]> err.no Git - linux-2.6/blobdiff - drivers/net/mv643xx_eth.c
at91_ether: use netstats in net_device structure
[linux-2.6] / drivers / net / mv643xx_eth.c
index 5bd4b38e953d9a0b87656fcac695d408ad673d1e..83a877f3a553ee94611dced9e87d28f2222aecaf 100644 (file)
 #include <asm/system.h>
 
 static char mv643xx_eth_driver_name[] = "mv643xx_eth";
-static char mv643xx_eth_driver_version[] = "1.0";
+static char mv643xx_eth_driver_version[] = "1.1";
 
 #define MV643XX_ETH_CHECKSUM_OFFLOAD_TX
 #define MV643XX_ETH_NAPI
 #define MV643XX_ETH_TX_FAST_REFILL
-#undef MV643XX_ETH_COAL
-
-#define MV643XX_ETH_TX_COAL 100
-#ifdef MV643XX_ETH_COAL
-#define MV643XX_ETH_RX_COAL 100
-#endif
 
 #ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
 #define MAX_DESCS_PER_SKB      (MAX_SKB_FRAGS + 1)
@@ -73,8 +67,6 @@ static char mv643xx_eth_driver_version[] = "1.0";
 #define MAX_DESCS_PER_SKB      1
 #endif
 
-#define ETH_HW_IP_ALIGN                2
-
 /*
  * Registers shared between all ports.
  */
@@ -99,22 +91,33 @@ static char mv643xx_eth_driver_version[] = "1.0";
 #define PORT_STATUS(p)                 (0x0444 + ((p) << 10))
 #define  TX_FIFO_EMPTY                 0x00000400
 #define TXQ_COMMAND(p)                 (0x0448 + ((p) << 10))
+#define TXQ_FIX_PRIO_CONF(p)           (0x044c + ((p) << 10))
+#define TX_BW_RATE(p)                  (0x0450 + ((p) << 10))
 #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_RX                                0x00000804
+#define  INT_TX_END                    0x07f80000
+#define  INT_RX                                0x0007fbfc
 #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                    0x00000101
+#define  INT_EXT_TX                    0x0000ffff
 #define INT_MASK(p)                    (0x0468 + ((p) << 10))
 #define INT_MASK_EXT(p)                        (0x046c + ((p) << 10))
 #define TX_FIFO_URGENT_THRESHOLD(p)    (0x0474 + ((p) << 10))
-#define RXQ_CURRENT_DESC_PTR(p)                (0x060c + ((p) << 10))
+#define TXQ_FIX_PRIO_CONF_MOVED(p)     (0x04dc + ((p) << 10))
+#define TX_BW_RATE_MOVED(p)            (0x04e0 + ((p) << 10))
+#define TX_BW_MTU_MOVED(p)             (0x04e8 + ((p) << 10))
+#define TX_BW_BURST_MOVED(p)           (0x04ec + ((p) << 10))
+#define RXQ_CURRENT_DESC_PTR(p, q)     (0x060c + ((p) << 10) + ((q) << 4))
 #define RXQ_COMMAND(p)                 (0x0680 + ((p) << 10))
-#define TXQ_CURRENT_DESC_PTR(p)                (0x06c0 + ((p) << 10))
+#define TXQ_CURRENT_DESC_PTR(p, q)     (0x06c0 + ((p) << 10) + ((q) << 2))
+#define TXQ_BW_TOKENS(p, q)            (0x0700 + ((p) << 10) + ((q) << 4))
+#define TXQ_BW_CONF(p, q)              (0x0704 + ((p) << 10) + ((q) << 4))
+#define TXQ_BW_WRR_CONF(p, q)          (0x0708 + ((p) << 10) + ((q) << 4))
 #define MIB_COUNTERS(p)                        (0x1000 + ((p) << 7))
 #define SPECIAL_MCAST_TABLE(p)         (0x1400 + ((p) << 10))
 #define OTHER_MCAST_TABLE(p)           (0x1500 + ((p) << 10))
@@ -164,12 +167,6 @@ static char mv643xx_eth_driver_version[] = "1.0";
 #define DEFAULT_RX_QUEUE_SIZE          400
 #define DEFAULT_TX_QUEUE_SIZE          800
 
-/* SMI reg */
-#define SMI_BUSY               0x10000000      /* 0 - Write, 1 - Read  */
-#define SMI_READ_VALID         0x08000000      /* 0 - Write, 1 - Read  */
-#define SMI_OPCODE_WRITE       0               /* Completion of Read   */
-#define SMI_OPCODE_READ                0x04000000      /* Operation is in progress */
-
 
 /*
  * RX/TX descriptors.
@@ -237,14 +234,27 @@ struct tx_desc {
 
 /* global *******************************************************************/
 struct mv643xx_eth_shared_private {
+       /*
+        * Ethernet controller base address.
+        */
        void __iomem *base;
 
-       /* used to protect SMI_REG, which is shared across ports */
+       /*
+        * Protects access to SMI_REG, which is shared between ports.
+        */
        spinlock_t phy_lock;
 
+       /*
+        * Per-port MBUS window access register value.
+        */
        u32 win_protect;
 
+       /*
+        * Hardware-specific parameters.
+        */
        unsigned int t_clk;
+       int extended_rx_coal_limit;
+       int tx_bw_control_moved;
 };
 
 
@@ -283,6 +293,8 @@ struct mib_counters {
 };
 
 struct rx_queue {
+       int index;
+
        int rx_ring_size;
 
        int rx_desc_count;
@@ -297,41 +309,34 @@ struct rx_queue {
        struct timer_list rx_oom;
 };
 
-struct mv643xx_eth_private {
-       struct mv643xx_eth_shared_private *shared;
-       int port_num;                   /* User Ethernet port number    */
-
-       struct mv643xx_eth_shared_private *shared_smi;
-
-       u32 tx_sram_addr;               /* Base address of tx sram area */
-       u32 tx_sram_size;               /* Size of tx sram area         */
+struct tx_queue {
+       int index;
 
-       /* Tx/Rx rings managment indexes fields. For driver use */
-
-       /* Next available and first returning Tx resource */
-       int tx_curr_desc, tx_used_desc;
+       int tx_ring_size;
 
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-       u32 tx_clean_threshold;
-#endif
+       int tx_desc_count;
+       int tx_curr_desc;
+       int tx_used_desc;
 
        struct tx_desc *tx_desc_area;
        dma_addr_t tx_desc_dma;
        int tx_desc_area_size;
        struct sk_buff **tx_skb;
+};
 
-       struct work_struct tx_timeout_task;
+struct mv643xx_eth_private {
+       struct mv643xx_eth_shared_private *shared;
+       int port_num;
 
        struct net_device *dev;
-       struct mib_counters mib_counters;
+
+       struct mv643xx_eth_shared_private *shared_smi;
+       int phy_addr;
+
        spinlock_t lock;
-       /* Size of Tx Ring per queue */
-       int tx_ring_size;
-       /* Number of tx descriptors in use */
-       int tx_desc_count;
 
-       u32 rx_int_coal;
-       u32 tx_int_coal;
+       struct mib_counters mib_counters;
+       struct work_struct tx_timeout_task;
        struct mii_if_info mii;
 
        /*
@@ -340,8 +345,23 @@ 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;
-       struct rx_queue rxq[1];
+       struct rx_queue rxq[8];
+
+       /*
+        * TX state.
+        */
+       int default_tx_ring_size;
+       unsigned long tx_desc_sram_addr;
+       int tx_desc_sram_size;
+       u8 txq_mask;
+       int txq_primary;
+       struct tx_queue txq[8];
+#ifdef MV643XX_ETH_TX_FAST_REFILL
+       int tx_clean_threshold;
+#endif
 };
 
 
@@ -360,58 +380,63 @@ static inline void wrl(struct mv643xx_eth_private *mp, int offset, u32 data)
 /* rxq/txq helper functions *************************************************/
 static struct mv643xx_eth_private *rxq_to_mp(struct rx_queue *rxq)
 {
-       return container_of(rxq, struct mv643xx_eth_private, rxq[0]);
+       return container_of(rxq, struct mv643xx_eth_private, rxq[rxq->index]);
+}
+
+static struct mv643xx_eth_private *txq_to_mp(struct tx_queue *txq)
+{
+       return container_of(txq, struct mv643xx_eth_private, txq[txq->index]);
 }
 
 static void rxq_enable(struct rx_queue *rxq)
 {
        struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
-       wrl(mp, RXQ_COMMAND(mp->port_num), 1);
+       wrl(mp, RXQ_COMMAND(mp->port_num), 1 << rxq->index);
 }
 
 static void rxq_disable(struct rx_queue *rxq)
 {
        struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
-       u8 mask = 1;
+       u8 mask = 1 << rxq->index;
 
        wrl(mp, RXQ_COMMAND(mp->port_num), mask << 8);
        while (rdl(mp, RXQ_COMMAND(mp->port_num)) & mask)
                udelay(10);
 }
 
-static void mv643xx_eth_port_enable_tx(struct mv643xx_eth_private *mp,
-                                       unsigned int queues)
+static void txq_enable(struct tx_queue *txq)
 {
-       wrl(mp, TXQ_COMMAND(mp->port_num), queues);
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       wrl(mp, TXQ_COMMAND(mp->port_num), 1 << txq->index);
 }
 
-static unsigned int mv643xx_eth_port_disable_tx(struct mv643xx_eth_private *mp)
+static void txq_disable(struct tx_queue *txq)
 {
-       unsigned int port_num = mp->port_num;
-       u32 queues;
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       u8 mask = 1 << txq->index;
 
-       /* Stop Tx port activity. Check port Tx activity. */
-       queues = rdl(mp, TXQ_COMMAND(port_num)) & 0xFF;
-       if (queues) {
-               /* Issue stop command for active queues only */
-               wrl(mp, TXQ_COMMAND(port_num), (queues << 8));
+       wrl(mp, TXQ_COMMAND(mp->port_num), mask << 8);
+       while (rdl(mp, TXQ_COMMAND(mp->port_num)) & mask)
+               udelay(10);
+}
 
-               /* Wait for all Tx activity to terminate. */
-               /* Check port cause register that all Tx queues are stopped */
-               while (rdl(mp, TXQ_COMMAND(port_num)) & 0xFF)
-                       udelay(10);
+static void __txq_maybe_wake(struct tx_queue *txq)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
 
-               /* Wait for Tx FIFO to empty */
-               while (rdl(mp, PORT_STATUS(port_num)) & TX_FIFO_EMPTY)
-                       udelay(10);
-       }
+       /*
+        * netif_{stop,wake}_queue() flow control only applies to
+        * the primary queue.
+        */
+       BUG_ON(txq->index != mp->txq_primary);
 
-       return queues;
+       if (txq->tx_ring_size - txq->tx_desc_count >= MAX_DESCS_PER_SKB)
+               netif_wake_queue(mp->dev);
 }
 
 
 /* rx ***********************************************************************/
-static void mv643xx_eth_free_completed_tx_descs(struct net_device *dev);
+static void txq_reclaim(struct tx_queue *txq, int force);
 
 static void rxq_refill(struct rx_queue *rxq)
 {
@@ -456,10 +481,15 @@ static void rxq_refill(struct rx_queue *rxq)
                                                RX_ENABLE_INTERRUPT;
                wmb();
 
-               skb_reserve(skb, ETH_HW_IP_ALIGN);
+               /*
+                * 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 (rxq->rx_desc_count == 0) {
+       if (rxq->rx_desc_count != rxq->rx_ring_size) {
                rxq->rx_oom.expires = jiffies + (HZ / 10);
                add_timer(&rxq->rx_oom);
        }
@@ -480,9 +510,9 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 
        rx = 0;
        while (rx < budget) {
-               struct sk_buff *skb;
-               volatile struct rx_desc *rx_desc;
+               struct rx_desc *rx_desc;
                unsigned int cmd_sts;
+               struct sk_buff *skb;
                unsigned long flags;
 
                spin_lock_irqsave(&mp->lock, flags);
@@ -503,34 +533,40 @@ static int rxq_process(struct rx_queue *rxq, int budget)
 
                spin_unlock_irqrestore(&mp->lock, flags);
 
