net: spread __net_init, __net_exit
[safe/jmp/linux-2.6] / drivers / net / mv643xx_eth.c
index 8c812c3..af67af5 100644 (file)
@@ -38,6 +38,7 @@
 #include <linux/init.h>
 #include <linux/dma-mapping.h>
 #include <linux/in.h>
+#include <linux/ip.h>
 #include <linux/tcp.h>
 #include <linux/udp.h>
 #include <linux/etherdevice.h>
 #include <linux/kernel.h>
 #include <linux/spinlock.h>
 #include <linux/workqueue.h>
-#include <linux/mii.h>
+#include <linux/phy.h>
 #include <linux/mv643xx_eth.h>
-#include <asm/io.h>
-#include <asm/types.h>
+#include <linux/io.h>
+#include <linux/types.h>
+#include <linux/inet_lro.h>
 #include <asm/system.h>
+#include <linux/list.h>
 
 static char mv643xx_eth_driver_name[] = "mv643xx_eth";
-static char mv643xx_eth_driver_version[] = "1.3";
+static char mv643xx_eth_driver_version[] = "1.4";
 
-#define MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-#define MV643XX_ETH_NAPI
-#define MV643XX_ETH_TX_FAST_REFILL
-
-#ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-#define MAX_DESCS_PER_SKB      (MAX_SKB_FRAGS + 1)
-#else
-#define MAX_DESCS_PER_SKB      1
-#endif
 
 /*
  * Registers shared between all ports.
  */
 #define PHY_ADDR                       0x0000
 #define SMI_REG                                0x0004
+#define  SMI_BUSY                      0x10000000
+#define  SMI_READ_VALID                        0x08000000
+#define  SMI_OPCODE_READ               0x04000000
+#define  SMI_OPCODE_WRITE              0x00000000
+#define ERR_INT_CAUSE                  0x0080
+#define  ERR_INT_SMI_DONE              0x00000010
+#define ERR_INT_MASK                   0x0084
 #define WINDOW_BASE(w)                 (0x0200 + ((w) << 3))
 #define WINDOW_SIZE(w)                 (0x0204 + ((w) << 3))
 #define WINDOW_REMAP_HIGH(w)           (0x0280 + ((w) << 2))
@@ -79,16 +80,34 @@ static char mv643xx_eth_driver_version[] = "1.3";
 #define WINDOW_PROTECT(w)              (0x0294 + ((w) << 4))
 
 /*
- * Per-port registers.
+ * Main per-port registers.  These live at offset 0x0400 for
+ * port #0, 0x0800 for port #1, and 0x0c00 for port #2.
  */
-#define PORT_CONFIG(p)                 (0x0400 + ((p) << 10))
+#define PORT_CONFIG                    0x0000
 #define  UNICAST_PROMISCUOUS_MODE      0x00000001
-#define PORT_CONFIG_EXT(p)             (0x0404 + ((p) << 10))
-#define MAC_ADDR_LOW(p)                        (0x0414 + ((p) << 10))
-#define MAC_ADDR_HIGH(p)               (0x0418 + ((p) << 10))
-#define SDMA_CONFIG(p)                 (0x041c + ((p) << 10))
-#define PORT_SERIAL_CONTROL(p)         (0x043c + ((p) << 10))
-#define PORT_STATUS(p)                 (0x0444 + ((p) << 10))
+#define PORT_CONFIG_EXT                        0x0004
+#define MAC_ADDR_LOW                   0x0014
+#define MAC_ADDR_HIGH                  0x0018
+#define SDMA_CONFIG                    0x001c
+#define  TX_BURST_SIZE_16_64BIT                0x01000000
+#define  TX_BURST_SIZE_4_64BIT         0x00800000
+#define  BLM_TX_NO_SWAP                        0x00000020
+#define  BLM_RX_NO_SWAP                        0x00000010
+#define  RX_BURST_SIZE_16_64BIT                0x00000008
+#define  RX_BURST_SIZE_4_64BIT         0x00000004
+#define PORT_SERIAL_CONTROL            0x003c
+#define  SET_MII_SPEED_TO_100          0x01000000
+#define  SET_GMII_SPEED_TO_1000                0x00800000
+#define  SET_FULL_DUPLEX_MODE          0x00200000
+#define  MAX_RX_PACKET_9700BYTE                0x000a0000
+#define  DISABLE_AUTO_NEG_SPEED_GMII   0x00002000
+#define  DO_NOT_FORCE_LINK_FAIL                0x00000400
+#define  SERIAL_PORT_CONTROL_RESERVED  0x00000200
+#define  DISABLE_AUTO_NEG_FOR_FLOW_CTRL        0x00000008
+#define  DISABLE_AUTO_NEG_FOR_DUPLEX   0x00000004
+#define  FORCE_LINK_PASS               0x00000002
+#define  SERIAL_PORT_ENABLE            0x00000001
+#define PORT_STATUS                    0x0044
 #define  TX_FIFO_EMPTY                 0x00000400
 #define  TX_IN_PROGRESS                        0x00000080
 #define  PORT_SPEED_MASK               0x00000030
@@ -98,35 +117,37 @@ static char mv643xx_eth_driver_version[] = "1.3";
 #define  FLOW_CONTROL_ENABLED          0x00000008
 #define  FULL_DUPLEX                   0x00000004
 #define  LINK_UP                       0x00000002
-#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_TX_END_0                  0x00080000
+#define TXQ_COMMAND                    0x0048
+#define TXQ_FIX_PRIO_CONF              0x004c
+#define TX_BW_RATE                     0x0050
+#define TX_BW_MTU                      0x0058
+#define TX_BW_BURST                    0x005c
+#define INT_CAUSE                      0x0060
 #define  INT_TX_END                    0x07f80000
-#define  INT_RX                                0x0007fbfc
+#define  INT_TX_END_0                  0x00080000
+#define  INT_RX                                0x000003fc
+#define  INT_RX_0                      0x00000004
 #define  INT_EXT                       0x00000002
-#define INT_CAUSE_EXT(p)               (0x0464 + ((p) << 10))
-#define  INT_EXT_LINK                  0x00100000
-#define  INT_EXT_PHY                   0x00010000
-#define  INT_EXT_TX_ERROR_0            0x00000100
-#define  INT_EXT_TX_0                  0x00000001
-#define  INT_EXT_TX                    0x0000ffff
-#define INT_MASK(p)                    (0x0468 + ((p) << 10))
-#define INT_MASK_EXT(p)                        (0x046c + ((p) << 10))
-#define TX_FIFO_URGENT_THRESHOLD(p)    (0x0474 + ((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, 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 INT_CAUSE_EXT                  0x0064
+#define  INT_EXT_LINK_PHY              0x00110000
+#define  INT_EXT_TX                    0x000000ff
+#define INT_MASK                       0x0068
+#define INT_MASK_EXT                   0x006c
+#define TX_FIFO_URGENT_THRESHOLD       0x0074
+#define TXQ_FIX_PRIO_CONF_MOVED                0x00dc
+#define TX_BW_RATE_MOVED               0x00e0
+#define TX_BW_MTU_MOVED                        0x00e8
+#define TX_BW_BURST_MOVED              0x00ec
+#define RXQ_CURRENT_DESC_PTR(q)                (0x020c + ((q) << 4))
+#define RXQ_COMMAND                    0x0280
+#define TXQ_CURRENT_DESC_PTR(q)                (0x02c0 + ((q) << 2))
+#define TXQ_BW_TOKENS(q)               (0x0300 + ((q) << 4))
+#define TXQ_BW_CONF(q)                 (0x0304 + ((q) << 4))
+#define TXQ_BW_WRR_CONF(q)             (0x0308 + ((q) << 4))
+
+/*
+ * Misc per-port registers.
+ */
 #define MIB_COUNTERS(p)                        (0x1000 + ((p) << 7))
 #define SPECIAL_MCAST_TABLE(p)         (0x1400 + ((p) << 10))
 #define OTHER_MCAST_TABLE(p)           (0x1500 + ((p) << 10))
@@ -134,45 +155,29 @@ static char mv643xx_eth_driver_version[] = "1.3";
 
 
 /*
- * SDMA configuration register.
+ * SDMA configuration register default value.
  */
-#define RX_BURST_SIZE_16_64BIT         (4 << 1)
-#define BLM_RX_NO_SWAP                 (1 << 4)
-#define BLM_TX_NO_SWAP                 (1 << 5)
-#define TX_BURST_SIZE_16_64BIT         (4 << 22)
-
 #if defined(__BIG_ENDIAN)
 #define PORT_SDMA_CONFIG_DEFAULT_VALUE         \
-               RX_BURST_SIZE_16_64BIT  |       \
-               TX_BURST_SIZE_16_64BIT
+               (RX_BURST_SIZE_4_64BIT  |       \
+                TX_BURST_SIZE_4_64BIT)
 #elif defined(__LITTLE_ENDIAN)
 #define PORT_SDMA_CONFIG_DEFAULT_VALUE         \
-               RX_BURST_SIZE_16_64BIT  |       \
-               BLM_RX_NO_SWAP          |       \
-               BLM_TX_NO_SWAP          |       \
-               TX_BURST_SIZE_16_64BIT
+               (RX_BURST_SIZE_4_64BIT  |       \
+                BLM_RX_NO_SWAP         |       \
+                BLM_TX_NO_SWAP         |       \
+                TX_BURST_SIZE_4_64BIT)
 #else
 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
 #endif
 
 
 /*
- * Port serial control register.
+ * Misc definitions.
  */
-#define SET_MII_SPEED_TO_100                   (1 << 24)
-#define SET_GMII_SPEED_TO_1000                 (1 << 23)
-#define SET_FULL_DUPLEX_MODE                   (1 << 21)
-#define MAX_RX_PACKET_9700BYTE                 (5 << 17)
-#define DISABLE_AUTO_NEG_SPEED_GMII            (1 << 13)
-#define DO_NOT_FORCE_LINK_FAIL                 (1 << 10)
-#define SERIAL_PORT_CONTROL_RESERVED           (1 << 9)
-#define DISABLE_AUTO_NEG_FOR_FLOW_CTRL         (1 << 3)
-#define DISABLE_AUTO_NEG_FOR_DUPLEX            (1 << 2)
-#define FORCE_LINK_PASS                                (1 << 1)
-#define SERIAL_PORT_ENABLE                     (1 << 0)
-
-#define DEFAULT_RX_QUEUE_SIZE          400
-#define DEFAULT_TX_QUEUE_SIZE          800
+#define DEFAULT_RX_QUEUE_SIZE  128
+#define DEFAULT_TX_QUEUE_SIZE  256
+#define SKB_DMA_REALIGN                ((PAGE_SIZE - NET_SKB_PAD) % SMP_CACHE_BYTES)
 
 
 /*
@@ -225,6 +230,12 @@ struct tx_desc {
 #define RX_ENABLE_INTERRUPT            0x20000000
 #define RX_FIRST_DESC                  0x08000000
 #define RX_LAST_DESC                   0x04000000
+#define RX_IP_HDR_OK                   0x02000000
+#define RX_PKT_IS_IPV4                 0x01000000
+#define RX_PKT_IS_ETHERNETV2           0x00800000
+#define RX_PKT_LAYER4_TYPE_MASK                0x00600000
+#define RX_PKT_LAYER4_TYPE_TCP_IPV4    0x00000000
+#define RX_PKT_IS_VLAN_TAGGED          0x00080000
 
 /* TX descriptor command */
 #define TX_ENABLE_INTERRUPT            0x00800000
@@ -249,9 +260,23 @@ struct mv643xx_eth_shared_private {
        void __iomem *base;
 
        /*
-        * Protects access to SMI_REG, which is shared between ports.
+        * Points at the right SMI instance to use.
+        */
+       struct mv643xx_eth_shared_private *smi;
+
+       /*
+        * Provides access to local SMI interface.
         */
-       spinlock_t phy_lock;
+       struct mii_bus *smi_bus;
+
+       /*
+        * If we have access to the error interrupt pin (which is
+        * somewhat misnamed as it not only reflects internal errors
+        * but also reflects SMI completion), use that to wait for
+        * SMI access completion instead of polling the SMI busy bit.
+        */
+       int err_interrupt;
+       wait_queue_head_t smi_busy_wait;
 
        /*
         * Per-port MBUS window access register value.
@@ -263,9 +288,16 @@ struct mv643xx_eth_shared_private {
         */
        unsigned int t_clk;
        int extended_rx_coal_limit;
-       int tx_bw_control_moved;
+       int tx_bw_control;
 };
 
+#define TX_BW_CONTROL_ABSENT           0
+#define TX_BW_CONTROL_OLD_LAYOUT       1
+#define TX_BW_CONTROL_NEW_LAYOUT       2
+
+static int mv643xx_eth_open(struct net_device *dev);
+static int mv643xx_eth_stop(struct net_device *dev);
+
 
 /* per-port *****************************************************************/
 struct mib_counters {
@@ -301,6 +333,12 @@ struct mib_counters {
        u32 late_collision;
 };
 
+struct lro_counters {
+       u32 lro_aggregated;
+       u32 lro_flushed;
+       u32 lro_no_desc;
+};
+
 struct rx_queue {
        int index;
 
@@ -315,7 +353,8 @@ struct rx_queue {
        int rx_desc_area_size;
        struct sk_buff **rx_skb;
 
-       struct timer_list rx_oom;
+       struct net_lro_mgr lro_mgr;
+       struct net_lro_desc lro_arr[8];
 };
 
 struct tx_queue {
@@ -330,47 +369,61 @@ struct tx_queue {
        struct tx_desc *tx_desc_area;
        dma_addr_t tx_desc_dma;
        int tx_desc_area_size;
-       struct sk_buff **tx_skb;
+
+       struct sk_buff_head tx_skb;
+
+       unsigned long tx_packets;
+       unsigned long tx_bytes;
+       unsigned long tx_dropped;
 };
 
 struct mv643xx_eth_private {
        struct mv643xx_eth_shared_private *shared;
+       void __iomem *base;
        int port_num;
 
        struct net_device *dev;
 
-       struct mv643xx_eth_shared_private *shared_smi;
-       int phy_addr;
-
-       spinlock_t lock;
+       struct phy_device *phy;
 
