]> err.no Git - linux-2.6/blobdiff - drivers/net/bnx2x_main.c
bnx2x: HW attention lock
[linux-2.6] / drivers / net / bnx2x_main.c
index fabde5555e326810c7059f3bd6256948967fc8d9..c8b61788abb8105ce89153f084b2960b9073435c 100644 (file)
@@ -60,8 +60,8 @@
 #include "bnx2x.h"
 #include "bnx2x_init.h"
 
-#define DRV_MODULE_VERSION      "1.42.4"
-#define DRV_MODULE_RELDATE      "2008/4/9"
+#define DRV_MODULE_VERSION      "1.45.6"
+#define DRV_MODULE_RELDATE      "2008/06/23"
 #define BNX2X_BC_VER           0x040200
 
 /* Time in jiffies before concluding the transmitter is hung */
@@ -76,23 +76,21 @@ MODULE_DESCRIPTION("Broadcom NetXtreme II BCM57710 Driver");
 MODULE_LICENSE("GPL");
 MODULE_VERSION(DRV_MODULE_VERSION);
 
+static int disable_tpa;
 static int use_inta;
 static int poll;
 static int debug;
-static int disable_tpa;
-static int nomcp;
 static int load_count[3]; /* 0-common, 1-port0, 2-port1 */
 static int use_multi;
 
+module_param(disable_tpa, int, 0);
 module_param(use_inta, int, 0);
 module_param(poll, int, 0);
 module_param(debug, int, 0);
-module_param(disable_tpa, int, 0);
-module_param(nomcp, int, 0);
+MODULE_PARM_DESC(disable_tpa, "disable the TPA (LRO) feature");
 MODULE_PARM_DESC(use_inta, "use INT#A instead of MSI-X");
 MODULE_PARM_DESC(poll, "use polling (for debug)");
 MODULE_PARM_DESC(debug, "default debug msglevel");
-MODULE_PARM_DESC(nomcp, "ignore management CPU");
 
 #ifdef BNX2X_MULTI
 module_param(use_multi, int, 0);
@@ -503,6 +501,9 @@ static void bnx2x_panic_dump(struct bnx2x *bp)
        int i;
        u16 j, start, end;
 
+       bp->stats_state = STATS_STATE_DISABLED;
+       DP(BNX2X_MSG_STATS, "stats_state - DISABLED\n");
+
        BNX2X_ERR("begin crash dump -----------------\n");
 
        for_each_queue(bp, i) {
@@ -513,17 +514,20 @@ static void bnx2x_panic_dump(struct bnx2x *bp)
                          "  tx_bd_prod(%x)  tx_bd_cons(%x)  *tx_cons_sb(%x)\n",
                          i, fp->tx_pkt_prod, fp->tx_pkt_cons, fp->tx_bd_prod,
                          fp->tx_bd_cons, le16_to_cpu(*fp->tx_cons_sb));
-               BNX2X_ERR("          rx_comp_prod(%x)  rx_comp_cons(%x)"
-                         "  *rx_cons_sb(%x)  *rx_bd_cons_sb(%x)"
-                         "  rx_sge_prod(%x)  last_max_sge(%x)\n",
-                         fp->rx_comp_prod, fp->rx_comp_cons,
-                         le16_to_cpu(*fp->rx_cons_sb),
-                         le16_to_cpu(*fp->rx_bd_cons_sb),
-                         fp->rx_sge_prod, fp->last_max_sge);
-               BNX2X_ERR("          fp_c_idx(%x)  fp_u_idx(%x)"
-                         "  bd data(%x,%x)  rx_alloc_failed(%lx)\n",
-                         fp->fp_c_idx, fp->fp_u_idx, hw_prods->packets_prod,
-                         hw_prods->bds_prod, fp->rx_alloc_failed);
+               BNX2X_ERR("          rx_bd_prod(%x)  rx_bd_cons(%x)"
+                         "  *rx_bd_cons_sb(%x)  rx_comp_prod(%x)"
+                         "  rx_comp_cons(%x)  *rx_cons_sb(%x)\n",
+                         fp->rx_bd_prod, fp->rx_bd_cons,
+                         le16_to_cpu(*fp->rx_bd_cons_sb), fp->rx_comp_prod,
+                         fp->rx_comp_cons, le16_to_cpu(*fp->rx_cons_sb));
+               BNX2X_ERR("          rx_sge_prod(%x)  last_max_sge(%x)"
+                         "  fp_c_idx(%x)  *sb_c_idx(%x)  fp_u_idx(%x)"
+                         "  *sb_u_idx(%x)  bd data(%x,%x)\n",
+                         fp->rx_sge_prod, fp->last_max_sge, fp->fp_c_idx,
+                         fp->status_blk->c_status_block.status_block_index,
+                         fp->fp_u_idx,
+                         fp->status_blk->u_status_block.status_block_index,
+                         hw_prods->packets_prod, hw_prods->bds_prod);
 
                start = TX_BD(le16_to_cpu(*fp->tx_cons_sb) - 10);
                end = TX_BD(le16_to_cpu(*fp->tx_cons_sb) + 245);
@@ -582,9 +586,6 @@ static void bnx2x_panic_dump(struct bnx2x *bp)
        bnx2x_fw_dump(bp);
        bnx2x_mc_assert(bp);
        BNX2X_ERR("end crash dump -----------------\n");
-
-       bp->stats_state = STATS_STATE_DISABLED;
-       DP(BNX2X_MSG_STATS, "stats_state - DISABLED\n");
 }
 
 static void bnx2x_int_enable(struct bnx2x *bp)
@@ -716,21 +717,6 @@ static inline u16 bnx2x_update_fpsb_idx(struct bnx2x_fastpath *fp)
        return rc;
 }
 
-static inline int bnx2x_has_work(struct bnx2x_fastpath *fp)
-{
-       u16 rx_cons_sb = le16_to_cpu(*fp->rx_cons_sb);
-
-       if ((rx_cons_sb & MAX_RCQ_DESC_CNT) == MAX_RCQ_DESC_CNT)
-               rx_cons_sb++;
-
-       if ((fp->rx_comp_cons != rx_cons_sb) ||
-           (fp->tx_pkt_prod != le16_to_cpu(*fp->tx_cons_sb)) ||
-           (fp->tx_pkt_prod != fp->tx_pkt_cons))
-               return 1;
-
-       return 0;
-}
-
 static u16 bnx2x_ack_int(struct bnx2x *bp)
 {
        u32 igu_addr = (IGU_ADDR_SIMD_MASK + IGU_FUNC_BASE * BP_FUNC(bp)) * 8;
@@ -814,7 +800,7 @@ static u16 bnx2x_free_tx_pkt(struct bnx2x *bp, struct bnx2x_fastpath *fp,
        }
 
        /* release skb */
-       BUG_TRAP(skb);
+       WARN_ON(!skb);
        dev_kfree_skb(skb);
        tx_buf->first_bd = 0;
        tx_buf->skb = NULL;
@@ -837,9 +823,9 @@ static inline u16 bnx2x_tx_avail(struct bnx2x_fastpath *fp)
        used = SUB_S16(prod, cons) + (s16)NUM_TX_RINGS;
 
 #ifdef BNX2X_STOP_ON_ERROR
-       BUG_TRAP(used >= 0);
-       BUG_TRAP(used <= fp->bp->tx_ring_size);
-       BUG_TRAP((fp->bp->tx_ring_size - used) <= MAX_TX_AVAIL);
+       WARN_ON(used < 0);
+       WARN_ON(used > fp->bp->tx_ring_size);
+       WARN_ON((fp->bp->tx_ring_size - used) > MAX_TX_AVAIL);
 #endif
 
        return (s16)(fp->bp->tx_ring_size) - used;
@@ -898,6 +884,7 @@ static void bnx2x_tx_int(struct bnx2x_fastpath *fp, int work)
                netif_tx_lock(bp->dev);
 
                if (netif_queue_stopped(bp->dev) &&
+                   (bp->state == BNX2X_STATE_OPEN) &&
                    (bnx2x_tx_avail(fp) >= MAX_SKB_FRAGS + 3))
                        netif_wake_queue(bp->dev);
 
@@ -1020,7 +1007,7 @@ static inline int bnx2x_alloc_rx_sge(struct bnx2x *bp,
 
        mapping = pci_map_page(bp->pdev, page, 0, BCM_PAGE_SIZE*PAGES_PER_SGE,
                               PCI_DMA_FROMDEVICE);
-       if (unlikely(dma_mapping_error(mapping))) {
+       if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
                __free_pages(page, PAGES_PER_SGE_SHIFT);
                return -ENOMEM;
        }
@@ -1048,7 +1035,7 @@ static inline int bnx2x_alloc_rx_skb(struct bnx2x *bp,
 
        mapping = pci_map_single(bp->pdev, skb->data, bp->rx_buf_use_size,
                                 PCI_DMA_FROMDEVICE);
-       if (unlikely(dma_mapping_error(mapping))) {
+       if (unlikely(dma_mapping_error(&bp->pdev->dev, mapping))) {
                dev_kfree_skb(skb);
                return -ENOMEM;
        }
@@ -1261,7 +1248,7 @@ static int bnx2x_fill_frag_skb(struct bnx2x *bp, struct bnx2x_fastpath *fp,
                   where we are and drop the whole packet */
                err = bnx2x_alloc_rx_sge(bp, fp, sge_idx);
                if (unlikely(err)) {
-                       fp->rx_alloc_failed++;
+                       bp->eth_stats.rx_skb_alloc_failed++;
                        return err;
                }
 
@@ -1297,14 +1284,13 @@ static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
        pci_unmap_single(bp->pdev, pci_unmap_addr(rx_buf, mapping),
                         bp->rx_buf_use_size, PCI_DMA_FROMDEVICE);
 
-       /* if alloc failed drop the packet and keep the buffer in the bin */
        if (likely(new_skb)) {
+               /* fix ip xsum and give it to the stack */
+               /* (no need to map the new skb) */
 
                prefetch(skb);
                prefetch(((char *)(skb)) + 128);
 
-               /* else fix ip xsum and give it to the stack */
-               /* (no need to map the new skb) */
 #ifdef BNX2X_STOP_ON_ERROR
                if (pad + len > bp->rx_buf_size) {
                        BNX2X_ERR("skb_put is about to fail...  "
@@ -1353,9 +1339,10 @@ static void bnx2x_tpa_stop(struct bnx2x *bp, struct bnx2x_fastpath *fp,
                fp->tpa_pool[queue].skb = new_skb;
 
        } else {
+               /* else drop the packet and keep the buffer in the bin */
                DP(NETIF_MSG_RX_STATUS,
                   "Failed to allocate new skb - dropping packet!\n");
-               fp->rx_alloc_failed++;
+               bp->eth_stats.rx_skb_alloc_failed++;
        }
 
        fp->tpa_state[queue] = BNX2X_TPA_STOP;
@@ -1503,11 +1490,10 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
 
                        /* is this an error packet? */
                        if (unlikely(cqe_fp_flags & ETH_RX_ERROR_FALGS)) {
-                       /* do we sometimes forward error packets anyway? */
                                DP(NETIF_MSG_RX_ERR,
                                   "ERROR  flags %x  rx packet %u\n",
                                   cqe_fp_flags, sw_comp_cons);
-                               /* TBD make sure MC counts this as a drop */
+                               bp->eth_stats.rx_err_discard_pkt++;
                                goto reuse_rx;
                        }
 
@@ -1524,7 +1510,7 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
                                        DP(NETIF_MSG_RX_ERR,
                                           "ERROR  packet dropped "
                                           "because of alloc failure\n");
-                                       fp->rx_alloc_failed++;
+                                       bp->eth_stats.rx_skb_alloc_failed++;
                                        goto reuse_rx;
                                }
 
@@ -1550,7 +1536,7 @@ static int bnx2x_rx_int(struct bnx2x_fastpath *fp, int budget)
                                DP(NETIF_MSG_RX_ERR,
                                   "ERROR  packet dropped because "
                                   "of alloc failure\n");
-                               fp->rx_alloc_failed++;
+                               bp->eth_stats.rx_skb_alloc_failed++;
 reuse_rx:
                                bnx2x_reuse_rx_skb(fp, skb, bd_cons, bd_prod);
                                goto next_rx;
@@ -1559,10 +1545,12 @@ reuse_rx:
                        skb->protocol = eth_type_trans(skb, bp->dev);
 
                        skb->ip_summed = CHECKSUM_NONE;
-                       if (bp->rx_csum && BNX2X_RX_SUM_OK(cqe))
-                               skb->ip_summed = CHECKSUM_UNNECESSARY;
-
-                       /* TBD do we pass bad csum packets in promisc */
+                       if (bp->rx_csum) {
+                               if (likely(BNX2X_RX_CSUM_OK(cqe)))
+                                       skb->ip_summed = CHECKSUM_UNNECESSARY;
+                               else
+                                       bp->eth_stats.hw_csum_err++;
+                       }
                }
 
 #ifdef BCM_VLAN
@@ -1615,6 +1603,12 @@ static irqreturn_t bnx2x_msix_fp_int(int irq, void *fp_cookie)
        struct net_device *dev = bp->dev;
        int index = FP_IDX(fp);
 
+       /* Return here if interrupt is disabled */
+       if (unlikely(atomic_read(&bp->intr_sem) != 0)) {
+               DP(NETIF_MSG_INTR, "called but intr_sem not 0, returning\n");
+               return IRQ_HANDLED;
+       }
+
        DP(BNX2X_MSG_FP, "got an MSI-X interrupt on IDX:SB [%d:%d]\n",
           index, FP_SB_ID(fp));
        bnx2x_ack_sb(bp, FP_SB_ID(fp), USTORM_ID, 0, IGU_INT_DISABLE, 0);
@@ -1699,11 +1693,12 @@ static void bnx2x_stats_handle(struct bnx2x *bp, enum bnx2x_stats_event event);
  * General service functions
  */
 
-static int bnx2x_hw_lock(struct bnx2x *bp, u32 resource)
+static int bnx2x_acquire_hw_lock(struct bnx2x *bp, u32 resource)
 {
        u32 lock_status;
        u32 resource_bit = (1 << resource);
-       u8 port = BP_PORT(bp);
+       int func = BP_FUNC(bp);
+       u32 hw_lock_control_reg;
        int cnt;
 
        /* Validating that the resource is within range */
@@ -1714,8 +1709,15 @@ static int bnx2x_hw_lock(struct bnx2x *bp, u32 resource)
                return -EINVAL;
        }
 
+       if (func <= 5) {
+               hw_lock_control_reg = (MISC_REG_DRIVER_CONTROL_1 + func*8);
+       } else {
+               hw_lock_control_reg =
+                               (MISC_REG_DRIVER_CONTROL_7 + (func - 6)*8);
+       }
+
        /* Validating that the resource is not already taken */
-       lock_status = REG_RD(bp, MISC_REG_DRIVER_CONTROL_1 + port*8);
+       lock_status = REG_RD(bp, hw_lock_control_reg);
        if (lock_status & resource_bit) {
                DP(NETIF_MSG_HW, "lock_status 0x%x  resource_bit 0x%x\n",
                   lock_status, resource_bit);
@@ -1725,9 +1727,8 @@ static int bnx2x_hw_lock(struct bnx2x *bp, u32 resource)
        /* Try for 1 second every 5ms */
        for (cnt = 0; cnt < 200; cnt++) {
                /* Try to acquire the lock */
-               REG_WR(bp, MISC_REG_DRIVER_CONTROL_1 + port*8 + 4,
-                      resource_bit);
-               lock_status = REG_RD(bp, MISC_REG_DRIVER_CONTROL_1 + port*8);
+               REG_WR(bp, hw_lock_control_reg + 4, resource_bit);
+               lock_status = REG_RD(bp, hw_lock_control_reg);
                if (lock_status & resource_bit)
                        return 0;
 
@@ -1737,11 +1738,12 @@ static int bnx2x_hw_lock(struct bnx2x *bp, u32 resource)
        return -EAGAIN;
 }
 
-static int bnx2x_hw_unlock(struct bnx2x *bp, u32 resource)
+static int bnx2x_release_hw_lock(struct bnx2x *bp, u32 resource)
 {
        u32 lock_status;
        u32 resource_bit = (1 << resource);
-       u8 port = BP_PORT(bp);
+       int func = BP_FUNC(bp);
+       u32 hw_lock_control_reg;
 
        /* Validating that the resource is within range */
        if (resource > HW_LOCK_MAX_RESOURCE_VALUE) {
@@ -1751,20 +1753,27 @@ static int bnx2x_hw_unlock(struct bnx2x *bp, u32 resource)
                return -EINVAL;
        }
 