-               dma_unmap_single(NULL, rx_desc->buf_ptr + ETH_HW_IP_ALIGN,
-                                       mp->dev->mtu + 24, DMA_FROM_DEVICE);
+               dma_unmap_single(NULL, rx_desc->buf_ptr + 2,
+                                mp->dev->mtu + 24, DMA_FROM_DEVICE);
                rxq->rx_desc_count--;
                rx++;
 
                /*
                 * Update statistics.
-                * Note byte count includes 4 byte CRC count
+                *
+                * Note that the descriptor byte count includes 2 dummy
+                * bytes automatically inserted by the hardware at the
+                * start of the packet (which we don't count), and a 4
+                * byte CRC at the end of the packet (which we do count).
                 */
                stats->rx_packets++;
-               stats->rx_bytes += rx_desc->byte_cnt - ETH_HW_IP_ALIGN;
+               stats->rx_bytes += rx_desc->byte_cnt - 2;
 
                /*
-                * In case received a packet without first / last bits on OR
-                * the error summary bit is on, the packets needs to be dropeed.
+                * In case we received a packet without first / last bits
+                * on, or the error summary bit is set, the packet needs
+                * to be dropped.
                 */
                if (((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
                                        (RX_FIRST_DESC | RX_LAST_DESC))
                                || (cmd_sts & ERROR_SUMMARY)) {
                        stats->rx_dropped++;
+
                        if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
                                (RX_FIRST_DESC | RX_LAST_DESC)) {
                                if (net_ratelimit())
-                                       printk(KERN_ERR
-                                               "%s: Received packet spread "
-                                               "on multiple descriptors\n",
-                                               mp->dev->name);
+                                       dev_printk(KERN_ERR, &mp->dev->dev,
+                                                  "received packet spanning "
+                                                  "multiple descriptors\n");
                        }
+
                        if (cmd_sts & ERROR_SUMMARY)
                                stats->rx_errors++;
 
@@ -540,7 +576,7 @@ 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 - ETH_HW_IP_ALIGN - 4);
+                       skb_put(skb, rx_desc->byte_cnt - 2 - 4);
 
                        if (cmd_sts & LAYER_4_CHECKSUM_OK) {
                                skb->ip_summed = CHECKSUM_UNNECESSARY;
@@ -554,8 +590,10 @@ static int rxq_process(struct rx_queue *rxq, int budget)
                        netif_rx(skb);
 #endif
                }
+
                mp->dev->last_rx = jiffies;
        }
+
        rxq_refill(rxq);
 
        return rx;
@@ -566,23 +604,29 @@ static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
 {
        struct mv643xx_eth_private *mp;
        int rx;
+       int i;
 
        mp = container_of(napi, struct mv643xx_eth_private, napi);
 
 #ifdef MV643XX_ETH_TX_FAST_REFILL
        if (++mp->tx_clean_threshold > 5) {
-               mv643xx_eth_free_completed_tx_descs(mp->dev);
                mp->tx_clean_threshold = 0;
+               for (i = 0; i < 8; i++)
+                       if (mp->txq_mask & (1 << i))
+                               txq_reclaim(mp->txq + i, 0);
        }
 #endif
 
-       rx = rxq_process(mp->rxq, budget);
+       rx = 0;
+       for (i = 7; rx < budget && i >= 0; i--)
+               if (mp->rxq_mask & (1 << i))
+                       rx += rxq_process(mp->rxq + i, budget - rx);
 
        if (rx < budget) {
                netif_rx_complete(mp->dev, napi);
                wrl(mp, INT_CAUSE(mp->port_num), 0);
                wrl(mp, INT_CAUSE_EXT(mp->port_num), 0);
-               wrl(mp, INT_MASK(mp->port_num), INT_RX | INT_EXT);
+               wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
        }
 
        return rx;
@@ -593,55 +637,59 @@ static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
 /* tx ***********************************************************************/
 static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
 {
-       unsigned int frag;
-       skb_frag_t *fragp;
+       int frag;
 
        for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
-               fragp = &skb_shinfo(skb)->frags[frag];
-               if (fragp->size <= 8 && fragp->page_offset & 0x7)
+               skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
+               if (fragp->size <= 8 && fragp->page_offset & 7)
                        return 1;
        }
+
        return 0;
 }
 
-static int alloc_tx_desc_index(struct mv643xx_eth_private *mp)
+static int txq_alloc_desc_index(struct tx_queue *txq)
 {
        int tx_desc_curr;
 
-       BUG_ON(mp->tx_desc_count >= mp->tx_ring_size);
+       BUG_ON(txq->tx_desc_count >= txq->tx_ring_size);
 
-       tx_desc_curr = mp->tx_curr_desc;
-       mp->tx_curr_desc = (tx_desc_curr + 1) % mp->tx_ring_size;
+       tx_desc_curr = txq->tx_curr_desc;
+       txq->tx_curr_desc = (tx_desc_curr + 1) % txq->tx_ring_size;
 
-       BUG_ON(mp->tx_curr_desc == mp->tx_used_desc);
+       BUG_ON(txq->tx_curr_desc == txq->tx_used_desc);
 
        return tx_desc_curr;
 }
 
-static void tx_fill_frag_descs(struct mv643xx_eth_private *mp,
-                                  struct sk_buff *skb)
+static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
+       int nr_frags = skb_shinfo(skb)->nr_frags;
        int frag;
-       int tx_index;
-       struct tx_desc *desc;
 
-       for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
-               skb_frag_t *this_frag = &skb_shinfo(skb)->frags[frag];
-
-               tx_index = alloc_tx_desc_index(mp);
-               desc = &mp->tx_desc_area[tx_index];
-
-               desc->cmd_sts = BUFFER_OWNED_BY_DMA;
-               /* Last Frag enables interrupt and frees the skb */
-               if (frag == (skb_shinfo(skb)->nr_frags - 1)) {
-                       desc->cmd_sts |= ZERO_PADDING |
-                                        TX_LAST_DESC |
-                                        TX_ENABLE_INTERRUPT;
-                       mp->tx_skb[tx_index] = skb;
-               } else
-                       mp->tx_skb[tx_index] = NULL;
-
-               desc = &mp->tx_desc_area[tx_index];
+       for (frag = 0; frag < nr_frags; frag++) {
+               skb_frag_t *this_frag;
+               int tx_index;
+               struct tx_desc *desc;
+
+               this_frag = &skb_shinfo(skb)->frags[frag];
+               tx_index = txq_alloc_desc_index(txq);
+               desc = &txq->tx_desc_area[tx_index];
+
+               /*
+                * The last fragment will generate an interrupt
+                * which will free the skb on TX completion.
+                */
+               if (frag == nr_frags - 1) {
+                       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;
                desc->byte_cnt = this_frag->size;
                desc->buf_ptr = dma_map_page(NULL, this_frag->page,
@@ -656,29 +704,28 @@ static inline __be16 sum16_as_be(__sum16 sum)
        return (__force __be16)sum;
 }
 
-static void tx_submit_descs_for_skb(struct mv643xx_eth_private *mp,
-                                       struct sk_buff *skb)
+static void txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
+       int nr_frags = skb_shinfo(skb)->nr_frags;
        int tx_index;
        struct tx_desc *desc;
        u32 cmd_sts;
        int length;
-       int nr_frags = skb_shinfo(skb)->nr_frags;
 
        cmd_sts = TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
 
-       tx_index = alloc_tx_desc_index(mp);
-       desc = &mp->tx_desc_area[tx_index];
+       tx_index = txq_alloc_desc_index(txq);
+       desc = &txq->tx_desc_area[tx_index];
 
        if (nr_frags) {
-               tx_fill_frag_descs(mp, skb);
+               txq_submit_frag_skb(txq, skb);
 
                length = skb_headlen(skb);
-               mp->tx_skb[tx_index] = NULL;
+               txq->tx_skb[tx_index] = NULL;
        } else {
                cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
                length = skb->len;
-               mp->tx_skb[tx_index] = skb;
+               txq->tx_skb[tx_index] = skb;
        }
 
        desc->byte_cnt = length;
@@ -714,42 +761,51 @@ static void tx_submit_descs_for_skb(struct mv643xx_eth_private *mp,
 
        /* ensure all descriptors are written before poking hardware */
        wmb();
-       mv643xx_eth_port_enable_tx(mp, 1);
+       txq_enable(txq);
 
-       mp->tx_desc_count += nr_frags + 1;
+       txq->tx_desc_count += nr_frags + 1;
 }
 
-static int mv643xx_eth_start_xmit(struct sk_buff *skb, struct net_device *dev)
+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;
+       struct tx_queue *txq;
        unsigned long flags;
 
-       BUG_ON(netif_queue_stopped(dev));
-
        if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
                stats->tx_dropped++;
-               printk(KERN_DEBUG "%s: failed to linearize tiny "
-                               "unaligned fragment\n", dev->name);
+               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);
 
-       if (mp->tx_ring_size - mp->tx_desc_count < MAX_DESCS_PER_SKB) {
-               printk(KERN_ERR "%s: transmit with queue full\n", dev->name);
-               netif_stop_queue(dev);
+       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);
-               return NETDEV_TX_BUSY;
+               if (txq->index == mp->txq_primary && net_ratelimit())
+                       dev_printk(KERN_ERR, &dev->dev,
+                                  "primary tx queue full?!\n");
+               kfree_skb(skb);
+               return NETDEV_TX_OK;
        }
 
-       tx_submit_descs_for_skb(mp, skb);
+       txq_submit_skb(txq, skb);
        stats->tx_bytes += skb->len;
        stats->tx_packets++;
        dev->trans_start = jiffies;
 
-       if (mp->tx_ring_size - mp->tx_desc_count < MAX_DESCS_PER_SKB)
-               netif_stop_queue(dev);
+       if (txq->index == mp->txq_primary) {
+               int entries_left;
+
+               entries_left = txq->tx_ring_size - txq->tx_desc_count;
+               if (entries_left < MAX_DESCS_PER_SKB)
+                       netif_stop_queue(dev);
+       }
 
        spin_unlock_irqrestore(&mp->lock, flags);
 
@@ -757,14 +813,117 @@ static int mv643xx_eth_start_xmit(struct sk_buff *skb, struct net_device *dev)
 }
 
 
+/* tx rate control **********************************************************/
+/*
+ * Set total maximum TX rate (shared by all TX queues for this port)
+ * to 'rate' bits per second, with a maximum burst of 'burst' bytes.
+ */
+static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
+{
+       int token_rate;
+       int mtu;
+       int bucket_size;
+
+       token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
+       if (token_rate > 1023)
+               token_rate = 1023;
+
+       mtu = (mp->dev->mtu + 255) >> 8;
+       if (mtu > 63)
+               mtu = 63;
+
+       bucket_size = (burst + 255) >> 8;
+       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 {
+               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);
+       }
+}
+
+static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int token_rate;
+       int bucket_size;
+
+       token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
+       if (token_rate > 1023)
+               token_rate = 1023;
+
+       bucket_size = (burst + 255) >> 8;
+       if (bucket_size > 65535)
+               bucket_size = 65535;
+
+       wrl(mp, TXQ_BW_TOKENS(mp->port_num, txq->index), token_rate << 14);
+       wrl(mp, TXQ_BW_CONF(mp->port_num, txq->index),
+                       (bucket_size << 10) | token_rate);
+}
+
+static void txq_set_fixed_prio_mode(struct tx_queue *txq)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int off;
+       u32 val;
+
+       /*
+        * Turn on fixed priority mode.
+        */
+       if (mp->shared->tx_bw_control_moved)
+               off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+       else
+               off = TXQ_FIX_PRIO_CONF(mp->port_num);
+
+       val = rdl(mp, off);
+       val |= 1 << txq->index;
+       wrl(mp, off, val);
+}
+
+static void txq_set_wrr(struct tx_queue *txq, int weight)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int off;
+       u32 val;
+
+       /*
+        * Turn off fixed priority mode.
+        */
+       if (mp->shared->tx_bw_control_moved)
+               off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+       else
+               off = TXQ_FIX_PRIO_CONF(mp->port_num);
+
+       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);
+
+       val = rdl(mp, off);
+       val = (val & ~0xff) | (weight & 0xff);
+       wrl(mp, off, val);
+}
+
+
 /* mii management interface *************************************************/
-static int phy_addr_get(struct mv643xx_eth_private *mp);
+#define SMI_BUSY               0x10000000
+#define SMI_READ_VALID         0x08000000
+#define SMI_OPCODE_READ                0x04000000
+#define SMI_OPCODE_WRITE       0x00000000
 