+       struct timer_list mib_counters_timer;
+       spinlock_t mib_counters_lock;
        struct mib_counters mib_counters;
+
+       struct lro_counters lro_counters;
+
        struct work_struct tx_timeout_task;
-       struct mii_if_info mii;
+
+       struct napi_struct napi;
+       u32 int_mask;
+       u8 oom;
+       u8 work_link;
+       u8 work_tx;
+       u8 work_tx_end;
+       u8 work_rx;
+       u8 work_rx_refill;
+
+       int skb_size;
+       struct sk_buff_head rx_recycle;
 
        /*
         * RX state.
         */
-       int default_rx_ring_size;
+       int rx_ring_size;
        unsigned long rx_desc_sram_addr;
        int rx_desc_sram_size;
-       u8 rxq_mask;
-       int rxq_primary;
-       struct napi_struct napi;
+       int rxq_count;
+       struct timer_list rx_oom;
        struct rx_queue rxq[8];
 
        /*
         * TX state.
         */
-       int default_tx_ring_size;
+       int tx_ring_size;
        unsigned long tx_desc_sram_addr;
        int tx_desc_sram_size;
-       u8 txq_mask;
-       int txq_primary;
+       int txq_count;
        struct tx_queue txq[8];
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-       int tx_clean_threshold;
-#endif
 };
 
 
@@ -380,11 +433,21 @@ static inline u32 rdl(struct mv643xx_eth_private *mp, int offset)
        return readl(mp->shared->base + offset);
 }
 
+static inline u32 rdlp(struct mv643xx_eth_private *mp, int offset)
+{
+       return readl(mp->base + offset);
+}
+
 static inline void wrl(struct mv643xx_eth_private *mp, int offset, u32 data)
 {
        writel(data, mp->shared->base + offset);
 }
 
+static inline void wrlp(struct mv643xx_eth_private *mp, int offset, u32 data)
+{
+       writel(data, mp->base + offset);
+}
+
 
 /* rxq/txq helper functions *************************************************/
 static struct mv643xx_eth_private *rxq_to_mp(struct rx_queue *rxq)
@@ -400,7 +463,7 @@ static struct mv643xx_eth_private *txq_to_mp(struct tx_queue *txq)
 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 << rxq->index);
+       wrlp(mp, RXQ_COMMAND, 1 << rxq->index);
 }
 
 static void rxq_disable(struct rx_queue *rxq)
@@ -408,26 +471,25 @@ static void rxq_disable(struct rx_queue *rxq)
        struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
        u8 mask = 1 << rxq->index;
 
-       wrl(mp, RXQ_COMMAND(mp->port_num), mask << 8);
-       while (rdl(mp, RXQ_COMMAND(mp->port_num)) & mask)
+       wrlp(mp, RXQ_COMMAND, mask << 8);
+       while (rdlp(mp, RXQ_COMMAND) & mask)
                udelay(10);
 }
 
 static void txq_reset_hw_ptr(struct tx_queue *txq)
 {
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
-       int off = TXQ_CURRENT_DESC_PTR(mp->port_num, txq->index);
        u32 addr;
 
        addr = (u32)txq->tx_desc_dma;
        addr += txq->tx_curr_desc * sizeof(struct tx_desc);
-       wrl(mp, off, addr);
+       wrlp(mp, TXQ_CURRENT_DESC_PTR(txq->index), addr);
 }
 
 static void txq_enable(struct tx_queue *txq)
 {
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
-       wrl(mp, TXQ_COMMAND(mp->port_num), 1 << txq->index);
+       wrlp(mp, TXQ_COMMAND, 1 << txq->index);
 }
 
 static void txq_disable(struct tx_queue *txq)
@@ -435,135 +497,90 @@ static void txq_disable(struct tx_queue *txq)
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
        u8 mask = 1 << txq->index;
 
-       wrl(mp, TXQ_COMMAND(mp->port_num), mask << 8);
-       while (rdl(mp, TXQ_COMMAND(mp->port_num)) & mask)
+       wrlp(mp, TXQ_COMMAND, mask << 8);
+       while (rdlp(mp, TXQ_COMMAND) & mask)
                udelay(10);
 }
 
-static void __txq_maybe_wake(struct tx_queue *txq)
+static void txq_maybe_wake(struct tx_queue *txq)
 {
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
 
-       /*
-        * netif_{stop,wake}_queue() flow control only applies to
-        * the primary queue.
-        */
-       BUG_ON(txq->index != mp->txq_primary);
-
-       if (txq->tx_ring_size - txq->tx_desc_count >= MAX_DESCS_PER_SKB)
-               netif_wake_queue(mp->dev);
+       if (netif_tx_queue_stopped(nq)) {
+               __netif_tx_lock(nq, smp_processor_id());
+               if (txq->tx_ring_size - txq->tx_desc_count >= MAX_SKB_FRAGS + 1)
+                       netif_tx_wake_queue(nq);
+               __netif_tx_unlock(nq);
+       }
 }
 
 
-/* rx ***********************************************************************/
-static void txq_reclaim(struct tx_queue *txq, int force);
-
-static void rxq_refill(struct rx_queue *rxq)
+/* rx napi ******************************************************************/
+static int
+mv643xx_get_skb_header(struct sk_buff *skb, void **iphdr, void **tcph,
+                      u64 *hdr_flags, void *priv)
 {
-       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
-       unsigned long flags;
-
-       spin_lock_irqsave(&mp->lock, flags);
-
-       while (rxq->rx_desc_count < rxq->rx_ring_size) {
-               int skb_size;
-               struct sk_buff *skb;
-               int unaligned;
-               int rx;
-
-               /*
-                * Reserve 2+14 bytes for an ethernet header (the
-                * hardware automatically prepends 2 bytes of dummy
-                * data to each received packet), 16 bytes for up to
-                * four VLAN tags, and 4 bytes for the trailing FCS
-                * -- 36 bytes total.
-                */
-               skb_size = mp->dev->mtu + 36;
-
-               /*
-                * Make sure that the skb size is a multiple of 8
-                * bytes, as the lower three bits of the receive
-                * descriptor's buffer size field are ignored by
-                * the hardware.
-                */
-               skb_size = (skb_size + 7) & ~7;
-
-               skb = dev_alloc_skb(skb_size + dma_get_cache_alignment() - 1);
-               if (skb == NULL)
-                       break;
-
-               unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
-               if (unaligned)
-                       skb_reserve(skb, dma_get_cache_alignment() - unaligned);
-
-               rxq->rx_desc_count++;
-               rx = rxq->rx_used_desc;
-               rxq->rx_used_desc = (rx + 1) % rxq->rx_ring_size;
+       unsigned long cmd_sts = (unsigned long)priv;
 
-               rxq->rx_desc_area[rx].buf_ptr = dma_map_single(NULL, skb->data,
-                                               skb_size, DMA_FROM_DEVICE);
-               rxq->rx_desc_area[rx].buf_size = skb_size;
-               rxq->rx_skb[rx] = skb;
-               wmb();
-               rxq->rx_desc_area[rx].cmd_sts = BUFFER_OWNED_BY_DMA |
-                                               RX_ENABLE_INTERRUPT;
-               wmb();
-
-               /*
-                * The hardware automatically prepends 2 bytes of
-                * dummy data to each received packet, so that the
-                * IP header ends up 16-byte aligned.
-                */
-               skb_reserve(skb, 2);
-       }
-
-       if (rxq->rx_desc_count != rxq->rx_ring_size)
-               mod_timer(&rxq->rx_oom, jiffies + (HZ / 10));
-
-       spin_unlock_irqrestore(&mp->lock, flags);
-}
+       /*
+        * Make sure that this packet is Ethernet II, is not VLAN
+        * tagged, is IPv4, has a valid IP header, and is TCP.
+        */
+       if ((cmd_sts & (RX_IP_HDR_OK | RX_PKT_IS_IPV4 |
+                      RX_PKT_IS_ETHERNETV2 | RX_PKT_LAYER4_TYPE_MASK |
+                      RX_PKT_IS_VLAN_TAGGED)) !=
+           (RX_IP_HDR_OK | RX_PKT_IS_IPV4 |
+            RX_PKT_IS_ETHERNETV2 | RX_PKT_LAYER4_TYPE_TCP_IPV4))
+               return -1;
+
+       skb_reset_network_header(skb);
+       skb_set_transport_header(skb, ip_hdrlen(skb));
+       *iphdr = ip_hdr(skb);
+       *tcph = tcp_hdr(skb);
+       *hdr_flags = LRO_IPV4 | LRO_TCP;
 
-static inline void rxq_refill_timer_wrapper(unsigned long data)
-{
-       rxq_refill((struct rx_queue *)data);
+       return 0;
 }
 
 static int rxq_process(struct rx_queue *rxq, int budget)
 {
        struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
        struct net_device_stats *stats = &mp->dev->stats;
+       int lro_flush_needed;
        int rx;
 
+       lro_flush_needed = 0;
        rx = 0;
        while (rx < budget && rxq->rx_desc_count) {
                struct rx_desc *rx_desc;
                unsigned int cmd_sts;
                struct sk_buff *skb;
-               unsigned long flags;
-
-               spin_lock_irqsave(&mp->lock, flags);
+               u16 byte_cnt;
 
                rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
 
                cmd_sts = rx_desc->cmd_sts;
-               if (cmd_sts & BUFFER_OWNED_BY_DMA) {
-                       spin_unlock_irqrestore(&mp->lock, flags);
+               if (cmd_sts & BUFFER_OWNED_BY_DMA)
                        break;
-               }
                rmb();
 
                skb = rxq->rx_skb[rxq->rx_curr_desc];
                rxq->rx_skb[rxq->rx_curr_desc] = NULL;
 
-               rxq->rx_curr_desc = (rxq->rx_curr_desc + 1) % rxq->rx_ring_size;
-
-               spin_unlock_irqrestore(&mp->lock, flags);
+               rxq->rx_curr_desc++;
+               if (rxq->rx_curr_desc == rxq->rx_ring_size)
+                       rxq->rx_curr_desc = 0;
 
-               dma_unmap_single(NULL, rx_desc->buf_ptr,
+               dma_unmap_single(mp->dev->dev.parent, rx_desc->buf_ptr,
                                 rx_desc->buf_size, DMA_FROM_DEVICE);
                rxq->rx_desc_count--;
                rx++;
 
+               mp->work_rx_refill |= 1 << rxq->index;
+
+               byte_cnt = rx_desc->byte_cnt;
+
                /*
                 * Update statistics.
                 *
@@ -573,95 +590,119 @@ static int rxq_process(struct rx_queue *rxq, int budget)
                 * byte CRC at the end of the packet (which we do count).
                 */
                stats->rx_packets++;
-               stats->rx_bytes += rx_desc->byte_cnt - 2;
+               stats->rx_bytes += byte_cnt - 2;
 
                /*
                 * In case we received a packet without first / last bits
                 * 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())
-                                       dev_printk(KERN_ERR, &mp->dev->dev,
-                                                  "received packet spanning "
-                                                  "multiple descriptors\n");
-                       }
+               if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC | ERROR_SUMMARY))
+                       != (RX_FIRST_DESC | RX_LAST_DESC))
+                       goto err;
 
-                       if (cmd_sts & ERROR_SUMMARY)
-                               stats->rx_errors++;
+               /*
+                * The -4 is for the CRC in the trailer of the
+                * received packet
+                */
+               skb_put(skb, byte_cnt - 2 - 4);
 
-                       dev_kfree_skb_irq(skb);
-               } else {
-                       /*
-                        * The -4 is for the CRC in the trailer of the
-                        * received packet
-                        */
-                       skb_put(skb, rx_desc->byte_cnt - 2 - 4);
-
-                       if (cmd_sts & LAYER_4_CHECKSUM_OK) {
-                               skb->ip_summed = CHECKSUM_UNNECESSARY;
-                               skb->csum = htons(
-                                       (cmd_sts & 0x0007fff8) >> 3);
-                       }
-                       skb->protocol = eth_type_trans(skb, mp->dev);
-#ifdef MV643XX_ETH_NAPI
+               if (cmd_sts & LAYER_4_CHECKSUM_OK)
+                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+               skb->protocol = eth_type_trans(skb, mp->dev);
+
+               if (skb->dev->features & NETIF_F_LRO &&
+                   skb->ip_summed == CHECKSUM_UNNECESSARY) {
+                       lro_receive_skb(&rxq->lro_mgr, skb, (void *)cmd_sts);
+                       lro_flush_needed = 1;
+               } else
                        netif_receive_skb(skb);
-#else
-                       netif_rx(skb);
-#endif
+
+               continue;
+
+err:
+               stats->rx_dropped++;
+
+               if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
+                       (RX_FIRST_DESC | RX_LAST_DESC)) {
+                       if (net_ratelimit())
+                               dev_printk(KERN_ERR, &mp->dev->dev,
+                                          "received packet spanning "
+                                          "multiple descriptors\n");
                }
 
-               mp->dev->last_rx = jiffies;
+               if (cmd_sts & ERROR_SUMMARY)
+                       stats->rx_errors++;
+
+               dev_kfree_skb(skb);
        }
 
-       rxq_refill(rxq);
+       if (lro_flush_needed)
+               lro_flush_all(&rxq->lro_mgr);
+
+       if (rx < budget)
+               mp->work_rx &= ~(1 << rxq->index);
 
        return rx;
 }
 
-#ifdef MV643XX_ETH_NAPI
-static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
+static int rxq_refill(struct rx_queue *rxq, int budget)
 {
-       struct mv643xx_eth_private *mp;
-       int rx;
-       int i;
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+       int refilled;
 
-       mp = container_of(napi, struct mv643xx_eth_private, napi);
+       refilled = 0;
+       while (refilled < budget && rxq->rx_desc_count < rxq->rx_ring_size) {
+               struct sk_buff *skb;
+               int rx;
+               struct rx_desc *rx_desc;
+               int size;
+
+               skb = __skb_dequeue(&mp->rx_recycle);
+               if (skb == NULL)
+                       skb = dev_alloc_skb(mp->skb_size);
 
