1 files changed, 753 insertions, 274 deletions

diff --git a/drivers/infiniband/ulp/ipoib/ipoib_ib.c b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
index eee82363167..6a7003ddb0b 100644
--- a/drivers/infiniband/ulp/ipoib/ipoib_ib.c
+++ b/drivers/infiniband/ulp/ipoib/ipoib_ib.c
@@ -1,5 +1,8 @@
 /*
  * Copyright (c) 2004, 2005 Topspin Communications.  All rights reserved.
+ * Copyright (c) 2005 Sun Microsystems, Inc. All rights reserved.
+ * Copyright (c) 2005 Mellanox Technologies. All rights reserved.
+ * Copyright (c) 2004, 2005 Voltaire, 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
@@ -28,14 +31,15 @@
  * 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.
- *
- * $Id: ipoib_ib.c 1386 2004-12-27 16:23:17Z roland $
  */
 
 #include <linux/delay.h>
+#include <linux/moduleparam.h>
 #include <linux/dma-mapping.h>
+#include <linux/slab.h>
 
-#include <ib_cache.h>
+#include <linux/ip.h>
+#include <linux/tcp.h>
 
 #include "ipoib.h"
 
@@ -47,29 +51,30 @@ MODULE_PARM_DESC(data_debug_level,
 		 "Enable data path debug tracing if > 0");
 #endif
 
-#define	IPOIB_OP_RECV	(1ul << 31)
-
-static DECLARE_MUTEX(pkey_sem);
+static DEFINE_MUTEX(pkey_mutex);
 
 struct ipoib_ah *ipoib_create_ah(struct net_device *dev,
 				 struct ib_pd *pd, struct ib_ah_attr *attr)
 {
 	struct ipoib_ah *ah;
+	struct ib_ah *vah;
 
 	ah = kmalloc(sizeof *ah, GFP_KERNEL);
 	if (!ah)
-		return NULL;
+		return ERR_PTR(-ENOMEM);
 
 	ah->dev       = dev;
 	ah->last_send = 0;
 	kref_init(&ah->ref);
 
-	ah->ah = ib_create_ah(pd, attr);
-	if (IS_ERR(ah->ah)) {
+	vah = ib_create_ah(pd, attr);
+	if (IS_ERR(vah)) {
 		kfree(ah);
-		ah = NULL;
-	} else
+		ah = (struct ipoib_ah *)vah;
+	} else {
+		ah->ah = vah;
 		ipoib_dbg(netdev_priv(dev), "Created ah %p\n", ah->ah);
+	}
 
 	return ah;
 }
@@ -81,76 +86,136 @@ void ipoib_free_ah(struct kref *kref)
 
 	unsigned long flags;
 
-	if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
-		ipoib_dbg(priv, "Freeing ah %p\n", ah->ah);
-		ib_destroy_ah(ah->ah);
-		kfree(ah);
-	} else {
-		spin_lock_irqsave(&priv->lock, flags);
-		list_add_tail(&ah->list, &priv->dead_ahs);
-		spin_unlock_irqrestore(&priv->lock, flags);
-	}
-}
-
-static inline int ipoib_ib_receive(struct ipoib_dev_priv *priv,
-				   unsigned int wr_id,
-				   dma_addr_t addr)
-{
-	struct ib_sge list = {
-		.addr    = addr,
-		.length  = IPOIB_BUF_SIZE,
-		.lkey    = priv->mr->lkey,
-	};
-	struct ib_recv_wr param = {
-		.wr_id 	    = wr_id | IPOIB_OP_RECV,
-		.sg_list    = &list,
-		.num_sge    = 1,
-	};
-	struct ib_recv_wr *bad_wr;
+	spin_lock_irqsave(&priv->lock, flags);
+	list_add_tail(&ah->list, &priv->dead_ahs);
+	spin_unlock_irqrestore(&priv->lock, flags);
+}
+
+static void ipoib_ud_dma_unmap_rx(struct ipoib_dev_priv *priv,
+				  u64 mapping[IPOIB_UD_RX_SG])
+{
+	if (ipoib_ud_need_sg(priv->max_ib_mtu)) {
+		ib_dma_unmap_single(priv->ca, mapping[0], IPOIB_UD_HEAD_SIZE,
+				    DMA_FROM_DEVICE);
+		ib_dma_unmap_page(priv->ca, mapping[1], PAGE_SIZE,
+				  DMA_FROM_DEVICE);
+	} else
+		ib_dma_unmap_single(priv->ca, mapping[0],
+				    IPOIB_UD_BUF_SIZE(priv->max_ib_mtu),
+				    DMA_FROM_DEVICE);
+}
+
+static void ipoib_ud_skb_put_frags(struct ipoib_dev_priv *priv,
+				   struct sk_buff *skb,
+				   unsigned int length)
+{
+	if (ipoib_ud_need_sg(priv->max_ib_mtu)) {
+		skb_frag_t *frag = &skb_shinfo(skb)->frags[0];
+		unsigned int size;
+		/*
+		 * There is only two buffers needed for max_payload = 4K,
+		 * first buf size is IPOIB_UD_HEAD_SIZE
+		 */
+		skb->tail += IPOIB_UD_HEAD_SIZE;
+		skb->len  += length;
+
+		size = length - IPOIB_UD_HEAD_SIZE;
+
+		skb_frag_size_set(frag, size);
+		skb->data_len += size;
+		skb->truesize += PAGE_SIZE;
+	} else
+		skb_put(skb, length);
 
-	return ib_post_recv(priv->qp, &param, &bad_wr);
 }
 
 static int ipoib_ib_post_receive(struct net_device *dev, int id)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	struct sk_buff *skb;
-	dma_addr_t addr;
+	struct ib_recv_wr *bad_wr;
 	int ret;
 
-	skb = dev_alloc_skb(IPOIB_BUF_SIZE + 4);
-	if (!skb) {
-		ipoib_warn(priv, "failed to allocate receive buffer\n");
+	priv->rx_wr.wr_id   = id | IPOIB_OP_RECV;
+	priv->rx_sge[0].addr = priv->rx_ring[id].mapping[0];
+	priv->rx_sge[1].addr = priv->rx_ring[id].mapping[1];
 