-static void read_smi_reg(struct mv643xx_eth_private *mp,
-                               unsigned int phy_reg, unsigned int *value)
+static void smi_reg_read(struct mv643xx_eth_private *mp, unsigned int addr,
+                        unsigned int reg, unsigned int *value)
 {
        void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
-       int phy_addr = phy_addr_get(mp);
        unsigned long flags;
        int i;
 
@@ -780,7 +939,7 @@ static void read_smi_reg(struct mv643xx_eth_private *mp,
                udelay(10);
        }
 
-       writel((phy_addr << 16) | (phy_reg << 21) | SMI_OPCODE_READ, smi_reg);
+       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++) {
@@ -796,11 +955,11 @@ out:
        spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
 }
 
-static void write_smi_reg(struct mv643xx_eth_private *mp,
-                                  unsigned int phy_reg, unsigned int value)
+static void smi_reg_write(struct mv643xx_eth_private *mp,
+                         unsigned int addr,
+                         unsigned int reg, unsigned int value)
 {
        void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
-       int phy_addr = phy_addr_get(mp);
        unsigned long flags;
        int i;
 
@@ -816,65 +975,63 @@ static void write_smi_reg(struct mv643xx_eth_private *mp,
                udelay(10);
        }
 
-       writel((phy_addr << 16) | (phy_reg << 21) |
-               SMI_OPCODE_WRITE | (value & 0xffff), smi_reg);
+       writel(SMI_OPCODE_WRITE | (reg << 21) |
+               (addr << 16) | (value & 0xffff), smi_reg);
 out:
        spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
 }
 
 
 /* mib counters *************************************************************/
-static void clear_mib_counters(struct mv643xx_eth_private *mp)
+static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
 {
-       unsigned int port_num = mp->port_num;
-       int i;
-
-       /* Perform dummy reads from MIB counters */
-       for (i = 0; i < 0x80; i += 4)
-               rdl(mp, MIB_COUNTERS(port_num) + i);
+       return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
 }
 
-static inline u32 read_mib(struct mv643xx_eth_private *mp, int offset)
+static void mib_counters_clear(struct mv643xx_eth_private *mp)
 {
-       return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
+       int i;
+
+       for (i = 0; i < 0x80; i += 4)
+               mib_read(mp, i);
 }
 
-static void update_mib_counters(struct mv643xx_eth_private *mp)
+static void mib_counters_update(struct mv643xx_eth_private *mp)
 {
        struct mib_counters *p = &mp->mib_counters;
 
-       p->good_octets_received += read_mib(mp, 0x00);
-       p->good_octets_received += (u64)read_mib(mp, 0x04) << 32;
-       p->bad_octets_received += read_mib(mp, 0x08);
-       p->internal_mac_transmit_err += read_mib(mp, 0x0c);
-       p->good_frames_received += read_mib(mp, 0x10);
-       p->bad_frames_received += read_mib(mp, 0x14);
-       p->broadcast_frames_received += read_mib(mp, 0x18);
-       p->multicast_frames_received += read_mib(mp, 0x1c);
-       p->frames_64_octets += read_mib(mp, 0x20);
-       p->frames_65_to_127_octets += read_mib(mp, 0x24);
-       p->frames_128_to_255_octets += read_mib(mp, 0x28);
-       p->frames_256_to_511_octets += read_mib(mp, 0x2c);
-       p->frames_512_to_1023_octets += read_mib(mp, 0x30);
-       p->frames_1024_to_max_octets += read_mib(mp, 0x34);
-       p->good_octets_sent += read_mib(mp, 0x38);
-       p->good_octets_sent += (u64)read_mib(mp, 0x3c) << 32;
-       p->good_frames_sent += read_mib(mp, 0x40);
-       p->excessive_collision += read_mib(mp, 0x44);
-       p->multicast_frames_sent += read_mib(mp, 0x48);
-       p->broadcast_frames_sent += read_mib(mp, 0x4c);
-       p->unrec_mac_control_received += read_mib(mp, 0x50);
-       p->fc_sent += read_mib(mp, 0x54);
-       p->good_fc_received += read_mib(mp, 0x58);
-       p->bad_fc_received += read_mib(mp, 0x5c);
-       p->undersize_received += read_mib(mp, 0x60);
-       p->fragments_received += read_mib(mp, 0x64);
-       p->oversize_received += read_mib(mp, 0x68);
-       p->jabber_received += read_mib(mp, 0x6c);
-       p->mac_receive_error += read_mib(mp, 0x70);
-       p->bad_crc_event += read_mib(mp, 0x74);
-       p->collision += read_mib(mp, 0x78);
-       p->late_collision += read_mib(mp, 0x7c);
+       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);
+       p->internal_mac_transmit_err += mib_read(mp, 0x0c);
+       p->good_frames_received += mib_read(mp, 0x10);
+       p->bad_frames_received += mib_read(mp, 0x14);
+       p->broadcast_frames_received += mib_read(mp, 0x18);
+       p->multicast_frames_received += mib_read(mp, 0x1c);
+       p->frames_64_octets += mib_read(mp, 0x20);
+       p->frames_65_to_127_octets += mib_read(mp, 0x24);
+       p->frames_128_to_255_octets += mib_read(mp, 0x28);
+       p->frames_256_to_511_octets += mib_read(mp, 0x2c);
+       p->frames_512_to_1023_octets += mib_read(mp, 0x30);
+       p->frames_1024_to_max_octets += mib_read(mp, 0x34);
+       p->good_octets_sent += mib_read(mp, 0x38);
+       p->good_octets_sent += (u64)mib_read(mp, 0x3c) << 32;
+       p->good_frames_sent += mib_read(mp, 0x40);
+       p->excessive_collision += mib_read(mp, 0x44);
+       p->multicast_frames_sent += mib_read(mp, 0x48);
+       p->broadcast_frames_sent += mib_read(mp, 0x4c);
+       p->unrec_mac_control_received += mib_read(mp, 0x50);
+       p->fc_sent += mib_read(mp, 0x54);
+       p->good_fc_received += mib_read(mp, 0x58);
+       p->bad_fc_received += mib_read(mp, 0x5c);
+       p->undersize_received += mib_read(mp, 0x60);
+       p->fragments_received += mib_read(mp, 0x64);
+       p->oversize_received += mib_read(mp, 0x68);
+       p->jabber_received += mib_read(mp, 0x6c);
+       p->mac_receive_error += mib_read(mp, 0x70);
+       p->bad_crc_event += mib_read(mp, 0x74);
+       p->collision += mib_read(mp, 0x78);
+       p->late_collision += mib_read(mp, 0x7c);
 }
 
 
@@ -944,18 +1101,41 @@ static int mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *
        err = mii_ethtool_gset(&mp->mii, cmd);
        spin_unlock_irq(&mp->lock);
 
-       /* The PHY may support 1000baseT_Half, but the mv643xx does not */
+       /*
+        * The MAC does not support 1000baseT_Half.
+        */
        cmd->supported &= ~SUPPORTED_1000baseT_Half;
        cmd->advertising &= ~ADVERTISED_1000baseT_Half;
 
        return err;
 }
 
+static int mv643xx_eth_get_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       cmd->supported = SUPPORTED_MII;
+       cmd->advertising = ADVERTISED_MII;
+       cmd->speed = SPEED_1000;
+       cmd->duplex = DUPLEX_FULL;
+       cmd->port = PORT_MII;
+       cmd->phy_address = 0;
+       cmd->transceiver = XCVR_INTERNAL;
+       cmd->autoneg = AUTONEG_DISABLE;
+       cmd->maxtxpkt = 1;
+       cmd->maxrxpkt = 1;
+
+       return 0;
+}
+
 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);
@@ -963,23 +1143,33 @@ static int mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *
        return err;
 }
 
-static void mv643xx_eth_get_drvinfo(struct net_device *netdev,
-                               struct ethtool_drvinfo *drvinfo)
+static int mv643xx_eth_set_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       return -EINVAL;
+}
+
+static void mv643xx_eth_get_drvinfo(struct net_device *dev,
+                                   struct ethtool_drvinfo *drvinfo)
 {
        strncpy(drvinfo->driver,  mv643xx_eth_driver_name, 32);
        strncpy(drvinfo->version, mv643xx_eth_driver_version, 32);
        strncpy(drvinfo->fw_version, "N/A", 32);
-       strncpy(drvinfo->bus_info, "mv643xx", 32);
+       strncpy(drvinfo->bus_info, "platform", 32);
        drvinfo->n_stats = ARRAY_SIZE(mv643xx_eth_stats);
 }
 
-static int mv643xx_eth_nway_restart(struct net_device *dev)
+static int mv643xx_eth_nway_reset(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
        return mii_nway_restart(&mp->mii);
 }
 
+static int mv643xx_eth_nway_reset_phyless(struct net_device *dev)
+{
+       return -EINVAL;
+}
+
 static u32 mv643xx_eth_get_link(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
@@ -987,29 +1177,33 @@ static u32 mv643xx_eth_get_link(struct net_device *dev)
        return mii_link_ok(&mp->mii);
 }
 
-static void mv643xx_eth_get_strings(struct net_device *netdev, uint32_t stringset,
-                               uint8_t *data)
+static u32 mv643xx_eth_get_link_phyless(struct net_device *dev)
+{
+       return 1;
+}
+
+static void mv643xx_eth_get_strings(struct net_device *dev,
+                                   uint32_t stringset, uint8_t *data)
 {
        int i;
 
-       switch(stringset) {
-       case ETH_SS_STATS:
-               for (i=0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
+       if (stringset == ETH_SS_STATS) {
+               for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
                        memcpy(data + i * ETH_GSTRING_LEN,
                                mv643xx_eth_stats[i].stat_string,
                                ETH_GSTRING_LEN);
                }
-               break;
        }
 }
 
-static void mv643xx_eth_get_ethtool_stats(struct net_device *netdev,
-                               struct ethtool_stats *stats, uint64_t *data)
+static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
+                                         struct ethtool_stats *stats,
+                                         uint64_t *data)
 {
-       struct mv643xx_eth_private *mp = netdev->priv;
+       struct mv643xx_eth_private *mp = dev->priv;
        int i;
 
-       update_mib_counters(mp);
+       mib_counters_update(mp);
 
        for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
                const struct mv643xx_eth_stats *stat;
@@ -1027,38 +1221,47 @@ static void mv643xx_eth_get_ethtool_stats(struct net_device *netdev,
        }
 }
 
-static int mv643xx_eth_get_sset_count(struct net_device *netdev, int sset)
+static int mv643xx_eth_get_sset_count(struct net_device *dev, int sset)
 {
-       switch (sset) {
-       case ETH_SS_STATS:
+       if (sset == ETH_SS_STATS)
                return ARRAY_SIZE(mv643xx_eth_stats);
-       default:
-               return -EOPNOTSUPP;
-       }
+
+       return -EOPNOTSUPP;
 }
 
 static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
-       .get_settings           = mv643xx_eth_get_settings,
-       .set_settings           = mv643xx_eth_set_settings,
-       .get_drvinfo            = mv643xx_eth_get_drvinfo,
-       .get_link               = mv643xx_eth_get_link,
+       .get_settings           = mv643xx_eth_get_settings,
+       .set_settings           = mv643xx_eth_set_settings,
+       .get_drvinfo            = mv643xx_eth_get_drvinfo,
+       .nway_reset             = mv643xx_eth_nway_reset,
+       .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,
+       .get_sset_count         = mv643xx_eth_get_sset_count,
+};
+
+static const struct ethtool_ops mv643xx_eth_ethtool_ops_phyless = {
+       .get_settings           = mv643xx_eth_get_settings_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,
        .set_sg                 = ethtool_op_set_sg,
+       .get_strings            = mv643xx_eth_get_strings,
+       .get_ethtool_stats      = mv643xx_eth_get_ethtool_stats,
        .get_sset_count         = mv643xx_eth_get_sset_count,
-       .get_ethtool_stats      = mv643xx_eth_get_ethtool_stats,
-       .get_strings            = mv643xx_eth_get_strings,
-       .nway_reset             = mv643xx_eth_nway_restart,
 };
 