+       if (func <= 5) {
+               hw_lock_control_reg = (MISC_REG_DRIVER_CONTROL_1 + func*8);
+       } else {
+               hw_lock_control_reg =
+                               (MISC_REG_DRIVER_CONTROL_7 + (func - 6)*8);
+       }
+
        /* Validating that the resource is currently taken */
-       lock_status = REG_RD(bp, MISC_REG_DRIVER_CONTROL_1 + port*8);
+       lock_status = REG_RD(bp, hw_lock_control_reg);
        if (!(lock_status & resource_bit)) {
                DP(NETIF_MSG_HW, "lock_status 0x%x  resource_bit 0x%x\n",
                   lock_status, resource_bit);
                return -EFAULT;
        }
 
-       REG_WR(bp, MISC_REG_DRIVER_CONTROL_1 + port*8, resource_bit);
+       REG_WR(bp, hw_lock_control_reg, resource_bit);
        return 0;
 }
 
 /* HW Lock for shared dual port PHYs */
-static void bnx2x_phy_hw_lock(struct bnx2x *bp)
+static void bnx2x_acquire_phy_lock(struct bnx2x *bp)
 {
        u32 ext_phy_type = XGXS_EXT_PHY_TYPE(bp->link_params.ext_phy_config);
 
@@ -1772,16 +1781,16 @@ static void bnx2x_phy_hw_lock(struct bnx2x *bp)
 
        if ((ext_phy_type == PORT_HW_CFG_XGXS_EXT_PHY_TYPE_BCM8072) ||
            (ext_phy_type == PORT_HW_CFG_XGXS_EXT_PHY_TYPE_BCM8073))
-               bnx2x_hw_lock(bp, HW_LOCK_RESOURCE_8072_MDIO);
+               bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_8072_MDIO);
 }
 
-static void bnx2x_phy_hw_unlock(struct bnx2x *bp)
+static void bnx2x_release_phy_lock(struct bnx2x *bp)
 {
        u32 ext_phy_type = XGXS_EXT_PHY_TYPE(bp->link_params.ext_phy_config);
 
        if ((ext_phy_type == PORT_HW_CFG_XGXS_EXT_PHY_TYPE_BCM8072) ||
            (ext_phy_type == PORT_HW_CFG_XGXS_EXT_PHY_TYPE_BCM8073))
-               bnx2x_hw_unlock(bp, HW_LOCK_RESOURCE_8072_MDIO);
+               bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_8072_MDIO);
 
        mutex_unlock(&bp->port.phy_mutex);
 }
@@ -1801,7 +1810,7 @@ int bnx2x_set_gpio(struct bnx2x *bp, int gpio_num, u32 mode)
                return -EINVAL;
        }
 
-       bnx2x_hw_lock(bp, HW_LOCK_RESOURCE_GPIO);
+       bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_GPIO);
        /* read GPIO and mask except the float bits */
        gpio_reg = (REG_RD(bp, MISC_REG_GPIO) & MISC_REGISTERS_GPIO_FLOAT);
 
@@ -1834,7 +1843,7 @@ int bnx2x_set_gpio(struct bnx2x *bp, int gpio_num, u32 mode)
        }
 
        REG_WR(bp, MISC_REG_GPIO, gpio_reg);
-       bnx2x_hw_unlock(bp, HW_LOCK_RESOURCE_GPIO);
+       bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_GPIO);
 
        return 0;
 }
@@ -1850,7 +1859,7 @@ static int bnx2x_set_spio(struct bnx2x *bp, int spio_num, u32 mode)
                return -EINVAL;
        }
 
-       bnx2x_hw_lock(bp, HW_LOCK_RESOURCE_SPIO);
+       bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_SPIO);
        /* read SPIO and mask except the float bits */
        spio_reg = (REG_RD(bp, MISC_REG_SPIO) & MISC_REGISTERS_SPIO_FLOAT);
 
@@ -1880,7 +1889,7 @@ static int bnx2x_set_spio(struct bnx2x *bp, int spio_num, u32 mode)
        }
 
        REG_WR(bp, MISC_REG_SPIO, spio_reg);
-       bnx2x_hw_unlock(bp, HW_LOCK_RESOURCE_SPIO);
+       bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_SPIO);
 
        return 0;
 }
@@ -1940,46 +1949,56 @@ static void bnx2x_link_report(struct bnx2x *bp)
 
 static u8 bnx2x_initial_phy_init(struct bnx2x *bp)
 {
-       u8 rc;
+       if (!BP_NOMCP(bp)) {
+               u8 rc;
 
-       /* Initialize link parameters structure variables */
-       bp->link_params.mtu = bp->dev->mtu;
+               /* Initialize link parameters structure variables */
+               bp->link_params.mtu = bp->dev->mtu;
 
-       bnx2x_phy_hw_lock(bp);
-       rc = bnx2x_phy_init(&bp->link_params, &bp->link_vars);
-       bnx2x_phy_hw_unlock(bp);
+               bnx2x_acquire_phy_lock(bp);
+               rc = bnx2x_phy_init(&bp->link_params, &bp->link_vars);
+               bnx2x_release_phy_lock(bp);
 
-       if (bp->link_vars.link_up)
-               bnx2x_link_report(bp);
+               if (bp->link_vars.link_up)
+                       bnx2x_link_report(bp);
 
-       bnx2x_calc_fc_adv(bp);
+               bnx2x_calc_fc_adv(bp);
 
-       return rc;
+               return rc;
+       }
+       BNX2X_ERR("Bootcode is missing -not initializing link\n");
+       return -EINVAL;
 }
 
 static void bnx2x_link_set(struct bnx2x *bp)
 {
-       bnx2x_phy_hw_lock(bp);
-       bnx2x_phy_init(&bp->link_params, &bp->link_vars);
-       bnx2x_phy_hw_unlock(bp);
+       if (!BP_NOMCP(bp)) {
+               bnx2x_acquire_phy_lock(bp);
+               bnx2x_phy_init(&bp->link_params, &bp->link_vars);
+               bnx2x_release_phy_lock(bp);
 
-       bnx2x_calc_fc_adv(bp);
+               bnx2x_calc_fc_adv(bp);
+       } else
+               BNX2X_ERR("Bootcode is missing -not setting link\n");
 }
 
 static void bnx2x__link_reset(struct bnx2x *bp)
 {
-       bnx2x_phy_hw_lock(bp);
-       bnx2x_link_reset(&bp->link_params, &bp->link_vars);
-       bnx2x_phy_hw_unlock(bp);
+       if (!BP_NOMCP(bp)) {
+               bnx2x_acquire_phy_lock(bp);
+               bnx2x_link_reset(&bp->link_params, &bp->link_vars);
+               bnx2x_release_phy_lock(bp);
+       } else
+               BNX2X_ERR("Bootcode is missing -not resetting link\n");
 }
 
 static u8 bnx2x_link_test(struct bnx2x *bp)
 {
        u8 rc;
 
-       bnx2x_phy_hw_lock(bp);
+       bnx2x_acquire_phy_lock(bp);
        rc = bnx2x_test_link(&bp->link_params, &bp->link_vars);
-       bnx2x_phy_hw_unlock(bp);
+       bnx2x_release_phy_lock(bp);
 
        return rc;
 }
@@ -2203,9 +2222,9 @@ static void bnx2x_link_attn(struct bnx2x *bp)
        /* Make sure that we are synced with the current statistics */
        bnx2x_stats_handle(bp, STATS_EVENT_STOP);
 
-       bnx2x_phy_hw_lock(bp);
+       bnx2x_acquire_phy_lock(bp);
        bnx2x_link_update(&bp->link_params, &bp->link_vars);
-       bnx2x_phy_hw_unlock(bp);
+       bnx2x_release_phy_lock(bp);
 
        if (bp->link_vars.link_up) {
 
@@ -2357,7 +2376,7 @@ static int bnx2x_sp_post(struct bnx2x *bp, int command, int cid,
 }
 
 /* acquire split MCP access lock register */
-static int bnx2x_lock_alr(struct bnx2x *bp)
+static int bnx2x_acquire_alr(struct bnx2x *bp)
 {
        u32 i, j, val;
        int rc = 0;
@@ -2374,15 +2393,15 @@ static int bnx2x_lock_alr(struct bnx2x *bp)
                msleep(5);
        }
        if (!(val & (1L << 31))) {
-               BNX2X_ERR("Cannot acquire nvram interface\n");
+               BNX2X_ERR("Cannot acquire MCP access lock register\n");
                rc = -EBUSY;
        }
 
        return rc;
 }
 
-/* Release split MCP access lock register */
-static void bnx2x_unlock_alr(struct bnx2x *bp)
+/* release split MCP access lock register */
+static void bnx2x_release_alr(struct bnx2x *bp)
 {
        u32 val = 0;
 
@@ -2395,7 +2414,6 @@ static inline u16 bnx2x_update_dsb_idx(struct bnx2x *bp)
        u16 rc = 0;
 
        barrier(); /* status block is written to by the chip */
-
        if (bp->def_att_idx != def_sb->atten_status_block.attn_bits_index) {
                bp->def_att_idx = def_sb->atten_status_block.attn_bits_index;
                rc |= 1;
@@ -2432,20 +2450,25 @@ static void bnx2x_attn_int_asserted(struct bnx2x *bp, u32 asserted)
                              MISC_REG_AEU_MASK_ATTN_FUNC_0;
        u32 nig_int_mask_addr = port ? NIG_REG_MASK_INTERRUPT_PORT1 :
                                       NIG_REG_MASK_INTERRUPT_PORT0;
+       u32 aeu_mask;
 
-       if (~bp->aeu_mask & (asserted & 0xff))
-               BNX2X_ERR("IGU ERROR\n");
        if (bp->attn_state & asserted)
                BNX2X_ERR("IGU ERROR\n");
 
+       bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_PORT0_ATT_MASK + port);
+       aeu_mask = REG_RD(bp, aeu_addr);
+
        DP(NETIF_MSG_HW, "aeu_mask %x  newly asserted %x\n",
-          bp->aeu_mask, asserted);
-       bp->aeu_mask &= ~(asserted & 0xff);
-       DP(NETIF_MSG_HW, "after masking: aeu_mask %x\n", bp->aeu_mask);
+          aeu_mask, asserted);
+       aeu_mask &= ~(asserted & 0xff);
+       DP(NETIF_MSG_HW, "new mask %x\n", aeu_mask);
 
-       REG_WR(bp, aeu_addr, bp->aeu_mask);
+       REG_WR(bp, aeu_addr, aeu_mask);
+       bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_PORT0_ATT_MASK + port);
 
+       DP(NETIF_MSG_HW, "attn_state %x\n", bp->attn_state);
        bp->attn_state |= asserted;
+       DP(NETIF_MSG_HW, "new state %x\n", bp->attn_state);
 
        if (asserted & ATTN_HARD_WIRED_MASK) {
                if (asserted & ATTN_NIG_FOR_FUNC) {
@@ -2699,10 +2722,11 @@ static void bnx2x_attn_int_deasserted(struct bnx2x *bp, u32 deasserted)
        int index;
        u32 reg_addr;
        u32 val;
+       u32 aeu_mask;
 
        /* need to take HW lock because MCP or other port might also
           try to handle this event */
-       bnx2x_lock_alr(bp);
+       bnx2x_acquire_alr(bp);
 
        attn.sig[0] = REG_RD(bp, MISC_REG_AEU_AFTER_INVERT_1_FUNC_0 + port*4);
        attn.sig[1] = REG_RD(bp, MISC_REG_AEU_AFTER_INVERT_2_FUNC_0 + port*4);
@@ -2738,28 +2762,31 @@ static void bnx2x_attn_int_deasserted(struct bnx2x *bp, u32 deasserted)
                }
        }
 
-       bnx2x_unlock_alr(bp);
+       bnx2x_release_alr(bp);
 
        reg_addr = (IGU_ADDR_ATTN_BITS_CLR + IGU_FUNC_BASE * BP_FUNC(bp)) * 8;
 
        val = ~deasserted;
-/*     DP(NETIF_MSG_INTR, "write 0x%08x to IGU addr 0x%x\n",
-          val, BAR_IGU_INTMEM + reg_addr); */
+       DP(NETIF_MSG_HW, "about to mask 0x%08x at HC addr 0x%x\n",
+          val, reg_addr);
        REG_WR(bp, BAR_IGU_INTMEM + reg_addr, val);
 
-       if (bp->aeu_mask & (deasserted & 0xff))
-               BNX2X_ERR("IGU BUG!\n");
        if (~bp->attn_state & deasserted)
-               BNX2X_ERR("IGU BUG!\n");
+               BNX2X_ERR("IGU ERROR\n");
 
        reg_addr = port ? MISC_REG_AEU_MASK_ATTN_FUNC_1 :
                          MISC_REG_AEU_MASK_ATTN_FUNC_0;
 
-       DP(NETIF_MSG_HW, "aeu_mask %x\n", bp->aeu_mask);
-       bp->aeu_mask |= (deasserted & 0xff);
+       bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_PORT0_ATT_MASK + port);
+       aeu_mask = REG_RD(bp, reg_addr);
+
+       DP(NETIF_MSG_HW, "aeu_mask %x  newly deasserted %x\n",
+          aeu_mask, deasserted);
+       aeu_mask |= (deasserted & 0xff);
+       DP(NETIF_MSG_HW, "new mask %x\n", aeu_mask);
 
-       DP(NETIF_MSG_HW, "new mask %x\n", bp->aeu_mask);
-       REG_WR(bp, reg_addr, bp->aeu_mask);
+       REG_WR(bp, reg_addr, aeu_mask);
+       bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_PORT0_ATT_MASK + port);
 
        DP(NETIF_MSG_HW, "attn_state %x\n", bp->attn_state);
        bp->attn_state &= ~deasserted;
@@ -2963,37 +2990,6 @@ static inline long bnx2x_hilo(u32 *hiref)
  * Init service functions
  */
 
-static void bnx2x_storm_stats_init(struct bnx2x *bp)
-{
-       int func = BP_FUNC(bp);
-
-       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_STATS_FLAGS_OFFSET(func), 1);
-       REG_WR(bp, BAR_XSTRORM_INTMEM +
-              XSTORM_STATS_FLAGS_OFFSET(func) + 4, 0);
-
-       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_STATS_FLAGS_OFFSET(func), 1);
-       REG_WR(bp, BAR_TSTRORM_INTMEM +
-              TSTORM_STATS_FLAGS_OFFSET(func) + 4, 0);
-
-       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_STATS_FLAGS_OFFSET(func), 0);
-       REG_WR(bp, BAR_CSTRORM_INTMEM +
-              CSTORM_STATS_FLAGS_OFFSET(func) + 4, 0);
-
-       REG_WR(bp, BAR_XSTRORM_INTMEM +
-              XSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func),
-              U64_LO(bnx2x_sp_mapping(bp, fw_stats)));
-       REG_WR(bp, BAR_XSTRORM_INTMEM +
-              XSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func) + 4,
-              U64_HI(bnx2x_sp_mapping(bp, fw_stats)));
-
-       REG_WR(bp, BAR_TSTRORM_INTMEM +
-              TSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func),
-              U64_LO(bnx2x_sp_mapping(bp, fw_stats)));
-       REG_WR(bp, BAR_TSTRORM_INTMEM +
-              TSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func) + 4,
-              U64_HI(bnx2x_sp_mapping(bp, fw_stats)));
-}
-
 static void bnx2x_storm_stats_post(struct bnx2x *bp)
 {
        if (!bp->stats_pending) {
@@ -3032,6 +3028,8 @@ static void bnx2x_stats_init(struct bnx2x *bp)
        memset(&(bp->port.old_nig_stats), 0, sizeof(struct nig_stats));
        bp->port.old_nig_stats.brb_discard =
                        REG_RD(bp, NIG_REG_STAT0_BRB_DISCARD + port*0x38);
+       bp->port.old_nig_stats.brb_truncate =
+                       REG_RD(bp, NIG_REG_STAT0_BRB_TRUNCATE + port*0x38);
        REG_RD_DMAE(bp, NIG_REG_STAT0_EGRESS_MAC_PKT0 + port*0x50,
                    &(bp->port.old_nig_stats.egress_mac_pkt0_lo), 2);
        REG_RD_DMAE(bp, NIG_REG_STAT0_EGRESS_MAC_PKT1 + port*0x50,
@@ -3451,8 +3449,7 @@ static void bnx2x_bmac_stats_update(struct bnx2x *bp)
        UPDATE_STAT64(rx_stat_grovr, rx_stat_dot3statsframestoolong);
        UPDATE_STAT64(rx_stat_grfrg, rx_stat_etherstatsfragments);
        UPDATE_STAT64(rx_stat_grjbr, rx_stat_etherstatsjabbers);
-       UPDATE_STAT64(rx_stat_grxpf, rx_stat_bmac_xpf);
-       UPDATE_STAT64(rx_stat_grxcf, rx_stat_bmac_xcf);
+       UPDATE_STAT64(rx_stat_grxcf, rx_stat_maccontrolframesreceived);
        UPDATE_STAT64(rx_stat_grxpf, rx_stat_xoffstateentered);
        UPDATE_STAT64(rx_stat_grxpf, rx_stat_xoffpauseframesreceived);
        UPDATE_STAT64(tx_stat_gtxpf, tx_stat_outxoffsent);
@@ -3536,6 +3533,8 @@ static int bnx2x_hw_stats_update(struct bnx2x *bp)
 
