1 files changed, 25 insertions, 24 deletions

diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c
index db73de0d2511..652eb05a6c2d 100644
--- a/drivers/net/bnx2.c
+++ b/drivers/net/bnx2.c
@@ -56,8 +56,8 @@
 
 #define DRV_MODULE_NAME         "bnx2"
 #define PFX DRV_MODULE_NAME     ": "
-#define DRV_MODULE_VERSION      "1.4.43"
-#define DRV_MODULE_RELDATE      "June 28, 2006"
+#define DRV_MODULE_VERSION      "1.4.44"
+#define DRV_MODULE_RELDATE      "August 10, 2006"
 
 #define RUN_AT(x) (jiffies + (x))
 
@@ -209,8 +209,10 @@ MODULE_DEVICE_TABLE(pci, bnx2_pci_tbl);
 
 static inline u32 bnx2_tx_avail(struct bnx2 *bp)
 {
-        u32 diff = TX_RING_IDX(bp->tx_prod) - TX_RING_IDX(bp->tx_cons);
+        u32 diff;
 
+        smp_mb();
+        diff = TX_RING_IDX(bp->tx_prod) - TX_RING_IDX(bp->tx_cons);
         if (diff > MAX_TX_DESC_CNT)
                 diff = (diff & MAX_TX_DESC_CNT) - 1;
         return (bp->tx_ring_size - diff);
@@ -1569,7 +1571,7 @@ bnx2_alloc_rx_skb(struct bnx2 *bp, u16 index)
         struct rx_bd *rxbd = &bp->rx_desc_ring[RX_RING(index)][RX_IDX(index)];
         unsigned long align;
 
-        skb = dev_alloc_skb(bp->rx_buf_size);
+        skb = netdev_alloc_skb(bp->dev, bp->rx_buf_size);
         if (skb == NULL) {
                 return -ENOMEM;
         }
@@ -1578,7 +1580,6 @@ bnx2_alloc_rx_skb(struct bnx2 *bp, u16 index)
                 skb_reserve(skb, 8 - align);
         }
 
-        skb->dev = bp->dev;
         mapping = pci_map_single(bp->pdev, skb->data, bp->rx_buf_use_size,
                 PCI_DMA_FROMDEVICE);
 
@@ -1686,15 +1687,20 @@ bnx2_tx_int(struct bnx2 *bp)
         }
 
         bp->tx_cons = sw_cons;
+        /* Need to make the tx_cons update visible to bnx2_start_xmit()
+         * before checking for netif_queue_stopped().  Without the
+         * memory barrier, there is a small possibility that bnx2_start_xmit()
+         * will miss it and cause the queue to be stopped forever.
+         */
+        smp_mb();
 
-        if (unlikely(netif_queue_stopped(bp->dev))) {
-                spin_lock(&bp->tx_lock);
+        if (unlikely(netif_queue_stopped(bp->dev)) &&
+                     (bnx2_tx_avail(bp) > bp->tx_wake_thresh)) {
+                netif_tx_lock(bp->dev);
                 if ((netif_queue_stopped(bp->dev)) &&
-                    (bnx2_tx_avail(bp) > MAX_SKB_FRAGS)) {
-
+                    (bnx2_tx_avail(bp) > bp->tx_wake_thresh))
                         netif_wake_queue(bp->dev);
-                }
-                spin_unlock(&bp->tx_lock);
+                netif_tx_unlock(bp->dev);
         }
 }
 
@@ -1786,7 +1792,7 @@ bnx2_rx_int(struct bnx2 *bp, int budget)
                 if ((bp->dev->mtu > 1500) && (len <= RX_COPY_THRESH)) {
                         struct sk_buff *new_skb;
 
-                        new_skb = dev_alloc_skb(len + 2);
+                        new_skb = netdev_alloc_skb(bp->dev, len + 2);
                         if (new_skb == NULL)
                                 goto reuse_rx;
 
@@ -1797,7 +1803,6 @@ bnx2_rx_int(struct bnx2 *bp, int budget)
 
                         skb_reserve(new_skb, 2);
                         skb_put(new_skb, len);
-                        new_skb->dev = bp->dev;
 
                         bnx2_reuse_rx_skb(bp, skb,
                                 sw_ring_cons, sw_ring_prod);
@@ -3503,6 +3508,8 @@ bnx2_init_tx_ring(struct bnx2 *bp)
         struct tx_bd *txbd;
         u32 val;
 
+        bp->tx_wake_thresh = bp->tx_ring_size / 2;
+
         txbd = &bp->tx_desc_ring[MAX_TX_DESC_CNT];
                 
         txbd->tx_bd_haddr_hi = (u64) bp->tx_desc_mapping >> 32;
@@ -3952,7 +3959,7 @@ bnx2_run_loopback(struct bnx2 *bp, int loopback_mode)
                 return -EINVAL;
 
         pkt_size = 1514;
-        skb = dev_alloc_skb(pkt_size);
+        skb = netdev_alloc_skb(bp->dev, pkt_size);
         if (!skb)
                 return -ENOMEM;
         packet = skb_put(skb, pkt_size);
@@ -4390,10 +4397,8 @@ bnx2_vlan_rx_kill_vid(struct net_device *dev, uint16_t vid)
 #endif
 
 /* Called with netif_tx_lock.
- * hard_start_xmit is pseudo-lockless - a lock is only required when
- * the tx queue is full. This way, we get the benefit of lockless
- * operations most of the time without the complexities to handle
- * netif_stop_queue/wake_queue race conditions.
+ * bnx2_tx_int() runs without netif_tx_lock unless it needs to call
+ * netif_wake_queue().
  */
 static int
 bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
@@ -4512,12 +4517,9 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
         dev->trans_start = jiffies;
 
         if (unlikely(bnx2_tx_avail(bp) <= MAX_SKB_FRAGS)) {
-                spin_lock(&bp->tx_lock);
                 netif_stop_queue(dev);
-                
-                if (bnx2_tx_avail(bp) > MAX_SKB_FRAGS)
+                if (bnx2_tx_avail(bp) > bp->tx_wake_thresh)
                         netif_wake_queue(dev);
-                spin_unlock(&bp->tx_lock);
         }
 
         return NETDEV_TX_OK;
@@ -5628,7 +5630,6 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
         bp->pdev = pdev;
 
         spin_lock_init(&bp->phy_lock);
-        spin_lock_init(&bp->tx_lock);
         INIT_WORK(&bp->reset_task, bnx2_reset_task, bp);
 
         dev->base_addr = dev->mem_start = pci_resource_start(pdev, 0);
@@ -5751,7 +5752,7 @@ bnx2_init_board(struct pci_dev *pdev, struct net_device *dev)
         bp->mac_addr[5] = (u8) reg;
 
         bp->tx_ring_size = MAX_TX_DESC_CNT;
-        bnx2_set_rx_ring_size(bp, 100);
+        bnx2_set_rx_ring_size(bp, 255);
 
         bp->rx_csum = 1;