 
 /* address handling *********************************************************/
 static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       unsigned int port_num = mp->port_num;
        unsigned int mac_h;
        unsigned int mac_l;
 
-       mac_h = rdl(mp, MAC_ADDR_HIGH(port_num));
-       mac_l = rdl(mp, MAC_ADDR_LOW(port_num));
+       mac_h = rdl(mp, MAC_ADDR_HIGH(mp->port_num));
+       mac_l = rdl(mp, MAC_ADDR_LOW(mp->port_num));
 
        addr[0] = (mac_h >> 24) & 0xff;
        addr[1] = (mac_h >> 16) & 0xff;
@@ -1070,72 +1273,54 @@ static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
 
 static void init_mac_tables(struct mv643xx_eth_private *mp)
 {
-       unsigned int port_num = mp->port_num;
-       int table_index;
-
-       /* Clear DA filter unicast table (Ex_dFUT) */
-       for (table_index = 0; table_index <= 0xC; table_index += 4)
-               wrl(mp, UNICAST_TABLE(port_num) + table_index, 0);
+       int i;
 
-       for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-               /* Clear DA filter special multicast table (Ex_dFSMT) */
-               wrl(mp, SPECIAL_MCAST_TABLE(port_num) + table_index, 0);
-               /* Clear DA filter other multicast table (Ex_dFOMT) */
-               wrl(mp, OTHER_MCAST_TABLE(port_num) + table_index, 0);
+       for (i = 0; i < 0x100; i += 4) {
+               wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, 0);
+               wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, 0);
        }
+
+       for (i = 0; i < 0x10; i += 4)
+               wrl(mp, UNICAST_TABLE(mp->port_num) + i, 0);
 }
 
 static void set_filter_table_entry(struct mv643xx_eth_private *mp,
-                                           int table, unsigned char entry)
+                                  int table, unsigned char entry)
 {
        unsigned int table_reg;
-       unsigned int tbl_offset;
-       unsigned int reg_offset;
-
-       tbl_offset = (entry / 4) * 4;   /* Register offset of DA table entry */
-       reg_offset = entry % 4;         /* Entry offset within the register */
 
        /* Set "accepts frame bit" at specified table entry */
-       table_reg = rdl(mp, table + tbl_offset);
-       table_reg |= 0x01 << (8 * reg_offset);
-       wrl(mp, table + tbl_offset, table_reg);
+       table_reg = rdl(mp, table + (entry & 0xfc));
+       table_reg |= 0x01 << (8 * (entry & 3));
+       wrl(mp, table + (entry & 0xfc), table_reg);
 }
 
 static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       unsigned int port_num = mp->port_num;
        unsigned int mac_h;
        unsigned int mac_l;
        int table;
 
-       mac_l = (addr[4] << 8) | (addr[5]);
-       mac_h = (addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) |
-                                                       (addr[3] << 0);
+       mac_l = (addr[4] << 8) | addr[5];
+       mac_h = (addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3];
 
-       wrl(mp, MAC_ADDR_LOW(port_num), mac_l);
-       wrl(mp, MAC_ADDR_HIGH(port_num), mac_h);
+       wrl(mp, MAC_ADDR_LOW(mp->port_num), mac_l);
+       wrl(mp, MAC_ADDR_HIGH(mp->port_num), mac_h);
 
-       /* Accept frames with this address */
-       table = UNICAST_TABLE(port_num);
+       table = UNICAST_TABLE(mp->port_num);
        set_filter_table_entry(mp, table, addr[5] & 0x0f);
 }
 
-static void mv643xx_eth_update_mac_address(struct net_device *dev)
+static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
+       /* +2 is for the offset of the HW addr type */
+       memcpy(dev->dev_addr, addr + 2, 6);
+
        init_mac_tables(mp);
        uc_addr_set(mp, dev->dev_addr);
-}
-
-static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
-{
-       int i;
 
-       for (i = 0; i < 6; i++)
-               /* +2 is for the offset of the HW addr type */
-               dev->dev_addr[i] = ((unsigned char *)addr)[i + 2];
-       mv643xx_eth_update_mac_address(dev);
        return 0;
 }
 
@@ -1157,106 +1342,66 @@ static int addr_crc(unsigned char *addr)
        return crc;
 }
 
-static void mc_addr(struct mv643xx_eth_private *mp, unsigned char *addr)
+static void mv643xx_eth_set_rx_mode(struct net_device *dev)
 {
-       unsigned int port_num = mp->port_num;
-       int table;
-       int crc;
-
-       if ((addr[0] == 0x01) && (addr[1] == 0x00) &&
-           (addr[2] == 0x5E) && (addr[3] == 0x00) && (addr[4] == 0x00)) {
-               table = SPECIAL_MCAST_TABLE(port_num);
-               set_filter_table_entry(mp, table, addr[5]);
-               return;
-       }
-
-       crc = addr_crc(addr);
-
-       table = OTHER_MCAST_TABLE(port_num);
-       set_filter_table_entry(mp, table, crc);
-}
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       u32 port_config;
+       struct dev_addr_list *addr;
+       int i;
 
-static void set_multicast_list(struct net_device *dev)
-{
+       port_config = rdl(mp, PORT_CONFIG(mp->port_num));
+       if (dev->flags & IFF_PROMISC)
+               port_config |= UNICAST_PROMISCUOUS_MODE;
+       else
+               port_config &= ~UNICAST_PROMISCUOUS_MODE;
+       wrl(mp, PORT_CONFIG(mp->port_num), port_config);
 
-       struct dev_mc_list      *mc_list;
-       int                     i;
-       int                     table_index;
-       struct mv643xx_eth_private      *mp = netdev_priv(dev);
-       unsigned int            port_num = mp->port_num;
+       if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
+               int port_num = mp->port_num;
+               u32 accept = 0x01010101;
 
-       /* If the device is in promiscuous mode or in all multicast mode,
-        * we will fully populate both multicast tables with accept.
-        * This is guaranteed to yield a match on all multicast addresses...
-        */
-       if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI)) {
-               for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-                       /* Set all entries in DA filter special multicast
-                        * table (Ex_dFSMT)
-                        * Set for ETH_Q0 for now
-                        * Bits
-                        * 0      Accept=1, Drop=0
-                        * 3-1  Queue    ETH_Q0=0
-                        * 7-4  Reserved = 0;
-                        */
-                       wrl(mp, SPECIAL_MCAST_TABLE(port_num) + table_index, 0x01010101);
-
-                       /* Set all entries in DA filter other multicast
-                        * table (Ex_dFOMT)
-                        * Set for ETH_Q0 for now
-                        * Bits
-                        * 0      Accept=1, Drop=0
-                        * 3-1  Queue    ETH_Q0=0
-                        * 7-4  Reserved = 0;
-                        */
-                       wrl(mp, OTHER_MCAST_TABLE(port_num) + table_index, 0x01010101);
+               for (i = 0; i < 0x100; i += 4) {
+                       wrl(mp, SPECIAL_MCAST_TABLE(port_num) + i, accept);
+                       wrl(mp, OTHER_MCAST_TABLE(port_num) + i, accept);
                }
                return;
        }
 
-       /* We will clear out multicast tables every time we get the list.
-        * Then add the entire new list...
-        */
-       for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-               /* Clear DA filter special multicast table (Ex_dFSMT) */
-               wrl(mp, SPECIAL_MCAST_TABLE(port_num) + table_index, 0);
-
-               /* Clear DA filter other multicast table (Ex_dFOMT) */
-               wrl(mp, OTHER_MCAST_TABLE(port_num) + table_index, 0);
+       for (i = 0; i < 0x100; i += 4) {
+               wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, 0);
+               wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, 0);
        }
 
-       /* Get pointer to net_device multicast list and add each one... */
-       for (i = 0, mc_list = dev->mc_list;
-                       (i < 256) && (mc_list != NULL) && (i < dev->mc_count);
-                       i++, mc_list = mc_list->next)
-               if (mc_list->dmi_addrlen == 6)
-                       mc_addr(mp, mc_list->dmi_addr);
-}
+       for (addr = dev->mc_list; addr != NULL; addr = addr->next) {
+               u8 *a = addr->da_addr;
+               int table;
 
-static void mv643xx_eth_set_rx_mode(struct net_device *dev)
-{
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       u32 config_reg;
+               if (addr->da_addrlen != 6)
+                       continue;
 
-       config_reg = rdl(mp, PORT_CONFIG(mp->port_num));
-       if (dev->flags & IFF_PROMISC)
-               config_reg |= UNICAST_PROMISCUOUS_MODE;
-       else
-               config_reg &= ~UNICAST_PROMISCUOUS_MODE;
-       wrl(mp, PORT_CONFIG(mp->port_num), config_reg);
+               if (memcmp(a, "\x01\x00\x5e\x00\x00", 5) == 0) {
+                       table = SPECIAL_MCAST_TABLE(mp->port_num);
+                       set_filter_table_entry(mp, table, a[5]);
+               } else {
+                       int crc = addr_crc(a);
 
-       set_multicast_list(dev);
+                       table = OTHER_MCAST_TABLE(mp->port_num);
+                       set_filter_table_entry(mp, table, crc);
+               }
+       }
 }
 
 
 /* rx/tx queue initialisation ***********************************************/
-static int rxq_init(struct mv643xx_eth_private *mp)
+static int rxq_init(struct mv643xx_eth_private *mp, int index)
 {
-       struct rx_queue *rxq = mp->rxq;
+       struct rx_queue *rxq = mp->rxq + index;
        struct rx_desc *rx_desc;
        int size;
        int i;
 
+       rxq->index = index;
+
        rxq->rx_ring_size = mp->default_rx_ring_size;
 
        rxq->rx_desc_count = 0;
@@ -1265,7 +1410,7 @@ static int rxq_init(struct mv643xx_eth_private *mp)
 
        size = rxq->rx_ring_size * sizeof(struct rx_desc);
 
-       if (size <= mp->rx_desc_sram_size) {
+       if (index == mp->rxq_primary && 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;
@@ -1306,7 +1451,7 @@ static int rxq_init(struct mv643xx_eth_private *mp)
 
 
 out_free:
-       if (size <= mp->rx_desc_sram_size)
+       if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size)
                iounmap(rxq->rx_desc_area);
        else
                dma_free_coherent(NULL, size,
@@ -1339,7 +1484,8 @@ static void rxq_deinit(struct rx_queue *rxq)
                           rxq->rx_desc_count);
        }
 
-       if (rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
+       if (rxq->index == mp->rxq_primary &&
+           rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
                iounmap(rxq->rx_desc_area);
        else
                dma_free_coherent(NULL, rxq->rx_desc_area_size,
@@ -1348,69 +1494,108 @@ static void rxq_deinit(struct rx_queue *rxq)
        kfree(rxq->rx_skb);
 }
 
-static void ether_init_tx_desc_ring(struct mv643xx_eth_private *mp)
+static int txq_init(struct mv643xx_eth_private *mp, int index)
 {
-       int tx_desc_num = mp->tx_ring_size;
-       struct tx_desc *p_tx_desc;
+       struct tx_queue *txq = mp->txq + index;
+       struct tx_desc *tx_desc;
+       int size;
        int i;
 
