1 files changed, 769 insertions, 702 deletions

diff --git a/drivers/net/mv643xx_eth.c b/drivers/net/mv643xx_eth.c
index 0a18b9e96da..a9c8c08044b 100644
--- a/drivers/net/mv643xx_eth.c
+++ b/drivers/net/mv643xx_eth.c
@@ -38,6 +38,7 @@
 #include <linux/init.h>
 #include <linux/dma-mapping.h>
 #include <linux/in.h>
+#include <linux/ip.h>
 #include <linux/tcp.h>
 #include <linux/udp.h>
 #include <linux/etherdevice.h>
@@ -48,30 +49,28 @@
 #include <linux/kernel.h>
 #include <linux/spinlock.h>
 #include <linux/workqueue.h>
-#include <linux/mii.h>
+#include <linux/phy.h>
 #include <linux/mv643xx_eth.h>
 #include <asm/io.h>
 #include <asm/types.h>
 #include <asm/system.h>
 
 static char mv643xx_eth_driver_name[] = "mv643xx_eth";
-static char mv643xx_eth_driver_version[] = "1.3";
+static char mv643xx_eth_driver_version[] = "1.4";
 
-#define MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-#define MV643XX_ETH_NAPI
-#define MV643XX_ETH_TX_FAST_REFILL
-
-#ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-#define MAX_DESCS_PER_SKB	(MAX_SKB_FRAGS + 1)
-#else
-#define MAX_DESCS_PER_SKB	1
-#endif
 
 /*
  * Registers shared between all ports.
  */
 #define PHY_ADDR			0x0000
 #define SMI_REG				0x0004
+#define  SMI_BUSY			0x10000000
+#define  SMI_READ_VALID			0x08000000
+#define  SMI_OPCODE_READ		0x04000000
+#define  SMI_OPCODE_WRITE		0x00000000
+#define ERR_INT_CAUSE			0x0080
+#define  ERR_INT_SMI_DONE		0x00000010
+#define ERR_INT_MASK			0x0084
 #define WINDOW_BASE(w)			(0x0200 + ((w) << 3))
 #define WINDOW_SIZE(w)			(0x0204 + ((w) << 3))
 #define WINDOW_REMAP_HIGH(w)		(0x0280 + ((w) << 2))
@@ -104,16 +103,12 @@ static char mv643xx_eth_driver_version[] = "1.3";
 #define TX_BW_MTU(p)			(0x0458 + ((p) << 10))
 #define TX_BW_BURST(p)			(0x045c + ((p) << 10))
 #define INT_CAUSE(p)			(0x0460 + ((p) << 10))
-#define  INT_TX_END_0			0x00080000
 #define  INT_TX_END			0x07f80000
-#define  INT_RX				0x0007fbfc
+#define  INT_RX				0x000003fc
 #define  INT_EXT			0x00000002
 #define INT_CAUSE_EXT(p)		(0x0464 + ((p) << 10))
-#define  INT_EXT_LINK			0x00100000
-#define  INT_EXT_PHY			0x00010000
-#define  INT_EXT_TX_ERROR_0		0x00000100
-#define  INT_EXT_TX_0			0x00000001
-#define  INT_EXT_TX			0x0000ffff
+#define  INT_EXT_LINK_PHY		0x00110000
+#define  INT_EXT_TX			0x000000ff
 #define INT_MASK(p)			(0x0468 + ((p) << 10))
 #define INT_MASK_EXT(p)			(0x046c + ((p) << 10))
 #define TX_FIFO_URGENT_THRESHOLD(p)	(0x0474 + ((p) << 10))
@@ -171,8 +166,8 @@ static char mv643xx_eth_driver_version[] = "1.3";
 #define FORCE_LINK_PASS				(1 << 1)
 #define SERIAL_PORT_ENABLE			(1 << 0)
 
-#define DEFAULT_RX_QUEUE_SIZE		400
-#define DEFAULT_TX_QUEUE_SIZE		800
+#define DEFAULT_RX_QUEUE_SIZE		128
+#define DEFAULT_TX_QUEUE_SIZE		256
 
 
 /*
@@ -249,9 +244,23 @@ struct mv643xx_eth_shared_private {
 	void __iomem *base;
 
 	/*
-	 * Protects access to SMI_REG, which is shared between ports.
+	 * Points at the right SMI instance to use.
+	 */
+	struct mv643xx_eth_shared_private *smi;
+
+	/*
+	 * Provides access to local SMI interface.
+	 */
+	struct mii_bus *smi_bus;
+
+	/*
+	 * If we have access to the error interrupt pin (which is
+	 * somewhat misnamed as it not only reflects internal errors
+	 * but also reflects SMI completion), use that to wait for
+	 * SMI access completion instead of polling the SMI busy bit.
 	 */
-	spinlock_t phy_lock;
+	int err_interrupt;
+	wait_queue_head_t smi_busy_wait;
 
 	/*
 	 * Per-port MBUS window access register value.
@@ -263,9 +272,13 @@ struct mv643xx_eth_shared_private {
 	 */
 	unsigned int t_clk;
 	int extended_rx_coal_limit;
-	int tx_bw_control_moved;
+	int tx_bw_control;
 };
 
+#define TX_BW_CONTROL_ABSENT		0
+#define TX_BW_CONTROL_OLD_LAYOUT	1
+#define TX_BW_CONTROL_NEW_LAYOUT	2
+
 
