Merge branch 'core/xen' into x86/xen
[safe/jmp/linux-2.6] / drivers / net / mv643xx_eth.c
index bcc891d..0a18b9e 100644 (file)
 #include <asm/types.h>
 #include <asm/system.h>
 
-static char mv643xx_driver_name[] = "mv643xx_eth";
-static char mv643xx_driver_version[] = "1.0";
+static char mv643xx_eth_driver_name[] = "mv643xx_eth";
+static char mv643xx_eth_driver_version[] = "1.3";
 
-#define MV643XX_CHECKSUM_OFFLOAD_TX
-#define MV643XX_NAPI
-#define MV643XX_TX_FAST_REFILL
-#undef MV643XX_COAL
+#define MV643XX_ETH_CHECKSUM_OFFLOAD_TX
+#define MV643XX_ETH_NAPI
+#define MV643XX_ETH_TX_FAST_REFILL
 
-#define MV643XX_TX_COAL 100
-#ifdef MV643XX_COAL
-#define MV643XX_RX_COAL 100
-#endif
-
-#ifdef MV643XX_CHECKSUM_OFFLOAD_TX
+#ifdef MV643XX_ETH_CHECKSUM_OFFLOAD_TX
 #define MAX_DESCS_PER_SKB      (MAX_SKB_FRAGS + 1)
 #else
 #define MAX_DESCS_PER_SKB      1
 #endif
 
-#define ETH_VLAN_HLEN          4
-#define ETH_FCS_LEN            4
-#define ETH_HW_IP_ALIGN                2               /* hw aligns IP header */
-#define ETH_WRAPPER_LEN                (ETH_HW_IP_ALIGN + ETH_HLEN + \
-                                       ETH_VLAN_HLEN + ETH_FCS_LEN)
-#define ETH_RX_SKB_SIZE                (dev->mtu + ETH_WRAPPER_LEN + \
-                                       dma_get_cache_alignment())
-
 /*
  * Registers shared between all ports.
  */
@@ -103,23 +89,44 @@ static char mv643xx_driver_version[] = "1.0";
 #define SDMA_CONFIG(p)                 (0x041c + ((p) << 10))
 #define PORT_SERIAL_CONTROL(p)         (0x043c + ((p) << 10))
 #define PORT_STATUS(p)                 (0x0444 + ((p) << 10))
+#define  TX_FIFO_EMPTY                 0x00000400
+#define  TX_IN_PROGRESS                        0x00000080
+#define  PORT_SPEED_MASK               0x00000030
+#define  PORT_SPEED_1000               0x00000010
+#define  PORT_SPEED_100                        0x00000020
+#define  PORT_SPEED_10                 0x00000000
+#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_RX                                0x00000804
+#define  INT_TX_END_0                  0x00080000
+#define  INT_TX_END                    0x07f80000
+#define  INT_RX                                0x0007fbfc
 #define  INT_EXT                       0x00000002
 #define INT_CAUSE_EXT(p)               (0x0464 + ((p) << 10))
 #define  INT_EXT_LINK                  0x00100000
 #define  INT_EXT_PHY                   0x00010000
 #define  INT_EXT_TX_ERROR_0            0x00000100
 #define  INT_EXT_TX_0                  0x00000001
-#define  INT_EXT_TX                    0x00000101
+#define  INT_EXT_TX                    0x0000ffff
 #define INT_MASK(p)                    (0x0468 + ((p) << 10))
 #define INT_MASK_EXT(p)                        (0x046c + ((p) << 10))
 #define TX_FIFO_URGENT_THRESHOLD(p)    (0x0474 + ((p) << 10))
-#define RXQ_CURRENT_DESC_PTR(p)                (0x060c + ((p) << 10))
+#define TXQ_FIX_PRIO_CONF_MOVED(p)     (0x04dc + ((p) << 10))
+#define TX_BW_RATE_MOVED(p)            (0x04e0 + ((p) << 10))
+#define TX_BW_MTU_MOVED(p)             (0x04e8 + ((p) << 10))
+#define TX_BW_BURST_MOVED(p)           (0x04ec + ((p) << 10))
+#define RXQ_CURRENT_DESC_PTR(p, q)     (0x060c + ((p) << 10) + ((q) << 4))
 #define RXQ_COMMAND(p)                 (0x0680 + ((p) << 10))
-#define TXQ_CURRENT_DESC_PTR(p)                (0x06c0 + ((p) << 10))
+#define TXQ_CURRENT_DESC_PTR(p, q)     (0x06c0 + ((p) << 10) + ((q) << 2))
+#define TXQ_BW_TOKENS(p, q)            (0x0700 + ((p) << 10) + ((q) << 4))
+#define TXQ_BW_CONF(p, q)              (0x0704 + ((p) << 10) + ((q) << 4))
+#define TXQ_BW_WRR_CONF(p, q)          (0x0708 + ((p) << 10) + ((q) << 4))
 #define MIB_COUNTERS(p)                        (0x1000 + ((p) << 7))
 #define SPECIAL_MCAST_TABLE(p)         (0x1400 + ((p) << 10))
 #define OTHER_MCAST_TABLE(p)           (0x1500 + ((p) << 10))
@@ -129,21 +136,21 @@ static char mv643xx_driver_version[] = "1.0";
 /*
  * SDMA configuration register.
  */
-#define RX_BURST_SIZE_4_64BIT          (2 << 1)
+#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_4_64BIT          (2 << 22)
+#define TX_BURST_SIZE_16_64BIT         (4 << 22)
 
 #if defined(__BIG_ENDIAN)
 #define PORT_SDMA_CONFIG_DEFAULT_VALUE         \
-               RX_BURST_SIZE_4_64BIT   |       \
-               TX_BURST_SIZE_4_64BIT
+               RX_BURST_SIZE_16_64BIT  |       \
+               TX_BURST_SIZE_16_64BIT
 #elif defined(__LITTLE_ENDIAN)
 #define PORT_SDMA_CONFIG_DEFAULT_VALUE         \
-               RX_BURST_SIZE_4_64BIT   |       \
+               RX_BURST_SIZE_16_64BIT  |       \
                BLM_RX_NO_SWAP          |       \
                BLM_TX_NO_SWAP          |       \
-               TX_BURST_SIZE_4_64BIT
+               TX_BURST_SIZE_16_64BIT
 #else
 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
 #endif
@@ -155,9 +162,7 @@ static char mv643xx_driver_version[] = "1.0";
 #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_1522BYTE                 (1 << 17)
 #define MAX_RX_PACKET_9700BYTE                 (5 << 17)
-#define MAX_RX_PACKET_MASK                     (7 << 17)
 #define DISABLE_AUTO_NEG_SPEED_GMII            (1 << 13)
 #define DO_NOT_FORCE_LINK_FAIL                 (1 << 10)
 #define SERIAL_PORT_CONTROL_RESERVED           (1 << 9)
@@ -166,94 +171,15 @@ static char mv643xx_driver_version[] = "1.0";
 #define FORCE_LINK_PASS                                (1 << 1)
 #define SERIAL_PORT_ENABLE                     (1 << 0)
 
-#define PORT_DEFAULT_TRANSMIT_QUEUE_SIZE       800
-#define PORT_DEFAULT_RECEIVE_QUEUE_SIZE                400
-
-/* Port serial status reg (PSR) */
-#define ETH_INTERFACE_PCM                      0x00000001
-#define ETH_LINK_IS_UP                         0x00000002
-#define ETH_PORT_AT_FULL_DUPLEX                        0x00000004
-#define ETH_RX_FLOW_CTRL_ENABLED               0x00000008
-#define ETH_GMII_SPEED_1000                    0x00000010
-#define ETH_MII_SPEED_100                      0x00000020
-#define ETH_TX_IN_PROGRESS                     0x00000080
-#define ETH_BYPASS_ACTIVE                      0x00000100
-#define ETH_PORT_AT_PARTITION_STATE            0x00000200
-#define ETH_PORT_TX_FIFO_EMPTY                 0x00000400
-
-/* SMI reg */
-#define ETH_SMI_BUSY           0x10000000      /* 0 - Write, 1 - Read  */
-#define ETH_SMI_READ_VALID     0x08000000      /* 0 - Write, 1 - Read  */
-#define ETH_SMI_OPCODE_WRITE   0               /* Completion of Read   */
-#define ETH_SMI_OPCODE_READ    0x04000000      /* Operation is in progress */
-
-/* Interrupt Cause Register Bit Definitions */
-
-/* SDMA command status fields macros */
-
-/* Tx & Rx descriptors status */
-#define ETH_ERROR_SUMMARY                      0x00000001
-
-/* Tx & Rx descriptors command */
-#define ETH_BUFFER_OWNED_BY_DMA                        0x80000000
-
-/* Tx descriptors status */
-#define ETH_LC_ERROR                           0
-#define ETH_UR_ERROR                           0x00000002
-#define ETH_RL_ERROR                           0x00000004
-#define ETH_LLC_SNAP_FORMAT                    0x00000200
-
-/* Rx descriptors status */
-#define ETH_OVERRUN_ERROR                      0x00000002
-#define ETH_MAX_FRAME_LENGTH_ERROR             0x00000004
-#define ETH_RESOURCE_ERROR                     0x00000006
-#define ETH_VLAN_TAGGED                                0x00080000
-#define ETH_BPDU_FRAME                         0x00100000
-#define ETH_UDP_FRAME_OVER_IP_V_4              0x00200000
-#define ETH_OTHER_FRAME_TYPE                   0x00400000
-#define ETH_LAYER_2_IS_ETH_V_2                 0x00800000
-#define ETH_FRAME_TYPE_IP_V_4                  0x01000000
-#define ETH_FRAME_HEADER_OK                    0x02000000
-#define ETH_RX_LAST_DESC                       0x04000000
-#define ETH_RX_FIRST_DESC                      0x08000000
-#define ETH_UNKNOWN_DESTINATION_ADDR           0x10000000
-#define ETH_RX_ENABLE_INTERRUPT                        0x20000000
-#define ETH_LAYER_4_CHECKSUM_OK                        0x40000000
-
-/* Rx descriptors byte count */
-#define ETH_FRAME_FRAGMENTED                   0x00000004
-
-/* Tx descriptors command */
-#define ETH_LAYER_4_CHECKSUM_FIRST_DESC                0x00000400
-#define ETH_FRAME_SET_TO_VLAN                  0x00008000
-#define ETH_UDP_FRAME                          0x00010000
-#define ETH_GEN_TCP_UDP_CHECKSUM               0x00020000
-#define ETH_GEN_IP_V_4_CHECKSUM                        0x00040000
-#define ETH_ZERO_PADDING                       0x00080000
-#define ETH_TX_LAST_DESC                       0x00100000
-#define ETH_TX_FIRST_DESC                      0x00200000
-#define ETH_GEN_CRC                            0x00400000
-#define ETH_TX_ENABLE_INTERRUPT                        0x00800000
-#define ETH_AUTO_MODE                          0x40000000
-
-#define ETH_TX_IHL_SHIFT                       11
-
-/* typedefs */
-
-typedef enum _eth_func_ret_status {
-       ETH_OK,                 /* Returned as expected.                */
-       ETH_ERROR,              /* Fundamental error.                   */
-       ETH_RETRY,              /* Could not process request. Try later.*/
-       ETH_END_OF_JOB,         /* Ring has nothing to process.         */
-       ETH_QUEUE_FULL,         /* Ring resource error.                 */
-       ETH_QUEUE_LAST_RESOURCE /* Ring resources about to exhaust.     */
-} ETH_FUNC_RET_STATUS;
-
-/* These are for big-endian machines.  Little endian needs different
- * definitions.
+#define DEFAULT_RX_QUEUE_SIZE          400
+#define DEFAULT_TX_QUEUE_SIZE          800
+
+
+/*
+ * RX/TX descriptors.
  */
 #if defined(__BIG_ENDIAN)
-struct eth_rx_desc {
+struct rx_desc {
        u16 byte_cnt;           /* Descriptor buffer byte count         */
        u16 buf_size;           /* Buffer size                          */
        u32 cmd_sts;            /* Descriptor command status            */
@@ -261,7 +187,7 @@ struct eth_rx_desc {
        u32 buf_ptr;            /* Descriptor buffer pointer            */
 };
 
-struct eth_tx_desc {
+struct tx_desc {
        u16 byte_cnt;           /* buffer byte count                    */
        u16 l4i_chk;            /* CPU provided TCP checksum            */
        u32 cmd_sts;            /* Command/status field                 */
@@ -269,7 +195,7 @@ struct eth_tx_desc {
        u32 buf_ptr;            /* pointer to buffer for this descriptor*/
 };
 #elif defined(__LITTLE_ENDIAN)
-struct eth_rx_desc {
+struct rx_desc {
        u32 cmd_sts;            /* Descriptor command status            */
        u16 buf_size;           /* Buffer size                          */
        u16 byte_cnt;           /* Descriptor buffer byte count         */
@@ -277,7 +203,7 @@ struct eth_rx_desc {
        u32 next_desc_ptr;      /* Next descriptor pointer              */
 };
 
-struct eth_tx_desc {
+struct tx_desc {
        u32 cmd_sts;            /* Command/status field                 */
        u16 l4i_chk;            /* CPU provided TCP checksum            */
        u16 byte_cnt;           /* buffer byte count                    */
@@ -288,32 +214,61 @@ struct eth_tx_desc {
 #error One of __BIG_ENDIAN or __LITTLE_ENDIAN must be defined
 #endif
 
-/* Unified struct for Rx and Tx operations. The user is not required to        */
-/* be familier with neither Tx nor Rx descriptors.                     */
-struct pkt_info {
-       unsigned short byte_cnt;        /* Descriptor buffer byte count */
-       unsigned short l4i_chk;         /* Tx CPU provided TCP Checksum */
-       unsigned int cmd_sts;           /* Descriptor command status    */
-       dma_addr_t buf_ptr;             /* Descriptor buffer pointer    */
-       struct sk_buff *return_info;    /* User resource return information */
-};
+/* RX & TX descriptor command */
+#define BUFFER_OWNED_BY_DMA            0x80000000
+
+/* RX & TX descriptor status */
+#define ERROR_SUMMARY                  0x00000001
+
+/* RX descriptor status */
+#define LAYER_4_CHECKSUM_OK            0x40000000
+#define RX_ENABLE_INTERRUPT            0x20000000
+#define RX_FIRST_DESC                  0x08000000
+#define RX_LAST_DESC                   0x04000000
+
+/* TX descriptor command */
+#define TX_ENABLE_INTERRUPT            0x00800000
+#define GEN_CRC                                0x00400000
+#define TX_FIRST_DESC                  0x00200000
+#define TX_LAST_DESC                   0x00100000
+#define ZERO_PADDING                   0x00080000
+#define GEN_IP_V4_CHECKSUM             0x00040000
+#define GEN_TCP_UDP_CHECKSUM           0x00020000
+#define UDP_FRAME                      0x00010000
+#define MAC_HDR_EXTRA_4_BYTES          0x00008000
+#define MAC_HDR_EXTRA_8_BYTES          0x00000200
+
+#define TX_IHL_SHIFT                   11
 
 
 /* global *******************************************************************/
-struct mv643xx_shared_private {
-       void __iomem *eth_base;
+struct mv643xx_eth_shared_private {
+       /*
+        * Ethernet controller base address.
+        */
+       void __iomem *base;
 
-       /* used to protect SMI_REG, which is shared across ports */
+       /*
+        * Protects access to SMI_REG, which is shared between ports.
+        */
        spinlock_t phy_lock;
 
+       /*
+        * Per-port MBUS window access register value.
+        */
        u32 win_protect;
 
+       /*
+        * Hardware-specific parameters.
+        */
        unsigned int t_clk;
+       int extended_rx_coal_limit;
+       int tx_bw_control_moved;
 };
 
 
 /* per-port *****************************************************************/
-struct mv643xx_mib_counters {
+struct mib_counters {
        u64 good_octets_received;
        u32 bad_octets_received;
        u32 internal_mac_transmit_err;
@@ -346,388 +301,299 @@ struct mv643xx_mib_counters {
        u32 late_collision;
 };
 
-struct mv643xx_private {
-       struct mv643xx_shared_private *shared;
-       int port_num;                   /* User Ethernet port number    */
-
-       struct mv643xx_shared_private *shared_smi;
+struct rx_queue {
+       int index;
 
-       u32 rx_sram_addr;               /* Base address of rx sram area */
-       u32 rx_sram_size;               /* Size of rx sram area         */
-       u32 tx_sram_addr;               /* Base address of tx sram area */
-       u32 tx_sram_size;               /* Size of tx sram area         */
+       int rx_ring_size;
 
-       int rx_resource_err;            /* Rx ring resource error flag */
+       int rx_desc_count;
+       int rx_curr_desc;
+       int rx_used_desc;
 
-       /* Tx/Rx rings managment indexes fields. For driver use */
+       struct rx_desc *rx_desc_area;
+       dma_addr_t rx_desc_dma;
+       int rx_desc_area_size;
+       struct sk_buff **rx_skb;
 
-       /* Next available and first returning Rx resource */
-       int rx_curr_desc_q, rx_used_desc_q;
+       struct timer_list rx_oom;
+};
 
-       /* Next available and first returning Tx resource */
-       int tx_curr_desc_q, tx_used_desc_q;
+struct tx_queue {
+       int index;
 
-#ifdef MV643XX_TX_FAST_REFILL
-       u32 tx_clean_threshold;
-#endif
+       int tx_ring_size;
 
-       struct eth_rx_desc *p_rx_desc_area;
-       dma_addr_t rx_desc_dma;
-       int rx_desc_area_size;
-       struct sk_buff **rx_skb;
+       int tx_desc_count;
+       int tx_curr_desc;
+       int tx_used_desc;
 
-       struct eth_tx_desc *p_tx_desc_area;
+       struct tx_desc *tx_desc_area;
        dma_addr_t tx_desc_dma;
        int tx_desc_area_size;
        struct sk_buff **tx_skb;
+};
 
-       struct work_struct tx_timeout_task;
+struct mv643xx_eth_private {
+       struct mv643xx_eth_shared_private *shared;
+       int port_num;
 
        struct net_device *dev;
-       struct napi_struct napi;
-       struct net_device_stats stats;
-       struct mv643xx_mib_counters mib_counters;
+
+       struct mv643xx_eth_shared_private *shared_smi;
+       int phy_addr;
+
        spinlock_t lock;
-       /* Size of Tx Ring per queue */
-       int tx_ring_size;
-       /* Number of tx descriptors in use */
-       int tx_desc_count;
-       /* Size of Rx Ring per queue */
-       int rx_ring_size;
-       /* Number of rx descriptors in use */
-       int rx_desc_count;
+
+       struct mib_counters mib_counters;
+       struct work_struct tx_timeout_task;
+       struct mii_if_info mii;
 
        /*
-        * Used in case RX Ring is empty, which can be caused when
-        * system does not have resources (skb's)
+        * RX state.
         */
-       struct timer_list timeout;
+       int default_rx_ring_size;
+       unsigned long rx_desc_sram_addr;
+       int rx_desc_sram_size;
+       u8 rxq_mask;
+       int rxq_primary;
+       struct napi_struct napi;
+       struct rx_queue rxq[8];
 
-       u32 rx_int_coal;
-       u32 tx_int_coal;
-       struct mii_if_info mii;
+       /*
+        * TX state.
+        */
+       int default_tx_ring_size;
+       unsigned long tx_desc_sram_addr;
+       int tx_desc_sram_size;
+       u8 txq_mask;
+       int txq_primary;
+       struct tx_queue txq[8];
+#ifdef MV643XX_ETH_TX_FAST_REFILL
+       int tx_clean_threshold;
+#endif
 };
 
 
 /* port register accessors **************************************************/
-static inline u32 rdl(struct mv643xx_private *mp, int offset)
+static inline u32 rdl(struct mv643xx_eth_private *mp, int offset)
 {
-       return readl(mp->shared->eth_base + offset);
+       return readl(mp->shared->base + offset);
 }
 
-static inline void wrl(struct mv643xx_private *mp, int offset, u32 data)
+static inline void wrl(struct mv643xx_eth_private *mp, int offset, u32 data)
 {
-       writel(data, mp->shared->eth_base + offset);
+       writel(data, mp->shared->base + offset);
 }
 
 
 /* rxq/txq helper functions *************************************************/
-static void mv643xx_eth_port_enable_rx(struct mv643xx_private *mp,
-                                       unsigned int queues)
+static struct mv643xx_eth_private *rxq_to_mp(struct rx_queue *rxq)
+{
+       return container_of(rxq, struct mv643xx_eth_private, rxq[rxq->index]);
+}
+
+static struct mv643xx_eth_private *txq_to_mp(struct tx_queue *txq)
 {
-       wrl(mp, RXQ_COMMAND(mp->port_num), queues);
+       return container_of(txq, struct mv643xx_eth_private, txq[txq->index]);
 }
 
-static unsigned int mv643xx_eth_port_disable_rx(struct mv643xx_private *mp)
+static void rxq_enable(struct rx_queue *rxq)
 {
-       unsigned int port_num = mp->port_num;
-       u32 queues;
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+       wrl(mp, RXQ_COMMAND(mp->port_num), 1 << rxq->index);
+}
 
-       /* Stop Rx port activity. Check port Rx activity. */
-       queues = rdl(mp, RXQ_COMMAND(port_num)) & 0xFF;
-       if (queues) {
-               /* Issue stop command for active queues only */
-               wrl(mp, RXQ_COMMAND(port_num), (queues << 8));
+static void rxq_disable(struct rx_queue *rxq)
+{
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+       u8 mask = 1 << rxq->index;
 