-		priv->rx_ring[id].skb = NULL;
-		return -ENOMEM;
-	}
-	skb_reserve(skb, 4);	/* 16 byte align IP header */
-	priv->rx_ring[id].skb = skb;
-	addr = dma_map_single(priv->ca->dma_device,
-			      skb->data, IPOIB_BUF_SIZE,
-			      DMA_FROM_DEVICE);
-	pci_unmap_addr_set(&priv->rx_ring[id], mapping, addr);
 
-	ret = ipoib_ib_receive(priv, id, addr);
-	if (ret) {
-		ipoib_warn(priv, "ipoib_ib_receive failed for buf %d (%d)\n",
-			   id, ret);
-		dma_unmap_single(priv->ca->dma_device, addr,
-				 IPOIB_BUF_SIZE, DMA_FROM_DEVICE);
-		dev_kfree_skb_any(skb);
+	ret = ib_post_recv(priv->qp, &priv->rx_wr, &bad_wr);
+	if (unlikely(ret)) {
+		ipoib_warn(priv, "receive failed for buf %d (%d)\n", id, ret);
+		ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[id].mapping);
+		dev_kfree_skb_any(priv->rx_ring[id].skb);
 		priv->rx_ring[id].skb = NULL;
 	}
 
 	return ret;
 }
 
+static struct sk_buff *ipoib_alloc_rx_skb(struct net_device *dev, int id)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct sk_buff *skb;
+	int buf_size;
+	int tailroom;
+	u64 *mapping;
+
+	if (ipoib_ud_need_sg(priv->max_ib_mtu)) {
+		buf_size = IPOIB_UD_HEAD_SIZE;
+		tailroom = 128; /* reserve some tailroom for IP/TCP headers */
+	} else {
+		buf_size = IPOIB_UD_BUF_SIZE(priv->max_ib_mtu);
+		tailroom = 0;
+	}
+
+	skb = dev_alloc_skb(buf_size + tailroom + 4);
+	if (unlikely(!skb))
+		return NULL;
+
+	/*
+	 * IB will leave a 40 byte gap for a GRH and IPoIB adds a 4 byte
+	 * header.  So we need 4 more bytes to get to 48 and align the
+	 * IP header to a multiple of 16.
+	 */
+	skb_reserve(skb, 4);
+
+	mapping = priv->rx_ring[id].mapping;
+	mapping[0] = ib_dma_map_single(priv->ca, skb->data, buf_size,
+				       DMA_FROM_DEVICE);
+	if (unlikely(ib_dma_mapping_error(priv->ca, mapping[0])))
+		goto error;
+
+	if (ipoib_ud_need_sg(priv->max_ib_mtu)) {
+		struct page *page = alloc_page(GFP_ATOMIC);
+		if (!page)
+			goto partial_error;
+		skb_fill_page_desc(skb, 0, page, 0, PAGE_SIZE);
+		mapping[1] =
+			ib_dma_map_page(priv->ca, page,
+					0, PAGE_SIZE, DMA_FROM_DEVICE);
+		if (unlikely(ib_dma_mapping_error(priv->ca, mapping[1])))
+			goto partial_error;
+	}
+
+	priv->rx_ring[id].skb = skb;
+	return skb;
+
+partial_error:
+	ib_dma_unmap_single(priv->ca, mapping[0], buf_size, DMA_FROM_DEVICE);
+error:
+	dev_kfree_skb_any(skb);
+	return NULL;
+}
+
 static int ipoib_ib_post_receives(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int i;
 
-	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i) {
+	for (i = 0; i < ipoib_recvq_size; ++i) {
+		if (!ipoib_alloc_rx_skb(dev, i)) {
+			ipoib_warn(priv, "failed to allocate receive buffer %d\n", i);
+			return -ENOMEM;
+		}
 		if (ipoib_ib_post_receive(dev, i)) {
 			ipoib_warn(priv, "ipoib_ib_post_receive failed for buf %d\n", i);
 			return -EIO;
@@ -160,137 +225,316 @@ static int ipoib_ib_post_receives(struct net_device *dev)
 	return 0;
 }
 