 /* per-port *****************************************************************/
 struct mib_counters {
@@ -314,8 +327,6 @@ struct rx_queue {
 	dma_addr_t rx_desc_dma;
 	int rx_desc_area_size;
 	struct sk_buff **rx_skb;
-
-	struct timer_list rx_oom;
 };
 
 struct tx_queue {
@@ -330,7 +341,12 @@ struct tx_queue {
 	struct tx_desc *tx_desc_area;
 	dma_addr_t tx_desc_dma;
 	int tx_desc_area_size;
-	struct sk_buff **tx_skb;
+
+	struct sk_buff_head tx_skb;
+
+	unsigned long tx_packets;
+	unsigned long tx_bytes;
+	unsigned long tx_dropped;
 };
 
 struct mv643xx_eth_private {
@@ -339,14 +355,24 @@ struct mv643xx_eth_private {
 
 	struct net_device *dev;
 
-	struct mv643xx_eth_shared_private *shared_smi;
-	int phy_addr;
-
-	spinlock_t lock;
+	struct phy_device *phy;
 
+	struct timer_list mib_counters_timer;
+	spinlock_t mib_counters_lock;
 	struct mib_counters mib_counters;
+
 	struct work_struct tx_timeout_task;
-	struct mii_if_info mii;
+
+	struct napi_struct napi;
+	u8 work_link;
+	u8 work_tx;
+	u8 work_tx_end;
+	u8 work_rx;
+	u8 work_rx_refill;
+	u8 work_rx_oom;
+
+	int skb_size;
+	struct sk_buff_head rx_recycle;
 
 	/*
 	 * RX state.
@@ -354,9 +380,8 @@ struct mv643xx_eth_private {
 	int default_rx_ring_size;
 	unsigned long rx_desc_sram_addr;
 	int rx_desc_sram_size;
-	u8 rxq_mask;
-	int rxq_primary;
-	struct napi_struct napi;
+	int rxq_count;
+	struct timer_list rx_oom;
 	struct rx_queue rxq[8];
 
 	/*
@@ -365,12 +390,8 @@ struct mv643xx_eth_private {
 	int default_tx_ring_size;
 	unsigned long tx_desc_sram_addr;
 	int tx_desc_sram_size;
-	u8 txq_mask;
-	int txq_primary;
+	int txq_count;
 	struct tx_queue txq[8];
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-	int tx_clean_threshold;
-#endif
 };
 
 
@@ -440,94 +461,21 @@ static void txq_disable(struct tx_queue *txq)
 		udelay(10);
 }
 
-static void __txq_maybe_wake(struct tx_queue *txq)
+static void txq_maybe_wake(struct tx_queue *txq)
 {
 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
+	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
 
-	/*
-	 * netif_{stop,wake}_queue() flow control only applies to
-	 * the primary queue.
-	 */
-	BUG_ON(txq->index != mp->txq_primary);
-
-	if (txq->tx_ring_size - txq->tx_desc_count >= MAX_DESCS_PER_SKB)
-		netif_wake_queue(mp->dev);
-}
-
-
-/* rx ***********************************************************************/
-static void txq_reclaim(struct tx_queue *txq, int force);
-
-static void rxq_refill(struct rx_queue *rxq)
-{
-	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
-	unsigned long flags;
-
-	spin_lock_irqsave(&mp->lock, flags);
-
-	while (rxq->rx_desc_count < rxq->rx_ring_size) {
-		int skb_size;
-		struct sk_buff *skb;
-		int unaligned;
-		int rx;
-
-		/*
-		 * Reserve 2+14 bytes for an ethernet header (the
-		 * hardware automatically prepends 2 bytes of dummy
-		 * data to each received packet), 16 bytes for up to
-		 * four VLAN tags, and 4 bytes for the trailing FCS
-		 * -- 36 bytes total.
-		 */
-		skb_size = mp->dev->mtu + 36;
-
-		/*
-		 * Make sure that the skb size is a multiple of 8
-		 * bytes, as the lower three bits of the receive
-		 * descriptor's buffer size field are ignored by
-		 * the hardware.
-		 */
-		skb_size = (skb_size + 7) & ~7;
-
-		skb = dev_alloc_skb(skb_size + dma_get_cache_alignment() - 1);
-		if (skb == NULL)
-			break;
-
-		unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
-		if (unaligned)
-			skb_reserve(skb, dma_get_cache_alignment() - unaligned);
-
-		rxq->rx_desc_count++;
-		rx = rxq->rx_used_desc;
-		rxq->rx_used_desc = (rx + 1) % rxq->rx_ring_size;
-
-		rxq->rx_desc_area[rx].buf_ptr = dma_map_single(NULL, skb->data,
-						skb_size, DMA_FROM_DEVICE);
-		rxq->rx_desc_area[rx].buf_size = skb_size;
-		rxq->rx_skb[rx] = skb;
-		wmb();
-		rxq->rx_desc_area[rx].cmd_sts = BUFFER_OWNED_BY_DMA |
-						RX_ENABLE_INTERRUPT;
-		wmb();
-
-		/*
-		 * The hardware automatically prepends 2 bytes of
-		 * dummy data to each received packet, so that the
-		 * IP header ends up 16-byte aligned.
-		 */
-		skb_reserve(skb, 2);
+	if (netif_tx_queue_stopped(nq)) {
+		__netif_tx_lock(nq, smp_processor_id());
+		if (txq->tx_ring_size - txq->tx_desc_count >= MAX_SKB_FRAGS + 1)
+			netif_tx_wake_queue(nq);
+		__netif_tx_unlock(nq);
 	}
-
-	if (rxq->rx_desc_count != rxq->rx_ring_size)
-		mod_timer(&rxq->rx_oom, jiffies + (HZ / 10));
-
-	spin_unlock_irqrestore(&mp->lock, flags);
 }
 