-               /* Wait for all Rx activity to terminate. */
-               /* Check port cause register that all Rx queues are stopped */
-               while (rdl(mp, RXQ_COMMAND(port_num)) & 0xFF)
-                       udelay(10);
-       }
+       wrl(mp, RXQ_COMMAND(mp->port_num), mask << 8);
+       while (rdl(mp, RXQ_COMMAND(mp->port_num)) & 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;
 
-       return queues;
+       addr = (u32)txq->tx_desc_dma;
+       addr += txq->tx_curr_desc * sizeof(struct tx_desc);
+       wrl(mp, off, addr);
 }
 
-static void mv643xx_eth_port_enable_tx(struct mv643xx_private *mp,
-                                       unsigned int queues)
+static void txq_enable(struct tx_queue *txq)
 {
-       wrl(mp, TXQ_COMMAND(mp->port_num), queues);
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       wrl(mp, TXQ_COMMAND(mp->port_num), 1 << txq->index);
 }
 
-static unsigned int mv643xx_eth_port_disable_tx(struct mv643xx_private *mp)
+static void txq_disable(struct tx_queue *txq)
 {
-       unsigned int port_num = mp->port_num;
-       u32 queues;
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       u8 mask = 1 << txq->index;
 
-       /* Stop Tx port activity. Check port Tx activity. */
-       queues = rdl(mp, TXQ_COMMAND(port_num)) & 0xFF;
-       if (queues) {
-               /* Issue stop command for active queues only */
-               wrl(mp, TXQ_COMMAND(port_num), (queues << 8));
+       wrl(mp, TXQ_COMMAND(mp->port_num), mask << 8);
+       while (rdl(mp, TXQ_COMMAND(mp->port_num)) & mask)
+               udelay(10);
+}
 
-               /* Wait for all Tx activity to terminate. */
-               /* Check port cause register that all Tx queues are stopped */
-               while (rdl(mp, TXQ_COMMAND(port_num)) & 0xFF)
-                       udelay(10);
+static void __txq_maybe_wake(struct tx_queue *txq)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
 
-               /* Wait for Tx FIFO to empty */
-               while (rdl(mp, PORT_STATUS(port_num)) & ETH_PORT_TX_FIFO_EMPTY)
-                       udelay(10);
-       }
+       /*
+        * netif_{stop,wake}_queue() flow control only applies to
+        * the primary queue.
+        */
+       BUG_ON(txq->index != mp->txq_primary);
 
-       return queues;
+       if (txq->tx_ring_size - txq->tx_desc_count >= MAX_DESCS_PER_SKB)
+               netif_wake_queue(mp->dev);
 }
 
 
 /* rx ***********************************************************************/
-static void mv643xx_eth_free_completed_tx_descs(struct net_device *dev);
+static void txq_reclaim(struct tx_queue *txq, int force);
 
-/*
- * eth_rx_return_buff - Returns a Rx buffer back to the Rx ring.
- *
- * DESCRIPTION:
- *     This routine returns a Rx buffer back to the Rx ring. It retrieves the
- *     next 'used' descriptor and attached the returned buffer to it.
- *     In case the Rx ring was in "resource error" condition, where there are
- *     no available Rx resources, the function resets the resource error flag.
- *
- * INPUT:
- *     struct mv643xx_private  *mp             Ethernet Port Control srtuct.
- *     struct pkt_info         *p_pkt_info     Information on returned buffer.
- *
- * OUTPUT:
- *     New available Rx resource in Rx descriptor ring.
- *
- * RETURN:
- *     ETH_ERROR in case the routine can not access Rx desc ring.
- *     ETH_OK otherwise.
- */
-static ETH_FUNC_RET_STATUS eth_rx_return_buff(struct mv643xx_private *mp,
-                                               struct pkt_info *p_pkt_info)
+static void rxq_refill(struct rx_queue *rxq)
 {
-       int used_rx_desc;       /* Where to return Rx resource */
-       volatile struct eth_rx_desc *p_used_rx_desc;
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
        unsigned long flags;
 
        spin_lock_irqsave(&mp->lock, flags);
 
-       /* Get 'used' Rx descriptor */
-       used_rx_desc = mp->rx_used_desc_q;
-       p_used_rx_desc = &mp->p_rx_desc_area[used_rx_desc];
+       while (rxq->rx_desc_count < rxq->rx_ring_size) {
+               int skb_size;
+               struct sk_buff *skb;
+               int unaligned;
+               int rx;
 
-       p_used_rx_desc->buf_ptr = p_pkt_info->buf_ptr;
-       p_used_rx_desc->buf_size = p_pkt_info->byte_cnt;
-       mp->rx_skb[used_rx_desc] = p_pkt_info->return_info;
+               /*
+                * 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;
 
-       /* Flush the write pipe */
+               /*
+                * 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;
 
-       /* Return the descriptor to DMA ownership */
-       wmb();
-       p_used_rx_desc->cmd_sts =
-                       ETH_BUFFER_OWNED_BY_DMA | ETH_RX_ENABLE_INTERRUPT;
-       wmb();
+               skb = dev_alloc_skb(skb_size + dma_get_cache_alignment() - 1);
+               if (skb == NULL)
+                       break;
 
-       /* Move the used descriptor pointer to the next descriptor */
-       mp->rx_used_desc_q = (used_rx_desc + 1) % mp->rx_ring_size;
+               unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
+               if (unaligned)
+                       skb_reserve(skb, dma_get_cache_alignment() - unaligned);
 
-       /* Any Rx return cancels the Rx resource error status */
-       mp->rx_resource_err = 0;
+               rxq->rx_desc_count++;
+               rx = rxq->rx_used_desc;
+               rxq->rx_used_desc = (rx + 1) % rxq->rx_ring_size;
 
-       spin_unlock_irqrestore(&mp->lock, flags);
+               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();
 
-       return ETH_OK;
-}
+               /*
+                * 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);
+       }
 
-/*
- * mv643xx_eth_rx_refill_descs
- *
- * Fills / refills RX queue on a certain gigabit ethernet port
- *
- * Input :     pointer to ethernet interface network device structure
- * Output :    N/A
- */
-static void mv643xx_eth_rx_refill_descs(struct net_device *dev)
-{
-       struct mv643xx_private *mp = netdev_priv(dev);
-       struct pkt_info pkt_info;
-       struct sk_buff *skb;
-       int unaligned;
+       if (rxq->rx_desc_count != rxq->rx_ring_size)
+               mod_timer(&rxq->rx_oom, jiffies + (HZ / 10));
 
-       while (mp->rx_desc_count < mp->rx_ring_size) {
-               skb = dev_alloc_skb(ETH_RX_SKB_SIZE + dma_get_cache_alignment());
-               if (!skb)
-                       break;
-               mp->rx_desc_count++;
-               unaligned = (u32)skb->data & (dma_get_cache_alignment() - 1);
-               if (unaligned)
-                       skb_reserve(skb, dma_get_cache_alignment() - unaligned);
-               pkt_info.cmd_sts = ETH_RX_ENABLE_INTERRUPT;
-               pkt_info.byte_cnt = ETH_RX_SKB_SIZE;
-               pkt_info.buf_ptr = dma_map_single(NULL, skb->data,
-                                       ETH_RX_SKB_SIZE, DMA_FROM_DEVICE);
-               pkt_info.return_info = skb;
-               if (eth_rx_return_buff(mp, &pkt_info) != ETH_OK) {
-                       printk(KERN_ERR
-                               "%s: Error allocating RX Ring\n", dev->name);
-                       break;
-               }
-               skb_reserve(skb, ETH_HW_IP_ALIGN);
-       }
-       /*
-        * If RX ring is empty of SKB, set a timer to try allocating
-        * again at a later time.
-        */
-       if (mp->rx_desc_count == 0) {
-               printk(KERN_INFO "%s: Rx ring is empty\n", dev->name);
-               mp->timeout.expires = jiffies + (HZ / 10);      /* 100 mSec */
-               add_timer(&mp->timeout);
-       }
+       spin_unlock_irqrestore(&mp->lock, flags);
 }
 
-/*
- * mv643xx_eth_rx_refill_descs_timer_wrapper
- *
- * Timer routine to wake up RX queue filling task. This function is
- * used only in case the RX queue is empty, and all alloc_skb has
- * failed (due to out of memory event).
- *
- * Input :     pointer to ethernet interface network device structure
- * Output :    N/A
- */
-static inline void mv643xx_eth_rx_refill_descs_timer_wrapper(unsigned long data)
+static inline void rxq_refill_timer_wrapper(unsigned long data)
 {
-       mv643xx_eth_rx_refill_descs((struct net_device *)data);
+       rxq_refill((struct rx_queue *)data);
 }
 
-/*
- * eth_port_receive - Get received information from Rx ring.
- *
- * DESCRIPTION:
- *     This routine returns the received data to the caller. There is no
- *     data copying during routine operation. All information is returned
- *     using pointer to packet information struct passed from the caller.
- *     If the routine exhausts Rx ring resources then the resource error flag
- *     is set.
- *
- * INPUT:
- *     struct mv643xx_private  *mp             Ethernet Port Control srtuct.
- *     struct pkt_info         *p_pkt_info     User packet buffer.
- *
- * OUTPUT:
- *     Rx ring current and used indexes are updated.
- *
- * RETURN:
- *     ETH_ERROR in case the routine can not access Rx desc ring.
- *     ETH_QUEUE_FULL if Rx ring resources are exhausted.
- *     ETH_END_OF_JOB if there is no received data.
- *     ETH_OK otherwise.
- */
-static ETH_FUNC_RET_STATUS eth_port_receive(struct mv643xx_private *mp,
-                                               struct pkt_info *p_pkt_info)
+static int rxq_process(struct rx_queue *rxq, int budget)
 {
-       int rx_next_curr_desc, rx_curr_desc, rx_used_desc;
-       volatile struct eth_rx_desc *p_rx_desc;
-       unsigned int command_status;
-       unsigned long flags;
-
-       /* Do not process Rx ring in case of Rx ring resource error */
-       if (mp->rx_resource_err)
-               return ETH_QUEUE_FULL;
-
-       spin_lock_irqsave(&mp->lock, flags);
-
-       /* Get the Rx Desc ring 'curr and 'used' indexes */
-       rx_curr_desc = mp->rx_curr_desc_q;
-       rx_used_desc = mp->rx_used_desc_q;
-
-       p_rx_desc = &mp->p_rx_desc_area[rx_curr_desc];
-
-       /* The following parameters are used to save readings from memory */
-       command_status = p_rx_desc->cmd_sts;
-       rmb();
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+       struct net_device_stats *stats = &mp->dev->stats;
+       int rx;
 
-       /* Nothing to receive... */
-       if (command_status & (ETH_BUFFER_OWNED_BY_DMA)) {
-               spin_unlock_irqrestore(&mp->lock, flags);
-               return ETH_END_OF_JOB;
-       }
+       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;
 
-       p_pkt_info->byte_cnt = p_rx_desc->byte_cnt - ETH_HW_IP_ALIGN;
-       p_pkt_info->cmd_sts = command_status;
-       p_pkt_info->buf_ptr = p_rx_desc->buf_ptr + ETH_HW_IP_ALIGN;
-       p_pkt_info->return_info = mp->rx_skb[rx_curr_desc];
-       p_pkt_info->l4i_chk = p_rx_desc->buf_size;
-
-       /*
-        * Clean the return info field to indicate that the
-        * packet has been moved to the upper layers
-        */
-       mp->rx_skb[rx_curr_desc] = NULL;
+               spin_lock_irqsave(&mp->lock, flags);
 
-       /* Update current index in data structure */
-       rx_next_curr_desc = (rx_curr_desc + 1) % mp->rx_ring_size;
-       mp->rx_curr_desc_q = rx_next_curr_desc;
+               rx_desc = &rxq->rx_desc_area[rxq->rx_curr_desc];
 
-       /* Rx descriptors exhausted. Set the Rx ring resource error flag */
-       if (rx_next_curr_desc == rx_used_desc)
-               mp->rx_resource_err = 1;
+               cmd_sts = rx_desc->cmd_sts;
+               if (cmd_sts & BUFFER_OWNED_BY_DMA) {
+                       spin_unlock_irqrestore(&mp->lock, flags);
+                       break;
+               }
+               rmb();
 
-       spin_unlock_irqrestore(&mp->lock, flags);
+               skb = rxq->rx_skb[rxq->rx_curr_desc];
+               rxq->rx_skb[rxq->rx_curr_desc] = NULL;
 
-       return ETH_OK;
-}
+               rxq->rx_curr_desc = (rxq->rx_curr_desc + 1) % rxq->rx_ring_size;
 
-/*
- * mv643xx_eth_receive
- *
- * This function is forward packets that are received from the port's
- * queues toward kernel core or FastRoute them to another interface.
- *
- * Input :     dev - a pointer to the required interface
- *             max - maximum number to receive (0 means unlimted)
- *
- * Output :    number of served packets
- */
-static int mv643xx_eth_receive_queue(struct net_device *dev, int budget)
-{
-       struct mv643xx_private *mp = netdev_priv(dev);
-       struct net_device_stats *stats = &dev->stats;
-       unsigned int received_packets = 0;
-       struct sk_buff *skb;
-       struct pkt_info pkt_info;
+               spin_unlock_irqrestore(&mp->lock, flags);
 
-       while (budget-- > 0 && eth_port_receive(mp, &pkt_info) == ETH_OK) {
-               dma_unmap_single(NULL, pkt_info.buf_ptr, ETH_RX_SKB_SIZE,
-                                                       DMA_FROM_DEVICE);
-               mp->rx_desc_count--;
-               received_packets++;
+               dma_unmap_single(NULL, rx_desc->buf_ptr + 2,
+                                rx_desc->buf_size, DMA_FROM_DEVICE);
+               rxq->rx_desc_count--;
+               rx++;
 
                /*
                 * Update statistics.
-                * Note byte count includes 4 byte CRC count
+                *
+                * Note that the descriptor byte count includes 2 dummy
+                * bytes automatically inserted by the hardware at the
+                * start of the packet (which we don't count), and a 4
+                * byte CRC at the end of the packet (which we do count).
                 */
                stats->rx_packets++;
-               stats->rx_bytes += pkt_info.byte_cnt;
-               skb = pkt_info.return_info;
+               stats->rx_bytes += rx_desc->byte_cnt - 2;
+
                /*
-                * In case received a packet without first / last bits on OR
-                * the error summary bit is on, the packets needs to be dropeed.
+                * In case we received a packet without first / last bits
+                * on, or the error summary bit is set, the packet needs
+                * to be dropped.
                 */
-               if (((pkt_info.cmd_sts
-                               & (ETH_RX_FIRST_DESC | ETH_RX_LAST_DESC)) !=
-                                       (ETH_RX_FIRST_DESC | ETH_RX_LAST_DESC))
-                               || (pkt_info.cmd_sts & ETH_ERROR_SUMMARY)) {
+               if (((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
+                                       (RX_FIRST_DESC | RX_LAST_DESC))
+                               || (cmd_sts & ERROR_SUMMARY)) {
                        stats->rx_dropped++;
-                       if ((pkt_info.cmd_sts & (ETH_RX_FIRST_DESC |
-                                                       ETH_RX_LAST_DESC)) !=
-                               (ETH_RX_FIRST_DESC | ETH_RX_LAST_DESC)) {
+
+                       if ((cmd_sts & (RX_FIRST_DESC | RX_LAST_DESC)) !=
+                               (RX_FIRST_DESC | RX_LAST_DESC)) {
                                if (net_ratelimit())
-                                       printk(KERN_ERR
-                                               "%s: Received packet spread "
-                                               "on multiple descriptors\n",
-                                               dev->name);
+                                       dev_printk(KERN_ERR, &mp->dev->dev,
+                                                  "received packet spanning "
+                                                  "multiple descriptors\n");
                        }
-                       if (pkt_info.cmd_sts & ETH_ERROR_SUMMARY)
+
+                       if (cmd_sts & ERROR_SUMMARY)
                                stats->rx_errors++;
 
                        dev_kfree_skb_irq(skb);
@@ -736,132 +602,124 @@ static int mv643xx_eth_receive_queue(struct net_device *dev, int budget)
                         * The -4 is for the CRC in the trailer of the
                         * received packet
                         */
-                       skb_put(skb, pkt_info.byte_cnt - 4);
+                       skb_put(skb, rx_desc->byte_cnt - 2 - 4);
 
-                       if (pkt_info.cmd_sts & ETH_LAYER_4_CHECKSUM_OK) {
+                       if (cmd_sts & LAYER_4_CHECKSUM_OK) {
                                skb->ip_summed = CHECKSUM_UNNECESSARY;
                                skb->csum = htons(
-                                       (pkt_info.cmd_sts & 0x0007fff8) >> 3);
+                                       (cmd_sts & 0x0007fff8) >> 3);
                        }
-                       skb->protocol = eth_type_trans(skb, dev);
-#ifdef MV643XX_NAPI
+                       skb->protocol = eth_type_trans(skb, mp->dev);
+#ifdef MV643XX_ETH_NAPI
                        netif_receive_skb(skb);
 #else
                        netif_rx(skb);
 #endif
                }
-               dev->last_rx = jiffies;
+
+               mp->dev->last_rx = jiffies;
        }
-       mv643xx_eth_rx_refill_descs(dev);       /* Fill RX ring with skb's */
 
-       return received_packets;
+       rxq_refill(rxq);
+
+       return rx;
 }
 
-#ifdef MV643XX_NAPI
-/*
- * mv643xx_poll
- *
- * This function is used in case of NAPI
- */
-static int mv643xx_poll(struct napi_struct *napi, int budget)
+#ifdef MV643XX_ETH_NAPI
+static int mv643xx_eth_poll(struct napi_struct *napi, int budget)
 {
-       struct mv643xx_private *mp = container_of(napi, struct mv643xx_private, napi);
-       struct net_device *dev = mp->dev;
-       unsigned int port_num = mp->port_num;
-       int work_done;
+       struct mv643xx_eth_private *mp;
+       int rx;
+       int i;
+
+       mp = container_of(napi, struct mv643xx_eth_private, napi);
 
-#ifdef MV643XX_TX_FAST_REFILL
+#ifdef MV643XX_ETH_TX_FAST_REFILL
        if (++mp->tx_clean_threshold > 5) {
-               mv643xx_eth_free_completed_tx_descs(dev);
                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);
+               }
        }
 #endif
 
-       work_done = 0;
-       if ((rdl(mp, RXQ_CURRENT_DESC_PTR(port_num)))
-           != (u32) mp->rx_used_desc_q)
-               work_done = mv643xx_eth_receive_queue(dev, budget);
+       rx = 0;
+       for (i = 7; rx < budget && i >= 0; i--)
+               if (mp->rxq_mask & (1 << i))
+                       rx += rxq_process(mp->rxq + i, budget - rx);
 
-       if (work_done < budget) {
-               netif_rx_complete(dev, napi);
-               wrl(mp, INT_CAUSE(port_num), 0);
-               wrl(mp, INT_CAUSE_EXT(port_num), 0);
-               wrl(mp, INT_MASK(port_num), INT_RX | INT_EXT);
+       if (rx < budget) {
+               netif_rx_complete(mp->dev, napi);
+               wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
        }
 
-       return work_done;
+       return rx;
 }
 #endif
 
 
 /* tx ***********************************************************************/
-/**
- * has_tiny_unaligned_frags - check if skb has any small, unaligned fragments
- *
- * Hardware can't handle unaligned fragments smaller than 9 bytes.
- * This helper function detects that case.
- */
-
 static inline unsigned int has_tiny_unaligned_frags(struct sk_buff *skb)
 {
-       unsigned int frag;
-       skb_frag_t *fragp;
+       int frag;
 
        for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
-               fragp = &skb_shinfo(skb)->frags[frag];
-               if (fragp->size <= 8 && fragp->page_offset & 0x7)
+               skb_frag_t *fragp = &skb_shinfo(skb)->frags[frag];
+               if (fragp->size <= 8 && fragp->page_offset & 7)
                        return 1;
        }
+
        return 0;
 }
 
-/**
- * eth_alloc_tx_desc_index - return the index of the next available tx desc
- */
-static int eth_alloc_tx_desc_index(struct mv643xx_private *mp)
+static int txq_alloc_desc_index(struct tx_queue *txq)
 {
        int tx_desc_curr;
 
-       BUG_ON(mp->tx_desc_count >= mp->tx_ring_size);
+       BUG_ON(txq->tx_desc_count >= txq->tx_ring_size);
 
-       tx_desc_curr = mp->tx_curr_desc_q;
-       mp->tx_curr_desc_q = (tx_desc_curr + 1) % mp->tx_ring_size;
+       tx_desc_curr = txq->tx_curr_desc;
+       txq->tx_curr_desc = (tx_desc_curr + 1) % txq->tx_ring_size;
 
-       BUG_ON(mp->tx_curr_desc_q == mp->tx_used_desc_q);
+       BUG_ON(txq->tx_curr_desc == txq->tx_used_desc);
 
        return tx_desc_curr;
 }
 
-/**
- * eth_tx_fill_frag_descs - fill tx hw descriptors for an skb's fragments.
- *
- * Ensure the data for each fragment to be transmitted is mapped properly,
- * then fill in descriptors in the tx hw queue.
- */
-static void eth_tx_fill_frag_descs(struct mv643xx_private *mp,
-                                  struct sk_buff *skb)
+static void txq_submit_frag_skb(struct tx_queue *txq, struct sk_buff *skb)
 {
+       int nr_frags = skb_shinfo(skb)->nr_frags;
        int frag;
-       int tx_index;
-       struct eth_tx_desc *desc;
 