-static void ipoib_ib_handle_wc(struct net_device *dev,
-			       struct ib_wc *wc)
+static void ipoib_ib_handle_rx_wc(struct net_device *dev, struct ib_wc *wc)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	unsigned int wr_id = wc->wr_id & ~IPOIB_OP_RECV;
+	struct sk_buff *skb;
+	u64 mapping[IPOIB_UD_RX_SG];
+	union ib_gid *dgid;
+
+	ipoib_dbg_data(priv, "recv completion: id %d, status: %d\n",
+		       wr_id, wc->status);
+
+	if (unlikely(wr_id >= ipoib_recvq_size)) {
+		ipoib_warn(priv, "recv completion event with wrid %d (> %d)\n",
+			   wr_id, ipoib_recvq_size);
+		return;
+	}
+
+	skb  = priv->rx_ring[wr_id].skb;
+
+	if (unlikely(wc->status != IB_WC_SUCCESS)) {
+		if (wc->status != IB_WC_WR_FLUSH_ERR)
+			ipoib_warn(priv, "failed recv event "
+				   "(status=%d, wrid=%d vend_err %x)\n",
+				   wc->status, wr_id, wc->vendor_err);
+		ipoib_ud_dma_unmap_rx(priv, priv->rx_ring[wr_id].mapping);
+		dev_kfree_skb_any(skb);
+		priv->rx_ring[wr_id].skb = NULL;
+		return;
+	}
+
+	/*
+	 * Drop packets that this interface sent, ie multicast packets
+	 * that the HCA has replicated.
+	 */
+	if (wc->slid == priv->local_lid && wc->src_qp == priv->qp->qp_num)
+		goto repost;
+
+	memcpy(mapping, priv->rx_ring[wr_id].mapping,
+	       IPOIB_UD_RX_SG * sizeof *mapping);
+
+	/*
+	 * If we can't allocate a new RX buffer, dump
+	 * this packet and reuse the old buffer.
+	 */
+	if (unlikely(!ipoib_alloc_rx_skb(dev, wr_id))) {
+		++dev->stats.rx_dropped;
+		goto repost;
+	}
+
+	ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
+		       wc->byte_len, wc->slid);
+
+	ipoib_ud_dma_unmap_rx(priv, mapping);
+	ipoib_ud_skb_put_frags(priv, skb, wc->byte_len);
+
+	/* First byte of dgid signals multicast when 0xff */
+	dgid = &((struct ib_grh *)skb->data)->dgid;
+
+	if (!(wc->wc_flags & IB_WC_GRH) || dgid->raw[0] != 0xff)
+		skb->pkt_type = PACKET_HOST;
+	else if (memcmp(dgid, dev->broadcast + 4, sizeof(union ib_gid)) == 0)
+		skb->pkt_type = PACKET_BROADCAST;
+	else
+		skb->pkt_type = PACKET_MULTICAST;
+
+	skb_pull(skb, IB_GRH_BYTES);
+
+	skb->protocol = ((struct ipoib_header *) skb->data)->proto;
+	skb_reset_mac_header(skb);
+	skb_pull(skb, IPOIB_ENCAP_LEN);
+
+	++dev->stats.rx_packets;
+	dev->stats.rx_bytes += skb->len;
+
+	skb->dev = dev;
+	if ((dev->features & NETIF_F_RXCSUM) &&
+			likely(wc->wc_flags & IB_WC_IP_CSUM_OK))
+		skb->ip_summed = CHECKSUM_UNNECESSARY;
+
+	napi_gro_receive(&priv->napi, skb);
+
+repost:
+	if (unlikely(ipoib_ib_post_receive(dev, wr_id)))
+		ipoib_warn(priv, "ipoib_ib_post_receive failed "
+			   "for buf %d\n", wr_id);
+}
+
+static int ipoib_dma_map_tx(struct ib_device *ca,
+			    struct ipoib_tx_buf *tx_req)
+{
+	struct sk_buff *skb = tx_req->skb;
+	u64 *mapping = tx_req->mapping;
+	int i;
+	int off;
+
+	if (skb_headlen(skb)) {
+		mapping[0] = ib_dma_map_single(ca, skb->data, skb_headlen(skb),
+					       DMA_TO_DEVICE);
+		if (unlikely(ib_dma_mapping_error(ca, mapping[0])))
+			return -EIO;
+
+		off = 1;
+	} else
+		off = 0;
+
+	for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) {
+		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+		mapping[i + off] = ib_dma_map_page(ca,
+						 skb_frag_page(frag),
+						 frag->page_offset, skb_frag_size(frag),
+						 DMA_TO_DEVICE);
+		if (unlikely(ib_dma_mapping_error(ca, mapping[i + off])))
+			goto partial_error;
+	}
+	return 0;
+
+partial_error:
+	for (; i > 0; --i) {
+		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1];
+
+		ib_dma_unmap_page(ca, mapping[i - !off], skb_frag_size(frag), DMA_TO_DEVICE);
+	}
+
+	if (off)
+		ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE);
+
+	return -EIO;
+}
+
+static void ipoib_dma_unmap_tx(struct ib_device *ca,
+			       struct ipoib_tx_buf *tx_req)
+{
+	struct sk_buff *skb = tx_req->skb;
+	u64 *mapping = tx_req->mapping;
+	int i;
+	int off;
+
+	if (skb_headlen(skb)) {
+		ib_dma_unmap_single(ca, mapping[0], skb_headlen(skb), DMA_TO_DEVICE);
+		off = 1;
+	} else
+		off = 0;
+
+	for (i = 0; i < skb_shinfo(skb)->nr_frags; ++i) {
+		const skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+
+		ib_dma_unmap_page(ca, mapping[i + off], skb_frag_size(frag),
+				  DMA_TO_DEVICE);
+	}
+}
+
+static void ipoib_ib_handle_tx_wc(struct net_device *dev, struct ib_wc *wc)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	unsigned int wr_id = wc->wr_id;
+	struct ipoib_tx_buf *tx_req;
 
-	ipoib_dbg_data(priv, "called: id %d, op %d, status: %d\n",
-		       wr_id, wc->opcode, wc->status);
+	ipoib_dbg_data(priv, "send completion: id %d, status: %d\n",
+		       wr_id, wc->status);
 
-	if (wr_id & IPOIB_OP_RECV) {
-		wr_id &= ~IPOIB_OP_RECV;
+	if (unlikely(wr_id >= ipoib_sendq_size)) {
+		ipoib_warn(priv, "send completion event with wrid %d (> %d)\n",
+			   wr_id, ipoib_sendq_size);
+		return;
+	}
 
-		if (wr_id < IPOIB_RX_RING_SIZE) {
-			struct sk_buff *skb = priv->rx_ring[wr_id].skb;
+	tx_req = &priv->tx_ring[wr_id];
 
-			priv->rx_ring[wr_id].skb = NULL;
+	ipoib_dma_unmap_tx(priv->ca, tx_req);
 
-			dma_unmap_single(priv->ca->dma_device,
-					 pci_unmap_addr(&priv->rx_ring[wr_id],
-							mapping),
-					 IPOIB_BUF_SIZE,
-					 DMA_FROM_DEVICE);
+	++dev->stats.tx_packets;
+	dev->stats.tx_bytes += tx_req->skb->len;
 
-			if (wc->status != IB_WC_SUCCESS) {
-				if (wc->status != IB_WC_WR_FLUSH_ERR)
-					ipoib_warn(priv, "failed recv event "
-						   "(status=%d, wrid=%d vend_err %x)\n",
-						   wc->status, wr_id, wc->vendor_err);
-				dev_kfree_skb_any(skb);
-				return;
-			}
+	dev_kfree_skb_any(tx_req->skb);
 
-			ipoib_dbg_data(priv, "received %d bytes, SLID 0x%04x\n",
-				       wc->byte_len, wc->slid);
-
-			skb_put(skb, wc->byte_len);
-			skb_pull(skb, IB_GRH_BYTES);
-
-			if (wc->slid != priv->local_lid ||
-			    wc->src_qp != priv->qp->qp_num) {
-				skb->protocol = ((struct ipoib_header *) skb->data)->proto;
-				skb->mac.raw = skb->data;
-				skb_pull(skb, IPOIB_ENCAP_LEN);
-
-				dev->last_rx = jiffies;
-				++priv->stats.rx_packets;
-				priv->stats.rx_bytes += skb->len;
-
-				skb->dev = dev;
-				/* XXX get correct PACKET_ type here */
-				skb->pkt_type = PACKET_HOST;
-				netif_rx_ni(skb);
-			} else {
-				ipoib_dbg_data(priv, "dropping loopback packet\n");
-				dev_kfree_skb_any(skb);
-			}
+	++priv->tx_tail;
+	if (unlikely(--priv->tx_outstanding == ipoib_sendq_size >> 1) &&
+	    netif_queue_stopped(dev) &&
+	    test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+		netif_wake_queue(dev);
+
+	if (wc->status != IB_WC_SUCCESS &&
+	    wc->status != IB_WC_WR_FLUSH_ERR)
+		ipoib_warn(priv, "failed send event "
+			   "(status=%d, wrid=%d vend_err %x)\n",
+			   wc->status, wr_id, wc->vendor_err);
+}
 
