rt2x00: Move start() and stop() handlers into rt2x00lib.c
[safe/jmp/linux-2.6] / drivers / net / bnx2.c
index 0300a75..3bb69d5 100644 (file)
@@ -56,8 +56,8 @@
 
 #define DRV_MODULE_NAME                "bnx2"
 #define PFX DRV_MODULE_NAME    ": "
-#define DRV_MODULE_VERSION     "1.7.0"
-#define DRV_MODULE_RELDATE     "December 11, 2007"
+#define DRV_MODULE_VERSION     "1.7.1"
+#define DRV_MODULE_RELDATE     "December 19, 2007"
 
 #define RUN_AT(x) (jiffies + (x))
 
@@ -399,44 +399,65 @@ bnx2_write_phy(struct bnx2 *bp, u32 reg, u32 val)
 static void
 bnx2_disable_int(struct bnx2 *bp)
 {
-       REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
-              BNX2_PCICFG_INT_ACK_CMD_MASK_INT);
+       int i;
+       struct bnx2_napi *bnapi;
+
+       for (i = 0; i < bp->irq_nvecs; i++) {
+               bnapi = &bp->bnx2_napi[i];
+               REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD, bnapi->int_num |
+                      BNX2_PCICFG_INT_ACK_CMD_MASK_INT);
+       }
        REG_RD(bp, BNX2_PCICFG_INT_ACK_CMD);
 }
 
 static void
 bnx2_enable_int(struct bnx2 *bp)
 {
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       int i;
+       struct bnx2_napi *bnapi;
 
-       REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
-              BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
-              BNX2_PCICFG_INT_ACK_CMD_MASK_INT | bnapi->last_status_idx);
+       for (i = 0; i < bp->irq_nvecs; i++) {
+               bnapi = &bp->bnx2_napi[i];
 
-       REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
-              BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID | bnapi->last_status_idx);
+               REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD, bnapi->int_num |
+                      BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
+                      BNX2_PCICFG_INT_ACK_CMD_MASK_INT |
+                      bnapi->last_status_idx);
 
+               REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD, bnapi->int_num |
+                      BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
+                      bnapi->last_status_idx);
+       }
        REG_WR(bp, BNX2_HC_COMMAND, bp->hc_cmd | BNX2_HC_COMMAND_COAL_NOW);
 }
 
 static void
 bnx2_disable_int_sync(struct bnx2 *bp)
 {
+       int i;
+
        atomic_inc(&bp->intr_sem);
        bnx2_disable_int(bp);
-       synchronize_irq(bp->pdev->irq);
+       for (i = 0; i < bp->irq_nvecs; i++)
+               synchronize_irq(bp->irq_tbl[i].vector);
 }
 
 static void
 bnx2_napi_disable(struct bnx2 *bp)
 {
-       napi_disable(&bp->bnx2_napi.napi);
+       int i;
+
+       for (i = 0; i < bp->irq_nvecs; i++)
+               napi_disable(&bp->bnx2_napi[i].napi);
 }
 
 static void
 bnx2_napi_enable(struct bnx2 *bp)
 {
-       napi_enable(&bp->bnx2_napi.napi);
+       int i;
+
+       for (i = 0; i < bp->irq_nvecs; i++)
+               napi_enable(&bp->bnx2_napi[i].napi);
 }
 
 static void
@@ -559,6 +580,9 @@ bnx2_alloc_mem(struct bnx2 *bp)
 
        /* Combine status and statistics blocks into one allocation. */
        status_blk_size = L1_CACHE_ALIGN(sizeof(struct status_block));
+       if (bp->flags & MSIX_CAP_FLAG)
+               status_blk_size = L1_CACHE_ALIGN(BNX2_MAX_MSIX_HW_VEC *
+                                                BNX2_SBLK_MSIX_ALIGN_SIZE);
        bp->status_stats_size = status_blk_size +
                                sizeof(struct statistics_block);
 
@@ -569,7 +593,17 @@ bnx2_alloc_mem(struct bnx2 *bp)
 
        memset(bp->status_blk, 0, bp->status_stats_size);
 
-       bp->bnx2_napi.status_blk = bp->status_blk;
+       bp->bnx2_napi[0].status_blk = bp->status_blk;
+       if (bp->flags & MSIX_CAP_FLAG) {
+               for (i = 1; i < BNX2_MAX_MSIX_VEC; i++) {
+                       struct bnx2_napi *bnapi = &bp->bnx2_napi[i];
+
+                       bnapi->status_blk_msix = (void *)
+                               ((unsigned long) bp->status_blk +
+                                BNX2_SBLK_MSIX_ALIGN_SIZE * i);
+                       bnapi->int_num = i << 24;
+               }
+       }
 
        bp->stats_blk = (void *) ((unsigned long) bp->status_blk +
                                  status_blk_size);
@@ -2276,7 +2310,7 @@ bnx2_free_rx_page(struct bnx2 *bp, u16 index)
 }
 
 static inline int
-bnx2_alloc_rx_skb(struct bnx2 *bp, u16 index)
+bnx2_alloc_rx_skb(struct bnx2 *bp, struct bnx2_napi *bnapi, u16 index)
 {
        struct sk_buff *skb;
        struct sw_bd *rx_buf = &bp->rx_buf_ring[index];
@@ -2301,7 +2335,7 @@ bnx2_alloc_rx_skb(struct bnx2 *bp, u16 index)
        rxbd->rx_bd_haddr_hi = (u64) mapping >> 32;
        rxbd->rx_bd_haddr_lo = (u64) mapping & 0xffffffff;
 