-       for (frag = 0; frag < skb_shinfo(skb)->nr_frags; frag++) {
-               skb_frag_t *this_frag = &skb_shinfo(skb)->frags[frag];
-
-               tx_index = eth_alloc_tx_desc_index(mp);
-               desc = &mp->p_tx_desc_area[tx_index];
-
-               desc->cmd_sts = ETH_BUFFER_OWNED_BY_DMA;
-               /* Last Frag enables interrupt and frees the skb */
-               if (frag == (skb_shinfo(skb)->nr_frags - 1)) {
-                       desc->cmd_sts |= ETH_ZERO_PADDING |
-                                        ETH_TX_LAST_DESC |
-                                        ETH_TX_ENABLE_INTERRUPT;
-                       mp->tx_skb[tx_index] = skb;
-               } else
-                       mp->tx_skb[tx_index] = NULL;
-
-               desc = &mp->p_tx_desc_area[tx_index];
+       for (frag = 0; frag < nr_frags; frag++) {
+               skb_frag_t *this_frag;
+               int tx_index;
+               struct tx_desc *desc;
+
+               this_frag = &skb_shinfo(skb)->frags[frag];
+               tx_index = txq_alloc_desc_index(txq);
+               desc = &txq->tx_desc_area[tx_index];
+
+               /*
+                * The last fragment will generate an interrupt
+                * which will free the skb on TX completion.
+                */
+               if (frag == nr_frags - 1) {
+                       desc->cmd_sts = BUFFER_OWNED_BY_DMA |
+                                       ZERO_PADDING | TX_LAST_DESC |
+                                       TX_ENABLE_INTERRUPT;
+                       txq->tx_skb[tx_index] = skb;
+               } else {
+                       desc->cmd_sts = BUFFER_OWNED_BY_DMA;
+                       txq->tx_skb[tx_index] = NULL;
+               }
+
                desc->l4i_chk = 0;
                desc->byte_cnt = this_frag->size;
                desc->buf_ptr = dma_map_page(NULL, this_frag->page,
@@ -876,52 +734,68 @@ static inline __be16 sum16_as_be(__sum16 sum)
        return (__force __be16)sum;
 }
 
-/**
- * eth_tx_submit_descs_for_skb - submit data from an skb to the tx hw
- *
- * Ensure the data for an skb to be transmitted is mapped properly,
- * then fill in descriptors in the tx hw queue and start the hardware.
- */
-static void eth_tx_submit_descs_for_skb(struct mv643xx_private *mp,
-                                       struct sk_buff *skb)
+static void 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 eth_tx_desc *desc;
+       struct tx_desc *desc;
        u32 cmd_sts;
        int length;
-       int nr_frags = skb_shinfo(skb)->nr_frags;
 
-       cmd_sts = ETH_TX_FIRST_DESC | ETH_GEN_CRC | ETH_BUFFER_OWNED_BY_DMA;
+       cmd_sts = TX_FIRST_DESC | GEN_CRC | BUFFER_OWNED_BY_DMA;
 
-       tx_index = eth_alloc_tx_desc_index(mp);
-       desc = &mp->p_tx_desc_area[tx_index];
+       tx_index = txq_alloc_desc_index(txq);
+       desc = &txq->tx_desc_area[tx_index];
 
        if (nr_frags) {
-               eth_tx_fill_frag_descs(mp, skb);
+               txq_submit_frag_skb(txq, skb);
 
                length = skb_headlen(skb);
-               mp->tx_skb[tx_index] = NULL;
+               txq->tx_skb[tx_index] = NULL;
        } else {
-               cmd_sts |= ETH_ZERO_PADDING |
-                          ETH_TX_LAST_DESC |
-                          ETH_TX_ENABLE_INTERRUPT;
+               cmd_sts |= ZERO_PADDING | TX_LAST_DESC | TX_ENABLE_INTERRUPT;
                length = skb->len;
-               mp->tx_skb[tx_index] = skb;
+               txq->tx_skb[tx_index] = skb;
        }
 
        desc->byte_cnt = length;
        desc->buf_ptr = dma_map_single(NULL, skb->data, length, DMA_TO_DEVICE);
 
        if (skb->ip_summed == CHECKSUM_PARTIAL) {
-               BUG_ON(skb->protocol != htons(ETH_P_IP));
+               int mac_hdr_len;
+
+               BUG_ON(skb->protocol != htons(ETH_P_IP) &&
+                      skb->protocol != htons(ETH_P_8021Q));
 
-               cmd_sts |= ETH_GEN_TCP_UDP_CHECKSUM |
-                          ETH_GEN_IP_V_4_CHECKSUM  |
-                          ip_hdr(skb)->ihl << ETH_TX_IHL_SHIFT;
+               cmd_sts |= GEN_TCP_UDP_CHECKSUM |
+                          GEN_IP_V4_CHECKSUM   |
+                          ip_hdr(skb)->ihl << TX_IHL_SHIFT;
+
+               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:
+                       cmd_sts |= MAC_HDR_EXTRA_4_BYTES;
+                       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;
+               }
 
                switch (ip_hdr(skb)->protocol) {
                case IPPROTO_UDP:
-                       cmd_sts |= ETH_UDP_FRAME;
+                       cmd_sts |= UDP_FRAME;
                        desc->l4i_chk = ntohs(sum16_as_be(udp_hdr(skb)->check));
                        break;
                case IPPROTO_TCP:
@@ -932,7 +806,7 @@ static void eth_tx_submit_descs_for_skb(struct mv643xx_private *mp,
                }
        } else {
                /* Errata BTS #50, IHL must be 5 if no HW checksum */
-               cmd_sts |= 5 << ETH_TX_IHL_SHIFT;
+               cmd_sts |= 5 << TX_IHL_SHIFT;
                desc->l4i_chk = 0;
        }
 
@@ -940,48 +814,57 @@ static void eth_tx_submit_descs_for_skb(struct mv643xx_private *mp,
        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));
+
        /* ensure all descriptors are written before poking hardware */
        wmb();
-       mv643xx_eth_port_enable_tx(mp, 1);
+       txq_enable(txq);
 
-       mp->tx_desc_count += nr_frags + 1;
+       txq->tx_desc_count += nr_frags + 1;
 }
 
-/**
- * mv643xx_eth_start_xmit - queue an skb to the hardware for transmission
- *
- */
-static int mv643xx_eth_start_xmit(struct sk_buff *skb, struct net_device *dev)
+static int mv643xx_eth_xmit(struct sk_buff *skb, struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
        struct net_device_stats *stats = &dev->stats;
+       struct tx_queue *txq;
        unsigned long flags;
 
-       BUG_ON(netif_queue_stopped(dev));
-
        if (has_tiny_unaligned_frags(skb) && __skb_linearize(skb)) {
                stats->tx_dropped++;
-               printk(KERN_DEBUG "%s: failed to linearize tiny "
-                               "unaligned fragment\n", dev->name);
+               dev_printk(KERN_DEBUG, &dev->dev,
+                          "failed to linearize skb with tiny "
+                          "unaligned fragment\n");
                return NETDEV_TX_BUSY;
        }
 
        spin_lock_irqsave(&mp->lock, flags);
 
-       if (mp->tx_ring_size - mp->tx_desc_count < MAX_DESCS_PER_SKB) {
-               printk(KERN_ERR "%s: transmit with queue full\n", dev->name);
-               netif_stop_queue(dev);
+       txq = mp->txq + mp->txq_primary;
+
+       if (txq->tx_ring_size - txq->tx_desc_count < MAX_DESCS_PER_SKB) {
                spin_unlock_irqrestore(&mp->lock, flags);
-               return NETDEV_TX_BUSY;
+               if (txq->index == mp->txq_primary && net_ratelimit())
+                       dev_printk(KERN_ERR, &dev->dev,
+                                  "primary tx queue full?!\n");
+               kfree_skb(skb);
+               return NETDEV_TX_OK;
        }
 
-       eth_tx_submit_descs_for_skb(mp, skb);
+       txq_submit_skb(txq, skb);
        stats->tx_bytes += skb->len;
        stats->tx_packets++;
        dev->trans_start = jiffies;
 
-       if (mp->tx_ring_size - mp->tx_desc_count < MAX_DESCS_PER_SKB)
-               netif_stop_queue(dev);
+       if (txq->index == mp->txq_primary) {
+               int entries_left;
+
+               entries_left = txq->tx_ring_size - txq->tx_desc_count;
+               if (entries_left < MAX_DESCS_PER_SKB)
+                       netif_stop_queue(dev);
+       }
 
        spin_unlock_irqrestore(&mp->lock, flags);
 
@@ -989,34 +872,117 @@ static int mv643xx_eth_start_xmit(struct sk_buff *skb, struct net_device *dev)
 }
 
 
-/* mii management interface *************************************************/
-static int ethernet_phy_get(struct mv643xx_private *mp);
-
+/* tx rate control **********************************************************/
 /*
- * eth_port_read_smi_reg - Read PHY registers
- *
- * DESCRIPTION:
- *     This routine utilize the SMI interface to interact with the PHY in
- *     order to perform PHY register read.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *     unsigned int    phy_reg         PHY register address offset.
- *     unsigned int    *value          Register value buffer.
- *
- * OUTPUT:
- *     Write the value of a specified PHY register into given buffer.
- *
- * RETURN:
- *     false if the PHY is busy or read data is not in valid state.
- *     true otherwise.
- *
+ * Set total maximum TX rate (shared by all TX queues for this port)
+ * to 'rate' bits per second, with a maximum burst of 'burst' bytes.
  */
-static void eth_port_read_smi_reg(struct mv643xx_private *mp,
-                               unsigned int phy_reg, unsigned int *value)
+static void tx_set_rate(struct mv643xx_eth_private *mp, int rate, int burst)
+{
+       int token_rate;
+       int mtu;
+       int bucket_size;
+
+       token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
+       if (token_rate > 1023)
+               token_rate = 1023;
+
+       mtu = (mp->dev->mtu + 255) >> 8;
+       if (mtu > 63)
+               mtu = 63;
+
+       bucket_size = (burst + 255) >> 8;
+       if (bucket_size > 65535)
+               bucket_size = 65535;
+
+       if (mp->shared->tx_bw_control_moved) {
+               wrl(mp, TX_BW_RATE_MOVED(mp->port_num), token_rate);
+               wrl(mp, TX_BW_MTU_MOVED(mp->port_num), mtu);
+               wrl(mp, TX_BW_BURST_MOVED(mp->port_num), bucket_size);
+       } else {
+               wrl(mp, TX_BW_RATE(mp->port_num), token_rate);
+               wrl(mp, TX_BW_MTU(mp->port_num), mtu);
+               wrl(mp, TX_BW_BURST(mp->port_num), bucket_size);
+       }
+}
+
+static void txq_set_rate(struct tx_queue *txq, int rate, int burst)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int token_rate;
+       int bucket_size;
+
+       token_rate = ((rate / 1000) * 64) / (mp->shared->t_clk / 1000);
+       if (token_rate > 1023)
+               token_rate = 1023;
+
+       bucket_size = (burst + 255) >> 8;
+       if (bucket_size > 65535)
+               bucket_size = 65535;
+
+       wrl(mp, TXQ_BW_TOKENS(mp->port_num, txq->index), token_rate << 14);
+       wrl(mp, TXQ_BW_CONF(mp->port_num, txq->index),
+                       (bucket_size << 10) | token_rate);
+}
+
+static void txq_set_fixed_prio_mode(struct tx_queue *txq)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int off;
+       u32 val;
+
+       /*
+        * Turn on fixed priority mode.
+        */
+       if (mp->shared->tx_bw_control_moved)
+               off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+       else
+               off = TXQ_FIX_PRIO_CONF(mp->port_num);
+
+       val = rdl(mp, off);
+       val |= 1 << txq->index;
+       wrl(mp, off, val);
+}
+
+static void txq_set_wrr(struct tx_queue *txq, int weight)
+{
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
+       int off;
+       u32 val;
+
+       /*
+        * Turn off fixed priority mode.
+        */
+       if (mp->shared->tx_bw_control_moved)
+               off = TXQ_FIX_PRIO_CONF_MOVED(mp->port_num);
+       else
+               off = TXQ_FIX_PRIO_CONF(mp->port_num);
+
+       val = rdl(mp, off);
+       val &= ~(1 << txq->index);
+       wrl(mp, off, val);
+
+       /*
+        * Configure WRR weight for this queue.
+        */
+       off = TXQ_BW_WRR_CONF(mp->port_num, txq->index);
+
+       val = rdl(mp, off);
+       val = (val & ~0xff) | (weight & 0xff);
+       wrl(mp, off, val);
+}
+
+
+/* mii management interface *************************************************/
+#define SMI_BUSY               0x10000000
+#define SMI_READ_VALID         0x08000000
+#define SMI_OPCODE_READ                0x04000000
+#define SMI_OPCODE_WRITE       0x00000000
+
+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->eth_base + SMI_REG;
-       int phy_addr = ethernet_phy_get(mp);
+       void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
        unsigned long flags;
        int i;
 
@@ -1024,7 +990,7 @@ static void eth_port_read_smi_reg(struct mv643xx_private *mp,
        spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
 
        /* wait for the SMI register to become available */
-       for (i = 0; readl(smi_reg) & ETH_SMI_BUSY; i++) {
+       for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
                if (i == 1000) {
                        printk("%s: PHY busy timeout\n", mp->dev->name);
                        goto out;
@@ -1032,11 +998,10 @@ static void eth_port_read_smi_reg(struct mv643xx_private *mp,
                udelay(10);
        }
 
-       writel((phy_addr << 16) | (phy_reg << 21) | ETH_SMI_OPCODE_READ,
-               smi_reg);
+       writel(SMI_OPCODE_READ | (reg << 21) | (addr << 16), smi_reg);
 
        /* now wait for the data to be valid */
-       for (i = 0; !(readl(smi_reg) & ETH_SMI_READ_VALID); i++) {
+       for (i = 0; !(readl(smi_reg) & SMI_READ_VALID); i++) {
                if (i == 1000) {
                        printk("%s: PHY read timeout\n", mp->dev->name);
                        goto out;
@@ -1049,31 +1014,11 @@ out:
        spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
 }
 
-/*
- * eth_port_write_smi_reg - Write to PHY registers
- *
- * DESCRIPTION:
- *     This routine utilize the SMI interface to interact with the PHY in
- *     order to perform writes to PHY registers.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *     unsigned int    phy_reg         PHY register address offset.
- *     unsigned int    value           Register value.
- *
- * OUTPUT:
- *     Write the given value to the specified PHY register.
- *
- * RETURN:
- *     false if the PHY is busy.
- *     true otherwise.
- *
- */
-static void eth_port_write_smi_reg(struct mv643xx_private *mp,
-                                  unsigned int phy_reg, unsigned int value)
+static void smi_reg_write(struct mv643xx_eth_private *mp,
+                         unsigned int addr,
+                         unsigned int reg, unsigned int value)
 {
-       void __iomem *smi_reg = mp->shared_smi->eth_base + SMI_REG;
-       int phy_addr = ethernet_phy_get(mp);
+       void __iomem *smi_reg = mp->shared_smi->base + SMI_REG;
        unsigned long flags;
        int i;
 
@@ -1081,7 +1026,7 @@ static void eth_port_write_smi_reg(struct mv643xx_private *mp,
        spin_lock_irqsave(&mp->shared_smi->phy_lock, flags);
 
        /* wait for the SMI register to become available */
-       for (i = 0; readl(smi_reg) & ETH_SMI_BUSY; i++) {
+       for (i = 0; readl(smi_reg) & SMI_BUSY; i++) {
                if (i == 1000) {
                        printk("%s: PHY busy timeout\n", mp->dev->name);
                        goto out;
@@ -1089,159 +1034,185 @@ static void eth_port_write_smi_reg(struct mv643xx_private *mp,
                udelay(10);
        }
 
-       writel((phy_addr << 16) | (phy_reg << 21) |
-               ETH_SMI_OPCODE_WRITE | (value & 0xffff), smi_reg);
+       writel(SMI_OPCODE_WRITE | (reg << 21) |
+               (addr << 16) | (value & 0xffff), smi_reg);
 out:
        spin_unlock_irqrestore(&mp->shared_smi->phy_lock, flags);
 }
 
 
 /* mib counters *************************************************************/
-/*
- * eth_clear_mib_counters - Clear all MIB counters
- *
- * DESCRIPTION:
- *     This function clears all MIB counters of a specific ethernet port.
- *     A read from the MIB counter will reset the counter.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     After reading all MIB counters, the counters resets.
- *
- * RETURN:
- *     MIB counter value.
- *
- */
-static void eth_clear_mib_counters(struct mv643xx_private *mp)
+static inline u32 mib_read(struct mv643xx_eth_private *mp, int offset)
+{
+       return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
+}
+
+static void mib_counters_clear(struct mv643xx_eth_private *mp)
 {
-       unsigned int port_num = mp->port_num;
        int i;
 
-       /* Perform dummy reads from MIB counters */
        for (i = 0; i < 0x80; i += 4)
-               rdl(mp, MIB_COUNTERS(port_num) + i);
-}
-
-static inline u32 read_mib(struct mv643xx_private *mp, int offset)
-{
-       return rdl(mp, MIB_COUNTERS(mp->port_num) + offset);
-}
-
-static void eth_update_mib_counters(struct mv643xx_private *mp)
-{
-       struct mv643xx_mib_counters *p = &mp->mib_counters;
-
-       p->good_octets_received += read_mib(mp, 0x00);
-       p->good_octets_received += (u64)read_mib(mp, 0x04) << 32;
-       p->bad_octets_received += read_mib(mp, 0x08);
-       p->internal_mac_transmit_err += read_mib(mp, 0x0c);
-       p->good_frames_received += read_mib(mp, 0x10);
-       p->bad_frames_received += read_mib(mp, 0x14);
-       p->broadcast_frames_received += read_mib(mp, 0x18);
-       p->multicast_frames_received += read_mib(mp, 0x1c);
-       p->frames_64_octets += read_mib(mp, 0x20);
-       p->frames_65_to_127_octets += read_mib(mp, 0x24);
-       p->frames_128_to_255_octets += read_mib(mp, 0x28);
-       p->frames_256_to_511_octets += read_mib(mp, 0x2c);
-       p->frames_512_to_1023_octets += read_mib(mp, 0x30);
-       p->frames_1024_to_max_octets += read_mib(mp, 0x34);
-       p->good_octets_sent += read_mib(mp, 0x38);
-       p->good_octets_sent += (u64)read_mib(mp, 0x3c) << 32;
-       p->good_frames_sent += read_mib(mp, 0x40);
-       p->excessive_collision += read_mib(mp, 0x44);
-       p->multicast_frames_sent += read_mib(mp, 0x48);
-       p->broadcast_frames_sent += read_mib(mp, 0x4c);
-       p->unrec_mac_control_received += read_mib(mp, 0x50);
-       p->fc_sent += read_mib(mp, 0x54);
-       p->good_fc_received += read_mib(mp, 0x58);
-       p->bad_fc_received += read_mib(mp, 0x5c);
-       p->undersize_received += read_mib(mp, 0x60);
-       p->fragments_received += read_mib(mp, 0x64);
-       p->oversize_received += read_mib(mp, 0x68);
-       p->jabber_received += read_mib(mp, 0x6c);
-       p->mac_receive_error += read_mib(mp, 0x70);
-       p->bad_crc_event += read_mib(mp, 0x74);
-       p->collision += read_mib(mp, 0x78);
-       p->late_collision += read_mib(mp, 0x7c);
+               mib_read(mp, i);
+}
+
+static void mib_counters_update(struct mv643xx_eth_private *mp)
+{
+       struct mib_counters *p = &mp->mib_counters;
+
+       p->good_octets_received += mib_read(mp, 0x00);
+       p->good_octets_received += (u64)mib_read(mp, 0x04) << 32;
+       p->bad_octets_received += mib_read(mp, 0x08);
+       p->internal_mac_transmit_err += mib_read(mp, 0x0c);
+       p->good_frames_received += mib_read(mp, 0x10);
+       p->bad_frames_received += mib_read(mp, 0x14);
+       p->broadcast_frames_received += mib_read(mp, 0x18);
+       p->multicast_frames_received += mib_read(mp, 0x1c);
+       p->frames_64_octets += mib_read(mp, 0x20);
+       p->frames_65_to_127_octets += mib_read(mp, 0x24);
+       p->frames_128_to_255_octets += mib_read(mp, 0x28);
+       p->frames_256_to_511_octets += mib_read(mp, 0x2c);
+       p->frames_512_to_1023_octets += mib_read(mp, 0x30);
+       p->frames_1024_to_max_octets += mib_read(mp, 0x34);
+       p->good_octets_sent += mib_read(mp, 0x38);
+       p->good_octets_sent += (u64)mib_read(mp, 0x3c) << 32;
+       p->good_frames_sent += mib_read(mp, 0x40);
+       p->excessive_collision += mib_read(mp, 0x44);
+       p->multicast_frames_sent += mib_read(mp, 0x48);
+       p->broadcast_frames_sent += mib_read(mp, 0x4c);
+       p->unrec_mac_control_received += mib_read(mp, 0x50);
+       p->fc_sent += mib_read(mp, 0x54);
+       p->good_fc_received += mib_read(mp, 0x58);
+       p->bad_fc_received += mib_read(mp, 0x5c);
+       p->undersize_received += mib_read(mp, 0x60);
+       p->fragments_received += mib_read(mp, 0x64);
+       p->oversize_received += mib_read(mp, 0x68);
+       p->jabber_received += mib_read(mp, 0x6c);
+       p->mac_receive_error += mib_read(mp, 0x70);
+       p->bad_crc_event += mib_read(mp, 0x74);
+       p->collision += mib_read(mp, 0x78);
+       p->late_collision += mib_read(mp, 0x7c);
 }
 
 
 /* ethtool ******************************************************************/
-struct mv643xx_stats {
+struct mv643xx_eth_stats {
        char stat_string[ETH_GSTRING_LEN];
        int sizeof_stat;
-       int stat_offset;
+       int netdev_off;
+       int mp_off;
 };
 