-			/* repost receive */
-			if (ipoib_ib_post_receive(dev, wr_id))
-				ipoib_warn(priv, "ipoib_ib_post_receive failed "
-					   "for buf %d\n", wr_id);
-		} else
-			ipoib_warn(priv, "completion event with wrid %d\n",
-				   wr_id);
+static int poll_tx(struct ipoib_dev_priv *priv)
+{
+	int n, i;
 
-	} else {
-		struct ipoib_buf *tx_req;
-		unsigned long flags;
+	n = ib_poll_cq(priv->send_cq, MAX_SEND_CQE, priv->send_wc);
+	for (i = 0; i < n; ++i)
+		ipoib_ib_handle_tx_wc(priv->dev, priv->send_wc + i);
 
-		if (wr_id >= IPOIB_TX_RING_SIZE) {
-			ipoib_warn(priv, "completion event with wrid %d (> %d)\n",
-				   wr_id, IPOIB_TX_RING_SIZE);
-			return;
-		}
+	return n == MAX_SEND_CQE;
+}
+
+int ipoib_poll(struct napi_struct *napi, int budget)
+{
+	struct ipoib_dev_priv *priv = container_of(napi, struct ipoib_dev_priv, napi);
+	struct net_device *dev = priv->dev;
+	int done;
+	int t;
+	int n, i;
 
-		ipoib_dbg_data(priv, "send complete, wrid %d\n", wr_id);
+	done  = 0;
 
-		tx_req = &priv->tx_ring[wr_id];
+poll_more:
+	while (done < budget) {
+		int max = (budget - done);
 
-		dma_unmap_single(priv->ca->dma_device,
-				 pci_unmap_addr(tx_req, mapping),
-				 tx_req->skb->len,
-				 DMA_TO_DEVICE);
+		t = min(IPOIB_NUM_WC, max);
+		n = ib_poll_cq(priv->recv_cq, t, priv->ibwc);
 
-		++priv->stats.tx_packets;
-		priv->stats.tx_bytes += tx_req->skb->len;
+		for (i = 0; i < n; i++) {
+			struct ib_wc *wc = priv->ibwc + i;
 
-		dev_kfree_skb_any(tx_req->skb);
+			if (wc->wr_id & IPOIB_OP_RECV) {
+				++done;
+				if (wc->wr_id & IPOIB_OP_CM)
+					ipoib_cm_handle_rx_wc(dev, wc);
+				else
+					ipoib_ib_handle_rx_wc(dev, wc);
+			} else
+				ipoib_cm_handle_tx_wc(priv->dev, wc);
+		}
 
-		spin_lock_irqsave(&priv->tx_lock, flags);
-		++priv->tx_tail;
-		if (netif_queue_stopped(dev) &&
-		    priv->tx_head - priv->tx_tail <= IPOIB_TX_RING_SIZE / 2)
-			netif_wake_queue(dev);
-		spin_unlock_irqrestore(&priv->tx_lock, flags);
+		if (n != t)
+			break;
+	}
 
-		if (wc->status != IB_WC_SUCCESS &&
-		    wc->status != IB_WC_WR_FLUSH_ERR)
-			ipoib_warn(priv, "failed send event "
-				   "(status=%d, wrid=%d vend_err %x)\n",
-				   wc->status, wr_id, wc->vendor_err);
+	if (done < budget) {
+		napi_complete(napi);
+		if (unlikely(ib_req_notify_cq(priv->recv_cq,
+					      IB_CQ_NEXT_COMP |
+					      IB_CQ_REPORT_MISSED_EVENTS)) &&
+		    napi_reschedule(napi))
+			goto poll_more;
 	}
+
+	return done;
 }
 
 void ipoib_ib_completion(struct ib_cq *cq, void *dev_ptr)
 {
-	struct net_device *dev = (struct net_device *) dev_ptr;
+	struct net_device *dev = dev_ptr;
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+
+	napi_schedule(&priv->napi);
+}
+
+static void drain_tx_cq(struct net_device *dev)
+{
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	int n, i;
 
-	ib_req_notify_cq(cq, IB_CQ_NEXT_COMP);
-	do {
-		n = ib_poll_cq(cq, IPOIB_NUM_WC, priv->ibwc);
-		for (i = 0; i < n; ++i)
-			ipoib_ib_handle_wc(dev, priv->ibwc + i);
-	} while (n == IPOIB_NUM_WC);
+	netif_tx_lock(dev);
+	while (poll_tx(priv))
+		; /* nothing */
+
+	if (netif_queue_stopped(dev))
+		mod_timer(&priv->poll_timer, jiffies + 1);
+
+	netif_tx_unlock(dev);
+}
+
+void ipoib_send_comp_handler(struct ib_cq *cq, void *dev_ptr)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev_ptr);
+
+	mod_timer(&priv->poll_timer, jiffies);
 }
 
 static inline int post_send(struct ipoib_dev_priv *priv,
 			    unsigned int wr_id,
 			    struct ib_ah *address, u32 qpn,
-			    dma_addr_t addr, int len)
+			    struct ipoib_tx_buf *tx_req,
+			    void *head, int hlen)
 {
 	struct ib_send_wr *bad_wr;
+	int i, off;
+	struct sk_buff *skb = tx_req->skb;
+	skb_frag_t *frags = skb_shinfo(skb)->frags;
+	int nr_frags = skb_shinfo(skb)->nr_frags;
+	u64 *mapping = tx_req->mapping;
+
+	if (skb_headlen(skb)) {
+		priv->tx_sge[0].addr         = mapping[0];
+		priv->tx_sge[0].length       = skb_headlen(skb);
+		off = 1;
+	} else
+		off = 0;
 