        ADD_EXTEND_64(pstats->brb_drop_hi, pstats->brb_drop_lo,
                      new->brb_discard - old->brb_discard);
+       ADD_EXTEND_64(estats->brb_truncate_hi, estats->brb_truncate_lo,
+                     new->brb_truncate - old->brb_truncate);
 
        UPDATE_STAT64_NIG(egress_mac_pkt0,
                                        etherstatspkts1024octetsto1522octets);
@@ -3713,8 +3712,7 @@ static void bnx2x_net_stats_update(struct bnx2x *bp)
        nstats->rx_length_errors =
                                estats->rx_stat_etherstatsundersizepkts_lo +
                                estats->jabber_packets_received;
-       nstats->rx_over_errors = estats->brb_drop_lo +
-                                estats->brb_truncate_discard;
+       nstats->rx_over_errors = estats->brb_drop_lo + estats->brb_truncate_lo;
        nstats->rx_crc_errors = estats->rx_stat_dot3statsfcserrors_lo;
        nstats->rx_frame_errors = estats->rx_stat_dot3statsalignmenterrors_lo;
        nstats->rx_fifo_errors = old_tclient->no_buff_discard;
@@ -4094,9 +4092,6 @@ static void bnx2x_init_def_sb(struct bnx2x *bp,
                                               reg_offset + 0xc + 0x10*index);
        }
 
-       bp->aeu_mask = REG_RD(bp, (port ? MISC_REG_AEU_MASK_ATTN_FUNC_1 :
-                                         MISC_REG_AEU_MASK_ATTN_FUNC_0));
-
        reg_offset = (port ? HC_REG_ATTN_MSG1_ADDR_L :
                             HC_REG_ATTN_MSG0_ADDR_L);
 
@@ -4194,6 +4189,7 @@ static void bnx2x_init_def_sb(struct bnx2x *bp,
                         XSTORM_DEF_SB_HC_DISABLE_OFFSET(func, index), 1);
 
        bp->stats_pending = 0;
+       bp->set_mac_pending = 0;
 
        bnx2x_ack_sb(bp, sb_id, CSTORM_ID, 0, IGU_INT_ENABLE, 0);
 }
@@ -4363,18 +4359,18 @@ static void bnx2x_init_rx_rings(struct bnx2x *bp)
                fp->rx_sge_prod = ring_prod;
 
                /* Allocate BDs and initialize BD ring */
-               fp->rx_comp_cons = fp->rx_alloc_failed = 0;
+               fp->rx_comp_cons = 0;
                cqe_ring_prod = ring_prod = 0;
                for (i = 0; i < bp->rx_ring_size; i++) {
                        if (bnx2x_alloc_rx_skb(bp, fp, ring_prod) < 0) {
                                BNX2X_ERR("was only able to allocate "
                                          "%d rx skbs\n", i);
-                               fp->rx_alloc_failed++;
+                               bp->eth_stats.rx_skb_alloc_failed++;
                                break;
                        }
                        ring_prod = NEXT_RX_IDX(ring_prod);
                        cqe_ring_prod = NEXT_RCQ_IDX(cqe_ring_prod);
-                       BUG_TRAP(ring_prod > i);
+                       WARN_ON(ring_prod <= i);
                }
 
                fp->rx_bd_prod = ring_prod;
@@ -4535,7 +4531,7 @@ static void bnx2x_set_client_config(struct bnx2x *bp)
        int i;
 
        tstorm_client.mtu = bp->dev->mtu + ETH_OVREHEAD;
-       tstorm_client.statistics_counter_id = 0;
+       tstorm_client.statistics_counter_id = BP_CL_ID(bp);
        tstorm_client.config_flags =
                                TSTORM_ETH_CLIENT_CONFIG_STATSITICS_ENABLE;
 #ifdef BCM_VLAN
@@ -4617,13 +4613,35 @@ static void bnx2x_set_storm_rx_mode(struct bnx2x *bp)
                bnx2x_set_client_config(bp);
 }
 
-static void bnx2x_init_internal(struct bnx2x *bp)
+static void bnx2x_init_internal_common(struct bnx2x *bp)
+{
+       int i;
+
+       /* Zero this manually as its initialization is
+          currently missing in the initTool */
+       for (i = 0; i < (USTORM_AGG_DATA_SIZE >> 2); i++)
+               REG_WR(bp, BAR_USTRORM_INTMEM +
+                      USTORM_AGG_DATA_OFFSET + i * 4, 0);
+}
+
+static void bnx2x_init_internal_port(struct bnx2x *bp)
+{
+       int port = BP_PORT(bp);
+
+       REG_WR(bp, BAR_USTRORM_INTMEM + USTORM_HC_BTR_OFFSET(port), BNX2X_BTR);
+       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_HC_BTR_OFFSET(port), BNX2X_BTR);
+       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_HC_BTR_OFFSET(port), BNX2X_BTR);
+       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_HC_BTR_OFFSET(port), BNX2X_BTR);
+}
+
+static void bnx2x_init_internal_func(struct bnx2x *bp)
 {
        struct tstorm_eth_function_common_config tstorm_config = {0};
        struct stats_indication_flags stats_flags = {0};
        int port = BP_PORT(bp);
        int func = BP_FUNC(bp);
        int i;
+       u16 max_agg_size;
 
        if (is_multi(bp)) {
                tstorm_config.config_flags = MULTI_FLAGS;
@@ -4636,31 +4654,53 @@ static void bnx2x_init_internal(struct bnx2x *bp)
               TSTORM_FUNCTION_COMMON_CONFIG_OFFSET(func),
               (*(u32 *)&tstorm_config));
 
-/*     DP(NETIF_MSG_IFUP, "tstorm_config: 0x%08x\n",
-          (*(u32 *)&tstorm_config)); */
-
        bp->rx_mode = BNX2X_RX_MODE_NONE; /* no rx until link is up */
        bnx2x_set_storm_rx_mode(bp);
 
+       /* reset xstorm per client statistics */
+       for (i = 0; i < sizeof(struct xstorm_per_client_stats) / 4; i++) {
+               REG_WR(bp, BAR_XSTRORM_INTMEM +
+                      XSTORM_PER_COUNTER_ID_STATS_OFFSET(port, BP_CL_ID(bp)) +
+                      i*4, 0);
+       }
+       /* reset tstorm per client statistics */
+       for (i = 0; i < sizeof(struct tstorm_per_client_stats) / 4; i++) {
+               REG_WR(bp, BAR_TSTRORM_INTMEM +
+                      TSTORM_PER_COUNTER_ID_STATS_OFFSET(port, BP_CL_ID(bp)) +
+                      i*4, 0);
+       }
+
+       /* Init statistics related context */
        stats_flags.collect_eth = 1;
 
-       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_STATS_FLAGS_OFFSET(port),
+       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_STATS_FLAGS_OFFSET(func),
               ((u32 *)&stats_flags)[0]);
-       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_STATS_FLAGS_OFFSET(port) + 4,
+       REG_WR(bp, BAR_XSTRORM_INTMEM + XSTORM_STATS_FLAGS_OFFSET(func) + 4,
               ((u32 *)&stats_flags)[1]);
 
-       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_STATS_FLAGS_OFFSET(port),
+       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_STATS_FLAGS_OFFSET(func),
               ((u32 *)&stats_flags)[0]);
-       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_STATS_FLAGS_OFFSET(port) + 4,
+       REG_WR(bp, BAR_TSTRORM_INTMEM + TSTORM_STATS_FLAGS_OFFSET(func) + 4,
               ((u32 *)&stats_flags)[1]);
 
-       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_STATS_FLAGS_OFFSET(port),
+       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_STATS_FLAGS_OFFSET(func),
               ((u32 *)&stats_flags)[0]);
-       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_STATS_FLAGS_OFFSET(port) + 4,
+       REG_WR(bp, BAR_CSTRORM_INTMEM + CSTORM_STATS_FLAGS_OFFSET(func) + 4,
               ((u32 *)&stats_flags)[1]);
 
-/*     DP(NETIF_MSG_IFUP, "stats_flags: 0x%08x 0x%08x\n",
-          ((u32 *)&stats_flags)[0], ((u32 *)&stats_flags)[1]); */
+       REG_WR(bp, BAR_XSTRORM_INTMEM +
+              XSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func),
+              U64_LO(bnx2x_sp_mapping(bp, fw_stats)));
+       REG_WR(bp, BAR_XSTRORM_INTMEM +
+              XSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func) + 4,
+              U64_HI(bnx2x_sp_mapping(bp, fw_stats)));
+
+       REG_WR(bp, BAR_TSTRORM_INTMEM +
+              TSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func),
+              U64_LO(bnx2x_sp_mapping(bp, fw_stats)));
+       REG_WR(bp, BAR_TSTRORM_INTMEM +
+              TSTORM_ETH_STATS_QUERY_ADDR_OFFSET(func) + 4,
+              U64_HI(bnx2x_sp_mapping(bp, fw_stats)));
 
        if (CHIP_IS_E1H(bp)) {
                REG_WR8(bp, BAR_XSTRORM_INTMEM + XSTORM_FUNCTION_MODE_OFFSET,
@@ -4676,15 +4716,12 @@ static void bnx2x_init_internal(struct bnx2x *bp)
                         bp->e1hov);
        }
 
-       /* Zero this manualy as its initialization is
-          currently missing in the initTool */
-       for (i = 0; i < USTORM_AGG_DATA_SIZE >> 2; i++)
-               REG_WR(bp, BAR_USTRORM_INTMEM +
-                      USTORM_AGG_DATA_OFFSET + 4*i, 0);
-
+       /* Init CQ ring mapping and aggregation size */
+       max_agg_size = min((u32)(bp->rx_buf_use_size +
+                                8*BCM_PAGE_SIZE*PAGES_PER_SGE),
+                          (u32)0xffff);
        for_each_queue(bp, i) {
                struct bnx2x_fastpath *fp = &bp->fp[i];
-               u16 max_agg_size;
 
                REG_WR(bp, BAR_USTRORM_INTMEM +
                       USTORM_CQE_PAGE_BASE_OFFSET(port, FP_CL_ID(fp)),
@@ -4693,16 +4730,34 @@ static void bnx2x_init_internal(struct bnx2x *bp)
                       USTORM_CQE_PAGE_BASE_OFFSET(port, FP_CL_ID(fp)) + 4,
                       U64_HI(fp->rx_comp_mapping));
 
-               max_agg_size = min((u32)(bp->rx_buf_use_size +
-                                        8*BCM_PAGE_SIZE*PAGES_PER_SGE),
-                                  (u32)0xffff);
                REG_WR16(bp, BAR_USTRORM_INTMEM +
                         USTORM_MAX_AGG_SIZE_OFFSET(port, FP_CL_ID(fp)),
                         max_agg_size);
        }
 }
 
-static void bnx2x_nic_init(struct bnx2x *bp)
+static void bnx2x_init_internal(struct bnx2x *bp, u32 load_code)
+{
+       switch (load_code) {
+       case FW_MSG_CODE_DRV_LOAD_COMMON:
+               bnx2x_init_internal_common(bp);
+               /* no break */
+
+       case FW_MSG_CODE_DRV_LOAD_PORT:
+               bnx2x_init_internal_port(bp);
+               /* no break */
+
+       case FW_MSG_CODE_DRV_LOAD_FUNCTION:
+               bnx2x_init_internal_func(bp);
+               break;
+
+       default:
+               BNX2X_ERR("Unknown load_code (0x%x) from MCP\n", load_code);
+               break;
+       }
+}
+
+static void bnx2x_nic_init(struct bnx2x *bp, u32 load_code)
 {
        int i;
 
@@ -4728,8 +4783,7 @@ static void bnx2x_nic_init(struct bnx2x *bp)
        bnx2x_init_tx_ring(bp);
        bnx2x_init_sp_ring(bp);
        bnx2x_init_context(bp);
-       bnx2x_init_internal(bp);
-       bnx2x_storm_stats_init(bp);
+       bnx2x_init_internal(bp, load_code);
        bnx2x_init_ind_table(bp);
        bnx2x_int_enable(bp);
 }
@@ -5638,18 +5692,23 @@ static u32 bnx2x_fw_command(struct bnx2x *bp, u32 command)
        int func = BP_FUNC(bp);
        u32 seq = ++bp->fw_seq;
        u32 rc = 0;
+       u32 cnt = 1;
+       u8 delay = CHIP_REV_IS_SLOW(bp) ? 100 : 10;
 
        SHMEM_WR(bp, func_mb[func].drv_mb_header, (command | seq));
        DP(BNX2X_MSG_MCP, "wrote command (%x) to FW MB\n", (command | seq));
 
-       /* let the FW do it's magic ... */
-       msleep(100); /* TBD */
+       do {
+               /* let the FW do it's magic ... */
+               msleep(delay);
 
-       if (CHIP_REV_IS_SLOW(bp))
-               msleep(900);
+               rc = SHMEM_RD(bp, func_mb[func].fw_mb_header);
 
-       rc = SHMEM_RD(bp, func_mb[func].fw_mb_header);
-       DP(BNX2X_MSG_MCP, "read (%x) seq is (%x) from FW MB\n", rc, seq);
+               /* Give the FW up to 2 second (200*10ms) */
+       } while ((seq != (rc & FW_MSG_SEQ_NUMBER_MASK)) && (cnt++ < 200));
+
+       DP(BNX2X_MSG_MCP, "[after %d ms] read (%x) seq is (%x) from FW MB\n",
+          cnt*delay, rc, seq);
 
        /* is this a reply to our command? */
        if (seq == (rc & FW_MSG_SEQ_NUMBER_MASK)) {
@@ -6183,22 +6242,24 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
        if (!BP_NOMCP(bp)) {
                load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_REQ);
                if (!load_code) {
-                       BNX2X_ERR("MCP response failure, unloading\n");
+                       BNX2X_ERR("MCP response failure, aborting\n");
                        return -EBUSY;
                }
                if (load_code == FW_MSG_CODE_DRV_LOAD_REFUSED)
                        return -EBUSY; /* other port in diagnostic mode */
 
        } else {
+               int port = BP_PORT(bp);
+
                DP(NETIF_MSG_IFUP, "NO MCP load counts before us %d, %d, %d\n",
                   load_count[0], load_count[1], load_count[2]);
                load_count[0]++;
-               load_count[1 + BP_PORT(bp)]++;
+               load_count[1 + port]++;
                DP(NETIF_MSG_IFUP, "NO MCP new load counts       %d, %d, %d\n",
                   load_count[0], load_count[1], load_count[2]);
                if (load_count[0] == 1)
                        load_code = FW_MSG_CODE_DRV_LOAD_COMMON;
-               else if (load_count[1 + BP_PORT(bp)] == 1)
+               else if (load_count[1 + port] == 1)
                        load_code = FW_MSG_CODE_DRV_LOAD_PORT;
                else
                        load_code = FW_MSG_CODE_DRV_LOAD_FUNCTION;
@@ -6247,9 +6308,6 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                bnx2x_fp(bp, i, disable_tpa) =
                                        ((bp->flags & TPA_ENABLE_FLAG) == 0);
 
-       /* Disable interrupt handling until HW is initialized */
-       atomic_set(&bp->intr_sem, 1);
-
        if (bp->flags & USING_MSIX_FLAG) {
                rc = bnx2x_req_msix_irqs(bp);
                if (rc) {
@@ -6276,17 +6334,14 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                goto load_error;
        }
 
-       /* Enable interrupt handling */
-       atomic_set(&bp->intr_sem, 0);
-
        /* Setup NIC internals and enable interrupts */
-       bnx2x_nic_init(bp);
+       bnx2x_nic_init(bp, load_code);
 