-static inline void rxq_refill_timer_wrapper(unsigned long data)
-{
-	rxq_refill((struct rx_queue *)data);
-}
 
+/* rx napi ******************************************************************/
 static int rxq_process(struct rx_queue *rxq, int budget)
 {
 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
@@ -539,31 +487,31 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 		struct rx_desc *rx_desc;
 		unsigned int cmd_sts;
 		struct sk_buff *skb;
-		unsigned long flags;
-
-		spin_lock_irqsave(&mp->lock, flags);
+		u16 byte_cnt;
 
 		rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
 
 		cmd_sts = rx_desc->cmd_sts;
-		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
-			spin_unlock_irqrestore(&mp->lock, flags);
+		if (cmd_sts & BUFFER_OWNED_BY_DMA)
 			break;
-		}
 		rmb();
 
 		skb = rxq->rx_skb[rxq->rx_curr_desc];
 		rxq->rx_skb[rxq->rx_curr_desc] = NULL;
 
-		rxq->rx_curr_desc = (rxq->rx_curr_desc + 1) % rxq->rx_ring_size;
-
-		spin_unlock_irqrestore(&mp->lock, flags);
+		rxq->rx_curr_desc++;
+		if (rxq->rx_curr_desc == rxq->rx_ring_size)
+			rxq->rx_curr_desc = 0;
 
-		dma_unmap_single(NULL, rx_desc->buf_ptr + 2,
+		dma_unmap_single(NULL, rx_desc->buf_ptr,
 				 rx_desc->buf_size, DMA_FROM_DEVICE);
 		rxq->rx_desc_count--;
 		rx++;
 
+		mp->work_rx_refill |= 1 << rxq->index;
+
+		byte_cnt = rx_desc->byte_cnt;
+
 		/*
 		 * Update statistics.
 		 *
@@ -573,7 +521,7 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 		 * byte CRC at the end of the packet (which we do count).
 		 */
 		stats->rx_packets++;
-		stats->rx_bytes += rx_desc->byte_cnt - 2;
+		stats->rx_bytes += byte_cnt - 2;
 
 		/*
 		 * In case we received a packet without first / last bits
@@ -596,72 +544,84 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 			if (cmd_sts & ERROR_SUMMARY)
 				stats->rx_errors++;
 
-			dev_kfree_skb_irq(skb);
+			dev_kfree_skb(skb);
 		} else {
 			/*
 			 * The -4 is for the CRC in the trailer of the
 			 * received packet
 			 */
-			skb_put(skb, rx_desc->byte_cnt - 2 - 4);
+			skb_put(skb, byte_cnt - 2 - 4);
 
-			if (cmd_sts & LAYER_4_CHECKSUM_OK) {
+			if (cmd_sts & LAYER_4_CHECKSUM_OK)
 				skb->ip_summed = CHECKSUM_UNNECESSARY;
-				skb->csum = htons(
-					(cmd_sts & 0x0007fff8) >> 3);
-			}
 			skb->protocol = eth_type_trans(skb, mp->dev);
-#ifdef MV643XX_ETH_NAPI
 			netif_receive_skb(skb);
-#else
-			netif_rx(skb);
-#endif
 		}
 
 		mp->dev->last_rx = jiffies;
 	}
 
-	rxq_refill(rxq);
+	if (rx < budget)
+		mp->work_rx &= ~(1 << rxq->index);
 
 	return rx;
 }
 
-#ifdef MV643XX_ETH_NAPI
-static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
+static int rxq_refill(struct rx_queue *rxq, int budget)
 {
-	struct mv643xx_eth_private *mp;
-	int rx;
-	int i;
+	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+	int refilled;
 
-	mp = container_of(napi, struct mv643xx_eth_private, napi);
+	refilled = 0;
+	while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
+		struct sk_buff *skb;
+		int unaligned;
+		int rx;
+
+		skb = __skb_dequeue(&mp->rx_recycle);
+		if (skb == NULL)
+			skb = dev_alloc_skb(mp->skb_size +
+					    dma_get_cache_alignment() - 1);
 
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-	if (++mp->tx_clean_threshold > 5) {
-		mp->tx_clean_threshold = 0;
-		for (i = 0; i < 8; i++)
-			if (mp->txq_mask & (1 << i))
-				txq_reclaim(mp->txq + i, 0);
-
-		if (netif_carrier_ok(mp->dev)) {
-			spin_lock_irq(&mp->lock);
-			__txq_maybe_wake(mp->txq + mp->txq_primary);
-			spin_unlock_irq(&mp->lock);
+		if (skb == NULL) {
+			mp->work_rx_oom |= 1 << rxq->index;
+			goto oom;
 		}
-	}
-#endif
 