-       /* Initialize the next_desc_ptr links in the Tx descriptors ring */
-       p_tx_desc = (struct tx_desc *)mp->tx_desc_area;
-       for (i = 0; i < tx_desc_num; i++) {
-               p_tx_desc[i].next_desc_ptr = mp->tx_desc_dma +
-                       ((i + 1) % tx_desc_num) * sizeof(struct tx_desc);
+       txq->index = index;
+
+       txq->tx_ring_size = mp->default_tx_ring_size;
+
+       txq->tx_desc_count = 0;
+       txq->tx_curr_desc = 0;
+       txq->tx_used_desc = 0;
+
+       size = txq->tx_ring_size * sizeof(struct tx_desc);
+
+       if (index == mp->txq_primary && size <= mp->tx_desc_sram_size) {
+               txq->tx_desc_area = ioremap(mp->tx_desc_sram_addr,
+                                               mp->tx_desc_sram_size);
+               txq->tx_desc_dma = mp->tx_desc_sram_addr;
+       } else {
+               txq->tx_desc_area = dma_alloc_coherent(NULL, size,
+                                                       &txq->tx_desc_dma,
+                                                       GFP_KERNEL);
+       }
+
+       if (txq->tx_desc_area == NULL) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "can't allocate tx ring (%d bytes)\n", size);
+               goto out;
+       }
+       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++) {
+               int nexti = (i + 1) % txq->tx_ring_size;
+               tx_desc[i].next_desc_ptr = txq->tx_desc_dma +
+                                       nexti * sizeof(struct tx_desc);
        }
 
-       mp->tx_curr_desc = 0;
-       mp->tx_used_desc = 0;
+       return 0;
+
+
+out_free:
+       if (index == mp->txq_primary && 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);
 
-       mp->tx_desc_area_size = tx_desc_num * sizeof(struct tx_desc);
+out:
+       return -ENOMEM;
 }
 
-static int mv643xx_eth_free_tx_descs(struct net_device *dev, int force)
+static void txq_reclaim(struct tx_queue *txq, int force)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       struct tx_desc *desc;
-       u32 cmd_sts;
-       struct sk_buff *skb;
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
        unsigned long flags;
-       int tx_index;
-       dma_addr_t addr;
-       int count;
-       int released = 0;
-
-       while (mp->tx_desc_count > 0) {
-               spin_lock_irqsave(&mp->lock, flags);
 
-               /* tx_desc_count might have changed before acquiring the lock */
-               if (mp->tx_desc_count <= 0) {
-                       spin_unlock_irqrestore(&mp->lock, flags);
-                       return released;
-               }
+       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 = mp->tx_used_desc;
-               desc = &mp->tx_desc_area[tx_index];
+               tx_index = txq->tx_used_desc;
+               desc = &txq->tx_desc_area[tx_index];
                cmd_sts = desc->cmd_sts;
 
-               if (!force && (cmd_sts & BUFFER_OWNED_BY_DMA)) {
-                       spin_unlock_irqrestore(&mp->lock, flags);
-                       return released;
-               }
+               if (!force && (cmd_sts & BUFFER_OWNED_BY_DMA))
+                       break;
 
-               mp->tx_used_desc = (tx_index + 1) % mp->tx_ring_size;
-               mp->tx_desc_count--;
+               txq->tx_used_desc = (tx_index + 1) % txq->tx_ring_size;
+               txq->tx_desc_count--;
 
                addr = desc->buf_ptr;
                count = desc->byte_cnt;
-               skb = mp->tx_skb[tx_index];
-               if (skb)
-                       mp->tx_skb[tx_index] = NULL;
+               skb = txq->tx_skb[tx_index];
+               txq->tx_skb[tx_index] = NULL;
 
                if (cmd_sts & ERROR_SUMMARY) {
-                       printk("%s: Error in TX\n", dev->name);
-                       dev->stats.tx_errors++;
+                       dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
+                       mp->dev->stats.tx_errors++;
                }
 
+               /*
+                * Drop mp->lock while we free the skb.
+                */
                spin_unlock_irqrestore(&mp->lock, flags);
 
                if (cmd_sts & TX_FIRST_DESC)
@@ -1421,123 +1606,116 @@ static int mv643xx_eth_free_tx_descs(struct net_device *dev, int force)
                if (skb)
                        dev_kfree_skb_irq(skb);
 
-               released = 1;
+               spin_lock_irqsave(&mp->lock, flags);
        }
-
-       return released;
-}
-
-static void mv643xx_eth_free_completed_tx_descs(struct net_device *dev)
-{
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-
-       if (mv643xx_eth_free_tx_descs(dev, 0) &&
-           mp->tx_ring_size - mp->tx_desc_count >= MAX_DESCS_PER_SKB)
-               netif_wake_queue(dev);
-}
-
-static void mv643xx_eth_free_all_tx_descs(struct net_device *dev)
-{
-       mv643xx_eth_free_tx_descs(dev, 1);
+       spin_unlock_irqrestore(&mp->lock, flags);
 }
 
-static void mv643xx_eth_free_tx_rings(struct net_device *dev)
+static void txq_deinit(struct tx_queue *txq)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-
-       /* Stop Tx Queues */
-       mv643xx_eth_port_disable_tx(mp);
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
 
-       /* Free outstanding skb's on TX ring */
-       mv643xx_eth_free_all_tx_descs(dev);
+       txq_disable(txq);
+       txq_reclaim(txq, 1);
 
-       BUG_ON(mp->tx_used_desc != mp->tx_curr_desc);
+       BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
 
-       /* Free TX ring */
-       if (mp->tx_sram_size)
-               iounmap(mp->tx_desc_area);
+       if (txq->index == mp->txq_primary &&
+           txq->tx_desc_area_size <= mp->tx_desc_sram_size)
+               iounmap(txq->tx_desc_area);
        else
-               dma_free_coherent(NULL, mp->tx_desc_area_size,
-                               mp->tx_desc_area, mp->tx_desc_dma);
+               dma_free_coherent(NULL, txq->tx_desc_area_size,
+                                 txq->tx_desc_area, txq->tx_desc_dma);
+
+       kfree(txq->tx_skb);
 }
 
 
 /* netdev ops and related ***************************************************/
-static void port_reset(struct mv643xx_eth_private *mp);
-
-static void mv643xx_eth_update_pscr(struct net_device *dev,
-                                   struct ethtool_cmd *ecmd)
+static void update_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       int port_num = mp->port_num;
-       u32 o_pscr, n_pscr;
-       unsigned int queues;
+       u32 pscr_o;
+       u32 pscr_n;
 
-       o_pscr = rdl(mp, PORT_SERIAL_CONTROL(port_num));
-       n_pscr = o_pscr;
+       pscr_o = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
 
        /* clear speed, duplex and rx buffer size fields */
-       n_pscr &= ~(SET_MII_SPEED_TO_100  |
-                  SET_GMII_SPEED_TO_1000 |
-                  SET_FULL_DUPLEX_MODE   |
-                  MAX_RX_PACKET_MASK);
-
-       if (ecmd->duplex == DUPLEX_FULL)
-               n_pscr |= SET_FULL_DUPLEX_MODE;
-
-       if (ecmd->speed == SPEED_1000)
-               n_pscr |= SET_GMII_SPEED_TO_1000 |
-                         MAX_RX_PACKET_9700BYTE;
-       else {
-               if (ecmd->speed == SPEED_100)
-                       n_pscr |= SET_MII_SPEED_TO_100;
-               n_pscr |= MAX_RX_PACKET_1522BYTE;
+       pscr_n = pscr_o & ~(SET_MII_SPEED_TO_100   |
+                           SET_GMII_SPEED_TO_1000 |
+                           SET_FULL_DUPLEX_MODE   |
+                           MAX_RX_PACKET_MASK);
+
+       if (speed == SPEED_1000) {
+               pscr_n |= SET_GMII_SPEED_TO_1000 | MAX_RX_PACKET_9700BYTE;
+       } else {
+               if (speed == SPEED_100)
+                       pscr_n |= SET_MII_SPEED_TO_100;
+               pscr_n |= MAX_RX_PACKET_1522BYTE;
        }
 
-       if (n_pscr != o_pscr) {
-               if ((o_pscr & SERIAL_PORT_ENABLE) == 0)
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
+       if (duplex == DUPLEX_FULL)
+               pscr_n |= SET_FULL_DUPLEX_MODE;
+
+       if (pscr_n != pscr_o) {
+               if ((pscr_o & SERIAL_PORT_ENABLE) == 0)
+                       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr_n);
                else {
-                       queues = mv643xx_eth_port_disable_tx(mp);
-
-                       o_pscr &= ~SERIAL_PORT_ENABLE;
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), o_pscr);
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
-                       if (queues)
-                               mv643xx_eth_port_enable_tx(mp, queues);
+                       int i;
+
+                       for (i = 0; i < 8; i++)
+                               if (mp->txq_mask & (1 << i))
+                                       txq_disable(mp->txq + i);
+
+                       pscr_o &= ~SERIAL_PORT_ENABLE;
+                       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr_o);
+                       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr_n);
+                       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr_n);
+
+                       for (i = 0; i < 8; i++)
+                               if (mp->txq_mask & (1 << i))
+                                       txq_enable(mp->txq + i);
                }
        }
 }
 
-static irqreturn_t mv643xx_eth_int_handler(int irq, void *dev_id)
+static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
 {
        struct net_device *dev = (struct net_device *)dev_id;
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       u32 int_cause, int_cause_ext = 0;
-       unsigned int port_num = mp->port_num;
+       u32 int_cause;
+       u32 int_cause_ext;
+       u32 txq_active;
+
+       int_cause = rdl(mp, INT_CAUSE(mp->port_num)) &
+                       (INT_TX_END | INT_RX | INT_EXT);
+       if (int_cause == 0)
+               return IRQ_NONE;
 
-       /* Read interrupt cause registers */
-       int_cause = rdl(mp, INT_CAUSE(port_num)) & (INT_RX | INT_EXT);
+       int_cause_ext = 0;
        if (int_cause & INT_EXT) {
-               int_cause_ext = rdl(mp, INT_CAUSE_EXT(port_num))
+               int_cause_ext = rdl(mp, INT_CAUSE_EXT(mp->port_num))
                                & (INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
-               wrl(mp, INT_CAUSE_EXT(port_num), ~int_cause_ext);
+               wrl(mp, INT_CAUSE_EXT(mp->port_num), ~int_cause_ext);
        }
 
-       /* PHY status changed */
-       if (int_cause_ext & (INT_EXT_LINK | INT_EXT_PHY)) {
-               struct ethtool_cmd cmd;
+       if (int_cause_ext & (INT_EXT_PHY | INT_EXT_LINK)) {
+               if (mp->phy_addr == -1 || mii_link_ok(&mp->mii)) {
+                       int i;
+
+                       if (mp->phy_addr != -1) {
+                               struct ethtool_cmd cmd;
+
+                               mii_ethtool_gset(&mp->mii, &cmd);
+                               update_pscr(mp, cmd.speed, cmd.duplex);
+                       }
+
+                       for (i = 0; i < 8; i++)
+                               if (mp->txq_mask & (1 << i))
+                                       txq_enable(mp->txq + i);
 
-               if (mii_link_ok(&mp->mii)) {
-                       mii_ethtool_gset(&mp->mii, &cmd);
-                       mv643xx_eth_update_pscr(dev, &cmd);
-                       mv643xx_eth_port_enable_tx(mp, 1);
                        if (!netif_carrier_ok(dev)) {
                                netif_carrier_on(dev);
-                               if (mp->tx_ring_size - mp->tx_desc_count >=
-                                                       MAX_DESCS_PER_SKB)
-                                       netif_wake_queue(dev);
+                               __txq_maybe_wake(mp->txq + mp->txq_primary);
                        }
                } else if (netif_carrier_ok(dev)) {
                        netif_stop_queue(dev);
@@ -1545,54 +1723,81 @@ static irqreturn_t mv643xx_eth_int_handler(int irq, void *dev_id)
                }
        }
 
+       /*
+        * RxBuffer or RxError set for any of the 8 queues?
+        */
 #ifdef MV643XX_ETH_NAPI
        if (int_cause & INT_RX) {
-               /* schedule the NAPI poll routine to maintain port */
-               wrl(mp, INT_MASK(port_num), 0x00000000);
-
-               /* wait for previous write to complete */
-               rdl(mp, INT_MASK(port_num));
+               wrl(mp, INT_MASK(mp->port_num), 0x00000000);
+               rdl(mp, INT_MASK(mp->port_num));
 
                netif_rx_schedule(dev, &mp->napi);
        }
 #else
-       if (int_cause & INT_RX)
-               rxq_process(mp->rxq, INT_MAX);
+       if (int_cause & INT_RX) {
+               int i;
+
+               for (i = 7; i >= 0; i--)
+                       if (mp->rxq_mask & (1 << i))
+                               rxq_process(mp->rxq + i, INT_MAX);
+       }
 #endif
-       if (int_cause_ext & INT_EXT_TX)
-               mv643xx_eth_free_completed_tx_descs(dev);
+
+       txq_active = rdl(mp, TXQ_COMMAND(mp->port_num));
 