-#define MV643XX_STAT(m) FIELD_SIZEOF(struct mv643xx_private, m), \
-                                       offsetof(struct mv643xx_private, m)
-
-static const struct mv643xx_stats mv643xx_gstrings_stats[] = {
-       { "rx_packets", MV643XX_STAT(stats.rx_packets) },
-       { "tx_packets", MV643XX_STAT(stats.tx_packets) },
-       { "rx_bytes", MV643XX_STAT(stats.rx_bytes) },
-       { "tx_bytes", MV643XX_STAT(stats.tx_bytes) },
-       { "rx_errors", MV643XX_STAT(stats.rx_errors) },
-       { "tx_errors", MV643XX_STAT(stats.tx_errors) },
-       { "rx_dropped", MV643XX_STAT(stats.rx_dropped) },
-       { "tx_dropped", MV643XX_STAT(stats.tx_dropped) },
-       { "good_octets_received", MV643XX_STAT(mib_counters.good_octets_received) },
-       { "bad_octets_received", MV643XX_STAT(mib_counters.bad_octets_received) },
-       { "internal_mac_transmit_err", MV643XX_STAT(mib_counters.internal_mac_transmit_err) },
-       { "good_frames_received", MV643XX_STAT(mib_counters.good_frames_received) },
-       { "bad_frames_received", MV643XX_STAT(mib_counters.bad_frames_received) },
-       { "broadcast_frames_received", MV643XX_STAT(mib_counters.broadcast_frames_received) },
-       { "multicast_frames_received", MV643XX_STAT(mib_counters.multicast_frames_received) },
-       { "frames_64_octets", MV643XX_STAT(mib_counters.frames_64_octets) },
-       { "frames_65_to_127_octets", MV643XX_STAT(mib_counters.frames_65_to_127_octets) },
-       { "frames_128_to_255_octets", MV643XX_STAT(mib_counters.frames_128_to_255_octets) },
-       { "frames_256_to_511_octets", MV643XX_STAT(mib_counters.frames_256_to_511_octets) },
-       { "frames_512_to_1023_octets", MV643XX_STAT(mib_counters.frames_512_to_1023_octets) },
-       { "frames_1024_to_max_octets", MV643XX_STAT(mib_counters.frames_1024_to_max_octets) },
-       { "good_octets_sent", MV643XX_STAT(mib_counters.good_octets_sent) },
-       { "good_frames_sent", MV643XX_STAT(mib_counters.good_frames_sent) },
-       { "excessive_collision", MV643XX_STAT(mib_counters.excessive_collision) },
-       { "multicast_frames_sent", MV643XX_STAT(mib_counters.multicast_frames_sent) },
-       { "broadcast_frames_sent", MV643XX_STAT(mib_counters.broadcast_frames_sent) },
-       { "unrec_mac_control_received", MV643XX_STAT(mib_counters.unrec_mac_control_received) },
-       { "fc_sent", MV643XX_STAT(mib_counters.fc_sent) },
-       { "good_fc_received", MV643XX_STAT(mib_counters.good_fc_received) },
-       { "bad_fc_received", MV643XX_STAT(mib_counters.bad_fc_received) },
-       { "undersize_received", MV643XX_STAT(mib_counters.undersize_received) },
-       { "fragments_received", MV643XX_STAT(mib_counters.fragments_received) },
-       { "oversize_received", MV643XX_STAT(mib_counters.oversize_received) },
-       { "jabber_received", MV643XX_STAT(mib_counters.jabber_received) },
-       { "mac_receive_error", MV643XX_STAT(mib_counters.mac_receive_error) },
-       { "bad_crc_event", MV643XX_STAT(mib_counters.bad_crc_event) },
-       { "collision", MV643XX_STAT(mib_counters.collision) },
-       { "late_collision", MV643XX_STAT(mib_counters.late_collision) },
+#define SSTAT(m)                                               \
+       { #m, FIELD_SIZEOF(struct net_device_stats, m),         \
+         offsetof(struct net_device, stats.m), -1 }
+
+#define MIBSTAT(m)                                             \
+       { #m, FIELD_SIZEOF(struct mib_counters, m),             \
+         -1, offsetof(struct mv643xx_eth_private, mib_counters.m) }
+
+static const struct mv643xx_eth_stats mv643xx_eth_stats[] = {
+       SSTAT(rx_packets),
+       SSTAT(tx_packets),
+       SSTAT(rx_bytes),
+       SSTAT(tx_bytes),
+       SSTAT(rx_errors),
+       SSTAT(tx_errors),
+       SSTAT(rx_dropped),
+       SSTAT(tx_dropped),
+       MIBSTAT(good_octets_received),
+       MIBSTAT(bad_octets_received),
+       MIBSTAT(internal_mac_transmit_err),
+       MIBSTAT(good_frames_received),
+       MIBSTAT(bad_frames_received),
+       MIBSTAT(broadcast_frames_received),
+       MIBSTAT(multicast_frames_received),
+       MIBSTAT(frames_64_octets),
+       MIBSTAT(frames_65_to_127_octets),
+       MIBSTAT(frames_128_to_255_octets),
+       MIBSTAT(frames_256_to_511_octets),
+       MIBSTAT(frames_512_to_1023_octets),
+       MIBSTAT(frames_1024_to_max_octets),
+       MIBSTAT(good_octets_sent),
+       MIBSTAT(good_frames_sent),
+       MIBSTAT(excessive_collision),
+       MIBSTAT(multicast_frames_sent),
+       MIBSTAT(broadcast_frames_sent),
+       MIBSTAT(unrec_mac_control_received),
+       MIBSTAT(fc_sent),
+       MIBSTAT(good_fc_received),
+       MIBSTAT(bad_fc_received),
+       MIBSTAT(undersize_received),
+       MIBSTAT(fragments_received),
+       MIBSTAT(oversize_received),
+       MIBSTAT(jabber_received),
+       MIBSTAT(mac_receive_error),
+       MIBSTAT(bad_crc_event),
+       MIBSTAT(collision),
+       MIBSTAT(late_collision),
 };
 
-#define MV643XX_STATS_LEN      ARRAY_SIZE(mv643xx_gstrings_stats)
-
-static int mv643xx_get_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_private *mp = netdev_priv(dev);
+       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);
 
-       /* The PHY may support 1000baseT_Half, but the mv643xx does not */
+       /*
+        * The MAC does not support 1000baseT_Half.
+        */
        cmd->supported &= ~SUPPORTED_1000baseT_Half;
        cmd->advertising &= ~ADVERTISED_1000baseT_Half;
 
        return err;
 }
 
-static int mv643xx_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
+static int mv643xx_eth_get_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       u32 port_status;
+
+       port_status = rdl(mp, PORT_STATUS(mp->port_num));
+
+       cmd->supported = SUPPORTED_MII;
+       cmd->advertising = ADVERTISED_MII;
+       switch (port_status & PORT_SPEED_MASK) {
+       case PORT_SPEED_10:
+               cmd->speed = SPEED_10;
+               break;
+       case PORT_SPEED_100:
+               cmd->speed = SPEED_100;
+               break;
+       case PORT_SPEED_1000:
+               cmd->speed = SPEED_1000;
+               break;
+       default:
+               cmd->speed = -1;
+               break;
+       }
+       cmd->duplex = (port_status & FULL_DUPLEX) ? DUPLEX_FULL : DUPLEX_HALF;
+       cmd->port = PORT_MII;
+       cmd->phy_address = 0;
+       cmd->transceiver = XCVR_INTERNAL;
+       cmd->autoneg = AUTONEG_DISABLE;
+       cmd->maxtxpkt = 1;
+       cmd->maxrxpkt = 1;
+
+       return 0;
+}
+
+static int mv643xx_eth_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
        int err;
 
+       /*
+        * The MAC does not support 1000baseT_Half.
+        */
+       cmd->advertising &= ~ADVERTISED_1000baseT_Half;
+
        spin_lock_irq(&mp->lock);
        err = mii_ethtool_sset(&mp->mii, cmd);
        spin_unlock_irq(&mp->lock);
@@ -1249,578 +1220,468 @@ static int mv643xx_set_settings(struct net_device *dev, struct ethtool_cmd *cmd)
        return err;
 }
 
-static void mv643xx_get_drvinfo(struct net_device *netdev,
-                               struct ethtool_drvinfo *drvinfo)
+static int mv643xx_eth_set_settings_phyless(struct net_device *dev, struct ethtool_cmd *cmd)
+{
+       return -EINVAL;
+}
+
+static void mv643xx_eth_get_drvinfo(struct net_device *dev,
+                                   struct ethtool_drvinfo *drvinfo)
 {
-       strncpy(drvinfo->driver,  mv643xx_driver_name, 32);
-       strncpy(drvinfo->version, mv643xx_driver_version, 32);
+       strncpy(drvinfo->driver,  mv643xx_eth_driver_name, 32);
+       strncpy(drvinfo->version, mv643xx_eth_driver_version, 32);
        strncpy(drvinfo->fw_version, "N/A", 32);
-       strncpy(drvinfo->bus_info, "mv643xx", 32);
-       drvinfo->n_stats = MV643XX_STATS_LEN;
+       strncpy(drvinfo->bus_info, "platform", 32);
+       drvinfo->n_stats = ARRAY_SIZE(mv643xx_eth_stats);
 }
 
-static int mv643xx_eth_nway_restart(struct net_device *dev)
+static int mv643xx_eth_nway_reset(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
        return mii_nway_restart(&mp->mii);
 }
 
+static int mv643xx_eth_nway_reset_phyless(struct net_device *dev)
+{
+       return -EINVAL;
+}
+
 static u32 mv643xx_eth_get_link(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
        return mii_link_ok(&mp->mii);
 }
 
-static void mv643xx_get_strings(struct net_device *netdev, uint32_t stringset,
-                               uint8_t *data)
+static u32 mv643xx_eth_get_link_phyless(struct net_device *dev)
+{
+       return 1;
+}
+
+static void mv643xx_eth_get_strings(struct net_device *dev,
+                                   uint32_t stringset, uint8_t *data)
 {
        int i;
 
-       switch(stringset) {
-       case ETH_SS_STATS:
-               for (i=0; i < MV643XX_STATS_LEN; i++) {
+       if (stringset == ETH_SS_STATS) {
+               for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
                        memcpy(data + i * ETH_GSTRING_LEN,
-                                       mv643xx_gstrings_stats[i].stat_string,
-                                       ETH_GSTRING_LEN);
+                               mv643xx_eth_stats[i].stat_string,
+                               ETH_GSTRING_LEN);
                }
-               break;
        }
 }
 
-static void mv643xx_get_ethtool_stats(struct net_device *netdev,
-                               struct ethtool_stats *stats, uint64_t *data)
+static void mv643xx_eth_get_ethtool_stats(struct net_device *dev,
+                                         struct ethtool_stats *stats,
+                                         uint64_t *data)
 {
-       struct mv643xx_private *mp = netdev->priv;
+       struct mv643xx_eth_private *mp = dev->priv;
        int i;
 
-       eth_update_mib_counters(mp);
+       mib_counters_update(mp);
+
+       for (i = 0; i < ARRAY_SIZE(mv643xx_eth_stats); i++) {
+               const struct mv643xx_eth_stats *stat;
+               void *p;
 
-       for (i = 0; i < MV643XX_STATS_LEN; i++) {
-               char *p = (char *)mp+mv643xx_gstrings_stats[i].stat_offset;
-               data[i] = (mv643xx_gstrings_stats[i].sizeof_stat ==
-                       sizeof(uint64_t)) ? *(uint64_t *)p : *(uint32_t *)p;
+               stat = mv643xx_eth_stats + i;
+
+               if (stat->netdev_off >= 0)
+                       p = ((void *)mp->dev) + stat->netdev_off;
+               else
+                       p = ((void *)mp) + stat->mp_off;
+
+               data[i] = (stat->sizeof_stat == 8) ?
+                               *(uint64_t *)p : *(uint32_t *)p;
        }
 }
 
-static int mv643xx_get_sset_count(struct net_device *netdev, int sset)
+static int mv643xx_eth_get_sset_count(struct net_device *dev, int sset)
 {
-       switch (sset) {
-       case ETH_SS_STATS:
-               return MV643XX_STATS_LEN;
-       default:
-               return -EOPNOTSUPP;
-       }
+       if (sset == ETH_SS_STATS)
+               return ARRAY_SIZE(mv643xx_eth_stats);
+
+       return -EOPNOTSUPP;
 }
 
-static const struct ethtool_ops mv643xx_ethtool_ops = {
-       .get_settings           = mv643xx_get_settings,
-       .set_settings           = mv643xx_set_settings,
-       .get_drvinfo            = mv643xx_get_drvinfo,
-       .get_link               = mv643xx_eth_get_link,
+static const struct ethtool_ops mv643xx_eth_ethtool_ops = {
+       .get_settings           = mv643xx_eth_get_settings,
+       .set_settings           = mv643xx_eth_set_settings,
+       .get_drvinfo            = mv643xx_eth_get_drvinfo,
+       .nway_reset             = mv643xx_eth_nway_reset,
+       .get_link               = mv643xx_eth_get_link,
+       .set_sg                 = ethtool_op_set_sg,
+       .get_strings            = mv643xx_eth_get_strings,
+       .get_ethtool_stats      = mv643xx_eth_get_ethtool_stats,
+       .get_sset_count         = mv643xx_eth_get_sset_count,
+};
+
+static const struct ethtool_ops mv643xx_eth_ethtool_ops_phyless = {
+       .get_settings           = mv643xx_eth_get_settings_phyless,
+       .set_settings           = mv643xx_eth_set_settings_phyless,
+       .get_drvinfo            = mv643xx_eth_get_drvinfo,
+       .nway_reset             = mv643xx_eth_nway_reset_phyless,
+       .get_link               = mv643xx_eth_get_link_phyless,
        .set_sg                 = ethtool_op_set_sg,
-       .get_sset_count         = mv643xx_get_sset_count,
-       .get_ethtool_stats      = mv643xx_get_ethtool_stats,
-       .get_strings            = mv643xx_get_strings,
-       .nway_reset             = mv643xx_eth_nway_restart,
+       .get_strings            = mv643xx_eth_get_strings,
+       .get_ethtool_stats      = mv643xx_eth_get_ethtool_stats,
+       .get_sset_count         = mv643xx_eth_get_sset_count,
 };
 
 
 /* address handling *********************************************************/
-/*
- * eth_port_uc_addr_get - Read the MAC address from the port's hw registers
- */
-static void eth_port_uc_addr_get(struct mv643xx_private *mp,
-                                unsigned char *p_addr)
+static void uc_addr_get(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       unsigned int port_num = mp->port_num;
        unsigned int mac_h;
        unsigned int mac_l;
 
-       mac_h = rdl(mp, MAC_ADDR_HIGH(port_num));
-       mac_l = rdl(mp, MAC_ADDR_LOW(port_num));
+       mac_h = rdl(mp, MAC_ADDR_HIGH(mp->port_num));
+       mac_l = rdl(mp, MAC_ADDR_LOW(mp->port_num));
 
-       p_addr[0] = (mac_h >> 24) & 0xff;
-       p_addr[1] = (mac_h >> 16) & 0xff;
-       p_addr[2] = (mac_h >> 8) & 0xff;
-       p_addr[3] = mac_h & 0xff;
-       p_addr[4] = (mac_l >> 8) & 0xff;
-       p_addr[5] = mac_l & 0xff;
+       addr[0] = (mac_h >> 24) & 0xff;
+       addr[1] = (mac_h >> 16) & 0xff;
+       addr[2] = (mac_h >> 8) & 0xff;
+       addr[3] = mac_h & 0xff;
+       addr[4] = (mac_l >> 8) & 0xff;
+       addr[5] = mac_l & 0xff;
 }
 
-/*
- * eth_port_init_mac_tables - Clear all entrance in the UC, SMC and OMC tables
- *
- * DESCRIPTION:
- *     Go through all the DA filter tables (Unicast, Special Multicast &
- *     Other Multicast) and set each entry to 0.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     Multicast and Unicast packets are rejected.
- *
- * RETURN:
- *     None.
- */
-static void eth_port_init_mac_tables(struct mv643xx_private *mp)
+static void init_mac_tables(struct mv643xx_eth_private *mp)
 {
-       unsigned int port_num = mp->port_num;
-       int table_index;
-
-       /* Clear DA filter unicast table (Ex_dFUT) */
-       for (table_index = 0; table_index <= 0xC; table_index += 4)
-               wrl(mp, UNICAST_TABLE(port_num) + table_index, 0);
+       int i;
 
-       for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-               /* Clear DA filter special multicast table (Ex_dFSMT) */
-               wrl(mp, SPECIAL_MCAST_TABLE(port_num) + table_index, 0);
-               /* Clear DA filter other multicast table (Ex_dFOMT) */
-               wrl(mp, OTHER_MCAST_TABLE(port_num) + table_index, 0);
+       for (i = 0; i < 0x100; i += 4) {
+               wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, 0);
+               wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, 0);
        }
+
+       for (i = 0; i < 0x10; i += 4)
+               wrl(mp, UNICAST_TABLE(mp->port_num) + i, 0);
 }
 
-/*
- * The entries in each table are indexed by a hash of a packet's MAC
- * address.  One bit in each entry determines whether the packet is
- * accepted.  There are 4 entries (each 8 bits wide) in each register
- * of the table.  The bits in each entry are defined as follows:
- *     0       Accept=1, Drop=0
- *     3-1     Queue                   (ETH_Q0=0)
- *     7-4     Reserved = 0;
- */
-static void eth_port_set_filter_table_entry(struct mv643xx_private *mp,
-                                           int table, unsigned char entry)
+static void set_filter_table_entry(struct mv643xx_eth_private *mp,
+                                  int table, unsigned char entry)
 {
        unsigned int table_reg;
-       unsigned int tbl_offset;
-       unsigned int reg_offset;
-
-       tbl_offset = (entry / 4) * 4;   /* Register offset of DA table entry */
-       reg_offset = entry % 4;         /* Entry offset within the register */
 
        /* Set "accepts frame bit" at specified table entry */
-       table_reg = rdl(mp, table + tbl_offset);
-       table_reg |= 0x01 << (8 * reg_offset);
-       wrl(mp, table + tbl_offset, table_reg);
+       table_reg = rdl(mp, table + (entry & 0xfc));
+       table_reg |= 0x01 << (8 * (entry & 3));
+       wrl(mp, table + (entry & 0xfc), table_reg);
 }
 
-/*
- * eth_port_uc_addr_set - Write a MAC address into the port's hw registers
- */
-static void eth_port_uc_addr_set(struct mv643xx_private *mp,
-                                unsigned char *p_addr)
+static void uc_addr_set(struct mv643xx_eth_private *mp, unsigned char *addr)
 {
-       unsigned int port_num = mp->port_num;
        unsigned int mac_h;
        unsigned int mac_l;
        int table;
 
-       mac_l = (p_addr[4] << 8) | (p_addr[5]);
-       mac_h = (p_addr[0] << 24) | (p_addr[1] << 16) | (p_addr[2] << 8) |
-                                                       (p_addr[3] << 0);
+       mac_l = (addr[4] << 8) | addr[5];
+       mac_h = (addr[0] << 24) | (addr[1] << 16) | (addr[2] << 8) | addr[3];
 
-       wrl(mp, MAC_ADDR_LOW(port_num), mac_l);
-       wrl(mp, MAC_ADDR_HIGH(port_num), mac_h);
+       wrl(mp, MAC_ADDR_LOW(mp->port_num), mac_l);
+       wrl(mp, MAC_ADDR_HIGH(mp->port_num), mac_h);
 
-       /* Accept frames with this address */
-       table = UNICAST_TABLE(port_num);
-       eth_port_set_filter_table_entry(mp, table, p_addr[5] & 0x0f);
+       table = UNICAST_TABLE(mp->port_num);
+       set_filter_table_entry(mp, table, addr[5] & 0x0f);
 }
 
-/*
- * mv643xx_eth_update_mac_address
- *
- * Update the MAC address of the port in the address table
- *
- * Input :     pointer to ethernet interface network device structure
- * Output :    N/A
- */
-static void mv643xx_eth_update_mac_address(struct net_device *dev)
+static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       eth_port_init_mac_tables(mp);
-       eth_port_uc_addr_set(mp, dev->dev_addr);
-}
+       /* +2 is for the offset of the HW addr type */
+       memcpy(dev->dev_addr, addr + 2, 6);
 
-/*
- * mv643xx_eth_set_mac_address
- *
- * Change the interface's mac address.
- * No special hardware thing should be done because interface is always
- * put in promiscuous mode.
- *
- * Input :     pointer to ethernet interface network device structure and
- *             a pointer to the designated entry to be added to the cache.
- * Output :    zero upon success, negative upon failure
- */
-static int mv643xx_eth_set_mac_address(struct net_device *dev, void *addr)
-{
-       int i;
+       init_mac_tables(mp);
+       uc_addr_set(mp, dev->dev_addr);
 
-       for (i = 0; i < 6; i++)
-               /* +2 is for the offset of the HW addr type */
-               dev->dev_addr[i] = ((unsigned char *)addr)[i + 2];
-       mv643xx_eth_update_mac_address(dev);
        return 0;
 }
 