        /* Send LOAD_DONE command to MCP */
        if (!BP_NOMCP(bp)) {
                load_code = bnx2x_fw_command(bp, DRV_MSG_CODE_LOAD_DONE);
                if (!load_code) {
-                       BNX2X_ERR("MCP response failure, unloading\n");
+                       BNX2X_ERR("MCP response failure, aborting\n");
                        rc = -EBUSY;
                        goto load_int_disable;
                }
@@ -6301,11 +6356,12 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
        for_each_queue(bp, i)
                napi_enable(&bnx2x_fp(bp, i, napi));
 
+       /* Enable interrupt handling */
+       atomic_set(&bp->intr_sem, 0);
+
        rc = bnx2x_setup_leading(bp);
        if (rc) {
-#ifdef BNX2X_STOP_ON_ERROR
-               bp->panic = 1;
-#endif
+               BNX2X_ERR("Setup leading failed!\n");
                goto load_stop_netif;
        }
 
@@ -6339,7 +6395,6 @@ static int bnx2x_nic_load(struct bnx2x *bp, int load_mode)
                break;
 
        case LOAD_OPEN:
-               /* IRQ is only requested from bnx2x_open */
                netif_start_queue(bp->dev);
                bnx2x_set_rx_mode(bp->dev);
                if (bp->flags & USING_MSIX_FLAG)
@@ -6411,7 +6466,7 @@ static int bnx2x_stop_multi(struct bnx2x *bp, int index)
        return rc;
 }
 
-static void bnx2x_stop_leading(struct bnx2x *bp)
+static int bnx2x_stop_leading(struct bnx2x *bp)
 {
        u16 dsb_sp_prod_idx;
        /* if the other port is handling traffic,
@@ -6429,7 +6484,7 @@ static void bnx2x_stop_leading(struct bnx2x *bp)
        rc = bnx2x_wait_ramrod(bp, BNX2X_FP_STATE_HALTED, 0,
                               &(bp->fp[0].state), 1);
        if (rc) /* timeout */
-               return;
+               return rc;
 
        dsb_sp_prod_idx = *bp->dsb_sp_prod;
 
@@ -6448,13 +6503,18 @@ static void bnx2x_stop_leading(struct bnx2x *bp)
                           *bp->dsb_sp_prod, dsb_sp_prod_idx);
 #ifdef BNX2X_STOP_ON_ERROR
                        bnx2x_panic();
+#else
+                       rc = -EBUSY;
 #endif
                        break;
                }
                cnt--;
+               msleep(1);
        }
        bp->state = BNX2X_STATE_CLOSING_WAIT4_UNLOAD;
        bp->fp[0].state = BNX2X_FP_STATE_CLOSED;
+
+       return rc;
 }
 
 static void bnx2x_reset_func(struct bnx2x *bp)
@@ -6539,8 +6599,9 @@ static void bnx2x_reset_chip(struct bnx2x *bp, u32 reset_code)
 /* msut be called with rtnl_lock */
 static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
 {
+       int port = BP_PORT(bp);
        u32 reset_code = 0;
-       int i, cnt;
+       int i, cnt, rc;
 
        bp->state = BNX2X_STATE_CLOSING_WAIT4_HALT;
 
@@ -6557,22 +6618,17 @@ static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
                 (DRV_PULSE_ALWAYS_ALIVE | bp->fw_drv_pulse_wr_seq));
        bnx2x_stats_handle(bp, STATS_EVENT_STOP);
 
-       /* Wait until all fast path tasks complete */
+       /* Wait until tx fast path tasks complete */
        for_each_queue(bp, i) {
                struct bnx2x_fastpath *fp = &bp->fp[i];
 
-#ifdef BNX2X_STOP_ON_ERROR
-#ifdef __powerpc64__
-               DP(NETIF_MSG_RX_STATUS, "fp->tpa_queue_used = 0x%lx\n",
-#else
-               DP(NETIF_MSG_IFDOWN, "fp->tpa_queue_used = 0x%llx\n",
-#endif
-                  fp->tpa_queue_used);
-#endif
                cnt = 1000;
                smp_rmb();
-               while (bnx2x_has_work(fp)) {
-                       msleep(1);
+               while (BNX2X_HAS_TX_WORK(fp)) {
+
+                       if (!netif_running(bp->dev))
+                               bnx2x_tx_int(fp, 1000);
+
                        if (!cnt) {
                                BNX2X_ERR("timeout waiting for queue[%d]\n",
                                          i);
@@ -6584,14 +6640,13 @@ static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
 #endif
                        }
                        cnt--;
+                       msleep(1);
                        smp_rmb();
                }
        }
 
-       /* Wait until all slow path tasks complete */
-       cnt = 1000;
-       while ((bp->spq_left != MAX_SPQ_PENDING) && cnt--)
-               msleep(1);
+       /* Give HW time to discard old tx messages */
+       msleep(1);
 
        for_each_queue(bp, i)
                napi_disable(&bnx2x_fp(bp, i, napi));
@@ -6601,52 +6656,51 @@ static int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode)
        /* Release IRQs */
        bnx2x_free_irq(bp);
 
-       if (bp->flags & NO_WOL_FLAG)
+       if (unload_mode == UNLOAD_NORMAL)
+               reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS;
+
+       else if (bp->flags & NO_WOL_FLAG) {
                reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_MCP;
+               if (CHIP_IS_E1H(bp))
+                       REG_WR(bp, MISC_REG_E1HMF_MODE, 0);
 
-       else if (bp->wol) {
-               u32 emac_base = BP_PORT(bp) ? GRCBASE_EMAC1 : GRCBASE_EMAC0;
+       else if (bp->wol) {
+               u32 emac_base = port ? GRCBASE_EMAC1 : GRCBASE_EMAC0;
                u8 *mac_addr = bp->dev->dev_addr;
                u32 val;
-
                /* The mac address is written to entries 1-4 to
                   preserve entry 0 which is used by the PMF */
+               u8 entry = (BP_E1HVN(bp) + 1)*8;
+
                val = (mac_addr[0] << 8) | mac_addr[1];
-               EMAC_WR(EMAC_REG_EMAC_MAC_MATCH + (BP_E1HVN(bp) + 1)*8, val);
+               EMAC_WR(EMAC_REG_EMAC_MAC_MATCH + entry, val);
 
                val = (mac_addr[2] << 24) | (mac_addr[3] << 16) |
                      (mac_addr[4] << 8) | mac_addr[5];
-               EMAC_WR(EMAC_REG_EMAC_MAC_MATCH + (BP_E1HVN(bp) + 1)*8 + 4,
-                       val);
+               EMAC_WR(EMAC_REG_EMAC_MAC_MATCH + entry + 4, val);
 
                reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_EN;
 
        } else
                reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS;
 
+       if (CHIP_IS_E1H(bp))
+               REG_WR(bp, NIG_REG_LLH0_FUNC_EN + port*8, 0);
+
        /* Close multi and leading connections
           Completions for ramrods are collected in a synchronous way */
        for_each_nondefault_queue(bp, i)
                if (bnx2x_stop_multi(bp, i))
                        goto unload_error;
 
-       if (CHIP_IS_E1H(bp))
-               REG_WR(bp, NIG_REG_LLH0_FUNC_EN + BP_PORT(bp)*8, 0);
-
-       bnx2x_stop_leading(bp);
-#ifdef BNX2X_STOP_ON_ERROR
-       /* If ramrod completion timed out - break here! */
-       if (bp->panic) {
+       rc = bnx2x_stop_leading(bp);
+       if (rc) {
                BNX2X_ERR("Stop leading failed!\n");
+#ifdef BNX2X_STOP_ON_ERROR
                return -EBUSY;
-       }
+#else
+               goto unload_error;
 #endif
-
-       if ((bp->state != BNX2X_STATE_CLOSING_WAIT4_UNLOAD) ||
-           (bp->fp[0].state != BNX2X_FP_STATE_CLOSED)) {
-               DP(NETIF_MSG_IFDOWN, "failed to close leading properly!  "
-                  "state 0x%x  fp[0].state 0x%x\n",
-                  bp->state, bp->fp[0].state);
        }
 
 unload_error:
@@ -6656,12 +6710,12 @@ unload_error:
                DP(NETIF_MSG_IFDOWN, "NO MCP load counts      %d, %d, %d\n",
                   load_count[0], load_count[1], load_count[2]);
                load_count[0]--;
-               load_count[1 + BP_PORT(bp)]--;
+               load_count[1 + port]--;
                DP(NETIF_MSG_IFDOWN, "NO MCP new load counts  %d, %d, %d\n",
                   load_count[0], load_count[1], load_count[2]);
                if (load_count[0] == 0)
                        reset_code = FW_MSG_CODE_DRV_UNLOAD_COMMON;
-               else if (load_count[1 + BP_PORT(bp)] == 0)
+               else if (load_count[1 + port] == 0)
                        reset_code = FW_MSG_CODE_DRV_UNLOAD_PORT;
                else
                        reset_code = FW_MSG_CODE_DRV_UNLOAD_FUNCTION;
@@ -6733,50 +6787,86 @@ static void __devinit bnx2x_undi_unload(struct bnx2x *bp)
                /* Check if it is the UNDI driver
                 * UNDI driver initializes CID offset for normal bell to 0x7
                 */
+               bnx2x_acquire_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
                val = REG_RD(bp, DORQ_REG_NORM_CID_OFST);
                if (val == 0x7) {
                        u32 reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS;
-                       /* save our func and fw_seq */
+                       /* save our func */
                        int func = BP_FUNC(bp);
-                       u16 fw_seq = bp->fw_seq;
+                       u32 swap_en;
+                       u32 swap_val;
 
                        BNX2X_DEV_INFO("UNDI is active! reset device\n");
 
                        /* try unload UNDI on port 0 */
                        bp->func = 0;
-                       bp->fw_seq = (SHMEM_RD(bp,
-                                            func_mb[bp->func].drv_mb_header) &
-                                     DRV_MSG_SEQ_NUMBER_MASK);
-
+                       bp->fw_seq =
+                              (SHMEM_RD(bp, func_mb[bp->func].drv_mb_header) &
+                               DRV_MSG_SEQ_NUMBER_MASK);
                        reset_code = bnx2x_fw_command(bp, reset_code);
-                       bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE);
 
                        /* if UNDI is loaded on the other port */
                        if (reset_code != FW_MSG_CODE_DRV_UNLOAD_COMMON) {
 
+                               /* send "DONE" for previous unload */
+                               bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE);
+
+                               /* unload UNDI on port 1 */
                                bp->func = 1;
-                               bp->fw_seq = (SHMEM_RD(bp,
-                                            func_mb[bp->func].drv_mb_header) &
-                                             DRV_MSG_SEQ_NUMBER_MASK);
-
-                               bnx2x_fw_command(bp,
-                                            DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS);
-                               bnx2x_fw_command(bp,
-                                                DRV_MSG_CODE_UNLOAD_DONE);
-
-                               /* restore our func and fw_seq */
-                               bp->func = func;
-                               bp->fw_seq = fw_seq;
+                               bp->fw_seq =
+                              (SHMEM_RD(bp, func_mb[bp->func].drv_mb_header) &
+                                       DRV_MSG_SEQ_NUMBER_MASK);
+                               reset_code = DRV_MSG_CODE_UNLOAD_REQ_WOL_DIS;
+
+                               bnx2x_fw_command(bp, reset_code);
                        }
 
+                       REG_WR(bp, (BP_PORT(bp) ? HC_REG_CONFIG_1 :
+                                   HC_REG_CONFIG_0), 0x1000);
+
+                       /* close input traffic and wait for it */
+                       /* Do not rcv packets to BRB */
+                       REG_WR(bp,
+                             (BP_PORT(bp) ? NIG_REG_LLH1_BRB1_DRV_MASK :
+                                            NIG_REG_LLH0_BRB1_DRV_MASK), 0x0);
+                       /* Do not direct rcv packets that are not for MCP to
+                        * the BRB */
+                       REG_WR(bp,
+                              (BP_PORT(bp) ? NIG_REG_LLH1_BRB1_NOT_MCP :
+                                             NIG_REG_LLH0_BRB1_NOT_MCP), 0x0);
+                       /* clear AEU */
+                       REG_WR(bp,
+                            (BP_PORT(bp) ? MISC_REG_AEU_MASK_ATTN_FUNC_1 :
+                                           MISC_REG_AEU_MASK_ATTN_FUNC_0), 0);
+                       msleep(10);
+
+                       /* save NIG port swap info */
+                       swap_val = REG_RD(bp, NIG_REG_PORT_SWAP);
+                       swap_en = REG_RD(bp, NIG_REG_STRAP_OVERRIDE);
                        /* reset device */
                        REG_WR(bp,
                               GRCBASE_MISC + MISC_REGISTERS_RESET_REG_1_CLEAR,
-                              0xd3ffff7f);
+                              0xd3ffffff);
                        REG_WR(bp,
                               GRCBASE_MISC + MISC_REGISTERS_RESET_REG_2_CLEAR,
                               0x1403);
+                       /* take the NIG out of reset and restore swap values */
+                       REG_WR(bp,
+                              GRCBASE_MISC + MISC_REGISTERS_RESET_REG_1_SET,
+                              MISC_REGISTERS_RESET_REG_1_RST_NIG);
+                       REG_WR(bp, NIG_REG_PORT_SWAP, swap_val);
+                       REG_WR(bp, NIG_REG_STRAP_OVERRIDE, swap_en);
+
+                       /* send unload done to the MCP */
+                       bnx2x_fw_command(bp, DRV_MSG_CODE_UNLOAD_DONE);
+
+                       /* restore our func and fw_seq */
+                       bp->func = func;
+                       bp->fw_seq =
+                              (SHMEM_RD(bp, func_mb[bp->func].drv_mb_header) &
+                               DRV_MSG_SEQ_NUMBER_MASK);
                }
+               bnx2x_release_hw_lock(bp, HW_LOCK_RESOURCE_UNDI);
        }
 }
 
@@ -7202,7 +7292,7 @@ static void __devinit bnx2x_link_settings_requested(struct bnx2x *bp)
        bp->link_params.req_flow_ctrl = (bp->port.link_config &
                                         PORT_FEATURE_FLOW_CONTROL_MASK);
        if ((bp->link_params.req_flow_ctrl == FLOW_CTRL_AUTO) &&
-           (!bp->port.supported & SUPPORTED_Autoneg))
+           !(bp->port.supported & SUPPORTED_Autoneg))
                bp->link_params.req_flow_ctrl = FLOW_CTRL_NONE;
 
        BNX2X_DEV_INFO("req_line_speed %d  req_duplex %d  req_flow_ctrl 0x%x"
@@ -7337,8 +7427,8 @@ static int __devinit bnx2x_init_bp(struct bnx2x *bp)
        int func = BP_FUNC(bp);
        int rc;
 
-       if (nomcp)
-               bp->flags |= NO_MCP_FLAG;
+       /* Disable interrupt handling until HW is initialized */
+       atomic_set(&bp->intr_sem, 1);
 
        mutex_init(&bp->port.phy_mutex);
 
@@ -7377,8 +7467,6 @@ static int __devinit bnx2x_init_bp(struct bnx2x *bp)
        bp->tx_ticks = 50;
        bp->rx_ticks = 25;
 
-       bp->stats_ticks = 1000000 & 0xffff00;
-
        bp->timer_interval = (CHIP_REV_IS_SLOW(bp) ? 5*HZ : HZ);
        bp->current_interval = (poll ? poll : bp->timer_interval);
 
@@ -7635,11 +7723,11 @@ static void bnx2x_get_drvinfo(struct net_device *dev,
 
        phy_fw_ver[0] = '\0';
        if (bp->port.pmf) {
-               bnx2x_phy_hw_lock(bp);
+               bnx2x_acquire_phy_lock(bp);
                bnx2x_get_ext_phy_fw_version(&bp->link_params,
                                             (bp->state != BNX2X_STATE_CLOSED),
                                             phy_fw_ver, PHY_FW_VER_LEN);
-               bnx2x_phy_hw_unlock(bp);
+               bnx2x_release_phy_lock(bp);
        }
 
        snprintf(info->fw_version, 32, "%d.%d.%d:%d BC:%x%s%s",
@@ -8097,7 +8185,7 @@ static int bnx2x_set_eeprom(struct net_device *dev,
        if (eeprom->magic == 0x00504859)
                if (bp->port.pmf) {
 
-                       bnx2x_phy_hw_lock(bp);
+                       bnx2x_acquire_phy_lock(bp);
                        rc = bnx2x_flash_download(bp, BP_PORT(bp),
                                             bp->link_params.ext_phy_config,
                                             (bp->state != BNX2X_STATE_CLOSED),
@@ -8109,7 +8197,7 @@ static int bnx2x_set_eeprom(struct net_device *dev,
                                rc |= bnx2x_phy_init(&bp->link_params,
                                                     &bp->link_vars);
                        }
-                       bnx2x_phy_hw_unlock(bp);
+                       bnx2x_release_phy_lock(bp);
 
