X-Git-Url: http://ftp.safe.ca/?a=blobdiff_plain;f=drivers%2Fnet%2Fmv643xx_eth.c;h=a9c8c08044b1847ebe560964433ad3ee4a564322;hb=9301975ec251bab1ad7cfcb84a688b26187e4e4a;hp=e592fac87068985b013287095b29e0fd6d8b0cd0;hpb=befefe2177d4bca07a64a2410333388c063a6f7d;p=safe%2Fjmp%2Flinux-2.6

diff --git a/drivers/net/mv643xx_eth.c b/drivers/net/mv643xx_eth.c
index e592fac..a9c8c08 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>
@@ -48,16 +49,15 @@
 #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_TX_FAST_REFILL
 
 /*
  * Registers shared between all ports.
@@ -103,7 +103,6 @@ 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				0x000003fc
 #define  INT_EXT			0x00000002
@@ -167,8 +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
 
 
 /*
@@ -250,9 +249,9 @@ struct mv643xx_eth_shared_private {
 	struct mv643xx_eth_shared_private *smi;
 
 	/*
-	 * Protects access to SMI_REG, which is shared between ports.
+	 * Provides access to local SMI interface.
 	 */
-	struct mutex phy_lock;
+	struct mii_bus *smi_bus;
 
 	/*
 	 * If we have access to the error interrupt pin (which is
@@ -273,9 +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 {
@@ -338,7 +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 {
@@ -347,13 +355,24 @@ struct mv643xx_eth_private {
 
 	struct net_device *dev;
 
-	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.
@@ -362,7 +381,6 @@ struct mv643xx_eth_private {
 	unsigned long rx_desc_sram_addr;
 	int rx_desc_sram_size;
 	int rxq_count;
-	struct napi_struct napi;
 	struct timer_list rx_oom;
 	struct rx_queue rxq[8];
 
@@ -374,9 +392,6 @@ struct mv643xx_eth_private {
 	int tx_desc_sram_size;
 	int txq_count;
 	struct tx_queue txq[8];
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-	int tx_clean_threshold;
-#endif
 };
 
 
@@ -446,87 +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 != 0);
-
-	if (txq->tx_ring_size - txq->tx_desc_count >= MAX_SKB_FRAGS + 1)
-		netif_wake_queue(mp->dev);
-}
-
-
-/* rx ***********************************************************************/
-static void txq_reclaim(struct tx_queue *txq, int force);
-
-static int rxq_refill(struct rx_queue *rxq, int budget, int *oom)
-{
-	int skb_size;
-	int refilled;
-
-	/*
-	 * 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 = rxq_to_mp(rxq)->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;
-
-	refilled = 0;
-	while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
-		struct sk_buff *skb;
-		int unaligned;
-		int rx;
-
-		skb = dev_alloc_skb(skb_size + dma_get_cache_alignment() - 1);
-		if (skb == NULL) {
-			*oom = 1;
-			break;
-		}
-
-		unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
-		if (unaligned)
-			skb_reserve(skb, dma_get_cache_alignment() - unaligned);
-
-		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,
-						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);
 	}
-
-	return refilled;
 }
 
+
+/* rx napi ******************************************************************/
 static int rxq_process(struct rx_queue *rxq, int budget)
 {
 	struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
@@ -538,6 +487,7 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 		struct rx_desc *rx_desc;
 		unsigned int cmd_sts;
 		struct sk_buff *skb;
+		u16 byte_cnt;
 
 		rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
 
@@ -558,6 +508,10 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 		rxq->rx_desc_count--;
 		rx++;
 
+		mp->work_rx_refill |= 1 << rxq->index;
+
+		byte_cnt = rx_desc->byte_cnt;
+
 		/*
 		 * Update statistics.
 		 *
@@ -567,7 +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
@@ -596,13 +550,10 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 			 * 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);
 			netif_receive_skb(skb);
 		}
@@ -610,54 +561,66 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 		mp->dev->last_rx = jiffies;
 	}
 
+	if (rx < budget)
+		mp->work_rx &= ~(1 << rxq->index);
+
 	return rx;
 }
 
-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 work_done;
-	int oom;
-	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;
 
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-	if (++mp->tx_clean_threshold > 5) {
-		mp->tx_clean_threshold = 0;
-		for (i = 0; i < mp->txq_count; i++)
-			txq_reclaim(mp->txq + i, 0);
+		skb = __skb_dequeue(&mp->rx_recycle);
+		if (skb == NULL)
+			skb = dev_alloc_skb(mp->skb_size +
+					    dma_get_cache_alignment() - 1);
 
-		spin_lock_irq(&mp->lock);
-		__txq_maybe_wake(mp->txq);
-		spin_unlock_irq(&mp->lock);
-	}
-#endif
+		if (skb == NULL) {
+			mp->work_rx_oom |= 1 << rxq->index;
+			goto oom;
+		}
 