-       bp->rx_prod_bseq += bp->rx_buf_use_size;
+       bnapi->rx_prod_bseq += bp->rx_buf_use_size;
 
        return 0;
 }
@@ -2344,18 +2378,21 @@ bnx2_get_hw_tx_cons(struct bnx2_napi *bnapi)
 {
        u16 cons;
 
-       cons = bnapi->status_blk->status_tx_quick_consumer_index0;
+       if (bnapi->int_num == 0)
+               cons = bnapi->status_blk->status_tx_quick_consumer_index0;
+       else
+               cons = bnapi->status_blk_msix->status_tx_quick_consumer_index;
 
        if (unlikely((cons & MAX_TX_DESC_CNT) == MAX_TX_DESC_CNT))
                cons++;
        return cons;
 }
 
-static void
-bnx2_tx_int(struct bnx2 *bp, struct bnx2_napi *bnapi)
+static int
+bnx2_tx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
 {
        u16 hw_cons, sw_cons, sw_ring_cons;
-       int tx_free_bd = 0;
+       int tx_pkt = 0;
 
        hw_cons = bnx2_get_hw_tx_cons(bnapi);
        sw_cons = bnapi->tx_cons;
@@ -2405,9 +2442,10 @@ bnx2_tx_int(struct bnx2 *bp, struct bnx2_napi *bnapi)
 
                sw_cons = NEXT_TX_BD(sw_cons);
 
-               tx_free_bd += last + 1;
-
                dev_kfree_skb(skb);
+               tx_pkt++;
+               if (tx_pkt == budget)
+                       break;
 
                hw_cons = bnx2_get_hw_tx_cons(bnapi);
        }
@@ -2429,17 +2467,19 @@ bnx2_tx_int(struct bnx2 *bp, struct bnx2_napi *bnapi)
                        netif_wake_queue(bp->dev);
                netif_tx_unlock(bp->dev);
        }
+       return tx_pkt;
 }
 
 static void
-bnx2_reuse_rx_skb_pages(struct bnx2 *bp, struct sk_buff *skb, int count)
+bnx2_reuse_rx_skb_pages(struct bnx2 *bp, struct bnx2_napi *bnapi,
+                       struct sk_buff *skb, int count)
 {
        struct sw_pg *cons_rx_pg, *prod_rx_pg;
        struct rx_bd *cons_bd, *prod_bd;
        dma_addr_t mapping;
        int i;
-       u16 hw_prod = bp->rx_pg_prod, prod;
-       u16 cons = bp->rx_pg_cons;
+       u16 hw_prod = bnapi->rx_pg_prod, prod;
+       u16 cons = bnapi->rx_pg_cons;
 
        for (i = 0; i < count; i++) {
                prod = RX_PG_RING_IDX(hw_prod);
@@ -2476,12 +2516,12 @@ bnx2_reuse_rx_skb_pages(struct bnx2 *bp, struct sk_buff *skb, int count)
                cons = RX_PG_RING_IDX(NEXT_RX_BD(cons));
                hw_prod = NEXT_RX_BD(hw_prod);
        }
-       bp->rx_pg_prod = hw_prod;
-       bp->rx_pg_cons = cons;
+       bnapi->rx_pg_prod = hw_prod;
+       bnapi->rx_pg_cons = cons;
 }
 
 static inline void
-bnx2_reuse_rx_skb(struct bnx2 *bp, struct sk_buff *skb,
+bnx2_reuse_rx_skb(struct bnx2 *bp, struct bnx2_napi *bnapi, struct sk_buff *skb,
        u16 cons, u16 prod)
 {
        struct sw_bd *cons_rx_buf, *prod_rx_buf;
@@ -2494,7 +2534,7 @@ bnx2_reuse_rx_skb(struct bnx2 *bp, struct sk_buff *skb,
                pci_unmap_addr(cons_rx_buf, mapping),
                bp->rx_offset + RX_COPY_THRESH, PCI_DMA_FROMDEVICE);
 
-       bp->rx_prod_bseq += bp->rx_buf_use_size;
+       bnapi->rx_prod_bseq += bp->rx_buf_use_size;
 
        prod_rx_buf->skb = skb;
 
@@ -2511,20 +2551,21 @@ bnx2_reuse_rx_skb(struct bnx2 *bp, struct sk_buff *skb,
 }
 
 static int
-bnx2_rx_skb(struct bnx2 *bp, struct sk_buff *skb, unsigned int len,
-           unsigned int hdr_len, dma_addr_t dma_addr, u32 ring_idx)
+bnx2_rx_skb(struct bnx2 *bp, struct bnx2_napi *bnapi, struct sk_buff *skb,
+           unsigned int len, unsigned int hdr_len, dma_addr_t dma_addr,
+           u32 ring_idx)
 {
        int err;
        u16 prod = ring_idx & 0xffff;
 
