[BNX2]: Fix tx race condition.
Fix a subtle race condition between bnx2_start_xmit() and bnx2_tx_int()
similar to the one in tg3 discovered by Herbert Xu:
CPU0 CPU1
bnx2_start_xmit()
if (tx_ring_full) {
tx_lock
bnx2_tx()
if (!netif_queue_stopped)
netif_stop_queue()
if (!tx_ring_full)
update_tx_ring
netif_wake_queue()
tx_unlock
}
Even though tx_ring is updated before the if statement in bnx2_tx_int() in
program order, it can be re-ordered by the CPU as shown above. This
scenario can cause the tx queue to be stopped forever if bnx2_tx_int() has
just freed up the entire tx_ring. The possibility of this happening
should be very rare though.
The following changes are made, very much identical to the tg3 fix:
1. Add memory barrier to fix the above race condition.
2. Eliminate the private tx_lock altogether and rely solely on
netif_tx_lock. This eliminates one spinlock in bnx2_start_xmit()
when the ring is full.
3. Because of 2, use netif_tx_lock in bnx2_tx_int() before calling
netif_wake_queue().
4. Add memory barrier to bnx2_tx_avail().
5. Add bp->tx_wake_thresh which is set to half the tx ring size.
6. Check for the full wake queue condition before getting
netif_tx_lock in tg3_tx(). This reduces the number of unnecessary
spinlocks when the tx ring is full in a steady-state condition.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
diff --git a/drivers/net/bnx2.h b/drivers/net/bnx2.h
index 658c5ee..fe80476 100644
--- a/drivers/net/bnx2.h
+++ b/drivers/net/bnx2.h
@@ -3890,10 +3890,6 @@
u32 tx_prod_bseq __attribute__((aligned(L1_CACHE_BYTES)));
u16 tx_prod;
- struct tx_bd *tx_desc_ring;
- struct sw_bd *tx_buf_ring;
- int tx_ring_size;
-
u16 tx_cons __attribute__((aligned(L1_CACHE_BYTES)));
u16 hw_tx_cons;
@@ -3916,9 +3912,11 @@
struct sw_bd *rx_buf_ring;
struct rx_bd *rx_desc_ring[MAX_RX_RINGS];
- /* Only used to synchronize netif_stop_queue/wake_queue when tx */
- /* ring is full */
- spinlock_t tx_lock;
+ /* TX constants */
+ struct tx_bd *tx_desc_ring;
+ struct sw_bd *tx_buf_ring;
+ int tx_ring_size;
+ u32 tx_wake_thresh;
/* End of fields used in the performance code paths. */