-	priv->tx_sge.addr             = addr;
-	priv->tx_sge.length           = len;
-
-	priv->tx_wr.wr_id 	      = wr_id;
-	priv->tx_wr.wr.ud.remote_qpn  = qpn;
-	priv->tx_wr.wr.ud.ah 	      = address;
+	for (i = 0; i < nr_frags; ++i) {
+		priv->tx_sge[i + off].addr = mapping[i + off];
+		priv->tx_sge[i + off].length = skb_frag_size(&frags[i]);
+	}
+	priv->tx_wr.num_sge	     = nr_frags + off;
+	priv->tx_wr.wr_id 	     = wr_id;
+	priv->tx_wr.wr.ud.remote_qpn = qpn;
+	priv->tx_wr.wr.ud.ah 	     = address;
+
+	if (head) {
+		priv->tx_wr.wr.ud.mss	 = skb_shinfo(skb)->gso_size;
+		priv->tx_wr.wr.ud.header = head;
+		priv->tx_wr.wr.ud.hlen	 = hlen;
+		priv->tx_wr.opcode	 = IB_WR_LSO;
+	} else
+		priv->tx_wr.opcode	 = IB_WR_SEND;
 
 	return ib_post_send(priv->qp, &priv->tx_wr, &bad_wr);
 }
@@ -299,16 +543,31 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 		struct ipoib_ah *address, u32 qpn)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	struct ipoib_buf *tx_req;
-	dma_addr_t addr;
-
-	if (skb->len > dev->mtu + INFINIBAND_ALEN) {
-		ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
-			   skb->len, dev->mtu + INFINIBAND_ALEN);
-		++priv->stats.tx_dropped;
-		++priv->stats.tx_errors;
-		dev_kfree_skb_any(skb);
-		return;
+	struct ipoib_tx_buf *tx_req;
+	int hlen, rc;
+	void *phead;
+
+	if (skb_is_gso(skb)) {
+		hlen = skb_transport_offset(skb) + tcp_hdrlen(skb);
+		phead = skb->data;
+		if (unlikely(!skb_pull(skb, hlen))) {
+			ipoib_warn(priv, "linear data too small\n");
+			++dev->stats.tx_dropped;
+			++dev->stats.tx_errors;
+			dev_kfree_skb_any(skb);
+			return;
+		}
+	} else {
+		if (unlikely(skb->len > priv->mcast_mtu + IPOIB_ENCAP_LEN)) {
+			ipoib_warn(priv, "packet len %d (> %d) too long to send, dropping\n",
+				   skb->len, priv->mcast_mtu + IPOIB_ENCAP_LEN);
+			++dev->stats.tx_dropped;
+			++dev->stats.tx_errors;
+			ipoib_cm_skb_too_long(dev, skb, priv->mcast_mtu);
+			return;
+		}
+		phead = NULL;
+		hlen  = 0;
 	}
 
 	ipoib_dbg_data(priv, "sending packet, length=%d address=%p qpn=0x%06x\n",
@@ -321,30 +580,49 @@ void ipoib_send(struct net_device *dev, struct sk_buff *skb,
 	 * means we have to make sure everything is properly recorded and
 	 * our state is consistent before we call post_send().
 	 */
-	tx_req = &priv->tx_ring[priv->tx_head & (IPOIB_TX_RING_SIZE - 1)];
+	tx_req = &priv->tx_ring[priv->tx_head & (ipoib_sendq_size - 1)];
 	tx_req->skb = skb;
-	addr = dma_map_single(priv->ca->dma_device, skb->data, skb->len,
-			      DMA_TO_DEVICE);
-	pci_unmap_addr_set(tx_req, mapping, addr);
-
-	if (unlikely(post_send(priv, priv->tx_head & (IPOIB_TX_RING_SIZE - 1),
-			       address->ah, qpn, addr, skb->len))) {
-		ipoib_warn(priv, "post_send failed\n");
-		++priv->stats.tx_errors;
-		dma_unmap_single(priv->ca->dma_device, addr, skb->len,
-				 DMA_TO_DEVICE);
+	if (unlikely(ipoib_dma_map_tx(priv->ca, tx_req))) {
+		++dev->stats.tx_errors;
+		dev_kfree_skb_any(skb);
+		return;
+	}
+
+	if (skb->ip_summed == CHECKSUM_PARTIAL)
+		priv->tx_wr.send_flags |= IB_SEND_IP_CSUM;
+	else
+		priv->tx_wr.send_flags &= ~IB_SEND_IP_CSUM;
+
+	if (++priv->tx_outstanding == ipoib_sendq_size) {
+		ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
+		if (ib_req_notify_cq(priv->send_cq, IB_CQ_NEXT_COMP))
+			ipoib_warn(priv, "request notify on send CQ failed\n");
+		netif_stop_queue(dev);
+	}
+
+	skb_orphan(skb);
+	skb_dst_drop(skb);
+
+	rc = post_send(priv, priv->tx_head & (ipoib_sendq_size - 1),
+		       address->ah, qpn, tx_req, phead, hlen);
+	if (unlikely(rc)) {
+		ipoib_warn(priv, "post_send failed, error %d\n", rc);
+		++dev->stats.tx_errors;
+		--priv->tx_outstanding;
+		ipoib_dma_unmap_tx(priv->ca, tx_req);
 		dev_kfree_skb_any(skb);
+		if (netif_queue_stopped(dev))
+			netif_wake_queue(dev);
 	} else {
 		dev->trans_start = jiffies;
 
 		address->last_send = priv->tx_head;
 		++priv->tx_head;
-
-		if (priv->tx_head - priv->tx_tail == IPOIB_TX_RING_SIZE) {
-			ipoib_dbg(priv, "TX ring full, stopping kernel net queue\n");
-			netif_stop_queue(dev);
-		}
 	}
+
+	if (unlikely(priv->tx_outstanding > MAX_SEND_CQE))
+		while (poll_tx(priv))
+			; /* nothing */
 }
 
 static void __ipoib_reap_ah(struct net_device *dev)
@@ -352,31 +630,38 @@ static void __ipoib_reap_ah(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ipoib_ah *ah, *tah;
 	LIST_HEAD(remove_list);
+	unsigned long flags;
+
+	netif_tx_lock_bh(dev);
+	spin_lock_irqsave(&priv->lock, flags);
 
-	spin_lock_irq(&priv->lock);
 	list_for_each_entry_safe(ah, tah, &priv->dead_ahs, list)
 		if ((int) priv->tx_tail - (int) ah->last_send >= 0) {
 			list_del(&ah->list);
-			list_add_tail(&ah->list, &remove_list);
+			ib_destroy_ah(ah->ah);
+			kfree(ah);
 		}
-	spin_unlock_irq(&priv->lock);
 