-#ifdef MV643XX_ETH_TX_FAST_REFILL
-       if (++mp->tx_clean_threshold > 5) {
-               mp->tx_clean_threshold = 0;
-               for (i = 0; i < 8; i++)
-                       if (mp->txq_mask & (1 << i))
-                               txq_reclaim(mp->txq + i, 0);
-
-               if (netif_carrier_ok(mp->dev)) {
-                       spin_lock_irq(&mp->lock);
-                       __txq_maybe_wake(mp->txq + mp->txq_primary);
-                       spin_unlock_irq(&mp->lock);
+               if (skb == NULL) {
+                       mp->oom = 1;
+                       goto oom;
                }
-       }
-#endif
 
-       rx = 0;
-       for (i = 7; rx < budget && i >= 0; i--)
-               if (mp->rxq_mask & (1 << i))
-                       rx += rxq_process(mp->rxq + i, budget - rx);
+               if (SKB_DMA_REALIGN)
+                       skb_reserve(skb, SKB_DMA_REALIGN);
+
+               refilled++;
+               rxq->rx_desc_count++;
 
-       if (rx < budget) {
-               netif_rx_complete(mp->dev, napi);
-               wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
+               rx = rxq->rx_used_desc++;
+               if (rxq->rx_used_desc == rxq->rx_ring_size)
+                       rxq->rx_used_desc = 0;
+
+               rx_desc = rxq->rx_desc_area + rx;
+
+               size = skb->end - skb->data;
+               rx_desc->buf_ptr = dma_map_single(mp->dev->dev.parent,
+                                                 skb->data, size,
+                                                 DMA_FROM_DEVICE);
+               rx_desc->buf_size = size;
+               rxq->rx_skb[rx] = skb;
+               wmb();
+               rx_desc->cmd_sts = BUFFER_OWNED_BY_DMA | RX_ENABLE_INTERRUPT;
+               wmb();
+
+               /*
+                * The hardware automatically prepends 2 bytes of
+                * dummy data to each received packet, so that the
+                * IP header ends up 16-byte aligned.
+                */
+               skb_reserve(skb, 2);
        }
 
-       return rx;
+       if (refilled < budget)
+               mp->work_rx_refill &= ~(1 << rxq->index);
+
+oom:
+       return refilled;
 }
-#endif
 
 
 /* tx ***********************************************************************/
@@ -678,22 +719,9 @@ static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
        return 0;
 }
 
-static int txq_alloc_desc_index(struct tx_queue *txq)
-{
-       int tx_desc_curr;
-
-       BUG_ON(txq->tx_desc_count >= txq->tx_ring_size);
-
-       tx_desc_curr = txq->tx_curr_desc;
-       txq->tx_curr_desc = (tx_desc_curr + 1) % txq->tx_ring_size;
-
-       BUG_ON(txq->tx_curr_desc == txq->tx_used_desc);
-
-       return tx_desc_curr;
-}
-
 static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
        int nr_frags = skb_shinfo(skb)->nr_frags;
        int frag;
 
@@ -703,7 +731,9 @@ static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
                struct tx_desc *desc;
 
                this_frag = &skb_shinfo(skb)->frags[frag];
-               tx_index = txq_alloc_desc_index(txq);
+               tx_index = txq->tx_curr_desc++;
+               if (txq->tx_curr_desc == txq->tx_ring_size)
+                       txq->tx_curr_desc = 0;
                desc = &txq->tx_desc_area[tx_index];
 
                /*
@@ -714,18 +744,16 @@ static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
                        desc->cmd_sts = BUFFER_OWNED_BY_DMA |
                                        ZERO_PADDING | TX_LAST_DESC |
                                        TX_ENABLE_INTERRUPT;
-                       txq->tx_skb[tx_index] = skb;
                } else {
                        desc->cmd_sts = BUFFER_OWNED_BY_DMA;
-                       txq->tx_skb[tx_index] = NULL;
                }
 
                desc->l4i_chk = 0;
                desc->byte_cnt = this_frag->size;
-               desc->buf_ptr = dma_map_page(NULL, this_frag->page,
-                                               this_frag->page_offset,
-                                               this_frag->size,
-                                               DMA_TO_DEVICE);
+               desc->buf_ptr = dma_map_page(mp->dev->dev.parent,
+                                            this_frag->page,
+                                            this_frag->page_offset,
+                                            this_frag->size, DMA_TO_DEVICE);
        }
 }
 
@@ -734,144 +762,231 @@ static inline __be16 sum16_as_be(__sum16 sum)
        return (__force __be16)sum;
 }
 
-static void txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
+static int txq_submit_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
        int nr_frags = skb_shinfo(skb)->nr_frags;
        int tx_index;
        struct tx_desc *desc;
        u32 cmd_sts;
+       u16 l4i_chk;
        int length;
 
        cmd_sts = TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
-
-       tx_index = txq_alloc_desc_index(txq);
-       desc = &txq->tx_desc_area[tx_index];
-
-       if (nr_frags) {
-               txq_submit_frag_skb(txq, skb);
-
-               length = skb_headlen(skb);
-               txq->tx_skb[tx_index] = NULL;
-       } else {
-               cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
-               length = skb->len;
-               txq->tx_skb[tx_index] = skb;
-       }
-
-       desc->byte_cnt = length;
-       desc->buf_ptr = dma_map_single(NULL, skb->data, length, DMA_TO_DEVICE);
+       l4i_chk = 0;
 
        if (skb->ip_summed == CHECKSUM_PARTIAL) {
-               int mac_hdr_len;
+               int tag_bytes;
 
                BUG_ON(skb->protocol != htons(ETH_P_IP) &&
                       skb->protocol != htons(ETH_P_8021Q));
 
-               cmd_sts |= GEN_TCP_UDP_CHECKSUM |
-                          GEN_IP_V4_CHECKSUM   |
-                          ip_hdr(skb)->ihl << TX_IHL_SHIFT;
+               tag_bytes = (void *)ip_hdr(skb) - (void *)skb->data - ETH_HLEN;
+               if (unlikely(tag_bytes & ~12)) {
+                       if (skb_checksum_help(skb) == 0)
+                               goto no_csum;
+                       kfree_skb(skb);
+                       return 1;
+               }
 
-               mac_hdr_len = (void *)ip_hdr(skb) - (void *)skb->data;
-               switch (mac_hdr_len - ETH_HLEN) {
-               case 0:
-                       break;
-               case 4:
-                       cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
-                       break;
-               case 8:
-                       cmd_sts |= MAC_HDR_EXTRA_8_BYTES;
-                       break;
-               case 12:
+               if (tag_bytes & 4)
                        cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
+               if (tag_bytes & 8)
                        cmd_sts |= MAC_HDR_EXTRA_8_BYTES;
-                       break;
-               default:
-                       if (net_ratelimit())
-                               dev_printk(KERN_ERR, &txq_to_mp(txq)->dev->dev,
-                                  "mac header length is %d?!\n", mac_hdr_len);
-                       break;
-               }
+
+               cmd_sts |= GEN_TCP_UDP_CHECKSUM |
+                          GEN_IP_V4_CHECKSUM   |
+                          ip_hdr(skb)->ihl << TX_IHL_SHIFT;
 
                switch (ip_hdr(skb)->protocol) {
                case IPPROTO_UDP:
                        cmd_sts |= UDP_FRAME;
-                       desc->l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
+                       l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
                        break;
                case IPPROTO_TCP:
-                       desc->l4i_chk = ntohs(sum16_as_be(tcp_hdr(skb)->check));
+                       l4i_chk = ntohs(sum16_as_be(tcp_hdr(skb)->check));
                        break;
                default:
                        BUG();
                }
        } else {
+no_csum:
                /* Errata BTS #50, IHL must be 5 if no HW checksum */
                cmd_sts |= 5 << TX_IHL_SHIFT;
-               desc->l4i_chk = 0;
        }
 
+       tx_index = txq->tx_curr_desc++;
+       if (txq->tx_curr_desc == txq->tx_ring_size)
+               txq->tx_curr_desc = 0;
+       desc = &txq->tx_desc_area[tx_index];
+
+       if (nr_frags) {
+               txq_submit_frag_skb(txq, skb);
+               length = skb_headlen(skb);
+       } else {
+               cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
+               length = skb->len;
+       }
+
+       desc->l4i_chk = l4i_chk;
+       desc->byte_cnt = length;
+       desc->buf_ptr = dma_map_single(mp->dev->dev.parent, skb->data,
+                                      length, DMA_TO_DEVICE);
+
+       __skb_queue_tail(&txq->tx_skb, skb);
+
        /* ensure all other descriptors are written before first cmd_sts */
        wmb();
        desc->cmd_sts = cmd_sts;
 
-       /* clear TX_END interrupt status */
-       wrl(mp, INT_CAUSE(mp->port_num), ~(INT_TX_END_0 << txq->index));
-       rdl(mp, INT_CAUSE(mp->port_num));
+       /* clear TX_END status */
+       mp->work_tx_end &= ~(1 << txq->index);
 
        /* ensure all descriptors are written before poking hardware */
        wmb();
        txq_enable(txq);
 
        txq->tx_desc_count += nr_frags + 1;
+
+       return 0;
 }
 
-static int mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
+static netdev_tx_t mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       struct net_device_stats *stats = &dev->stats;
+       int queue;
        struct tx_queue *txq;
-       unsigned long flags;
+       struct netdev_queue *nq;
+
+       queue = skb_get_queue_mapping(skb);
+       txq = mp->txq + queue;
+       nq = netdev_get_tx_queue(dev, queue);
 
        if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
-               stats->tx_dropped++;
+               txq->tx_dropped++;
                dev_printk(KERN_DEBUG, &dev->dev,
                           "failed to linearize skb with tiny "
                           "unaligned fragment\n");
                return NETDEV_TX_BUSY;
        }
 
-       spin_lock_irqsave(&mp->lock, flags);
-
-       txq = mp->txq + mp->txq_primary;
-
-       if (txq->tx_ring_size - txq->tx_desc_count < MAX_DESCS_PER_SKB) {
-               spin_unlock_irqrestore(&mp->lock, flags);
-               if (txq->index == mp->txq_primary && net_ratelimit())
-                       dev_printk(KERN_ERR, &dev->dev,
-                                  "primary tx queue full?!\n");
+       if (txq->tx_ring_size - txq->tx_desc_count < MAX_SKB_FRAGS + 1) {
+               if (net_ratelimit())
+                       dev_printk(KERN_ERR, &dev->dev, "tx queue full?!\n");
                kfree_skb(skb);
                return NETDEV_TX_OK;
        }
 
-       txq_submit_skb(txq, skb);
-       stats->tx_bytes += skb->len;
-       stats->tx_packets++;
-       dev->trans_start = jiffies;
-
-       if (txq->index == mp->txq_primary) {
+       if (!txq_submit_skb(txq, skb)) {
                int entries_left;
 
+               txq->tx_bytes += skb->len;
+               txq->tx_packets++;
+               dev->trans_start = jiffies;
+
                entries_left = txq->tx_ring_size - txq->tx_desc_count;
-               if (entries_left < MAX_DESCS_PER_SKB)
-                       netif_stop_queue(dev);
+               if (entries_left < MAX_SKB_FRAGS + 1)
+                       netif_tx_stop_queue(nq);
        }
 
-       spin_unlock_irqrestore(&mp->lock, flags);
-
        return NETDEV_TX_OK;
 }
 
 
+/* tx napi ******************************************************************/
+static void txq_kick(struct tx_queue *txq)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
+       u32 hw_desc_ptr;
+       u32 expected_ptr;
+
+       __netif_tx_lock(nq, smp_processor_id());
+
+       if (rdlp(mp, TXQ_COMMAND) & (1 << txq->index))
+               goto out;
+
+       hw_desc_ptr = rdlp(mp, TXQ_CURRENT_DESC_PTR(txq->index));
+       expected_ptr = (u32)txq->tx_desc_dma +
+                               txq->tx_curr_desc * sizeof(struct tx_desc);
+
+       if (hw_desc_ptr != expected_ptr)
+               txq_enable(txq);
+
+out:
+       __netif_tx_unlock(nq);
+
+       mp->work_tx_end &= ~(1 << txq->index);
+}
+
+static int txq_reclaim(struct tx_queue *txq, int budget, int force)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       struct netdev_queue *nq = netdev_get_tx_queue(mp->dev, txq->index);
+       int reclaimed;
+
+       __netif_tx_lock(nq, smp_processor_id());
+
+       reclaimed = 0;
+       while (reclaimed < budget && txq->tx_desc_count > 0) {
+               int tx_index;
+               struct tx_desc *desc;
+               u32 cmd_sts;
+               struct sk_buff *skb;
+
+               tx_index = txq->tx_used_desc;
+               desc = &txq->tx_desc_area[tx_index];
+               cmd_sts = desc->cmd_sts;
+
+               if (cmd_sts & BUFFER_OWNED_BY_DMA) {
+                       if (!force)
+                               break;
+                       desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
+               }
+
+               txq->tx_used_desc = tx_index + 1;
+               if (txq->tx_used_desc == txq->tx_ring_size)
+                       txq->tx_used_desc = 0;
+
+               reclaimed++;
+               txq->tx_desc_count--;
+
+               skb = NULL;
+               if (cmd_sts & TX_LAST_DESC)
+                       skb = __skb_dequeue(&txq->tx_skb);
+
+               if (cmd_sts & ERROR_SUMMARY) {
+                       dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
+                       mp->dev->stats.tx_errors++;
+               }
+
+               if (cmd_sts & TX_FIRST_DESC) {
+                       dma_unmap_single(mp->dev->dev.parent, desc->buf_ptr,
+                                        desc->byte_cnt, DMA_TO_DEVICE);
+               } else {
+                       dma_unmap_page(mp->dev->dev.parent, desc->buf_ptr,
+                                      desc->byte_cnt, DMA_TO_DEVICE);
+               }
+
+               if (skb != NULL) {
+                       if (skb_queue_len(&mp->rx_recycle) <
+                                       mp->rx_ring_size &&
+                           skb_recycle_check(skb, mp->skb_size))
+                               __skb_queue_head(&mp->rx_recycle, skb);
+                       else
+                               dev_kfree_skb(skb);
+               }
+       }
+
+       __netif_tx_unlock(nq);
+
+       if (reclaimed < budget)
+               mp->work_tx &= ~(1 << txq->index);
+
+       return reclaimed;
+}
+
+
 /* tx rate control **********************************************************/
 /*
  * Set total maximum TX rate (shared by all TX queues for this port)
@@ -895,14 +1010,17 @@ static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
        if (bucket_size > 65535)
                bucket_size = 65535;
 