-       err = bnx2_alloc_rx_skb(bp, prod);
+       err = bnx2_alloc_rx_skb(bp, bnapi, prod);
        if (unlikely(err)) {
-               bnx2_reuse_rx_skb(bp, skb, (u16) (ring_idx >> 16), prod);
+               bnx2_reuse_rx_skb(bp, bnapi, skb, (u16) (ring_idx >> 16), prod);
                if (hdr_len) {
                        unsigned int raw_len = len + 4;
                        int pages = PAGE_ALIGN(raw_len - hdr_len) >> PAGE_SHIFT;
 
-                       bnx2_reuse_rx_skb_pages(bp, NULL, pages);
+                       bnx2_reuse_rx_skb_pages(bp, bnapi, NULL, pages);
                }
                return err;
        }
@@ -2539,8 +2580,8 @@ bnx2_rx_skb(struct bnx2 *bp, struct sk_buff *skb, unsigned int len,
        } else {
                unsigned int i, frag_len, frag_size, pages;
                struct sw_pg *rx_pg;
-               u16 pg_cons = bp->rx_pg_cons;
-               u16 pg_prod = bp->rx_pg_prod;
+               u16 pg_cons = bnapi->rx_pg_cons;
+               u16 pg_prod = bnapi->rx_pg_prod;
 
                frag_size = len + 4 - hdr_len;
                pages = PAGE_ALIGN(frag_size) >> PAGE_SHIFT;
@@ -2551,9 +2592,10 @@ bnx2_rx_skb(struct bnx2 *bp, struct sk_buff *skb, unsigned int len,
                        if (unlikely(frag_len <= 4)) {
                                unsigned int tail = 4 - frag_len;
 
-                               bp->rx_pg_cons = pg_cons;
-                               bp->rx_pg_prod = pg_prod;
-                               bnx2_reuse_rx_skb_pages(bp, NULL, pages - i);
+                               bnapi->rx_pg_cons = pg_cons;
+                               bnapi->rx_pg_prod = pg_prod;
+                               bnx2_reuse_rx_skb_pages(bp, bnapi, NULL,
+                                                       pages - i);
                                skb->len -= tail;
                                if (i == 0) {
                                        skb->tail -= tail;
@@ -2579,9 +2621,10 @@ bnx2_rx_skb(struct bnx2 *bp, struct sk_buff *skb, unsigned int len,
 
                        err = bnx2_alloc_rx_page(bp, RX_PG_RING_IDX(pg_prod));
                        if (unlikely(err)) {
-                               bp->rx_pg_cons = pg_cons;
-                               bp->rx_pg_prod = pg_prod;
-                               bnx2_reuse_rx_skb_pages(bp, skb, pages - i);
+                               bnapi->rx_pg_cons = pg_cons;
+                               bnapi->rx_pg_prod = pg_prod;
+                               bnx2_reuse_rx_skb_pages(bp, bnapi, skb,
+                                                       pages - i);
                                return err;
                        }
 
@@ -2593,8 +2636,8 @@ bnx2_rx_skb(struct bnx2 *bp, struct sk_buff *skb, unsigned int len,
                        pg_prod = NEXT_RX_BD(pg_prod);
                        pg_cons = RX_PG_RING_IDX(NEXT_RX_BD(pg_cons));
                }
-               bp->rx_pg_prod = pg_prod;
-               bp->rx_pg_cons = pg_cons;
+               bnapi->rx_pg_prod = pg_prod;
+               bnapi->rx_pg_cons = pg_cons;
        }
        return 0;
 }
@@ -2617,8 +2660,8 @@ bnx2_rx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
        int rx_pkt = 0, pg_ring_used = 0;
 
        hw_cons = bnx2_get_hw_rx_cons(bnapi);
-       sw_cons = bp->rx_cons;
-       sw_prod = bp->rx_prod;
+       sw_cons = bnapi->rx_cons;
+       sw_prod = bnapi->rx_prod;
 
        /* Memory barrier necessary as speculative reads of the rx
         * buffer can be ahead of the index in the status block
@@ -2654,7 +2697,8 @@ bnx2_rx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
                        L2_FHDR_ERRORS_TOO_SHORT |
                        L2_FHDR_ERRORS_GIANT_FRAME)) {
 
-                       bnx2_reuse_rx_skb(bp, skb, sw_ring_cons, sw_ring_prod);
+                       bnx2_reuse_rx_skb(bp, bnapi, skb, sw_ring_cons,
+                                         sw_ring_prod);
                        goto next_rx;
                }
                hdr_len = 0;
@@ -2673,7 +2717,7 @@ bnx2_rx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
 