-	work_done = 0;
-	oom = 0;
-	for (i = mp->rxq_count - 1; work_done < budget && i >= 0; i--) {
-		struct rx_queue *rxq = mp->rxq + i;
+		unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
+		if (unaligned)
+			skb_reserve(skb, dma_get_cache_alignment() - unaligned);
 
-		work_done += rxq_process(rxq, budget - work_done);
-		work_done += rxq_refill(rxq, budget - work_done, &oom);
-	}
+		refilled++;
+		rxq->rx_desc_count++;
 
-	if (work_done < budget) {
-		if (oom)
-			mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
-		netif_rx_complete(mp->dev, napi);
-		wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
-	}
+		rx = rxq->rx_used_desc++;
+		if (rxq->rx_used_desc == rxq->rx_ring_size)
+			rxq->rx_used_desc = 0;
 
-	return work_done;
-}
+		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();
 
-static inline void oom_timer_wrapper(unsigned long data)
-{
-	struct mv643xx_eth_private *mp = (void *)data;
+		/*
+		 * 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);
+	}
 
-	napi_schedule(&mp->napi);
+	if (refilled < budget)
+		mp->work_rx_refill &= ~(1 << rxq->index);
+
+oom:
+	return refilled;
 }
 
 
@@ -712,10 +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;
@@ -732,144 +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;
-
 	if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
-		spin_unlock_irqrestore(&mp->lock, flags);
-		if (txq->index == 0 && net_ratelimit())
-			dev_printk(KERN_ERR, &dev->dev,
-				   "primary tx queue full?!\n");
+		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 == 0) {
+	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_SKB_FRAGS + 1)
-			netif_stop_queue(dev);
+			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)
@@ -893,14 +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;
 	}
 }
 
@@ -932,14 +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)
@@ -951,23 +1006,30 @@ 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);
+	}
 }
 
 
@@ -1011,68 +1073,80 @@ static int smi_wait_ready(struct mv643xx_eth_shared_private *msp)
 	return 0;
 }
 
-static int smi_reg_read(struct mv643xx_eth_private *mp,
-			unsigned int addr, unsigned int reg)
+static int smi_bus_read(struct mii_bus *bus, int addr, int reg)
 {
-	struct mv643xx_eth_shared_private *msp = mp->shared->smi;
+	struct mv643xx_eth_shared_private *msp = bus->priv;
 	void __iomem *smi_reg = msp->base + SMI_REG;
 	int ret;
 
-	mutex_lock(&msp->phy_lock);
-
 	if (smi_wait_ready(msp)) {
-		printk("%s: SMI bus busy timeout\n", mp->dev->name);
-		ret = -ETIMEDOUT;
-		goto out;
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
 	}
 
 	writel(SMI_OPCODE_READ | (reg << 21) | (addr << 16), smi_reg);
 
 	if (smi_wait_ready(msp)) {
-		printk("%s: SMI bus busy timeout\n", mp->dev->name);
-		ret = -ETIMEDOUT;
-		goto out;
+		printk("mv643xx_eth: SMI bus busy timeout\n");
+		return -ETIMEDOUT;
 	}
 
 	ret = readl(smi_reg);
 	if (!(ret & SMI_READ_VALID)) {
-		printk("%s: SMI bus read not valid\n", mp->dev->name);
-		ret = -ENODEV;
-		goto out;
+		printk("mv643xx_eth: SMI bus read not valid\n");
+		return -ENODEV;
 	}
 
-	ret &= 0xffff;
-
-out:
-	mutex_unlock(&msp->phy_lock);
-
-	return ret;
+	return ret & 0xffff;
 }
 
-static int 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)
 {
-	struct mv643xx_eth_shared_private *msp = mp->shared->smi;
+	struct mv643xx_eth_shared_private *msp = bus->priv;
 	void __iomem *smi_reg = msp->base + SMI_REG;
 
-	mutex_lock(&msp->phy_lock);
-
 	if (smi_wait_ready(msp)) {
-		printk("%s: SMI bus busy timeout\n", mp->dev->name);
-		mutex_unlock(&msp->phy_lock);
+		printk("mv643xx_eth: SMI bus busy timeout\n");
 		return -ETIMEDOUT;
 	}
 
 	writel(SMI_OPCODE_WRITE | (reg << 21) |
-		(addr << 16) | (value & 0xffff), smi_reg);
+		(addr << 16) | (val & 0xffff), smi_reg);
 
-	mutex_unlock(&msp->phy_lock);
+	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);
@@ -1090,6 +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);
@@ -1122,6 +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);
 }
 
 
@@ -1187,7 +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;
 
-	err = mii_ethtool_gset(&mp->mii, cmd);
+	err = phy_read_status(mp->phy);
+	if (err == 0)
+		err = phy_ethtool_gset(mp->phy, cmd);
 