        /*
-        * If no real interrupt occured, exit.
-        * This can happen when using gigE interrupt coalescing mechanism.
+        * TxBuffer or TxError set for any of the 8 queues?
         */
-       if ((int_cause == 0x0) && (int_cause_ext == 0x0))
-               return IRQ_NONE;
+       if (int_cause_ext & INT_EXT_TX) {
+               int i;
+
+               for (i = 0; i < 8; i++)
+                       if (mp->txq_mask & (1 << i))
+                               txq_reclaim(mp->txq + i, 0);
+       }
+
+       /*
+        * Any TxEnd interrupts?
+        */
+       if (int_cause & INT_TX_END) {
+               int i;
+
+               wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_TX_END));
+               for (i = 0; i < 8; i++) {
+                       struct tx_queue *txq = mp->txq + i;
+                       if (txq->tx_desc_count && !((txq_active >> i) & 1))
+                               txq_enable(txq);
+               }
+       }
+
+       /*
+        * Enough space again in the primary TX queue for a full packet?
+        */
+       if (int_cause_ext & INT_EXT_TX) {
+               struct tx_queue *txq = mp->txq + mp->txq_primary;
+               __txq_maybe_wake(txq);
+       }
 
        return IRQ_HANDLED;
 }
 
 static void phy_reset(struct mv643xx_eth_private *mp)
 {
-       unsigned int phy_reg_data;
+       unsigned int data;
 
-       /* Reset the PHY */
-       read_smi_reg(mp, 0, &phy_reg_data);
-       phy_reg_data |= 0x8000; /* Set bit 15 to reset the PHY */
-       write_smi_reg(mp, 0, phy_reg_data);
+       smi_reg_read(mp, mp->phy_addr, 0, &data);
+       data |= 0x8000;
+       smi_reg_write(mp, mp->phy_addr, 0, data);
 
-       /* wait for PHY to come out of reset */
        do {
                udelay(1);
-               read_smi_reg(mp, 0, &phy_reg_data);
-       } while (phy_reg_data & 0x8000);
+               smi_reg_read(mp, mp->phy_addr, 0, &data);
+       } while (data & 0x8000);
 }
 
-static void port_start(struct net_device *dev)
+static void port_start(struct mv643xx_eth_private *mp)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
        u32 pscr;
-       struct ethtool_cmd ethtool_cmd;
        int i;
 
        /*
@@ -1612,12 +1817,41 @@ static void port_start(struct net_device *dev)
 
        wrl(mp, SDMA_CONFIG(mp->port_num), PORT_SDMA_CONFIG_DEFAULT_VALUE);
 
-       mv643xx_eth_get_settings(dev, &ethtool_cmd);
-       phy_reset(mp);
-       mv643xx_eth_set_settings(dev, &ethtool_cmd);
+       /*
+        * Perform PHY reset, if there is a PHY.
+        */
+       if (mp->phy_addr != -1) {
+               struct ethtool_cmd cmd;
 
-       /* Add the assigned Ethernet address to the port's address table */
-       uc_addr_set(mp, dev->dev_addr);
+               mv643xx_eth_get_settings(mp->dev, &cmd);
+               phy_reset(mp);
+               mv643xx_eth_set_settings(mp->dev, &cmd);
+       }
+
+       /*
+        * Configure TX path and queues.
+        */
+       tx_set_rate(mp, 1000000000, 16777216);
+       for (i = 0; i < 8; i++) {
+               struct tx_queue *txq = mp->txq + i;
+               int off = TXQ_CURRENT_DESC_PTR(mp->port_num, i);
+               u32 addr;
+
+               if ((mp->txq_mask & (1 << i)) == 0)
+                       continue;
+
+               addr = (u32)txq->tx_desc_dma;
+               addr += txq->tx_curr_desc * sizeof(struct tx_desc);
+               wrl(mp, off, addr);
+
+               txq_set_rate(txq, 1000000000, 16777216);
+               txq_set_fixed_prio_mode(txq);
+       }
+
+       /*
+        * Add configured unicast address to address filter table.
+        */
+       uc_addr_set(mp, mp->dev->dev_addr);
 
        /*
         * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
@@ -1631,150 +1865,124 @@ static void port_start(struct net_device *dev)
        wrl(mp, PORT_CONFIG_EXT(mp->port_num), 0x00000000);
 
        /*
-        * Enable the receive queue.
+        * Enable the receive queues.
         */
-       for (i = 0; i < 1; i++) {
-               struct rx_queue *rxq = mp->rxq;
-               int off = RXQ_CURRENT_DESC_PTR(mp->port_num);
+       for (i = 0; i < 8; i++) {
+               struct rx_queue *rxq = mp->rxq + i;
+               int off = RXQ_CURRENT_DESC_PTR(mp->port_num, i);
                u32 addr;
 
+               if ((mp->rxq_mask & (1 << i)) == 0)
+                       continue;
+
                addr = (u32)rxq->rx_desc_dma;
                addr += rxq->rx_curr_desc * sizeof(struct rx_desc);
                wrl(mp, off, addr);
 
                rxq_enable(rxq);
        }
-
-
-       wrl(mp, TXQ_CURRENT_DESC_PTR(mp->port_num),
-               (u32)((struct tx_desc *)mp->tx_desc_dma + mp->tx_curr_desc));
-
-       /* Disable port bandwidth limits by clearing MTU register */
-       wrl(mp, TX_BW_MTU(mp->port_num), 0);
 }
 
-#ifdef MV643XX_ETH_COAL
-static unsigned int set_rx_coal(struct mv643xx_eth_private *mp,
-                                       unsigned int delay)
+static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
 {
-       unsigned int port_num = mp->port_num;
        unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
-
-       /* Set RX Coalescing mechanism */
-       wrl(mp, SDMA_CONFIG(port_num),
-               ((coal & 0x3fff) << 8) |
-               (rdl(mp, SDMA_CONFIG(port_num))
-                       & 0xffc000ff));
-
-       return coal;
+       u32 val;
+
+       val = rdl(mp, SDMA_CONFIG(mp->port_num));
+       if (mp->shared->extended_rx_coal_limit) {
+               if (coal > 0xffff)
+                       coal = 0xffff;
+               val &= ~0x023fff80;
+               val |= (coal & 0x8000) << 10;
+               val |= (coal & 0x7fff) << 7;
+       } else {
+               if (coal > 0x3fff)
+                       coal = 0x3fff;
+               val &= ~0x003fff00;
+               val |= (coal & 0x3fff) << 8;
+       }
+       wrl(mp, SDMA_CONFIG(mp->port_num), val);
 }
-#endif
 
-static unsigned int set_tx_coal(struct mv643xx_eth_private *mp,
-                                       unsigned int delay)
+static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
 {
        unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
 
-       /* Set TX Coalescing mechanism */
-       wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), coal << 4);
-
-       return coal;
-}
-
-static void port_init(struct mv643xx_eth_private *mp)
-{
-       port_reset(mp);
-
-       init_mac_tables(mp);
+       if (coal > 0x3fff)
+               coal = 0x3fff;
+       wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), (coal & 0x3fff) << 4);
 }
 
 static int mv643xx_eth_open(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
-       unsigned int size;
        int err;
+       int i;
 
-       /* Clear any pending ethernet port interrupts */
-       wrl(mp, INT_CAUSE(port_num), 0);
-       wrl(mp, INT_CAUSE_EXT(port_num), 0);
-       /* wait for previous write to complete */
-       rdl(mp, INT_CAUSE_EXT(port_num));
+       wrl(mp, INT_CAUSE(mp->port_num), 0);
+       wrl(mp, INT_CAUSE_EXT(mp->port_num), 0);
+       rdl(mp, INT_CAUSE_EXT(mp->port_num));
 
-       err = request_irq(dev->irq, mv643xx_eth_int_handler,
-                       IRQF_SHARED | IRQF_SAMPLE_RANDOM, dev->name, dev);
+       err = request_irq(dev->irq, mv643xx_eth_irq,
+                         IRQF_SHARED | IRQF_SAMPLE_RANDOM,
+                         dev->name, dev);
        if (err) {
-               printk(KERN_ERR "%s: Can not assign IRQ\n", dev->name);
+               dev_printk(KERN_ERR, &dev->dev, "can't assign irq\n");
                return -EAGAIN;
        }
 
-       port_init(mp);
+       init_mac_tables(mp);
 
-       err = rxq_init(mp);
-       if (err)
-               goto out_free_irq;
-       rxq_refill(mp->rxq);
+       for (i = 0; i < 8; i++) {
+               if ((mp->rxq_mask & (1 << i)) == 0)
+                       continue;
 
-       mp->tx_skb = kmalloc(sizeof(*mp->tx_skb) * mp->tx_ring_size,
-                                                               GFP_KERNEL);
-       if (!mp->tx_skb) {
-               printk(KERN_ERR "%s: Cannot allocate Tx skb ring\n", dev->name);
-               err = -ENOMEM;
-               goto out_free_rx_skb;
+               err = rxq_init(mp, i);
+               if (err) {
+                       while (--i >= 0)
+                               if (mp->rxq_mask & (1 << i))
+                                       rxq_deinit(mp->rxq + i);
+                       goto out;
+               }
+
+               rxq_refill(mp->rxq + i);
        }
 
-       /* Allocate TX ring */
-       mp->tx_desc_count = 0;
-       size = mp->tx_ring_size * sizeof(struct tx_desc);
-       mp->tx_desc_area_size = size;
-
-       if (mp->tx_sram_size) {
-               mp->tx_desc_area = ioremap(mp->tx_sram_addr,
-                                                       mp->tx_sram_size);
-               mp->tx_desc_dma = mp->tx_sram_addr;
-       } else
-               mp->tx_desc_area = dma_alloc_coherent(NULL, size,
-                                                       &mp->tx_desc_dma,
-                                                       GFP_KERNEL);
+       for (i = 0; i < 8; i++) {
+               if ((mp->txq_mask & (1 << i)) == 0)
+                       continue;
 
-       if (!mp->tx_desc_area) {
-               printk(KERN_ERR "%s: Cannot allocate Tx Ring (size %d bytes)\n",
-                                                       dev->name, size);
-               err = -ENOMEM;
-               goto out_free_tx_skb;
+               err = txq_init(mp, i);
+               if (err) {
+                       while (--i >= 0)
+                               if (mp->txq_mask & (1 << i))
+                                       txq_deinit(mp->txq + i);
+                       goto out_free;
+               }
        }
-       BUG_ON((u32) mp->tx_desc_area & 0xf);   /* check 16-byte alignment */
-       memset((void *)mp->tx_desc_area, 0, mp->tx_desc_area_size);
-
-       ether_init_tx_desc_ring(mp);
 
 #ifdef MV643XX_ETH_NAPI
        napi_enable(&mp->napi);
 #endif
 
-       port_start(dev);
-
-       /* Interrupt Coalescing */
-
-#ifdef MV643XX_ETH_COAL
-       mp->rx_int_coal = set_rx_coal(mp, MV643XX_ETH_RX_COAL);
-#endif
+       port_start(mp);
 
-       mp->tx_int_coal = set_tx_coal(mp, MV643XX_ETH_TX_COAL);
+       set_rx_coal(mp, 0);
+       set_tx_coal(mp, 0);
 
-       /* Unmask phy and link status changes interrupts */
-       wrl(mp, INT_MASK_EXT(port_num), INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
+       wrl(mp, INT_MASK_EXT(mp->port_num),
+           INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
 
-       /* Unmask RX buffer and TX end interrupt */
-       wrl(mp, INT_MASK(port_num), INT_RX | INT_EXT);
+       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
 
        return 0;
 