                        new_skb = netdev_alloc_skb(bp->dev, len + 2);
                        if (new_skb == NULL) {
-                               bnx2_reuse_rx_skb(bp, skb, sw_ring_cons,
+                               bnx2_reuse_rx_skb(bp, bnapi, skb, sw_ring_cons,
                                                  sw_ring_prod);
                                goto next_rx;
                        }
@@ -2684,12 +2728,12 @@ bnx2_rx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
                        skb_reserve(new_skb, 2);
                        skb_put(new_skb, len);
 
-                       bnx2_reuse_rx_skb(bp, skb,
+                       bnx2_reuse_rx_skb(bp, bnapi, skb,
                                sw_ring_cons, sw_ring_prod);
 
                        skb = new_skb;
-               } else if (unlikely(bnx2_rx_skb(bp, skb, len, hdr_len, dma_addr,
-                                   (sw_ring_cons << 16) | sw_ring_prod)))
+               } else if (unlikely(bnx2_rx_skb(bp, bnapi, skb, len, hdr_len,
+                          dma_addr, (sw_ring_cons << 16) | sw_ring_prod)))
                        goto next_rx;
 
                skb->protocol = eth_type_trans(skb, bp->dev);
@@ -2713,7 +2757,7 @@ bnx2_rx_int(struct bnx2 *bp, struct bnx2_napi *bnapi, int budget)
                }
 
 #ifdef BCM_VLAN
-               if ((status & L2_FHDR_STATUS_L2_VLAN_TAG) && (bp->vlgrp != 0)) {
+               if ((status & L2_FHDR_STATUS_L2_VLAN_TAG) && bp->vlgrp) {
                        vlan_hwaccel_receive_skb(skb, bp->vlgrp,
                                rx_hdr->l2_fhdr_vlan_tag);
                }
@@ -2737,16 +2781,16 @@ next_rx:
                        rmb();
                }
        }
-       bp->rx_cons = sw_cons;
-       bp->rx_prod = sw_prod;
+       bnapi->rx_cons = sw_cons;
+       bnapi->rx_prod = sw_prod;
 
        if (pg_ring_used)
                REG_WR16(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_PG_BDIDX,
-                        bp->rx_pg_prod);
+                        bnapi->rx_pg_prod);
 
        REG_WR16(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BDIDX, sw_prod);
 
-       REG_WR(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BSEQ, bp->rx_prod_bseq);
+       REG_WR(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BSEQ, bnapi->rx_prod_bseq);
 
        mmiowb();
 
@@ -2762,7 +2806,7 @@ bnx2_msi(int irq, void *dev_instance)
 {
        struct net_device *dev = dev_instance;
        struct bnx2 *bp = netdev_priv(dev);
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[0];
 
        prefetch(bnapi->status_blk);
        REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
@@ -2783,7 +2827,7 @@ bnx2_msi_1shot(int irq, void *dev_instance)
 {
        struct net_device *dev = dev_instance;
        struct bnx2 *bp = netdev_priv(dev);
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[0];
 
        prefetch(bnapi->status_blk);
 
@@ -2801,7 +2845,7 @@ bnx2_interrupt(int irq, void *dev_instance)
 {
        struct net_device *dev = dev_instance;
        struct bnx2 *bp = netdev_priv(dev);
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[0];
        struct status_block *sblk = bnapi->status_blk;
 
        /* When using INTx, it is possible for the interrupt to arrive
@@ -2836,6 +2880,23 @@ bnx2_interrupt(int irq, void *dev_instance)
        return IRQ_HANDLED;
 }
 
+static irqreturn_t
+bnx2_tx_msix(int irq, void *dev_instance)
+{
+       struct net_device *dev = dev_instance;
+       struct bnx2 *bp = netdev_priv(dev);
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[BNX2_TX_VEC];
+
+       prefetch(bnapi->status_blk_msix);
+
+       /* Return here if interrupt is disabled. */
+       if (unlikely(atomic_read(&bp->intr_sem) != 0))
+               return IRQ_HANDLED;
+
+       netif_rx_schedule(dev, &bnapi->napi);
+       return IRQ_HANDLED;
+}
+
 #define STATUS_ATTN_EVENTS     (STATUS_ATTN_BITS_LINK_STATE | \
                                 STATUS_ATTN_BITS_TIMER_ABORT)
 
@@ -2845,7 +2906,7 @@ bnx2_has_work(struct bnx2_napi *bnapi)
        struct bnx2 *bp = bnapi->bp;
        struct status_block *sblk = bp->status_blk;
 
-       if ((bnx2_get_hw_rx_cons(bnapi) != bp->rx_cons) ||
+       if ((bnx2_get_hw_rx_cons(bnapi) != bnapi->rx_cons) ||
            (bnx2_get_hw_tx_cons(bnapi) != bnapi->hw_tx_cons))
                return 1;
 
@@ -2856,6 +2917,29 @@ bnx2_has_work(struct bnx2_napi *bnapi)
        return 0;
 }
 