 	/*
 	 * The MAC does not support 1000baseT_Half.
@@ -1241,7 +1328,7 @@ static int mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *
 	 */
 	cmd->advertising &= ~ADVERTISED_1000baseT_Half;
 
-	return mii_ethtool_sset(&mp->mii, cmd);
+	return phy_ethtool_sset(mp->phy, cmd);
 }
 
 static int mv643xx_eth_set_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
@@ -1263,7 +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)
@@ -1273,14 +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,
@@ -1304,6 +1384,7 @@ static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
 	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++) {
@@ -1347,7 +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,
@@ -1624,18 +1705,11 @@ static int txq_init(struct mv643xx_eth_private *mp, int index)
 	if (txq->tx_desc_area == NULL) {
 		dev_printk(KERN_ERR, &mp->dev->dev,
 			   "can't allocate tx ring (%d bytes)\n", size);
-		goto out;
+		return -ENOMEM;
 	}
 	memset(txq->tx_desc_area, 0, size);
 
 	txq->tx_desc_area_size = size;
-	txq->tx_skb = kmalloc(txq->tx_ring_size * sizeof(*txq->tx_skb),
-								GFP_KERNEL);
-	if (txq->tx_skb == NULL) {
-		dev_printk(KERN_ERR, &mp->dev->dev,
-			   "can't allocate tx skb ring\n");
-		goto out_free;
-	}
 
 	tx_desc = (struct tx_desc *)txq->tx_desc_area;
 	for (i = 0; i < txq->tx_ring_size; i++) {
@@ -1651,99 +1725,77 @@ static int txq_init(struct mv643xx_eth_private *mp, int index)
 					nexti * sizeof(struct tx_desc);
 	}
 
-	return 0;
-
-
-out_free:
-	if (index == 0 && size <= mp->tx_desc_sram_size)
-		iounmap(txq->tx_desc_area);
-	else
-		dma_free_coherent(NULL, size,
-				  txq->tx_desc_area,
-				  txq->tx_desc_dma);
+	skb_queue_head_init(&txq->tx_skb);
 
-out:
-	return -ENOMEM;
+	return 0;
 }
 
-static void txq_reclaim(struct tx_queue *txq, int force)
+static void txq_deinit(struct tx_queue *txq)
 {
 	struct mv643xx_eth_private *mp = txq_to_mp(txq);
-	unsigned long flags;
 
-	spin_lock_irqsave(&mp->lock, flags);
-	while (txq->tx_desc_count > 0) {
-		int tx_index;
-		struct tx_desc *desc;
-		u32 cmd_sts;
-		struct sk_buff *skb;
-		dma_addr_t addr;
-		int count;
-
-		tx_index = txq->tx_used_desc;
-		desc = &txq->tx_desc_area[tx_index];
-		cmd_sts = desc->cmd_sts;
+	txq_disable(txq);
+	txq_reclaim(txq, txq->tx_ring_size, 1);
 
-		if (cmd_sts & BUFFER_OWNED_BY_DMA) {
-			if (!force)
-				break;
-			desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
-		}
+	BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
 
-		txq->tx_used_desc = tx_index + 1;
-		if (txq->tx_used_desc == txq->tx_ring_size)
-			txq->tx_used_desc = 0;
-		txq->tx_desc_count--;
+	if (txq->index == 0 &&
+	    txq->tx_desc_area_size <= mp->tx_desc_sram_size)
+		iounmap(txq->tx_desc_area);
+	else
+		dma_free_coherent(NULL, txq->tx_desc_area_size,
+				  txq->tx_desc_area, txq->tx_desc_dma);
+}
 
-		addr = desc->buf_ptr;
-		count = desc->byte_cnt;
-		skb = txq->tx_skb[tx_index];
-		txq->tx_skb[tx_index] = NULL;
 
-		if (cmd_sts & ERROR_SUMMARY) {
-			dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
-			mp->dev->stats.tx_errors++;
-		}
+/* netdev ops and related ***************************************************/
+static int mv643xx_eth_collect_events(struct mv643xx_eth_private *mp)
+{
+	u32 int_cause;
+	u32 int_cause_ext;
 
-		/*
-		 * Drop mp->lock while we free the skb.
-		 */
-		spin_unlock_irqrestore(&mp->lock, flags);
+	int_cause = rdl(mp, INT_CAUSE(mp->port_num)) &
+			(INT_TX_END | INT_RX | INT_EXT);
+	if (int_cause == 0)
+		return 0;
 
-		if (cmd_sts & TX_FIRST_DESC)
-			dma_unmap_single(NULL, addr, count, DMA_TO_DEVICE);
-		else
-			dma_unmap_page(NULL, addr, count, DMA_TO_DEVICE);
+	int_cause_ext = 0;
+	if (int_cause & INT_EXT)
+		int_cause_ext = rdl(mp, INT_CAUSE_EXT(mp->port_num));
 