-	list_for_each_entry_safe(ah, tah, &remove_list, list) {
-		ipoib_dbg(priv, "Reaping ah %p\n", ah->ah);
-		ib_destroy_ah(ah->ah);
-		kfree(ah);
-	}
+	spin_unlock_irqrestore(&priv->lock, flags);
+	netif_tx_unlock_bh(dev);
 }
 
-void ipoib_reap_ah(void *dev_ptr)
+void ipoib_reap_ah(struct work_struct *work)
 {
-	struct net_device *dev = dev_ptr;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, ah_reap_task.work);
+	struct net_device *dev = priv->dev;
 
 	__ipoib_reap_ah(dev);
 
 	if (!test_bit(IPOIB_STOP_REAPER, &priv->flags))
-		queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
+		queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
+				   round_jiffies_relative(HZ));
+}
+
+static void ipoib_ib_tx_timer_func(unsigned long ctx)
+{
+	drain_tx_cq((struct net_device *)ctx);
 }
 
 int ipoib_ib_dev_open(struct net_device *dev)
@@ -384,34 +669,74 @@ int ipoib_ib_dev_open(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	int ret;
 
-	ret = ipoib_qp_create(dev);
+	if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &priv->pkey_index)) {
+		ipoib_warn(priv, "P_Key 0x%04x not found\n", priv->pkey);
+		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+		return -1;
+	}
+	set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+
+	ret = ipoib_init_qp(dev);
 	if (ret) {
-		ipoib_warn(priv, "ipoib_qp_create returned %d\n", ret);
+		ipoib_warn(priv, "ipoib_init_qp returned %d\n", ret);
 		return -1;
 	}
 
 	ret = ipoib_ib_post_receives(dev);
 	if (ret) {
 		ipoib_warn(priv, "ipoib_ib_post_receives returned %d\n", ret);
-		return -1;
+		goto dev_stop;
+	}
+
+	ret = ipoib_cm_dev_open(dev);
+	if (ret) {
+		ipoib_warn(priv, "ipoib_cm_dev_open returned %d\n", ret);
+		goto dev_stop;
 	}
 
 	clear_bit(IPOIB_STOP_REAPER, &priv->flags);
-	queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task, HZ);
+	queue_delayed_work(ipoib_workqueue, &priv->ah_reap_task,
+			   round_jiffies_relative(HZ));
+
+	if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
+		napi_enable(&priv->napi);
 
 	return 0;
+dev_stop:
+	if (!test_and_set_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
+		napi_enable(&priv->napi);
+	ipoib_ib_dev_stop(dev, 1);
+	return -1;
+}
+
+static void ipoib_pkey_dev_check_presence(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	u16 pkey_index = 0;
+
+	if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
+		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+	else
+		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
 }
 
 int ipoib_ib_dev_up(struct net_device *dev)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
+	ipoib_pkey_dev_check_presence(dev);
+
+	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
+		ipoib_dbg(priv, "PKEY is not assigned.\n");
+		return 0;
+	}
+
 	set_bit(IPOIB_FLAG_OPER_UP, &priv->flags);
 
 	return ipoib_mcast_start_thread(dev);
 }
 
-int ipoib_ib_dev_down(struct net_device *dev)
+int ipoib_ib_dev_down(struct net_device *dev, int flush)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
@@ -422,25 +747,15 @@ int ipoib_ib_dev_down(struct net_device *dev)
 
 	/* Shutdown the P_Key thread if still active */
 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		set_bit(IPOIB_PKEY_STOP, &priv->flags);
-		cancel_delayed_work(&priv->pkey_task);
-		up(&pkey_sem);
-		flush_workqueue(ipoib_workqueue);
+		cancel_delayed_work_sync(&priv->pkey_poll_task);
+		mutex_unlock(&pkey_mutex);
 	}
 
-	ipoib_mcast_stop_thread(dev);
-
-	/*
-	 * Flush the multicast groups first so we stop any multicast joins. The
-	 * completion thread may have already died and we may deadlock waiting
-	 * for the completion thread to finish some multicast joins.
-	 */
+	ipoib_mcast_stop_thread(dev, flush);
 	ipoib_mcast_dev_flush(dev);
 
-	/* Delete broadcast and local addresses since they will be recreated */
-	ipoib_mcast_dev_down(dev);
-
 	ipoib_flush_paths(dev);
 
 	return 0;
@@ -452,26 +767,71 @@ static int recvs_pending(struct net_device *dev)
 	int pending = 0;
 	int i;
 
-	for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
+	for (i = 0; i < ipoib_recvq_size; ++i)
 		if (priv->rx_ring[i].skb)
 			++pending;
 
 	return pending;
 }
 
-int ipoib_ib_dev_stop(struct net_device *dev)
+void ipoib_drain_cq(struct net_device *dev)
+{
+	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	int i, n;
+
+	/*
+	 * We call completion handling routines that expect to be
+	 * called from the BH-disabled NAPI poll context, so disable
+	 * BHs here too.
+	 */
+	local_bh_disable();
+
+	do {
+		n = ib_poll_cq(priv->recv_cq, IPOIB_NUM_WC, priv->ibwc);
+		for (i = 0; i < n; ++i) {
+			/*
+			 * Convert any successful completions to flush
+			 * errors to avoid passing packets up the
+			 * stack after bringing the device down.
+			 */
+			if (priv->ibwc[i].status == IB_WC_SUCCESS)
+				priv->ibwc[i].status = IB_WC_WR_FLUSH_ERR;
+
+			if (priv->ibwc[i].wr_id & IPOIB_OP_RECV) {
+				if (priv->ibwc[i].wr_id & IPOIB_OP_CM)
+					ipoib_cm_handle_rx_wc(dev, priv->ibwc + i);
+				else
+					ipoib_ib_handle_rx_wc(dev, priv->ibwc + i);
+			} else
+				ipoib_cm_handle_tx_wc(dev, priv->ibwc + i);
+		}
+	} while (n == IPOIB_NUM_WC);
+
+	while (poll_tx(priv))
+		; /* nothing */
+
+	local_bh_enable();
+}
+
+int ipoib_ib_dev_stop(struct net_device *dev, int flush)
 {
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 	struct ib_qp_attr qp_attr;
-	int attr_mask;
 	unsigned long begin;
-	struct ipoib_buf *tx_req;
+	struct ipoib_tx_buf *tx_req;
 	int i;
 
