct usnic_ib_dev *us_ibdev)
+{
+	int i;
+	for (i = 0; i < ARRAY_SIZE(usnic_class_attributes); ++i) {
+		device_remove_file(&us_ibdev->ib_dev.dev,
+					usnic_class_attributes[i]);
+	}
+
+	kobject_put(us_ibdev->qpn_kobj);
+}
+
+void usnic_ib_sysfs_qpn_add(struct usnic_ib_qp_grp *qp_grp)
+{
+	struct usnic_ib_dev *us_ibdev;
+	int err;
+
+	us_ibdev = qp_grp->vf->pf;
+
+	err = kobject_init_and_add(&qp_grp->kobj, &usnic_ib_qpn_type,
+			kobject_get(us_ibdev->qpn_kobj),
+			"%d", qp_grp->grp_id);
+	if (err) {
+		kobject_put(us_ibdev->qpn_kobj);
+		return;
+	}
+}
+
+void usnic_ib_sysfs_qpn_remove(struct usnic_ib_qp_grp *qp_grp)
+{
+	struct usnic_ib_dev *us_ibdev;
+
+	us_ibdev = qp_grp->vf->pf;
+
+	kobject_put(&qp_grp->kobj);
+	kobject_put(us_ibdev->qpn_kobj);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_sysfs.h b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.h
new file mode 100644
index 00000000000..0d09b493cd0
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_sysfs.h
@@ -0,0 +1,29 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_SYSFS_H_
+#define USNIC_IB_SYSFS_H_
+
+#include "usnic_ib.h"
+
+int usnic_ib_sysfs_register_usdev(struct usnic_ib_dev *us_ibdev);
+void usnic_ib_sysfs_unregister_usdev(struct usnic_ib_dev *us_ibdev);
+void usnic_ib_sysfs_qpn_add(struct usnic_ib_qp_grp *qp_grp);
+void usnic_ib_sysfs_qpn_remove(struct usnic_ib_qp_grp *qp_grp);
+
+#endif /* !USNIC_IB_SYSFS_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.c b/drivers/infiniband/hw/usnic/usnic_ib_verbs.c
new file mode 100644
index 00000000000..53bd6a2d9cd
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_verbs.c
@@ -0,0 +1,768 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/module.h>
+#include <linux/init.h>
+#include <linux/slab.h>
+#include <linux/errno.h>
+
+#include <rdma/ib_user_verbs.h>
+#include <rdma/ib_addr.h>
+
+#include "usnic_abi.h"
+#include "usnic_ib.h"
+#include "usnic_common_util.h"
+#include "usnic_ib_qp_grp.h"
+#include "usnic_fwd.h"
+#include "usnic_log.h"
+#include "usnic_uiom.h"
+#include "usnic_transport.h"
+
+#define USNIC_DEFAULT_TRANSPORT USNIC_TRANSPORT_ROCE_CUSTOM
+
+static void usnic_ib_fw_string_to_u64(char *fw_ver_str, u64 *fw_ver)
+{
+	*fw_ver = (u64) *fw_ver_str;
+}
+
+static int usnic_ib_fill_create_qp_resp(struct usnic_ib_qp_grp *qp_grp,
+					struct ib_udata *udata)
+{
+	struct usnic_ib_dev *us_ibdev;
+	struct usnic_ib_create_qp_resp resp;
+	struct pci_dev *pdev;
+	struct vnic_dev_bar *bar;
+	struct usnic_vnic_res_chunk *chunk;
+	struct usnic_ib_qp_grp_flow *default_flow;
+	int i, err;
+
+	memset(&resp, 0, sizeof(resp));
+
+	us_ibdev = qp_grp->vf->pf;
+	pdev = usnic_vnic_get_pdev(qp_grp->vf->vnic);
+	if (!pdev) {
+		usnic_err("Failed to get pdev of qp_grp %d\n",
+				qp_grp->grp_id);
+		return -EFAULT;
+	}
+
+	bar = usnic_vnic_get_bar(qp_grp->vf->vnic, 0);
+	if (!bar) {
+		usnic_err("Failed to get bar0 of qp_grp %d vf %s",
+				qp_grp->grp_id, pci_name(pdev));
+		return -EFAULT;
+	}
+
+	resp.vfid = usnic_vnic_get_index(qp_grp->vf->vnic);
+	resp.bar_bus_addr = bar->bus_addr;
+	resp.bar_len = bar->len;
+
+	chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_RQ);
+	if (IS_ERR_OR_NULL(chunk)) {
+		usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+			usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_RQ),
+			qp_grp->grp_id,
+			PTR_ERR(chunk));
+		return chunk ? PTR_ERR(chunk) : -ENOMEM;
+	}
+
+	WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_RQ);
+	resp.rq_cnt = chunk->cnt;
+	for (i = 0; i < chunk->cnt; i++)
+		resp.rq_idx[i] = chunk->res[i]->vnic_idx;
+
+	chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_WQ);
+	if (IS_ERR_OR_NULL(chunk)) {
+		usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+			usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_WQ),
+			qp_grp->grp_id,
+			PTR_ERR(chunk));
+		return chunk ? PTR_ERR(chunk) : -ENOMEM;
+	}
+
+	WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_WQ);
+	resp.wq_cnt = chunk->cnt;
+	for (i = 0; i < chunk->cnt; i++)
+		resp.wq_idx[i] = chunk->res[i]->vnic_idx;
+
+	chunk = usnic_ib_qp_grp_get_chunk(qp_grp, USNIC_VNIC_RES_TYPE_CQ);
+	if (IS_ERR_OR_NULL(chunk)) {
+		usnic_err("Failed to get chunk %s for qp_grp %d with err %ld\n",
+			usnic_vnic_res_type_to_str(USNIC_VNIC_RES_TYPE_CQ),
+			qp_grp->grp_id,
+			PTR_ERR(chunk));
+		return chunk ? PTR_ERR(chunk) : -ENOMEM;
+	}
+
+	WARN_ON(chunk->type != USNIC_VNIC_RES_TYPE_CQ);
+	resp.cq_cnt = chunk->cnt;
+	for (i = 0; i < chunk->cnt; i++)
+		resp.cq_idx[i] = chunk->res[i]->vnic_idx;
+
+	default_flow = list_first_entry(&qp_grp->flows_lst,
+					struct usnic_ib_qp_grp_flow, link);
+	resp.transport = default_flow->trans_type;
+
+	err = ib_copy_to_udata(udata, &resp, sizeof(resp));
+	if (err) {
+		usnic_err("Failed to copy udata for %s", us_ibdev->ib_dev.name);
+		return err;
+	}
+
+	return 0;
+}
+
+static struct usnic_ib_qp_grp*
+find_free_vf_and_create_qp_grp(struct usnic_ib_dev *us_ibdev,
+				struct usnic_ib_pd *pd,
+				struct usnic_transport_spec *trans_spec,
+				struct usnic_vnic_res_spec *res_spec)
+{
+	struct usnic_ib_vf *vf;
+	struct usnic_vnic *vnic;
+	struct usnic_ib_qp_grp *qp_grp;
+	struct device *dev, **dev_list;
+	int i, found = 0;
+
+	BUG_ON(!mutex_is_locked(&us_ibdev->usdev_lock));
+
+	if (list_empty(&us_ibdev->vf_dev_list)) {
+		usnic_info("No vfs to allocate\n");
+		return NULL;
+	}
+
+	if (usnic_ib_share_vf) {
+		/* Try to find resouces on a used vf which is in pd */
+		dev_list = usnic_uiom_get_dev_list(pd->umem_pd);
+		for (i = 0; dev_list[i]; i++) {
+			dev = dev_list[i];
+			vf = pci_get_drvdata(to_pci_dev(dev));
+			spin_lock(&vf->lock);
+			vnic = vf->vnic;
+			if (!usnic_vnic_check_room(vnic, res_spec)) {
+				usnic_dbg("Found used vnic %s from %s\n",
+						us_ibdev->ib_dev.name,
+						pci_name(usnic_vnic_get_pdev(
+									vnic)));
+				found = 1;
+				break;
+			}
+			spin_unlock(&vf->lock);
+
+		}
+		usnic_uiom_free_dev_list(dev_list);
+	}
+
+	if (!found) {
+		/* Try to find resources on an unused vf */
+		list_for_each_entry(vf, &us_ibdev->vf_dev_list, link) {
+			spin_lock(&vf->lock);
+			vnic = vf->vnic;
+			if (vf->qp_grp_ref_cnt == 0 &&
+				usnic_vnic_check_room(vnic, res_spec) == 0) {
+				found = 1;
+				break;
+			}
+			spin_unlock(&vf->lock);
+		}
+	}
+
+	if (!found) {
+		usnic_info("No free qp grp found on %s\n",
+				us_ibdev->ib_dev.name);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	qp_grp = usnic_ib_qp_grp_create(us_ibdev->ufdev, vf, pd, res_spec,
+						trans_spec);
+	spin_unlock(&vf->lock);
+	if (IS_ERR_OR_NULL(qp_grp)) {
+		usnic_err("Failed to allocate qp_grp\n");
+		return ERR_PTR(qp_grp ? PTR_ERR(qp_grp) : -ENOMEM);
+	}
+
+	return qp_grp;
+}
+
+static void qp_grp_destroy(struct usnic_ib_qp_grp *qp_grp)
+{
+	struct usnic_ib_vf *vf = qp_grp->vf;
+
+	WARN_ON(qp_grp->state != IB_QPS_RESET);
+
+	spin_lock(&vf->lock);
+	usnic_ib_qp_grp_destroy(qp_grp);
+	spin_unlock(&vf->lock);
+}
+
+static void eth_speed_to_ib_speed(int speed, u8 *active_speed,
+					u8 *active_width)
+{
+	if (speed <= 10000) {
+		*active_width = IB_WIDTH_1X;
+		*active_speed = IB_SPEED_FDR10;
+	} else if (speed <= 20000) {
+		*active_width = IB_WIDTH_4X;
+		*active_speed = IB_SPEED_DDR;
+	} else if (speed <= 30000) {
+		*active_width = IB_WIDTH_4X;
+		*active_speed = IB_SPEED_QDR;
+	} else if (speed <= 40000) {
+		*active_width = IB_WIDTH_4X;
+		*active_speed = IB_SPEED_FDR10;
+	} else {
+		*active_width = IB_WIDTH_4X;
+		*active_speed = IB_SPEED_EDR;
+	}
+}
+
+static int create_qp_validate_user_data(struct usnic_ib_create_qp_cmd cmd)
+{
+	if (cmd.spec.trans_type <= USNIC_TRANSPORT_UNKNOWN ||
+			cmd.spec.trans_type >= USNIC_TRANSPORT_MAX)
+		return -EINVAL;
+
+	return 0;
+}
+
+/* Start of ib callback functions */
+
+enum rdma_link_layer usnic_ib_port_link_layer(struct ib_device *device,
+						u8 port_num)
+{
+	return IB_LINK_LAYER_ETHERNET;
+}
+
+int usnic_ib_query_device(struct ib_device *ibdev,
+				struct ib_device_attr *props)
+{
+	struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+	union ib_gid gid;
+	struct ethtool_drvinfo info;
+	struct ethtool_cmd cmd;
+	int qp_per_vf;
+
+	usnic_dbg("\n");
+	mutex_lock(&us_ibdev->usdev_lock);
+	us_ibdev->netdev->ethtool_ops->get_drvinfo(us_ibdev->netdev, &info);
+	us_ibdev->netdev->ethtool_ops->get_settings(us_ibdev->netdev, &cmd);
+	memset(props, 0, sizeof(*props));
+	usnic_mac_ip_to_gid(us_ibdev->ufdev->mac, us_ibdev->ufdev->inaddr,
+			&gid.raw[0]);
+	memcpy(&props->sys_image_guid, &gid.global.interface_id,
+		sizeof(gid.global.interface_id));
+	usnic_ib_fw_string_to_u64(&info.fw_version[0], &props->fw_ver);
+	props->max_mr_size = USNIC_UIOM_MAX_MR_SIZE;
+	props->page_size_cap = USNIC_UIOM_PAGE_SIZE;
+	props->vendor_id = PCI_VENDOR_ID_CISCO;