+static int bnx2_tx_poll(struct napi_struct *napi, int budget)
+{
+       struct bnx2_napi *bnapi = container_of(napi, struct bnx2_napi, napi);
+       struct bnx2 *bp = bnapi->bp;
+       int work_done = 0;
+       struct status_block_msix *sblk = bnapi->status_blk_msix;
+
+       do {
+               work_done += bnx2_tx_int(bp, bnapi, budget - work_done);
+               if (unlikely(work_done >= budget))
+                       return work_done;
+
+               bnapi->last_status_idx = sblk->status_idx;
+               rmb();
+       } while (bnx2_get_hw_tx_cons(bnapi) != bnapi->hw_tx_cons);
+
+       netif_rx_complete(bp->dev, napi);
+       REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD, bnapi->int_num |
+              BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
+              bnapi->last_status_idx);
+       return work_done;
+}
+
 static int bnx2_poll_work(struct bnx2 *bp, struct bnx2_napi *bnapi,
                          int work_done, int budget)
 {
@@ -2877,9 +2961,9 @@ static int bnx2_poll_work(struct bnx2 *bp, struct bnx2_napi *bnapi,
        }
 
        if (bnx2_get_hw_tx_cons(bnapi) != bnapi->hw_tx_cons)
-               bnx2_tx_int(bp, bnapi);
+               bnx2_tx_int(bp, bnapi, 0);
 
-       if (bnx2_get_hw_rx_cons(bnapi) != bp->rx_cons)
+       if (bnx2_get_hw_rx_cons(bnapi) != bnapi->rx_cons)
                work_done += bnx2_rx_int(bp, bnapi, budget - work_done);
 
        return work_done;
@@ -2906,7 +2990,7 @@ static int bnx2_poll(struct napi_struct *napi, int budget)
                rmb();
                if (likely(!bnx2_has_work(bnapi))) {
                        netif_rx_complete(bp->dev, napi);
-                       if (likely(bp->flags & USING_MSI_FLAG)) {
+                       if (likely(bp->flags & USING_MSI_OR_MSIX_FLAG)) {
                                REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD,
                                       BNX2_PCICFG_INT_ACK_CMD_INDEX_VALID |
                                       bnapi->last_status_idx);
@@ -4067,6 +4151,15 @@ bnx2_init_remote_phy(struct bnx2 *bp)
        }
 }
 
+static void
+bnx2_setup_msix_tbl(struct bnx2 *bp)
+{
+       REG_WR(bp, BNX2_PCI_GRC_WINDOW_ADDR, BNX2_PCI_GRC_WINDOW_ADDR_SEP_WIN);
+
+       REG_WR(bp, BNX2_PCI_GRC_WINDOW2_ADDR, BNX2_MSIX_TABLE_ADDR);
+       REG_WR(bp, BNX2_PCI_GRC_WINDOW3_ADDR, BNX2_MSIX_PBA_ADDR);
+}
+
 static int
 bnx2_reset_chip(struct bnx2 *bp, u32 reset_code)
 {
@@ -4166,6 +4259,9 @@ bnx2_reset_chip(struct bnx2 *bp, u32 reset_code)
                rc = bnx2_alloc_bad_rbuf(bp);
        }
 
+       if (bp->flags & USING_MSIX_FLAG)
+               bnx2_setup_msix_tbl(bp);
+
        return rc;
 }
 
@@ -4173,7 +4269,7 @@ static int
 bnx2_init_chip(struct bnx2 *bp)
 {
        u32 val;
-       int rc;
+       int rc, i;
 
        /* Make sure the interrupt is not active. */
        REG_WR(bp, BNX2_PCICFG_INT_ACK_CMD, BNX2_PCICFG_INT_ACK_CMD_MASK_INT);
@@ -4269,7 +4365,9 @@ bnx2_init_chip(struct bnx2 *bp)
                val |= BNX2_EMAC_RX_MTU_SIZE_JUMBO_ENA;
        REG_WR(bp, BNX2_EMAC_RX_MTU_SIZE, val);
 
-       bp->bnx2_napi.last_status_idx = 0;
+       for (i = 0; i < BNX2_MAX_MSIX_VEC; i++)
+               bp->bnx2_napi[i].last_status_idx = 0;
+
        bp->rx_mode = BNX2_EMAC_RX_MODE_SORT_MODE;
 
        /* Set up how to generate a link change interrupt. */
@@ -4316,6 +4414,24 @@ bnx2_init_chip(struct bnx2 *bp)
                      BNX2_HC_CONFIG_COLLECT_STATS;
        }
 
+       if (bp->flags & USING_MSIX_FLAG) {
+               REG_WR(bp, BNX2_HC_MSIX_BIT_VECTOR,
+                      BNX2_HC_MSIX_BIT_VECTOR_VAL);
+
+               REG_WR(bp, BNX2_HC_SB_CONFIG_1,
+                       BNX2_HC_SB_CONFIG_1_TX_TMR_MODE |
+                       BNX2_HC_SB_CONFIG_1_ONE_SHOT);
+
+               REG_WR(bp, BNX2_HC_TX_QUICK_CONS_TRIP_1,
+                       (bp->tx_quick_cons_trip_int << 16) |
+                        bp->tx_quick_cons_trip);
+
+               REG_WR(bp, BNX2_HC_TX_TICKS_1,
+                       (bp->tx_ticks_int << 16) | bp->tx_ticks);
+
+               val |= BNX2_HC_CONFIG_SB_ADDR_INC_128B;
+       }
+
        if (bp->flags & ONE_SHOT_MSI_FLAG)
                val |= BNX2_HC_CONFIG_ONE_SHOT;
 