-	/* Kill the existing QP and allocate a new one */
+	if (test_and_clear_bit(IPOIB_FLAG_INITIALIZED, &priv->flags))
+		napi_disable(&priv->napi);
+
+	ipoib_cm_dev_stop(dev);
+
+	/*
+	 * Move our QP to the error state and then reinitialize in
+	 * when all work requests have completed or have been flushed.
+	 */
 	qp_attr.qp_state = IB_QPS_ERR;
-	attr_mask        = IB_QP_STATE;
-	if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
 		ipoib_warn(priv, "Failed to modify QP to ERROR state\n");
 
 	/* Wait for all sends and receives to complete */
@@ -488,44 +848,46 @@ int ipoib_ib_dev_stop(struct net_device *dev)
 			 */
 			while ((int) priv->tx_tail - (int) priv->tx_head < 0) {
 				tx_req = &priv->tx_ring[priv->tx_tail &
-							(IPOIB_TX_RING_SIZE - 1)];
-				dma_unmap_single(priv->ca->dma_device,
-						 pci_unmap_addr(tx_req, mapping),
-						 tx_req->skb->len,
-						 DMA_TO_DEVICE);
+							(ipoib_sendq_size - 1)];
+				ipoib_dma_unmap_tx(priv->ca, tx_req);
 				dev_kfree_skb_any(tx_req->skb);
 				++priv->tx_tail;
+				--priv->tx_outstanding;
 			}
 
-			for (i = 0; i < IPOIB_RX_RING_SIZE; ++i)
-				if (priv->rx_ring[i].skb) {
-					dma_unmap_single(priv->ca->dma_device,
-							 pci_unmap_addr(&priv->rx_ring[i],
-									mapping),
-							 IPOIB_BUF_SIZE,
-							 DMA_FROM_DEVICE);
-					dev_kfree_skb_any(priv->rx_ring[i].skb);
-					priv->rx_ring[i].skb = NULL;
-				}
+			for (i = 0; i < ipoib_recvq_size; ++i) {
+				struct ipoib_rx_buf *rx_req;
+
+				rx_req = &priv->rx_ring[i];
+				if (!rx_req->skb)
+					continue;
+				ipoib_ud_dma_unmap_rx(priv,
+						      priv->rx_ring[i].mapping);
+				dev_kfree_skb_any(rx_req->skb);
+				rx_req->skb = NULL;
+			}
 
 			goto timeout;
 		}
 
+		ipoib_drain_cq(dev);
+
 		msleep(1);
 	}
 
 	ipoib_dbg(priv, "All sends and receives done.\n");
 
 timeout:
+	del_timer_sync(&priv->poll_timer);
 	qp_attr.qp_state = IB_QPS_RESET;
-	attr_mask        = IB_QP_STATE;
-	if (ib_modify_qp(priv->qp, &qp_attr, attr_mask))
+	if (ib_modify_qp(priv->qp, &qp_attr, IB_QP_STATE))
 		ipoib_warn(priv, "Failed to modify QP to RESET state\n");
 
 	/* Wait for all AHs to be reaped */
 	set_bit(IPOIB_STOP_REAPER, &priv->flags);
 	cancel_delayed_work(&priv->ah_reap_task);
-	flush_workqueue(ipoib_workqueue);
+	if (flush)
+		flush_workqueue(ipoib_workqueue);
 
 	begin = jiffies;
 
@@ -540,6 +902,8 @@ timeout:
 		msleep(1);
 	}
 
+	ib_req_notify_cq(priv->recv_cq, IB_CQ_NEXT_COMP);
+
 	return 0;
 }
 
@@ -556,6 +920,9 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 		return -ENODEV;
 	}
 
+	setup_timer(&priv->poll_timer, ipoib_ib_tx_timer_func,
+		    (unsigned long) dev);
+
 	if (dev->flags & IFF_UP) {
 		if (ipoib_ib_dev_open(dev)) {
 			ipoib_transport_dev_cleanup(dev);
@@ -566,28 +933,148 @@ int ipoib_ib_dev_init(struct net_device *dev, struct ib_device *ca, int port)
 	return 0;
 }
 
-void ipoib_ib_dev_flush(void *_dev)
+/*
+ * Takes whatever value which is in pkey index 0 and updates priv->pkey
+ * returns 0 if the pkey value was changed.
+ */
+static inline int update_parent_pkey(struct ipoib_dev_priv *priv)
+{
+	int result;
+	u16 prev_pkey;
+
+	prev_pkey = priv->pkey;
+	result = ib_query_pkey(priv->ca, priv->port, 0, &priv->pkey);
+	if (result) {
+		ipoib_warn(priv, "ib_query_pkey port %d failed (ret = %d)\n",
+			   priv->port, result);
+		return result;
+	}
+
+	priv->pkey |= 0x8000;
+
+	if (prev_pkey != priv->pkey) {
+		ipoib_dbg(priv, "pkey changed from 0x%x to 0x%x\n",
+			  prev_pkey, priv->pkey);
+		/*
+		 * Update the pkey in the broadcast address, while making sure to set
+		 * the full membership bit, so that we join the right broadcast group.
+		 */
+		priv->dev->broadcast[8] = priv->pkey >> 8;
+		priv->dev->broadcast[9] = priv->pkey & 0xff;
+		return 0;
+	}
+
+	return 1;
+}
+
+static void __ipoib_ib_dev_flush(struct ipoib_dev_priv *priv,
+				enum ipoib_flush_level level)
 {
-	struct net_device *dev = (struct net_device *)_dev;
-	struct ipoib_dev_priv *priv = netdev_priv(dev), *cpriv;
+	struct ipoib_dev_priv *cpriv;
+	struct net_device *dev = priv->dev;
+	u16 new_index;
+	int result;
+
+	down_read(&priv->vlan_rwsem);
+
+	/*
+	 * Flush any child interfaces too -- they might be up even if
+	 * the parent is down.
+	 */
+	list_for_each_entry(cpriv, &priv->child_intfs, list)
+		__ipoib_ib_dev_flush(cpriv, level);
+
+	up_read(&priv->vlan_rwsem);
 
