bnx2x: don't wait for Tx completion on recovery
authorYuval Mintz <Yuval.Mintz@qlogic.com>
Sun, 13 Mar 2016 19:21:48 +0000 (21:21 +0200)
committerDavid S. Miller <davem@davemloft.net>
Wed, 16 Mar 2016 23:23:35 +0000 (19:23 -0400)
When driver has hit a parity event, HW can no longer write to host memory.
As a result, Tx completions cannot be written to the host SB memory, and
waiting for Tx completions eventually timeout.
As driver is willing to delay as much as 1-2 seconds per Tx queue for its
draining and this delay is sequential, the time to recover might greatly
lengthen needlessly in case the recovery is done under multi-connection
traffic.

Signed-off-by: Yuval Mintz <Yuval.Mintz@qlogic.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c

index a949783c8fc3bbb4c2ba1926492bf4de376ca166..0a9108cd4c4583e397cbc51d2aace18a8d96fb16 100644 (file)
@@ -3042,8 +3042,12 @@ int bnx2x_nic_unload(struct bnx2x *bp, int unload_mode, bool keep_link)
                bnx2x_save_statistics(bp);
        }
 
-       /* wait till consumers catch up with producers in all queues */
-       bnx2x_drain_tx_queues(bp);
+       /* wait till consumers catch up with producers in all queues.
+        * If we're recovering, FW can't write to host so no reason
+        * to wait for the queues to complete all Tx.
+        */
+       if (unload_mode != UNLOAD_RECOVERY)
+               bnx2x_drain_tx_queues(bp);
 
        /* if VF indicate to PF this function is going down (PF will delete sp
         * elements and clear initializations