@@ -4348,6 +4464,25 @@ bnx2_init_chip(struct bnx2 *bp)
 }
 
 static void
+bnx2_clear_ring_states(struct bnx2 *bp)
+{
+       struct bnx2_napi *bnapi;
+       int i;
+
+       for (i = 0; i < BNX2_MAX_MSIX_VEC; i++) {
+               bnapi = &bp->bnx2_napi[i];
+
+               bnapi->tx_cons = 0;
+               bnapi->hw_tx_cons = 0;
+               bnapi->rx_prod_bseq = 0;
+               bnapi->rx_prod = 0;
+               bnapi->rx_cons = 0;
+               bnapi->rx_pg_prod = 0;
+               bnapi->rx_pg_cons = 0;
+       }
+}
+
+static void
 bnx2_init_tx_context(struct bnx2 *bp, u32 cid)
 {
        u32 val, offset0, offset1, offset2, offset3;
@@ -4380,8 +4515,17 @@ static void
 bnx2_init_tx_ring(struct bnx2 *bp)
 {
        struct tx_bd *txbd;
-       u32 cid;
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       u32 cid = TX_CID;
+       struct bnx2_napi *bnapi;
+
+       bp->tx_vec = 0;
+       if (bp->flags & USING_MSIX_FLAG) {
+               cid = TX_TSS_CID;
+               bp->tx_vec = BNX2_TX_VEC;
+               REG_WR(bp, BNX2_TSCH_TSS_CFG, BNX2_TX_INT_NUM |
+                      (TX_TSS_CID << 7));
+       }
+       bnapi = &bp->bnx2_napi[bp->tx_vec];
 
        bp->tx_wake_thresh = bp->tx_ring_size / 2;
 
@@ -4391,11 +4535,8 @@ bnx2_init_tx_ring(struct bnx2 *bp)
        txbd->tx_bd_haddr_lo = (u64) bp->tx_desc_mapping & 0xffffffff;
 
        bp->tx_prod = 0;
-       bnapi->tx_cons = 0;
-       bnapi->hw_tx_cons = 0;
        bp->tx_prod_bseq = 0;
 
-       cid = TX_CID;
        bp->tx_bidx_addr = MB_GET_CID_ADDR(cid) + BNX2_L2CTX_TX_HOST_BIDX;
        bp->tx_bseq_addr = MB_GET_CID_ADDR(cid) + BNX2_L2CTX_TX_HOST_BSEQ;
 
@@ -4432,12 +4573,7 @@ bnx2_init_rx_ring(struct bnx2 *bp)
        int i;
        u16 prod, ring_prod;
        u32 val, rx_cid_addr = GET_CID_ADDR(RX_CID);
-
-       bp->rx_prod = 0;
-       bp->rx_cons = 0;
-       bp->rx_prod_bseq = 0;
-       bp->rx_pg_prod = 0;
-       bp->rx_pg_cons = 0;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[0];
 
        bnx2_init_rxbd_rings(bp->rx_desc_ring, bp->rx_desc_mapping,
                             bp->rx_buf_use_size, bp->rx_max_ring);
@@ -4473,29 +4609,30 @@ bnx2_init_rx_ring(struct bnx2 *bp)
        val = (u64) bp->rx_desc_mapping[0] & 0xffffffff;
        CTX_WR(bp, rx_cid_addr, BNX2_L2CTX_NX_BDHADDR_LO, val);
 
-       ring_prod = prod = bp->rx_pg_prod;
+       ring_prod = prod = bnapi->rx_pg_prod;
        for (i = 0; i < bp->rx_pg_ring_size; i++) {
                if (bnx2_alloc_rx_page(bp, ring_prod) < 0)
                        break;
                prod = NEXT_RX_BD(prod);
                ring_prod = RX_PG_RING_IDX(prod);
        }
-       bp->rx_pg_prod = prod;
+       bnapi->rx_pg_prod = prod;
 
-       ring_prod = prod = bp->rx_prod;
+       ring_prod = prod = bnapi->rx_prod;
        for (i = 0; i < bp->rx_ring_size; i++) {
-               if (bnx2_alloc_rx_skb(bp, ring_prod) < 0) {
+               if (bnx2_alloc_rx_skb(bp, bnapi, ring_prod) < 0) {
                        break;
                }
                prod = NEXT_RX_BD(prod);
                ring_prod = RX_RING_IDX(prod);
        }
-       bp->rx_prod = prod;
+       bnapi->rx_prod = prod;
 
-       REG_WR16(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_PG_BDIDX, bp->rx_pg_prod);
+       REG_WR16(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_PG_BDIDX,
+                bnapi->rx_pg_prod);
        REG_WR16(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BDIDX, prod);
 