-		if (skb)
-			dev_kfree_skb_irq(skb);
+	int_cause &= INT_TX_END | INT_RX;
+	if (int_cause) {
+		wrl(mp, INT_CAUSE(mp->port_num), ~int_cause);
+		mp->work_tx_end |= ((int_cause & INT_TX_END) >> 19) &
+				~(rdl(mp, TXQ_COMMAND(mp->port_num)) & 0xff);
+		mp->work_rx |= (int_cause & INT_RX) >> 2;
+	}
 
-		spin_lock_irqsave(&mp->lock, flags);
+	int_cause_ext &= INT_EXT_LINK_PHY | INT_EXT_TX;
+	if (int_cause_ext) {
+		wrl(mp, INT_CAUSE_EXT(mp->port_num), ~int_cause_ext);
+		if (int_cause_ext & INT_EXT_LINK_PHY)
+			mp->work_link = 1;
+		mp->work_tx |= int_cause_ext & INT_EXT_TX;
 	}
-	spin_unlock_irqrestore(&mp->lock, flags);
+
+	return 1;
 }
 
-static void txq_deinit(struct tx_queue *txq)
+static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
 {
-	struct mv643xx_eth_private *mp = txq_to_mp(txq);
-
-	txq_disable(txq);
-	txq_reclaim(txq, 1);
+	struct net_device *dev = (struct net_device *)dev_id;
+	struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-	BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
+	if (unlikely(!mv643xx_eth_collect_events(mp)))
+		return IRQ_NONE;
 
-	if (txq->index == 0 &&
-	    txq->tx_desc_area_size <= mp->tx_desc_sram_size)
-		iounmap(txq->tx_desc_area);
-	else
-		dma_free_coherent(NULL, txq->tx_desc_area_size,
-				  txq->tx_desc_area, txq->tx_desc_dma);
+	wrl(mp, INT_MASK(mp->port_num), 0);
+	napi_schedule(&mp->napi);
 
-	kfree(txq->tx_skb);
+	return IRQ_HANDLED;
 }
 
-
-/* netdev ops and related ***************************************************/
 static void handle_link_event(struct mv643xx_eth_private *mp)
 {
 	struct net_device *dev = mp->dev;
@@ -1764,7 +1816,7 @@ static void handle_link_event(struct mv643xx_eth_private *mp)
 			for (i = 0; i < mp->txq_count; i++) {
 				struct tx_queue *txq = mp->txq + i;
 
-				txq_reclaim(txq, 1);
+				txq_reclaim(txq, txq->tx_ring_size, 1);
 				txq_reset_hw_ptr(txq);
 			}
 		}
@@ -1797,102 +1849,88 @@ static void handle_link_event(struct mv643xx_eth_private *mp)
 		netif_carrier_on(dev);
 }
 
-static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
+static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
 {
-	struct net_device *dev = (struct net_device *)dev_id;
-	struct mv643xx_eth_private *mp = netdev_priv(dev);
-	u32 int_cause;
-	u32 int_cause_ext;
+	struct mv643xx_eth_private *mp;
+	int work_done;
 
-	int_cause = rdl(mp, INT_CAUSE(mp->port_num)) &
-			(INT_TX_END | INT_RX | INT_EXT);
-	if (int_cause == 0)
-		return IRQ_NONE;
+	mp = container_of(napi, struct mv643xx_eth_private, napi);
 
-	int_cause_ext = 0;
-	if (int_cause & INT_EXT) {
-		int_cause_ext = rdl(mp, INT_CAUSE_EXT(mp->port_num))
-				& (INT_EXT_LINK_PHY | INT_EXT_TX);
-		wrl(mp, INT_CAUSE_EXT(mp->port_num), ~int_cause_ext);
-	}
+	mp->work_rx_refill |= mp->work_rx_oom;
+	mp->work_rx_oom = 0;
 
-	if (int_cause_ext & INT_EXT_LINK_PHY)
-		handle_link_event(mp);
+	work_done = 0;
+	while (work_done < budget) {
+		u8 queue_mask;
+		int queue;
+		int work_tbd;
+
+		if (mp->work_link) {
+			mp->work_link = 0;
+			handle_link_event(mp);
+			continue;
+		}
 
-	/*
-	 * RxBuffer or RxError set for any of the 8 queues?
-	 */
-	if (int_cause & INT_RX) {
-		wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_RX));
-		wrl(mp, INT_MASK(mp->port_num), 0x00000000);
-		rdl(mp, INT_MASK(mp->port_num));
+		queue_mask = mp->work_tx | mp->work_tx_end |
+				mp->work_rx | mp->work_rx_refill;
+		if (!queue_mask) {
+			if (mv643xx_eth_collect_events(mp))
+				continue;
+			break;
+		}
 
