e1000: Redo netpoll fix to address community concerns
authorAuke Kok <auke\-jan.h.kok@intel.com>
Fri, 14 Jul 2006 23:14:23 +0000 (16:14 -0700)
committerAuke Kok <juke-jan.h.kok@intel.com>
Fri, 14 Jul 2006 23:14:23 +0000 (16:14 -0700)
The original suggested fix for netpoll was found to be racy on SMP
kernels. While it is highly unlikely that this race would ever be seen
in the real world due to current netpoll usage models, we implemented
this updated fix to address concerns.

Signed-off-by: Mitch Williams <mitch.a.williams@intel.com>
Signed-off-by: Auke Kok <auke-jan.h.kok@intel.com>
drivers/net/e1000/e1000_main.c

index 6d3d41934503811332657ad6ee27e02f394e8c5a..1c6bcad5b910d050b153f0ea755ee300fd29bbf9 100644 (file)
@@ -3387,8 +3387,8 @@ e1000_intr(int irq, void *data, struct pt_regs *regs)
                E1000_WRITE_REG(hw, IMC, ~0);
                E1000_WRITE_FLUSH(hw);
        }
-       if (likely(netif_rx_schedule_prep(&adapter->polling_netdev[0])))
-               __netif_rx_schedule(&adapter->polling_netdev[0]);
+       if (likely(netif_rx_schedule_prep(netdev)))
+               __netif_rx_schedule(netdev);
        else
                e1000_irq_enable(adapter);
 #else
@@ -3431,34 +3431,26 @@ e1000_clean(struct net_device *poll_dev, int *budget)
 {
        struct e1000_adapter *adapter;
        int work_to_do = min(*budget, poll_dev->quota);
-       int tx_cleaned = 0, i = 0, work_done = 0;
+       int tx_cleaned = 0, work_done = 0;
 
        /* Must NOT use netdev_priv macro here. */
        adapter = poll_dev->priv;
 
        /* Keep link state information with original netdev */
-       if (!netif_carrier_ok(adapter->netdev))
+       if (!netif_carrier_ok(poll_dev))
                goto quit_polling;
 
-       while (poll_dev != &adapter->polling_netdev[i]) {
-               i++;
-               BUG_ON(i == adapter->num_rx_queues);
+       /* e1000_clean is called per-cpu.  This lock protects
+        * tx_ring[0] from being cleaned by multiple cpus
+        * simultaneously.  A failure obtaining the lock means
+        * tx_ring[0] is currently being cleaned anyway. */
+       if (spin_trylock(&adapter->tx_queue_lock)) {
+               tx_cleaned = e1000_clean_tx_irq(adapter,
+                                               &adapter->tx_ring[0]);
+               spin_unlock(&adapter->tx_queue_lock);
        }
 
-       if (likely(adapter->num_tx_queues == 1)) {
-               /* e1000_clean is called per-cpu.  This lock protects
-                * tx_ring[0] from being cleaned by multiple cpus
-                * simultaneously.  A failure obtaining the lock means
-                * tx_ring[0] is currently being cleaned anyway. */
-               if (spin_trylock(&adapter->tx_queue_lock)) {
-                       tx_cleaned = e1000_clean_tx_irq(adapter,
-                                                       &adapter->tx_ring[0]);
-                       spin_unlock(&adapter->tx_queue_lock);
-               }
-       } else
-               tx_cleaned = e1000_clean_tx_irq(adapter, &adapter->tx_ring[i]);
-
-       adapter->clean_rx(adapter, &adapter->rx_ring[i],
+       adapter->clean_rx(adapter, &adapter->rx_ring[0],
                          &work_done, work_to_do);
 
        *budget -= work_done;
@@ -3466,7 +3458,7 @@ e1000_clean(struct net_device *poll_dev, int *budget)
 
        /* If no Tx and not enough Rx work done, exit the polling mode */
        if ((!tx_cleaned && (work_done == 0)) ||
-          !netif_running(adapter->netdev)) {
+          !netif_running(poll_dev)) {
 quit_polling:
                netif_rx_complete(poll_dev);
                e1000_irq_enable(adapter);
@@ -4752,6 +4744,7 @@ static void
 e1000_netpoll(struct net_device *netdev)
 {
        struct e1000_adapter *adapter = netdev_priv(netdev);
+
        disable_irq(adapter->pdev->irq);
        e1000_intr(adapter->pdev->irq, netdev, NULL);
        e1000_clean_tx_irq(adapter, adapter->tx_ring);