+	props->vendor_part_id = PCI_DEVICE_ID_CISCO_VIC_USPACE_NIC;
+	props->hw_ver = us_ibdev->pdev->subsystem_device;
+	qp_per_vf = max(us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_WQ],
+			us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_RQ]);
+	props->max_qp = qp_per_vf *
+		atomic_read(&us_ibdev->vf_cnt.refcount);
+	props->device_cap_flags = IB_DEVICE_PORT_ACTIVE_EVENT |
+		IB_DEVICE_SYS_IMAGE_GUID | IB_DEVICE_BLOCK_MULTICAST_LOOPBACK;
+	props->max_cq = us_ibdev->vf_res_cnt[USNIC_VNIC_RES_TYPE_CQ] *
+		atomic_read(&us_ibdev->vf_cnt.refcount);
+	props->max_pd = USNIC_UIOM_MAX_PD_CNT;
+	props->max_mr = USNIC_UIOM_MAX_MR_CNT;
+	props->local_ca_ack_delay = 0;
+	props->max_pkeys = 0;
+	props->atomic_cap = IB_ATOMIC_NONE;
+	props->masked_atomic_cap = props->atomic_cap;
+	props->max_qp_rd_atom = 0;
+	props->max_qp_init_rd_atom = 0;
+	props->max_res_rd_atom = 0;
+	props->max_srq = 0;
+	props->max_srq_wr = 0;
+	props->max_srq_sge = 0;
+	props->max_fast_reg_page_list_len = 0;
+	props->max_mcast_grp = 0;
+	props->max_mcast_qp_attach = 0;
+	props->max_total_mcast_qp_attach = 0;
+	props->max_map_per_fmr = 0;
+	/* Owned by Userspace
+	 * max_qp_wr, max_sge, max_sge_rd, max_cqe */
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return 0;
+}
+
+int usnic_ib_query_port(struct ib_device *ibdev, u8 port,
+				struct ib_port_attr *props)
+{
+	struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+	struct ethtool_cmd cmd;
+
+	usnic_dbg("\n");
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	us_ibdev->netdev->ethtool_ops->get_settings(us_ibdev->netdev, &cmd);
+	memset(props, 0, sizeof(*props));
+
+	props->lid = 0;
+	props->lmc = 1;
+	props->sm_lid = 0;
+	props->sm_sl = 0;
+
+	if (!us_ibdev->ufdev->link_up) {
+		props->state = IB_PORT_DOWN;
+		props->phys_state = 3;
+	} else if (!us_ibdev->ufdev->inaddr) {
+		props->state = IB_PORT_INIT;
+		props->phys_state = 4;
+	} else {
+		props->state = IB_PORT_ACTIVE;
+		props->phys_state = 5;
+	}
+
+	props->port_cap_flags = 0;
+	props->gid_tbl_len = 1;
+	props->pkey_tbl_len = 1;
+	props->bad_pkey_cntr = 0;
+	props->qkey_viol_cntr = 0;
+	eth_speed_to_ib_speed(cmd.speed, &props->active_speed,
+				&props->active_width);
+	props->max_mtu = IB_MTU_4096;
+	props->active_mtu = iboe_get_mtu(us_ibdev->ufdev->mtu);
+	/* Userspace will adjust for hdrs */
+	props->max_msg_sz = us_ibdev->ufdev->mtu;
+	props->max_vl_num = 1;
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return 0;
+}
+
+int usnic_ib_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
+				int qp_attr_mask,
+				struct ib_qp_init_attr *qp_init_attr)
+{
+	struct usnic_ib_qp_grp *qp_grp;
+	struct usnic_ib_vf *vf;
+	int err;
+
+	usnic_dbg("\n");
+
+	memset(qp_attr, 0, sizeof(*qp_attr));
+	memset(qp_init_attr, 0, sizeof(*qp_init_attr));
+
+	qp_grp = to_uqp_grp(qp);
+	vf = qp_grp->vf;
+	mutex_lock(&vf->pf->usdev_lock);
+	usnic_dbg("\n");
+	qp_attr->qp_state = qp_grp->state;
+	qp_attr->cur_qp_state = qp_grp->state;
+
+	switch (qp_grp->ibqp.qp_type) {
+	case IB_QPT_UD:
+		qp_attr->qkey = 0;
+		break;
+	default:
+		usnic_err("Unexpected qp_type %d\n", qp_grp->ibqp.qp_type);
+		err = -EINVAL;
+		goto err_out;
+	}
+
+	mutex_unlock(&vf->pf->usdev_lock);
+	return 0;
+
+err_out:
+	mutex_unlock(&vf->pf->usdev_lock);
+	return err;
+}
+
+int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+				union ib_gid *gid)
+{
+
+	struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+	usnic_dbg("\n");
+
+	if (index > 1)
+		return -EINVAL;
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	memset(&(gid->raw[0]), 0, sizeof(gid->raw));
+	usnic_mac_ip_to_gid(us_ibdev->ufdev->mac, us_ibdev->ufdev->inaddr,
+			&gid->raw[0]);
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return 0;
+}
+
+int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+				u16 *pkey)
+{
+	if (index > 1)
+		return -EINVAL;
+
+	*pkey = 0xffff;
+	return 0;
+}
+
+struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev,
+					struct ib_ucontext *context,
+					struct ib_udata *udata)
+{
+	struct usnic_ib_pd *pd;
+	void *umem_pd;
+
+	usnic_dbg("\n");
+
+	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	umem_pd = pd->umem_pd = usnic_uiom_alloc_pd();
+	if (IS_ERR_OR_NULL(umem_pd)) {
+		kfree(pd);
+		return ERR_PTR(umem_pd ? PTR_ERR(umem_pd) : -ENOMEM);
+	}
+
+	usnic_info("domain 0x%p allocated for context 0x%p and device %s\n",
+			pd, context, ibdev->name);
+	return &pd->ibpd;
+}
+
+int usnic_ib_dealloc_pd(struct ib_pd *pd)
+{
+	usnic_info("freeing domain 0x%p\n", pd);
+
+	usnic_uiom_dealloc_pd((to_upd(pd))->umem_pd);
+	kfree(pd);
+	return 0;
+}
+
+struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd,
+					struct ib_qp_init_attr *init_attr,
+					struct ib_udata *udata)
+{
+	int err;
+	struct usnic_ib_dev *us_ibdev;
+	struct usnic_ib_qp_grp *qp_grp;
+	struct usnic_ib_ucontext *ucontext;
+	int cq_cnt;
+	struct usnic_vnic_res_spec res_spec;
+	struct usnic_ib_create_qp_cmd cmd;
+	struct usnic_transport_spec trans_spec;
+
+	usnic_dbg("\n");
+
+	ucontext = to_uucontext(pd->uobject->context);
+	us_ibdev = to_usdev(pd->device);
+
+	if (init_attr->create_flags)
+		return ERR_PTR(-EINVAL);
+
+	err = ib_copy_from_udata(&cmd, udata, sizeof(cmd));
+	if (err) {
+		usnic_err("%s: cannot copy udata for create_qp\n",
+				us_ibdev->ib_dev.name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	err = create_qp_validate_user_data(cmd);
+	if (err) {
+		usnic_err("%s: Failed to validate user data\n",
+				us_ibdev->ib_dev.name);
+		return ERR_PTR(-EINVAL);
+	}
+
+	if (init_attr->qp_type != IB_QPT_UD) {
+		usnic_err("%s asked to make a non-UD QP: %d\n",
+				us_ibdev->ib_dev.name, init_attr->qp_type);
+		return ERR_PTR(-EINVAL);
+	}
+
+	trans_spec = cmd.spec;
+	mutex_lock(&us_ibdev->usdev_lock);
+	cq_cnt = (init_attr->send_cq == init_attr->recv_cq) ? 1 : 2;
+	res_spec = min_transport_spec[trans_spec.trans_type];
+	usnic_vnic_res_spec_update(&res_spec, USNIC_VNIC_RES_TYPE_CQ, cq_cnt);
+	qp_grp = find_free_vf_and_create_qp_grp(us_ibdev, to_upd(pd),
+						&trans_spec,
+						&res_spec);
+	if (IS_ERR_OR_NULL(qp_grp)) {
+		err = qp_grp ? PTR_ERR(qp_grp) : -ENOMEM;
+		goto out_release_mutex;
+	}
+
+	err = usnic_ib_fill_create_qp_resp(qp_grp, udata);
+	if (err) {
+		err = -EBUSY;
+		goto out_release_qp_grp;
+	}
+
+	qp_grp->ctx = ucontext;
+	list_add_tail(&qp_grp->link, &ucontext->qp_grp_list);
+	usnic_ib_log_vf(qp_grp->vf);
+	mutex_unlock(&us_ibdev->usdev_lock);
+	return &qp_grp->ibqp;
+
+out_release_qp_grp:
+	qp_grp_destroy(qp_grp);
+out_release_mutex:
+	mutex_unlock(&us_ibdev->usdev_lock);
+	return ERR_PTR(err);
+}
+
+int usnic_ib_destroy_qp(struct ib_qp *qp)
+{
+	struct usnic_ib_qp_grp *qp_grp;
+	struct usnic_ib_vf *vf;
+
+	usnic_dbg("\n");
+
+	qp_grp = to_uqp_grp(qp);
+	vf = qp_grp->vf;
+	mutex_lock(&vf->pf->usdev_lock);
+	if (usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RESET, NULL)) {
+		usnic_err("Failed to move qp grp %u to reset\n",
+				qp_grp->grp_id);
+	}
+
+	list_del(&qp_grp->link);
+	qp_grp_destroy(qp_grp);
+	mutex_unlock(&vf->pf->usdev_lock);
+
+	return 0;
+}
+
+int usnic_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+				int attr_mask, struct ib_udata *udata)
+{
+	struct usnic_ib_qp_grp *qp_grp;
+	int status;
+	usnic_dbg("\n");
+
+	qp_grp = to_uqp_grp(ibqp);
+
+	/* TODO: Future Support All States */
+	mutex_lock(&qp_grp->vf->pf->usdev_lock);
+	if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_INIT) {
+		status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_INIT, NULL);
+	} else if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_RTR) {
+		status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RTR, NULL);
+	} else if ((attr_mask & IB_QP_STATE) && attr->qp_state == IB_QPS_RTS) {
+		status = usnic_ib_qp_grp_modify(qp_grp, IB_QPS_RTS, NULL);
+	} else {
+		usnic_err("Unexpected combination mask: %u state: %u\n",
+				attr_mask & IB_QP_STATE, attr->qp_state);
+		status = -EINVAL;
+	}
+
+	mutex_unlock(&qp_grp->vf->pf->usdev_lock);
+	return status;
+}
+
+struct ib_cq *usnic_ib_create_cq(struct ib_device *ibdev, int entries,
+					int vector, struct ib_ucontext *context,
+					struct ib_udata *udata)
+{
+	struct ib_cq *cq;
+
+	usnic_dbg("\n");
+	cq = kzalloc(sizeof(*cq), GFP_KERNEL);
+	if (!cq)
+		return ERR_PTR(-EBUSY);
+
+	return cq;
+}
+
+int usnic_ib_destroy_cq(struct ib_cq *cq)
+{
+	usnic_dbg("\n");
+	kfree(cq);
+	return 0;
+}
+
+struct ib_mr *usnic_ib_reg_mr(struct ib_pd *pd, u64 start, u64 length,
+					u64 virt_addr, int access_flags,
+					struct ib_udata *udata)
+{
+	struct usnic_ib_mr *mr;
+	int err;
+
+	usnic_dbg("start 0x%llx va 0x%llx length 0x%llx\n", start,
+			virt_addr, length);
+
+	mr = kzalloc(sizeof(*mr), GFP_KERNEL);
+	if (IS_ERR_OR_NULL(mr))