-		napi_schedule(&mp->napi);
+		queue = fls(queue_mask) - 1;
+		queue_mask = 1 << queue;
+
+		work_tbd = budget - work_done;
+		if (work_tbd > 16)
+			work_tbd = 16;
+
+		if (mp->work_tx_end & queue_mask) {
+			txq_kick(mp->txq + queue);
+		} else if (mp->work_tx & queue_mask) {
+			work_done += txq_reclaim(mp->txq + queue, work_tbd, 0);
+			txq_maybe_wake(mp->txq + queue);
+		} else if (mp->work_rx & queue_mask) {
+			work_done += rxq_process(mp->rxq + queue, work_tbd);
+		} else if (mp->work_rx_refill & queue_mask) {
+			work_done += rxq_refill(mp->rxq + queue, work_tbd);
+		} else {
+			BUG();
+		}
 	}
 
-	/*
-	 * TxBuffer or TxError set for any of the 8 queues?
-	 */
-	if (int_cause_ext & INT_EXT_TX) {
-		int i;
-
-		for (i = 0; i < mp->txq_count; i++)
-			txq_reclaim(mp->txq + i, 0);
-
-		/*
-		 * Enough space again in the primary TX queue for a
-		 * full packet?
-		 */
-		spin_lock(&mp->lock);
-		__txq_maybe_wake(mp->txq);
-		spin_unlock(&mp->lock);
+	if (work_done < budget) {
+		if (mp->work_rx_oom)
+			mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
+		napi_complete(napi);
+		wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
 	}
 
-	/*
-	 * Any TxEnd interrupts?
-	 */
-	if (int_cause & INT_TX_END) {
-		int i;
-
-		wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_TX_END));
-
-		spin_lock(&mp->lock);
-		for (i = 0; i < 8; i++) {
-			struct tx_queue *txq = mp->txq + i;
-			u32 hw_desc_ptr;
-			u32 expected_ptr;
-
-			if ((int_cause & (INT_TX_END_0 << i)) == 0)
-				continue;
-
-			hw_desc_ptr =
-				rdl(mp, TXQ_CURRENT_DESC_PTR(mp->port_num, i));
-			expected_ptr = (u32)txq->tx_desc_dma +
-				txq->tx_curr_desc * sizeof(struct tx_desc);
+	return work_done;
+}
 
-			if (hw_desc_ptr != expected_ptr)
-				txq_enable(txq);
-		}
-		spin_unlock(&mp->lock);
-	}
+static inline void oom_timer_wrapper(unsigned long data)
+{
+	struct mv643xx_eth_private *mp = (void *)data;
 
-	return IRQ_HANDLED;
+	napi_schedule(&mp->napi);
 }
 
 static void phy_reset(struct mv643xx_eth_private *mp)
 {
 	int data;
 
-	data = smi_reg_read(mp, mp->phy_addr, MII_BMCR);
+	data = phy_read(mp->phy, MII_BMCR);
 	if (data < 0)
 		return;
 
 	data |= BMCR_RESET;
-	if (smi_reg_write(mp, mp->phy_addr, MII_BMCR, data) < 0)
+	if (phy_write(mp->phy, MII_BMCR, data) < 0)
 		return;
 
 	do {
-		data = smi_reg_read(mp, mp->phy_addr, MII_BMCR);
+		data = phy_read(mp->phy, MII_BMCR);
 	} while (data >= 0 && data & BMCR_RESET);
 }
 
@@ -1904,7 +1942,7 @@ static void port_start(struct mv643xx_eth_private *mp)
 	/*
 	 * Perform PHY reset, if there is a PHY.
 	 */
-	if (mp->phy_addr != -1) {
+	if (mp->phy != NULL) {
 		struct ethtool_cmd cmd;
 
 		mv643xx_eth_get_settings(mp->dev, &cmd);
@@ -1921,7 +1959,7 @@ static void port_start(struct mv643xx_eth_private *mp)
 	wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
 
 	pscr |= DO_NOT_FORCE_LINK_FAIL;
-	if (mp->phy_addr == -1)
+	if (mp->phy == NULL)
 		pscr |= FORCE_LINK_PASS;
 	wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
 
@@ -1946,9 +1984,10 @@ static void port_start(struct mv643xx_eth_private *mp)
 
 	/*
 	 * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
-	 * frames to RX queue #0.
+	 * frames to RX queue #0, and include the pseudo-header when
+	 * calculating receive checksums.
 	 */
-	wrl(mp, PORT_CONFIG(mp->port_num), 0x00000000);
+	wrl(mp, PORT_CONFIG(mp->port_num), 0x02000000);
 
 	/*
 	 * Treat BPDUs as normal multicasts, and disable partition mode.
@@ -2001,11 +2040,30 @@ static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
 	wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), (coal & 0x3fff) << 4);
 }
 
+static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
+{
+	int skb_size;
+
+	/*
+	 * 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.
+	 */
+	mp->skb_size = (skb_size + 7) & ~7;
+}
+
 static int mv643xx_eth_open(struct net_device *dev)
 {
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
 	int err;
-	int oom;
 	int i;
 
 	wrl(mp, INT_CAUSE(mp->port_num), 0);
@@ -2021,9 +2079,12 @@ static int mv643xx_eth_open(struct net_device *dev)
 
 	init_mac_tables(mp);
 