-/*
- * eth_port_mc_addr - Multicast address settings.
- *
- * The MV device supports multicast using two tables:
- * 1) Special Multicast Table for MAC addresses of the form
- *    0x01-00-5E-00-00-XX (where XX is between 0x00 and 0x_FF).
- *    The MAC DA[7:0] bits are used as a pointer to the Special Multicast
- *    Table entries in the DA-Filter table.
- * 2) Other Multicast Table for multicast of another type. A CRC-8bit
- *    is used as an index to the Other Multicast Table entries in the
- *    DA-Filter table.  This function calculates the CRC-8bit value.
- * In either case, eth_port_set_filter_table_entry() is then called
- * to set to set the actual table entry.
- */
-static void eth_port_mc_addr(struct mv643xx_private *mp, unsigned char *p_addr)
+static int addr_crc(unsigned char *addr)
 {
-       unsigned int port_num = mp->port_num;
-       unsigned int mac_h;
-       unsigned int mac_l;
-       unsigned char crc_result = 0;
-       int table;
-       int mac_array[48];
-       int crc[8];
+       int crc = 0;
        int i;
 
-       if ((p_addr[0] == 0x01) && (p_addr[1] == 0x00) &&
-           (p_addr[2] == 0x5E) && (p_addr[3] == 0x00) && (p_addr[4] == 0x00)) {
-               table = SPECIAL_MCAST_TABLE(port_num);
-               eth_port_set_filter_table_entry(mp, table, p_addr[5]);
-               return;
-       }
-
-       /* Calculate CRC-8 out of the given address */
-       mac_h = (p_addr[0] << 8) | (p_addr[1]);
-       mac_l = (p_addr[2] << 24) | (p_addr[3] << 16) |
-                       (p_addr[4] << 8) | (p_addr[5] << 0);
-
-       for (i = 0; i < 32; i++)
-               mac_array[i] = (mac_l >> i) & 0x1;
-       for (i = 32; i < 48; i++)
-               mac_array[i] = (mac_h >> (i - 32)) & 0x1;
-
-       crc[0] = mac_array[45] ^ mac_array[43] ^ mac_array[40] ^ mac_array[39] ^
-                mac_array[35] ^ mac_array[34] ^ mac_array[31] ^ mac_array[30] ^
-                mac_array[28] ^ mac_array[23] ^ mac_array[21] ^ mac_array[19] ^
-                mac_array[18] ^ mac_array[16] ^ mac_array[14] ^ mac_array[12] ^
-                mac_array[8]  ^ mac_array[7]  ^ mac_array[6]  ^ mac_array[0];
-
-       crc[1] = mac_array[46] ^ mac_array[45] ^ mac_array[44] ^ mac_array[43] ^
-                mac_array[41] ^ mac_array[39] ^ mac_array[36] ^ mac_array[34] ^
-                mac_array[32] ^ mac_array[30] ^ mac_array[29] ^ mac_array[28] ^
-                mac_array[24] ^ mac_array[23] ^ mac_array[22] ^ mac_array[21] ^
-                mac_array[20] ^ mac_array[18] ^ mac_array[17] ^ mac_array[16] ^
-                mac_array[15] ^ mac_array[14] ^ mac_array[13] ^ mac_array[12] ^
-                mac_array[9]  ^ mac_array[6]  ^ mac_array[1]  ^ mac_array[0];
-
-       crc[2] = mac_array[47] ^ mac_array[46] ^ mac_array[44] ^ mac_array[43] ^
-                mac_array[42] ^ mac_array[39] ^ mac_array[37] ^ mac_array[34] ^
-                mac_array[33] ^ mac_array[29] ^ mac_array[28] ^ mac_array[25] ^
-                mac_array[24] ^ mac_array[22] ^ mac_array[17] ^ mac_array[15] ^
-                mac_array[13] ^ mac_array[12] ^ mac_array[10] ^ mac_array[8]  ^
-                mac_array[6]  ^ mac_array[2]  ^ mac_array[1]  ^ mac_array[0];
-
-       crc[3] = mac_array[47] ^ mac_array[45] ^ mac_array[44] ^ mac_array[43] ^
-                mac_array[40] ^ mac_array[38] ^ mac_array[35] ^ mac_array[34] ^
-                mac_array[30] ^ mac_array[29] ^ mac_array[26] ^ mac_array[25] ^
-                mac_array[23] ^ mac_array[18] ^ mac_array[16] ^ mac_array[14] ^
-                mac_array[13] ^ mac_array[11] ^ mac_array[9]  ^ mac_array[7]  ^
-                mac_array[3]  ^ mac_array[2]  ^ mac_array[1];
-
-       crc[4] = mac_array[46] ^ mac_array[45] ^ mac_array[44] ^ mac_array[41] ^
-                mac_array[39] ^ mac_array[36] ^ mac_array[35] ^ mac_array[31] ^
-                mac_array[30] ^ mac_array[27] ^ mac_array[26] ^ mac_array[24] ^
-                mac_array[19] ^ mac_array[17] ^ mac_array[15] ^ mac_array[14] ^
-                mac_array[12] ^ mac_array[10] ^ mac_array[8]  ^ mac_array[4]  ^
-                mac_array[3]  ^ mac_array[2];
-
-       crc[5] = mac_array[47] ^ mac_array[46] ^ mac_array[45] ^ mac_array[42] ^
-                mac_array[40] ^ mac_array[37] ^ mac_array[36] ^ mac_array[32] ^
-                mac_array[31] ^ mac_array[28] ^ mac_array[27] ^ mac_array[25] ^
-                mac_array[20] ^ mac_array[18] ^ mac_array[16] ^ mac_array[15] ^
-                mac_array[13] ^ mac_array[11] ^ mac_array[9]  ^ mac_array[5]  ^
-                mac_array[4]  ^ mac_array[3];
-
-       crc[6] = mac_array[47] ^ mac_array[46] ^ mac_array[43] ^ mac_array[41] ^
-                mac_array[38] ^ mac_array[37] ^ mac_array[33] ^ mac_array[32] ^
-                mac_array[29] ^ mac_array[28] ^ mac_array[26] ^ mac_array[21] ^
-                mac_array[19] ^ mac_array[17] ^ mac_array[16] ^ mac_array[14] ^
-                mac_array[12] ^ mac_array[10] ^ mac_array[6]  ^ mac_array[5]  ^
-                mac_array[4];
-
-       crc[7] = mac_array[47] ^ mac_array[44] ^ mac_array[42] ^ mac_array[39] ^
-                mac_array[38] ^ mac_array[34] ^ mac_array[33] ^ mac_array[30] ^
-                mac_array[29] ^ mac_array[27] ^ mac_array[22] ^ mac_array[20] ^
-                mac_array[18] ^ mac_array[17] ^ mac_array[15] ^ mac_array[13] ^
-                mac_array[11] ^ mac_array[7]  ^ mac_array[6]  ^ mac_array[5];
+       for (i = 0; i < 6; i++) {
+               int j;
 
-       for (i = 0; i < 8; i++)
-               crc_result = crc_result | (crc[i] << i);
+               crc = (crc ^ addr[i]) << 8;
+               for (j = 7; j >= 0; j--) {
+                       if (crc & (0x100 << j))
+                               crc ^= 0x107 << j;
+               }
+       }
 
-       table = OTHER_MCAST_TABLE(port_num);
-       eth_port_set_filter_table_entry(mp, table, crc_result);
+       return crc;
 }
 
-/*
- * Set the entire multicast list based on dev->mc_list.
- */
-static void eth_port_set_multicast_list(struct net_device *dev)
+static void mv643xx_eth_set_rx_mode(struct net_device *dev)
 {
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       u32 port_config;
+       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);
 
-       struct dev_mc_list      *mc_list;
-       int                     i;
-       int                     table_index;
-       struct mv643xx_private  *mp = netdev_priv(dev);
-       unsigned int            eth_port_num = mp->port_num;
+       if (dev->flags & (IFF_PROMISC | IFF_ALLMULTI)) {
+               int port_num = mp->port_num;
+               u32 accept = 0x01010101;
 
-       /* If the device is in promiscuous mode or in all multicast mode,
-        * we will fully populate both multicast tables with accept.
-        * This is guaranteed to yield a match on all multicast addresses...
-        */
-       if ((dev->flags & IFF_PROMISC) || (dev->flags & IFF_ALLMULTI)) {
-               for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-                       /* Set all entries in DA filter special multicast
-                        * table (Ex_dFSMT)
-                        * Set for ETH_Q0 for now
-                        * Bits
-                        * 0      Accept=1, Drop=0
-                        * 3-1  Queue    ETH_Q0=0
-                        * 7-4  Reserved = 0;
-                        */
-                       wrl(mp, SPECIAL_MCAST_TABLE(eth_port_num) + table_index, 0x01010101);
-
-                       /* Set all entries in DA filter other multicast
-                        * table (Ex_dFOMT)
-                        * Set for ETH_Q0 for now
-                        * Bits
-                        * 0      Accept=1, Drop=0
-                        * 3-1  Queue    ETH_Q0=0
-                        * 7-4  Reserved = 0;
-                        */
-                       wrl(mp, OTHER_MCAST_TABLE(eth_port_num) + table_index, 0x01010101);
+               for (i = 0; i < 0x100; i += 4) {
+                       wrl(mp, SPECIAL_MCAST_TABLE(port_num) + i, accept);
+                       wrl(mp, OTHER_MCAST_TABLE(port_num) + i, accept);
                }
                return;
        }
 
-       /* We will clear out multicast tables every time we get the list.
-        * Then add the entire new list...
-        */
-       for (table_index = 0; table_index <= 0xFC; table_index += 4) {
-               /* Clear DA filter special multicast table (Ex_dFSMT) */
-               wrl(mp, SPECIAL_MCAST_TABLE(eth_port_num) + table_index, 0);
-
-               /* Clear DA filter other multicast table (Ex_dFOMT) */
-               wrl(mp, OTHER_MCAST_TABLE(eth_port_num) + table_index, 0);
+       for (i = 0; i < 0x100; i += 4) {
+               wrl(mp, SPECIAL_MCAST_TABLE(mp->port_num) + i, 0);
+               wrl(mp, OTHER_MCAST_TABLE(mp->port_num) + i, 0);
        }
 
-       /* Get pointer to net_device multicast list and add each one... */
-       for (i = 0, mc_list = dev->mc_list;
-                       (i < 256) && (mc_list != NULL) && (i < dev->mc_count);
-                       i++, mc_list = mc_list->next)
-               if (mc_list->dmi_addrlen == 6)
-                       eth_port_mc_addr(mp, mc_list->dmi_addr);
-}
+       for (addr = dev->mc_list; addr != NULL; addr = addr->next) {
+               u8 *a = addr->da_addr;
+               int table;
 
-/*
- * mv643xx_eth_set_rx_mode
- *
- * Change from promiscuos to regular rx mode
- *
- * Input :     pointer to ethernet interface network device structure
- * Output :    N/A
- */
-static void mv643xx_eth_set_rx_mode(struct net_device *dev)
-{
-       struct mv643xx_private *mp = netdev_priv(dev);
-       u32 config_reg;
+               if (addr->da_addrlen != 6)
+                       continue;
 
-       config_reg = rdl(mp, PORT_CONFIG(mp->port_num));
-       if (dev->flags & IFF_PROMISC)
-               config_reg |= UNICAST_PROMISCUOUS_MODE;
-       else
-               config_reg &= ~UNICAST_PROMISCUOUS_MODE;
-       wrl(mp, PORT_CONFIG(mp->port_num), config_reg);
+               if (memcmp(a, "\x01\x00\x5e\x00\x00", 5) == 0) {
+                       table = SPECIAL_MCAST_TABLE(mp->port_num);
+                       set_filter_table_entry(mp, table, a[5]);
+               } else {
+                       int crc = addr_crc(a);
 
-       eth_port_set_multicast_list(dev);
+                       table = OTHER_MCAST_TABLE(mp->port_num);
+                       set_filter_table_entry(mp, table, crc);
+               }
+       }
 }
 
 
 /* rx/tx queue initialisation ***********************************************/
-/*
- * ether_init_rx_desc_ring - Curve a Rx chain desc list and buffer in memory.
- *
- * DESCRIPTION:
- *     This function prepares a Rx chained list of descriptors and packet
- *     buffers in a form of a ring. The routine must be called after port
- *     initialization routine and before port start routine.
- *     The Ethernet SDMA engine uses CPU bus addresses to access the various
- *     devices in the system (i.e. DRAM). This function uses the ethernet
- *     struct 'virtual to physical' routine (set by the user) to set the ring
- *     with physical addresses.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port Control srtuct.
- *
- * OUTPUT:
- *     The routine updates the Ethernet port control struct with information
- *     regarding the Rx descriptors and buffers.
- *
- * RETURN:
- *     None.
- */
-static void ether_init_rx_desc_ring(struct mv643xx_private *mp)
+static int rxq_init(struct mv643xx_eth_private *mp, int index)
 {
-       volatile struct eth_rx_desc *p_rx_desc;
-       int rx_desc_num = mp->rx_ring_size;
+       struct rx_queue *rxq = mp->rxq + index;
+       struct rx_desc *rx_desc;
+       int size;
        int i;
 
-       /* initialize the next_desc_ptr links in the Rx descriptors ring */
-       p_rx_desc = (struct eth_rx_desc *)mp->p_rx_desc_area;
-       for (i = 0; i < rx_desc_num; i++) {
-               p_rx_desc[i].next_desc_ptr = mp->rx_desc_dma +
-                       ((i + 1) % rx_desc_num) * sizeof(struct eth_rx_desc);
+       rxq->index = index;
+
+       rxq->rx_ring_size = mp->default_rx_ring_size;
+
+       rxq->rx_desc_count = 0;
+       rxq->rx_curr_desc = 0;
+       rxq->rx_used_desc = 0;
+
+       size = rxq->rx_ring_size * sizeof(struct rx_desc);
+
+       if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size) {
+               rxq->rx_desc_area = ioremap(mp->rx_desc_sram_addr,
+                                               mp->rx_desc_sram_size);
+               rxq->rx_desc_dma = mp->rx_desc_sram_addr;
+       } else {
+               rxq->rx_desc_area = dma_alloc_coherent(NULL, size,
+                                                       &rxq->rx_desc_dma,
+                                                       GFP_KERNEL);
+       }
+
+       if (rxq->rx_desc_area == NULL) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "can't allocate rx ring (%d bytes)\n", size);
+               goto out;
+       }
+       memset(rxq->rx_desc_area, 0, size);
+
+       rxq->rx_desc_area_size = size;
+       rxq->rx_skb = kmalloc(rxq->rx_ring_size * sizeof(*rxq->rx_skb),
+                                                               GFP_KERNEL);
+       if (rxq->rx_skb == NULL) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "can't allocate rx skb ring\n");
+               goto out_free;
+       }
+
+       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;
+               rx_desc[i].next_desc_ptr = rxq->rx_desc_dma +
+                                       nexti * sizeof(struct rx_desc);
        }
 
-       /* Save Rx desc pointer to driver struct. */
-       mp->rx_curr_desc_q = 0;
-       mp->rx_used_desc_q = 0;
+       init_timer(&rxq->rx_oom);
+       rxq->rx_oom.data = (unsigned long)rxq;
+       rxq->rx_oom.function = rxq_refill_timer_wrapper;
 
-       mp->rx_desc_area_size = rx_desc_num * sizeof(struct eth_rx_desc);
+       return 0;
+
+
+out_free:
+       if (index == mp->rxq_primary && size <= mp->rx_desc_sram_size)
+               iounmap(rxq->rx_desc_area);
+       else
+               dma_free_coherent(NULL, size,
+                                 rxq->rx_desc_area,
+                                 rxq->rx_desc_dma);
+
+out:
+       return -ENOMEM;
 }
 
-static void mv643xx_eth_free_rx_rings(struct net_device *dev)
+static void rxq_deinit(struct rx_queue *rxq)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       int curr;
+       struct mv643xx_eth_private *mp = rxq_to_mp(rxq);
+       int i;
 
-       /* Stop RX Queues */
-       mv643xx_eth_port_disable_rx(mp);
+       rxq_disable(rxq);
 
-       /* Free preallocated skb's on RX rings */
-       for (curr = 0; mp->rx_desc_count && curr < mp->rx_ring_size; curr++) {
-               if (mp->rx_skb[curr]) {
-                       dev_kfree_skb(mp->rx_skb[curr]);
-                       mp->rx_desc_count--;
+       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]);
+                       rxq->rx_desc_count--;
                }
        }
 
-       if (mp->rx_desc_count)
-               printk(KERN_ERR
-                       "%s: Error in freeing Rx Ring. %d skb's still"
-                       " stuck in RX Ring - ignoring them\n", dev->name,
-                       mp->rx_desc_count);
-       /* Free RX ring */
-       if (mp->rx_sram_size)
-               iounmap(mp->p_rx_desc_area);
+       if (rxq->rx_desc_count) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "error freeing rx ring -- %d skbs stuck\n",
+                          rxq->rx_desc_count);
+       }
+
+       if (rxq->index == mp->rxq_primary &&
+           rxq->rx_desc_area_size <= mp->rx_desc_sram_size)
+               iounmap(rxq->rx_desc_area);
        else
-               dma_free_coherent(NULL, mp->rx_desc_area_size,
-                               mp->p_rx_desc_area, mp->rx_desc_dma);
+               dma_free_coherent(NULL, rxq->rx_desc_area_size,
+                                 rxq->rx_desc_area, rxq->rx_desc_dma);
+
+       kfree(rxq->rx_skb);
 }
 
-/*
- * ether_init_tx_desc_ring - Curve a Tx chain desc list and buffer in memory.
- *
- * DESCRIPTION:
- *     This function prepares a Tx chained list of descriptors and packet
- *     buffers in a form of a ring. The routine must be called after port
- *     initialization routine and before port start routine.
- *     The Ethernet SDMA engine uses CPU bus addresses to access the various
- *     devices in the system (i.e. DRAM). This function uses the ethernet
- *     struct 'virtual to physical' routine (set by the user) to set the ring
- *     with physical addresses.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port Control srtuct.
- *
- * OUTPUT:
- *     The routine updates the Ethernet port control struct with information
- *     regarding the Tx descriptors and buffers.
- *
- * RETURN:
- *     None.
- */
-static void ether_init_tx_desc_ring(struct mv643xx_private *mp)
+static int txq_init(struct mv643xx_eth_private *mp, int index)
 {
-       int tx_desc_num = mp->tx_ring_size;
-       struct eth_tx_desc *p_tx_desc;
+       struct tx_queue *txq = mp->txq + index;
+       struct tx_desc *tx_desc;
+       int size;
        int i;
 
-       /* Initialize the next_desc_ptr links in the Tx descriptors ring */
-       p_tx_desc = (struct eth_tx_desc *)mp->p_tx_desc_area;
-       for (i = 0; i < tx_desc_num; i++) {
-               p_tx_desc[i].next_desc_ptr = mp->tx_desc_dma +
-                       ((i + 1) % tx_desc_num) * sizeof(struct eth_tx_desc);
+       txq->index = index;
+
+       txq->tx_ring_size = mp->default_tx_ring_size;
+
+       txq->tx_desc_count = 0;
+       txq->tx_curr_desc = 0;
+       txq->tx_used_desc = 0;
+
+       size = txq->tx_ring_size * sizeof(struct tx_desc);
+
+       if (index == mp->txq_primary && size <= mp->tx_desc_sram_size) {
+               txq->tx_desc_area = ioremap(mp->tx_desc_sram_addr,
+                                               mp->tx_desc_sram_size);
+               txq->tx_desc_dma = mp->tx_desc_sram_addr;
+       } else {
+               txq->tx_desc_area = dma_alloc_coherent(NULL, size,
+                                                       &txq->tx_desc_dma,
+                                                       GFP_KERNEL);
+       }
+
+       if (txq->tx_desc_area == NULL) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "can't allocate tx ring (%d bytes)\n", size);
+               goto out;
+       }
+       memset(txq->tx_desc_area, 0, size);
+
+       txq->tx_desc_area_size = size;
+       txq->tx_skb = kmalloc(txq->tx_ring_size * sizeof(*txq->tx_skb),
+                                                               GFP_KERNEL);
+       if (txq->tx_skb == NULL) {
+               dev_printk(KERN_ERR, &mp->dev->dev,
+                          "can't allocate tx skb ring\n");
+               goto out_free;
+       }
+
+       tx_desc = (struct tx_desc *)txq->tx_desc_area;
+       for (i = 0; i < txq->tx_ring_size; i++) {
+               struct tx_desc *txd = tx_desc + i;
+               int nexti = (i + 1) % txq->tx_ring_size;
+
+               txd->cmd_sts = 0;
+               txd->next_desc_ptr = txq->tx_desc_dma +
+                                       nexti * sizeof(struct tx_desc);
        }
 
-       mp->tx_curr_desc_q = 0;
-       mp->tx_used_desc_q = 0;
+       return 0;
+
+
+out_free:
+       if (index == mp->txq_primary && size <= mp->tx_desc_sram_size)
+               iounmap(txq->tx_desc_area);
+       else
+               dma_free_coherent(NULL, size,
+                                 txq->tx_desc_area,
+                                 txq->tx_desc_dma);
 
-       mp->tx_desc_area_size = tx_desc_num * sizeof(struct eth_tx_desc);
+out:
+       return -ENOMEM;
 }
 
-/**
- * mv643xx_eth_free_tx_descs - Free the tx desc data for completed descriptors
- *
- * If force is non-zero, frees uncompleted descriptors as well
- */
-static int mv643xx_eth_free_tx_descs(struct net_device *dev, int force)
+static void txq_reclaim(struct tx_queue *txq, int force)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       struct eth_tx_desc *desc;
-       u32 cmd_sts;
-       struct sk_buff *skb;
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
        unsigned long flags;
-       int tx_index;
-       dma_addr_t addr;
-       int count;
-       int released = 0;
 
-       while (mp->tx_desc_count > 0) {
-               spin_lock_irqsave(&mp->lock, flags);
-
-               /* tx_desc_count might have changed before acquiring the lock */
-               if (mp->tx_desc_count <= 0) {
-                       spin_unlock_irqrestore(&mp->lock, flags);
-                       return released;
-               }
-
-               tx_index = mp->tx_used_desc_q;
-               desc = &mp->p_tx_desc_area[tx_index];
+       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;
 