+		return ERR_PTR(mr ? PTR_ERR(mr) : -ENOMEM);
+
+	mr->umem = usnic_uiom_reg_get(to_upd(pd)->umem_pd, start, length,
+					access_flags, 0);
+	if (IS_ERR_OR_NULL(mr->umem)) {
+		err = mr->umem ? PTR_ERR(mr->umem) : -EFAULT;
+		goto err_free;
+	}
+
+	mr->ibmr.lkey = mr->ibmr.rkey = 0;
+	return &mr->ibmr;
+
+err_free:
+	kfree(mr);
+	return ERR_PTR(err);
+}
+
+int usnic_ib_dereg_mr(struct ib_mr *ibmr)
+{
+	struct usnic_ib_mr *mr = to_umr(ibmr);
+
+	usnic_dbg("va 0x%lx length 0x%zx\n", mr->umem->va, mr->umem->length);
+
+	usnic_uiom_reg_release(mr->umem, ibmr->pd->uobject->context->closing);
+	kfree(mr);
+	return 0;
+}
+
+struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev,
+							struct ib_udata *udata)
+{
+	struct usnic_ib_ucontext *context;
+	struct usnic_ib_dev *us_ibdev = to_usdev(ibdev);
+	usnic_dbg("\n");
+
+	context = kmalloc(sizeof(*context), GFP_KERNEL);
+	if (!context)
+		return ERR_PTR(-ENOMEM);
+
+	INIT_LIST_HEAD(&context->qp_grp_list);
+	mutex_lock(&us_ibdev->usdev_lock);
+	list_add_tail(&context->link, &us_ibdev->ctx_list);
+	mutex_unlock(&us_ibdev->usdev_lock);
+
+	return &context->ibucontext;
+}
+
+int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext)
+{
+	struct usnic_ib_ucontext *context = to_uucontext(ibcontext);
+	struct usnic_ib_dev *us_ibdev = to_usdev(ibcontext->device);
+	usnic_dbg("\n");
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	BUG_ON(!list_empty(&context->qp_grp_list));
+	list_del(&context->link);
+	mutex_unlock(&us_ibdev->usdev_lock);
+	kfree(context);
+	return 0;
+}
+
+int usnic_ib_mmap(struct ib_ucontext *context,
+				struct vm_area_struct *vma)
+{
+	struct usnic_ib_ucontext *uctx = to_ucontext(context);
+	struct usnic_ib_dev *us_ibdev;
+	struct usnic_ib_qp_grp *qp_grp;
+	struct usnic_ib_vf *vf;
+	struct vnic_dev_bar *bar;
+	dma_addr_t bus_addr;
+	unsigned int len;
+	unsigned int vfid;
+
+	usnic_dbg("\n");
+
+	us_ibdev = to_usdev(context->device);
+	vma->vm_flags |= VM_IO;
+	vma->vm_page_prot = pgprot_noncached(vma->vm_page_prot);
+	vfid = vma->vm_pgoff;
+	usnic_dbg("Page Offset %lu PAGE_SHIFT %u VFID %u\n",
+			vma->vm_pgoff, PAGE_SHIFT, vfid);
+
+	mutex_lock(&us_ibdev->usdev_lock);
+	list_for_each_entry(qp_grp, &uctx->qp_grp_list, link) {
+		vf = qp_grp->vf;
+		if (usnic_vnic_get_index(vf->vnic) == vfid) {
+			bar = usnic_vnic_get_bar(vf->vnic, 0);
+			if ((vma->vm_end - vma->vm_start) != bar->len) {
+				usnic_err("Bar0 Len %lu - Request map %lu\n",
+						bar->len,
+						vma->vm_end - vma->vm_start);
+				mutex_unlock(&us_ibdev->usdev_lock);
+				return -EINVAL;
+			}
+			bus_addr = bar->bus_addr;
+			len = bar->len;
+			usnic_dbg("bus: %pa vaddr: %p size: %ld\n",
+					&bus_addr, bar->vaddr, bar->len);
+			mutex_unlock(&us_ibdev->usdev_lock);
+
+			return remap_pfn_range(vma,
+						vma->vm_start,
+						bus_addr >> PAGE_SHIFT,
+						len, vma->vm_page_prot);
+		}
+	}
+
+	mutex_unlock(&us_ibdev->usdev_lock);
+	usnic_err("No VF %u found\n", vfid);
+	return -EINVAL;
+}
+
+/* In ib callbacks section -  Start of stub funcs */
+struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd,
+					struct ib_ah_attr *ah_attr)
+{
+	usnic_dbg("\n");
+	return ERR_PTR(-EPERM);
+}
+
+int usnic_ib_destroy_ah(struct ib_ah *ah)
+{
+	usnic_dbg("\n");
+	return -EINVAL;
+}
+
+int usnic_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+				struct ib_send_wr **bad_wr)
+{
+	usnic_dbg("\n");
+	return -EINVAL;
+}
+
+int usnic_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+				struct ib_recv_wr **bad_wr)
+{
+	usnic_dbg("\n");
+	return -EINVAL;
+}
+
+int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries,
+				struct ib_wc *wc)
+{
+	usnic_dbg("\n");
+	return -EINVAL;
+}
+
+int usnic_ib_req_notify_cq(struct ib_cq *cq,
+					enum ib_cq_notify_flags flags)
+{
+	usnic_dbg("\n");
+	return -EINVAL;
+}
+
+struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc)
+{
+	usnic_dbg("\n");
+	return ERR_PTR(-ENOMEM);
+}
+
+
+/* In ib callbacks section - End of stub funcs */
+/* End of ib callbacks section */
diff --git a/drivers/infiniband/hw/usnic/usnic_ib_verbs.h b/drivers/infiniband/hw/usnic/usnic_ib_verbs.h
new file mode 100644
index 00000000000..bb864f5aed7
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_ib_verbs.h
@@ -0,0 +1,72 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_IB_VERBS_H_
+#define USNIC_IB_VERBS_H_
+
+#include "usnic_ib.h"
+
+enum rdma_link_layer usnic_ib_port_link_layer(struct ib_device *device,
+						u8 port_num);
+int usnic_ib_query_device(struct ib_device *ibdev,
+				struct ib_device_attr *props);
+int usnic_ib_query_port(struct ib_device *ibdev, u8 port,
+				struct ib_port_attr *props);
+int usnic_ib_query_qp(struct ib_qp *qp, struct ib_qp_attr *qp_attr,
+				int qp_attr_mask,
+				struct ib_qp_init_attr *qp_init_attr);
+int usnic_ib_query_gid(struct ib_device *ibdev, u8 port, int index,
+				union ib_gid *gid);
+int usnic_ib_query_pkey(struct ib_device *ibdev, u8 port, u16 index,
+				u16 *pkey);
+struct ib_pd *usnic_ib_alloc_pd(struct ib_device *ibdev,
+				struct ib_ucontext *context,
+				struct ib_udata *udata);
+int usnic_ib_dealloc_pd(struct ib_pd *pd);
+struct ib_qp *usnic_ib_create_qp(struct ib_pd *pd,
+					struct ib_qp_init_attr *init_attr,
+					struct ib_udata *udata);
+int usnic_ib_destroy_qp(struct ib_qp *qp);
+int usnic_ib_modify_qp(struct ib_qp *ibqp, struct ib_qp_attr *attr,
+				int attr_mask, struct ib_udata *udata);
+struct ib_cq *usnic_ib_create_cq(struct ib_device *ibdev, int entries,
+					int vector, struct ib_ucontext *context,
+					struct ib_udata *udata);
+int usnic_ib_destroy_cq(struct ib_cq *cq);
+struct ib_mr *usnic_ib_reg_mr(struct ib_pd *pd, u64 start, u64 length,
+				u64 virt_addr, int access_flags,
+				struct ib_udata *udata);
+int usnic_ib_dereg_mr(struct ib_mr *ibmr);
+struct ib_ucontext *usnic_ib_alloc_ucontext(struct ib_device *ibdev,
+						struct ib_udata *udata);
+int usnic_ib_dealloc_ucontext(struct ib_ucontext *ibcontext);
+int usnic_ib_mmap(struct ib_ucontext *context,
+			struct vm_area_struct *vma);
+struct ib_ah *usnic_ib_create_ah(struct ib_pd *pd,
+					struct ib_ah_attr *ah_attr);
+int usnic_ib_destroy_ah(struct ib_ah *ah);
+int usnic_ib_post_send(struct ib_qp *ibqp, struct ib_send_wr *wr,
+			struct ib_send_wr **bad_wr);
+int usnic_ib_post_recv(struct ib_qp *ibqp, struct ib_recv_wr *wr,
+			struct ib_recv_wr **bad_wr);
+int usnic_ib_poll_cq(struct ib_cq *ibcq, int num_entries,
+			struct ib_wc *wc);
+int usnic_ib_req_notify_cq(struct ib_cq *cq,
+				enum ib_cq_notify_flags flags);
+struct ib_mr *usnic_ib_get_dma_mr(struct ib_pd *pd, int acc);
+#endif /* !USNIC_IB_VERBS_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_log.h b/drivers/infiniband/hw/usnic/usnic_log.h
new file mode 100644
index 00000000000..75777a66c68
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_log.h
@@ -0,0 +1,58 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_LOG_H_
+#define USNIC_LOG_H_
+
+#include "usnic.h"
+
+extern unsigned int usnic_log_lvl;
+
+#define USNIC_LOG_LVL_NONE		(0)
+#define USNIC_LOG_LVL_ERR		(1)
+#define USNIC_LOG_LVL_INFO		(2)
+#define USNIC_LOG_LVL_DBG		(3)
+
+#define usnic_printk(lvl, args...) \
+	do { \
+		printk(lvl "%s:%s:%d: ", DRV_NAME, __func__, \
+				__LINE__); \
+		printk(args); \
+	} while (0)
+
+#define usnic_dbg(args...) \
+	do { \
+		if (unlikely(usnic_log_lvl >= USNIC_LOG_LVL_DBG)) { \
+			usnic_printk(KERN_INFO, args); \
+	} \
+} while (0)
+
+#define usnic_info(args...) \
+do { \
+	if (usnic_log_lvl >= USNIC_LOG_LVL_INFO) { \
+			usnic_printk(KERN_INFO, args); \
+	} \
+} while (0)
+
+#define usnic_err(args...) \
+	do { \
+		if (usnic_log_lvl >= USNIC_LOG_LVL_ERR) { \
+			usnic_printk(KERN_ERR, args); \
+		} \
+	} while (0)
+#endif /* !USNIC_LOG_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_transport.c b/drivers/infiniband/hw/usnic/usnic_transport.c
new file mode 100644
index 00000000000..ddef6f77a78
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_transport.c
@@ -0,0 +1,202 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/bitmap.h>
+#include <linux/file.h>
+#include <linux/module.h>
+#include <linux/slab.h>
+#include <net/inet_sock.h>
+
+#include "usnic_transport.h"
+#include "usnic_log.h"
+
+/* ROCE */
+static unsigned long *roce_bitmap;
+static u16 roce_next_port = 1;
+#define ROCE_BITMAP_SZ ((1 << (8 /*CHAR_BIT*/ * sizeof(u16)))/8 /*CHAR BIT*/)
+static DEFINE_SPINLOCK(roce_bitmap_lock);
+
+const char *usnic_transport_to_str(enum usnic_transport_type type)
+{
+	switch (type) {
+	case USNIC_TRANSPORT_UNKNOWN:
+		return "Unknown";
+	case USNIC_TRANSPORT_ROCE_CUSTOM:
+		return "roce custom";
+	case USNIC_TRANSPORT_IPV4_UDP:
+		return "IPv4 UDP";
+	case USNIC_TRANSPORT_MAX:
+		return "Max?";
+	default:
+		return "Not known";