                } else /* Only the PMF can access the PHY */
                        return -EINVAL;
@@ -8128,7 +8216,6 @@ static int bnx2x_get_coalesce(struct net_device *dev,
 
        coal->rx_coalesce_usecs = bp->rx_ticks;
        coal->tx_coalesce_usecs = bp->tx_ticks;
-       coal->stats_block_coalesce_usecs = bp->stats_ticks;
 
        return 0;
 }
@@ -8146,11 +8233,6 @@ static int bnx2x_set_coalesce(struct net_device *dev,
        if (bp->tx_ticks > 0x3000)
                bp->tx_ticks = 0x3000;
 
-       bp->stats_ticks = coal->stats_block_coalesce_usecs;
-       if (bp->stats_ticks > 0xffff00)
-               bp->stats_ticks = 0xffff00;
-       bp->stats_ticks &= 0xffff00;
-
        if (netif_running(dev))
                bnx2x_update_coalesce(bp);
 
@@ -8302,17 +8384,28 @@ static int bnx2x_set_rx_csum(struct net_device *dev, u32 data)
 
 static int bnx2x_set_tso(struct net_device *dev, u32 data)
 {
-       if (data)
+       if (data) {
                dev->features |= (NETIF_F_TSO | NETIF_F_TSO_ECN);
-       else
+               dev->features |= NETIF_F_TSO6;
+       } else {
                dev->features &= ~(NETIF_F_TSO | NETIF_F_TSO_ECN);
+               dev->features &= ~NETIF_F_TSO6;
+       }
+
        return 0;
 }
 
-static struct {
+static const struct {
        char string[ETH_GSTRING_LEN];
 } bnx2x_tests_str_arr[BNX2X_NUM_TESTS] = {
-       { "MC Errors  (online)" }
+       { "register_test (offline)" },
+       { "memory_test (offline)" },
+       { "loopback_test (offline)" },
+       { "nvram_test (online)" },
+       { "interrupt_test (online)" },
+       { "link_test (online)" },
+       { "idle check (online)" },
+       { "MC errors (online)" }
 };
 
 static int bnx2x_self_test_count(struct net_device *dev)
@@ -8320,105 +8413,595 @@ static int bnx2x_self_test_count(struct net_device *dev)
        return BNX2X_NUM_TESTS;
 }
 
-static void bnx2x_self_test(struct net_device *dev,
-                           struct ethtool_test *etest, u64 *buf)
+static int bnx2x_test_registers(struct bnx2x *bp)
+{
+       int idx, i, rc = -ENODEV;
+       u32 wr_val = 0;
+       static const struct {
+               u32  offset0;
+               u32  offset1;
+               u32  mask;
+       } reg_tbl[] = {
+/* 0 */                { BRB1_REG_PAUSE_LOW_THRESHOLD_0,      4, 0x000003ff },
+               { DORQ_REG_DB_ADDR0,                   4, 0xffffffff },
+               { HC_REG_AGG_INT_0,                    4, 0x000003ff },
+               { PBF_REG_MAC_IF0_ENABLE,              4, 0x00000001 },
+               { PBF_REG_P0_INIT_CRD,                 4, 0x000007ff },
+               { PRS_REG_CID_PORT_0,                  4, 0x00ffffff },
+               { PXP2_REG_PSWRQ_CDU0_L2P,             4, 0x000fffff },
+               { PXP2_REG_RQ_CDU0_EFIRST_MEM_ADDR,    8, 0x0003ffff },
+               { PXP2_REG_PSWRQ_TM0_L2P,              4, 0x000fffff },
+               { PXP2_REG_RQ_USDM0_EFIRST_MEM_ADDR,   8, 0x0003ffff },
+/* 10 */       { PXP2_REG_PSWRQ_TSDM0_L2P,            4, 0x000fffff },
+               { QM_REG_CONNNUM_0,                    4, 0x000fffff },
+               { TM_REG_LIN0_MAX_ACTIVE_CID,          4, 0x0003ffff },
+               { SRC_REG_KEYRSS0_0,                  40, 0xffffffff },
+               { SRC_REG_KEYRSS0_7,                  40, 0xffffffff },
+               { XCM_REG_WU_DA_SET_TMR_CNT_FLG_CMD00, 4, 0x00000001 },
+               { XCM_REG_WU_DA_CNT_CMD00,             4, 0x00000003 },
+               { XCM_REG_GLB_DEL_ACK_MAX_CNT_0,       4, 0x000000ff },
+               { NIG_REG_EGRESS_MNG0_FIFO,           20, 0xffffffff },
+               { NIG_REG_LLH0_T_BIT,                  4, 0x00000001 },
+/* 20 */       { NIG_REG_EMAC0_IN_EN,                 4, 0x00000001 },
+               { NIG_REG_BMAC0_IN_EN,                 4, 0x00000001 },
+               { NIG_REG_XCM0_OUT_EN,                 4, 0x00000001 },
+               { NIG_REG_BRB0_OUT_EN,                 4, 0x00000001 },
+               { NIG_REG_LLH0_XCM_MASK,               4, 0x00000007 },
+               { NIG_REG_LLH0_ACPI_PAT_6_LEN,        68, 0x000000ff },
+               { NIG_REG_LLH0_ACPI_PAT_0_CRC,        68, 0xffffffff },
+               { NIG_REG_LLH0_DEST_MAC_0_0,         160, 0xffffffff },
+               { NIG_REG_LLH0_DEST_IP_0_1,          160, 0xffffffff },
+               { NIG_REG_LLH0_IPV4_IPV6_0,          160, 0x00000001 },
+/* 30 */       { NIG_REG_LLH0_DEST_UDP_0,           160, 0x0000ffff },
+               { NIG_REG_LLH0_DEST_TCP_0,           160, 0x0000ffff },
+               { NIG_REG_LLH0_VLAN_ID_0,            160, 0x00000fff },
+               { NIG_REG_XGXS_SERDES0_MODE_SEL,       4, 0x00000001 },
+               { NIG_REG_LED_CONTROL_OVERRIDE_TRAFFIC_P0, 4, 0x00000001 },
+               { NIG_REG_STATUS_INTERRUPT_PORT0,      4, 0x07ffffff },
+               { NIG_REG_XGXS0_CTRL_EXTREMOTEMDIOST, 24, 0x00000001 },
+               { NIG_REG_SERDES0_CTRL_PHY_ADDR,      16, 0x0000001f },
+
+               { 0xffffffff, 0, 0x00000000 }
+       };
+
+       if (!netif_running(bp->dev))
+               return rc;
+
+       /* Repeat the test twice:
+          First by writing 0x00000000, second by writing 0xffffffff */
+       for (idx = 0; idx < 2; idx++) {
+
+               switch (idx) {
+               case 0:
+                       wr_val = 0;
+                       break;
+               case 1:
+                       wr_val = 0xffffffff;
+                       break;
+               }
+
+               for (i = 0; reg_tbl[i].offset0 != 0xffffffff; i++) {
+                       u32 offset, mask, save_val, val;
+                       int port = BP_PORT(bp);
+
+                       offset = reg_tbl[i].offset0 + port*reg_tbl[i].offset1;
+                       mask = reg_tbl[i].mask;
+
+                       save_val = REG_RD(bp, offset);
+
+                       REG_WR(bp, offset, wr_val);
+                       val = REG_RD(bp, offset);
+
+                       /* Restore the original register's value */
+                       REG_WR(bp, offset, save_val);
+
+                       /* verify that value is as expected value */
+                       if ((val & mask) != (wr_val & mask))
+                               goto test_reg_exit;
+               }
+       }
+
+       rc = 0;
+
+test_reg_exit:
+       return rc;
+}
+
+static int bnx2x_test_memory(struct bnx2x *bp)
 {
-       struct bnx2x *bp = netdev_priv(dev);
-       int stats_state;
+       int i, j, rc = -ENODEV;
+       u32 val;
+       static const struct {
+               u32 offset;
+               int size;
+       } mem_tbl[] = {
+               { CCM_REG_XX_DESCR_TABLE,   CCM_REG_XX_DESCR_TABLE_SIZE },
+               { CFC_REG_ACTIVITY_COUNTER, CFC_REG_ACTIVITY_COUNTER_SIZE },
+               { CFC_REG_LINK_LIST,        CFC_REG_LINK_LIST_SIZE },
+               { DMAE_REG_CMD_MEM,         DMAE_REG_CMD_MEM_SIZE },
+               { TCM_REG_XX_DESCR_TABLE,   TCM_REG_XX_DESCR_TABLE_SIZE },
+               { UCM_REG_XX_DESCR_TABLE,   UCM_REG_XX_DESCR_TABLE_SIZE },
+               { XCM_REG_XX_DESCR_TABLE,   XCM_REG_XX_DESCR_TABLE_SIZE },
+
+               { 0xffffffff, 0 }
+       };
+       static const struct {
+               char *name;
+               u32 offset;
+               u32 mask;
+       } prty_tbl[] = {
+               { "CCM_REG_CCM_PRTY_STS",     CCM_REG_CCM_PRTY_STS,     0 },
+               { "CFC_REG_CFC_PRTY_STS",     CFC_REG_CFC_PRTY_STS,     0 },
+               { "DMAE_REG_DMAE_PRTY_STS",   DMAE_REG_DMAE_PRTY_STS,   0 },
+               { "TCM_REG_TCM_PRTY_STS",     TCM_REG_TCM_PRTY_STS,     0 },
+               { "UCM_REG_UCM_PRTY_STS",     UCM_REG_UCM_PRTY_STS,     0 },
+               { "XCM_REG_XCM_PRTY_STS",     XCM_REG_XCM_PRTY_STS,     0x1 },
+
+               { NULL, 0xffffffff, 0 }
+       };
 
-       memset(buf, 0, sizeof(u64) * BNX2X_NUM_TESTS);
+       if (!netif_running(bp->dev))
+               return rc;
 
-       if (bp->state != BNX2X_STATE_OPEN) {
-               DP(NETIF_MSG_PROBE, "state is %x, returning\n", bp->state);
-               return;
+       /* Go through all the memories */
+       for (i = 0; mem_tbl[i].offset != 0xffffffff; i++)
+               for (j = 0; j < mem_tbl[i].size; j++)
+                       REG_RD(bp, mem_tbl[i].offset + j*4);
+
+       /* Check the parity status */
+       for (i = 0; prty_tbl[i].offset != 0xffffffff; i++) {
+               val = REG_RD(bp, prty_tbl[i].offset);
+               if (val & ~(prty_tbl[i].mask)) {
+                       DP(NETIF_MSG_HW,
+                          "%s is 0x%x\n", prty_tbl[i].name, val);
+                       goto test_mem_exit;
+               }
        }
 
-       stats_state = bp->stats_state;
+       rc = 0;
 
-       if (bnx2x_mc_assert(bp) != 0) {
-               buf[0] = 1;
-               etest->flags |= ETH_TEST_FL_FAILED;
+test_mem_exit:
+       return rc;
+}
+
+static void bnx2x_netif_start(struct bnx2x *bp)
+{
+       int i;
+
+       if (atomic_dec_and_test(&bp->intr_sem)) {
+               if (netif_running(bp->dev)) {
+                       bnx2x_int_enable(bp);
+                       for_each_queue(bp, i)
+                               napi_enable(&bnx2x_fp(bp, i, napi));
+                       if (bp->state == BNX2X_STATE_OPEN)
+                               netif_wake_queue(bp->dev);
+               }
        }
+}
 
-#ifdef BNX2X_EXTRA_DEBUG
-       bnx2x_panic_dump(bp);
-#endif
+static void bnx2x_netif_stop(struct bnx2x *bp)
+{
+       int i;
+
+       if (netif_running(bp->dev)) {
+               netif_tx_disable(bp->dev);
+               bp->dev->trans_start = jiffies; /* prevent tx timeout */
+               for_each_queue(bp, i)
+                       napi_disable(&bnx2x_fp(bp, i, napi));
+       }
+       bnx2x_int_disable_sync(bp);
 }
 