-	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
+	if (!test_bit(IPOIB_FLAG_INITIALIZED, &priv->flags)) {
+		/* for non-child devices must check/update the pkey value here */
+		if (level == IPOIB_FLUSH_HEAVY &&
+		    !test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags))
+			update_parent_pkey(priv);
+		ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_INITIALIZED not set.\n");
 		return;
+	}
+
+	if (!test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
+		ipoib_dbg(priv, "Not flushing - IPOIB_FLAG_ADMIN_UP not set.\n");
+		return;
+	}
+
+	if (level == IPOIB_FLUSH_HEAVY) {
+		/* child devices chase their origin pkey value, while non-child
+		 * (parent) devices should always takes what present in pkey index 0
+		 */
+		if (test_bit(IPOIB_FLAG_SUBINTERFACE, &priv->flags)) {
+			if (ib_find_pkey(priv->ca, priv->port, priv->pkey, &new_index)) {
+				clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
+				ipoib_ib_dev_down(dev, 0);
+				ipoib_ib_dev_stop(dev, 0);
+				if (ipoib_pkey_dev_delay_open(dev))
+					return;
+			}
+			/* restart QP only if P_Key index is changed */
+			if (test_and_set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags) &&
+			    new_index == priv->pkey_index) {
+				ipoib_dbg(priv, "Not flushing - P_Key index not changed.\n");
+				return;
+			}
+			priv->pkey_index = new_index;
+		} else {
+			result = update_parent_pkey(priv);
+			/* restart QP only if P_Key value changed */
+			if (result) {
+				ipoib_dbg(priv, "Not flushing - P_Key value not changed.\n");
+				return;
+			}
+		}
+	}
 
-	ipoib_dbg(priv, "flushing\n");
+	if (level == IPOIB_FLUSH_LIGHT) {
+		ipoib_mark_paths_invalid(dev);
+		ipoib_mcast_dev_flush(dev);
+	}
 
-	ipoib_ib_dev_down(dev);
+	if (level >= IPOIB_FLUSH_NORMAL)
+		ipoib_ib_dev_down(dev, 0);
+
+	if (level == IPOIB_FLUSH_HEAVY) {
+		ipoib_ib_dev_stop(dev, 0);
+		ipoib_ib_dev_open(dev);
+	}
 
 	/*
 	 * The device could have been brought down between the start and when
 	 * we get here, don't bring it back up if it's not configured up
 	 */
-	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags))
-		ipoib_ib_dev_up(dev);
+	if (test_bit(IPOIB_FLAG_ADMIN_UP, &priv->flags)) {
+		if (level >= IPOIB_FLUSH_NORMAL)
+			ipoib_ib_dev_up(dev);
+		ipoib_mcast_restart_task(&priv->restart_task);
+	}
+}
 
-	/* Flush any child interfaces too */
-	list_for_each_entry(cpriv, &priv->child_intfs, list)
-		ipoib_ib_dev_flush(&cpriv->dev);
+void ipoib_ib_dev_flush_light(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, flush_light);
+
+	__ipoib_ib_dev_flush(priv, IPOIB_FLUSH_LIGHT);
+}
+
+void ipoib_ib_dev_flush_normal(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, flush_normal);
+
+	__ipoib_ib_dev_flush(priv, IPOIB_FLUSH_NORMAL);
+}
+
+void ipoib_ib_dev_flush_heavy(struct work_struct *work)
+{
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, flush_heavy);
+
+	__ipoib_ib_dev_flush(priv, IPOIB_FLUSH_HEAVY);
 }
 
 void ipoib_ib_dev_cleanup(struct net_device *dev)
@@ -595,11 +1082,14 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
 	struct ipoib_dev_priv *priv = netdev_priv(dev);
 
 	ipoib_dbg(priv, "cleaning up ib_dev\n");
+	/*
+	 * We must make sure there are no more (path) completions
+	 * that may wish to touch priv fields that are no longer valid
+	 */
+	ipoib_flush_paths(dev);
 
-	ipoib_mcast_stop_thread(dev);
-
-	/* Delete the broadcast address and the local address */
-	ipoib_mcast_dev_down(dev);
+	ipoib_mcast_stop_thread(dev, 1);
+	ipoib_mcast_dev_flush(dev);
 
 	ipoib_transport_dev_cleanup(dev);
 }
@@ -613,35 +1103,24 @@ void ipoib_ib_dev_cleanup(struct net_device *dev)
  * Bug #2507. This implementation will probably be removed when the P_Key
  * change async notification is available.
  */
-int ipoib_open(struct net_device *dev);
-
-static void ipoib_pkey_dev_check_presence(struct net_device *dev)
-{
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
-	u16 pkey_index = 0;
 
-	if (ib_find_cached_pkey(priv->ca, priv->port, priv->pkey, &pkey_index))
-		clear_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
-	else
-		set_bit(IPOIB_PKEY_ASSIGNED, &priv->flags);
-}
-
-void ipoib_pkey_poll(void *dev_ptr)
+void ipoib_pkey_poll(struct work_struct *work)
 {
-	struct net_device *dev = dev_ptr;
-	struct ipoib_dev_priv *priv = netdev_priv(dev);
+	struct ipoib_dev_priv *priv =
+		container_of(work, struct ipoib_dev_priv, pkey_poll_task.work);
+	struct net_device *dev = priv->dev;
 
 	ipoib_pkey_dev_check_presence(dev);
 
 	if (test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags))
 		ipoib_open(dev);
 	else {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		if (!test_bit(IPOIB_PKEY_STOP, &priv->flags))
 			queue_delayed_work(ipoib_workqueue,
-					   &priv->pkey_task,
+					   &priv->pkey_poll_task,
 					   HZ);
-		up(&pkey_sem);
+		mutex_unlock(&pkey_mutex);
 	}
 }
 
@@ -655,12 +1134,12 @@ int ipoib_pkey_dev_delay_open(struct net_device *dev)
 
 	/* P_Key value not assigned yet - start polling */
 	if (!test_bit(IPOIB_PKEY_ASSIGNED, &priv->flags)) {
-		down(&pkey_sem);
+		mutex_lock(&pkey_mutex);
 		clear_bit(IPOIB_PKEY_STOP, &priv->flags);
 		queue_delayed_work(ipoib_workqueue,
-				   &priv->pkey_task,
+				   &priv->pkey_poll_task,
 				   HZ);
-		up(&pkey_sem);
+		mutex_unlock(&pkey_mutex);
 		return 1;
 	}