+	}
+}
+
+int usnic_transport_sock_to_str(char *buf, int buf_sz,
+					struct socket *sock)
+{
+	int err;
+	uint32_t addr;
+	uint16_t port;
+	int proto;
+
+	memset(buf, 0, buf_sz);
+	err = usnic_transport_sock_get_addr(sock, &proto, &addr, &port);
+	if (err)
+		return 0;
+
+	return scnprintf(buf, buf_sz, "Proto:%u Addr:%pI4h Port:%hu",
+			proto, &addr, port);
+}
+
+/*
+ * reserve a port number.  if "0" specified, we will try to pick one
+ * starting at roce_next_port.  roce_next_port will take on the values
+ * 1..4096
+ */
+u16 usnic_transport_rsrv_port(enum usnic_transport_type type, u16 port_num)
+{
+	if (type == USNIC_TRANSPORT_ROCE_CUSTOM) {
+		spin_lock(&roce_bitmap_lock);
+		if (!port_num) {
+			port_num = bitmap_find_next_zero_area(roce_bitmap,
+						ROCE_BITMAP_SZ,
+						roce_next_port /* start */,
+						1 /* nr */,
+						0 /* align */);
+			roce_next_port = (port_num & 4095) + 1;
+		} else if (test_bit(port_num, roce_bitmap)) {
+			usnic_err("Failed to allocate port for %s\n",
+					usnic_transport_to_str(type));
+			spin_unlock(&roce_bitmap_lock);
+			goto out_fail;
+		}
+		bitmap_set(roce_bitmap, port_num, 1);
+		spin_unlock(&roce_bitmap_lock);
+	} else {
+		usnic_err("Failed to allocate port - transport %s unsupported\n",
+				usnic_transport_to_str(type));
+		goto out_fail;
+	}
+
+	usnic_dbg("Allocating port %hu for %s\n", port_num,
+			usnic_transport_to_str(type));
+	return port_num;
+
+out_fail:
+	return 0;
+}
+
+void usnic_transport_unrsrv_port(enum usnic_transport_type type, u16 port_num)
+{
+	if (type == USNIC_TRANSPORT_ROCE_CUSTOM) {
+		spin_lock(&roce_bitmap_lock);
+		if (!port_num) {
+			usnic_err("Unreserved unvalid port num 0 for %s\n",
+					usnic_transport_to_str(type));
+			goto out_roce_custom;
+		}
+
+		if (!test_bit(port_num, roce_bitmap)) {
+			usnic_err("Unreserving invalid %hu for %s\n",
+					port_num,
+					usnic_transport_to_str(type));
+			goto out_roce_custom;
+		}
+		bitmap_clear(roce_bitmap, port_num, 1);
+		usnic_dbg("Freeing port %hu for %s\n", port_num,
+				usnic_transport_to_str(type));
+out_roce_custom:
+		spin_unlock(&roce_bitmap_lock);
+	} else {
+		usnic_err("Freeing invalid port %hu for %d\n", port_num, type);
+	}
+}
+
+struct socket *usnic_transport_get_socket(int sock_fd)
+{
+	struct socket *sock;
+	int err;
+	char buf[25];
+
+	/* sockfd_lookup will internally do a fget */
+	sock = sockfd_lookup(sock_fd, &err);
+	if (!sock) {
+		usnic_err("Unable to lookup socket for fd %d with err %d\n",
+				sock_fd, err);
+		return ERR_PTR(-ENOENT);
+	}
+
+	usnic_transport_sock_to_str(buf, sizeof(buf), sock);
+	usnic_dbg("Get sock %s\n", buf);
+
+	return sock;
+}
+
+void usnic_transport_put_socket(struct socket *sock)
+{
+	char buf[100];
+
+	usnic_transport_sock_to_str(buf, sizeof(buf), sock);
+	usnic_dbg("Put sock %s\n", buf);
+	sockfd_put(sock);
+}
+
+int usnic_transport_sock_get_addr(struct socket *sock, int *proto,
+					uint32_t *addr, uint16_t *port)
+{
+	int len;
+	int err;
+	struct sockaddr_in sock_addr;
+
+	err = sock->ops->getname(sock,
+				(struct sockaddr *)&sock_addr,
+				&len, 0);
+	if (err)
+		return err;
+
+	if (sock_addr.sin_family != AF_INET)
+		return -EINVAL;
+
+	if (proto)
+		*proto = sock->sk->sk_protocol;
+	if (port)
+		*port = ntohs(((struct sockaddr_in *)&sock_addr)->sin_port);
+	if (addr)
+		*addr = ntohl(((struct sockaddr_in *)
+					&sock_addr)->sin_addr.s_addr);
+
+	return 0;
+}
+
+int usnic_transport_init(void)
+{
+	roce_bitmap = kzalloc(ROCE_BITMAP_SZ, GFP_KERNEL);
+	if (!roce_bitmap) {
+		usnic_err("Failed to allocate bit map");
+		return -ENOMEM;
+	}
+
+	/* Do not ever allocate bit 0, hence set it here */
+	bitmap_set(roce_bitmap, 0, 1);
+	return 0;
+}
+
+void usnic_transport_fini(void)
+{
+	kfree(roce_bitmap);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_transport.h b/drivers/infiniband/hw/usnic/usnic_transport.h
new file mode 100644
index 00000000000..7e5dc6d9f46
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_transport.h
@@ -0,0 +1,51 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_TRANSPORT_H_
+#define USNIC_TRANSPORT_H_
+
+#include "usnic_abi.h"
+
+const char *usnic_transport_to_str(enum usnic_transport_type trans_type);
+/*
+ * Returns number of bytes written, excluding null terminator. If
+ * nothing was written, the function returns 0.
+ */
+int usnic_transport_sock_to_str(char *buf, int buf_sz,
+					struct socket *sock);
+/*
+ * Reserve a port. If "port_num" is set, then the function will try
+ * to reserve that particular port.
+ */
+u16 usnic_transport_rsrv_port(enum usnic_transport_type type, u16 port_num);
+void usnic_transport_unrsrv_port(enum usnic_transport_type type, u16 port_num);
+/*
+ * Do a fget on the socket refered to by sock_fd and returns the socket.
+ * Socket will not be destroyed before usnic_transport_put_socket has
+ * been called.
+ */
+struct socket *usnic_transport_get_socket(int sock_fd);
+void usnic_transport_put_socket(struct socket *sock);
+/*
+ * Call usnic_transport_get_socket before calling *_sock_get_addr
+ */
+int usnic_transport_sock_get_addr(struct socket *sock, int *proto,
+					uint32_t *addr, uint16_t *port);
+int usnic_transport_init(void);
+void usnic_transport_fini(void);
+#endif /* !USNIC_TRANSPORT_H */
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.c b/drivers/infiniband/hw/usnic/usnic_uiom.c
new file mode 100644
index 00000000000..801a1d6937e
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_uiom.c
@@ -0,0 +1,604 @@
+/*
+ * Copyright (c) 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2013 Cisco Systems.  All rights reserved.
+ *
+ * This software is available to you under a choice of one of two
+ * licenses.  You may choose to be licensed under the terms of the GNU
+ * General Public License (GPL) Version 2, available from the file
+ * COPYING in the main directory of this source tree, or the
+ * OpenIB.org BSD license below:
+ *
+ *     Redistribution and use in source and binary forms, with or
+ *     without modification, are permitted provided that the following
+ *     conditions are met:
+ *
+ *      - Redistributions of source code must retain the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer.
+ *
+ *      - Redistributions in binary form must reproduce the above
+ *        copyright notice, this list of conditions and the following
+ *        disclaimer in the documentation and/or other materials
+ *        provided with the distribution.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ */
+
+#include <linux/mm.h>
+#include <linux/dma-mapping.h>
+#include <linux/sched.h>
+#include <linux/hugetlb.h>
+#include <linux/dma-attrs.h>
+#include <linux/iommu.h>
+#include <linux/workqueue.h>
+#include <linux/list.h>
+#include <linux/pci.h>
+
+#include "usnic_log.h"
+#include "usnic_uiom.h"
+#include "usnic_uiom_interval_tree.h"
+
+static struct workqueue_struct *usnic_uiom_wq;
+
+#define USNIC_UIOM_PAGE_CHUNK						\
+	((PAGE_SIZE - offsetof(struct usnic_uiom_chunk, page_list))	/\
+	((void *) &((struct usnic_uiom_chunk *) 0)->page_list[1] -	\
+	(void *) &((struct usnic_uiom_chunk *) 0)->page_list[0]))
+
+static void usnic_uiom_reg_account(struct work_struct *work)
+{
+	struct usnic_uiom_reg *umem = container_of(work,
+						struct usnic_uiom_reg, work);
+
+	down_write(&umem->mm->mmap_sem);
+	umem->mm->locked_vm -= umem->diff;
+	up_write(&umem->mm->mmap_sem);
+	mmput(umem->mm);
+	kfree(umem);
+}
+
+static int usnic_uiom_dma_fault(struct iommu_domain *domain,
+				struct device *dev,
+				unsigned long iova, int flags,
+				void *token)
+{
+	usnic_err("Device %s iommu fault domain 0x%pK va 0x%lx flags 0x%x\n",
+		dev_name(dev),
+		domain, iova, flags);
+	return -ENOSYS;
+}
+
+static void usnic_uiom_put_pages(struct list_head *chunk_list, int dirty)
+{
+	struct usnic_uiom_chunk *chunk, *tmp;
+	struct page *page;
+	struct scatterlist *sg;
+	int i;
+	dma_addr_t pa;
+
+	list_for_each_entry_safe(chunk, tmp, chunk_list, list) {
+		for_each_sg(chunk->page_list, sg, chunk->nents, i) {
+			page = sg_page(sg);
+			pa = sg_phys(sg);
+			if (dirty)
+				set_page_dirty_lock(page);
+			put_page(page);
+			usnic_dbg("pa: %pa\n", &pa);
+		}
+		kfree(chunk);
+	}
+}
+
+static int usnic_uiom_get_pages(unsigned long addr, size_t size, int writable,
+				int dmasync, struct list_head *chunk_list)
+{
+	struct page **page_list;
+	struct scatterlist *sg;
+	struct usnic_uiom_chunk *chunk;
+	unsigned long locked;
+	unsigned long lock_limit;
+	unsigned long cur_base;
+	unsigned long npages;
+	int ret;
+	int off;
+	int i;
+	int flags;
+	dma_addr_t pa;
+	DEFINE_DMA_ATTRS(attrs);
+
+	if (dmasync)
+		dma_set_attr(DMA_ATTR_WRITE_BARRIER, &attrs);
+
+	if (!can_do_mlock())
+		return -EPERM;
+
+	INIT_LIST_HEAD(chunk_list);