-out_free_tx_skb:
-       kfree(mp->tx_skb);
-out_free_rx_skb:
-       rxq_deinit(mp->rxq);
-out_free_irq:
+
+out_free:
+       for (i = 0; i < 8; i++)
+               if (mp->rxq_mask & (1 << i))
+                       rxq_deinit(mp->rxq + i);
+out:
        free_irq(dev->irq, dev);
 
        return err;
@@ -1782,32 +1990,33 @@ out_free_irq:
 
 static void port_reset(struct mv643xx_eth_private *mp)
 {
-       unsigned int port_num = mp->port_num;
-       unsigned int reg_data;
-
-       mv643xx_eth_port_disable_tx(mp);
-       rxq_disable(mp->rxq);
+       unsigned int data;
+       int i;
 
-       /* Clear all MIB counters */
-       clear_mib_counters(mp);
+       for (i = 0; i < 8; i++) {
+               if (mp->rxq_mask & (1 << i))
+                       rxq_disable(mp->rxq + i);
+               if (mp->txq_mask & (1 << i))
+                       txq_disable(mp->txq + i);
+       }
+       while (!(rdl(mp, PORT_STATUS(mp->port_num)) & TX_FIFO_EMPTY))
+               udelay(10);
 
        /* Reset the Enable bit in the Configuration Register */
-       reg_data = rdl(mp, PORT_SERIAL_CONTROL(port_num));
-       reg_data &= ~(SERIAL_PORT_ENABLE                |
-                       DO_NOT_FORCE_LINK_FAIL  |
-                       FORCE_LINK_PASS);
-       wrl(mp, PORT_SERIAL_CONTROL(port_num), reg_data);
+       data = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       data &= ~(SERIAL_PORT_ENABLE            |
+                 DO_NOT_FORCE_LINK_FAIL        |
+                 FORCE_LINK_PASS);
+       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), data);
 }
 
 static int mv643xx_eth_stop(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
+       int i;
 
-       /* Mask all interrupts on ethernet port */
-       wrl(mp, INT_MASK(port_num), 0x00000000);
-       /* wait for previous write to complete */
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
+       rdl(mp, INT_MASK(mp->port_num));
 
 #ifdef MV643XX_ETH_NAPI
        napi_disable(&mp->napi);
@@ -1815,29 +2024,41 @@ static int mv643xx_eth_stop(struct net_device *dev)
        netif_carrier_off(dev);
        netif_stop_queue(dev);
 
-       port_reset(mp);
+       free_irq(dev->irq, dev);
 
-       mv643xx_eth_free_tx_rings(dev);
-       rxq_deinit(mp->rxq);
+       port_reset(mp);
+       mib_counters_update(mp);
 
-       free_irq(dev->irq, dev);
+       for (i = 0; i < 8; i++) {
+               if (mp->rxq_mask & (1 << i))
+                       rxq_deinit(mp->rxq + i);
+               if (mp->txq_mask & (1 << i))
+                       txq_deinit(mp->txq + i);
+       }
 
        return 0;
 }
 
-static int mv643xx_eth_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
+static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
+       if (mp->phy_addr != -1)
+               return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
+
+       return -EOPNOTSUPP;
 }
 
 static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
 {
-       if ((new_mtu > 9500) || (new_mtu < 64))
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       if (new_mtu < 64 || new_mtu > 9500)
                return -EINVAL;
 
        dev->mtu = new_mtu;
+       tx_set_rate(mp, 1000000000, 16777216);
+
        if (!netif_running(dev))
                return 0;
 
@@ -1845,74 +2066,70 @@ static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
         * Stop and then re-open the interface. This will allocate RX
         * skbs of the new MTU.
         * There is a possible danger that the open will not succeed,
-        * due to memory being full, which might fail the open function.
+        * due to memory being full.
         */
        mv643xx_eth_stop(dev);
        if (mv643xx_eth_open(dev)) {
-               printk(KERN_ERR "%s: Fatal error on opening device\n",
-                       dev->name);
+               dev_printk(KERN_ERR, &dev->dev,
+                          "fatal error on re-opening device after "
+                          "MTU change\n");
        }
 
        return 0;
 }
 
-static void mv643xx_eth_tx_timeout_task(struct work_struct *ugly)
+static void tx_timeout_task(struct work_struct *ugly)
 {
-       struct mv643xx_eth_private *mp = container_of(ugly, struct mv643xx_eth_private,
-                                                 tx_timeout_task);
-       struct net_device *dev = mp->dev;
-
-       if (!netif_running(dev))
-               return;
+       struct mv643xx_eth_private *mp;
 
-       netif_stop_queue(dev);
+       mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
+       if (netif_running(mp->dev)) {
+               netif_stop_queue(mp->dev);
 
-       port_reset(mp);
-       port_start(dev);
+               port_reset(mp);
+               port_start(mp);
 
-       if (mp->tx_ring_size - mp->tx_desc_count >= MAX_DESCS_PER_SKB)
-               netif_wake_queue(dev);
+               __txq_maybe_wake(mp->txq + mp->txq_primary);
+       }
 }
 
 static void mv643xx_eth_tx_timeout(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       printk(KERN_INFO "%s: TX timeout  ", dev->name);
+       dev_printk(KERN_INFO, &dev->dev, "tx timeout\n");
 
-       /* Do the reset outside of interrupt context */
        schedule_work(&mp->tx_timeout_task);
 }
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
-static void mv643xx_eth_netpoll(struct net_device *netdev)
+static void mv643xx_eth_netpoll(struct net_device *dev)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(netdev);
-       int port_num = mp->port_num;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       wrl(mp, INT_MASK(port_num), 0x00000000);
-       /* wait for previous write to complete */
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
+       rdl(mp, INT_MASK(mp->port_num));
 
-       mv643xx_eth_int_handler(netdev->irq, netdev);
+       mv643xx_eth_irq(dev->irq, dev);
 
-       wrl(mp, INT_MASK(port_num), INT_RX | INT_CAUSE_EXT);
+       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_CAUSE_EXT);
 }
 #endif
 
-static int mv643xx_eth_mdio_read(struct net_device *dev, int phy_id, int location)
+static int mv643xx_eth_mdio_read(struct net_device *dev, int addr, int reg)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
        int val;
 
-       read_smi_reg(mp, location, &val);
+       smi_reg_read(mp, addr, reg, &val);
+
        return val;
 }
 
-static void mv643xx_eth_mdio_write(struct net_device *dev, int phy_id, int location, int val)
+static void mv643xx_eth_mdio_write(struct net_device *dev, int addr, int reg, int val)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       write_smi_reg(mp, location, val);
+       smi_reg_write(mp, addr, reg, val);
 }
 
 
@@ -1952,6 +2169,30 @@ mv643xx_eth_conf_mbus_windows(struct mv643xx_eth_shared_private *msp,
        msp->win_protect = win_protect;
 }
 
+static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
+{
+       /*
+        * Check whether we have a 14-bit coal limit field in bits
+        * [21:8], or a 16-bit coal limit in bits [25,21:7] of the
+        * SDMA config register.
+        */
+       writel(0x02000000, msp->base + SDMA_CONFIG(0));
+       if (readl(msp->base + SDMA_CONFIG(0)) & 0x02000000)
+               msp->extended_rx_coal_limit = 1;
+       else
+               msp->extended_rx_coal_limit = 0;
+
+       /*
+        * Check whether the TX rate control 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;
+}
+
 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 {
        static int mv643xx_eth_version_printed = 0;
@@ -1979,9 +2220,6 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
                goto out_free;
 
        spin_lock_init(&msp->phy_lock);
-       msp->t_clk = (pd != NULL && pd->t_clk != 0) ? pd->t_clk : 133000000;
-
-       platform_set_drvdata(pdev, msp);
 
        /*
         * (Re-)program MBUS remapping windows if we are asked to.
@@ -1989,6 +2227,14 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
        if (pd != NULL && pd->dram != NULL)
                mv643xx_eth_conf_mbus_windows(msp, pd->dram);
 
+       /*
+        * Detect hardware parameters.
+        */
+       msp->t_clk = (pd != NULL && pd->t_clk != 0) ? pd->t_clk : 133000000;
+       infer_hw_params(msp);
+
+       platform_set_drvdata(pdev, msp);
+
        return 0;
 
 out_free:
@@ -2008,104 +2254,170 @@ static int mv643xx_eth_shared_remove(struct platform_device *pdev)
 }
 
 static struct platform_driver mv643xx_eth_shared_driver = {
-       .probe = mv643xx_eth_shared_probe,
-       .remove = mv643xx_eth_shared_remove,
+       .probe          = mv643xx_eth_shared_probe,
+       .remove         = mv643xx_eth_shared_remove,
        .driver = {
-               .name = MV643XX_ETH_SHARED_NAME,
+               .name   = MV643XX_ETH_SHARED_NAME,
                .owner  = THIS_MODULE,
        },
 };
 
 static void phy_addr_set(struct mv643xx_eth_private *mp, int phy_addr)
 {
-       u32 reg_data;
        int addr_shift = 5 * mp->port_num;
+       u32 data;
 
-       reg_data = rdl(mp, PHY_ADDR);
-       reg_data &= ~(0x1f << addr_shift);
-       reg_data |= (phy_addr & 0x1f) << addr_shift;
-       wrl(mp, PHY_ADDR, reg_data);
+       data = rdl(mp, PHY_ADDR);
+       data &= ~(0x1f << addr_shift);
+       data |= (phy_addr & 0x1f) << addr_shift;
+       wrl(mp, PHY_ADDR, data);
 }
 
 static int phy_addr_get(struct mv643xx_eth_private *mp)
 {
-       unsigned int reg_data;
+       unsigned int data;
+
+       data = rdl(mp, PHY_ADDR);
+
+       return (data >> (5 * mp->port_num)) & 0x1f;
+}
+
+static void set_params(struct mv643xx_eth_private *mp,
+                      struct mv643xx_eth_platform_data *pd)
+{
+       struct net_device *dev = mp->dev;
+
+       if (is_valid_ether_addr(pd->mac_addr))
+               memcpy(dev->dev_addr, pd->mac_addr, 6);
+       else
+               uc_addr_get(mp, dev->dev_addr);
+
+       if (pd->phy_addr == -1) {
+               mp->shared_smi = NULL;
+               mp->phy_addr = -1;
+       } else {
+               mp->shared_smi = mp->shared;
+               if (pd->shared_smi != NULL)
+                       mp->shared_smi = platform_get_drvdata(pd->shared_smi);
+
+               if (pd->force_phy_addr || pd->phy_addr) {
+                       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;
+       mp->rx_desc_sram_addr = pd->rx_sram_addr;
+       mp->rx_desc_sram_size = pd->rx_sram_size;
+
+       if (pd->rx_queue_mask)
+               mp->rxq_mask = pd->rx_queue_mask;
+       else
+               mp->rxq_mask = 0x01;
+       mp->rxq_primary = fls(mp->rxq_mask) - 1;
 
-       reg_data = rdl(mp, PHY_ADDR);
+       mp->default_tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
+       if (pd->tx_queue_size)
+               mp->default_tx_ring_size = pd->tx_queue_size;
+       mp->tx_desc_sram_addr = pd->tx_sram_addr;
+       mp->tx_desc_sram_size = pd->tx_sram_size;
 
-       return ((reg_data >> (5 * mp->port_num)) & 0x1f);
+       if (pd->tx_queue_mask)
+               mp->txq_mask = pd->tx_queue_mask;
+       else
+               mp->txq_mask = 0x01;
+       mp->txq_primary = fls(mp->txq_mask) - 1;
 }
 
 static int phy_detect(struct mv643xx_eth_private *mp)
 {
-       unsigned int phy_reg_data0;
-       int auto_neg;
+       unsigned int data;
+       unsigned int data2;
+
+       smi_reg_read(mp, mp->phy_addr, 0, &data);
+       smi_reg_write(mp, mp->phy_addr, 0, data ^ 0x1000);
 
-       read_smi_reg(mp, 0, &phy_reg_data0);
-       auto_neg = phy_reg_data0 & 0x1000;
-       phy_reg_data0 ^= 0x1000;        /* invert auto_neg */
-       write_smi_reg(mp, 0, phy_reg_data0);
+       smi_reg_read(mp, mp->phy_addr, 0, &data2);
+       if (((data ^ data2) & 0x1000) == 0)
+               return -ENODEV;
 
-       read_smi_reg(mp, 0, &phy_reg_data0);
-       if ((phy_reg_data0 & 0x1000) == auto_neg)
-               return -ENODEV;                         /* change didn't take */
+       smi_reg_write(mp, mp->phy_addr, 0, data);
 