-               if (!force && (cmd_sts & ETH_BUFFER_OWNED_BY_DMA)) {
-                       spin_unlock_irqrestore(&mp->lock, flags);
-                       return released;
+               if (cmd_sts & BUFFER_OWNED_BY_DMA) {
+                       if (!force)
+                               break;
+                       desc->cmd_sts = cmd_sts & ~BUFFER_OWNED_BY_DMA;
                }
 
-               mp->tx_used_desc_q = (tx_index + 1) % mp->tx_ring_size;
-               mp->tx_desc_count--;
+               txq->tx_used_desc = (tx_index + 1) % txq->tx_ring_size;
+               txq->tx_desc_count--;
 
                addr = desc->buf_ptr;
                count = desc->byte_cnt;
-               skb = mp->tx_skb[tx_index];
-               if (skb)
-                       mp->tx_skb[tx_index] = NULL;
+               skb = txq->tx_skb[tx_index];
+               txq->tx_skb[tx_index] = NULL;
 
-               if (cmd_sts & ETH_ERROR_SUMMARY) {
-                       printk("%s: Error in TX\n", dev->name);
-                       dev->stats.tx_errors++;
+               if (cmd_sts & ERROR_SUMMARY) {
+                       dev_printk(KERN_INFO, &mp->dev->dev, "tx error\n");
+                       mp->dev->stats.tx_errors++;
                }
 
+               /*
+                * Drop mp->lock while we free the skb.
+                */
                spin_unlock_irqrestore(&mp->lock, flags);
 
-               if (cmd_sts & ETH_TX_FIRST_DESC)
+               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);
@@ -1828,637 +1689,465 @@ static int mv643xx_eth_free_tx_descs(struct net_device *dev, int force)
                if (skb)
                        dev_kfree_skb_irq(skb);
 
-               released = 1;
+               spin_lock_irqsave(&mp->lock, flags);
        }
-
-       return released;
+       spin_unlock_irqrestore(&mp->lock, flags);
 }
 
-static void mv643xx_eth_free_completed_tx_descs(struct net_device *dev)
+static void txq_deinit(struct tx_queue *txq)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = txq_to_mp(txq);
 
-       if (mv643xx_eth_free_tx_descs(dev, 0) &&
-           mp->tx_ring_size - mp->tx_desc_count >= MAX_DESCS_PER_SKB)
-               netif_wake_queue(dev);
-}
+       txq_disable(txq);
+       txq_reclaim(txq, 1);
 
-static void mv643xx_eth_free_all_tx_descs(struct net_device *dev)
-{
-       mv643xx_eth_free_tx_descs(dev, 1);
+       BUG_ON(txq->tx_used_desc != txq->tx_curr_desc);
+
+       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);
+
+       kfree(txq->tx_skb);
 }
 
-static void mv643xx_eth_free_tx_rings(struct net_device *dev)
-{
-       struct mv643xx_private *mp = netdev_priv(dev);
 
-       /* Stop Tx Queues */
-       mv643xx_eth_port_disable_tx(mp);
+/* netdev ops and related ***************************************************/
+static void handle_link_event(struct mv643xx_eth_private *mp)
+{
+       struct net_device *dev = mp->dev;
+       u32 port_status;
+       int speed;
+       int duplex;
+       int fc;
 
-       /* Free outstanding skb's on TX ring */
-       mv643xx_eth_free_all_tx_descs(dev);
+       port_status = rdl(mp, PORT_STATUS(mp->port_num));
+       if (!(port_status & LINK_UP)) {
+               if (netif_carrier_ok(dev)) {
+                       int i;
 
-       BUG_ON(mp->tx_used_desc_q != mp->tx_curr_desc_q);
+                       printk(KERN_INFO "%s: link down\n", dev->name);
 
-       /* Free TX ring */
-       if (mp->tx_sram_size)
-               iounmap(mp->p_tx_desc_area);
-       else
-               dma_free_coherent(NULL, mp->tx_desc_area_size,
-                               mp->p_tx_desc_area, mp->tx_desc_dma);
-}
+                       netif_carrier_off(dev);
+                       netif_stop_queue(dev);
 
+                       for (i = 0; i < 8; i++) {
+                               struct tx_queue *txq = mp->txq + i;
 
-/* netdev ops and related ***************************************************/
-static void eth_port_reset(struct mv643xx_private *mp);
-
-/* Set the mv643xx port configuration register for the speed/duplex mode. */
-static void mv643xx_eth_update_pscr(struct net_device *dev,
-                                   struct ethtool_cmd *ecmd)
-{
-       struct mv643xx_private *mp = netdev_priv(dev);
-       int port_num = mp->port_num;
-       u32 o_pscr, n_pscr;
-       unsigned int queues;
-
-       o_pscr = rdl(mp, PORT_SERIAL_CONTROL(port_num));
-       n_pscr = o_pscr;
-
-       /* clear speed, duplex and rx buffer size fields */
-       n_pscr &= ~(SET_MII_SPEED_TO_100  |
-                  SET_GMII_SPEED_TO_1000 |
-                  SET_FULL_DUPLEX_MODE   |
-                  MAX_RX_PACKET_MASK);
-
-       if (ecmd->duplex == DUPLEX_FULL)
-               n_pscr |= SET_FULL_DUPLEX_MODE;
-
-       if (ecmd->speed == SPEED_1000)
-               n_pscr |= SET_GMII_SPEED_TO_1000 |
-                         MAX_RX_PACKET_9700BYTE;
-       else {
-               if (ecmd->speed == SPEED_100)
-                       n_pscr |= SET_MII_SPEED_TO_100;
-               n_pscr |= MAX_RX_PACKET_1522BYTE;
-       }
-
-       if (n_pscr != o_pscr) {
-               if ((o_pscr & SERIAL_PORT_ENABLE) == 0)
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
-               else {
-                       queues = mv643xx_eth_port_disable_tx(mp);
-
-                       o_pscr &= ~SERIAL_PORT_ENABLE;
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), o_pscr);
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
-                       wrl(mp, PORT_SERIAL_CONTROL(port_num), n_pscr);
-                       if (queues)
-                               mv643xx_eth_port_enable_tx(mp, queues);
+                               if (mp->txq_mask & (1 << i)) {
+                                       txq_reclaim(txq, 1);
+                                       txq_reset_hw_ptr(txq);
+                               }
+                       }
                }
+               return;
        }
-}
 
-/*
- * mv643xx_eth_int_handler
- *
- * Main interrupt handler for the gigbit ethernet ports
- *
- * Input :     irq     - irq number (not used)
- *             dev_id  - a pointer to the required interface's data structure
- *             regs    - not used
- * Output :    N/A
- */
+       switch (port_status & PORT_SPEED_MASK) {
+       case PORT_SPEED_10:
+               speed = 10;
+               break;
+       case PORT_SPEED_100:
+               speed = 100;
+               break;
+       case PORT_SPEED_1000:
+               speed = 1000;
+               break;
+       default:
+               speed = -1;
+               break;
+       }
+       duplex = (port_status & FULL_DUPLEX) ? 1 : 0;
+       fc = (port_status & FLOW_CONTROL_ENABLED) ? 1 : 0;
+
+       printk(KERN_INFO "%s: link up, %d Mb/s, %s duplex, "
+                        "flow control %sabled\n", dev->name,
+                        speed, duplex ? "full" : "half",
+                        fc ? "en" : "dis");
+
+       if (!netif_carrier_ok(dev)) {
+               netif_carrier_on(dev);
+               netif_wake_queue(dev);
+       }
+}
 
-static irqreturn_t mv643xx_eth_int_handler(int irq, void *dev_id)
+static irqreturn_t mv643xx_eth_irq(int irq, void *dev_id)
 {
        struct net_device *dev = (struct net_device *)dev_id;
-       struct mv643xx_private *mp = netdev_priv(dev);
-       u32 eth_int_cause, eth_int_cause_ext = 0;
-       unsigned int port_num = mp->port_num;
-
-       /* Read interrupt cause registers */
-       eth_int_cause = rdl(mp, INT_CAUSE(port_num)) & (INT_RX | INT_EXT);
-       if (eth_int_cause & INT_EXT) {
-               eth_int_cause_ext = rdl(mp, INT_CAUSE_EXT(port_num))
-                               & (INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
-               wrl(mp, INT_CAUSE_EXT(port_num), ~eth_int_cause_ext);
-       }
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       u32 int_cause;
+       u32 int_cause_ext;
 
-       /* PHY status changed */
-       if (eth_int_cause_ext & (INT_EXT_LINK | INT_EXT_PHY)) {
-               struct ethtool_cmd cmd;
+       int_cause = rdl(mp, INT_CAUSE(mp->port_num)) &
+                       (INT_TX_END | INT_RX | INT_EXT);
+       if (int_cause == 0)
+               return IRQ_NONE;
 
-               if (mii_link_ok(&mp->mii)) {
-                       mii_ethtool_gset(&mp->mii, &cmd);
-                       mv643xx_eth_update_pscr(dev, &cmd);
-                       mv643xx_eth_port_enable_tx(mp, 1);
-                       if (!netif_carrier_ok(dev)) {
-                               netif_carrier_on(dev);
-                               if (mp->tx_ring_size - mp->tx_desc_count >=
-                                                       MAX_DESCS_PER_SKB)
-                                       netif_wake_queue(dev);
-                       }
-               } else if (netif_carrier_ok(dev)) {
-                       netif_stop_queue(dev);
-                       netif_carrier_off(dev);
-               }
+       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);
        }
 
-#ifdef MV643XX_NAPI
-       if (eth_int_cause & INT_RX) {
-               /* schedule the NAPI poll routine to maintain port */
-               wrl(mp, INT_MASK(port_num), 0x00000000);
+       if (int_cause_ext & (INT_EXT_PHY | INT_EXT_LINK))
+               handle_link_event(mp);
 
-               /* wait for previous write to complete */
-               rdl(mp, INT_MASK(port_num));
+       /*
+        * 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 (eth_int_cause & INT_RX)
-               mv643xx_eth_receive_queue(dev, INT_MAX);
+       if (int_cause & INT_RX) {
+               int i;
+
+               for (i = 7; i >= 0; i--)
+                       if (mp->rxq_mask & (1 << i))
+                               rxq_process(mp->rxq + i, INT_MAX);
+       }
 #endif
-       if (eth_int_cause_ext & INT_EXT_TX)
-               mv643xx_eth_free_completed_tx_descs(dev);
 
        /*
-        * If no real interrupt occured, exit.
-        * This can happen when using gigE interrupt coalescing mechanism.
+        * TxBuffer or TxError set for any of the 8 queues?
         */
-       if ((eth_int_cause == 0x0) && (eth_int_cause_ext == 0x0))
-               return IRQ_NONE;
+       if (int_cause_ext & INT_EXT_TX) {
+               int i;
+
+               for (i = 0; i < 8; i++)
+                       if (mp->txq_mask & (1 << i))
+                               txq_reclaim(mp->txq + i, 0);
+
+               /*
+                * 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);
+               }
+       }
+
+       /*
+        * Any TxEnd interrupts?
+        */
+       if (int_cause & INT_TX_END) {
+               int i;
+
+               wrl(mp, INT_CAUSE(mp->port_num), ~(int_cause & INT_TX_END));
+
+               spin_lock(&mp->lock);
+               for (i = 0; i < 8; i++) {
+                       struct tx_queue *txq = mp->txq + i;
+                       u32 hw_desc_ptr;
+                       u32 expected_ptr;
+
+                       if ((int_cause & (INT_TX_END_0 << i)) == 0)
+                               continue;
+
+                       hw_desc_ptr =
+                               rdl(mp, TXQ_CURRENT_DESC_PTR(mp->port_num, i));
+                       expected_ptr = (u32)txq->tx_desc_dma +
+                               txq->tx_curr_desc * sizeof(struct tx_desc);
+
+                       if (hw_desc_ptr != expected_ptr)
+                               txq_enable(txq);
+               }
+               spin_unlock(&mp->lock);
+       }
 
        return IRQ_HANDLED;
 }
 
-/*
- * ethernet_phy_reset - Reset Ethernet port PHY.
- *
- * DESCRIPTION:
- *     This routine utilizes the SMI interface to reset the ethernet port PHY.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     The PHY is reset.
- *
- * RETURN:
- *     None.
- *
- */
-static void ethernet_phy_reset(struct mv643xx_private *mp)
+static void phy_reset(struct mv643xx_eth_private *mp)
 {
-       unsigned int phy_reg_data;
+       unsigned int data;
 
-       /* Reset the PHY */
-       eth_port_read_smi_reg(mp, 0, &phy_reg_data);
-       phy_reg_data |= 0x8000; /* Set bit 15 to reset the PHY */
-       eth_port_write_smi_reg(mp, 0, phy_reg_data);
+       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
+       data |= BMCR_RESET;
+       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data);
 
-       /* wait for PHY to come out of reset */
        do {
                udelay(1);
-               eth_port_read_smi_reg(mp, 0, &phy_reg_data);
-       } while (phy_reg_data & 0x8000);
+               smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
+       } while (data & BMCR_RESET);
 }
 
-/*
- * eth_port_start - Start the Ethernet port activity.
- *
- * DESCRIPTION:
- *     This routine prepares the Ethernet port for Rx and Tx activity:
- *      1. Initialize Tx and Rx Current Descriptor Pointer for each queue that
- *         has been initialized a descriptor's ring (using
- *         ether_init_tx_desc_ring for Tx and ether_init_rx_desc_ring for Rx)
- *      2. Initialize and enable the Ethernet configuration port by writing to
- *         the port's configuration and command registers.
- *      3. Initialize and enable the SDMA by writing to the SDMA's
- *         configuration and command registers.  After completing these steps,
- *         the ethernet port SDMA can starts to perform Rx and Tx activities.
- *
- *     Note: Each Rx and Tx queue descriptor's list must be initialized prior
- *     to calling this function (use ether_init_tx_desc_ring for Tx queues
- *     and ether_init_rx_desc_ring for Rx queues).
- *
- * INPUT:
- *     dev - a pointer to the required interface
- *
- * OUTPUT:
- *     Ethernet port is ready to receive and transmit.
- *
- * RETURN:
- *     None.
- */
-static void eth_port_start(struct net_device *dev)
+static void port_start(struct mv643xx_eth_private *mp)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
-       int tx_curr_desc, rx_curr_desc;
        u32 pscr;
-       struct ethtool_cmd ethtool_cmd;
-
-       /* Assignment of Tx CTRP of given queue */
-       tx_curr_desc = mp->tx_curr_desc_q;
-       wrl(mp, TXQ_CURRENT_DESC_PTR(port_num),
-               (u32)((struct eth_tx_desc *)mp->tx_desc_dma + tx_curr_desc));
-
-       /* Assignment of Rx CRDP of given queue */
-       rx_curr_desc = mp->rx_curr_desc_q;
-       wrl(mp, RXQ_CURRENT_DESC_PTR(port_num),
-               (u32)((struct eth_rx_desc *)mp->rx_desc_dma + rx_curr_desc));
-
-       /* Add the assigned Ethernet address to the port's address table */
-       eth_port_uc_addr_set(mp, dev->dev_addr);
+       int i;
 
        /*
-        * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
-        * frames to RX queue #0.
+        * Perform PHY reset, if there is a PHY.
         */
-       wrl(mp, PORT_CONFIG(port_num), 0x00000000);
+       if (mp->phy_addr != -1) {
+               struct ethtool_cmd cmd;
+
+               mv643xx_eth_get_settings(mp->dev, &cmd);
+               phy_reset(mp);
+               mv643xx_eth_set_settings(mp->dev, &cmd);
+       }
 
        /*
-        * Treat BPDUs as normal multicasts, and disable partition mode.
+        * Configure basic link parameters.
         */
-       wrl(mp, PORT_CONFIG_EXT(port_num), 0x00000000);
-
-       pscr = rdl(mp, PORT_SERIAL_CONTROL(port_num));
-
-       pscr &= ~(SERIAL_PORT_ENABLE | FORCE_LINK_PASS);
-       wrl(mp, PORT_SERIAL_CONTROL(port_num), pscr);
-
-       pscr |= DISABLE_AUTO_NEG_FOR_FLOW_CTRL |
-               DISABLE_AUTO_NEG_SPEED_GMII    |
-               DISABLE_AUTO_NEG_FOR_DUPLEX    |
-               DO_NOT_FORCE_LINK_FAIL     |
-               SERIAL_PORT_CONTROL_RESERVED;
-
-       wrl(mp, PORT_SERIAL_CONTROL(port_num), pscr);
+       pscr = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
 
        pscr |= SERIAL_PORT_ENABLE;
-       wrl(mp, PORT_SERIAL_CONTROL(port_num), pscr);
+       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
 
-       /* Assign port SDMA configuration */
-       wrl(mp, SDMA_CONFIG(port_num), PORT_SDMA_CONFIG_DEFAULT_VALUE);
+       pscr |= DO_NOT_FORCE_LINK_FAIL;
+       if (mp->phy_addr == -1)
+               pscr |= FORCE_LINK_PASS;
+       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
 
-       /* Enable port Rx. */
-       mv643xx_eth_port_enable_rx(mp, 1);
+       wrl(mp, SDMA_CONFIG(mp->port_num), PORT_SDMA_CONFIG_DEFAULT_VALUE);
 
-       /* Disable port bandwidth limits by clearing MTU register */
-       wrl(mp, TX_BW_MTU(port_num), 0);
+       /*
+        * Configure TX path and queues.
+        */
+       tx_set_rate(mp, 1000000000, 16777216);
+       for (i = 0; i < 8; i++) {
+               struct tx_queue *txq = mp->txq + i;
 
-       /* save phy settings across reset */
-       mv643xx_get_settings(dev, &ethtool_cmd);
-       ethernet_phy_reset(mp);
-       mv643xx_set_settings(dev, &ethtool_cmd);
-}
+               if ((mp->txq_mask & (1 << i)) == 0)
+                       continue;
 
-#ifdef MV643XX_COAL
+               txq_reset_hw_ptr(txq);
+               txq_set_rate(txq, 1000000000, 16777216);
+               txq_set_fixed_prio_mode(txq);
+       }
 
-/*
- * eth_port_set_rx_coal - Sets coalescing interrupt mechanism on RX path
- *
- * DESCRIPTION:
- *     This routine sets the RX coalescing interrupt mechanism parameter.
- *     This parameter is a timeout counter, that counts in 64 t_clk
- *     chunks ; that when timeout event occurs a maskable interrupt
- *     occurs.
- *     The parameter is calculated using the tClk of the MV-643xx chip
- *     , and the required delay of the interrupt in usec.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet port
- *     unsigned int delay              Delay in usec
- *
- * OUTPUT:
- *     Interrupt coalescing mechanism value is set in MV-643xx chip.
- *
- * RETURN:
- *     The interrupt coalescing value set in the gigE port.
- *
- */
-static unsigned int eth_port_set_rx_coal(struct mv643xx_private *mp,
-                                       unsigned int delay)
-{
-       unsigned int port_num = mp->port_num;
-       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
+       /*
+        * Add configured unicast address to address filter table.
+        */
+       uc_addr_set(mp, mp->dev->dev_addr);
 
-       /* Set RX Coalescing mechanism */
-       wrl(mp, SDMA_CONFIG(port_num),
-               ((coal & 0x3fff) << 8) |
-               (rdl(mp, SDMA_CONFIG(port_num))
-                       & 0xffc000ff));
+       /*
+        * Receive all unmatched unicast, TCP, UDP, BPDU and broadcast
+        * frames to RX queue #0.
+        */
+       wrl(mp, PORT_CONFIG(mp->port_num), 0x00000000);
 
-       return coal;
-}
-#endif
+       /*
+        * Treat BPDUs as normal multicasts, and disable partition mode.
+        */
+       wrl(mp, PORT_CONFIG_EXT(mp->port_num), 0x00000000);
 
-/*
- * eth_port_set_tx_coal - Sets coalescing interrupt mechanism on TX path
- *
- * DESCRIPTION:
- *     This routine sets the TX coalescing interrupt mechanism parameter.
- *     This parameter is a timeout counter, that counts in 64 t_clk
- *     chunks ; that when timeout event occurs a maskable interrupt
- *     occurs.
- *     The parameter is calculated using the t_cLK frequency of the
- *     MV-643xx chip and the required delay in the interrupt in uSec
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet port
- *     unsigned int delay              Delay in uSeconds
- *
- * OUTPUT:
- *     Interrupt coalescing mechanism value is set in MV-643xx chip.
- *
- * RETURN:
- *     The interrupt coalescing value set in the gigE port.
- *
- */
-static unsigned int eth_port_set_tx_coal(struct mv643xx_private *mp,
-                                       unsigned int delay)
-{
-       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
+       /*
+        * Enable the receive queues.
+        */
+       for (i = 0; i < 8; i++) {
+               struct rx_queue *rxq = mp->rxq + i;
+               int off = RXQ_CURRENT_DESC_PTR(mp->port_num, i);
+               u32 addr;
+
+               if ((mp->rxq_mask & (1 << i)) == 0)
+                       continue;
 
-       /* Set TX Coalescing mechanism */
-       wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), coal << 4);
+               addr = (u32)rxq->rx_desc_dma;
+               addr += rxq->rx_curr_desc * sizeof(struct rx_desc);
+               wrl(mp, off, addr);
 