+
+	page_list = (struct page **) __get_free_page(GFP_KERNEL);
+	if (!page_list)
+		return -ENOMEM;
+
+	npages = PAGE_ALIGN(size + (addr & ~PAGE_MASK)) >> PAGE_SHIFT;
+
+	down_write(&current->mm->mmap_sem);
+
+	locked = npages + current->mm->locked_vm;
+	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+
+	if ((locked > lock_limit) && !capable(CAP_IPC_LOCK)) {
+		ret = -ENOMEM;
+		goto out;
+	}
+
+	flags = IOMMU_READ | IOMMU_CACHE;
+	flags |= (writable) ? IOMMU_WRITE : 0;
+	cur_base = addr & PAGE_MASK;
+	ret = 0;
+
+	while (npages) {
+		ret = get_user_pages(current, current->mm, cur_base,
+					min_t(unsigned long, npages,
+					PAGE_SIZE / sizeof(struct page *)),
+					1, !writable, page_list, NULL);
+
+		if (ret < 0)
+			goto out;
+
+		npages -= ret;
+		off = 0;
+
+		while (ret) {
+			chunk = kmalloc(sizeof(*chunk) +
+					sizeof(struct scatterlist) *
+					min_t(int, ret, USNIC_UIOM_PAGE_CHUNK),
+					GFP_KERNEL);
+			if (!chunk) {
+				ret = -ENOMEM;
+				goto out;
+			}
+
+			chunk->nents = min_t(int, ret, USNIC_UIOM_PAGE_CHUNK);
+			sg_init_table(chunk->page_list, chunk->nents);
+			for_each_sg(chunk->page_list, sg, chunk->nents, i) {
+				sg_set_page(sg, page_list[i + off],
+						PAGE_SIZE, 0);
+				pa = sg_phys(sg);
+				usnic_dbg("va: 0x%lx pa: %pa\n",
+						cur_base + i*PAGE_SIZE, &pa);
+			}
+			cur_base += chunk->nents * PAGE_SIZE;
+			ret -= chunk->nents;
+			off += chunk->nents;
+			list_add_tail(&chunk->list, chunk_list);
+		}
+
+		ret = 0;
+	}
+
+out:
+	if (ret < 0)
+		usnic_uiom_put_pages(chunk_list, 0);
+	else
+		current->mm->locked_vm = locked;
+
+	up_write(&current->mm->mmap_sem);
+	free_page((unsigned long) page_list);
+	return ret;
+}
+
+static void usnic_uiom_unmap_sorted_intervals(struct list_head *intervals,
+						struct usnic_uiom_pd *pd)
+{
+	struct usnic_uiom_interval_node *interval, *tmp;
+	long unsigned va, size;
+
+	list_for_each_entry_safe(interval, tmp, intervals, link) {
+		va = interval->start << PAGE_SHIFT;
+		size = ((interval->last - interval->start) + 1) << PAGE_SHIFT;
+		while (size > 0) {
+			/* Workaround for RH 970401 */
+			usnic_dbg("va 0x%lx size 0x%lx", va, PAGE_SIZE);
+			iommu_unmap(pd->domain, va, PAGE_SIZE);
+			va += PAGE_SIZE;
+			size -= PAGE_SIZE;
+		}
+	}
+}
+
+static void __usnic_uiom_reg_release(struct usnic_uiom_pd *pd,
+					struct usnic_uiom_reg *uiomr,
+					int dirty)
+{
+	int npages;
+	unsigned long vpn_start, vpn_last;
+	struct usnic_uiom_interval_node *interval, *tmp;
+	int writable = 0;
+	LIST_HEAD(rm_intervals);
+
+	npages = PAGE_ALIGN(uiomr->length + uiomr->offset) >> PAGE_SHIFT;
+	vpn_start = (uiomr->va & PAGE_MASK) >> PAGE_SHIFT;
+	vpn_last = vpn_start + npages - 1;
+
+	spin_lock(&pd->lock);
+	usnic_uiom_remove_interval(&pd->rb_root, vpn_start,
+					vpn_last, &rm_intervals);
+	usnic_uiom_unmap_sorted_intervals(&rm_intervals, pd);
+
+	list_for_each_entry_safe(interval, tmp, &rm_intervals, link) {
+		if (interval->flags & IOMMU_WRITE)
+			writable = 1;
+		list_del(&interval->link);
+		kfree(interval);
+	}
+
+	usnic_uiom_put_pages(&uiomr->chunk_list, dirty & writable);
+	spin_unlock(&pd->lock);
+}
+
+static int usnic_uiom_map_sorted_intervals(struct list_head *intervals,
+						struct usnic_uiom_reg *uiomr)
+{
+	int i, err;
+	size_t size;
+	struct usnic_uiom_chunk *chunk;
+	struct usnic_uiom_interval_node *interval_node;
+	dma_addr_t pa;
+	dma_addr_t pa_start = 0;
+	dma_addr_t pa_end = 0;
+	long int va_start = -EINVAL;
+	struct usnic_uiom_pd *pd = uiomr->pd;
+	long int va = uiomr->va & PAGE_MASK;
+	int flags = IOMMU_READ | IOMMU_CACHE;
+
+	flags |= (uiomr->writable) ? IOMMU_WRITE : 0;
+	chunk = list_first_entry(&uiomr->chunk_list, struct usnic_uiom_chunk,
+									list);
+	list_for_each_entry(interval_node, intervals, link) {
+iter_chunk:
+		for (i = 0; i < chunk->nents; i++, va += PAGE_SIZE) {
+			pa = sg_phys(&chunk->page_list[i]);
+			if ((va >> PAGE_SHIFT) < interval_node->start)
+				continue;
+
+			if ((va >> PAGE_SHIFT) == interval_node->start) {
+				/* First page of the interval */
+				va_start = va;
+				pa_start = pa;
+				pa_end = pa;
+			}
+
+			WARN_ON(va_start == -EINVAL);
+
+			if ((pa_end + PAGE_SIZE != pa) &&
+					(pa != pa_start)) {
+				/* PAs are not contiguous */
+				size = pa_end - pa_start + PAGE_SIZE;
+				usnic_dbg("va 0x%lx pa %pa size 0x%zx flags 0x%x",
+					va_start, &pa_start, size, flags);
+				err = iommu_map(pd->domain, va_start, pa_start,
+							size, flags);
+				if (err) {
+					usnic_err("Failed to map va 0x%lx pa %pa size 0x%zx with err %d\n",
+						va_start, &pa_start, size, err);
+					goto err_out;
+				}
+				va_start = va;
+				pa_start = pa;
+				pa_end = pa;
+			}
+
+			if ((va >> PAGE_SHIFT) == interval_node->last) {
+				/* Last page of the interval */
+				size = pa - pa_start + PAGE_SIZE;
+				usnic_dbg("va 0x%lx pa %pa size 0x%zx flags 0x%x\n",
+					va_start, &pa_start, size, flags);
+				err = iommu_map(pd->domain, va_start, pa_start,
+						size, flags);
+				if (err) {
+					usnic_err("Failed to map va 0x%lx pa %pa size 0x%zx with err %d\n",
+						va_start, &pa_start, size, err);
+					goto err_out;
+				}
+				break;
+			}
+
+			if (pa != pa_start)
+				pa_end += PAGE_SIZE;
+		}
+
+		if (i == chunk->nents) {
+			/*
+			 * Hit last entry of the chunk,
+			 * hence advance to next chunk
+			 */
+			chunk = list_first_entry(&chunk->list,
+							struct usnic_uiom_chunk,
+							list);
+			goto iter_chunk;
+		}
+	}
+
+	return 0;
+
+err_out:
+	usnic_uiom_unmap_sorted_intervals(intervals, pd);
+	return err;
+}
+
+struct usnic_uiom_reg *usnic_uiom_reg_get(struct usnic_uiom_pd *pd,
+						unsigned long addr, size_t size,
+						int writable, int dmasync)
+{
+	struct usnic_uiom_reg *uiomr;
+	unsigned long va_base, vpn_start, vpn_last;
+	unsigned long npages;
+	int offset, err;
+	LIST_HEAD(sorted_diff_intervals);
+
+	/*
+	 * Intel IOMMU map throws an error if a translation entry is
+	 * changed from read to write.  This module may not unmap
+	 * and then remap the entry after fixing the permission
+	 * b/c this open up a small windows where hw DMA may page fault
+	 * Hence, make all entries to be writable.
+	 */
+	writable = 1;
+
+	va_base = addr & PAGE_MASK;
+	offset = addr & ~PAGE_MASK;
+	npages = PAGE_ALIGN(size + offset) >> PAGE_SHIFT;
+	vpn_start = (addr & PAGE_MASK) >> PAGE_SHIFT;
+	vpn_last = vpn_start + npages - 1;
+
+	uiomr = kmalloc(sizeof(*uiomr), GFP_KERNEL);
+	if (!uiomr)
+		return ERR_PTR(-ENOMEM);
+
+	uiomr->va = va_base;
+	uiomr->offset = offset;
+	uiomr->length = size;
+	uiomr->writable = writable;
+	uiomr->pd = pd;
+
+	err = usnic_uiom_get_pages(addr, size, writable, dmasync,
+					&uiomr->chunk_list);
+	if (err) {
+		usnic_err("Failed get_pages vpn [0x%lx,0x%lx] err %d\n",
+				vpn_start, vpn_last, err);
+		goto out_free_uiomr;
+	}
+
+	spin_lock(&pd->lock);
+	err = usnic_uiom_get_intervals_diff(vpn_start, vpn_last,
+						(writable) ? IOMMU_WRITE : 0,
+						IOMMU_WRITE,
+						&pd->rb_root,
+						&sorted_diff_intervals);
+	if (err) {
+		usnic_err("Failed disjoint interval vpn [0x%lx,0x%lx] err %d\n",
+						vpn_start, vpn_last, err);
+		goto out_put_pages;
+	}
+
+	err = usnic_uiom_map_sorted_intervals(&sorted_diff_intervals, uiomr);
+	if (err) {
+		usnic_err("Failed map interval vpn [0x%lx,0x%lx] err %d\n",
+						vpn_start, vpn_last, err);
+		goto out_put_intervals;
+
+	}
+
+	err = usnic_uiom_insert_interval(&pd->rb_root, vpn_start, vpn_last,
+					(writable) ? IOMMU_WRITE : 0);
+	if (err) {
+		usnic_err("Failed insert interval vpn [0x%lx,0x%lx] err %d\n",
+						vpn_start, vpn_last, err);
+		goto out_unmap_intervals;
+	}
+
+	usnic_uiom_put_interval_set(&sorted_diff_intervals);
+	spin_unlock(&pd->lock);
+
+	return uiomr;
+
+out_unmap_intervals:
+	usnic_uiom_unmap_sorted_intervals(&sorted_diff_intervals, pd);
+out_put_intervals:
+	usnic_uiom_put_interval_set(&sorted_diff_intervals);
+out_put_pages:
+	usnic_uiom_put_pages(&uiomr->chunk_list, 0);
+	spin_unlock(&pd->lock);
+out_free_uiomr:
+	kfree(uiomr);
+	return ERR_PTR(err);
+}
+
+void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, int closing)
+{
+	struct mm_struct *mm;
+	unsigned long diff;
+
+	__usnic_uiom_reg_release(uiomr->pd, uiomr, 1);
+
+	mm = get_task_mm(current);
+	if (!mm) {
+		kfree(uiomr);
+		return;
+	}
+
+	diff = PAGE_ALIGN(uiomr->length + uiomr->offset) >> PAGE_SHIFT;
+
+	/*
+	 * We may be called with the mm's mmap_sem already held.  This
+	 * can happen when a userspace munmap() is the call that drops
+	 * the last reference to our file and calls our release
+	 * method.  If there are memory regions to destroy, we'll end
+	 * up here and not be able to take the mmap_sem.  In that case
+	 * we defer the vm_locked accounting to the system workqueue.