-	rx = 0;
-	for (i = 7; rx < budget && i >= 0; i--)
-		if (mp->rxq_mask & (1 << i))
-			rx += rxq_process(mp->rxq + i, budget - rx);
+		unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
+		if (unaligned)
+			skb_reserve(skb, dma_get_cache_alignment() - unaligned);
 
-	if (rx < budget) {
-		netif_rx_complete(mp->dev, napi);
-		wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
+		refilled++;
+		rxq->rx_desc_count++;
+
+		rx = rxq->rx_used_desc++;
+		if (rxq->rx_used_desc == rxq->rx_ring_size)
+			rxq->rx_used_desc = 0;
+
+		rxq->rx_desc_area[rx].buf_ptr = dma_map_single(NULL, skb->data,
+						mp->skb_size, DMA_FROM_DEVICE);
+		rxq->rx_desc_area[rx].buf_size = mp->skb_size;
+		rxq->rx_skb[rx] = skb;
+		wmb();
+		rxq->rx_desc_area[rx].cmd_sts = BUFFER_OWNED_BY_DMA |
+						RX_ENABLE_INTERRUPT;
+		wmb();
+
+		/*
+		 * The hardware automatically prepends 2 bytes of
+		 * dummy data to each received packet, so that the
+		 * IP header ends up 16-byte aligned.
+		 */
+		skb_reserve(skb, 2);
 	}
 
-	return rx;
+	if (refilled < budget)
+		mp->work_rx_refill &= ~(1 << rxq->index);
+
+oom:
+	return refilled;
 }
-#endif
 
 
 /* tx ***********************************************************************/
@@ -684,8 +644,9 @@ static int txq_alloc_desc_index(struct tx_queue *txq)
 
 	BUG_ON(txq->tx_desc_count >= txq->tx_ring_size);
 
-	tx_desc_curr = txq->tx_curr_desc;
-	txq->tx_curr_desc = (tx_desc_curr + 1) % txq->tx_ring_size;
+	tx_desc_curr = txq->tx_curr_desc++;
+	if (txq->tx_curr_desc == txq->tx_ring_size)
+		txq->tx_curr_desc = 0;
 
 	BUG_ON(txq->tx_curr_desc == txq->tx_used_desc);
 
@@ -714,10 +675,8 @@ static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
 			desc->cmd_sts = BUFFER_OWNED_BY_DMA |
 					ZERO_PADDING | TX_LAST_DESC |
 					TX_ENABLE_INTERRUPT;
-			txq->tx_skb[tx_index] = skb;
 		} else {
 			desc->cmd_sts = BUFFER_OWNED_BY_DMA;
-			txq->tx_skb[tx_index] = NULL;
 		}
 
 		desc->l4i_chk = 0;
@@ -734,144 +693,228 @@ static inline __be16 sum16_as_be(__sum16 sum)
 	return (__force __be16)sum;
 }
 
-static void txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
+static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
 	int nr_frags = skb_shinfo(skb)->nr_frags;
 	int tx_index;
 	struct tx_desc *desc;
 	u32 cmd_sts;
+	u16 l4i_chk;
 	int length;
 
 	cmd_sts = TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
-
-	tx_index = txq_alloc_desc_index(txq);
-	desc = &txq->tx_desc_area[tx_index];
-
-	if (nr_frags) {
-		txq_submit_frag_skb(txq, skb);
-
-		length = skb_headlen(skb);
-		txq->tx_skb[tx_index] = NULL;
-	} else {
-		cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
-		length = skb->len;
-		txq->tx_skb[tx_index] = skb;
-	}
-
-	desc->byte_cnt = length;
-	desc->buf_ptr = dma_map_single(NULL, skb->data, length, DMA_TO_DEVICE);
+	l4i_chk = 0;
 
 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
-		int mac_hdr_len;
+		int tag_bytes;
 
 		BUG_ON(skb->protocol != htons(ETH_P_IP) &&
 		       skb->protocol != htons(ETH_P_8021Q));
 
-		cmd_sts |= GEN_TCP_UDP_CHECKSUM |
-			   GEN_IP_V4_CHECKSUM   |
-			   ip_hdr(skb)->ihl << TX_IHL_SHIFT;
+		tag_bytes = (void *)ip_hdr(skb) - (void *)skb->data - ETH_HLEN;
+		if (unlikely(tag_bytes & ~12)) {
+			if (skb_checksum_help(skb) == 0)
+				goto no_csum;
+			kfree_skb(skb);
+			return 1;
+		}
 