-       if (mp->shared->tx_bw_control_moved) {
-               wrl(mp, TX_BW_RATE_MOVED(mp->port_num), token_rate);
-               wrl(mp, TX_BW_MTU_MOVED(mp->port_num), mtu);
-               wrl(mp, TX_BW_BURST_MOVED(mp->port_num), bucket_size);
-       } else {
-               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);
+       switch (mp->shared->tx_bw_control) {
+       case TX_BW_CONTROL_OLD_LAYOUT:
+               wrlp(mp, TX_BW_RATE, token_rate);
+               wrlp(mp, TX_BW_MTU, mtu);
+               wrlp(mp, TX_BW_BURST, bucket_size);
+               break;
+       case TX_BW_CONTROL_NEW_LAYOUT:
+               wrlp(mp, TX_BW_RATE_MOVED, token_rate);
+               wrlp(mp, TX_BW_MTU_MOVED, mtu);
+               wrlp(mp, TX_BW_BURST_MOVED, bucket_size);
+               break;
        }
 }
 
@@ -920,9 +1038,8 @@ static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
        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);
+       wrlp(mp, TXQ_BW_TOKENS(txq->index), token_rate << 14);
+       wrlp(mp, TXQ_BW_CONF(txq->index), (bucket_size << 10) | token_rate);
 }
 
 static void txq_set_fixed_prio_mode(struct tx_queue *txq)
@@ -934,114 +1051,161 @@ static void txq_set_fixed_prio_mode(struct tx_queue *txq)
        /*
         * Turn on fixed priority mode.
         */
-       if (mp->shared->tx_bw_control_moved)
-               off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
-       else
-               off = TXQ_FIX_PRIO_CONF(mp->port_num);
+       off = 0;
+       switch (mp->shared->tx_bw_control) {
+       case TX_BW_CONTROL_OLD_LAYOUT:
+               off = TXQ_FIX_PRIO_CONF;
+               break;
+       case TX_BW_CONTROL_NEW_LAYOUT:
+               off = TXQ_FIX_PRIO_CONF_MOVED;
+               break;
+       }
 
-       val = rdl(mp, off);
-       val |= 1 << txq->index;
-       wrl(mp, off, val);
+       if (off) {
+               val = rdlp(mp, off);
+               val |= 1 << txq->index;
+               wrlp(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);
+/* mii management interface *************************************************/
+static irqreturn_t mv643xx_eth_err_irq(int irq, void *dev_id)
+{
+       struct mv643xx_eth_shared_private *msp = dev_id;
 
-       val = rdl(mp, off);
-       val &= ~(1 << txq->index);
-       wrl(mp, off, val);
+       if (readl(msp->base + ERR_INT_CAUSE) & ERR_INT_SMI_DONE) {
+               writel(~ERR_INT_SMI_DONE, msp->base + ERR_INT_CAUSE);
+               wake_up(&msp->smi_busy_wait);
+               return IRQ_HANDLED;
+       }
 
-       /*
-        * Configure WRR weight for this queue.
-        */
-       off = TXQ_BW_WRR_CONF(mp->port_num, txq->index);
+       return IRQ_NONE;
+}
 
-       val = rdl(mp, off);
-       val = (val & ~0xff) | (weight & 0xff);
-       wrl(mp, off, val);
+static int smi_is_done(struct mv643xx_eth_shared_private *msp)
+{
+       return !(readl(msp->base + SMI_REG) & SMI_BUSY);
 }
 
+static int smi_wait_ready(struct mv643xx_eth_shared_private *msp)
+{
+       if (msp->err_interrupt == NO_IRQ) {
+               int i;
 
-/* mii management interface *************************************************/
-#define SMI_BUSY               0x10000000
-#define SMI_READ_VALID         0x08000000
-#define SMI_OPCODE_READ                0x04000000
-#define SMI_OPCODE_WRITE       0x00000000
+               for (i = 0; !smi_is_done(msp); i++) {
+                       if (i == 10)
+                               return -ETIMEDOUT;
+                       msleep(10);
+               }
 
-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;
-       unsigned long flags;
-       int i;
+               return 0;
+       }
 
-       /* the SMI register is a shared resource */
-       spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
+       if (!smi_is_done(msp)) {
+               wait_event_timeout(msp->smi_busy_wait, smi_is_done(msp),
+                                  msecs_to_jiffies(100));
+               if (!smi_is_done(msp))
+                       return -ETIMEDOUT;
+       }
 
-       /* wait for the SMI register to become available */
-       for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
-               if (i == 1000) {
-                       printk("%s: PHY busy timeout\n", mp->dev->name);
-                       goto out;
-               }
-               udelay(10);
+       return 0;
+}
+
+static int smi_bus_read(struct mii_bus *bus, int addr, int reg)
+{
+       struct mv643xx_eth_shared_private *msp = bus->priv;
+       void __iomem *smi_reg = msp->base + SMI_REG;
+       int ret;
+
+       if (smi_wait_ready(msp)) {
+               printk(KERN_WARNING "mv643xx_eth: SMI bus busy timeout\n");
+               return -ETIMEDOUT;
        }
 
        writel(SMI_OPCODE_READ | (reg << 21) | (addr << 16), smi_reg);
 
-       /* now wait for the data to be valid */
-       for (i = 0; !(readl(smi_reg) & SMI_READ_VALID); i++) {
-               if (i == 1000) {
-                       printk("%s: PHY read timeout\n", mp->dev->name);
-                       goto out;
-               }
-               udelay(10);
+       if (smi_wait_ready(msp)) {
+               printk(KERN_WARNING "mv643xx_eth: SMI bus busy timeout\n");
+               return -ETIMEDOUT;
        }
 
-       *value = readl(smi_reg) & 0xffff;
-out:
-       spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
+       ret = readl(smi_reg);
+       if (!(ret & SMI_READ_VALID)) {
+               printk(KERN_WARNING "mv643xx_eth: SMI bus read not valid\n");
+               return -ENODEV;
+       }
+
+       return ret & 0xffff;
 }
 
-static void smi_reg_write(struct mv643xx_eth_private *mp,
-                         unsigned int addr,
-                         unsigned int reg, unsigned int value)
+static int smi_bus_write(struct mii_bus *bus, int addr, int reg, u16 val)
 {
-       void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
-       unsigned long flags;
+       struct mv643xx_eth_shared_private *msp = bus->priv;
+       void __iomem *smi_reg = msp->base + SMI_REG;
+
+       if (smi_wait_ready(msp)) {
+               printk(KERN_WARNING "mv643xx_eth: SMI bus busy timeout\n");
+               return -ETIMEDOUT;
+       }
+
+       writel(SMI_OPCODE_WRITE | (reg << 21) |
+               (addr << 16) | (val & 0xffff), smi_reg);
+
+       if (smi_wait_ready(msp)) {
+               printk(KERN_WARNING "mv643xx_eth: SMI bus busy timeout\n");
+               return -ETIMEDOUT;
+       }
+
+       return 0;
+}
+
+
+/* statistics ***************************************************************/
+static struct net_device_stats *mv643xx_eth_get_stats(struct net_device *dev)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       struct net_device_stats *stats = &dev->stats;
+       unsigned long tx_packets = 0;
+       unsigned long tx_bytes = 0;
+       unsigned long tx_dropped = 0;
        int i;
 
-       /* the SMI register is a shared resource */
-       spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
+       for (i = 0; i < mp->txq_count; i++) {
+               struct tx_queue *txq = mp->txq + i;
 
-       /* wait for the SMI register to become available */
-       for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
-               if (i == 1000) {
-                       printk("%s: PHY busy timeout\n", mp->dev->name);
-                       goto out;
-               }
-               udelay(10);
+               tx_packets += txq->tx_packets;
+               tx_bytes += txq->tx_bytes;
+               tx_dropped += txq->tx_dropped;
        }
 
-       writel(SMI_OPCODE_WRITE | (reg << 21) |
-               (addr << 16) | (value & 0xffff), smi_reg);
-out:
-       spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
+       stats->tx_packets = tx_packets;
+       stats->tx_bytes = tx_bytes;
+       stats->tx_dropped = tx_dropped;
+
+       return stats;
 }
 
+static void mv643xx_eth_grab_lro_stats(struct mv643xx_eth_private *mp)
+{
+       u32 lro_aggregated = 0;
+       u32 lro_flushed = 0;
+       u32 lro_no_desc = 0;
+       int i;
+
+       for (i = 0; i < mp->rxq_count; i++) {
+               struct rx_queue *rxq = mp->rxq + i;
+
+               lro_aggregated += rxq->lro_mgr.stats.aggregated;
+               lro_flushed += rxq->lro_mgr.stats.flushed;
+               lro_no_desc += rxq->lro_mgr.stats.no_desc;
+       }
+
+       mp->lro_counters.lro_aggregated = lro_aggregated;
+       mp->lro_counters.lro_flushed = lro_flushed;
+       mp->lro_counters.lro_no_desc = lro_no_desc;
+}
 
-/* mib counters *************************************************************/
 static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
 {
        return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
@@ -1059,8 +1223,8 @@ static void mib_counters_update(struct mv643xx_eth_private *mp)
 {
        struct mib_counters *p = &mp->mib_counters;
 
+       spin_lock_bh(&mp->mib_counters_lock);
        p->good_octets_received += mib_read(mp, 0x00);
-       p->good_octets_received += (u64)mib_read(mp, 0x04) << 32;
        p->bad_octets_received += mib_read(mp, 0x08);
        p->internal_mac_transmit_err += mib_read(mp, 0x0c);
        p->good_frames_received += mib_read(mp, 0x10);
@@ -1074,7 +1238,6 @@ static void mib_counters_update(struct mv643xx_eth_private *mp)
        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);
@@ -1091,6 +1254,95 @@ static void mib_counters_update(struct mv643xx_eth_private *mp)
        p->bad_crc_event += mib_read(mp, 0x74);
        p->collision += mib_read(mp, 0x78);
        p->late_collision += mib_read(mp, 0x7c);
+       spin_unlock_bh(&mp->mib_counters_lock);
+
+       mod_timer(&mp->mib_counters_timer, jiffies + 30 * HZ);
+}
+
+static void mib_counters_timer_wrapper(unsigned long _mp)
+{
+       struct mv643xx_eth_private *mp = (void *)_mp;
+
+       mib_counters_update(mp);
+}
+
+
+/* interrupt coalescing *****************************************************/
+/*
+ * Hardware coalescing parameters are set in units of 64 t_clk
+ * cycles.  I.e.:
+ *
+ *     coal_delay_in_usec = 64000000 * register_value / t_clk_rate
+ *
+ *     register_value = coal_delay_in_usec * t_clk_rate / 64000000
+ *
+ * In the ->set*() methods, we round the computed register value
+ * to the nearest integer.
+ */
+static unsigned int get_rx_coal(struct mv643xx_eth_private *mp)
+{
+       u32 val = rdlp(mp, SDMA_CONFIG);
+       u64 temp;
+
+       if (mp->shared->extended_rx_coal_limit)
+               temp = ((val & 0x02000000) >> 10) | ((val & 0x003fff80) >> 7);
+       else
+               temp = (val & 0x003fff00) >> 8;
+
+       temp *= 64000000;
+       do_div(temp, mp->shared->t_clk);
+
+       return (unsigned int)temp;
+}
+
+static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
+{
+       u64 temp;
+       u32 val;
+
+       temp = (u64)usec * mp->shared->t_clk;
+       temp += 31999999;
+       do_div(temp, 64000000);
+
+       val = rdlp(mp, SDMA_CONFIG);
+       if (mp->shared->extended_rx_coal_limit) {
+               if (temp > 0xffff)
+                       temp = 0xffff;
+               val &= ~0x023fff80;
+               val |= (temp & 0x8000) << 10;
+               val |= (temp & 0x7fff) << 7;
+       } else {
+               if (temp > 0x3fff)
+                       temp = 0x3fff;
+               val &= ~0x003fff00;
+               val |= (temp & 0x3fff) << 8;
+       }
+       wrlp(mp, SDMA_CONFIG, val);
+}
+
+static unsigned int get_tx_coal(struct mv643xx_eth_private *mp)
+{
+       u64 temp;
+
+       temp = (rdlp(mp, TX_FIFO_URGENT_THRESHOLD) & 0x3fff0) >> 4;
+       temp *= 64000000;
+       do_div(temp, mp->shared->t_clk);
+
+       return (unsigned int)temp;
+}
+
+static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int usec)
+{
+       u64 temp;
+
+       temp = (u64)usec * mp->shared->t_clk;
+       temp += 31999999;
+       do_div(temp, 64000000);
+
+       if (temp > 0x3fff)
+               temp = 0x3fff;
+
+       wrlp(mp, TX_FIFO_URGENT_THRESHOLD, temp << 4);
 }
 
 
@@ -1110,6 +1362,10 @@ struct mv643xx_eth_stats {
        { #m, FIELD_SIZEOF(struct mib_counters, m),             \
          -1, offsetof(struct mv643xx_eth_private, mib_counters.m) }
 
+#define LROSTAT(m)                                             \
+       { #m, FIELD_SIZEOF(struct lro_counters, m),             \
+         -1, offsetof(struct mv643xx_eth_private, lro_counters.m) }
+
 static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
        SSTAT(rx_packets),
        SSTAT(tx_packets),
@@ -1149,16 +1405,20 @@ static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
        MIBSTAT(bad_crc_event),
        MIBSTAT(collision),
        MIBSTAT(late_collision),
+       LROSTAT(lro_aggregated),
+       LROSTAT(lro_flushed),
+       LROSTAT(lro_no_desc),
 };
 
-static int mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+static int
+mv643xx_eth_get_settings_phy(struct mv643xx_eth_private *mp,
+                            struct ethtool_cmd *cmd)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
        int err;
 