-static const struct {
-       long offset;
-       int size;
-       u32 flags;
-       char string[ETH_GSTRING_LEN];
+static void bnx2x_wait_for_link(struct bnx2x *bp, u8 link_up)
+{
+       int cnt = 1000;
+
+       if (link_up)
+               while (bnx2x_link_test(bp) && cnt--)
+                       msleep(10);
+}
+
+static int bnx2x_run_loopback(struct bnx2x *bp, int loopback_mode, u8 link_up)
+{
+       unsigned int pkt_size, num_pkts, i;
+       struct sk_buff *skb;
+       unsigned char *packet;
+       struct bnx2x_fastpath *fp = &bp->fp[0];
+       u16 tx_start_idx, tx_idx;
+       u16 rx_start_idx, rx_idx;
+       u16 pkt_prod;
+       struct sw_tx_bd *tx_buf;
+       struct eth_tx_bd *tx_bd;
+       dma_addr_t mapping;
+       union eth_rx_cqe *cqe;
+       u8 cqe_fp_flags;
+       struct sw_rx_bd *rx_buf;
+       u16 len;
+       int rc = -ENODEV;
+
+       if (loopback_mode == BNX2X_MAC_LOOPBACK) {
+               bp->link_params.loopback_mode = LOOPBACK_BMAC;
+               bnx2x_acquire_phy_lock(bp);
+               bnx2x_phy_init(&bp->link_params, &bp->link_vars);
+               bnx2x_release_phy_lock(bp);
+
+       } else if (loopback_mode == BNX2X_PHY_LOOPBACK) {
+               bp->link_params.loopback_mode = LOOPBACK_XGXS_10;
+               bnx2x_acquire_phy_lock(bp);
+               bnx2x_phy_init(&bp->link_params, &bp->link_vars);
+               bnx2x_release_phy_lock(bp);
+               /* wait until link state is restored */
+               bnx2x_wait_for_link(bp, link_up);
+
+       } else
+               return -EINVAL;
+
+       pkt_size = 1514;
+       skb = netdev_alloc_skb(bp->dev, bp->rx_buf_size);
+       if (!skb) {
+               rc = -ENOMEM;
+               goto test_loopback_exit;
+       }
+       packet = skb_put(skb, pkt_size);
+       memcpy(packet, bp->dev->dev_addr, ETH_ALEN);
+       memset(packet + ETH_ALEN, 0, (ETH_HLEN - ETH_ALEN));
+       for (i = ETH_HLEN; i < pkt_size; i++)
+               packet[i] = (unsigned char) (i & 0xff);
+
+       num_pkts = 0;
+       tx_start_idx = le16_to_cpu(*fp->tx_cons_sb);
+       rx_start_idx = le16_to_cpu(*fp->rx_cons_sb);
+
+       pkt_prod = fp->tx_pkt_prod++;
+       tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
+       tx_buf->first_bd = fp->tx_bd_prod;
+       tx_buf->skb = skb;
+
+       tx_bd = &fp->tx_desc_ring[TX_BD(fp->tx_bd_prod)];
+       mapping = pci_map_single(bp->pdev, skb->data,
+                                skb_headlen(skb), PCI_DMA_TODEVICE);
+       tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
+       tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
+       tx_bd->nbd = cpu_to_le16(1);
+       tx_bd->nbytes = cpu_to_le16(skb_headlen(skb));
+       tx_bd->vlan = cpu_to_le16(pkt_prod);
+       tx_bd->bd_flags.as_bitfield = (ETH_TX_BD_FLAGS_START_BD |
+                                      ETH_TX_BD_FLAGS_END_BD);
+       tx_bd->general_data = ((UNICAST_ADDRESS <<
+                               ETH_TX_BD_ETH_ADDR_TYPE_SHIFT) | 1);
+
+       fp->hw_tx_prods->bds_prod =
+               cpu_to_le16(le16_to_cpu(fp->hw_tx_prods->bds_prod) + 1);
+       mb(); /* FW restriction: must not reorder writing nbd and packets */
+       fp->hw_tx_prods->packets_prod =
+               cpu_to_le32(le32_to_cpu(fp->hw_tx_prods->packets_prod) + 1);
+       DOORBELL(bp, FP_IDX(fp), 0);
+
+       mmiowb();
+
+       num_pkts++;
+       fp->tx_bd_prod++;
+       bp->dev->trans_start = jiffies;
+
+       udelay(100);
+
+       tx_idx = le16_to_cpu(*fp->tx_cons_sb);
+       if (tx_idx != tx_start_idx + num_pkts)
+               goto test_loopback_exit;
+
+       rx_idx = le16_to_cpu(*fp->rx_cons_sb);
+       if (rx_idx != rx_start_idx + num_pkts)
+               goto test_loopback_exit;
+
+       cqe = &fp->rx_comp_ring[RCQ_BD(fp->rx_comp_cons)];
+       cqe_fp_flags = cqe->fast_path_cqe.type_error_flags;
+       if (CQE_TYPE(cqe_fp_flags) || (cqe_fp_flags & ETH_RX_ERROR_FALGS))
+               goto test_loopback_rx_exit;
+
+       len = le16_to_cpu(cqe->fast_path_cqe.pkt_len);
+       if (len != pkt_size)
+               goto test_loopback_rx_exit;
+
+       rx_buf = &fp->rx_buf_ring[RX_BD(fp->rx_bd_cons)];
+       skb = rx_buf->skb;
+       skb_reserve(skb, cqe->fast_path_cqe.placement_offset);
+       for (i = ETH_HLEN; i < pkt_size; i++)
+               if (*(skb->data + i) != (unsigned char) (i & 0xff))
+                       goto test_loopback_rx_exit;
+
+       rc = 0;
+
+test_loopback_rx_exit:
+       bp->dev->last_rx = jiffies;
+
+       fp->rx_bd_cons = NEXT_RX_IDX(fp->rx_bd_cons);
+       fp->rx_bd_prod = NEXT_RX_IDX(fp->rx_bd_prod);
+       fp->rx_comp_cons = NEXT_RCQ_IDX(fp->rx_comp_cons);
+       fp->rx_comp_prod = NEXT_RCQ_IDX(fp->rx_comp_prod);
+
+       /* Update producers */
+       bnx2x_update_rx_prod(bp, fp, fp->rx_bd_prod, fp->rx_comp_prod,
+                            fp->rx_sge_prod);
+       mmiowb(); /* keep prod updates ordered */
+
+test_loopback_exit:
+       bp->link_params.loopback_mode = LOOPBACK_NONE;
+
+       return rc;
+}
+
+static int bnx2x_test_loopback(struct bnx2x *bp, u8 link_up)
+{
+       int rc = 0;
+
+       if (!netif_running(bp->dev))
+               return BNX2X_LOOPBACK_FAILED;
+
+       bnx2x_netif_stop(bp);
+
+       if (bnx2x_run_loopback(bp, BNX2X_MAC_LOOPBACK, link_up)) {
+               DP(NETIF_MSG_PROBE, "MAC loopback failed\n");
+               rc |= BNX2X_MAC_LOOPBACK_FAILED;
+       }
+
+       if (bnx2x_run_loopback(bp, BNX2X_PHY_LOOPBACK, link_up)) {
+               DP(NETIF_MSG_PROBE, "PHY loopback failed\n");
+               rc |= BNX2X_PHY_LOOPBACK_FAILED;
+       }
+
+       bnx2x_netif_start(bp);
+
+       return rc;
+}
+
+#define CRC32_RESIDUAL                 0xdebb20e3
+
+static int bnx2x_test_nvram(struct bnx2x *bp)
+{
+       static const struct {
+               int offset;
+               int size;
+       } nvram_tbl[] = {
+               {     0,  0x14 }, /* bootstrap */
+               {  0x14,  0xec }, /* dir */
+               { 0x100, 0x350 }, /* manuf_info */
+               { 0x450,  0xf0 }, /* feature_info */
+               { 0x640,  0x64 }, /* upgrade_key_info */
+               { 0x6a4,  0x64 },
+               { 0x708,  0x70 }, /* manuf_key_info */
+               { 0x778,  0x70 },
+               {     0,     0 }
+       };
+       u32 buf[0x350 / 4];
+       u8 *data = (u8 *)buf;
+       int i, rc;
+       u32 magic, csum;
+
+       rc = bnx2x_nvram_read(bp, 0, data, 4);
+       if (rc) {
+               DP(NETIF_MSG_PROBE, "magic value read (rc -%d)\n", -rc);
+               goto test_nvram_exit;
+       }
+
+       magic = be32_to_cpu(buf[0]);
+       if (magic != 0x669955aa) {
+               DP(NETIF_MSG_PROBE, "magic value (0x%08x)\n", magic);
+               rc = -ENODEV;
+               goto test_nvram_exit;
+       }
+
+       for (i = 0; nvram_tbl[i].size; i++) {
+
+               rc = bnx2x_nvram_read(bp, nvram_tbl[i].offset, data,
+                                     nvram_tbl[i].size);
+               if (rc) {
+                       DP(NETIF_MSG_PROBE,
+                          "nvram_tbl[%d] read data (rc -%d)\n", i, -rc);
+                       goto test_nvram_exit;
+               }
+
+               csum = ether_crc_le(nvram_tbl[i].size, data);
+               if (csum != CRC32_RESIDUAL) {
+                       DP(NETIF_MSG_PROBE,
+                          "nvram_tbl[%d] csum value (0x%08x)\n", i, csum);
+                       rc = -ENODEV;
+                       goto test_nvram_exit;
+               }
+       }
+
+test_nvram_exit:
+       return rc;
+}
+
+static int bnx2x_test_intr(struct bnx2x *bp)
+{
+       struct mac_configuration_cmd *config = bnx2x_sp(bp, mac_config);
+       int i, rc;
+
+       if (!netif_running(bp->dev))
+               return -ENODEV;
+
+       config->hdr.length_6b = 0;
+       config->hdr.offset = 0;
+       config->hdr.client_id = BP_CL_ID(bp);
+       config->hdr.reserved1 = 0;
+
+       rc = bnx2x_sp_post(bp, RAMROD_CMD_ID_ETH_SET_MAC, 0,
+                          U64_HI(bnx2x_sp_mapping(bp, mac_config)),
+                          U64_LO(bnx2x_sp_mapping(bp, mac_config)), 0);
+       if (rc == 0) {
+               bp->set_mac_pending++;
+               for (i = 0; i < 10; i++) {
+                       if (!bp->set_mac_pending)
+                               break;
+                       msleep_interruptible(10);
+               }
+               if (i == 10)
+                       rc = -ENODEV;
+       }
+
+       return rc;
+}
+
+static void bnx2x_self_test(struct net_device *dev,
+                           struct ethtool_test *etest, u64 *buf)
+{
+       struct bnx2x *bp = netdev_priv(dev);
+
+       memset(buf, 0, sizeof(u64) * BNX2X_NUM_TESTS);
+
+       if (!netif_running(dev))
+               return;
+
+       /* offline tests are not suppoerted in MF mode */
+       if (IS_E1HMF(bp))
+               etest->flags &= ~ETH_TEST_FL_OFFLINE;
+
+       if (etest->flags & ETH_TEST_FL_OFFLINE) {
+               u8 link_up;
+
+               link_up = bp->link_vars.link_up;
+               bnx2x_nic_unload(bp, UNLOAD_NORMAL);
+               bnx2x_nic_load(bp, LOAD_DIAG);
+               /* wait until link state is restored */
+               bnx2x_wait_for_link(bp, link_up);
+
+               if (bnx2x_test_registers(bp) != 0) {
+                       buf[0] = 1;
+                       etest->flags |= ETH_TEST_FL_FAILED;
+               }
+               if (bnx2x_test_memory(bp) != 0) {
+                       buf[1] = 1;
+                       etest->flags |= ETH_TEST_FL_FAILED;
+               }
+               buf[2] = bnx2x_test_loopback(bp, link_up);
+               if (buf[2] != 0)
+                       etest->flags |= ETH_TEST_FL_FAILED;
+
+               bnx2x_nic_unload(bp, UNLOAD_NORMAL);
+               bnx2x_nic_load(bp, LOAD_NORMAL);
+               /* wait until link state is restored */
+               bnx2x_wait_for_link(bp, link_up);
+       }
+       if (bnx2x_test_nvram(bp) != 0) {
+               buf[3] = 1;
+               etest->flags |= ETH_TEST_FL_FAILED;
+       }
+       if (bnx2x_test_intr(bp) != 0) {
+               buf[4] = 1;
+               etest->flags |= ETH_TEST_FL_FAILED;
+       }
+       if (bp->port.pmf)
+               if (bnx2x_link_test(bp) != 0) {
+                       buf[5] = 1;
+                       etest->flags |= ETH_TEST_FL_FAILED;
+               }
+       buf[7] = bnx2x_mc_assert(bp);
+       if (buf[7] != 0)
+               etest->flags |= ETH_TEST_FL_FAILED;
+
+#ifdef BNX2X_EXTRA_DEBUG
+       bnx2x_panic_dump(bp);
+#endif
+}
+
+static const struct {
+       long offset;
+       int size;
+       u32 flags;
+#define STATS_FLAGS_PORT               1
+#define STATS_FLAGS_FUNC               2
+       u8 string[ETH_GSTRING_LEN];
 } bnx2x_stats_arr[BNX2X_NUM_STATS] = {
-/* 1 */        { STATS_OFFSET32(valid_bytes_received_hi),     8, 1, "rx_bytes" },
-       { STATS_OFFSET32(error_bytes_received_hi),     8, 1, "rx_error_bytes" },
-       { STATS_OFFSET32(total_bytes_transmitted_hi),  8, 1, "tx_bytes" },
-       { STATS_OFFSET32(tx_stat_ifhcoutbadoctets_hi), 8, 0, "tx_error_bytes" },
+/* 1 */        { STATS_OFFSET32(valid_bytes_received_hi),
+                               8, STATS_FLAGS_FUNC, "rx_bytes" },
+       { STATS_OFFSET32(error_bytes_received_hi),
+                               8, STATS_FLAGS_FUNC, "rx_error_bytes" },
+       { STATS_OFFSET32(total_bytes_transmitted_hi),
+                               8, STATS_FLAGS_FUNC, "tx_bytes" },
+       { STATS_OFFSET32(tx_stat_ifhcoutbadoctets_hi),
+                               8, STATS_FLAGS_PORT, "tx_error_bytes" },
        { STATS_OFFSET32(total_unicast_packets_received_hi),
-                                               8, 1, "rx_ucast_packets" },
+                               8, STATS_FLAGS_FUNC, "rx_ucast_packets" },
        { STATS_OFFSET32(total_multicast_packets_received_hi),
-                                               8, 1, "rx_mcast_packets" },
+                               8, STATS_FLAGS_FUNC, "rx_mcast_packets" },
        { STATS_OFFSET32(total_broadcast_packets_received_hi),
-                                               8, 1, "rx_bcast_packets" },
+                               8, STATS_FLAGS_FUNC, "rx_bcast_packets" },
        { STATS_OFFSET32(total_unicast_packets_transmitted_hi),
-                                               8, 1, "tx_packets" },
+                               8, STATS_FLAGS_FUNC, "tx_packets" },
        { STATS_OFFSET32(tx_stat_dot3statsinternalmactransmiterrors_hi),
-                                               8, 0, "tx_mac_errors" },
+                               8, STATS_FLAGS_PORT, "tx_mac_errors" },
 /* 10 */{ STATS_OFFSET32(rx_stat_dot3statscarriersenseerrors_hi),
-                                               8, 0, "tx_carrier_errors" },
+                               8, STATS_FLAGS_PORT, "tx_carrier_errors" },
        { STATS_OFFSET32(rx_stat_dot3statsfcserrors_hi),
-                                               8, 0, "rx_crc_errors" },
+                               8, STATS_FLAGS_PORT, "rx_crc_errors" },
        { STATS_OFFSET32(rx_stat_dot3statsalignmenterrors_hi),
-                                               8, 0, "rx_align_errors" },
+                               8, STATS_FLAGS_PORT, "rx_align_errors" },
        { STATS_OFFSET32(tx_stat_dot3statssinglecollisionframes_hi),
-                                               8, 0, "tx_single_collisions" },
+                               8, STATS_FLAGS_PORT, "tx_single_collisions" },
        { STATS_OFFSET32(tx_stat_dot3statsmultiplecollisionframes_hi),
-                                               8, 0, "tx_multi_collisions" },
+                               8, STATS_FLAGS_PORT, "tx_multi_collisions" },
        { STATS_OFFSET32(tx_stat_dot3statsdeferredtransmissions_hi),
-                                               8, 0, "tx_deferred" },
+                               8, STATS_FLAGS_PORT, "tx_deferred" },
        { STATS_OFFSET32(tx_stat_dot3statsexcessivecollisions_hi),
-                                               8, 0, "tx_excess_collisions" },
+                               8, STATS_FLAGS_PORT, "tx_excess_collisions" },
        { STATS_OFFSET32(tx_stat_dot3statslatecollisions_hi),
-                                               8, 0, "tx_late_collisions" },
+                               8, STATS_FLAGS_PORT, "tx_late_collisions" },
        { STATS_OFFSET32(tx_stat_etherstatscollisions_hi),
-                                               8, 0, "tx_total_collisions" },
+                               8, STATS_FLAGS_PORT, "tx_total_collisions" },
        { STATS_OFFSET32(rx_stat_etherstatsfragments_hi),
-                                               8, 0, "rx_fragments" },
-/* 20 */{ STATS_OFFSET32(rx_stat_etherstatsjabbers_hi), 8, 0, "rx_jabbers" },
+                               8, STATS_FLAGS_PORT, "rx_fragments" },
+/* 20 */{ STATS_OFFSET32(rx_stat_etherstatsjabbers_hi),
+                               8, STATS_FLAGS_PORT, "rx_jabbers" },
        { STATS_OFFSET32(rx_stat_etherstatsundersizepkts_hi),
-                                               8, 0, "rx_undersize_packets" },
+                               8, STATS_FLAGS_PORT, "rx_undersize_packets" },
        { STATS_OFFSET32(jabber_packets_received),
-                                               4, 1, "rx_oversize_packets" },
+                               4, STATS_FLAGS_FUNC, "rx_oversize_packets" },
        { STATS_OFFSET32(tx_stat_etherstatspkts64octets_hi),
-                                               8, 0, "tx_64_byte_packets" },
+                               8, STATS_FLAGS_PORT, "tx_64_byte_packets" },
        { STATS_OFFSET32(tx_stat_etherstatspkts65octetsto127octets_hi),
-                                       8, 0, "tx_65_to_127_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_65_to_127_byte_packets" },
        { STATS_OFFSET32(tx_stat_etherstatspkts128octetsto255octets_hi),
-                                       8, 0, "tx_128_to_255_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_128_to_255_byte_packets" },
        { STATS_OFFSET32(tx_stat_etherstatspkts256octetsto511octets_hi),
-                                       8, 0, "tx_256_to_511_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_256_to_511_byte_packets" },
        { STATS_OFFSET32(tx_stat_etherstatspkts512octetsto1023octets_hi),
-                                       8, 0, "tx_512_to_1023_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_512_to_1023_byte_packets" },
        { STATS_OFFSET32(etherstatspkts1024octetsto1522octets_hi),
-                                       8, 0, "tx_1024_to_1522_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_1024_to_1522_byte_packets" },
        { STATS_OFFSET32(etherstatspktsover1522octets_hi),
-                                       8, 0, "tx_1523_to_9022_byte_packets" },
+                       8, STATS_FLAGS_PORT, "tx_1523_to_9022_byte_packets" },
 /* 30 */{ STATS_OFFSET32(rx_stat_xonpauseframesreceived_hi),
-                                               8, 0, "rx_xon_frames" },
+                               8, STATS_FLAGS_PORT, "rx_xon_frames" },
        { STATS_OFFSET32(rx_stat_xoffpauseframesreceived_hi),
-                                               8, 0, "rx_xoff_frames" },
-       { STATS_OFFSET32(tx_stat_outxonsent_hi),  8, 0, "tx_xon_frames" },
-       { STATS_OFFSET32(tx_stat_outxoffsent_hi), 8, 0, "tx_xoff_frames" },
+                               8, STATS_FLAGS_PORT, "rx_xoff_frames" },
+       { STATS_OFFSET32(tx_stat_outxonsent_hi),
+                               8, STATS_FLAGS_PORT, "tx_xon_frames" },
+       { STATS_OFFSET32(tx_stat_outxoffsent_hi),
+                               8, STATS_FLAGS_PORT, "tx_xoff_frames" },
        { STATS_OFFSET32(rx_stat_maccontrolframesreceived_hi),
-                                               8, 0, "rx_mac_ctrl_frames" },
-       { STATS_OFFSET32(mac_filter_discard),   4, 1, "rx_filtered_packets" },
-       { STATS_OFFSET32(no_buff_discard),      4, 1, "rx_discards" },
-       { STATS_OFFSET32(xxoverflow_discard),   4, 1, "rx_fw_discards" },
-       { STATS_OFFSET32(brb_drop_hi),          8, 1, "brb_discard" },
-/* 39 */{ STATS_OFFSET32(brb_truncate_discard), 8, 1, "brb_truncate" }
+                               8, STATS_FLAGS_PORT, "rx_mac_ctrl_frames" },
+       { STATS_OFFSET32(mac_filter_discard),
+                               4, STATS_FLAGS_PORT, "rx_filtered_packets" },
+       { STATS_OFFSET32(no_buff_discard),
+                               4, STATS_FLAGS_FUNC, "rx_discards" },
+       { STATS_OFFSET32(xxoverflow_discard),
+                               4, STATS_FLAGS_PORT, "rx_fw_discards" },
+       { STATS_OFFSET32(brb_drop_hi),
+                               8, STATS_FLAGS_PORT, "brb_discard" },
+       { STATS_OFFSET32(brb_truncate_hi),
+                               8, STATS_FLAGS_PORT, "brb_truncate" },
+/* 40 */{ STATS_OFFSET32(rx_err_discard_pkt),
+                               4, STATS_FLAGS_FUNC, "rx_phy_ip_err_discards"},
+       { STATS_OFFSET32(rx_skb_alloc_failed),
+                               4, STATS_FLAGS_FUNC, "rx_skb_alloc_discard" },
+/* 42 */{ STATS_OFFSET32(hw_csum_err),
+                               4, STATS_FLAGS_FUNC, "rx_csum_offload_errors" }
 };
 