+	 */
+	if (closing) {
+		if (!down_write_trylock(&mm->mmap_sem)) {
+			INIT_WORK(&uiomr->work, usnic_uiom_reg_account);
+			uiomr->mm = mm;
+			uiomr->diff = diff;
+
+			queue_work(usnic_uiom_wq, &uiomr->work);
+			return;
+		}
+	} else
+		down_write(&mm->mmap_sem);
+
+	current->mm->locked_vm -= diff;
+	up_write(&mm->mmap_sem);
+	mmput(mm);
+	kfree(uiomr);
+}
+
+struct usnic_uiom_pd *usnic_uiom_alloc_pd(void)
+{
+	struct usnic_uiom_pd *pd;
+	void *domain;
+
+	pd = kzalloc(sizeof(*pd), GFP_KERNEL);
+	if (!pd)
+		return ERR_PTR(-ENOMEM);
+
+	pd->domain = domain = iommu_domain_alloc(&pci_bus_type);
+	if (IS_ERR_OR_NULL(domain)) {
+		usnic_err("Failed to allocate IOMMU domain with err %ld\n",
+				PTR_ERR(pd->domain));
+		kfree(pd);
+		return ERR_PTR(domain ? PTR_ERR(domain) : -ENOMEM);
+	}
+
+	iommu_set_fault_handler(pd->domain, usnic_uiom_dma_fault, NULL);
+
+	spin_lock_init(&pd->lock);
+	INIT_LIST_HEAD(&pd->devs);
+
+	return pd;
+}
+
+void usnic_uiom_dealloc_pd(struct usnic_uiom_pd *pd)
+{
+	iommu_domain_free(pd->domain);
+	kfree(pd);
+}
+
+int usnic_uiom_attach_dev_to_pd(struct usnic_uiom_pd *pd, struct device *dev)
+{
+	struct usnic_uiom_dev *uiom_dev;
+	int err;
+
+	uiom_dev = kzalloc(sizeof(*uiom_dev), GFP_ATOMIC);
+	if (!uiom_dev)
+		return -ENOMEM;
+	uiom_dev->dev = dev;
+
+	err = iommu_attach_device(pd->domain, dev);
+	if (err)
+		goto out_free_dev;
+
+	if (!iommu_domain_has_cap(pd->domain, IOMMU_CAP_CACHE_COHERENCY)) {
+		usnic_err("IOMMU of %s does not support cache coherency\n",
+				dev_name(dev));
+		err = -EINVAL;
+		goto out_detach_device;
+	}
+
+	spin_lock(&pd->lock);
+	list_add_tail(&uiom_dev->link, &pd->devs);
+	pd->dev_cnt++;
+	spin_unlock(&pd->lock);
+
+	return 0;
+
+out_detach_device:
+	iommu_detach_device(pd->domain, dev);
+out_free_dev:
+	kfree(uiom_dev);
+	return err;
+}
+
+void usnic_uiom_detach_dev_from_pd(struct usnic_uiom_pd *pd, struct device *dev)
+{
+	struct usnic_uiom_dev *uiom_dev;
+	int found = 0;
+
+	spin_lock(&pd->lock);
+	list_for_each_entry(uiom_dev, &pd->devs, link) {
+		if (uiom_dev->dev == dev) {
+			found = 1;
+			break;
+		}
+	}
+
+	if (!found) {
+		usnic_err("Unable to free dev %s - not found\n",
+				dev_name(dev));
+		spin_unlock(&pd->lock);
+		return;
+	}
+
+	list_del(&uiom_dev->link);
+	pd->dev_cnt--;
+	spin_unlock(&pd->lock);
+
+	return iommu_detach_device(pd->domain, dev);
+}
+
+struct device **usnic_uiom_get_dev_list(struct usnic_uiom_pd *pd)
+{
+	struct usnic_uiom_dev *uiom_dev;
+	struct device **devs;
+	int i = 0;
+
+	spin_lock(&pd->lock);
+	devs = kcalloc(pd->dev_cnt + 1, sizeof(*devs), GFP_ATOMIC);
+	if (!devs) {
+		devs = ERR_PTR(-ENOMEM);
+		goto out;
+	}
+
+	list_for_each_entry(uiom_dev, &pd->devs, link) {
+		devs[i++] = uiom_dev->dev;
+	}
+out:
+	spin_unlock(&pd->lock);
+	return devs;
+}
+
+void usnic_uiom_free_dev_list(struct device **devs)
+{
+	kfree(devs);
+}
+
+int usnic_uiom_init(char *drv_name)
+{
+	if (!iommu_present(&pci_bus_type)) {
+		usnic_err("IOMMU required but not present or enabled.  USNIC QPs will not function w/o enabling IOMMU\n");
+		return -EPERM;
+	}
+
+	usnic_uiom_wq = create_workqueue(drv_name);
+	if (!usnic_uiom_wq) {
+		usnic_err("Unable to alloc wq for drv %s\n", drv_name);
+		return -ENOMEM;
+	}
+
+	return 0;
+}
+
+void usnic_uiom_fini(void)
+{
+	flush_workqueue(usnic_uiom_wq);
+	destroy_workqueue(usnic_uiom_wq);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom.h b/drivers/infiniband/hw/usnic/usnic_uiom.h
new file mode 100644
index 00000000000..70440996e8f
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_uiom.h
@@ -0,0 +1,80 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_UIOM_H_
+#define USNIC_UIOM_H_
+
+#include <linux/list.h>
+#include <linux/scatterlist.h>
+
+#include "usnic_uiom_interval_tree.h"
+
+#define USNIC_UIOM_READ			(1)
+#define USNIC_UIOM_WRITE		(2)
+
+#define USNIC_UIOM_MAX_PD_CNT		(1000)
+#define USNIC_UIOM_MAX_MR_CNT		(1000000)
+#define USNIC_UIOM_MAX_MR_SIZE		(~0UL)
+#define USNIC_UIOM_PAGE_SIZE		(PAGE_SIZE)
+
+struct usnic_uiom_dev {
+	struct device			*dev;
+	struct list_head		link;
+};
+
+struct usnic_uiom_pd {
+	struct iommu_domain		*domain;
+	spinlock_t			lock;
+	struct rb_root			rb_root;
+	struct list_head		devs;
+	int				dev_cnt;
+};
+
+struct usnic_uiom_reg {
+	struct usnic_uiom_pd		*pd;
+	unsigned long			va;
+	size_t				length;
+	int				offset;
+	int				page_size;
+	int				writable;
+	struct list_head		chunk_list;
+	struct work_struct		work;
+	struct mm_struct		*mm;
+	unsigned long			diff;
+};
+
+struct usnic_uiom_chunk {
+	struct list_head		list;
+	int				nents;
+	struct scatterlist		page_list[0];
+};
+
+struct usnic_uiom_pd *usnic_uiom_alloc_pd(void);
+void usnic_uiom_dealloc_pd(struct usnic_uiom_pd *pd);
+int usnic_uiom_attach_dev_to_pd(struct usnic_uiom_pd *pd, struct device *dev);
+void usnic_uiom_detach_dev_from_pd(struct usnic_uiom_pd *pd,
+					struct device *dev);
+struct device **usnic_uiom_get_dev_list(struct usnic_uiom_pd *pd);
+void usnic_uiom_free_dev_list(struct device **devs);
+struct usnic_uiom_reg *usnic_uiom_reg_get(struct usnic_uiom_pd *pd,
+						unsigned long addr, size_t size,
+						int access, int dmasync);
+void usnic_uiom_reg_release(struct usnic_uiom_reg *uiomr, int closing);
+int usnic_uiom_init(char *drv_name);
+void usnic_uiom_fini(void);
+#endif /* USNIC_UIOM_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c
new file mode 100644
index 00000000000..3a4288e0fba
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.c
@@ -0,0 +1,254 @@
+/*
+ * Copyright (c) 2014, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#include <linux/init.h>
+#include <linux/list.h>
+#include <linux/slab.h>
+#include <linux/list_sort.h>
+
+#include <linux/interval_tree_generic.h>
+#include "usnic_uiom_interval_tree.h"
+
+#define START(node) ((node)->start)
+#define LAST(node) ((node)->last)
+
+#define MAKE_NODE(node, start, end, ref_cnt, flags, err, err_out)	\
+		do {							\
+			node = usnic_uiom_interval_node_alloc(start,	\
+					end, ref_cnt, flags);		\
+				if (!node) {				\
+					err = -ENOMEM;			\
+					goto err_out;			\
+				}					\
+		} while (0)
+
+#define MARK_FOR_ADD(node, list) (list_add_tail(&node->link, list))
+
+#define MAKE_NODE_AND_APPEND(node, start, end, ref_cnt, flags, err,	\
+				err_out, list)				\
+				do {					\
+					MAKE_NODE(node, start, end,	\
+						ref_cnt, flags, err,	\
+						err_out);		\
+					MARK_FOR_ADD(node, list);	\
+				} while (0)
+
+#define FLAGS_EQUAL(flags1, flags2, mask)				\
+			(((flags1) & (mask)) == ((flags2) & (mask)))
+
+static struct usnic_uiom_interval_node*
+usnic_uiom_interval_node_alloc(long int start, long int last, int ref_cnt,
+				int flags)
+{
+	struct usnic_uiom_interval_node *interval = kzalloc(sizeof(*interval),
+								GFP_ATOMIC);
+	if (!interval)
+		return NULL;
+
+	interval->start = start;
+	interval->last = last;
+	interval->flags = flags;
+	interval->ref_cnt = ref_cnt;
+
+	return interval;
+}
+
+static int interval_cmp(void *priv, struct list_head *a, struct list_head *b)
+{
+	struct usnic_uiom_interval_node *node_a, *node_b;
+
+	node_a = list_entry(a, struct usnic_uiom_interval_node, link);
+	node_b = list_entry(b, struct usnic_uiom_interval_node, link);
+
+	/* long to int */
+	if (node_a->start < node_b->start)
+		return -1;
+	else if (node_a->start > node_b->start)
+		return 1;
+
+	return 0;
+}
+
+static void
+find_intervals_intersection_sorted(struct rb_root *root, unsigned long start,
+					unsigned long last,
+					struct list_head *list)
+{
+	struct usnic_uiom_interval_node *node;
+
+	INIT_LIST_HEAD(list);
+
+	for (node = usnic_uiom_interval_tree_iter_first(root, start, last);
+		node;
+		node = usnic_uiom_interval_tree_iter_next(node, start, last))
+		list_add_tail(&node->link, list);
+
+	list_sort(NULL, list, interval_cmp);
+}
+
+int usnic_uiom_get_intervals_diff(unsigned long start, unsigned long last,
+					int flags, int flag_mask,
+					struct rb_root *root,
+					struct list_head *diff_set)
+{
+	struct usnic_uiom_interval_node *interval, *tmp;
+	int err = 0;
+	long int pivot = start;
+	LIST_HEAD(intersection_set);
+
+	INIT_LIST_HEAD(diff_set);
+
+	find_intervals_intersection_sorted(root, start, last,
+						&intersection_set);
+
+	list_for_each_entry(interval, &intersection_set, link) {
+		if (pivot < interval->start) {
+			MAKE_NODE_AND_APPEND(tmp, pivot, interval->start - 1,
+						1, flags, err, err_out,
+						diff_set);
+			pivot = interval->start;
+		}
+
+		/*
+		 * Invariant: Set [start, pivot] is either in diff_set or root,
+		 * but not in both.