-       return coal;
+               rxq_enable(rxq);
+       }
 }
 
-/*
- * eth_port_init - Initialize the Ethernet port driver
- *
- * DESCRIPTION:
- *     This function prepares the ethernet port to start its activity:
- *     1) Completes the ethernet port driver struct initialization toward port
- *             start routine.
- *     2) Resets the device to a quiescent state in case of warm reboot.
- *     3) Enable SDMA access to all four DRAM banks as well as internal SRAM.
- *     4) Clean MAC tables. The reset status of those tables is unknown.
- *     5) Set PHY address.
- *     Note: Call this routine prior to eth_port_start routine and after
- *     setting user values in the user fields of Ethernet port control
- *     struct.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet port control struct
- *
- * OUTPUT:
- *     See description.
- *
- * RETURN:
- *     None.
- */
-static void eth_port_init(struct mv643xx_private *mp)
+static void set_rx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
 {
-       mp->rx_resource_err = 0;
+       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
+       u32 val;
+
+       val = rdl(mp, SDMA_CONFIG(mp->port_num));
+       if (mp->shared->extended_rx_coal_limit) {
+               if (coal > 0xffff)
+                       coal = 0xffff;
+               val &= ~0x023fff80;
+               val |= (coal & 0x8000) << 10;
+               val |= (coal & 0x7fff) << 7;
+       } else {
+               if (coal > 0x3fff)
+                       coal = 0x3fff;
+               val &= ~0x003fff00;
+               val |= (coal & 0x3fff) << 8;
+       }
+       wrl(mp, SDMA_CONFIG(mp->port_num), val);
+}
 
-       eth_port_reset(mp);
+static void set_tx_coal(struct mv643xx_eth_private *mp, unsigned int delay)
+{
+       unsigned int coal = ((mp->shared->t_clk / 1000000) * delay) / 64;
 
-       eth_port_init_mac_tables(mp);
+       if (coal > 0x3fff)
+               coal = 0x3fff;
+       wrl(mp, TX_FIFO_URGENT_THRESHOLD(mp->port_num), (coal & 0x3fff) << 4);
 }
 
-/*
- * mv643xx_eth_open
- *
- * This function is called when openning the network device. The function
- * should initialize all the hardware, initialize cyclic Rx/Tx
- * descriptors chain and buffers and allocate an IRQ to the network
- * device.
- *
- * Input :     a pointer to the network device structure
- *
- * Output :    zero of success , nonzero if fails.
- */
-
 static int mv643xx_eth_open(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
-       unsigned int size;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
        int err;
+       int i;
 
-       /* Clear any pending ethernet port interrupts */
-       wrl(mp, INT_CAUSE(port_num), 0);
-       wrl(mp, INT_CAUSE_EXT(port_num), 0);
-       /* wait for previous write to complete */
-       rdl(mp, INT_CAUSE_EXT(port_num));
+       wrl(mp, INT_CAUSE(mp->port_num), 0);
+       wrl(mp, INT_CAUSE_EXT(mp->port_num), 0);
+       rdl(mp, INT_CAUSE_EXT(mp->port_num));
 
-       err = request_irq(dev->irq, mv643xx_eth_int_handler,
-                       IRQF_SHARED | IRQF_SAMPLE_RANDOM, dev->name, dev);
+       err = request_irq(dev->irq, mv643xx_eth_irq,
+                         IRQF_SHARED | IRQF_SAMPLE_RANDOM,
+                         dev->name, dev);
        if (err) {
-               printk(KERN_ERR "%s: Can not assign IRQ\n", dev->name);
+               dev_printk(KERN_ERR, &dev->dev, "can't assign irq\n");
                return -EAGAIN;
        }
 
-       eth_port_init(mp);
+       init_mac_tables(mp);
 
-       memset(&mp->timeout, 0, sizeof(struct timer_list));
-       mp->timeout.function = mv643xx_eth_rx_refill_descs_timer_wrapper;
-       mp->timeout.data = (unsigned long)dev;
-
-       /* Allocate RX and TX skb rings */
-       mp->rx_skb = kmalloc(sizeof(*mp->rx_skb) * mp->rx_ring_size,
-                                                               GFP_KERNEL);
-       if (!mp->rx_skb) {
-               printk(KERN_ERR "%s: Cannot allocate Rx skb ring\n", dev->name);
-               err = -ENOMEM;
-               goto out_free_irq;
-       }
-       mp->tx_skb = kmalloc(sizeof(*mp->tx_skb) * mp->tx_ring_size,
-                                                               GFP_KERNEL);
-       if (!mp->tx_skb) {
-               printk(KERN_ERR "%s: Cannot allocate Tx skb ring\n", dev->name);
-               err = -ENOMEM;
-               goto out_free_rx_skb;
-       }
-
-       /* Allocate TX ring */
-       mp->tx_desc_count = 0;
-       size = mp->tx_ring_size * sizeof(struct eth_tx_desc);
-       mp->tx_desc_area_size = size;
-
-       if (mp->tx_sram_size) {
-               mp->p_tx_desc_area = ioremap(mp->tx_sram_addr,
-                                                       mp->tx_sram_size);
-               mp->tx_desc_dma = mp->tx_sram_addr;
-       } else
-               mp->p_tx_desc_area = dma_alloc_coherent(NULL, size,
-                                                       &mp->tx_desc_dma,
-                                                       GFP_KERNEL);
+       for (i = 0; i < 8; i++) {
+               if ((mp->rxq_mask & (1 << i)) == 0)
+                       continue;
 
-       if (!mp->p_tx_desc_area) {
-               printk(KERN_ERR "%s: Cannot allocate Tx Ring (size %d bytes)\n",
-                                                       dev->name, size);
-               err = -ENOMEM;
-               goto out_free_tx_skb;
-       }
-       BUG_ON((u32) mp->p_tx_desc_area & 0xf); /* check 16-byte alignment */
-       memset((void *)mp->p_tx_desc_area, 0, mp->tx_desc_area_size);
-
-       ether_init_tx_desc_ring(mp);
-
-       /* Allocate RX ring */
-       mp->rx_desc_count = 0;
-       size = mp->rx_ring_size * sizeof(struct eth_rx_desc);
-       mp->rx_desc_area_size = size;
-
-       if (mp->rx_sram_size) {
-               mp->p_rx_desc_area = ioremap(mp->rx_sram_addr,
-                                                       mp->rx_sram_size);
-               mp->rx_desc_dma = mp->rx_sram_addr;
-       } else
-               mp->p_rx_desc_area = dma_alloc_coherent(NULL, size,
-                                                       &mp->rx_desc_dma,
-                                                       GFP_KERNEL);
+               err = rxq_init(mp, i);
+               if (err) {
+                       while (--i >= 0)
+                               if (mp->rxq_mask & (1 << i))
+                                       rxq_deinit(mp->rxq + i);
+                       goto out;
+               }
 
-       if (!mp->p_rx_desc_area) {
-               printk(KERN_ERR "%s: Cannot allocate Rx ring (size %d bytes)\n",
-                                                       dev->name, size);
-               printk(KERN_ERR "%s: Freeing previously allocated TX queues...",
-                                                       dev->name);
-               if (mp->rx_sram_size)
-                       iounmap(mp->p_tx_desc_area);
-               else
-                       dma_free_coherent(NULL, mp->tx_desc_area_size,
-                                       mp->p_tx_desc_area, mp->tx_desc_dma);
-               err = -ENOMEM;
-               goto out_free_tx_skb;
+               rxq_refill(mp->rxq + i);
        }
-       memset((void *)mp->p_rx_desc_area, 0, size);
 
-       ether_init_rx_desc_ring(mp);
+       for (i = 0; i < 8; i++) {
+               if ((mp->txq_mask & (1 << i)) == 0)
+                       continue;
 
-       mv643xx_eth_rx_refill_descs(dev);       /* Fill RX ring with skb's */
+               err = txq_init(mp, i);
+               if (err) {
+                       while (--i >= 0)
+                               if (mp->txq_mask & (1 << i))
+                                       txq_deinit(mp->txq + i);
+                       goto out_free;
+               }
+       }
 
-#ifdef MV643XX_NAPI
+#ifdef MV643XX_ETH_NAPI
        napi_enable(&mp->napi);
 #endif
 
-       eth_port_start(dev);
-
-       /* Interrupt Coalescing */
+       netif_carrier_off(dev);
+       netif_stop_queue(dev);
 
-#ifdef MV643XX_COAL
-       mp->rx_int_coal =
-               eth_port_set_rx_coal(mp, MV643XX_RX_COAL);
-#endif
+       port_start(mp);
 
-       mp->tx_int_coal =
-               eth_port_set_tx_coal(mp, MV643XX_TX_COAL);
+       set_rx_coal(mp, 0);
+       set_tx_coal(mp, 0);
 
-       /* Unmask phy and link status changes interrupts */
-       wrl(mp, INT_MASK_EXT(port_num), INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
+       wrl(mp, INT_MASK_EXT(mp->port_num),
+           INT_EXT_LINK | INT_EXT_PHY | INT_EXT_TX);
 
-       /* Unmask RX buffer and TX end interrupt */
-       wrl(mp, INT_MASK(port_num), INT_RX | INT_EXT);
+       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
 
        return 0;
 
-out_free_tx_skb:
-       kfree(mp->tx_skb);
-out_free_rx_skb:
-       kfree(mp->rx_skb);
-out_free_irq:
+
+out_free:
+       for (i = 0; i < 8; i++)
+               if (mp->rxq_mask & (1 << i))
+                       rxq_deinit(mp->rxq + i);
+out:
        free_irq(dev->irq, dev);
 
        return err;
 }
 
-/*
- * eth_port_reset - Reset Ethernet port
- *
- * DESCRIPTION:
- *     This routine resets the chip by aborting any SDMA engine activity and
- *     clearing the MIB counters. The Receiver and the Transmit unit are in
- *     idle state after this command is performed and the port is disabled.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     Channel activity is halted.
- *
- * RETURN:
- *     None.
- *
- */
-static void eth_port_reset(struct mv643xx_private *mp)
+static void port_reset(struct mv643xx_eth_private *mp)
 {
-       unsigned int port_num = mp->port_num;
-       unsigned int reg_data;
+       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);
+       }
 
-       mv643xx_eth_port_disable_tx(mp);
-       mv643xx_eth_port_disable_rx(mp);
+       while (1) {
+               u32 ps = rdl(mp, PORT_STATUS(mp->port_num));
 
-       /* Clear all MIB counters */
-       eth_clear_mib_counters(mp);
+               if ((ps & (TX_IN_PROGRESS | TX_FIFO_EMPTY)) == TX_FIFO_EMPTY)
+                       break;
+               udelay(10);
+       }
 
        /* Reset the Enable bit in the Configuration Register */
-       reg_data = rdl(mp, PORT_SERIAL_CONTROL(port_num));
-       reg_data &= ~(SERIAL_PORT_ENABLE                |
-                       DO_NOT_FORCE_LINK_FAIL  |
-                       FORCE_LINK_PASS);
-       wrl(mp, PORT_SERIAL_CONTROL(port_num), reg_data);
+       data = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       data &= ~(SERIAL_PORT_ENABLE            |
+                 DO_NOT_FORCE_LINK_FAIL        |
+                 FORCE_LINK_PASS);
+       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), data);
 }
 
-/*
- * mv643xx_eth_stop
- *
- * This function is used when closing the network device.
- * It updates the hardware,
- * release all memory that holds buffers and descriptors and release the IRQ.
- * Input :     a pointer to the device structure
- * Output :    zero if success , nonzero if fails
- */
-
 static int mv643xx_eth_stop(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       int i;
 
-       /* Mask all interrupts on ethernet port */
-       wrl(mp, INT_MASK(port_num), 0x00000000);
-       /* wait for previous write to complete */
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
+       rdl(mp, INT_MASK(mp->port_num));
 
-#ifdef MV643XX_NAPI
+#ifdef MV643XX_ETH_NAPI
        napi_disable(&mp->napi);
 #endif
        netif_carrier_off(dev);
        netif_stop_queue(dev);
 
-       eth_port_reset(mp);
+       free_irq(dev->irq, dev);
 
-       mv643xx_eth_free_tx_rings(dev);
-       mv643xx_eth_free_rx_rings(dev);
+       port_reset(mp);
+       mib_counters_update(mp);
 
-       free_irq(dev->irq, dev);
+       for (i = 0; i < 8; i++) {
+               if (mp->rxq_mask & (1 << i))
+                       rxq_deinit(mp->rxq + i);
+               if (mp->txq_mask & (1 << i))
+                       txq_deinit(mp->txq + i);
+       }
 
        return 0;
 }
 
-static int mv643xx_eth_do_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
+static int mv643xx_eth_ioctl(struct net_device *dev, struct ifreq *ifr, int cmd)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       if (mp->phy_addr != -1)
+               return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
 
-       return generic_mii_ioctl(&mp->mii, if_mii(ifr), cmd, NULL);
+       return -EOPNOTSUPP;
 }
 
-/*
- * Changes MTU (maximum transfer unit) of the gigabit ethenret port
- *
- * Input :     pointer to ethernet interface network device structure
- *             new mtu size
- * Output :    0 upon success, -EINVAL upon failure
- */
 static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
 {
-       if ((new_mtu > 9500) || (new_mtu < 64))
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+
+       if (new_mtu < 64 || new_mtu > 9500)
                return -EINVAL;
 
        dev->mtu = new_mtu;
+       tx_set_rate(mp, 1000000000, 16777216);
+
        if (!netif_running(dev))
                return 0;
 
@@ -2466,98 +2155,79 @@ static int mv643xx_eth_change_mtu(struct net_device *dev, int new_mtu)
         * Stop and then re-open the interface. This will allocate RX
         * skbs of the new MTU.
         * There is a possible danger that the open will not succeed,
-        * due to memory being full, which might fail the open function.
+        * due to memory being full.
         */
        mv643xx_eth_stop(dev);
        if (mv643xx_eth_open(dev)) {
-               printk(KERN_ERR "%s: Fatal error on opening device\n",
-                       dev->name);
+               dev_printk(KERN_ERR, &dev->dev,
+                          "fatal error on re-opening device after "
+                          "MTU change\n");
        }
 
        return 0;
 }
 
-/*
- * mv643xx_eth_tx_timeout_task
- *
- * Actual routine to reset the adapter when a timeout on Tx has occurred
- */
-static void mv643xx_eth_tx_timeout_task(struct work_struct *ugly)
+static void tx_timeout_task(struct work_struct *ugly)
 {
-       struct mv643xx_private *mp = container_of(ugly, struct mv643xx_private,
-                                                 tx_timeout_task);
-       struct net_device *dev = mp->dev;
-
-       if (!netif_running(dev))
-               return;
+       struct mv643xx_eth_private *mp;
 
-       netif_stop_queue(dev);
+       mp = container_of(ugly, struct mv643xx_eth_private, tx_timeout_task);
+       if (netif_running(mp->dev)) {
+               netif_stop_queue(mp->dev);
 
-       eth_port_reset(mp);
-       eth_port_start(dev);
+               port_reset(mp);
+               port_start(mp);
 
-       if (mp->tx_ring_size - mp->tx_desc_count >= MAX_DESCS_PER_SKB)
-               netif_wake_queue(dev);
+               __txq_maybe_wake(mp->txq + mp->txq_primary);
+       }
 }
 
-/*
- * mv643xx_eth_tx_timeout
- *
- * Called upon a timeout on transmitting a packet
- *
- * Input :     pointer to ethernet interface network device structure.
- * Output :    N/A
- */
 static void mv643xx_eth_tx_timeout(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       printk(KERN_INFO "%s: TX timeout  ", dev->name);
+       dev_printk(KERN_INFO, &dev->dev, "tx timeout\n");
 
-       /* Do the reset outside of interrupt context */
        schedule_work(&mp->tx_timeout_task);
 }
 
 #ifdef CONFIG_NET_POLL_CONTROLLER
-static void mv643xx_netpoll(struct net_device *netdev)
+static void mv643xx_eth_netpoll(struct net_device *dev)
 {
-       struct mv643xx_private *mp = netdev_priv(netdev);
-       int port_num = mp->port_num;
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
 
-       wrl(mp, INT_MASK(port_num), 0x00000000);
-       /* wait for previous write to complete */
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0x00000000);
+       rdl(mp, INT_MASK(mp->port_num));
 
-       mv643xx_eth_int_handler(netdev->irq, netdev);
+       mv643xx_eth_irq(dev->irq, dev);
 
-       wrl(mp, INT_MASK(port_num), INT_RX | INT_CAUSE_EXT);
+       wrl(mp, INT_MASK(mp->port_num), INT_TX_END | INT_RX | INT_EXT);
 }
 #endif
 
-/*
- * Wrappers for MII support library.
- */
-static int mv643xx_mdio_read(struct net_device *dev, int phy_id, int location)
+static int mv643xx_eth_mdio_read(struct net_device *dev, int addr, int reg)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
        int val;
 
-       eth_port_read_smi_reg(mp, location, &val);
+       smi_reg_read(mp, addr, reg, &val);
+
        return val;
 }
 
-static void mv643xx_mdio_write(struct net_device *dev, int phy_id, int location, int val)
+static void mv643xx_eth_mdio_write(struct net_device *dev, int addr, int reg, int val)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
-       eth_port_write_smi_reg(mp, location, val);
+       struct mv643xx_eth_private *mp = netdev_priv(dev);
+       smi_reg_write(mp, addr, reg, val);
 }
 
 
 /* platform glue ************************************************************/
-static void mv643xx_eth_conf_mbus_windows(struct mv643xx_shared_private *msp,
-                                         struct mbus_dram_target_info *dram)
+static void
+mv643xx_eth_conf_mbus_windows(struct mv643xx_eth_shared_private *msp,
+                             struct mbus_dram_target_info *dram)
 {
-       void __iomem *base = msp->eth_base;
+       void __iomem *base = msp->base;
        u32 win_enable;
        u32 win_protect;
        int i;
@@ -2588,16 +2258,41 @@ static void mv643xx_eth_conf_mbus_windows(struct mv643xx_shared_private *msp,
        msp->win_protect = win_protect;
 }
 
+static void infer_hw_params(struct mv643xx_eth_shared_private *msp)
+{
+       /*
+        * Check whether we have a 14-bit coal limit field in bits
+        * [21:8], or a 16-bit coal limit in bits [25,21:7] of the
+        * SDMA config register.
+        */
+       writel(0x02000000, msp->base + SDMA_CONFIG(0));
+       if (readl(msp->base + SDMA_CONFIG(0)) & 0x02000000)
+               msp->extended_rx_coal_limit = 1;
+       else
+               msp->extended_rx_coal_limit = 0;
+
+       /*
+        * Check whether the TX rate control registers are in the
+        * old or the new place.
+        */
+       writel(1, msp->base + TX_BW_MTU_MOVED(0));
+       if (readl(msp->base + TX_BW_MTU_MOVED(0)) & 1)
+               msp->tx_bw_control_moved = 1;
+       else
+               msp->tx_bw_control_moved = 0;
+}
+
 static int mv643xx_eth_shared_probe(struct platform_device *pdev)
 {
-       static int mv643xx_version_printed = 0;
+       static int mv643xx_eth_version_printed = 0;
        struct mv643xx_eth_shared_platform_data *pd = pdev->dev.platform_data;
-       struct mv643xx_shared_private *msp;
+       struct mv643xx_eth_shared_private *msp;
        struct resource *res;
        int ret;
 
-       if (!mv643xx_version_printed++)
-               printk(KERN_NOTICE "MV-643xx 10/100/1000 Ethernet Driver\n");
+       if (!mv643xx_eth_version_printed++)
+               printk(KERN_NOTICE "MV-643xx 10/100/1000 ethernet "
+                       "driver version %s\n", mv643xx_eth_driver_version);
 
        ret = -EINVAL;
        res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
@@ -2610,14 +2305,11 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
                goto out;
        memset(msp, 0, sizeof(*msp));
 
-       msp->eth_base = ioremap(res->start, res->end - res->start + 1);
-       if (msp->eth_base == NULL)
+       msp->base = ioremap(res->start, res->end - res->start + 1);
+       if (msp->base == NULL)
                goto out_free;
 
        spin_lock_init(&msp->phy_lock);
-       msp->t_clk = (pd != NULL && pd->t_clk != 0) ? pd->t_clk : 133000000;
-
-       platform_set_drvdata(pdev, msp);
 
        /*
         * (Re-)program MBUS remapping windows if we are asked to.
@@ -2625,6 +2317,14 @@ static int mv643xx_eth_shared_probe(struct platform_device *pdev)
        if (pd != NULL && pd->dram != NULL)
                mv643xx_eth_conf_mbus_windows(msp, pd->dram);
 
+       /*
+        * Detect hardware parameters.
+        */
+       msp->t_clk = (pd != NULL && pd->t_clk != 0) ? pd->t_clk : 133000000;
+       infer_hw_params(msp);
+
+       platform_set_drvdata(pdev, msp);
+
        return 0;
 
 out_free:
@@ -2635,321 +2335,294 @@ out:
 
 static int mv643xx_eth_shared_remove(struct platform_device *pdev)
 {
-       struct mv643xx_shared_private *msp = platform_get_drvdata(pdev);
+       struct mv643xx_eth_shared_private *msp = platform_get_drvdata(pdev);
 
-       iounmap(msp->eth_base);
+       iounmap(msp->base);
        kfree(msp);
 
        return 0;
 }
 
 static struct platform_driver mv643xx_eth_shared_driver = {
-       .probe = mv643xx_eth_shared_probe,
-       .remove = mv643xx_eth_shared_remove,
+       .probe          = mv643xx_eth_shared_probe,
+       .remove         = mv643xx_eth_shared_remove,
        .driver = {
-               .name = MV643XX_ETH_SHARED_NAME,
+               .name   = MV643XX_ETH_SHARED_NAME,
                .owner  = THIS_MODULE,
        },
 };
 