-       phy_reg_data0 ^= 0x1000;
-       write_smi_reg(mp, 0, phy_reg_data0);
        return 0;
 }
 
-static void mv643xx_init_ethtool_cmd(struct net_device *dev, int phy_address,
-                                    int speed, int duplex,
-                                    struct ethtool_cmd *cmd)
+static int phy_init(struct mv643xx_eth_private *mp,
+                   struct mv643xx_eth_platform_data *pd)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       struct ethtool_cmd cmd;
+       int err;
 
-       memset(cmd, 0, sizeof(*cmd));
+       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);
 
-       cmd->port = PORT_MII;
-       cmd->transceiver = XCVR_INTERNAL;
-       cmd->phy_address = phy_address;
-
-       if (speed == 0) {
-               cmd->autoneg = AUTONEG_ENABLE;
-               /* mii lib checks, but doesn't use speed on AUTONEG_ENABLE */
-               cmd->speed = SPEED_100;
-               cmd->advertising = ADVERTISED_10baseT_Half  |
-                                  ADVERTISED_10baseT_Full  |
-                                  ADVERTISED_100baseT_Half |
-                                  ADVERTISED_100baseT_Full;
+       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;
+                       cmd.advertising |= ADVERTISED_1000baseT_Full;
        } else {
-               cmd->autoneg = AUTONEG_DISABLE;
-               cmd->speed = speed;
-               cmd->duplex = duplex;
+               cmd.autoneg = AUTONEG_DISABLE;
+               cmd.speed = pd->speed;
+               cmd.duplex = pd->duplex;
        }
+
+       update_pscr(mp, cmd.speed, cmd.duplex);
+       mv643xx_eth_set_settings(mp->dev, &cmd);
+
+       return 0;
 }
 
 static int mv643xx_eth_probe(struct platform_device *pdev)
 {
        struct mv643xx_eth_platform_data *pd;
-       int port_num;
        struct mv643xx_eth_private *mp;
        struct net_device *dev;
-       u8 *p;
        struct resource *res;
-       int err;
-       struct ethtool_cmd cmd;
-       int duplex = DUPLEX_HALF;
-       int speed = 0;                  /* default to auto-negotiation */
        DECLARE_MAC_BUF(mac);
+       int err;
 
        pd = pdev->dev.platform_data;
        if (pd == NULL) {
-               printk(KERN_ERR "No mv643xx_eth_platform_data\n");
+               dev_printk(KERN_ERR, &pdev->dev,
+                          "no mv643xx_eth_platform_data\n");
                return -ENODEV;
        }
 
        if (pd->shared == NULL) {
-               printk(KERN_ERR "No mv643xx_eth_platform_data->shared\n");
+               dev_printk(KERN_ERR, &pdev->dev,
+                          "no mv643xx_eth_platform_data->shared\n");
                return -ENODEV;
        }
 
@@ -2113,145 +2425,85 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
        if (!dev)
                return -ENOMEM;
 
-       platform_set_drvdata(pdev, dev);
-
        mp = netdev_priv(dev);
+       platform_set_drvdata(pdev, mp);
+
+       mp->shared = platform_get_drvdata(pd->shared);
+       mp->port_num = pd->port_number;
+
        mp->dev = dev;
 #ifdef MV643XX_ETH_NAPI
        netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, 64);
 #endif
 
+       set_params(mp, pd);
+
+       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;
+
+               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
+       } else {
+               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops_phyless);
+       }
+
+
        res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
        BUG_ON(!res);
        dev->irq = res->start;
 
+       dev->hard_start_xmit = mv643xx_eth_xmit;
        dev->open = mv643xx_eth_open;
        dev->stop = mv643xx_eth_stop;
-       dev->hard_start_xmit = mv643xx_eth_start_xmit;
-       dev->set_mac_address = mv643xx_eth_set_mac_address;
        dev->set_multicast_list = mv643xx_eth_set_rx_mode;
-
-       /* No need to Tx Timeout */
+       dev->set_mac_address = mv643xx_eth_set_mac_address;
+       dev->do_ioctl = mv643xx_eth_ioctl;
+       dev->change_mtu = mv643xx_eth_change_mtu;
        dev->tx_timeout = mv643xx_eth_tx_timeout;
-
 #ifdef CONFIG_NET_POLL_CONTROLLER
        dev->poll_controller = mv643xx_eth_netpoll;
 #endif
-
        dev->watchdog_timeo = 2 * HZ;
        dev->base_addr = 0;
-       dev->change_mtu = mv643xx_eth_change_mtu;
-       dev->do_ioctl = mv643xx_eth_do_ioctl;
-       SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
 
 #ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-#ifdef MAX_SKB_FRAGS
        /*
         * Zero copy can only work if we use Discovery II memory. Else, we will
         * have to map the buffers to ISA memory which is only 16 MB
         */
        dev->features = NETIF_F_SG | NETIF_F_IP_CSUM;
 #endif
-#endif
-
-       /* Configure the timeout task */
-       INIT_WORK(&mp->tx_timeout_task, mv643xx_eth_tx_timeout_task);
-
-       spin_lock_init(&mp->lock);
 
-       mp->shared = platform_get_drvdata(pd->shared);
-       port_num = mp->port_num = pd->port_number;
+       SET_NETDEV_DEV(dev, &pdev->dev);
 
        if (mp->shared->win_protect)
-               wrl(mp, WINDOW_PROTECT(port_num), mp->shared->win_protect);
-
-       mp->shared_smi = mp->shared;
-       if (pd->shared_smi != NULL)
-               mp->shared_smi = platform_get_drvdata(pd->shared_smi);
-
-       /* set default config values */
-       uc_addr_get(mp, dev->dev_addr);
-       mp->tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
-
-       if (is_valid_ether_addr(pd->mac_addr))
-               memcpy(dev->dev_addr, pd->mac_addr, 6);
-
-       if (pd->phy_addr || pd->force_phy_addr)
-               phy_addr_set(mp, pd->phy_addr);
-
-       mp->default_rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
-       if (pd->rx_queue_size)
-               mp->default_rx_ring_size = pd->rx_queue_size;
-
-       if (pd->tx_queue_size)
-               mp->tx_ring_size = pd->tx_queue_size;
-
-       if (pd->tx_sram_size) {
-               mp->tx_sram_size = pd->tx_sram_size;
-               mp->tx_sram_addr = pd->tx_sram_addr;
-       }
-
-       if (pd->rx_sram_size) {
-               mp->rx_desc_sram_addr = pd->rx_sram_addr;
-               mp->rx_desc_sram_size = pd->rx_sram_size;
-       }
-
-       duplex = pd->duplex;
-       speed = pd->speed;
+               wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
 
-       /* Hook up MII support for ethtool */
-       mp->mii.dev = dev;
-       mp->mii.mdio_read = mv643xx_eth_mdio_read;
-       mp->mii.mdio_write = mv643xx_eth_mdio_write;
-       mp->mii.phy_id = phy_addr_get(mp);
-       mp->mii.phy_id_mask = 0x3f;
-       mp->mii.reg_num_mask = 0x1f;
-
-       err = phy_detect(mp);
-       if (err) {
-               pr_debug("%s: No PHY detected at addr %d\n",
-                               dev->name, phy_addr_get(mp));
-               goto out;
-       }
-
-       phy_reset(mp);
-       mp->mii.supports_gmii = mii_check_gmii_support(&mp->mii);
-       mv643xx_init_ethtool_cmd(dev, mp->mii.phy_id, speed, duplex, &cmd);
-       mv643xx_eth_update_pscr(dev, &cmd);
-       mv643xx_eth_set_settings(dev, &cmd);
-
-       SET_NETDEV_DEV(dev, &pdev->dev);
        err = register_netdev(dev);
        if (err)
                goto out;
 
-       p = dev->dev_addr;
-       printk(KERN_NOTICE
-               "%s: port %d with MAC address %s\n",
-               dev->name, port_num, print_mac(mac, p));
+       dev_printk(KERN_NOTICE, &dev->dev, "port %d with MAC address %s\n",
+                  mp->port_num, print_mac(mac, dev->dev_addr));
 
        if (dev->features & NETIF_F_SG)
-               printk(KERN_NOTICE "%s: Scatter Gather Enabled\n", dev->name);
+               dev_printk(KERN_NOTICE, &dev->dev, "scatter/gather enabled\n");
 
        if (dev->features & NETIF_F_IP_CSUM)
-               printk(KERN_NOTICE "%s: TX TCP/IP Checksumming Supported\n",
-                                                               dev->name);
-
-#ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-       printk(KERN_NOTICE "%s: RX TCP/UDP Checksum Offload ON \n", dev->name);
-#endif
-
-#ifdef MV643XX_ETH_COAL
-       printk(KERN_NOTICE "%s: TX and RX Interrupt Coalescing ON \n",
-                                                               dev->name);
-#endif
+               dev_printk(KERN_NOTICE, &dev->dev, "tx checksum offload\n");
 
 #ifdef MV643XX_ETH_NAPI
-       printk(KERN_NOTICE "%s: RX NAPI Enabled \n", dev->name);
+       dev_printk(KERN_NOTICE, &dev->dev, "napi enabled\n");
 #endif
 
-       if (mp->tx_sram_size > 0)
-               printk(KERN_NOTICE "%s: Using SRAM\n", dev->name);
+       if (mp->tx_desc_sram_size > 0)
+               dev_printk(KERN_NOTICE, &dev->dev, "configured with sram\n");
 
        return 0;
 
@@ -2263,35 +2515,35 @@ out:
 
 static int mv643xx_eth_remove(struct platform_device *pdev)
 {
-       struct net_device *dev = platform_get_drvdata(pdev);
+       struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
-       unregister_netdev(dev);
+       unregister_netdev(mp->dev);
        flush_scheduled_work();
+       free_netdev(mp->dev);
 
-       free_netdev(dev);
        platform_set_drvdata(pdev, NULL);
+
        return 0;
 }
 
 static void mv643xx_eth_shutdown(struct platform_device *pdev)
 {
-       struct net_device *dev = platform_get_drvdata(pdev);
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
+       struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
        /* Mask all interrupts on ethernet port */
-       wrl(mp, INT_MASK(port_num), 0);
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0);
+       rdl(mp, INT_MASK(mp->port_num));
 
-       port_reset(mp);
+       if (netif_running(mp->dev))
+               port_reset(mp);
 }
 
 static struct platform_driver mv643xx_eth_driver = {
-       .probe = mv643xx_eth_probe,
-       .remove = mv643xx_eth_remove,
-       .shutdown = mv643xx_eth_shutdown,
+       .probe          = mv643xx_eth_probe,
+       .remove         = mv643xx_eth_remove,
+       .shutdown       = mv643xx_eth_shutdown,
        .driver = {
-               .name = MV643XX_ETH_NAME,
+               .name   = MV643XX_ETH_NAME,
                .owner  = THIS_MODULE,
        },
 };
@@ -2306,21 +2558,21 @@ static int __init mv643xx_eth_init_module(void)
                if (rc)
                        platform_driver_unregister(&mv643xx_eth_shared_driver);
        }
+
        return rc;
 }
+module_init(mv643xx_eth_init_module);
 
 static void __exit mv643xx_eth_cleanup_module(void)
 {
        platform_driver_unregister(&mv643xx_eth_driver);
        platform_driver_unregister(&mv643xx_eth_shared_driver);
 }
-
-module_init(mv643xx_eth_init_module);
 module_exit(mv643xx_eth_cleanup_module);
 
-MODULE_LICENSE("GPL");
-MODULE_AUTHOR( "Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, Manish Lachwani"
-               " and Dale Farnsworth");
+MODULE_AUTHOR("Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, "
+             "Manish Lachwani, Dale Farnsworth and Lennert Buytenhek");
 MODULE_DESCRIPTION("Ethernet driver for Marvell MV643XX");
-MODULE_ALIAS("platform:" MV643XX_ETH_NAME);
+MODULE_LICENSE("GPL");
 MODULE_ALIAS("platform:" MV643XX_ETH_SHARED_NAME);
+MODULE_ALIAS("platform:" MV643XX_ETH_NAME);