+#define IS_NOT_E1HMF_STAT(bp, i) \
+               (IS_E1HMF(bp) && (bnx2x_stats_arr[i].flags & STATS_FLAGS_PORT))
+
 static void bnx2x_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
 {
        struct bnx2x *bp = netdev_priv(dev);
@@ -8427,7 +9010,7 @@ static void bnx2x_get_strings(struct net_device *dev, u32 stringset, u8 *buf)
        switch (stringset) {
        case ETH_SS_STATS:
                for (i = 0, j = 0; i < BNX2X_NUM_STATS; i++) {
-                       if (IS_E1HMF(bp) && (!bnx2x_stats_arr[i].flags))
+                       if (IS_NOT_E1HMF_STAT(bp, i))
                                continue;
                        strcpy(buf + j*ETH_GSTRING_LEN,
                               bnx2x_stats_arr[i].string);
@@ -8447,7 +9030,7 @@ static int bnx2x_get_stats_count(struct net_device *dev)
        int i, num_stats = 0;
 
        for (i = 0; i < BNX2X_NUM_STATS; i++) {
-               if (IS_E1HMF(bp) && (!bnx2x_stats_arr[i].flags))
+               if (IS_NOT_E1HMF_STAT(bp, i))
                        continue;
                num_stats++;
        }
@@ -8462,7 +9045,7 @@ static void bnx2x_get_ethtool_stats(struct net_device *dev,
        int i, j;
 
        for (i = 0, j = 0; i < BNX2X_NUM_STATS; i++) {
-               if (IS_E1HMF(bp) && (!bnx2x_stats_arr[i].flags))
+               if (IS_NOT_E1HMF_STAT(bp, i))
                        continue;
 
                if (bnx2x_stats_arr[i].size == 0) {
@@ -8545,7 +9128,7 @@ static struct ethtool_ops bnx2x_ethtool_ops = {
        .get_rx_csum            = bnx2x_get_rx_csum,
        .set_rx_csum            = bnx2x_set_rx_csum,
        .get_tx_csum            = ethtool_op_get_tx_csum,
-       .set_tx_csum            = ethtool_op_set_tx_csum,
+       .set_tx_csum            = ethtool_op_set_tx_hw_csum,
        .set_flags              = bnx2x_set_flags,
        .get_flags              = ethtool_op_get_flags,
        .get_sg                 = ethtool_op_get_sg,
@@ -8626,17 +9209,16 @@ static int bnx2x_poll(struct napi_struct *napi, int budget)
 
        bnx2x_update_fpsb_idx(fp);
 
-       if ((fp->tx_pkt_prod != le16_to_cpu(*fp->tx_cons_sb)) ||
-           (fp->tx_pkt_prod != fp->tx_pkt_cons))
+       if (BNX2X_HAS_TX_WORK(fp))
                bnx2x_tx_int(fp, budget);
 
-       if (le16_to_cpu(*fp->rx_cons_sb) != fp->rx_comp_cons)
+       if (BNX2X_HAS_RX_WORK(fp))
                work_done = bnx2x_rx_int(fp, budget);
 
-       rmb(); /* bnx2x_has_work() reads the status block */
+       rmb(); /* BNX2X_HAS_WORK() reads the status block */
 
        /* must not complete if we consumed full budget */
-       if ((work_done < budget) && !bnx2x_has_work(fp)) {
+       if ((work_done < budget) && !BNX2X_HAS_WORK(fp)) {
 
 #ifdef BNX2X_STOP_ON_ERROR
 poll_panic:
@@ -8651,9 +9233,180 @@ poll_panic:
        return work_done;
 }
 
-/* Called with netif_tx_lock.
+
+/* we split the first BD into headers and data BDs
+ * to ease the pain of our fellow micocode engineers
+ * we use one mapping for both BDs
+ * So far this has only been observed to happen
+ * in Other Operating Systems(TM)
+ */
+static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
+                                  struct bnx2x_fastpath *fp,
+                                  struct eth_tx_bd **tx_bd, u16 hlen,
+                                  u16 bd_prod, int nbd)
+{
+       struct eth_tx_bd *h_tx_bd = *tx_bd;
+       struct eth_tx_bd *d_tx_bd;
+       dma_addr_t mapping;
+       int old_len = le16_to_cpu(h_tx_bd->nbytes);
+
+       /* first fix first BD */
+       h_tx_bd->nbd = cpu_to_le16(nbd);
+       h_tx_bd->nbytes = cpu_to_le16(hlen);
+
+       DP(NETIF_MSG_TX_QUEUED, "TSO split header size is %d "
+          "(%x:%x) nbd %d\n", h_tx_bd->nbytes, h_tx_bd->addr_hi,
+          h_tx_bd->addr_lo, h_tx_bd->nbd);
+
+       /* now get a new data BD
+        * (after the pbd) and fill it */
+       bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
+       d_tx_bd = &fp->tx_desc_ring[bd_prod];
+
+       mapping = HILO_U64(le32_to_cpu(h_tx_bd->addr_hi),
+                          le32_to_cpu(h_tx_bd->addr_lo)) + hlen;
+
+       d_tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
+       d_tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
+       d_tx_bd->nbytes = cpu_to_le16(old_len - hlen);
+       d_tx_bd->vlan = 0;
+       /* this marks the BD as one that has no individual mapping
+        * the FW ignores this flag in a BD not marked start
+        */
+       d_tx_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_SW_LSO;
+       DP(NETIF_MSG_TX_QUEUED,
+          "TSO split data size is %d (%x:%x)\n",
+          d_tx_bd->nbytes, d_tx_bd->addr_hi, d_tx_bd->addr_lo);
+
+       /* update tx_bd for marking the last BD flag */
+       *tx_bd = d_tx_bd;
+
+       return bd_prod;
+}
+
+static inline u16 bnx2x_csum_fix(unsigned char *t_header, u16 csum, s8 fix)
+{
+       if (fix > 0)
+               csum = (u16) ~csum_fold(csum_sub(csum,
+                               csum_partial(t_header - fix, fix, 0)));
+
+       else if (fix < 0)
+               csum = (u16) ~csum_fold(csum_add(csum,
+                               csum_partial(t_header, -fix, 0)));
+
+       return swab16(csum);
+}
+
+static inline u32 bnx2x_xmit_type(struct bnx2x *bp, struct sk_buff *skb)
+{
+       u32 rc;
+
+       if (skb->ip_summed != CHECKSUM_PARTIAL)
+               rc = XMIT_PLAIN;
+
+       else {
+               if (skb->protocol == ntohs(ETH_P_IPV6)) {
+                       rc = XMIT_CSUM_V6;
+                       if (ipv6_hdr(skb)->nexthdr == IPPROTO_TCP)
+                               rc |= XMIT_CSUM_TCP;
+
+               } else {
+                       rc = XMIT_CSUM_V4;
+                       if (ip_hdr(skb)->protocol == IPPROTO_TCP)
+                               rc |= XMIT_CSUM_TCP;
+               }
+       }
+
+       if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV4)
+               rc |= XMIT_GSO_V4;
+
+       else if (skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6)
+               rc |= XMIT_GSO_V6;
+
+       return rc;
+}
+
+/* check if packet requires linearization (packet is too fragmented) */
+static int bnx2x_pkt_req_lin(struct bnx2x *bp, struct sk_buff *skb,
+                            u32 xmit_type)
+{
+       int to_copy = 0;
+       int hlen = 0;
+       int first_bd_sz = 0;
+
+       /* 3 = 1 (for linear data BD) + 2 (for PBD and last BD) */
+       if (skb_shinfo(skb)->nr_frags >= (MAX_FETCH_BD - 3)) {
+
+               if (xmit_type & XMIT_GSO) {
+                       unsigned short lso_mss = skb_shinfo(skb)->gso_size;
+                       /* Check if LSO packet needs to be copied:
+                          3 = 1 (for headers BD) + 2 (for PBD and last BD) */
+                       int wnd_size = MAX_FETCH_BD - 3;
+                       /* Number of widnows to check */
+                       int num_wnds = skb_shinfo(skb)->nr_frags - wnd_size;
+                       int wnd_idx = 0;
+                       int frag_idx = 0;
+                       u32 wnd_sum = 0;
+
+                       /* Headers length */
+                       hlen = (int)(skb_transport_header(skb) - skb->data) +
+                               tcp_hdrlen(skb);
+
+                       /* Amount of data (w/o headers) on linear part of SKB*/
+                       first_bd_sz = skb_headlen(skb) - hlen;
+
+                       wnd_sum  = first_bd_sz;
+
+                       /* Calculate the first sum - it's special */
+                       for (frag_idx = 0; frag_idx < wnd_size - 1; frag_idx++)
+                               wnd_sum +=
+                                       skb_shinfo(skb)->frags[frag_idx].size;
+
+                       /* If there was data on linear skb data - check it */
+                       if (first_bd_sz > 0) {
+                               if (unlikely(wnd_sum < lso_mss)) {
+                                       to_copy = 1;
+                                       goto exit_lbl;
+                               }
+
+                               wnd_sum -= first_bd_sz;
+                       }
+
+                       /* Others are easier: run through the frag list and
+                          check all windows */
+                       for (wnd_idx = 0; wnd_idx <= num_wnds; wnd_idx++) {
+                               wnd_sum +=
+                         skb_shinfo(skb)->frags[wnd_idx + wnd_size - 1].size;
+
+                               if (unlikely(wnd_sum < lso_mss)) {
+                                       to_copy = 1;
+                                       break;
+                               }
+                               wnd_sum -=
+                                       skb_shinfo(skb)->frags[wnd_idx].size;
+                       }
+
+               } else {
+                       /* in non-LSO too fragmented packet should always
+                          be linearized */
+                       to_copy = 1;
+               }
+       }
+
+exit_lbl:
+       if (unlikely(to_copy))
+               DP(NETIF_MSG_TX_QUEUED,
+                  "Linearization IS REQUIRED for %s packet. "
+                  "num_frags %d  hlen %d  first_bd_sz %d\n",
+                  (xmit_type & XMIT_GSO) ? "LSO" : "non-LSO",
+                  skb_shinfo(skb)->nr_frags, hlen, first_bd_sz);
+
+       return to_copy;
+}
+
+/* called with netif_tx_lock
  * bnx2x_tx_int() runs without netif_tx_lock unless it needs to call
- * netif_wake_queue().
+ * netif_wake_queue()
  */
 static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 {
@@ -8663,17 +9416,21 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
        struct eth_tx_bd *tx_bd;
        struct eth_tx_parse_bd *pbd = NULL;
        u16 pkt_prod, bd_prod;
-       int nbd, fp_index = 0;
+       int nbd, fp_index;
        dma_addr_t mapping;
+       u32 xmit_type = bnx2x_xmit_type(bp, skb);
+       int vlan_off = (bp->e1hov ? 4 : 0);
+       int i;
+       u8 hlen = 0;
 
 #ifdef BNX2X_STOP_ON_ERROR
        if (unlikely(bp->panic))
                return NETDEV_TX_BUSY;
 #endif
 
-       fp_index = smp_processor_id() % (bp->num_queues);
-
+       fp_index = (smp_processor_id() % bp->num_queues);
        fp = &bp->fp[fp_index];
+
        if (unlikely(bnx2x_tx_avail(bp->fp) <
                                        (skb_shinfo(skb)->nr_frags + 3))) {
                bp->eth_stats.driver_xoff++,
@@ -8682,20 +9439,37 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
                return NETDEV_TX_BUSY;
        }
 
+       DP(NETIF_MSG_TX_QUEUED, "SKB: summed %x  protocol %x  protocol(%x,%x)"
+          "  gso type %x  xmit_type %x\n",
+          skb->ip_summed, skb->protocol, ipv6_hdr(skb)->nexthdr,
+          ip_hdr(skb)->protocol, skb_shinfo(skb)->gso_type, xmit_type);
+
+       /* First, check if we need to linearaize the skb
+          (due to FW restrictions) */
+       if (bnx2x_pkt_req_lin(bp, skb, xmit_type)) {
+               /* Statistics of linearization */
+               bp->lin_cnt++;
+               if (skb_linearize(skb) != 0) {
+                       DP(NETIF_MSG_TX_QUEUED, "SKB linearization failed - "
+                          "silently dropping this SKB\n");
+                       dev_kfree_skb_any(skb);
+                       return NETDEV_TX_OK;
+               }
+       }
+
        /*
-       This is a bit ugly. First we use one BD which we mark as start,
+       Please read carefully. First we use one BD which we mark as start,
        then for TSO or xsum we have a parsing info BD,
-       and only then we have the rest of the TSO bds.
+       and only then we have the rest of the TSO BDs.
        (don't forget to mark the last one as last,
        and to unmap only AFTER you write to the BD ...)
-       I would like to thank DovH for this mess.
+       And above all, all pdb sizes are in words - NOT DWORDS!
        */
 
        pkt_prod = fp->tx_pkt_prod++;
-       bd_prod = fp->tx_bd_prod;
-       bd_prod = TX_BD(bd_prod);
+       bd_prod = TX_BD(fp->tx_bd_prod);
 
-       /* get a tx_buff and first bd */
+       /* get a tx_buf and first BD */
        tx_buf = &fp->tx_buf_ring[TX_BD(pkt_prod)];
        tx_bd = &fp->tx_desc_ring[bd_prod];
 
@@ -8704,65 +9478,80 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
                               ETH_TX_BD_ETH_ADDR_TYPE_SHIFT);
        tx_bd->general_data |= 1; /* header nbd */
 
-       /* remember the first bd of the packet */
-       tx_buf->first_bd = bd_prod;
+       /* remember the first BD of the packet */
+       tx_buf->first_bd = fp->tx_bd_prod;
+       tx_buf->skb = skb;
 
        DP(NETIF_MSG_TX_QUEUED,
           "sending pkt %u @%p  next_idx %u  bd %u @%p\n",
           pkt_prod, tx_buf, fp->tx_pkt_prod, bd_prod, tx_bd);
 
-       if (skb->ip_summed == CHECKSUM_PARTIAL) {
-               struct iphdr *iph = ip_hdr(skb);
-               u8 len;
+       if ((bp->vlgrp != NULL) && vlan_tx_tag_present(skb)) {
+               tx_bd->vlan = cpu_to_le16(vlan_tx_tag_get(skb));
+               tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_VLAN_TAG;
+               vlan_off += 4;
+       } else
+               tx_bd->vlan = cpu_to_le16(pkt_prod);
 
-               tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IP_CSUM;
+       if (xmit_type) {
 
-               /* turn on parsing and get a bd */
+               /* turn on parsing and get a BD */
                bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
                pbd = (void *)&fp->tx_desc_ring[bd_prod];
-               len = ((u8 *)iph - (u8 *)skb->data) / 2;
+
+               memset(pbd, 0, sizeof(struct eth_tx_parse_bd));
+       }
+
+       if (xmit_type & XMIT_CSUM) {
+               hlen = (skb_network_header(skb) - skb->data + vlan_off) / 2;
 
                /* for now NS flag is not used in Linux */
-               pbd->global_data = (len |
+               pbd->global_data = (hlen |
                                    ((skb->protocol == ntohs(ETH_P_8021Q)) <<
                                     ETH_TX_PARSE_BD_LLC_SNAP_EN_SHIFT));
-               pbd->ip_hlen = ip_hdrlen(skb) / 2;
-               pbd->total_hlen = cpu_to_le16(len + pbd->ip_hlen);
-               if (iph->protocol == IPPROTO_TCP) {
-                       struct tcphdr *th = tcp_hdr(skb);
 