-		mac_hdr_len = (void *)ip_hdr(skb) - (void *)skb->data;
-		switch (mac_hdr_len - ETH_HLEN) {
-		case 0:
-			break;
-		case 4:
-			cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
-			break;
-		case 8:
-			cmd_sts |= MAC_HDR_EXTRA_8_BYTES;
-			break;
-		case 12:
+		if (tag_bytes & 4)
 			cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
+		if (tag_bytes & 8)
 			cmd_sts |= MAC_HDR_EXTRA_8_BYTES;
-			break;
-		default:
-			if (net_ratelimit())
-				dev_printk(KERN_ERR, &txq_to_mp(txq)->dev->dev,
-				   "mac header length is %d?!\n", mac_hdr_len);
-			break;
-		}
+
+		cmd_sts |= GEN_TCP_UDP_CHECKSUM |
+			   GEN_IP_V4_CHECKSUM   |
+			   ip_hdr(skb)->ihl << TX_IHL_SHIFT;
 
 		switch (ip_hdr(skb)->protocol) {
 		case IPPROTO_UDP:
 			cmd_sts |= UDP_FRAME;
-			desc->l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
+			l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
 			break;
 		case IPPROTO_TCP:
-			desc->l4i_chk = ntohs(sum16_as_be(tcp_hdr(skb)->check));
+			l4i_chk = ntohs(sum16_as_be(tcp_hdr(skb)->check));
 			break;
 		default:
 			BUG();
 		}
 	} else {
+no_csum:
 		/* Errata BTS #50, IHL must be 5 if no HW checksum */
 		cmd_sts |= 5 << TX_IHL_SHIFT;
-		desc->l4i_chk = 0;
 	}
 
+	tx_index = txq_alloc_desc_index(txq);
+	desc = &txq->tx_desc_area[tx_index];
+
+	if (nr_frags) {
+		txq_submit_frag_skb(txq, skb);
+		length = skb_headlen(skb);
+	} else {
+		cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
+		length = skb->len;
+	}
+
+	desc->l4i_chk = l4i_chk;
+	desc->byte_cnt = length;
+	desc->buf_ptr = dma_map_single(NULL, skb->data, length, DMA_TO_DEVICE);
+
+	__skb_queue_tail(&txq->tx_skb, skb);
+
 	/* ensure all other descriptors are written before first cmd_sts */
 	wmb();
 	desc->cmd_sts = cmd_sts;
 
-	/* clear TX_END interrupt status */
-	wrl(mp, INT_CAUSE(mp->port_num), ~(INT_TX_END_0 << txq->index));
-	rdl(mp, INT_CAUSE(mp->port_num));
+	/* clear TX_END status */
+	mp->work_tx_end &= ~(1 << txq->index);
 
 	/* ensure all descriptors are written before poking hardware */
 	wmb();
 	txq_enable(txq);
 
 	txq->tx_desc_count += nr_frags + 1;
+
+	return 0;
 }
 
 static int mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
 {
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
-	struct net_device_stats *stats = &dev->stats;
+	int queue;
 	struct tx_queue *txq;
-	unsigned long flags;
+	struct netdev_queue *nq;
+
+	queue = skb_get_queue_mapping(skb);
+	txq = mp->txq + queue;
+	nq = netdev_get_tx_queue(dev, queue);
 
 	if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
-		stats->tx_dropped++;
+		txq->tx_dropped++;
 		dev_printk(KERN_DEBUG, &dev->dev,
 			   "failed to linearize skb with tiny "
 			   "unaligned fragment\n");
 		return NETDEV_TX_BUSY;
 	}
 
-	spin_lock_irqsave(&mp->lock, flags);
-
-	txq = mp->txq + mp->txq_primary;
-
-	if (txq->tx_ring_size - txq->tx_desc_count < MAX_DESCS_PER_SKB) {
-		spin_unlock_irqrestore(&mp->lock, flags);
-		if (txq->index == mp->txq_primary && net_ratelimit())
-			dev_printk(KERN_ERR, &dev->dev,
-				   "primary tx queue full?!\n");
+	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
+		if (net_ratelimit())
+			dev_printk(KERN_ERR, &dev->dev, "tx queue full?!\n");
 		kfree_skb(skb);
 		return NETDEV_TX_OK;
 	}
 
-	txq_submit_skb(txq, skb);
-	stats->tx_bytes += skb->len;
-	stats->tx_packets++;
-	dev->trans_start = jiffies;
-
-	if (txq->index == mp->txq_primary) {
+	if (!txq_submit_skb(txq, skb)) {
 		int entries_left;
 
+		txq->tx_bytes += skb->len;
+		txq->tx_packets++;
+		dev->trans_start = jiffies;
+
 		entries_left = txq->tx_ring_size - txq->tx_desc_count;
-		if (entries_left < MAX_DESCS_PER_SKB)
-			netif_stop_queue(dev);
+		if (entries_left < MAX_SKB_FRAGS + 1)
+			netif_tx_stop_queue(nq);
 	}
 