+	mv643xx_eth_recalc_skb_size(mp);
+
 	napi_enable(&mp->napi);
 
-	oom = 0;
+	skb_queue_head_init(&mp->rx_recycle);
+
 	for (i = 0; i < mp->rxq_count; i++) {
 		err = rxq_init(mp, i);
 		if (err) {
@@ -2032,10 +2093,10 @@ static int mv643xx_eth_open(struct net_device *dev)
 			goto out;
 		}
 
-		rxq_refill(mp->rxq + i, INT_MAX, &oom);
+		rxq_refill(mp->rxq + i, INT_MAX);
 	}
 
-	if (oom) {
+	if (mp->work_rx_oom) {
 		mp->rx_oom.expires = jiffies + (HZ / 10);
 		add_timer(&mp->rx_oom);
 	}
@@ -2105,6 +2166,8 @@ static int mv643xx_eth_stop(struct net_device *dev)
 	wrl(mp, INT_MASK(mp->port_num), 0x00000000);
 	rdl(mp, INT_MASK(mp->port_num));
 
+	del_timer_sync(&mp->mib_counters_timer);
+
 	napi_disable(&mp->napi);
 
 	del_timer_sync(&mp->rx_oom);
@@ -2114,8 +2177,11 @@ static int mv643xx_eth_stop(struct net_device *dev)
 	free_irq(dev->irq, dev);
 
 	port_reset(mp);
+	mv643xx_eth_get_stats(dev);
 	mib_counters_update(mp);
 
+	skb_queue_purge(&mp->rx_recycle);
+
 	for (i = 0; i < mp->rxq_count; i++)
 		rxq_deinit(mp->rxq + i);
 	for (i = 0; i < mp->txq_count; i++)
@@ -2128,8 +2194,8 @@ static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 {
 	struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-	if (mp->phy_addr != -1)
-		return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
+	if (mp->phy != NULL)
+		return phy_mii_ioctl(mp->phy, if_mii(ifr), cmd);
 
 	return -EOPNOTSUPP;
 }
@@ -2142,6 +2208,7 @@ static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
 		return -EINVAL;
 
 	dev->mtu = new_mtu;
+	mv643xx_eth_recalc_skb_size(mp);
 	tx_set_rate(mp, 1000000000, 16777216);
 
 	if (!netif_running(dev))
@@ -2169,10 +2236,10 @@ static void tx_timeout_task(struct work_struct *ugly)
 
 	mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
 	if (netif_running(mp->dev)) {
-		netif_stop_queue(mp->dev);
+		netif_tx_stop_all_queues(mp->dev);
 		port_reset(mp);
 		port_start(mp);
-		netif_wake_queue(mp->dev);
+		netif_tx_wake_all_queues(mp->dev);
 	}
 }
 
@@ -2199,18 +2266,6 @@ static void mv643xx_eth_netpoll(struct net_device *dev)
 }
 #endif
 
-static int mv643xx_eth_mdio_read(struct net_device *dev, int addr, int reg)
-{
-	struct mv643xx_eth_private *mp = netdev_priv(dev);
-	return smi_reg_read(mp, addr, reg);
-}
-
-static void mv643xx_eth_mdio_write(struct net_device *dev, int addr, int reg, int val)
-{
-	struct mv643xx_eth_private *mp = netdev_priv(dev);
-	smi_reg_write(mp, addr, reg, val);
-}
-
 
 /* platform glue ************************************************************/
 static void
@@ -2262,14 +2317,20 @@ static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
 		msp->extended_rx_coal_limit = 0;
 
 	/*
-	 * Check whether the TX rate control registers are in the
-	 * old or the new place.
+	 * Check whether the MAC supports TX rate control, and if
+	 * yes, whether its associated registers are in the old or
+	 * the new place.
 	 */
 	writel(1, msp->base + TX_BW_MTU_MOVED(0));
-	if (readl(msp->base + TX_BW_MTU_MOVED(0)) & 1)
-		msp->tx_bw_control_moved = 1;
-	else
-		msp->tx_bw_control_moved = 0;
+	if (readl(msp->base + TX_BW_MTU_MOVED(0)) & 1) {
+		msp->tx_bw_control = TX_BW_CONTROL_NEW_LAYOUT;
+	} else {
+		writel(7, msp->base + TX_BW_RATE(0));
+		if (readl(msp->base + TX_BW_RATE(0)) & 7)
+			msp->tx_bw_control = TX_BW_CONTROL_OLD_LAYOUT;
+		else
+			msp->tx_bw_control = TX_BW_CONTROL_ABSENT;
+	}
 }
 
 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
@@ -2299,11 +2360,27 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 	if (msp->base == NULL)
 		goto out_free;
 