+		 */
+
+		if (pivot > interval->last) {
+			continue;
+		} else if (pivot <= interval->last &&
+				FLAGS_EQUAL(interval->flags, flags,
+				flag_mask)) {
+			pivot = interval->last + 1;
+		}
+	}
+
+	if (pivot <= last)
+		MAKE_NODE_AND_APPEND(tmp, pivot, last, 1, flags, err, err_out,
+					diff_set);
+
+	return 0;
+
+err_out:
+	list_for_each_entry_safe(interval, tmp, diff_set, link) {
+		list_del(&interval->link);
+		kfree(interval);
+	}
+
+	return err;
+}
+
+void usnic_uiom_put_interval_set(struct list_head *intervals)
+{
+	struct usnic_uiom_interval_node *interval, *tmp;
+	list_for_each_entry_safe(interval, tmp, intervals, link)
+		kfree(interval);
+}
+
+int usnic_uiom_insert_interval(struct rb_root *root, unsigned long start,
+				unsigned long last, int flags)
+{
+	struct usnic_uiom_interval_node *interval, *tmp;
+	unsigned long istart, ilast;
+	int iref_cnt, iflags;
+	unsigned long lpivot = start;
+	int err = 0;
+	LIST_HEAD(to_add);
+	LIST_HEAD(intersection_set);
+
+	find_intervals_intersection_sorted(root, start, last,
+						&intersection_set);
+
+	list_for_each_entry(interval, &intersection_set, link) {
+		/*
+		 * Invariant - lpivot is the left edge of next interval to be
+		 * inserted
+		 */
+		istart = interval->start;
+		ilast = interval->last;
+		iref_cnt = interval->ref_cnt;
+		iflags = interval->flags;
+
+		if (istart < lpivot) {
+			MAKE_NODE_AND_APPEND(tmp, istart, lpivot - 1, iref_cnt,
+						iflags, err, err_out, &to_add);
+		} else if (istart > lpivot) {
+			MAKE_NODE_AND_APPEND(tmp, lpivot, istart - 1, 1, flags,
+						err, err_out, &to_add);
+			lpivot = istart;
+		} else {
+			lpivot = istart;
+		}
+
+		if (ilast > last) {
+			MAKE_NODE_AND_APPEND(tmp, lpivot, last, iref_cnt + 1,
+						iflags | flags, err, err_out,
+						&to_add);
+			MAKE_NODE_AND_APPEND(tmp, last + 1, ilast, iref_cnt,
+						iflags, err, err_out, &to_add);
+		} else {
+			MAKE_NODE_AND_APPEND(tmp, lpivot, ilast, iref_cnt + 1,
+						iflags | flags, err, err_out,
+						&to_add);
+		}
+
+		lpivot = ilast + 1;
+	}
+
+	if (lpivot <= last)
+		MAKE_NODE_AND_APPEND(tmp, lpivot, last, 1, flags, err, err_out,
+					&to_add);
+
+	list_for_each_entry_safe(interval, tmp, &intersection_set, link) {
+		usnic_uiom_interval_tree_remove(interval, root);
+		kfree(interval);
+	}
+
+	list_for_each_entry(interval, &to_add, link)
+		usnic_uiom_interval_tree_insert(interval, root);
+
+	return 0;
+
+err_out:
+	list_for_each_entry_safe(interval, tmp, &to_add, link)
+		kfree(interval);
+
+	return err;
+}
+
+void usnic_uiom_remove_interval(struct rb_root *root, unsigned long start,
+				unsigned long last, struct list_head *removed)
+{
+	struct usnic_uiom_interval_node *interval;
+
+	for (interval = usnic_uiom_interval_tree_iter_first(root, start, last);
+			interval;
+			interval = usnic_uiom_interval_tree_iter_next(interval,
+									start,
+									last)) {
+		if (--interval->ref_cnt == 0)
+			list_add_tail(&interval->link, removed);
+	}
+
+	list_for_each_entry(interval, removed, link)
+		usnic_uiom_interval_tree_remove(interval, root);
+}
+
+INTERVAL_TREE_DEFINE(struct usnic_uiom_interval_node, rb,
+			unsigned long, __subtree_last,
+			START, LAST, , usnic_uiom_interval_tree)
diff --git a/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h
new file mode 100644
index 00000000000..d4f752e258f
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_uiom_interval_tree.h
@@ -0,0 +1,73 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_UIOM_INTERVAL_TREE_H_
+#define USNIC_UIOM_INTERVAL_TREE_H_
+
+#include <linux/rbtree.h>
+
+struct usnic_uiom_interval_node {
+	struct rb_node			rb;
+	struct list_head		link;
+	unsigned long			start;
+	unsigned long			last;
+	unsigned long			__subtree_last;
+	unsigned int			ref_cnt;
+	int				flags;
+};
+
+extern void
+usnic_uiom_interval_tree_insert(struct usnic_uiom_interval_node *node,
+					struct rb_root *root);
+extern void
+usnic_uiom_interval_tree_remove(struct usnic_uiom_interval_node *node,
+					struct rb_root *root);
+extern struct usnic_uiom_interval_node *
+usnic_uiom_interval_tree_iter_first(struct rb_root *root,
+					unsigned long start,
+					unsigned long last);
+extern struct usnic_uiom_interval_node *
+usnic_uiom_interval_tree_iter_next(struct usnic_uiom_interval_node *node,
+			unsigned long start, unsigned long last);
+/*
+ * Inserts {start...last} into {root}.  If there are overlaps,
+ * nodes will be broken up and merged
+ */
+int usnic_uiom_insert_interval(struct rb_root *root,
+				unsigned long start, unsigned long last,
+				int flags);
+/*
+ * Removed {start...last} from {root}.  The nodes removed are returned in
+ * 'removed.' The caller is responsibile for freeing memory of nodes in
+ * 'removed.'
+ */
+void usnic_uiom_remove_interval(struct rb_root *root,
+				unsigned long start, unsigned long last,
+				struct list_head *removed);
+/*
+ * Returns {start...last} - {root} (relative complement of {start...last} in
+ * {root}) in diff_set sorted ascendingly
+ */
+int usnic_uiom_get_intervals_diff(unsigned long start,
+					unsigned long last, int flags,
+					int flag_mask,
+					struct rb_root *root,
+					struct list_head *diff_set);
+/* Call this to free diff_set returned by usnic_uiom_get_intervals_diff */
+void usnic_uiom_put_interval_set(struct list_head *intervals);
+#endif /* USNIC_UIOM_INTERVAL_TREE_H_ */
diff --git a/drivers/infiniband/hw/usnic/usnic_vnic.c b/drivers/infiniband/hw/usnic/usnic_vnic.c
new file mode 100644
index 00000000000..656b88c39ed
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_vnic.c
@@ -0,0 +1,467 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+#include <linux/errno.h>
+#include <linux/module.h>
+#include <linux/pci.h>
+
+#include "usnic_ib.h"
+#include "vnic_resource.h"
+#include "usnic_log.h"
+#include "usnic_vnic.h"
+
+struct usnic_vnic {
+	struct vnic_dev			*vdev;
+	struct vnic_dev_bar		bar[PCI_NUM_RESOURCES];
+	struct usnic_vnic_res_chunk	chunks[USNIC_VNIC_RES_TYPE_MAX];
+	spinlock_t			res_lock;
+};
+
+static enum vnic_res_type _to_vnic_res_type(enum usnic_vnic_res_type res_type)
+{
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+		vnic_res_type,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+		vnic_res_type,
+	static enum vnic_res_type usnic_vnic_type_2_vnic_type[] = {
+						USNIC_VNIC_RES_TYPES};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+	if (res_type >= USNIC_VNIC_RES_TYPE_MAX)
+		return RES_TYPE_MAX;
+
+	return usnic_vnic_type_2_vnic_type[res_type];
+}
+
+const char *usnic_vnic_res_type_to_str(enum usnic_vnic_res_type res_type)
+{
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+		desc,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+		desc,
+	static const char * const usnic_vnic_res_type_desc[] = {
+						USNIC_VNIC_RES_TYPES};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+	if (res_type >= USNIC_VNIC_RES_TYPE_MAX)
+		return "unknown";
+
+	return usnic_vnic_res_type_desc[res_type];
+
+}
+
+const char *usnic_vnic_pci_name(struct usnic_vnic *vnic)
+{
+	return pci_name(usnic_vnic_get_pdev(vnic));
+}
+
+int usnic_vnic_dump(struct usnic_vnic *vnic, char *buf,
+			int buf_sz,
+			void *hdr_obj,
+			int (*printtitle)(void *, char*, int),
+			int (*printcols)(char *, int),
+			int (*printrow)(void *, char *, int))
+{
+	struct usnic_vnic_res_chunk *chunk;
+	struct usnic_vnic_res *res;
+	struct vnic_dev_bar *bar0;
+	int i, j, offset;
+
+	offset = 0;
+	bar0 = usnic_vnic_get_bar(vnic, 0);
+	offset += scnprintf(buf + offset, buf_sz - offset,
+			"VF:%hu BAR0 bus_addr=%pa vaddr=0x%p size=%ld ",
+			usnic_vnic_get_index(vnic),
+			&bar0->bus_addr,
+			bar0->vaddr, bar0->len);
+	if (printtitle)
+		offset += printtitle(hdr_obj, buf + offset, buf_sz - offset);
+	offset += scnprintf(buf + offset, buf_sz - offset, "\n");
+	offset += scnprintf(buf + offset, buf_sz - offset,
+			"|RES\t|CTRL_PIN\t\t|IN_USE\t");
+	if (printcols)
+		offset += printcols(buf + offset, buf_sz - offset);
+	offset += scnprintf(buf + offset, buf_sz - offset, "\n");
+
+	spin_lock(&vnic->res_lock);
+	for (i = 0; i < ARRAY_SIZE(vnic->chunks); i++) {
+		chunk = &vnic->chunks[i];
+		for (j = 0; j < chunk->cnt; j++) {
+			res = chunk->res[j];
+			offset += scnprintf(buf + offset, buf_sz - offset,
+					"|%s[%u]\t|0x%p\t|%u\t",
+					usnic_vnic_res_type_to_str(res->type),
+					res->vnic_idx, res->ctrl, !!res->owner);
+			if (printrow) {
+				offset += printrow(res->owner, buf + offset,
+							buf_sz - offset);
+			}
+			offset += scnprintf(buf + offset, buf_sz - offset,
+						"\n");
+		}
+	}
+	spin_unlock(&vnic->res_lock);
+	return offset;
+}
+
+void usnic_vnic_res_spec_update(struct usnic_vnic_res_spec *spec,
+				enum usnic_vnic_res_type trgt_type,
+				u16 cnt)
+{
+	int i;
+
+	for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+		if (spec->resources[i].type == trgt_type) {
+			spec->resources[i].cnt = cnt;
+			return;
+		}
+	}
+
+	WARN_ON(1);
+}
+
+int usnic_vnic_res_spec_satisfied(const struct usnic_vnic_res_spec *min_spec,
+					struct usnic_vnic_res_spec *res_spec)
+{
+	int found, i, j;
+
+	for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+		found = 0;
+
+		for (j = 0; j < USNIC_VNIC_RES_TYPE_MAX; j++) {
+			if (res_spec->resources[i].type !=
+				min_spec->resources[i].type)
+				continue;
+			found = 1;
+			if (min_spec->resources[i].cnt >
+					res_spec->resources[i].cnt)
+				return -EINVAL;
+			break;
+		}
+
+		if (!found)
+			return -EINVAL;
+	}
+	return 0;
+}
+
+int usnic_vnic_spec_dump(char *buf, int buf_sz,
+				struct usnic_vnic_res_spec *res_spec)
+{
+	enum usnic_vnic_res_type res_type;
+	int res_cnt;
+	int i;
+	int offset = 0;
+
+	for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+		res_type = res_spec->resources[i].type;
+		res_cnt = res_spec->resources[i].cnt;
+		offset += scnprintf(buf + offset, buf_sz - offset,
+				"Res: %s Cnt: %d ",
+				usnic_vnic_res_type_to_str(res_type),
+				res_cnt);
+	}
+
+	return offset;
+}
+
+int usnic_vnic_check_room(struct usnic_vnic *vnic,
+				struct usnic_vnic_res_spec *res_spec)
+{
+	int i;
+	enum usnic_vnic_res_type res_type;
+	int res_cnt;
+