-       REG_WR(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BSEQ, bp->rx_prod_bseq);
+       REG_WR(bp, MB_RX_CID_ADDR + BNX2_L2CTX_HOST_BSEQ, bnapi->rx_prod_bseq);
 }
 
 static u32 bnx2_find_max_ring(u32 ring_size, u32 max_size)
@@ -4639,6 +4776,7 @@ bnx2_reset_nic(struct bnx2 *bp, u32 reset_code)
        if ((rc = bnx2_init_chip(bp)) != 0)
                return rc;
 
+       bnx2_clear_ring_states(bp);
        bnx2_init_tx_ring(bp);
        bnx2_init_rx_ring(bp);
        return 0;
@@ -4910,7 +5048,11 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
        struct sw_bd *rx_buf;
        struct l2_fhdr *rx_hdr;
        int ret = -ENODEV;
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[0], *tx_napi;
+
+       tx_napi = bnapi;
+       if (bp->flags & USING_MSIX_FLAG)
+               tx_napi = &bp->bnx2_napi[BNX2_TX_VEC];
 
        if (loopback_mode == BNX2_MAC_LOOPBACK) {
                bp->loopback = MAC_LOOPBACK;
@@ -4975,7 +5117,7 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
        pci_unmap_single(bp->pdev, map, pkt_size, PCI_DMA_TODEVICE);
        dev_kfree_skb(skb);
 
-       if (bnx2_get_hw_tx_cons(bnapi) != bp->tx_prod)
+       if (bnx2_get_hw_tx_cons(tx_napi) != bp->tx_prod)
                goto loopback_test_done;
 
        rx_idx = bnx2_get_hw_rx_cons(bnapi);
@@ -5259,14 +5401,22 @@ bnx2_request_irq(struct bnx2 *bp)
 {
        struct net_device *dev = bp->dev;
        unsigned long flags;
-       struct bnx2_irq *irq = &bp->irq_tbl[0];
-       int rc;
+       struct bnx2_irq *irq;
+       int rc = 0, i;
 
-       if (bp->flags & USING_MSI_FLAG)
+       if (bp->flags & USING_MSI_OR_MSIX_FLAG)
                flags = 0;
        else
                flags = IRQF_SHARED;
-       rc = request_irq(irq->vector, irq->handler, flags, dev->name, dev);
+
+       for (i = 0; i < bp->irq_nvecs; i++) {
+               irq = &bp->irq_tbl[i];
+               rc = request_irq(irq->vector, irq->handler, flags, irq->name,
+                                dev);
+               if (rc)
+                       break;
+               irq->requested = 1;
+       }
        return rc;
 }
 
@@ -5274,12 +5424,55 @@ static void
 bnx2_free_irq(struct bnx2 *bp)
 {
        struct net_device *dev = bp->dev;
+       struct bnx2_irq *irq;
+       int i;
 
-       free_irq(bp->irq_tbl[0].vector, dev);
-       if (bp->flags & USING_MSI_FLAG) {
+       for (i = 0; i < bp->irq_nvecs; i++) {
+               irq = &bp->irq_tbl[i];
+               if (irq->requested)
+                       free_irq(irq->vector, dev);
+               irq->requested = 0;
+       }
+       if (bp->flags & USING_MSI_FLAG)
                pci_disable_msi(bp->pdev);
-               bp->flags &= ~(USING_MSI_FLAG | ONE_SHOT_MSI_FLAG);
+       else if (bp->flags & USING_MSIX_FLAG)
+               pci_disable_msix(bp->pdev);
+
+       bp->flags &= ~(USING_MSI_OR_MSIX_FLAG | ONE_SHOT_MSI_FLAG);
+}
+
+static void
+bnx2_enable_msix(struct bnx2 *bp)
+{
+       int i, rc;
+       struct msix_entry msix_ent[BNX2_MAX_MSIX_VEC];
+
+       bnx2_setup_msix_tbl(bp);
+       REG_WR(bp, BNX2_PCI_MSIX_CONTROL, BNX2_MAX_MSIX_HW_VEC - 1);
+       REG_WR(bp, BNX2_PCI_MSIX_TBL_OFF_BIR, BNX2_PCI_GRC_WINDOW2_BASE);
+       REG_WR(bp, BNX2_PCI_MSIX_PBA_OFF_BIT, BNX2_PCI_GRC_WINDOW3_BASE);
+
+       for (i = 0; i < BNX2_MAX_MSIX_VEC; i++) {
+               msix_ent[i].entry = i;
+               msix_ent[i].vector = 0;
        }
+
+       rc = pci_enable_msix(bp->pdev, msix_ent, BNX2_MAX_MSIX_VEC);
+       if (rc != 0)
+               return;
+
+       bp->irq_tbl[BNX2_BASE_VEC].handler = bnx2_msi_1shot;
+       bp->irq_tbl[BNX2_TX_VEC].handler = bnx2_tx_msix;
+
+       strcpy(bp->irq_tbl[BNX2_BASE_VEC].name, bp->dev->name);
+       strcat(bp->irq_tbl[BNX2_BASE_VEC].name, "-base");
+       strcpy(bp->irq_tbl[BNX2_TX_VEC].name, bp->dev->name);
+       strcat(bp->irq_tbl[BNX2_TX_VEC].name, "-tx");
+
+       bp->irq_nvecs = BNX2_MAX_MSIX_VEC;
+       bp->flags |= USING_MSIX_FLAG | ONE_SHOT_MSI_FLAG;
+       for (i = 0; i < BNX2_MAX_MSIX_VEC; i++)
+               bp->irq_tbl[i].vector = msix_ent[i].vector;
 }
 