-       spin_lock_irq(&mp->lock);
-       err = mii_ethtool_gset(&mp->mii, cmd);
-       spin_unlock_irq(&mp->lock);
+       err = phy_read_status(mp->phy);
+       if (err == 0)
+               err = phy_ethtool_gset(mp->phy, cmd);
 
        /*
         * The MAC does not support 1000baseT_Half.
@@ -1169,12 +1429,13 @@ static int mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *
        return err;
 }
 
-static int mv643xx_eth_get_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
+static int
+mv643xx_eth_get_settings_phyless(struct mv643xx_eth_private *mp,
+                                struct ethtool_cmd *cmd)
 {
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
        u32 port_status;
 
-       port_status = rdl(mp, PORT_STATUS(mp->port_num));
+       port_status = rdlp(mp, PORT_STATUS);
 
        cmd->supported = SUPPORTED_MII;
        cmd->advertising = ADVERTISED_MII;
@@ -1203,26 +1464,31 @@ static int mv643xx_eth_get_settings_phyless(struct net_device *dev, struct ethto
        return 0;
 }
 
-static int mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+static int
+mv643xx_eth_get_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       int err;
+
+       if (mp->phy != NULL)
+               return mv643xx_eth_get_settings_phy(mp, cmd);
+       else
+               return mv643xx_eth_get_settings_phyless(mp, cmd);
+}
+
+static int
+mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       if (mp->phy == NULL)
+               return -EINVAL;
 
        /*
         * The MAC does not support 1000baseT_Half.
         */
        cmd->advertising &= ~ADVERTISED_1000baseT_Half;
 
-       spin_lock_irq(&mp->lock);
-       err = mii_ethtool_sset(&mp->mii, cmd);
-       spin_unlock_irq(&mp->lock);
-
-       return err;
-}
-
-static int mv643xx_eth_set_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
-{
-       return -EINVAL;
+       return phy_ethtool_sset(mp->phy, cmd);
 }
 
 static void mv643xx_eth_get_drvinfo(struct net_device *dev,
@@ -1239,24 +1505,95 @@ static int mv643xx_eth_nway_reset(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       return mii_nway_restart(&mp->mii);
+       if (mp->phy == NULL)
+               return -EINVAL;
+
+       return genphy_restart_aneg(mp->phy);
 }
 
-static int mv643xx_eth_nway_reset_phyless(struct net_device *dev)
+static u32 mv643xx_eth_get_link(struct net_device *dev)
 {
-       return -EINVAL;
+       return !!netif_carrier_ok(dev);
 }
 
-static u32 mv643xx_eth_get_link(struct net_device *dev)
+static int
+mv643xx_eth_get_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       return mii_link_ok(&mp->mii);
+       ec->rx_coalesce_usecs = get_rx_coal(mp);
+       ec->tx_coalesce_usecs = get_tx_coal(mp);
+
+       return 0;
 }
 
-static u32 mv643xx_eth_get_link_phyless(struct net_device *dev)
+static int
+mv643xx_eth_set_coalesce(struct net_device *dev, struct ethtool_coalesce *ec)
 {
-       return 1;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       set_rx_coal(mp, ec->rx_coalesce_usecs);
+       set_tx_coal(mp, ec->tx_coalesce_usecs);
+
+       return 0;
+}
+
+static void
+mv643xx_eth_get_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       er->rx_max_pending = 4096;
+       er->tx_max_pending = 4096;
+       er->rx_mini_max_pending = 0;
+       er->rx_jumbo_max_pending = 0;
+
+       er->rx_pending = mp->rx_ring_size;
+       er->tx_pending = mp->tx_ring_size;
+       er->rx_mini_pending = 0;
+       er->rx_jumbo_pending = 0;
+}
+
+static int
+mv643xx_eth_set_ringparam(struct net_device *dev, struct ethtool_ringparam *er)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       if (er->rx_mini_pending || er->rx_jumbo_pending)
+               return -EINVAL;
+
+       mp->rx_ring_size = er->rx_pending < 4096 ? er->rx_pending : 4096;
+       mp->tx_ring_size = er->tx_pending < 4096 ? er->tx_pending : 4096;
+
+       if (netif_running(dev)) {
+               mv643xx_eth_stop(dev);
+               if (mv643xx_eth_open(dev)) {
+                       dev_printk(KERN_ERR, &dev->dev,
+                                  "fatal error on re-opening device after "
+                                  "ring param change\n");
+                       return -ENOMEM;
+               }
+       }
+
+       return 0;
+}
+
+static u32
+mv643xx_eth_get_rx_csum(struct net_device *dev)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       return !!(rdlp(mp, PORT_CONFIG) & 0x02000000);
+}
+
+static int
+mv643xx_eth_set_rx_csum(struct net_device *dev, u32 rx_csum)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       wrlp(mp, PORT_CONFIG, rx_csum ? 0x02000000 : 0x00000000);
+
+       return 0;
 }
 
 static void mv643xx_eth_get_strings(struct net_device *dev,
@@ -1280,7 +1617,9 @@ static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
        struct mv643xx_eth_private *mp = netdev_priv(dev);
        int i;
 
+       mv643xx_eth_get_stats(dev);
        mib_counters_update(mp);
+       mv643xx_eth_grab_lro_stats(mp);
 
        for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
                const struct mv643xx_eth_stats *stat;
@@ -1312,21 +1651,18 @@ static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
        .get_drvinfo            = mv643xx_eth_get_drvinfo,
        .nway_reset             = mv643xx_eth_nway_reset,
        .get_link               = mv643xx_eth_get_link,
+       .get_coalesce           = mv643xx_eth_get_coalesce,
+       .set_coalesce           = mv643xx_eth_set_coalesce,
+       .get_ringparam          = mv643xx_eth_get_ringparam,
+       .set_ringparam          = mv643xx_eth_set_ringparam,
+       .get_rx_csum            = mv643xx_eth_get_rx_csum,
+       .set_rx_csum            = mv643xx_eth_set_rx_csum,
+       .set_tx_csum            = ethtool_op_set_tx_csum,
        .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_flags              = ethtool_op_get_flags,
+       .set_flags              = ethtool_op_set_flags,
        .get_sset_count         = mv643xx_eth_get_sset_count,
 };
 
@@ -1334,11 +1670,8 @@ static const struct ethtool_ops mv643xx_eth_ethtool_ops_phyless = {
 /* address handling *********************************************************/
 static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       unsigned int mac_h;
-       unsigned int mac_l;
-
-       mac_h = rdl(mp, MAC_ADDR_HIGH(mp->port_num));
-       mac_l = rdl(mp, MAC_ADDR_LOW(mp->port_num));
+       unsigned int mac_h = rdlp(mp, MAC_ADDR_HIGH);
+       unsigned int mac_l = rdlp(mp, MAC_ADDR_LOW);
 
        addr[0] = (mac_h >> 24) & 0xff;
        addr[1] = (mac_h >> 16) & 0xff;
@@ -1348,57 +1681,70 @@ static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
        addr[5] = mac_l & 0xff;
 }
 
-static void init_mac_tables(struct mv643xx_eth_private *mp)
+static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       int i;
-
-       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);
+       wrlp(mp, MAC_ADDR_HIGH,
+               (addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3]);
+       wrlp(mp, MAC_ADDR_LOW, (addr[4] << 8) | addr[5]);
 }
 
-static void set_filter_table_entry(struct mv643xx_eth_private *mp,
-                                  int table, unsigned char entry)
+static u32 uc_addr_filter_mask(struct net_device *dev)
 {
-       unsigned int table_reg;
-
-       /* Set "accepts frame bit" at specified table entry */
-       table_reg = rdl(mp, table + (entry & 0xfc));
-       table_reg |= 0x01 << (8 * (entry & 3));
-       wrl(mp, table + (entry & 0xfc), table_reg);
-}
+       struct netdev_hw_addr *ha;
+       u32 nibbles;
 
-static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
-{
-       unsigned int mac_h;
-       unsigned int mac_l;
-       int table;
+       if (dev->flags & IFF_PROMISC)
+               return 0;
 
-       mac_l = (addr[4] << 8) | addr[5];
-       mac_h = (addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3];
+       nibbles = 1 << (dev->dev_addr[5] & 0x0f);
+       list_for_each_entry(ha, &dev->uc.list, list) {
+               if (memcmp(dev->dev_addr, ha->addr, 5))
+                       return 0;
+               if ((dev->dev_addr[5] ^ ha->addr[5]) & 0xf0)
+                       return 0;
 
-       wrl(mp, MAC_ADDR_LOW(mp->port_num), mac_l);
-       wrl(mp, MAC_ADDR_HIGH(mp->port_num), mac_h);
+               nibbles |= 1 << (ha->addr[5] & 0x0f);
+       }
 
-       table = UNICAST_TABLE(mp->port_num);
-       set_filter_table_entry(mp, table, addr[5] & 0x0f);
+       return nibbles;
 }
 
-static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
+static void mv643xx_eth_program_unicast_filter(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
+       u32 port_config;
+       u32 nibbles;
+       int i;
 
-       /* +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);
 
-       return 0;
+       port_config = rdlp(mp, PORT_CONFIG) & ~UNICAST_PROMISCUOUS_MODE;
+
+       nibbles = uc_addr_filter_mask(dev);
+       if (!nibbles) {
+               port_config |= UNICAST_PROMISCUOUS_MODE;
+               nibbles = 0xffff;
+       }
+
+       for (i = 0; i < 16; i += 4) {
+               int off = UNICAST_TABLE(mp->port_num) + i;
+               u32 v;
+
+               v = 0;
+               if (nibbles & 1)
+                       v |= 0x00000001;
+               if (nibbles & 2)
+                       v |= 0x00000100;
+               if (nibbles & 4)
+                       v |= 0x00010000;
+               if (nibbles & 8)
+                       v |= 0x01000000;
+               nibbles >>= 4;
+
+               wrl(mp, off, v);
+       }
+
+       wrlp(mp, PORT_CONFIG, port_config);
 }
 
 static int addr_crc(unsigned char *addr)
@@ -1419,24 +1765,21 @@ static int addr_crc(unsigned char *addr)
        return crc;
 }
 
-static void mv643xx_eth_set_rx_mode(struct net_device *dev)
+static void mv643xx_eth_program_multicast_filter(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
-       u32 port_config;
+       u32 *mc_spec;
+       u32 *mc_other;
        struct dev_addr_list *addr;
        int i;
 
-       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);
-
        if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
-               int port_num = mp->port_num;
-               u32 accept = 0x01010101;
+               int port_num;
+               u32 accept;
 
+oom:
+               port_num = mp->port_num;
+               accept = 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);
@@ -1444,28 +1787,58 @@ static void mv643xx_eth_set_rx_mode(struct net_device *dev)
                return;
        }
 
-       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);
-       }
+       mc_spec = kmalloc(0x200, GFP_ATOMIC);
+       if (mc_spec == NULL)
+               goto oom;
+       mc_other = mc_spec + (0x100 >> 2);
+
+       memset(mc_spec, 0, 0x100);
+       memset(mc_other, 0, 0x100);
 
        for (addr = dev->mc_list; addr != NULL; addr = addr->next) {
                u8 *a = addr->da_addr;
-               int table;
-
-               if (addr->da_addrlen != 6)
-                       continue;
+               u32 *table;
+               int entry;
 
                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]);
+                       table = mc_spec;
+                       entry = a[5];
                } else {
-                       int crc = addr_crc(a);
-
-                       table = OTHER_MCAST_TABLE(mp->port_num);
-                       set_filter_table_entry(mp, table, crc);
+                       table = mc_other;
+                       entry = addr_crc(a);
                }
+
+               table[entry >> 2] |= 1 << (8 * (entry & 3));
+       }
+
+       for (i = 0; i < 0x100; i += 4) {
+               wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, mc_spec[i >> 2]);
+               wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, mc_other[i >> 2]);
        }
+
+       kfree(mc_spec);
+}
+
+static void mv643xx_eth_set_rx_mode(struct net_device *dev)
+{
+       mv643xx_eth_program_unicast_filter(dev);
+       mv643xx_eth_program_multicast_filter(dev);
+}
+
+static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
+{
+       struct sockaddr *sa = addr;
+
+       if (!is_valid_ether_addr(sa->sa_data))
+               return -EINVAL;
+
+       memcpy(dev->dev_addr, sa->sa_data, ETH_ALEN);
+
+       netif_addr_lock_bh(dev);
+       mv643xx_eth_program_unicast_filter(dev);
+       netif_addr_unlock_bh(dev);
+
+       return 0;
 }
 
 
@@ -1479,7 +1852,7 @@ static int rxq_init(struct mv643xx_eth_private *mp, int index)
 
        rxq->index = index;
 
-       rxq->rx_ring_size = mp->default_rx_ring_size;
+       rxq->rx_ring_size = mp->rx_ring_size;
 
        rxq->rx_desc_count = 0;
        rxq->rx_curr_desc = 0;
@@ -1487,14 +1860,14 @@ static int rxq_init(struct mv643xx_eth_private *mp, int index)
 
        size = rxq->rx_ring_size * sizeof(struct rx_desc);
 
-       if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size) {
+       if (index == 0 && size <= mp->rx_desc_sram_size) {
                rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
                                                mp->rx_desc_sram_size);
                rxq->rx_desc_dma = mp->rx_desc_sram_addr;
        } else {
-               rxq->rx_desc_area = dma_alloc_coherent(NULL, size,
-                                                       &rxq->rx_desc_dma,
-                                                       GFP_KERNEL);
+               rxq->rx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
+                                                      size, &rxq->rx_desc_dma,
+                                                      GFP_KERNEL);
        }
 
        if (rxq->rx_desc_area == NULL) {
@@ -1515,23 +1888,37 @@ static int rxq_init(struct mv643xx_eth_private *mp, int index)
 
        rx_desc = (struct rx_desc *)rxq->rx_desc_area;
        for (i = 0; i < rxq->rx_ring_size; i++) {
-               int nexti = (i + 1) % rxq->rx_ring_size;
+               int nexti;
+
+               nexti = i + 1;
+               if (nexti == rxq->rx_ring_size)
+                       nexti = 0;
+
                rx_desc[i].next_desc_ptr = rxq->rx_desc_dma +
                                        nexti * sizeof(struct rx_desc);
        }
 