+	for (i = 0; i < USNIC_VNIC_RES_TYPE_MAX; i++) {
+		res_type = res_spec->resources[i].type;
+		res_cnt = res_spec->resources[i].cnt;
+
+		if (res_type == USNIC_VNIC_RES_TYPE_EOL)
+			break;
+
+		if (res_cnt > usnic_vnic_res_free_cnt(vnic, res_type))
+			return -EBUSY;
+	}
+
+	return 0;
+}
+
+int usnic_vnic_res_cnt(struct usnic_vnic *vnic,
+				enum usnic_vnic_res_type type)
+{
+	return vnic->chunks[type].cnt;
+}
+
+int usnic_vnic_res_free_cnt(struct usnic_vnic *vnic,
+				enum usnic_vnic_res_type type)
+{
+	return vnic->chunks[type].free_cnt;
+}
+
+struct usnic_vnic_res_chunk *
+usnic_vnic_get_resources(struct usnic_vnic *vnic, enum usnic_vnic_res_type type,
+				int cnt, void *owner)
+{
+	struct usnic_vnic_res_chunk *src, *ret;
+	struct usnic_vnic_res *res;
+	int i;
+
+	if (usnic_vnic_res_free_cnt(vnic, type) < cnt || cnt < 1 || !owner)
+		return ERR_PTR(-EINVAL);
+
+	ret = kzalloc(sizeof(*ret), GFP_ATOMIC);
+	if (!ret) {
+		usnic_err("Failed to allocate chunk for %s - Out of memory\n",
+				usnic_vnic_pci_name(vnic));
+		return ERR_PTR(-ENOMEM);
+	}
+
+	ret->res = kzalloc(sizeof(*(ret->res))*cnt, GFP_ATOMIC);
+	if (!ret->res) {
+		usnic_err("Failed to allocate resources for %s. Out of memory\n",
+				usnic_vnic_pci_name(vnic));
+		kfree(ret);
+		return ERR_PTR(-ENOMEM);
+	}
+
+	spin_lock(&vnic->res_lock);
+	src = &vnic->chunks[type];
+	for (i = 0; i < src->cnt && ret->cnt < cnt; i++) {
+		res = src->res[i];
+		if (!res->owner) {
+			src->free_cnt--;
+			res->owner = owner;
+			ret->res[ret->cnt++] = res;
+		}
+	}
+
+	spin_unlock(&vnic->res_lock);
+	ret->type = type;
+	ret->vnic = vnic;
+	WARN_ON(ret->cnt != cnt);
+
+	return ret;
+}
+
+void usnic_vnic_put_resources(struct usnic_vnic_res_chunk *chunk)
+{
+
+	struct usnic_vnic_res *res;
+	int i;
+	struct usnic_vnic *vnic = chunk->vnic;
+
+	spin_lock(&vnic->res_lock);
+	while ((i = --chunk->cnt) >= 0) {
+		res = chunk->res[i];
+		chunk->res[i] = NULL;
+		res->owner = NULL;
+		vnic->chunks[res->type].free_cnt++;
+	}
+	spin_unlock(&vnic->res_lock);
+
+	kfree(chunk->res);
+	kfree(chunk);
+}
+
+u16 usnic_vnic_get_index(struct usnic_vnic *vnic)
+{
+	return usnic_vnic_get_pdev(vnic)->devfn - 1;
+}
+
+static int usnic_vnic_alloc_res_chunk(struct usnic_vnic *vnic,
+					enum usnic_vnic_res_type type,
+					struct usnic_vnic_res_chunk *chunk)
+{
+	int cnt, err, i;
+	struct usnic_vnic_res *res;
+
+	cnt = vnic_dev_get_res_count(vnic->vdev, _to_vnic_res_type(type));
+	if (cnt < 1)
+		return -EINVAL;
+
+	chunk->cnt = chunk->free_cnt = cnt;
+	chunk->res = kzalloc(sizeof(*(chunk->res))*cnt, GFP_KERNEL);
+	if (!chunk->res)
+		return -ENOMEM;
+
+	for (i = 0; i < cnt; i++) {
+		res = kzalloc(sizeof(*res), GFP_KERNEL);
+		if (!res) {
+			err = -ENOMEM;
+			goto fail;
+		}
+		res->type = type;
+		res->vnic_idx = i;
+		res->vnic = vnic;
+		res->ctrl = vnic_dev_get_res(vnic->vdev,
+						_to_vnic_res_type(type), i);
+		chunk->res[i] = res;
+	}
+
+	chunk->vnic = vnic;
+	return 0;
+fail:
+	for (i--; i >= 0; i--)
+		kfree(chunk->res[i]);
+	kfree(chunk->res);
+	return err;
+}
+
+static void usnic_vnic_free_res_chunk(struct usnic_vnic_res_chunk *chunk)
+{
+	int i;
+	for (i = 0; i < chunk->cnt; i++)
+		kfree(chunk->res[i]);
+	kfree(chunk->res);
+}
+
+static int usnic_vnic_discover_resources(struct pci_dev *pdev,
+						struct usnic_vnic *vnic)
+{
+	enum usnic_vnic_res_type res_type;
+	int i;
+	int err = 0;
+
+	for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+			continue;
+		vnic->bar[i].len = pci_resource_len(pdev, i);
+		vnic->bar[i].vaddr = pci_iomap(pdev, i, vnic->bar[i].len);
+		if (!vnic->bar[i].vaddr) {
+			usnic_err("Cannot memory-map BAR %d, aborting\n",
+					i);
+			err = -ENODEV;
+			goto out_clean_bar;
+		}
+		vnic->bar[i].bus_addr = pci_resource_start(pdev, i);
+	}
+
+	vnic->vdev = vnic_dev_register(NULL, pdev, pdev, vnic->bar,
+			ARRAY_SIZE(vnic->bar));
+	if (!vnic->vdev) {
+		usnic_err("Failed to register device %s\n",
+				pci_name(pdev));
+		err = -EINVAL;
+		goto out_clean_bar;
+	}
+
+	for (res_type = USNIC_VNIC_RES_TYPE_EOL + 1;
+			res_type < USNIC_VNIC_RES_TYPE_MAX; res_type++) {
+		err = usnic_vnic_alloc_res_chunk(vnic, res_type,
+						&vnic->chunks[res_type]);
+		if (err) {
+			usnic_err("Failed to alloc res %s with err %d\n",
+					usnic_vnic_res_type_to_str(res_type),
+					err);
+			goto out_clean_chunks;
+		}
+	}
+
+	return 0;
+
+out_clean_chunks:
+	for (res_type--; res_type > USNIC_VNIC_RES_TYPE_EOL; res_type--)
+		usnic_vnic_free_res_chunk(&vnic->chunks[res_type]);
+	vnic_dev_unregister(vnic->vdev);
+out_clean_bar:
+	for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+			continue;
+		if (!vnic->bar[i].vaddr)
+			break;
+
+		iounmap(vnic->bar[i].vaddr);
+	}
+
+	return err;
+}
+
+struct pci_dev *usnic_vnic_get_pdev(struct usnic_vnic *vnic)
+{
+	return vnic_dev_get_pdev(vnic->vdev);
+}
+
+struct vnic_dev_bar *usnic_vnic_get_bar(struct usnic_vnic *vnic,
+				int bar_num)
+{
+	return (bar_num < ARRAY_SIZE(vnic->bar)) ? &vnic->bar[bar_num] : NULL;
+}
+
+static void usnic_vnic_release_resources(struct usnic_vnic *vnic)
+{
+	int i;
+	struct pci_dev *pdev;
+	enum usnic_vnic_res_type res_type;
+
+	pdev = usnic_vnic_get_pdev(vnic);
+
+	for (res_type = USNIC_VNIC_RES_TYPE_EOL + 1;
+			res_type < USNIC_VNIC_RES_TYPE_MAX; res_type++)
+		usnic_vnic_free_res_chunk(&vnic->chunks[res_type]);
+
+	vnic_dev_unregister(vnic->vdev);
+
+	for (i = 0; i < ARRAY_SIZE(vnic->bar); i++) {
+		if (!(pci_resource_flags(pdev, i) & IORESOURCE_MEM))
+			continue;
+		iounmap(vnic->bar[i].vaddr);
+	}
+}
+
+struct usnic_vnic *usnic_vnic_alloc(struct pci_dev *pdev)
+{
+	struct usnic_vnic *vnic;
+	int err = 0;
+
+	if (!pci_is_enabled(pdev)) {
+		usnic_err("PCI dev %s is disabled\n", pci_name(pdev));
+		return ERR_PTR(-EINVAL);
+	}
+
+	vnic = kzalloc(sizeof(*vnic), GFP_KERNEL);
+	if (!vnic) {
+		usnic_err("Failed to alloc vnic for %s - out of memory\n",
+				pci_name(pdev));
+		return ERR_PTR(-ENOMEM);
+	}
+
+	spin_lock_init(&vnic->res_lock);
+
+	err = usnic_vnic_discover_resources(pdev, vnic);
+	if (err) {
+		usnic_err("Failed to discover %s resources with err %d\n",
+				pci_name(pdev), err);
+		goto out_free_vnic;
+	}
+
+	usnic_dbg("Allocated vnic for %s\n", usnic_vnic_pci_name(vnic));
+
+	return vnic;
+
+out_free_vnic:
+	kfree(vnic);
+
+	return ERR_PTR(err);
+}
+
+void usnic_vnic_free(struct usnic_vnic *vnic)
+{
+	usnic_vnic_release_resources(vnic);
+	kfree(vnic);
+}
diff --git a/drivers/infiniband/hw/usnic/usnic_vnic.h b/drivers/infiniband/hw/usnic/usnic_vnic.h
new file mode 100644
index 00000000000..14d931a8829
--- /dev/null
+++ b/drivers/infiniband/hw/usnic/usnic_vnic.h
@@ -0,0 +1,103 @@
+/*
+ * Copyright (c) 2013, Cisco Systems, Inc. All rights reserved.
+ *
+ * This program is free software; you may redistribute it and/or modify
+ * it under the terms of the GNU General Public License as published by
+ * the Free Software Foundation; version 2 of the License.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+ * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
+ * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
+ * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
+ * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+ * SOFTWARE.
+ *
+ */
+
+#ifndef USNIC_VNIC_H_
+#define USNIC_VNIC_H_
+
+#include <linux/pci.h>
+
+#include "vnic_dev.h"
+
+/*                      =USNIC_VNIC_RES_TYPE= =VNIC_RES=   =DESC= */
+#define USNIC_VNIC_RES_TYPES \
+	DEFINE_USNIC_VNIC_RES_AT(EOL, RES_TYPE_EOL, "EOL", 0) \
+	DEFINE_USNIC_VNIC_RES(WQ, RES_TYPE_WQ, "WQ") \
+	DEFINE_USNIC_VNIC_RES(RQ, RES_TYPE_RQ, "RQ") \
+	DEFINE_USNIC_VNIC_RES(CQ, RES_TYPE_CQ, "CQ") \
+	DEFINE_USNIC_VNIC_RES(INTR, RES_TYPE_INTR_CTRL, "INT") \
+	DEFINE_USNIC_VNIC_RES(MAX, RES_TYPE_MAX, "MAX")\
+
+#define DEFINE_USNIC_VNIC_RES_AT(usnic_vnic_res_t, vnic_res_type, desc, val) \
+	USNIC_VNIC_RES_TYPE_##usnic_vnic_res_t = val,
+#define DEFINE_USNIC_VNIC_RES(usnic_vnic_res_t, vnic_res_type, desc) \
+	USNIC_VNIC_RES_TYPE_##usnic_vnic_res_t,
+enum usnic_vnic_res_type {
+	USNIC_VNIC_RES_TYPES
+};
+#undef DEFINE_USNIC_VNIC_RES
+#undef DEFINE_USNIC_VNIC_RES_AT
+
+struct usnic_vnic_res {
+	enum usnic_vnic_res_type	type;
+	unsigned int			vnic_idx;
+	struct usnic_vnic		*vnic;
+	void __iomem			*ctrl;
+	void				*owner;
+};
+
+struct usnic_vnic_res_chunk {
+	enum usnic_vnic_res_type	type;
+	int				cnt;
+	int				free_cnt;
+	struct usnic_vnic_res		**res;
+	struct usnic_vnic		*vnic;
+};
+
+struct usnic_vnic_res_desc {
+	enum usnic_vnic_res_type	type;
+	uint16_t			cnt;
+};
+
+struct usnic_vnic_res_spec {
+	struct usnic_vnic_res_desc resources[USNIC_VNIC_RES_TYPE_MAX];
+};
+
+const char *usnic_vnic_res_type_to_str(enum usnic_vnic_res_type res_type);
+const char *usnic_vnic_pci_name(struct usnic_vnic *vnic);
+int usnic_vnic_dump(struct usnic_vnic *vnic, char *buf, int buf_sz,
+			void *hdr_obj,
+			int (*printtitle)(void *, char*, int),
+			int (*printcols)(char *, int),
+			int (*printrow)(void *, char *, int));
+void usnic_vnic_res_spec_update(struct usnic_vnic_res_spec *spec,
+				enum usnic_vnic_res_type trgt_type,
+				u16 cnt);
+int usnic_vnic_res_spec_satisfied(const struct usnic_vnic_res_spec *min_spec,
+					struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_spec_dump(char *buf, int buf_sz,
+				struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_check_room(struct usnic_vnic *vnic,
+				struct usnic_vnic_res_spec *res_spec);
+int usnic_vnic_res_cnt(struct usnic_vnic *vnic,
+				enum usnic_vnic_res_type type);
+int usnic_vnic_res_free_cnt(struct usnic_vnic *vnic,
+				enum usnic_vnic_res_type type);
+struct usnic_vnic_res_chunk *
+usnic_vnic_get_resources(struct usnic_vnic *vnic,
+				enum usnic_vnic_res_type type,
+				int cnt,
+				void *owner);
+void usnic_vnic_put_resources(struct usnic_vnic_res_chunk *chunk);
+struct pci_dev *usnic_vnic_get_pdev(struct usnic_vnic *vnic);
+struct vnic_dev_bar *usnic_vnic_get_bar(struct usnic_vnic *vnic,
+				int bar_num);
+struct usnic_vnic *usnic_vnic_alloc(struct pci_dev *pdev);
+void usnic_vnic_free(struct usnic_vnic *vnic);
+u16 usnic_vnic_get_index(struct usnic_vnic *vnic);
+
+#endif /*!USNIC_VNIC_H_*/