-	spin_unlock_irqrestore(&mp->lock, flags);
-
 	return NETDEV_TX_OK;
 }
 
 
+/* tx napi ******************************************************************/
+static void txq_kick(struct tx_queue *txq)
+{
+	struct mv643xx_eth_private *mp = txq_to_mp(txq);
+	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
+	u32 hw_desc_ptr;
+	u32 expected_ptr;
+
+	__netif_tx_lock(nq, smp_processor_id());
+
+	if (rdl(mp, TXQ_COMMAND(mp->port_num)) & (1 << txq->index))
+		goto out;
+
+	hw_desc_ptr = rdl(mp, TXQ_CURRENT_DESC_PTR(mp->port_num, txq->index));
+	expected_ptr = (u32)txq->tx_desc_dma +
+				txq->tx_curr_desc * sizeof(struct tx_desc);
+
+	if (hw_desc_ptr != expected_ptr)
+		txq_enable(txq);
+
+out:
+	__netif_tx_unlock(nq);
+
+	mp->work_tx_end &= ~(1 << txq->index);
+}
+
+static int txq_reclaim(struct tx_queue *txq, int budget, int force)
+{
+	struct mv643xx_eth_private *mp = txq_to_mp(txq);
+	struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
+	int reclaimed;
+
+	__netif_tx_lock(nq, smp_processor_id());
+
+	reclaimed = 0;
+	while (reclaimed < budget && txq->tx_desc_count > 0) {
+		int tx_index;
+		struct tx_desc *desc;
+		u32 cmd_sts;
+		struct sk_buff *skb;
+
+		tx_index = txq->tx_used_desc;
+		desc = &txq->tx_desc_area[tx_index];
+		cmd_sts = desc->cmd_sts;
+
+		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
+			if (!force)
+				break;
+			desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
+		}
+
+		txq->tx_used_desc = tx_index + 1;
+		if (txq->tx_used_desc == txq->tx_ring_size)
+			txq->tx_used_desc = 0;
+
+		reclaimed++;
+		txq->tx_desc_count--;
+
+		skb = NULL;
+		if (cmd_sts & TX_LAST_DESC)
+			skb = __skb_dequeue(&txq->tx_skb);
+
+		if (cmd_sts & ERROR_SUMMARY) {
+			dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
+			mp->dev->stats.tx_errors++;
+		}
+
+		if (cmd_sts & TX_FIRST_DESC) {
+			dma_unmap_single(NULL, desc->buf_ptr,
+					 desc->byte_cnt, DMA_TO_DEVICE);
+		} else {
+			dma_unmap_page(NULL, desc->buf_ptr,
+				       desc->byte_cnt, DMA_TO_DEVICE);
+		}
+
+		if (skb != NULL) {
+			if (skb_queue_len(&mp->rx_recycle) <
+					mp->default_rx_ring_size &&
+			    skb_recycle_check(skb, mp->skb_size))
+				__skb_queue_head(&mp->rx_recycle, skb);
+			else
+				dev_kfree_skb(skb);
+		}
+	}
+
+	__netif_tx_unlock(nq);
+
+	if (reclaimed < budget)
+		mp->work_tx &= ~(1 << txq->index);
+
+	return reclaimed;
+}
+
+
 /* tx rate control **********************************************************/
 /*
  * Set total maximum TX rate (shared by all TX queues for this port)
@@ -895,14 +938,17 @@ static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
 	if (bucket_size > 65535)
 		bucket_size = 65535;
 
-	if (mp->shared->tx_bw_control_moved) {
-		wrl(mp, TX_BW_RATE_MOVED(mp->port_num), token_rate);
-		wrl(mp, TX_BW_MTU_MOVED(mp->port_num), mtu);
-		wrl(mp, TX_BW_BURST_MOVED(mp->port_num), bucket_size);
-	} else {
+	switch (mp->shared->tx_bw_control) {
+	case TX_BW_CONTROL_OLD_LAYOUT:
 		wrl(mp, TX_BW_RATE(mp->port_num), token_rate);
 		wrl(mp, TX_BW_MTU(mp->port_num), mtu);
 		wrl(mp, TX_BW_BURST(mp->port_num), bucket_size);
+		break;
+	case TX_BW_CONTROL_NEW_LAYOUT:
+		wrl(mp, TX_BW_RATE_MOVED(mp->port_num), token_rate);
+		wrl(mp, TX_BW_MTU_MOVED(mp->port_num), mtu);
+		wrl(mp, TX_BW_BURST_MOVED(mp->port_num), bucket_size);
+		break;
 	}
 }
 
@@ -934,14 +980,21 @@ static void txq_set_fixed_prio_mode(struct tx_queue *txq)
 	/*
 	 * Turn on fixed priority mode.
 	 */
-	if (mp->shared->tx_bw_control_moved)
-		off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
-	else
+	off = 0;
+	switch (mp->shared->tx_bw_control) {
+	case TX_BW_CONTROL_OLD_LAYOUT:
 		off = TXQ_FIX_PRIO_CONF(mp->port_num);
+		break;
+	case TX_BW_CONTROL_NEW_LAYOUT:
+		off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+		break;
+	}
 
-	val = rdl(mp, off);
-	val |= 1 << txq->index;
-	wrl(mp, off, val);
+	if (off) {
+		val = rdl(mp, off);
+		val |= 1 << txq->index;
+		wrl(mp, off, val);
+	}
 }
 
 static void txq_set_wrr(struct tx_queue *txq, int weight)
@@ -953,95 +1006,147 @@ static void txq_set_wrr(struct tx_queue *txq, int weight)
 	/*
 	 * Turn off fixed priority mode.
 	 */
-	if (mp->shared->tx_bw_control_moved)
-		off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
-	else
+	off = 0;
+	switch (mp->shared->tx_bw_control) {
+	case TX_BW_CONTROL_OLD_LAYOUT:
 		off = TXQ_FIX_PRIO_CONF(mp->port_num);
+		break;
+	case TX_BW_CONTROL_NEW_LAYOUT:
+		off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+		break;
+	}
 
-	val = rdl(mp, off);
-	val &= ~(1 << txq->index);
-	wrl(mp, off, val);
+	if (off) {
+		val = rdl(mp, off);
+		val &= ~(1 << txq->index);
+		wrl(mp, off, val);
 