-/*
- * ethernet_phy_set - Set the ethernet port PHY address.
- *
- * DESCRIPTION:
- *     This routine sets the given ethernet port PHY address.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *     int             phy_addr        PHY address.
- *
- * OUTPUT:
- *     None.
- *
- * RETURN:
- *     None.
- *
- */
-static void ethernet_phy_set(struct mv643xx_private *mp, int phy_addr)
+static void phy_addr_set(struct mv643xx_eth_private *mp, int phy_addr)
 {
-       u32 reg_data;
        int addr_shift = 5 * mp->port_num;
+       u32 data;
 
-       reg_data = rdl(mp, PHY_ADDR);
-       reg_data &= ~(0x1f << addr_shift);
-       reg_data |= (phy_addr & 0x1f) << addr_shift;
-       wrl(mp, PHY_ADDR, reg_data);
+       data = rdl(mp, PHY_ADDR);
+       data &= ~(0x1f << addr_shift);
+       data |= (phy_addr & 0x1f) << addr_shift;
+       wrl(mp, PHY_ADDR, data);
 }
 
-/*
- * ethernet_phy_get - Get the ethernet port PHY address.
- *
- * DESCRIPTION:
- *     This routine returns the given ethernet port PHY address.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     None.
- *
- * RETURN:
- *     PHY address.
- *
- */
-static int ethernet_phy_get(struct mv643xx_private *mp)
+static int phy_addr_get(struct mv643xx_eth_private *mp)
 {
-       unsigned int reg_data;
+       unsigned int data;
 
-       reg_data = rdl(mp, PHY_ADDR);
+       data = rdl(mp, PHY_ADDR);
 
-       return ((reg_data >> (5 * mp->port_num)) & 0x1f);
+       return (data >> (5 * mp->port_num)) & 0x1f;
 }
 
-/*
- * ethernet_phy_detect - Detect whether a phy is present
- *
- * DESCRIPTION:
- *     This function tests whether there is a PHY present on
- *     the specified port.
- *
- * INPUT:
- *     struct mv643xx_private *mp      Ethernet Port.
- *
- * OUTPUT:
- *     None
- *
- * RETURN:
- *     0 on success
- *     -ENODEV on failure
- *
- */
-static int ethernet_phy_detect(struct mv643xx_private *mp)
+static void set_params(struct mv643xx_eth_private *mp,
+                      struct mv643xx_eth_platform_data *pd)
+{
+       struct net_device *dev = mp->dev;
+
+       if (is_valid_ether_addr(pd->mac_addr))
+               memcpy(dev->dev_addr, pd->mac_addr, 6);
+       else
+               uc_addr_get(mp, dev->dev_addr);
+
+       if (pd->phy_addr == -1) {
+               mp->shared_smi = NULL;
+               mp->phy_addr = -1;
+       } else {
+               mp->shared_smi = mp->shared;
+               if (pd->shared_smi != NULL)
+                       mp->shared_smi = platform_get_drvdata(pd->shared_smi);
+
+               if (pd->force_phy_addr || pd->phy_addr) {
+                       mp->phy_addr = pd->phy_addr & 0x3f;
+                       phy_addr_set(mp, mp->phy_addr);
+               } else {
+                       mp->phy_addr = phy_addr_get(mp);
+               }
+       }
+
+       mp->default_rx_ring_size = DEFAULT_RX_QUEUE_SIZE;
+       if (pd->rx_queue_size)
+               mp->default_rx_ring_size = pd->rx_queue_size;
+       mp->rx_desc_sram_addr = pd->rx_sram_addr;
+       mp->rx_desc_sram_size = pd->rx_sram_size;
+
+       if (pd->rx_queue_mask)
+               mp->rxq_mask = pd->rx_queue_mask;
+       else
+               mp->rxq_mask = 0x01;
+       mp->rxq_primary = fls(mp->rxq_mask) - 1;
+
+       mp->default_tx_ring_size = DEFAULT_TX_QUEUE_SIZE;
+       if (pd->tx_queue_size)
+               mp->default_tx_ring_size = pd->tx_queue_size;
+       mp->tx_desc_sram_addr = pd->tx_sram_addr;
+       mp->tx_desc_sram_size = pd->tx_sram_size;
+
+       if (pd->tx_queue_mask)
+               mp->txq_mask = pd->tx_queue_mask;
+       else
+               mp->txq_mask = 0x01;
+       mp->txq_primary = fls(mp->txq_mask) - 1;
+}
+
+static int phy_detect(struct mv643xx_eth_private *mp)
 {
-       unsigned int phy_reg_data0;
-       int auto_neg;
+       unsigned int data;
+       unsigned int data2;
 
-       eth_port_read_smi_reg(mp, 0, &phy_reg_data0);
-       auto_neg = phy_reg_data0 & 0x1000;
-       phy_reg_data0 ^= 0x1000;        /* invert auto_neg */
-       eth_port_write_smi_reg(mp, 0, phy_reg_data0);
+       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data);
+       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data ^ BMCR_ANENABLE);
 
-       eth_port_read_smi_reg(mp, 0, &phy_reg_data0);
-       if ((phy_reg_data0 & 0x1000) == auto_neg)
-               return -ENODEV;                         /* change didn't take */
+       smi_reg_read(mp, mp->phy_addr, MII_BMCR, &data2);
+       if (((data ^ data2) & BMCR_ANENABLE) == 0)
+               return -ENODEV;
+
+       smi_reg_write(mp, mp->phy_addr, MII_BMCR, data);
 
-       phy_reg_data0 ^= 0x1000;
-       eth_port_write_smi_reg(mp, 0, phy_reg_data0);
        return 0;
 }
 
-static void mv643xx_init_ethtool_cmd(struct net_device *dev, int phy_address,
-                                    int speed, int duplex,
-                                    struct ethtool_cmd *cmd)
+static int phy_init(struct mv643xx_eth_private *mp,
+                   struct mv643xx_eth_platform_data *pd)
 {
-       struct mv643xx_private *mp = netdev_priv(dev);
+       struct ethtool_cmd cmd;
+       int err;
 
-       memset(cmd, 0, sizeof(*cmd));
+       err = phy_detect(mp);
+       if (err) {
+               dev_printk(KERN_INFO, &mp->dev->dev,
+                          "no PHY detected at addr %d\n", mp->phy_addr);
+               return err;
+       }
+       phy_reset(mp);
 
-       cmd->port = PORT_MII;
-       cmd->transceiver = XCVR_INTERNAL;
-       cmd->phy_address = phy_address;
+       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;
 
-       if (speed == 0) {
-               cmd->autoneg = AUTONEG_ENABLE;
-               /* mii lib checks, but doesn't use speed on AUTONEG_ENABLE */
-               cmd->speed = SPEED_100;
-               cmd->advertising = ADVERTISED_10baseT_Half  |
-                                  ADVERTISED_10baseT_Full  |
-                                  ADVERTISED_100baseT_Half |
-                                  ADVERTISED_100baseT_Full;
+       mp->mii.supports_gmii = mii_check_gmii_support(&mp->mii);
+
+       memset(&cmd, 0, sizeof(cmd));
+
+       cmd.port = PORT_MII;
+       cmd.transceiver = XCVR_INTERNAL;
+       cmd.phy_address = mp->phy_addr;
+       if (pd->speed == 0) {
+               cmd.autoneg = AUTONEG_ENABLE;
+               cmd.speed = SPEED_100;
+               cmd.advertising = ADVERTISED_10baseT_Half  |
+                                 ADVERTISED_10baseT_Full  |
+                                 ADVERTISED_100baseT_Half |
+                                 ADVERTISED_100baseT_Full;
                if (mp->mii.supports_gmii)
-                       cmd->advertising |= ADVERTISED_1000baseT_Full;
+                       cmd.advertising |= ADVERTISED_1000baseT_Full;
        } else {
-               cmd->autoneg = AUTONEG_DISABLE;
-               cmd->speed = speed;
-               cmd->duplex = duplex;
+               cmd.autoneg = AUTONEG_DISABLE;
+               cmd.speed = pd->speed;
+               cmd.duplex = pd->duplex;
        }
+
+       mv643xx_eth_set_settings(mp->dev, &cmd);
+
+       return 0;
+}
+
+static void init_pscr(struct mv643xx_eth_private *mp, int speed, int duplex)
+{
+       u32 pscr;
+
+       pscr = rdl(mp, PORT_SERIAL_CONTROL(mp->port_num));
+       if (pscr & SERIAL_PORT_ENABLE) {
+               pscr &= ~SERIAL_PORT_ENABLE;
+               wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
+       }
+
+       pscr = MAX_RX_PACKET_9700BYTE | SERIAL_PORT_CONTROL_RESERVED;
+       if (mp->phy_addr == -1) {
+               pscr |= DISABLE_AUTO_NEG_SPEED_GMII;
+               if (speed == SPEED_1000)
+                       pscr |= SET_GMII_SPEED_TO_1000;
+               else if (speed == SPEED_100)
+                       pscr |= SET_MII_SPEED_TO_100;
+
+               pscr |= DISABLE_AUTO_NEG_FOR_FLOW_CTRL;
+
+               pscr |= DISABLE_AUTO_NEG_FOR_DUPLEX;
+               if (duplex == DUPLEX_FULL)
+                       pscr |= SET_FULL_DUPLEX_MODE;
+       }
+
+       wrl(mp, PORT_SERIAL_CONTROL(mp->port_num), pscr);
 }
 
-/*/
- * mv643xx_eth_probe
- *
- * First function called after registering the network device.
- * It's purpose is to initialize the device as an ethernet device,
- * fill the ethernet device structure with pointers * to functions,
- * and set the MAC address of the interface
- *
- * Input :     struct device *
- * Output :    -ENOMEM if failed , 0 if success
- */
 static int mv643xx_eth_probe(struct platform_device *pdev)
 {
        struct mv643xx_eth_platform_data *pd;
-       int port_num;
-       struct mv643xx_private *mp;
+       struct mv643xx_eth_private *mp;
        struct net_device *dev;
-       u8 *p;
        struct resource *res;
-       int err;
-       struct ethtool_cmd cmd;
-       int duplex = DUPLEX_HALF;
-       int speed = 0;                  /* default to auto-negotiation */
        DECLARE_MAC_BUF(mac);
+       int err;
 
        pd = pdev->dev.platform_data;
        if (pd == NULL) {
-               printk(KERN_ERR "No mv643xx_eth_platform_data\n");
+               dev_printk(KERN_ERR, &pdev->dev,
+                          "no mv643xx_eth_platform_data\n");
                return -ENODEV;
        }
 
        if (pd->shared == NULL) {
-               printk(KERN_ERR "No mv643xx_eth_platform_data->shared\n");
+               dev_printk(KERN_ERR, &pdev->dev,
+                          "no mv643xx_eth_platform_data->shared\n");
                return -ENODEV;
        }
 
-       dev = alloc_etherdev(sizeof(struct mv643xx_private));
+       dev = alloc_etherdev(sizeof(struct mv643xx_eth_private));
        if (!dev)
                return -ENOMEM;
 
-       platform_set_drvdata(pdev, dev);
-
        mp = netdev_priv(dev);
+       platform_set_drvdata(pdev, mp);
+
+       mp->shared = platform_get_drvdata(pd->shared);
+       mp->port_num = pd->port_number;
+
        mp->dev = dev;
-#ifdef MV643XX_NAPI
-       netif_napi_add(dev, &mp->napi, mv643xx_poll, 64);
+#ifdef MV643XX_ETH_NAPI
+       netif_napi_add(dev, &mp->napi, mv643xx_eth_poll, 64);
 #endif
 
+       set_params(mp, pd);
+
+       spin_lock_init(&mp->lock);
+
+       mib_counters_clear(mp);
+       INIT_WORK(&mp->tx_timeout_task, tx_timeout_task);
+
+       if (mp->phy_addr != -1) {
+               err = phy_init(mp, pd);
+               if (err)
+                       goto out;
+
+               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops);
+       } else {
+               SET_ETHTOOL_OPS(dev, &mv643xx_eth_ethtool_ops_phyless);
+       }
+       init_pscr(mp, pd->speed, pd->duplex);
+
+
        res = platform_get_resource(pdev, IORESOURCE_IRQ, 0);
        BUG_ON(!res);
        dev->irq = res->start;
 
+       dev->hard_start_xmit = mv643xx_eth_xmit;
        dev->open = mv643xx_eth_open;
        dev->stop = mv643xx_eth_stop;
-       dev->hard_start_xmit = mv643xx_eth_start_xmit;
-       dev->set_mac_address = mv643xx_eth_set_mac_address;
        dev->set_multicast_list = mv643xx_eth_set_rx_mode;
-
-       /* No need to Tx Timeout */
+       dev->set_mac_address = mv643xx_eth_set_mac_address;
+       dev->do_ioctl = mv643xx_eth_ioctl;
+       dev->change_mtu = mv643xx_eth_change_mtu;
        dev->tx_timeout = mv643xx_eth_tx_timeout;
-
 #ifdef CONFIG_NET_POLL_CONTROLLER
-       dev->poll_controller = mv643xx_netpoll;
+       dev->poll_controller = mv643xx_eth_netpoll;
 #endif
-
        dev->watchdog_timeo = 2 * HZ;
        dev->base_addr = 0;
-       dev->change_mtu = mv643xx_eth_change_mtu;
-       dev->do_ioctl = mv643xx_eth_do_ioctl;
-       SET_ETHTOOL_OPS(dev, &mv643xx_ethtool_ops);
 
-#ifdef MV643XX_CHECKSUM_OFFLOAD_TX
-#ifdef MAX_SKB_FRAGS
+#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;
-#endif
+       dev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM;
 #endif
 
-       /* Configure the timeout task */
-       INIT_WORK(&mp->tx_timeout_task, mv643xx_eth_tx_timeout_task);
-
-       spin_lock_init(&mp->lock);
-
-       mp->shared = platform_get_drvdata(pd->shared);
-       port_num = mp->port_num = pd->port_number;
+       SET_NETDEV_DEV(dev, &pdev->dev);
 
        if (mp->shared->win_protect)
-               wrl(mp, WINDOW_PROTECT(port_num), mp->shared->win_protect);
-
-       mp->shared_smi = mp->shared;
-       if (pd->shared_smi != NULL)
-               mp->shared_smi = platform_get_drvdata(pd->shared_smi);
-
-       /* set default config values */
-       eth_port_uc_addr_get(mp, dev->dev_addr);
-       mp->rx_ring_size = PORT_DEFAULT_RECEIVE_QUEUE_SIZE;
-       mp->tx_ring_size = PORT_DEFAULT_TRANSMIT_QUEUE_SIZE;
-
-       if (is_valid_ether_addr(pd->mac_addr))
-               memcpy(dev->dev_addr, pd->mac_addr, 6);
-
-       if (pd->phy_addr || pd->force_phy_addr)
-               ethernet_phy_set(mp, pd->phy_addr);
-
-       if (pd->rx_queue_size)
-               mp->rx_ring_size = pd->rx_queue_size;
-
-       if (pd->tx_queue_size)
-               mp->tx_ring_size = pd->tx_queue_size;
-
-       if (pd->tx_sram_size) {
-               mp->tx_sram_size = pd->tx_sram_size;
-               mp->tx_sram_addr = pd->tx_sram_addr;
-       }
-
-       if (pd->rx_sram_size) {
-               mp->rx_sram_size = pd->rx_sram_size;
-               mp->rx_sram_addr = pd->rx_sram_addr;
-       }
-
-       duplex = pd->duplex;
-       speed = pd->speed;
+               wrl(mp, WINDOW_PROTECT(mp->port_num), mp->shared->win_protect);
 
-       /* Hook up MII support for ethtool */
-       mp->mii.dev = dev;
-       mp->mii.mdio_read = mv643xx_mdio_read;
-       mp->mii.mdio_write = mv643xx_mdio_write;
-       mp->mii.phy_id = ethernet_phy_get(mp);
-       mp->mii.phy_id_mask = 0x3f;
-       mp->mii.reg_num_mask = 0x1f;
-
-       err = ethernet_phy_detect(mp);
-       if (err) {
-               pr_debug("%s: No PHY detected at addr %d\n",
-                               dev->name, ethernet_phy_get(mp));
-               goto out;
-       }
-
-       ethernet_phy_reset(mp);
-       mp->mii.supports_gmii = mii_check_gmii_support(&mp->mii);
-       mv643xx_init_ethtool_cmd(dev, mp->mii.phy_id, speed, duplex, &cmd);
-       mv643xx_eth_update_pscr(dev, &cmd);
-       mv643xx_set_settings(dev, &cmd);
-
-       SET_NETDEV_DEV(dev, &pdev->dev);
        err = register_netdev(dev);
        if (err)
                goto out;
 
-       p = dev->dev_addr;
-       printk(KERN_NOTICE
-               "%s: port %d with MAC address %s\n",
-               dev->name, port_num, print_mac(mac, p));
+       dev_printk(KERN_NOTICE, &dev->dev, "port %d with MAC address %s\n",
+                  mp->port_num, print_mac(mac, dev->dev_addr));
 
        if (dev->features & NETIF_F_SG)
-               printk(KERN_NOTICE "%s: Scatter Gather Enabled\n", dev->name);
+               dev_printk(KERN_NOTICE, &dev->dev, "scatter/gather enabled\n");
 
        if (dev->features & NETIF_F_IP_CSUM)
-               printk(KERN_NOTICE "%s: TX TCP/IP Checksumming Supported\n",
-                                                               dev->name);
-
-#ifdef MV643XX_CHECKSUM_OFFLOAD_TX
-       printk(KERN_NOTICE "%s: RX TCP/UDP Checksum Offload ON \n", dev->name);
-#endif
+               dev_printk(KERN_NOTICE, &dev->dev, "tx checksum offload\n");
 
-#ifdef MV643XX_COAL
-       printk(KERN_NOTICE "%s: TX and RX Interrupt Coalescing ON \n",
-                                                               dev->name);
+#ifdef MV643XX_ETH_NAPI
+       dev_printk(KERN_NOTICE, &dev->dev, "napi enabled\n");
 #endif
 
-#ifdef MV643XX_NAPI
-       printk(KERN_NOTICE "%s: RX NAPI Enabled \n", dev->name);
-#endif
-
-       if (mp->tx_sram_size > 0)
-               printk(KERN_NOTICE "%s: Using SRAM\n", dev->name);
+       if (mp->tx_desc_sram_size > 0)
+               dev_printk(KERN_NOTICE, &dev->dev, "configured with sram\n");
 
        return 0;
 
@@ -2961,49 +2634,40 @@ out:
 
 static int mv643xx_eth_remove(struct platform_device *pdev)
 {
-       struct net_device *dev = platform_get_drvdata(pdev);
+       struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
-       unregister_netdev(dev);
+       unregister_netdev(mp->dev);
        flush_scheduled_work();
+       free_netdev(mp->dev);
 
-       free_netdev(dev);
        platform_set_drvdata(pdev, NULL);
+
        return 0;
 }
 
 static void mv643xx_eth_shutdown(struct platform_device *pdev)
 {
-       struct net_device *dev = platform_get_drvdata(pdev);
-       struct mv643xx_private *mp = netdev_priv(dev);
-       unsigned int port_num = mp->port_num;
+       struct mv643xx_eth_private *mp = platform_get_drvdata(pdev);
 
        /* Mask all interrupts on ethernet port */
-       wrl(mp, INT_MASK(port_num), 0);
-       rdl(mp, INT_MASK(port_num));
+       wrl(mp, INT_MASK(mp->port_num), 0);
+       rdl(mp, INT_MASK(mp->port_num));
 
-       eth_port_reset(mp);
+       if (netif_running(mp->dev))
+               port_reset(mp);
 }
 
 static struct platform_driver mv643xx_eth_driver = {
-       .probe = mv643xx_eth_probe,
-       .remove = mv643xx_eth_remove,
-       .shutdown = mv643xx_eth_shutdown,
+       .probe          = mv643xx_eth_probe,
+       .remove         = mv643xx_eth_remove,
+       .shutdown       = mv643xx_eth_shutdown,
        .driver = {
-               .name = MV643XX_ETH_NAME,
+               .name   = MV643XX_ETH_NAME,
                .owner  = THIS_MODULE,
        },
 };
 
-/*
- * mv643xx_init_module
- *
- * Registers the network drivers into the Linux kernel
- *
- * Input :     N/A
- *
- * Output :    N/A
- */
-static int __init mv643xx_init_module(void)
+static int __init mv643xx_eth_init_module(void)
 {
        int rc;
 
@@ -3013,30 +2677,21 @@ static int __init mv643xx_init_module(void)
                if (rc)
                        platform_driver_unregister(&mv643xx_eth_shared_driver);
        }
+
        return rc;
 }
+module_init(mv643xx_eth_init_module);
 
-/*
- * mv643xx_cleanup_module
- *
- * Registers the network drivers into the Linux kernel
- *
- * Input :     N/A
- *
- * Output :    N/A
- */
-static void __exit mv643xx_cleanup_module(void)
+static void __exit mv643xx_eth_cleanup_module(void)
 {
        platform_driver_unregister(&mv643xx_eth_driver);
        platform_driver_unregister(&mv643xx_eth_shared_driver);
 }
+module_exit(mv643xx_eth_cleanup_module);
 
-module_init(mv643xx_init_module);
-module_exit(mv643xx_cleanup_module);
-
-MODULE_LICENSE("GPL");
-MODULE_AUTHOR( "Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, Manish Lachwani"
-               " and Dale Farnsworth");
+MODULE_AUTHOR("Rabeeh Khoury, Assaf Hoffman, Matthew Dharm, "
+             "Manish Lachwani, Dale Farnsworth and Lennert Buytenhek");
 MODULE_DESCRIPTION("Ethernet driver for Marvell MV643XX");
-MODULE_ALIAS("platform:" MV643XX_ETH_NAME);
+MODULE_LICENSE("GPL");
 MODULE_ALIAS("platform:" MV643XX_ETH_SHARED_NAME);
+MODULE_ALIAS("platform:" MV643XX_ETH_NAME);