diff options
author | Michael Chan <mchan@broadcom.com> | 2006-08-15 04:39:10 -0400 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2006-08-17 19:29:51 -0400 |
commit | 2f8af120a159a843948749ea88bcacda9779b132 (patch) | |
tree | e633037a72db856b8078e814e3451dd8f44c14de /drivers/net/bnx2.h | |
parent | fb33f82568d32016b3b3c00819574f9526e52be3 (diff) |
[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>
Diffstat (limited to 'drivers/net/bnx2.h')
-rw-r--r-- | drivers/net/bnx2.h | 12 |
1 files changed, 5 insertions, 7 deletions
diff --git a/drivers/net/bnx2.h b/drivers/net/bnx2.h index 658c5ee95c73..fe804763c607 100644 --- a/drivers/net/bnx2.h +++ b/drivers/net/bnx2.h | |||
@@ -3890,10 +3890,6 @@ struct bnx2 { | |||
3890 | u32 tx_prod_bseq __attribute__((aligned(L1_CACHE_BYTES))); | 3890 | u32 tx_prod_bseq __attribute__((aligned(L1_CACHE_BYTES))); |
3891 | u16 tx_prod; | 3891 | u16 tx_prod; |
3892 | 3892 | ||
3893 | struct tx_bd *tx_desc_ring; | ||
3894 | struct sw_bd *tx_buf_ring; | ||
3895 | int tx_ring_size; | ||
3896 | |||
3897 | u16 tx_cons __attribute__((aligned(L1_CACHE_BYTES))); | 3893 | u16 tx_cons __attribute__((aligned(L1_CACHE_BYTES))); |
3898 | u16 hw_tx_cons; | 3894 | u16 hw_tx_cons; |
3899 | 3895 | ||
@@ -3916,9 +3912,11 @@ struct bnx2 { | |||
3916 | struct sw_bd *rx_buf_ring; | 3912 | struct sw_bd *rx_buf_ring; |
3917 | struct rx_bd *rx_desc_ring[MAX_RX_RINGS]; | 3913 | struct rx_bd *rx_desc_ring[MAX_RX_RINGS]; |
3918 | 3914 | ||
3919 | /* Only used to synchronize netif_stop_queue/wake_queue when tx */ | 3915 | /* TX constants */ |
3920 | /* ring is full */ | 3916 | struct tx_bd *tx_desc_ring; |
3921 | spinlock_t tx_lock; | 3917 | struct sw_bd *tx_buf_ring; |
3918 | int tx_ring_size; | ||
3919 | u32 tx_wake_thresh; | ||
3922 | 3920 | ||
3923 | /* End of fields used in the performance code paths. */ | 3921 | /* End of fields used in the performance code paths. */ |
3924 | 3922 | ||