-                       tx_bd->bd_flags.as_bitfield |=
-                                               ETH_TX_BD_FLAGS_TCP_CSUM;
-                       pbd->tcp_flags = pbd_tcp_flags(skb);
-                       pbd->total_hlen += cpu_to_le16(tcp_hdrlen(skb) / 2);
-                       pbd->tcp_pseudo_csum = swab16(th->check);
+               pbd->ip_hlen = (skb_transport_header(skb) -
+                               skb_network_header(skb)) / 2;
 
-               } else if (iph->protocol == IPPROTO_UDP) {
-                       struct udphdr *uh = udp_hdr(skb);
+               hlen += pbd->ip_hlen + tcp_hdrlen(skb) / 2;
 
+               pbd->total_hlen = cpu_to_le16(hlen);
+               hlen = hlen*2 - vlan_off;
+
+               tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_TCP_CSUM;
+
+               if (xmit_type & XMIT_CSUM_V4)
                        tx_bd->bd_flags.as_bitfield |=
-                                               ETH_TX_BD_FLAGS_TCP_CSUM;
-                       pbd->total_hlen += cpu_to_le16(4);
+                                               ETH_TX_BD_FLAGS_IP_CSUM;
+               else
+                       tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_IPV6;
+
+               if (xmit_type & XMIT_CSUM_TCP) {
+                       pbd->tcp_pseudo_csum = swab16(tcp_hdr(skb)->check);
+
+               } else {
+                       s8 fix = SKB_CS_OFF(skb); /* signed! */
+
                        pbd->global_data |= ETH_TX_PARSE_BD_CS_ANY_FLG;
-                       pbd->cs_offset = 5; /* 10 >> 1 */
-                       pbd->tcp_pseudo_csum = 0;
-                       /* HW bug: we need to subtract 10 bytes before the
-                        * UDP header from the csum
-                        */
-                       uh->check = (u16) ~csum_fold(csum_sub(uh->check,
-                               csum_partial(((u8 *)(uh)-10), 10, 0)));
-               }
-       }
+                       pbd->cs_offset = fix / 2;
 
-       if ((bp->vlgrp != NULL) && vlan_tx_tag_present(skb)) {
-               tx_bd->vlan = cpu_to_le16(vlan_tx_tag_get(skb));
-               tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_VLAN_TAG;
-       } else {
-               tx_bd->vlan = cpu_to_le16(pkt_prod);
+                       DP(NETIF_MSG_TX_QUEUED,
+                          "hlen %d  offset %d  fix %d  csum before fix %x\n",
+                          le16_to_cpu(pbd->total_hlen), pbd->cs_offset, fix,
+                          SKB_CS(skb));
+
+                       /* HW bug: fixup the CSUM */
+                       pbd->tcp_pseudo_csum =
+                               bnx2x_csum_fix(skb_transport_header(skb),
+                                              SKB_CS(skb), fix);
+
+                       DP(NETIF_MSG_TX_QUEUED, "csum after fix %x\n",
+                          pbd->tcp_pseudo_csum);
+               }
        }
 
        mapping = pci_map_single(bp->pdev, skb->data,
-                                skb->len, PCI_DMA_TODEVICE);
+                                skb_headlen(skb), PCI_DMA_TODEVICE);
 
        tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
        tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
@@ -8771,13 +9560,12 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
        tx_bd->nbytes = cpu_to_le16(skb_headlen(skb));
 
        DP(NETIF_MSG_TX_QUEUED, "first bd @%p  addr (%x:%x)  nbd %d"
-          "  nbytes %d  flags %x  vlan %u\n",
-          tx_bd, tx_bd->addr_hi, tx_bd->addr_lo, tx_bd->nbd,
-          tx_bd->nbytes, tx_bd->bd_flags.as_bitfield, tx_bd->vlan);
+          "  nbytes %d  flags %x  vlan %x\n",
+          tx_bd, tx_bd->addr_hi, tx_bd->addr_lo, le16_to_cpu(tx_bd->nbd),
+          le16_to_cpu(tx_bd->nbytes), tx_bd->bd_flags.as_bitfield,
+          le16_to_cpu(tx_bd->vlan));
 
-       if (skb_shinfo(skb)->gso_size &&
-           (skb->len > (bp->dev->mtu + ETH_HLEN))) {
-               int hlen = 2 * le16_to_cpu(pbd->total_hlen);
+       if (xmit_type & XMIT_GSO) {
 
                DP(NETIF_MSG_TX_QUEUED,
                   "TSO packet len %d  hlen %d  total len %d  tso size %d\n",
@@ -8786,99 +9574,60 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
                tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_SW_LSO;
 
-               if (tx_bd->nbytes > cpu_to_le16(hlen)) {
-                       /* we split the first bd into headers and data bds
-                        * to ease the pain of our fellow micocode engineers
-                        * we use one mapping for both bds
-                        * So far this has only been observed to happen
-                        * in Other Operating Systems(TM)
-                        */
-
-                       /* first fix first bd */
-                       nbd++;
-                       tx_bd->nbd = cpu_to_le16(nbd);
-                       tx_bd->nbytes = cpu_to_le16(hlen);
-
-                       /* we only print this as an error
-                        * because we don't think this will ever happen.
-                        */
-                       BNX2X_ERR("TSO split header size is %d (%x:%x)"
-                                 "  nbd %d\n", tx_bd->nbytes, tx_bd->addr_hi,
-                                 tx_bd->addr_lo, tx_bd->nbd);
-
-                       /* now get a new data bd
-                        * (after the pbd) and fill it */
-                       bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
-                       tx_bd = &fp->tx_desc_ring[bd_prod];
-
-                       tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
-                       tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping) + hlen);
-                       tx_bd->nbytes = cpu_to_le16(skb_headlen(skb) - hlen);
-                       tx_bd->vlan = cpu_to_le16(pkt_prod);
-                       /* this marks the bd
-                        * as one that has no individual mapping
-                        * the FW ignores this flag in a bd not marked start
-                        */
-                       tx_bd->bd_flags.as_bitfield = ETH_TX_BD_FLAGS_SW_LSO;
-                       DP(NETIF_MSG_TX_QUEUED,
-                          "TSO split data size is %d (%x:%x)\n",
-                          tx_bd->nbytes, tx_bd->addr_hi, tx_bd->addr_lo);
-               }
-
-               if (!pbd) {
-                       /* supposed to be unreached
-                        * (and therefore not handled properly...)
-                        */
-                       BNX2X_ERR("LSO with no PBD\n");
-                       BUG();
-               }
+               if (unlikely(skb_headlen(skb) > hlen))
+                       bd_prod = bnx2x_tx_split(bp, fp, &tx_bd, hlen,
+                                                bd_prod, ++nbd);
 
                pbd->lso_mss = cpu_to_le16(skb_shinfo(skb)->gso_size);
                pbd->tcp_send_seq = swab32(tcp_hdr(skb)->seq);
-               pbd->ip_id = swab16(ip_hdr(skb)->id);
-               pbd->tcp_pseudo_csum =
+               pbd->tcp_flags = pbd_tcp_flags(skb);
+
+               if (xmit_type & XMIT_GSO_V4) {
+                       pbd->ip_id = swab16(ip_hdr(skb)->id);
+                       pbd->tcp_pseudo_csum =
                                swab16(~csum_tcpudp_magic(ip_hdr(skb)->saddr,
                                                          ip_hdr(skb)->daddr,
                                                          0, IPPROTO_TCP, 0));
+
+               } else
+                       pbd->tcp_pseudo_csum =
+                               swab16(~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
+                                                       &ipv6_hdr(skb)->daddr,
+                                                       0, IPPROTO_TCP, 0));
+
                pbd->global_data |= ETH_TX_PARSE_BD_PSEUDO_CS_WITHOUT_LEN;
        }
 
-       {
-               int i;
+       for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
+               skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
 
-               for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
-                       skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
+               bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
+               tx_bd = &fp->tx_desc_ring[bd_prod];
 
-                       bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
-                       tx_bd = &fp->tx_desc_ring[bd_prod];
+               mapping = pci_map_page(bp->pdev, frag->page, frag->page_offset,
+                                      frag->size, PCI_DMA_TODEVICE);
 
-                       mapping = pci_map_page(bp->pdev, frag->page,
-                                              frag->page_offset,
-                                              frag->size, PCI_DMA_TODEVICE);
+               tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
+               tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
+               tx_bd->nbytes = cpu_to_le16(frag->size);
+               tx_bd->vlan = cpu_to_le16(pkt_prod);
+               tx_bd->bd_flags.as_bitfield = 0;
 
-                       tx_bd->addr_hi = cpu_to_le32(U64_HI(mapping));
-                       tx_bd->addr_lo = cpu_to_le32(U64_LO(mapping));
-                       tx_bd->nbytes = cpu_to_le16(frag->size);
-                       tx_bd->vlan = cpu_to_le16(pkt_prod);
-                       tx_bd->bd_flags.as_bitfield = 0;
-                       DP(NETIF_MSG_TX_QUEUED, "frag %d  bd @%p"
-                          "  addr (%x:%x)  nbytes %d  flags %x\n",
-                          i, tx_bd, tx_bd->addr_hi, tx_bd->addr_lo,
-                          tx_bd->nbytes, tx_bd->bd_flags.as_bitfield);
-               } /* for */
+               DP(NETIF_MSG_TX_QUEUED,
+                  "frag %d  bd @%p  addr (%x:%x)  nbytes %d  flags %x\n",
+                  i, tx_bd, tx_bd->addr_hi, tx_bd->addr_lo,
+                  le16_to_cpu(tx_bd->nbytes), tx_bd->bd_flags.as_bitfield);
        }
 
-       /* now at last mark the bd as the last bd */
+       /* now at last mark the BD as the last BD */
        tx_bd->bd_flags.as_bitfield |= ETH_TX_BD_FLAGS_END_BD;
 
        DP(NETIF_MSG_TX_QUEUED, "last bd @%p  flags %x\n",
           tx_bd, tx_bd->bd_flags.as_bitfield);
 
-       tx_buf->skb = skb;
-
        bd_prod = TX_BD(NEXT_TX_IDX(bd_prod));
 
-       /* now send a tx doorbell, counting the next bd
+       /* now send a tx doorbell, counting the next BD
         * if the packet contains or ends with it
         */
        if (TX_BD_POFF(bd_prod) < nbd)
@@ -8890,20 +9639,20 @@ static int bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
                   "  tcp_flags %x  xsum %x  seq %u  hlen %u\n",
                   pbd, pbd->global_data, pbd->ip_hlen, pbd->ip_id,
                   pbd->lso_mss, pbd->tcp_flags, pbd->tcp_pseudo_csum,
-                  pbd->tcp_send_seq, pbd->total_hlen);
+                  pbd->tcp_send_seq, le16_to_cpu(pbd->total_hlen));
 
-       DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %u  bd %d\n", nbd, bd_prod);
+       DP(NETIF_MSG_TX_QUEUED, "doorbell: nbd %d  bd %u\n", nbd, bd_prod);
 
        fp->hw_tx_prods->bds_prod =
                cpu_to_le16(le16_to_cpu(fp->hw_tx_prods->bds_prod) + nbd);
        mb(); /* FW restriction: must not reorder writing nbd and packets */
        fp->hw_tx_prods->packets_prod =
                cpu_to_le32(le32_to_cpu(fp->hw_tx_prods->packets_prod) + 1);
-       DOORBELL(bp, fp_index, 0);
+       DOORBELL(bp, FP_IDX(fp), 0);
 
        mmiowb();
 
-       fp->tx_bd_prod = bd_prod;
+       fp->tx_bd_prod += nbd;
        dev->trans_start = jiffies;
 
        if (unlikely(bnx2x_tx_avail(fp) < MAX_SKB_FRAGS + 3)) {
@@ -9331,10 +10080,7 @@ static int __devinit bnx2x_init_dev(struct pci_dev *pdev,
        dev->features |= (NETIF_F_HW_VLAN_TX | NETIF_F_HW_VLAN_RX);
 #endif
        dev->features |= (NETIF_F_TSO | NETIF_F_TSO_ECN);
-
-       bp->timer_interval = HZ;
-       bp->current_interval = (poll ? poll : HZ);
-
+       dev->features |= NETIF_F_TSO6;
 
        return 0;
 
@@ -9499,7 +10245,7 @@ static int bnx2x_suspend(struct pci_dev *pdev, pm_message_t state)
 
        netif_device_detach(dev);
 
-       bnx2x_nic_unload(bp, UNLOAD_NORMAL);
+       bnx2x_nic_unload(bp, UNLOAD_CLOSE);
 
        bnx2x_set_power_state(bp, pci_choose_state(pdev, state));
 
@@ -9532,20 +10278,109 @@ static int bnx2x_resume(struct pci_dev *pdev)
        bnx2x_set_power_state(bp, PCI_D0);
        netif_device_attach(dev);
 
-       rc = bnx2x_nic_load(bp, LOAD_NORMAL);
+       rc = bnx2x_nic_load(bp, LOAD_OPEN);
 
        rtnl_unlock();
 
        return rc;
 }
 
+/**
+ * bnx2x_io_error_detected - called when PCI error is detected
+ * @pdev: Pointer to PCI device
+ * @state: The current pci connection state
+ *
+ * This function is called after a PCI bus error affecting
+ * this device has been detected.
+ */
+static pci_ers_result_t bnx2x_io_error_detected(struct pci_dev *pdev,
+                                               pci_channel_state_t state)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct bnx2x *bp = netdev_priv(dev);
+
+       rtnl_lock();
+
+       netif_device_detach(dev);
+
+       if (netif_running(dev))
+               bnx2x_nic_unload(bp, UNLOAD_CLOSE);
+
+       pci_disable_device(pdev);
+
+       rtnl_unlock();
+
+       /* Request a slot reset */
+       return PCI_ERS_RESULT_NEED_RESET;
+}
+
+/**
+ * bnx2x_io_slot_reset - called after the PCI bus has been reset
+ * @pdev: Pointer to PCI device
+ *
+ * Restart the card from scratch, as if from a cold-boot.
+ */
+static pci_ers_result_t bnx2x_io_slot_reset(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct bnx2x *bp = netdev_priv(dev);
+
+       rtnl_lock();
+
+       if (pci_enable_device(pdev)) {
+               dev_err(&pdev->dev,
+                       "Cannot re-enable PCI device after reset\n");
+               rtnl_unlock();
+               return PCI_ERS_RESULT_DISCONNECT;
+       }
+
+       pci_set_master(pdev);
+       pci_restore_state(pdev);
+
+       if (netif_running(dev))
+               bnx2x_set_power_state(bp, PCI_D0);
+
+       rtnl_unlock();
+
+       return PCI_ERS_RESULT_RECOVERED;
+}
+
+/**
+ * bnx2x_io_resume - called when traffic can start flowing again
+ * @pdev: Pointer to PCI device
+ *
+ * This callback is called when the error recovery driver tells us that
+ * its OK to resume normal operation.
+ */
+static void bnx2x_io_resume(struct pci_dev *pdev)
+{
+       struct net_device *dev = pci_get_drvdata(pdev);
+       struct bnx2x *bp = netdev_priv(dev);
+
+       rtnl_lock();
+
+       if (netif_running(dev))
+               bnx2x_nic_load(bp, LOAD_OPEN);
+
+       netif_device_attach(dev);
+
+       rtnl_unlock();
+}
+
+static struct pci_error_handlers bnx2x_err_handler = {
+       .error_detected = bnx2x_io_error_detected,
+       .slot_reset = bnx2x_io_slot_reset,
+       .resume = bnx2x_io_resume,
+};
+
 static struct pci_driver bnx2x_pci_driver = {
-       .name       = DRV_MODULE_NAME,
-       .id_table   = bnx2x_pci_tbl,
-       .probe      = bnx2x_init_one,
-       .remove     = __devexit_p(bnx2x_remove_one),
-       .suspend    = bnx2x_suspend,
-       .resume     = bnx2x_resume,
+       .name        = DRV_MODULE_NAME,
+       .id_table    = bnx2x_pci_tbl,
+       .probe       = bnx2x_init_one,
+       .remove      = __devexit_p(bnx2x_remove_one),
+       .suspend     = bnx2x_suspend,
+       .resume      = bnx2x_resume,
+       .err_handler = &bnx2x_err_handler,
 };
 
 static int __init bnx2x_init(void)