-	msp->smi = msp;
-	if (pd != NULL && pd->shared_smi != NULL)
+	/*
+	 * Set up and register SMI bus.
+	 */
+	if (pd == NULL || pd->shared_smi == NULL) {
+		msp->smi_bus = mdiobus_alloc();
+		if (msp->smi_bus == NULL)
+			goto out_unmap;
+
+		msp->smi_bus->priv = msp;
+		msp->smi_bus->name = "mv643xx_eth smi";
+		msp->smi_bus->read = smi_bus_read;
+		msp->smi_bus->write = smi_bus_write,
+		snprintf(msp->smi_bus->id, MII_BUS_ID_SIZE, "%d", pdev->id);
+		msp->smi_bus->parent = &pdev->dev;
+		msp->smi_bus->phy_mask = 0xffffffff;
+		if (mdiobus_register(msp->smi_bus) < 0)
+			goto out_free_mii_bus;
+		msp->smi = msp;
+	} else {
 		msp->smi = platform_get_drvdata(pd->shared_smi);
-
-	mutex_init(&msp->phy_lock);
+	}
 
 	msp->err_interrupt = NO_IRQ;
 	init_waitqueue_head(&msp->smi_busy_wait);
@@ -2339,6 +2416,10 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 
 	return 0;
 
+out_free_mii_bus:
+	mdiobus_free(msp->smi_bus);
+out_unmap:
+	iounmap(msp->base);
 out_free:
 	kfree(msp);
 out:
@@ -2348,7 +2429,12 @@ out:
 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
 {
 	struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
+	struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
 
+	if (pd == NULL || pd->shared_smi == NULL) {
+		mdiobus_free(msp->smi_bus);
+		mdiobus_unregister(msp->smi_bus);
+	}
 	if (msp->err_interrupt != NO_IRQ)
 		free_irq(msp->err_interrupt, msp);
 	iounmap(msp->base);
@@ -2396,17 +2482,6 @@ static void set_params(struct mv643xx_eth_private *mp,
 	else
 		uc_addr_get(mp, dev->dev_addr);
 
-	if (pd->phy_addr == MV643XX_ETH_PHY_NONE) {
-		mp->phy_addr = -1;
-	} else {
-		if (pd->phy_addr != MV643XX_ETH_PHY_ADDR_DEFAULT) {
-			mp->phy_addr = pd->phy_addr & 0x3f;
-			phy_addr_set(mp, mp->phy_addr);
-		} else {
-			mp->phy_addr = phy_addr_get(mp);
-		}
-	}
-
 	mp->default_rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
 	if (pd->rx_queue_size)
 		mp->default_rx_ring_size = pd->rx_queue_size;
@@ -2424,76 +2499,60 @@ static void set_params(struct mv643xx_eth_private *mp,
 	mp->txq_count = pd->tx_queue_count ? : 1;
 }
 
-static int phy_detect(struct mv643xx_eth_private *mp)
+static struct phy_device *phy_scan(struct mv643xx_eth_private *mp,
+				   int phy_addr)
 {
-	int data;
-	int data2;
-
-	data = smi_reg_read(mp, mp->phy_addr, MII_BMCR);
-	if (data < 0)
-		return -ENODEV;
+	struct mii_bus *bus = mp->shared->smi->smi_bus;
+	struct phy_device *phydev;
+	int start;
+	int num;
+	int i;
 
-	if (smi_reg_write(mp, mp->phy_addr, MII_BMCR, data ^ BMCR_ANENABLE) < 0)
-		return -ENODEV;
+	if (phy_addr == MV643XX_ETH_PHY_ADDR_DEFAULT) {
+		start = phy_addr_get(mp) & 0x1f;
+		num = 32;
+	} else {
+		start = phy_addr & 0x1f;
+		num = 1;
+	}
 
-	data2 = smi_reg_read(mp, mp->phy_addr, MII_BMCR);
-	if (data2 < 0)
-		return -ENODEV;
+	phydev = NULL;
+	for (i = 0; i < num; i++) {
+		int addr = (start + i) & 0x1f;
 
-	if (((data ^ data2) & BMCR_ANENABLE) == 0)
-		return -ENODEV;
+		if (bus->phy_map[addr] == NULL)
+			mdiobus_scan(bus, addr);
 
-	smi_reg_write(mp, mp->phy_addr, MII_BMCR, data);
+		if (phydev == NULL) {
+			phydev = bus->phy_map[addr];
+			if (phydev != NULL)
+				phy_addr_set(mp, addr);
+		}
+	}
 
-	return 0;
+	return phydev;
 }
 
-static int phy_init(struct mv643xx_eth_private *mp,
-		    struct mv643xx_eth_platform_data *pd)
+static void phy_init(struct mv643xx_eth_private *mp, int speed, int duplex)
 {
-	struct ethtool_cmd cmd;
-	int err;
+	struct phy_device *phy = mp->phy;
 