-	/*
-	 * Configure WRR weight for this queue.
-	 */
-	off = TXQ_BW_WRR_CONF(mp->port_num, txq->index);
+		/*
+		 * Configure WRR weight for this queue.
+		 */
+		off = TXQ_BW_WRR_CONF(mp->port_num, txq->index);
 
-	val = rdl(mp, off);
-	val = (val & ~0xff) | (weight & 0xff);
-	wrl(mp, off, val);
+		val = rdl(mp, off);
+		val = (val & ~0xff) | (weight & 0xff);
+		wrl(mp, off, val);
+	}
 }
 
 
 /* mii management interface *************************************************/
-#define SMI_BUSY		0x10000000
-#define SMI_READ_VALID		0x08000000
-#define SMI_OPCODE_READ		0x04000000
-#define SMI_OPCODE_WRITE	0x00000000
+static irqreturn_t mv643xx_eth_err_irq(int irq, void *dev_id)
+{
+	struct mv643xx_eth_shared_private *msp = dev_id;
 
-static void smi_reg_read(struct mv643xx_eth_private *mp, unsigned int addr,
-			 unsigned int reg, unsigned int *value)
+	if (readl(msp->base + ERR_INT_CAUSE) & ERR_INT_SMI_DONE) {
+		writel(~ERR_INT_SMI_DONE, msp->base + ERR_INT_CAUSE);
+		wake_up(&msp->smi_busy_wait);
+		return IRQ_HANDLED;
+	}
+
+	return IRQ_NONE;
+}
+
+static int smi_is_done(struct mv643xx_eth_shared_private *msp)
 {
-	void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
-	unsigned long flags;
-	int i;
+	return !(readl(msp->base + SMI_REG) & SMI_BUSY);
+}
 
-	/* the SMI register is a shared resource */
-	spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
+static int smi_wait_ready(struct mv643xx_eth_shared_private *msp)
+{
+	if (msp->err_interrupt == NO_IRQ) {
+		int i;
 
-	/* wait for the SMI register to become available */
-	for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
-		if (i == 1000) {
-			printk("%s: PHY busy timeout\n", mp->dev->name);
-			goto out;
+		for (i = 0; !smi_is_done(msp); i++) {
+			if (i == 10)
+				return -ETIMEDOUT;
+			msleep(10);
 		}
-		udelay(10);
+
+		return 0;
+	}
+
+	if (!wait_event_timeout(msp->smi_busy_wait, smi_is_done(msp),
+				msecs_to_jiffies(100)))
+		return -ETIMEDOUT;
+
+	return 0;
+}
+
+static int smi_bus_read(struct mii_bus *bus, int addr, int reg)
+{
+	struct mv643xx_eth_shared_private *msp = bus->priv;
+	void __iomem *smi_reg = msp->base + SMI_REG;
+	int ret;
+
+	if (smi_wait_ready(msp)) {
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
 	}
 
 	writel(SMI_OPCODE_READ | (reg << 21) | (addr << 16), smi_reg);
 
-	/* now wait for the data to be valid */
-	for (i = 0; !(readl(smi_reg) & SMI_READ_VALID); i++) {
-		if (i == 1000) {
-			printk("%s: PHY read timeout\n", mp->dev->name);
-			goto out;
-		}
-		udelay(10);
+	if (smi_wait_ready(msp)) {
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
 	}
 
-	*value = readl(smi_reg) & 0xffff;
-out:
-	spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
+	ret = readl(smi_reg);
+	if (!(ret & SMI_READ_VALID)) {
+		printk("mv643xx_eth: SMI bus read not valid\n");
+		return -ENODEV;
+	}
+
+	return ret & 0xffff;
 }
 
-static void smi_reg_write(struct mv643xx_eth_private *mp,
-			  unsigned int addr,
-			  unsigned int reg, unsigned int value)
+static int smi_bus_write(struct mii_bus *bus, int addr, int reg, u16 val)
 {
-	void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
-	unsigned long flags;
-	int i;
+	struct mv643xx_eth_shared_private *msp = bus->priv;
+	void __iomem *smi_reg = msp->base + SMI_REG;
 
-	/* the SMI register is a shared resource */
-	spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
-
-	/* wait for the SMI register to become available */
-	for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
-		if (i == 1000) {
-			printk("%s: PHY busy timeout\n", mp->dev->name);
-			goto out;
-		}
-		udelay(10);
+	if (smi_wait_ready(msp)) {
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
 	}
 
 	writel(SMI_OPCODE_WRITE | (reg << 21) |
-		(addr << 16) | (value & 0xffff), smi_reg);
-out:
-	spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
+		(addr << 16) | (val & 0xffff), smi_reg);
+
+	if (smi_wait_ready(msp)) {
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
+	}
+
+	return 0;
 }
 
 
-/* mib counters *************************************************************/
+/* statistics ***************************************************************/
+static struct net_device_stats *mv643xx_eth_get_stats(struct net_device *dev)
+{
+	struct mv643xx_eth_private *mp = netdev_priv(dev);
+	struct net_device_stats *stats = &dev->stats;
+	unsigned long tx_packets = 0;
+	unsigned long tx_bytes = 0;
+	unsigned long tx_dropped = 0;
+	int i;
+
+	for (i = 0; i < mp->txq_count; i++) {
+		struct tx_queue *txq = mp->txq + i;
+
+		tx_packets += txq->tx_packets;
+		tx_bytes += txq->tx_bytes;
+		tx_dropped += txq->tx_dropped;
+	}
+
+	stats->tx_packets = tx_packets;
+	stats->tx_bytes = tx_bytes;
+	stats->tx_dropped = tx_dropped;
+
+	return stats;
+}
+
 static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
 {
 	return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
@@ -1059,6 +1164,7 @@ static void mib_counters_update(struct mv643xx_eth_private *mp)
 {
 	struct mib_counters *p = &mp->mib_counters;
 