-       init_timer(&rxq->rx_oom);
-       rxq->rx_oom.data = (unsigned long)rxq;
-       rxq->rx_oom.function = rxq_refill_timer_wrapper;
+       rxq->lro_mgr.dev = mp->dev;
+       memset(&rxq->lro_mgr.stats, 0, sizeof(rxq->lro_mgr.stats));
+       rxq->lro_mgr.features = LRO_F_NAPI;
+       rxq->lro_mgr.ip_summed = CHECKSUM_UNNECESSARY;
+       rxq->lro_mgr.ip_summed_aggr = CHECKSUM_UNNECESSARY;
+       rxq->lro_mgr.max_desc = ARRAY_SIZE(rxq->lro_arr);
+       rxq->lro_mgr.max_aggr = 32;
+       rxq->lro_mgr.frag_align_pad = 0;
+       rxq->lro_mgr.lro_arr = rxq->lro_arr;
+       rxq->lro_mgr.get_skb_header = mv643xx_get_skb_header;
+
+       memset(&rxq->lro_arr, 0, sizeof(rxq->lro_arr));
 
        return 0;
 
 
 out_free:
-       if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size)
+       if (index == 0 && size <= mp->rx_desc_sram_size)
                iounmap(rxq->rx_desc_area);
        else
-               dma_free_coherent(NULL, size,
+               dma_free_coherent(mp->dev->dev.parent, size,
                                  rxq->rx_desc_area,
                                  rxq->rx_desc_dma);
 
@@ -1546,8 +1933,6 @@ static void rxq_deinit(struct rx_queue *rxq)
 
        rxq_disable(rxq);
 
-       del_timer_sync(&rxq->rx_oom);
-
        for (i = 0; i < rxq->rx_ring_size; i++) {
                if (rxq->rx_skb[i]) {
                        dev_kfree_skb(rxq->rx_skb[i]);
@@ -1561,11 +1946,11 @@ static void rxq_deinit(struct rx_queue *rxq)
                           rxq->rx_desc_count);
        }
 
-       if (rxq->index == mp->rxq_primary &&
+       if (rxq->index == 0 &&
            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,
+               dma_free_coherent(mp->dev->dev.parent, rxq->rx_desc_area_size,
                                  rxq->rx_desc_area, rxq->rx_desc_dma);
 
        kfree(rxq->rx_skb);
@@ -1580,7 +1965,7 @@ static int txq_init(struct mv643xx_eth_private *mp, int index)
 
        txq->index = index;
 
-       txq->tx_ring_size = mp->default_tx_ring_size;
+       txq->tx_ring_size = mp->tx_ring_size;
 
        txq->tx_desc_count = 0;
        txq->tx_curr_desc = 0;
@@ -1588,133 +1973,110 @@ static int txq_init(struct mv643xx_eth_private *mp, int index)
 
        size = txq->tx_ring_size * sizeof(struct tx_desc);
 
-       if (index == mp->txq_primary && size <= mp->tx_desc_sram_size) {
+       if (index == 0 && 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);
+               txq->tx_desc_area = dma_alloc_coherent(mp->dev->dev.parent,
+                                                      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;
+               return -ENOMEM;
        }
        memset(txq->tx_desc_area, 0, size);
 
        txq->tx_desc_area_size = size;
-       txq->tx_skb = kmalloc(txq->tx_ring_size * sizeof(*txq->tx_skb),
-                                                               GFP_KERNEL);
-       if (txq->tx_skb == NULL) {
-               dev_printk(KERN_ERR, &mp->dev->dev,
-                          "can't allocate tx skb ring\n");
-               goto out_free;
-       }
 
        tx_desc = (struct tx_desc *)txq->tx_desc_area;
        for (i = 0; i < txq->tx_ring_size; i++) {
                struct tx_desc *txd = tx_desc + i;
-               int nexti = (i + 1) % txq->tx_ring_size;
+               int nexti;
+
+               nexti = i + 1;
+               if (nexti == txq->tx_ring_size)
+                       nexti = 0;
 
                txd->cmd_sts = 0;
                txd->next_desc_ptr = txq->tx_desc_dma +
                                        nexti * sizeof(struct tx_desc);
        }
 
-       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);
+       skb_queue_head_init(&txq->tx_skb);
 
-out:
-       return -ENOMEM;
+       return 0;
 }
 
-static void txq_reclaim(struct tx_queue *txq, int force)
+static void txq_deinit(struct tx_queue *txq)
 {
        struct mv643xx_eth_private *mp = txq_to_mp(txq);
-       unsigned long flags;
-
-       spin_lock_irqsave(&mp->lock, flags);
-       while (txq->tx_desc_count > 0) {
-               int tx_index;
-               struct tx_desc *desc;
-               u32 cmd_sts;
-               struct sk_buff *skb;
-               dma_addr_t addr;
-               int count;
 
-               tx_index = txq->tx_used_desc;
-               desc = &txq->tx_desc_area[tx_index];
-               cmd_sts = desc->cmd_sts;
+       txq_disable(txq);
+       txq_reclaim(txq, txq->tx_ring_size, 1);
 
-               if (cmd_sts & BUFFER_OWNED_BY_DMA) {
-                       if (!force)
-                               break;
-                       desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
-               }
+       BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
 
-               txq->tx_used_desc = (tx_index + 1) % txq->tx_ring_size;
-               txq->tx_desc_count--;
+       if (txq->index == 0 &&
+           txq->tx_desc_area_size <= mp->tx_desc_sram_size)
+               iounmap(txq->tx_desc_area);
+       else
+               dma_free_coherent(mp->dev->dev.parent, txq->tx_desc_area_size,
+                                 txq->tx_desc_area, txq->tx_desc_dma);
+}
 
-               addr = desc->buf_ptr;
-               count = desc->byte_cnt;
-               skb = txq->tx_skb[tx_index];
-               txq->tx_skb[tx_index] = NULL;
 
-               if (cmd_sts & ERROR_SUMMARY) {
-                       dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
-                       mp->dev->stats.tx_errors++;
-               }
+/* netdev ops and related ***************************************************/
+static int mv643xx_eth_collect_events(struct mv643xx_eth_private *mp)
+{
+       u32 int_cause;
+       u32 int_cause_ext;
 
-               /*
-                * Drop mp->lock while we free the skb.
-                */
-               spin_unlock_irqrestore(&mp->lock, flags);
+       int_cause = rdlp(mp, INT_CAUSE) & mp->int_mask;
+       if (int_cause == 0)
+               return 0;
 
-               if (cmd_sts & TX_FIRST_DESC)
-                       dma_unmap_single(NULL, addr, count, DMA_TO_DEVICE);
-               else
-                       dma_unmap_page(NULL, addr, count, DMA_TO_DEVICE);
+       int_cause_ext = 0;
+       if (int_cause & INT_EXT) {
+               int_cause &= ~INT_EXT;
+               int_cause_ext = rdlp(mp, INT_CAUSE_EXT);
+       }
 
-               if (skb)
-                       dev_kfree_skb_irq(skb);
+       if (int_cause) {
+               wrlp(mp, INT_CAUSE, ~int_cause);
+               mp->work_tx_end |= ((int_cause & INT_TX_END) >> 19) &
+                               ~(rdlp(mp, TXQ_COMMAND) & 0xff);
+               mp->work_rx |= (int_cause & INT_RX) >> 2;
+       }
 
-               spin_lock_irqsave(&mp->lock, flags);
+       int_cause_ext &= INT_EXT_LINK_PHY | INT_EXT_TX;
+       if (int_cause_ext) {
+               wrlp(mp, INT_CAUSE_EXT, ~int_cause_ext);
+               if (int_cause_ext & INT_EXT_LINK_PHY)
+                       mp->work_link = 1;
+               mp->work_tx |= int_cause_ext & INT_EXT_TX;
        }
-       spin_unlock_irqrestore(&mp->lock, flags);
+
+       return 1;
 }
 
-static void txq_deinit(struct tx_queue *txq)
+static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
 {
-       struct mv643xx_eth_private *mp = txq_to_mp(txq);
-
-       txq_disable(txq);
-       txq_reclaim(txq, 1);
+       struct net_device *dev = (struct net_device *)dev_id;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
+       if (unlikely(!mv643xx_eth_collect_events(mp)))
+               return IRQ_NONE;
 
-       if (txq->index == mp->txq_primary &&
-           txq->tx_desc_area_size <= mp->tx_desc_sram_size)
-               iounmap(txq->tx_desc_area);
-       else
-               dma_free_coherent(NULL, txq->tx_desc_area_size,
-                                 txq->tx_desc_area, txq->tx_desc_dma);
+       wrlp(mp, INT_MASK, 0);
+       napi_schedule(&mp->napi);
 
-       kfree(txq->tx_skb);
+       return IRQ_HANDLED;
 }
 
-
-/* netdev ops and related ***************************************************/
 static void handle_link_event(struct mv643xx_eth_private *mp)
 {
        struct net_device *dev = mp->dev;
@@ -1723,7 +2085,7 @@ static void handle_link_event(struct mv643xx_eth_private *mp)
        int duplex;
        int fc;
 
-       port_status = rdl(mp, PORT_STATUS(mp->port_num));
+       port_status = rdlp(mp, PORT_STATUS);
        if (!(port_status & LINK_UP)) {
                if (netif_carrier_ok(dev)) {
                        int i;
@@ -1731,15 +2093,12 @@ static void handle_link_event(struct mv643xx_eth_private *mp)
                        printk(KERN_INFO "%s: link down\n", dev->name);
 
                        netif_carrier_off(dev);
-                       netif_stop_queue(dev);
 
-                       for (i = 0; i < 8; i++) {
+                       for (i = 0; i < mp->txq_count; i++) {
                                struct tx_queue *txq = mp->txq + i;
 
-                               if (mp->txq_mask & (1 << i)) {
-                                       txq_reclaim(txq, 1);
-                                       txq_reset_hw_ptr(txq);
-                               }
+                               txq_reclaim(txq, txq->tx_ring_size, 1);
+                               txq_reset_hw_ptr(txq);
                        }
                }
                return;
@@ -1767,119 +2126,98 @@ static void handle_link_event(struct mv643xx_eth_private *mp)
                         speed, duplex ? "full" : "half",
                         fc ? "en" : "dis");
 
-       if (!netif_carrier_ok(dev)) {
+       if (!netif_carrier_ok(dev))
                netif_carrier_on(dev);
-               netif_wake_queue(dev);
-       }
 }
 
-static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
+static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
 {
-       struct net_device *dev = (struct net_device *)dev_id;
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       u32 int_cause;
-       u32 int_cause_ext;
+       struct mv643xx_eth_private *mp;
+       int work_done;
 
-       int_cause = rdl(mp, INT_CAUSE(mp->port_num)) &
-                       (INT_TX_END | INT_RX | INT_EXT);
-       if (int_cause == 0)
-               return IRQ_NONE;
+       mp = container_of(napi, struct mv643xx_eth_private, napi);
 
-       int_cause_ext = 0;
-       if (int_cause & INT_EXT) {
-               int_cause_ext = rdl(mp, INT_CAUSE_EXT(mp->port_num))
-                               & (INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
-               wrl(mp, INT_CAUSE_EXT(mp->port_num), ~int_cause_ext);
+       if (unlikely(mp->oom)) {
+               mp->oom = 0;
+               del_timer(&mp->rx_oom);
        }
 
-       if (int_cause_ext & (INT_EXT_PHY | INT_EXT_LINK))
-               handle_link_event(mp);
-
-       /*
-        * RxBuffer or RxError set for any of the 8 queues?
-        */
-#ifdef MV643XX_ETH_NAPI
-       if (int_cause & INT_RX) {
-               wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_RX));
-               wrl(mp, INT_MASK(mp->port_num), 0x00000000);
-               rdl(mp, INT_MASK(mp->port_num));
-
-               netif_rx_schedule(dev, &mp->napi);
-       }
-#else
-       if (int_cause & INT_RX) {
-               int i;
+       work_done = 0;
+       while (work_done < budget) {
+               u8 queue_mask;
+               int queue;
+               int work_tbd;
 
-               for (i = 7; i >= 0; i--)
-                       if (mp->rxq_mask & (1 << i))
-                               rxq_process(mp->rxq + i, INT_MAX);
-       }
-#endif
+               if (mp->work_link) {
+                       mp->work_link = 0;
+                       handle_link_event(mp);
+                       work_done++;
+                       continue;
+               }
 
-       /*
-        * TxBuffer or TxError set for any of the 8 queues?
-        */
-       if (int_cause_ext & INT_EXT_TX) {
-               int i;
+               queue_mask = mp->work_tx | mp->work_tx_end | mp->work_rx;
+               if (likely(!mp->oom))
+                       queue_mask |= mp->work_rx_refill;
 
-               for (i = 0; i < 8; i++)
-                       if (mp->txq_mask & (1 << i))
-                               txq_reclaim(mp->txq + i, 0);
+               if (!queue_mask) {
+                       if (mv643xx_eth_collect_events(mp))
+                               continue;
+                       break;
+               }
 
-               /*
-                * Enough space again in the primary TX queue for a
-                * full packet?
-                */
-               if (netif_carrier_ok(dev)) {
-                       spin_lock(&mp->lock);
-                       __txq_maybe_wake(mp->txq + mp->txq_primary);
-                       spin_unlock(&mp->lock);
+               queue = fls(queue_mask) - 1;
+               queue_mask = 1 << queue;
+
+               work_tbd = budget - work_done;
+               if (work_tbd > 16)
+                       work_tbd = 16;
+
+               if (mp->work_tx_end & queue_mask) {
+                       txq_kick(mp->txq + queue);
+               } else if (mp->work_tx & queue_mask) {
+                       work_done += txq_reclaim(mp->txq + queue, work_tbd, 0);
+                       txq_maybe_wake(mp->txq + queue);
+               } else if (mp->work_rx & queue_mask) {
+                       work_done += rxq_process(mp->rxq + queue, work_tbd);
+               } else if (!mp->oom && (mp->work_rx_refill & queue_mask)) {
+                       work_done += rxq_refill(mp->rxq + queue, work_tbd);
+               } else {
+                       BUG();
                }
        }
 