 static void
@@ -5287,8 +5480,14 @@ bnx2_setup_int_mode(struct bnx2 *bp, int dis_msi)
 {
        bp->irq_tbl[0].handler = bnx2_interrupt;
        strcpy(bp->irq_tbl[0].name, bp->dev->name);
+       bp->irq_nvecs = 1;
+       bp->irq_tbl[0].vector = bp->pdev->irq;
+
+       if ((bp->flags & MSIX_CAP_FLAG) && !dis_msi)
+               bnx2_enable_msix(bp);
 
-       if ((bp->flags & MSI_CAP_FLAG) && !dis_msi) {
+       if ((bp->flags & MSI_CAP_FLAG) && !dis_msi &&
+           !(bp->flags & USING_MSIX_FLAG)) {
                if (pci_enable_msi(bp->pdev) == 0) {
                        bp->flags |= USING_MSI_FLAG;
                        if (CHIP_NUM(bp) == CHIP_NUM_5709) {
@@ -5296,10 +5495,10 @@ bnx2_setup_int_mode(struct bnx2 *bp, int dis_msi)
                                bp->irq_tbl[0].handler = bnx2_msi_1shot;
                        } else
                                bp->irq_tbl[0].handler = bnx2_msi;
+
+                       bp->irq_tbl[0].vector = bp->pdev->irq;
                }
        }
-
-       bp->irq_tbl[0].vector = bp->pdev->irq;
 }
 
 /* Called with rtnl_lock */
@@ -5375,9 +5574,10 @@ bnx2_open(struct net_device *dev)
                        bnx2_enable_int(bp);
                }
        }
-       if (bp->flags & USING_MSI_FLAG) {
+       if (bp->flags & USING_MSI_FLAG)
                printk(KERN_INFO PFX "%s: using MSI\n", dev->name);
-       }
+       else if (bp->flags & USING_MSIX_FLAG)
+               printk(KERN_INFO PFX "%s: using MSIX\n", dev->name);
 
        netif_start_queue(dev);
 
@@ -5441,7 +5641,7 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
        u32 len, vlan_tag_flags, last_frag, mss;
        u16 prod, ring_prod;
        int i;
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       struct bnx2_napi *bnapi = &bp->bnx2_napi[bp->tx_vec];
 
        if (unlikely(bnx2_tx_avail(bp, bnapi) <
            (skb_shinfo(skb)->nr_frags + 1))) {
@@ -5460,7 +5660,7 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
                vlan_tag_flags |= TX_BD_FLAGS_TCP_UDP_CKSUM;
        }
 
-       if (bp->vlgrp != 0 && vlan_tx_tag_present(skb)) {
+       if (bp->vlgrp && vlan_tx_tag_present(skb)) {
                vlan_tag_flags |=
                        (TX_BD_FLAGS_VLAN_TAG | (vlan_tx_tag_get(skb) << 16));
        }
@@ -6841,6 +7041,11 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
                }
        }
 
+       if (CHIP_NUM(bp) == CHIP_NUM_5709 && CHIP_REV(bp) != CHIP_REV_Ax) {
+               if (pci_find_capability(pdev, PCI_CAP_ID_MSIX))
+                       bp->flags |= MSIX_CAP_FLAG;
+       }
+
        if (CHIP_ID(bp) != CHIP_ID_5706_A0 && CHIP_ID(bp) != CHIP_ID_5706_A1) {
                if (pci_find_capability(pdev, PCI_CAP_ID_MSI))
                        bp->flags |= MSI_CAP_FLAG;
@@ -7108,13 +7313,19 @@ bnx2_bus_string(struct bnx2 *bp, char *str)
        return str;
 }
 
-static int __devinit
+static void __devinit
 bnx2_init_napi(struct bnx2 *bp)
 {
-       struct bnx2_napi *bnapi = &bp->bnx2_napi;
+       int i;
+       struct bnx2_napi *bnapi;
 
-       bnapi->bp = bp;
-       netif_napi_add(bp->dev, &bnapi->napi, bnx2_poll, 64);
+       for (i = 0; i < BNX2_MAX_MSIX_VEC; i++) {
+               bnapi = &bp->bnx2_napi[i];
+               bnapi->bp = bp;
+       }
+       netif_napi_add(bp->dev, &bp->bnx2_napi[0].napi, bnx2_poll, 64);
+       netif_napi_add(bp->dev, &bp->bnx2_napi[BNX2_TX_VEC].napi, bnx2_tx_poll,
+                      64);
 }
 
 static int __devinit