+	spin_lock(&mp->mib_counters_lock);
 	p->good_octets_received += mib_read(mp, 0x00);
 	p->good_octets_received += (u64)mib_read(mp, 0x04) << 32;
 	p->bad_octets_received += mib_read(mp, 0x08);
@@ -1091,6 +1197,16 @@ static void mib_counters_update(struct mv643xx_eth_private *mp)
 	p->bad_crc_event += mib_read(mp, 0x74);
 	p->collision += mib_read(mp, 0x78);
 	p->late_collision += mib_read(mp, 0x7c);
+	spin_unlock(&mp->mib_counters_lock);
+
+	mod_timer(&mp->mib_counters_timer, jiffies + 30 * HZ);
+}
+
+static void mib_counters_timer_wrapper(unsigned long _mp)
+{
+	struct mv643xx_eth_private *mp = (void *)_mp;
+
+	mib_counters_update(mp);
 }
 
 
@@ -1156,9 +1272,9 @@ static int mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
 	int err;
 
-	spin_lock_irq(&mp->lock);
-	err = mii_ethtool_gset(&mp->mii, cmd);
-	spin_unlock_irq(&mp->lock);
+	err = phy_read_status(mp->phy);
+	if (err == 0)
+		err = phy_ethtool_gset(mp->phy, cmd);
 
 	/*
 	 * The MAC does not support 1000baseT_Half.
@@ -1206,18 +1322,13 @@ static int mv643xx_eth_get_settings_phyless(struct net_device *dev, struct ethto
 static int mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 {
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
-	int err;
 
 	/*
 	 * The MAC does not support 1000baseT_Half.
 	 */
 	cmd->advertising &= ~ADVERTISED_1000baseT_Half;
 
-	spin_lock_irq(&mp->lock);
-	err = mii_ethtool_sset(&mp->mii, cmd);
-	spin_unlock_irq(&mp->lock);
-
-	return err;
+	return phy_ethtool_sset(mp->phy, cmd);
 }
 
 static int mv643xx_eth_set_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
@@ -1239,7 +1350,7 @@ static int mv643xx_eth_nway_reset(struct net_device *dev)
 {
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-	return mii_nway_restart(&mp->mii);
+	return genphy_restart_aneg(mp->phy);
 }
 
 static int mv643xx_eth_nway_reset_phyless(struct net_device *dev)
@@ -1249,14 +1360,7 @@ static int mv643xx_eth_nway_reset_phyless(struct net_device *dev)
 
 static u32 mv643xx_eth_get_link(struct net_device *dev)
 {
-	struct mv643xx_eth_private *mp = netdev_priv(dev);
-
-	return mii_link_ok(&mp->mii);
-}
-
-static u32 mv643xx_eth_get_link_phyless(struct net_device *dev)
-{
-	return 1;
+	return !!netif_carrier_ok(dev);
 }
 
 static void mv643xx_eth_get_strings(struct net_device *dev,
@@ -1277,9 +1381,10 @@ static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
 					  struct ethtool_stats *stats,
 					  uint64_t *data)
 {
-	struct mv643xx_eth_private *mp = dev->priv;
+	struct mv643xx_eth_private *mp = netdev_priv(dev);
 	int i;
 
+	mv643xx_eth_get_stats(dev);
 	mib_counters_update(mp);
 
 	for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
@@ -1323,7 +1428,7 @@ static const struct ethtool_ops mv643xx_eth_ethtool_ops_phyless = {
 	.set_settings		= mv643xx_eth_set_settings_phyless,
 	.get_drvinfo		= mv643xx_eth_get_drvinfo,
 	.nway_reset		= mv643xx_eth_nway_reset_phyless,
-	.get_link		= mv643xx_eth_get_link_phyless,
+	.get_link		= mv643xx_eth_get_link,
 	.set_sg			= ethtool_op_set_sg,
 	.get_strings		= mv643xx_eth_get_strings,
 	.get_ethtool_stats	= mv643xx_eth_get_ethtool_stats,
@@ -1487,7 +1592,7 @@ static int rxq_init(struct mv643xx_eth_private *mp, int index)
 
 	size = rxq->rx_ring_size * sizeof(struct rx_desc);
 
-	if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size) {
+	if (index == 0 && size <= mp->rx_desc_sram_size) {
 		rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
 						mp->rx_desc_sram_size);
 		rxq->rx_desc_dma = mp->rx_desc_sram_addr;
@@ -1515,20 +1620,21 @@ static int rxq_init(struct mv643xx_eth_private *mp, int index)
 
 	rx_desc = (struct rx_desc *)rxq->rx_desc_area;
 	for (i = 0; i < rxq->rx_ring_size; i++) {
-		int nexti = (i + 1) % rxq->rx_ring_size;
+		int nexti;
+