-       /*
-        * Any TxEnd interrupts?
-        */
-       if (int_cause & INT_TX_END) {
-               int i;
-
-               wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_TX_END));
-
-               spin_lock(&mp->lock);
-               for (i = 0; i < 8; i++) {
-                       struct tx_queue *txq = mp->txq + i;
-                       u32 hw_desc_ptr;
-                       u32 expected_ptr;
-
-                       if ((int_cause & (INT_TX_END_0 << i)) == 0)
-                               continue;
+       if (work_done < budget) {
+               if (mp->oom)
+                       mod_timer(&mp->rx_oom, jiffies + (HZ / 10));
+               napi_complete(napi);
+               wrlp(mp, INT_MASK, mp->int_mask);
+       }
 
-                       hw_desc_ptr =
-                               rdl(mp, TXQ_CURRENT_DESC_PTR(mp->port_num, i));
-                       expected_ptr = (u32)txq->tx_desc_dma +
-                               txq->tx_curr_desc * sizeof(struct tx_desc);
+       return work_done;
+}
 
-                       if (hw_desc_ptr != expected_ptr)
-                               txq_enable(txq);
-               }
-               spin_unlock(&mp->lock);
-       }
+static inline void oom_timer_wrapper(unsigned long data)
+{
+       struct mv643xx_eth_private *mp = (void *)data;
 
-       return IRQ_HANDLED;
+       napi_schedule(&mp->napi);
 }
 
 static void phy_reset(struct mv643xx_eth_private *mp)
 {
-       unsigned int data;
+       int data;
+
+       data = phy_read(mp->phy, MII_BMCR);
+       if (data < 0)
+               return;
 
-       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
        data |= BMCR_RESET;
-       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data);
+       if (phy_write(mp->phy, MII_BMCR, data) < 0)
+               return;
 
        do {
-               udelay(1);
-               smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
-       } while (data & BMCR_RESET);
+               data = phy_read(mp->phy, MII_BMCR);
+       } while (data >= 0 && data & BMCR_RESET);
 }
 
 static void port_start(struct mv643xx_eth_private *mp)
@@ -1890,7 +2228,7 @@ static void port_start(struct mv643xx_eth_private *mp)
        /*
         * Perform PHY reset, if there is a PHY.
         */
-       if (mp->phy_addr != -1) {
+       if (mp->phy != NULL) {
                struct ethtool_cmd cmd;
 
                mv643xx_eth_get_settings(mp->dev, &cmd);
@@ -1901,96 +2239,86 @@ static void port_start(struct mv643xx_eth_private *mp)
        /*
         * Configure basic link parameters.
         */
-       pscr = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       pscr = rdlp(mp, PORT_SERIAL_CONTROL);
 
        pscr |= SERIAL_PORT_ENABLE;
-       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
+       wrlp(mp, PORT_SERIAL_CONTROL, pscr);
 
        pscr |= DO_NOT_FORCE_LINK_FAIL;
-       if (mp->phy_addr == -1)
+       if (mp->phy == NULL)
                pscr |= FORCE_LINK_PASS;
-       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
-
-       wrl(mp, SDMA_CONFIG(mp->port_num), PORT_SDMA_CONFIG_DEFAULT_VALUE);
+       wrlp(mp, PORT_SERIAL_CONTROL, pscr);
 
        /*
         * Configure TX path and queues.
         */
        tx_set_rate(mp, 1000000000, 16777216);
-       for (i = 0; i < 8; i++) {
+       for (i = 0; i < mp->txq_count; i++) {
                struct tx_queue *txq = mp->txq + i;
 
-               if ((mp->txq_mask & (1 << i)) == 0)
-                       continue;
-
                txq_reset_hw_ptr(txq);
                txq_set_rate(txq, 1000000000, 16777216);
                txq_set_fixed_prio_mode(txq);
        }
 
        /*
-        * Add configured unicast address to address filter table.
+        * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
+        * frames to RX queue #0, and include the pseudo-header when
+        * calculating receive checksums.
         */
-       uc_addr_set(mp, mp->dev->dev_addr);
+       wrlp(mp, PORT_CONFIG, 0x02000000);
 
        /*
-        * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
-        * frames to RX queue #0.
+        * Treat BPDUs as normal multicasts, and disable partition mode.
         */
-       wrl(mp, PORT_CONFIG(mp->port_num), 0x00000000);
+       wrlp(mp, PORT_CONFIG_EXT, 0x00000000);
 
        /*
-        * Treat BPDUs as normal multicasts, and disable partition mode.
+        * Add configured unicast addresses to address filter table.
         */
-       wrl(mp, PORT_CONFIG_EXT(mp->port_num), 0x00000000);
+       mv643xx_eth_program_unicast_filter(mp->dev);
 
        /*
         * Enable the receive queues.
         */
-       for (i = 0; i < 8; i++) {
+       for (i = 0; i < mp->rxq_count; 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);
+               wrlp(mp, RXQ_CURRENT_DESC_PTR(i), addr);
 
                rxq_enable(rxq);
        }
 }
 
-static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
+static void mv643xx_eth_recalc_skb_size(struct mv643xx_eth_private *mp)
 {
-       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
-       u32 val;
+       int skb_size;
 
-       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);
-}
+       /*
+        * Reserve 2+14 bytes for an ethernet header (the hardware
+        * automatically prepends 2 bytes of dummy data to each
+        * received packet), 16 bytes for up to four VLAN tags, and
+        * 4 bytes for the trailing FCS -- 36 bytes total.
+        */
+       skb_size = mp->dev->mtu + 36;
 
-static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
-{
-       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
+       /*
+        * Make sure that the skb size is a multiple of 8 bytes, as
+        * the lower three bits of the receive descriptor's buffer
+        * size field are ignored by the hardware.
+        */
+       mp->skb_size = (skb_size + 7) & ~7;
 
-       if (coal > 0x3fff)
-               coal = 0x3fff;
-       wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), (coal & 0x3fff) << 4);
+       /*
+        * If NET_SKB_PAD is smaller than a cache line,
+        * netdev_alloc_skb() will cause skb->data to be misaligned
+        * to a cache line boundary.  If this is the case, include
+        * some extra space to allow re-aligning the data area.
+        */
+       mp->skb_size += SKB_DMA_REALIGN;
 }
 
 static int mv643xx_eth_open(struct net_device *dev)
@@ -1999,9 +2327,9 @@ static int mv643xx_eth_open(struct net_device *dev)
        int err;
        int i;
 
-       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));
+       wrlp(mp, INT_CAUSE, 0);
+       wrlp(mp, INT_CAUSE_EXT, 0);
+       rdlp(mp, INT_CAUSE_EXT);
 
        err = request_irq(dev->irq, mv643xx_eth_irq,
                          IRQF_SHARED, dev->name, dev);
@@ -2010,60 +2338,52 @@ static int mv643xx_eth_open(struct net_device *dev)
                return -EAGAIN;
        }
 
-       init_mac_tables(mp);
+       mv643xx_eth_recalc_skb_size(mp);
 
-       for (i = 0; i < 8; i++) {
-               if ((mp->rxq_mask & (1 << i)) == 0)
-                       continue;
+       napi_enable(&mp->napi);
 
+       skb_queue_head_init(&mp->rx_recycle);
+
+       mp->int_mask = INT_EXT;
+
+       for (i = 0; i < mp->rxq_count; i++) {
                err = rxq_init(mp, i);
                if (err) {
                        while (--i >= 0)
-                               if (mp->rxq_mask & (1 << i))
-                                       rxq_deinit(mp->rxq + i);
+                               rxq_deinit(mp->rxq + i);
                        goto out;
                }
 
-               rxq_refill(mp->rxq + i);
+               rxq_refill(mp->rxq + i, INT_MAX);
+               mp->int_mask |= INT_RX_0 << i;
        }
 
-       for (i = 0; i < 8; i++) {
-               if ((mp->txq_mask & (1 << i)) == 0)
-                       continue;
+       if (mp->oom) {
+               mp->rx_oom.expires = jiffies + (HZ / 10);
+               add_timer(&mp->rx_oom);
+       }
 
+       for (i = 0; i < mp->txq_count; i++) {
                err = txq_init(mp, i);
                if (err) {
                        while (--i >= 0)
-                               if (mp->txq_mask & (1 << i))
-                                       txq_deinit(mp->txq + i);
+                               txq_deinit(mp->txq + i);
                        goto out_free;
                }
+               mp->int_mask |= INT_TX_END_0 << i;
        }
 
-#ifdef MV643XX_ETH_NAPI
-       napi_enable(&mp->napi);
-#endif
-
-       netif_carrier_off(dev);
-       netif_stop_queue(dev);
-
        port_start(mp);
 
-       set_rx_coal(mp, 0);
-       set_tx_coal(mp, 0);
-
-       wrl(mp, INT_MASK_EXT(mp->port_num),
-           INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
-
-       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
+       wrlp(mp, INT_MASK_EXT, INT_EXT_LINK_PHY | INT_EXT_TX);
+       wrlp(mp, INT_MASK, mp->int_mask);
 
        return 0;
 
 
 out_free:
-       for (i = 0; i < 8; i++)
-               if (mp->rxq_mask & (1 << i))
-                       rxq_deinit(mp->rxq + i);
+       for (i = 0; i < mp->rxq_count; i++)
+               rxq_deinit(mp->rxq + i);
 out:
        free_irq(dev->irq, dev);
 
@@ -2075,15 +2395,13 @@ static void port_reset(struct mv643xx_eth_private *mp)
        unsigned int data;
        int i;
 
-       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);
-       }
+       for (i = 0; i < mp->rxq_count; i++)
+               rxq_disable(mp->rxq + i);
+       for (i = 0; i < mp->txq_count; i++)
+               txq_disable(mp->txq + i);
 
        while (1) {
-               u32 ps = rdl(mp, PORT_STATUS(mp->port_num));
+               u32 ps = rdlp(mp, PORT_STATUS);
 
                if ((ps & (TX_IN_PROGRESS | TX_FIFO_EMPTY)) == TX_FIFO_EMPTY)
                        break;
@@ -2091,11 +2409,11 @@ static void port_reset(struct mv643xx_eth_private *mp)
        }
 
        /* Reset the Enable bit in the Configuration Register */
-       data = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       data = rdlp(mp, PORT_SERIAL_CONTROL);
        data &= ~(SERIAL_PORT_ENABLE            |
                  DO_NOT_FORCE_LINK_FAIL        |
                  FORCE_LINK_PASS);
-       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), data);
+       wrlp(mp, PORT_SERIAL_CONTROL, data);
 }
 
 static int mv643xx_eth_stop(struct net_device *dev)
@@ -2103,26 +2421,29 @@ static int mv643xx_eth_stop(struct net_device *dev)
        struct mv643xx_eth_private *mp = netdev_priv(dev);
        int i;
 
-       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
-       rdl(mp, INT_MASK(mp->port_num));
+       wrlp(mp, INT_MASK_EXT, 0x00000000);
+       wrlp(mp, INT_MASK, 0x00000000);
+       rdlp(mp, INT_MASK);
 
-#ifdef MV643XX_ETH_NAPI
        napi_disable(&mp->napi);
-#endif
+
+       del_timer_sync(&mp->rx_oom);
+
        netif_carrier_off(dev);
-       netif_stop_queue(dev);
 
        free_irq(dev->irq, dev);
 
        port_reset(mp);
+       mv643xx_eth_get_stats(dev);
        mib_counters_update(mp);
+       del_timer_sync(&mp->mib_counters_timer);
 
-       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);
-       }
+       skb_queue_purge(&mp->rx_recycle);
+
+       for (i = 0; i < mp->rxq_count; i++)
+               rxq_deinit(mp->rxq + i);
+       for (i = 0; i < mp->txq_count; i++)
+               txq_deinit(mp->txq + i);
 
        return 0;
 }
@@ -2131,8 +2452,8 @@ static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       if (mp->phy_addr != -1)
-               return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
+       if (mp->phy != NULL)
+               return phy_mii_ioctl(mp->phy, if_mii(ifr), cmd);
 
        return -EOPNOTSUPP;
 }
@@ -2145,6 +2466,7 @@ static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
                return -EINVAL;
 
        dev->mtu = new_mtu;
+       mv643xx_eth_recalc_skb_size(mp);
        tx_set_rate(mp, 1000000000, 16777216);
 
        if (!netif_running(dev))
@@ -2172,12 +2494,10 @@ static void tx_timeout_task(struct work_struct *ugly)
 
        mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
        if (netif_running(mp->dev)) {
-               netif_stop_queue(mp->dev);
-
+               netif_tx_stop_all_queues(mp->dev);
                port_reset(mp);
                port_start(mp);
-
-               __txq_maybe_wake(mp->txq + mp->txq_primary);
+               netif_tx_wake_all_queues(mp->dev);
        }
 }
 
@@ -2195,31 +2515,15 @@ static void mv643xx_eth_netpoll(struct net_device *dev)
 {
        struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
-       rdl(mp, INT_MASK(mp->port_num));
+       wrlp(mp, INT_MASK, 0x00000000);
+       rdlp(mp, INT_MASK);
 
        mv643xx_eth_irq(dev->irq, dev);
 
-       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
+       wrlp(mp, INT_MASK, mp->int_mask);
 }
 #endif
 
-static int mv643xx_eth_mdio_read(struct net_device *dev, int addr, int reg)
-{
-       struct mv643xx_eth_private *mp = netdev_priv(dev);
-       int val;
-
-       smi_reg_read(mp, addr, reg, &val);
-
-       return 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);
-       smi_reg_write(mp, addr, reg, val);
-}
-
 