-	err = phy_detect(mp);
-	if (err) {
-		dev_printk(KERN_INFO, &mp->dev->dev,
-			   "no PHY detected at addr %d\n", mp->phy_addr);
-		return err;
-	}
 	phy_reset(mp);
 
-	mp->mii.phy_id = mp->phy_addr;
-	mp->mii.phy_id_mask = 0x3f;
-	mp->mii.reg_num_mask = 0x1f;
-	mp->mii.dev = mp->dev;
-	mp->mii.mdio_read = mv643xx_eth_mdio_read;
-	mp->mii.mdio_write = mv643xx_eth_mdio_write;
-
-	mp->mii.supports_gmii = mii_check_gmii_support(&mp->mii);
-
-	memset(&cmd, 0, sizeof(cmd));
-
-	cmd.port = PORT_MII;
-	cmd.transceiver = XCVR_INTERNAL;
-	cmd.phy_address = mp->phy_addr;
-	if (pd->speed == 0) {
-		cmd.autoneg = AUTONEG_ENABLE;
-		cmd.speed = SPEED_100;
-		cmd.advertising = ADVERTISED_10baseT_Half  |
-				  ADVERTISED_10baseT_Full  |
-				  ADVERTISED_100baseT_Half |
-				  ADVERTISED_100baseT_Full;
-		if (mp->mii.supports_gmii)
-			cmd.advertising |= ADVERTISED_1000baseT_Full;
+	phy_attach(mp->dev, phy->dev.bus_id, 0, PHY_INTERFACE_MODE_GMII);
+
+	if (speed == 0) {
+		phy->autoneg = AUTONEG_ENABLE;
+		phy->speed = 0;
+		phy->duplex = 0;
+		phy->advertising = phy->supported | ADVERTISED_Autoneg;
 	} else {
-		cmd.autoneg = AUTONEG_DISABLE;
-		cmd.speed = pd->speed;
-		cmd.duplex = pd->duplex;
+		phy->autoneg = AUTONEG_DISABLE;
+		phy->advertising = 0;
+		phy->speed = speed;
+		phy->duplex = duplex;
 	}
-
-	mv643xx_eth_set_settings(mp->dev, &cmd);
-
-	return 0;
+	phy_start_aneg(phy);
 }
 
 static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
@@ -2507,7 +2566,7 @@ static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
 	}
 
 	pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
-	if (mp->phy_addr == -1) {
+	if (mp->phy == NULL) {
 		pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
 		if (speed == SPEED_1000)
 			pscr |= SET_GMII_SPEED_TO_1000;
@@ -2546,7 +2605,7 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
 		return -ENODEV;
 	}
 
-	dev = alloc_etherdev(sizeof(struct mv643xx_eth_private));
+	dev = alloc_etherdev_mq(sizeof(struct mv643xx_eth_private), 8);
 	if (!dev)
 		return -ENOMEM;
 
@@ -2559,23 +2618,33 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
 	mp->dev = dev;
 
 	set_params(mp, pd);
+	dev->real_num_tx_queues = mp->txq_count;
 
-	spin_lock_init(&mp->lock);
-
-	mib_counters_clear(mp);
-	INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
-
-	if (mp->phy_addr != -1) {
-		err = phy_init(mp, pd);
-		if (err)
-			goto out;
+	if (pd->phy_addr != MV643XX_ETH_PHY_NONE)
+		mp->phy = phy_scan(mp, pd->phy_addr);
 
+	if (mp->phy != NULL) {
+		phy_init(mp, pd->speed, pd->duplex);
 		SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
 	} else {
 		SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops_phyless);
 	}
+
 	init_pscr(mp, pd->speed, pd->duplex);
 
+
+	mib_counters_clear(mp);
+
+	init_timer(&mp->mib_counters_timer);
+	mp->mib_counters_timer.data = (unsigned long)mp;
+	mp->mib_counters_timer.function = mib_counters_timer_wrapper;
+	mp->mib_counters_timer.expires = jiffies + 30 * HZ;
+	add_timer(&mp->mib_counters_timer);
+
+	spin_lock_init(&mp->mib_counters_lock);
+
+	INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
+
 	netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, 128);
 
 	init_timer(&mp->rx_oom);
@@ -2587,6 +2656,7 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
 	BUG_ON(!res);
 	dev->irq = res->start;
 
+	dev->get_stats = mv643xx_eth_get_stats;
 	dev->hard_start_xmit = mv643xx_eth_xmit;
 	dev->open = mv643xx_eth_open;
 	dev->stop = mv643xx_eth_stop;
@@ -2632,6 +2702,8 @@ static int mv643xx_eth_remove(struct platform_device *pdev)
 	struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
 	unregister_netdev(mp->dev);
+	if (mp->phy != NULL)
+		phy_detach(mp->phy);
 	flush_scheduled_work();
 	free_netdev(mp->dev);