 /* platform glue ************************************************************/
 static void
@@ -2264,26 +2568,32 @@ static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
         * [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)
+       writel(0x02000000, msp->base + 0x0400 + SDMA_CONFIG);
+       if (readl(msp->base + 0x0400 + SDMA_CONFIG) & 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.
+        * Check whether the MAC supports TX rate control, and if
+        * yes, whether its associated registers are in the old or
+        * the new place.
         */
-       writel(1, msp->base + TX_BW_MTU_MOVED(0));
-       if (readl(msp->base + TX_BW_MTU_MOVED(0)) & 1)
-               msp->tx_bw_control_moved = 1;
-       else
-               msp->tx_bw_control_moved = 0;
+       writel(1, msp->base + 0x0400 + TX_BW_MTU_MOVED);
+       if (readl(msp->base + 0x0400 + TX_BW_MTU_MOVED) & 1) {
+               msp->tx_bw_control = TX_BW_CONTROL_NEW_LAYOUT;
+       } else {
+               writel(7, msp->base + 0x0400 + TX_BW_RATE);
+               if (readl(msp->base + 0x0400 + TX_BW_RATE) & 7)
+                       msp->tx_bw_control = TX_BW_CONTROL_OLD_LAYOUT;
+               else
+                       msp->tx_bw_control = TX_BW_CONTROL_ABSENT;
+       }
 }
 
 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 {
-       static int mv643xx_eth_version_printed = 0;
+       static int mv643xx_eth_version_printed;
        struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
        struct mv643xx_eth_shared_private *msp;
        struct resource *res;
@@ -2308,7 +2618,45 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
        if (msp->base == NULL)
                goto out_free;
 
-       spin_lock_init(&msp->phy_lock);
+       /*
+        * Set up and register SMI bus.
+        */
+       if (pd == NULL || pd->shared_smi == NULL) {
+               msp->smi_bus = mdiobus_alloc();
+               if (msp->smi_bus == NULL)
+                       goto out_unmap;
+
+               msp->smi_bus->priv = msp;
+               msp->smi_bus->name = "mv643xx_eth smi";
+               msp->smi_bus->read = smi_bus_read;
+               msp->smi_bus->write = smi_bus_write,
+               snprintf(msp->smi_bus->id, MII_BUS_ID_SIZE, "%d", pdev->id);
+               msp->smi_bus->parent = &pdev->dev;
+               msp->smi_bus->phy_mask = 0xffffffff;
+               if (mdiobus_register(msp->smi_bus) < 0)
+                       goto out_free_mii_bus;
+               msp->smi = msp;
+       } else {
+               msp->smi = platform_get_drvdata(pd->shared_smi);
+       }
+
+       msp->err_interrupt = NO_IRQ;
+       init_waitqueue_head(&msp->smi_busy_wait);
+
+       /*
+        * Check whether the error interrupt is hooked up.
+        */
+       res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
+       if (res != NULL) {
+               int err;
+
+               err = request_irq(res->start, mv643xx_eth_err_irq,
+                                 IRQF_SHARED, "mv643xx_eth", msp);
+               if (!err) {
+                       writel(ERR_INT_SMI_DONE, msp->base + ERR_INT_MASK);
+                       msp->err_interrupt = res->start;
+               }
+       }
 
        /*
         * (Re-)program MBUS remapping windows if we are asked to.
@@ -2326,6 +2674,10 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 
        return 0;
 
+out_free_mii_bus:
+       mdiobus_free(msp->smi_bus);
+out_unmap:
+       iounmap(msp->base);
 out_free:
        kfree(msp);
 out:
@@ -2335,7 +2687,14 @@ out:
 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
 {
        struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
+       struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
 
+       if (pd == NULL || pd->shared_smi == NULL) {
+               mdiobus_unregister(msp->smi_bus);
+               mdiobus_free(msp->smi_bus);
+       }
+       if (msp->err_interrupt != NO_IRQ)
+               free_irq(msp->err_interrupt, msp);
        iounmap(msp->base);
        kfree(msp);
 
@@ -2381,124 +2740,91 @@ static void set_params(struct mv643xx_eth_private *mp,
        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;
+       mp->rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
        if (pd->rx_queue_size)
-               mp->default_rx_ring_size = pd->rx_queue_size;
+               mp->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;
+       mp->rxq_count = pd->rx_queue_count ? : 1;
 
-       mp->default_tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
+       mp->tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
        if (pd->tx_queue_size)
-               mp->default_tx_ring_size = pd->tx_queue_size;
+               mp->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;
 
-       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;
+       mp->txq_count = pd->tx_queue_count ? : 1;
 }
 
-static int phy_detect(struct mv643xx_eth_private *mp)
+static struct phy_device *phy_scan(struct mv643xx_eth_private *mp,
+                                  int phy_addr)
 {
-       unsigned int data;
-       unsigned int data2;
+       struct mii_bus *bus = mp->shared->smi->smi_bus;
+       struct phy_device *phydev;
+       int start;
+       int num;
+       int i;
 
-       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
-       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data ^ BMCR_ANENABLE);
+       if (phy_addr == MV643XX_ETH_PHY_ADDR_DEFAULT) {
+               start = phy_addr_get(mp) & 0x1f;
+               num = 32;
+       } else {
+               start = phy_addr & 0x1f;
+               num = 1;
+       }
 
-       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data2);
-       if (((data ^ data2) & BMCR_ANENABLE) == 0)
-               return -ENODEV;
+       phydev = NULL;
+       for (i = 0; i < num; i++) {
+               int addr = (start + i) & 0x1f;
 
-       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data);
+               if (bus->phy_map[addr] == NULL)
+                       mdiobus_scan(bus, addr);
 
-       return 0;
+               if (phydev == NULL) {
+                       phydev = bus->phy_map[addr];
+                       if (phydev != NULL)
+                               phy_addr_set(mp, addr);
+               }
+       }
+
+       return phydev;
 }
 
-static int phy_init(struct mv643xx_eth_private *mp,
-                   struct mv643xx_eth_platform_data *pd)
+static void phy_init(struct mv643xx_eth_private *mp, int speed, int duplex)
 {
-       struct ethtool_cmd cmd;
-       int err;
+       struct phy_device *phy = mp->phy;
 
-       err = phy_detect(mp);
-       if (err) {
-               dev_printk(KERN_INFO, &mp->dev->dev,
-                          "no PHY detected at addr %d\n", mp->phy_addr);
-               return err;
-       }
        phy_reset(mp);
 
-       mp->mii.phy_id = mp->phy_addr;
-       mp->mii.phy_id_mask = 0x3f;
-       mp->mii.reg_num_mask = 0x1f;
-       mp->mii.dev = mp->dev;
-       mp->mii.mdio_read = mv643xx_eth_mdio_read;
-       mp->mii.mdio_write = mv643xx_eth_mdio_write;
-
-       mp->mii.supports_gmii = mii_check_gmii_support(&mp->mii);
-
-       memset(&cmd, 0, sizeof(cmd));
-
-       cmd.port = PORT_MII;
-       cmd.transceiver = XCVR_INTERNAL;
-       cmd.phy_address = mp->phy_addr;
-       if (pd->speed == 0) {
-               cmd.autoneg = AUTONEG_ENABLE;
-               cmd.speed = SPEED_100;
-               cmd.advertising = ADVERTISED_10baseT_Half  |
-                                 ADVERTISED_10baseT_Full  |
-                                 ADVERTISED_100baseT_Half |
-                                 ADVERTISED_100baseT_Full;
-               if (mp->mii.supports_gmii)
-                       cmd.advertising |= ADVERTISED_1000baseT_Full;
+       phy_attach(mp->dev, dev_name(&phy->dev), 0, PHY_INTERFACE_MODE_GMII);
+
+       if (speed == 0) {
+               phy->autoneg = AUTONEG_ENABLE;
+               phy->speed = 0;
+               phy->duplex = 0;
+               phy->advertising = phy->supported | ADVERTISED_Autoneg;
        } else {
-               cmd.autoneg = AUTONEG_DISABLE;
-               cmd.speed = pd->speed;
-               cmd.duplex = pd->duplex;
+               phy->autoneg = AUTONEG_DISABLE;
+               phy->advertising = 0;
+               phy->speed = speed;
+               phy->duplex = duplex;
        }
-
-       mv643xx_eth_set_settings(mp->dev, &cmd);
-
-       return 0;
+       phy_start_aneg(phy);
 }
 
 static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
 {
        u32 pscr;
 
-       pscr = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       pscr = rdlp(mp, PORT_SERIAL_CONTROL);
        if (pscr & SERIAL_PORT_ENABLE) {
                pscr &= ~SERIAL_PORT_ENABLE;
-               wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
+               wrlp(mp, PORT_SERIAL_CONTROL, pscr);
        }
 
        pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
-       if (mp->phy_addr == -1) {
+       if (mp->phy == NULL) {
                pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
                if (speed == SPEED_1000)
                        pscr |= SET_GMII_SPEED_TO_1000;
@@ -2512,16 +2838,30 @@ static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
                        pscr |= SET_FULL_DUPLEX_MODE;
        }
 
-       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
+       wrlp(mp, PORT_SERIAL_CONTROL, pscr);
 }
 
+static const struct net_device_ops mv643xx_eth_netdev_ops = {
+       .ndo_open               = mv643xx_eth_open,
+       .ndo_stop               = mv643xx_eth_stop,
+       .ndo_start_xmit         = mv643xx_eth_xmit,
+       .ndo_set_rx_mode        = mv643xx_eth_set_rx_mode,
+       .ndo_set_mac_address    = mv643xx_eth_set_mac_address,
+       .ndo_do_ioctl           = mv643xx_eth_ioctl,
+       .ndo_change_mtu         = mv643xx_eth_change_mtu,
+       .ndo_tx_timeout         = mv643xx_eth_tx_timeout,
+       .ndo_get_stats          = mv643xx_eth_get_stats,
+#ifdef CONFIG_NET_POLL_CONTROLLER
+       .ndo_poll_controller    = mv643xx_eth_netpoll,
+#endif
+};
+
 static int mv643xx_eth_probe(struct platform_device *pdev)
 {
        struct mv643xx_eth_platform_data *pd;
        struct mv643xx_eth_private *mp;
        struct net_device *dev;
        struct resource *res;
-       DECLARE_MAC_BUF(mac);
        int err;
 
        pd = pdev->dev.platform_data;
@@ -2537,7 +2877,7 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
                return -ENODEV;
        }
 
-       dev = alloc_etherdev(sizeof(struct mv643xx_eth_private));
+       dev = alloc_etherdev_mq(sizeof(struct mv643xx_eth_private), 8);
        if (!dev)
                return -ENOMEM;
 
@@ -2545,80 +2885,74 @@ static int mv643xx_eth_probe(struct platform_device *pdev)
        platform_set_drvdata(pdev, mp);
 
        mp->shared = platform_get_drvdata(pd->shared);
+       mp->base = mp->shared->base + 0x0400 + (pd->port_number << 10);
        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);
+       dev->real_num_tx_queues = mp->txq_count;
+
+       if (pd->phy_addr != MV643XX_ETH_PHY_NONE)
+               mp->phy = phy_scan(mp, pd->phy_addr);
+
+       if (mp->phy != NULL)
+               phy_init(mp, pd->speed, pd->duplex);
+
+       SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
+
+       init_pscr(mp, pd->speed, pd->duplex);
 
-       spin_lock_init(&mp->lock);
 
        mib_counters_clear(mp);
+
+       init_timer(&mp->mib_counters_timer);
+       mp->mib_counters_timer.data = (unsigned long)mp;
+       mp->mib_counters_timer.function = mib_counters_timer_wrapper;
+       mp->mib_counters_timer.expires = jiffies + 30 * HZ;
+       add_timer(&mp->mib_counters_timer);
+
+       spin_lock_init(&mp->mib_counters_lock);
+
        INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
 
-       if (mp->phy_addr != -1) {
-               err = phy_init(mp, pd);
-               if (err)
-                       goto out;
+       netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, 128);
 
-               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
-       } else {
-               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops_phyless);
-       }
-       init_pscr(mp, pd->speed, pd->duplex);
+       init_timer(&mp->rx_oom);
+       mp->rx_oom.data = (unsigned long)mp;
+       mp->rx_oom.function = oom_timer_wrapper;
 
 
        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->set_multicast_list = mv643xx_eth_set_rx_mode;
-       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->netdev_ops = &mv643xx_eth_netdev_ops;
+
        dev->watchdog_timeo = 2 * HZ;
        dev->base_addr = 0;
 
-#ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
-       /*
-        * 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;
        dev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM;
-#endif
 
        SET_NETDEV_DEV(dev, &pdev->dev);
 
        if (mp->shared->win_protect)
                wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
 
-       err = register_netdev(dev);
-       if (err)
-               goto out;
+       netif_carrier_off(dev);
 
-       dev_printk(KERN_NOTICE, &dev->dev, "port %d with MAC address %s\n",
-                  mp->port_num, print_mac(mac, dev->dev_addr));
+       wrlp(mp, SDMA_CONFIG, PORT_SDMA_CONFIG_DEFAULT_VALUE);
 
-       if (dev->features & NETIF_F_SG)
-               dev_printk(KERN_NOTICE, &dev->dev, "scatter/gather enabled\n");
+       set_rx_coal(mp, 250);
+       set_tx_coal(mp, 0);
 
-       if (dev->features & NETIF_F_IP_CSUM)
-               dev_printk(KERN_NOTICE, &dev->dev, "tx checksum offload\n");
+       err = register_netdev(dev);
+       if (err)
+               goto out;
 
-#ifdef MV643XX_ETH_NAPI
-       dev_printk(KERN_NOTICE, &dev->dev, "napi enabled\n");
-#endif
+       dev_printk(KERN_NOTICE, &dev->dev, "port %d with MAC address %pM\n",
+                  mp->port_num, dev->dev_addr);
 
        if (mp->tx_desc_sram_size > 0)
                dev_printk(KERN_NOTICE, &dev->dev, "configured with sram\n");
@@ -2636,6 +2970,8 @@ static int mv643xx_eth_remove(struct platform_device *pdev)
        struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
        unregister_netdev(mp->dev);
+       if (mp->phy != NULL)
+               phy_detach(mp->phy);
        flush_scheduled_work();
        free_netdev(mp->dev);
 
@@ -2649,8 +2985,8 @@ static void mv643xx_eth_shutdown(struct platform_device *pdev)
        struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
        /* Mask all interrupts on ethernet port */
-       wrl(mp, INT_MASK(mp->port_num), 0);
-       rdl(mp, INT_MASK(mp->port_num));
+       wrlp(mp, INT_MASK, 0);
+       rdlp(mp, INT_MASK);
 
        if (netif_running(mp->dev))
                port_reset(mp);