bnx2x: fix panic when TX ring is full

There is a off by one error in the minimal number of BD in
bnx2x_start_xmit() and bnx2x_tx_int() before stopping/resuming tx queue.

A full size GSO packet, with data included in skb->head really needs
(MAX_SKB_FRAGS + 4) BDs, because of bnx2x_tx_split()

This error triggers if BQL is disabled and heavy TCP transmit traffic
occurs.

bnx2x_tx_split() definitely can be called, remove a wrong comment.

Reported-by: Tomas Hruby <thruby@google.com>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Eilon Greenstein <eilong@broadcom.com>
Cc: Yaniv Rosner <yanivr@broadcom.com>
Cc: Merav Sicron <meravs@broadcom.com>
Cc: Tom Herbert <therbert@google.com>
Cc: Robert Evans <evansr@google.com>
Cc: Willem de Bruijn <willemb@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
index cbc56f274e0cd33298007c7cb457884ad9831580..8098eea9704df6ffea4b904ab173ce88ef6f1016 100644 (file)
--- a/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
+++ b/drivers/net/ethernet/broadcom/bnx2x/bnx2x_cmn.c
@@ -190,7 +190,7 @@ int bnx2x_tx_int(struct bnx2x *bp, struct bnx2x_fp_txdata *txdata)
 
                if ((netif_tx_queue_stopped(txq)) &&
                    (bp->state == BNX2X_STATE_OPEN) &&
-                   (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3))
+                   (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 4))
                        netif_tx_wake_queue(txq);
 
                __netif_tx_unlock(txq);
@@ -2516,8 +2516,6 @@ int bnx2x_poll(struct napi_struct *napi, int budget)
 /* we split the first BD into headers and data BDs
  * to ease the pain of our fellow microcode engineers
  * we use one mapping for both BDs
- * So far this has only been observed to happen
- * in Other Operating Systems(TM)
  */
 static noinline u16 bnx2x_tx_split(struct bnx2x *bp,
                                   struct bnx2x_fp_txdata *txdata,
@@ -3171,7 +3169,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
 
        txdata->tx_bd_prod += nbd;
 
-       if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 3)) {
+       if (unlikely(bnx2x_tx_avail(bp, txdata) < MAX_SKB_FRAGS + 4)) {
                netif_tx_stop_queue(txq);
 
                /* paired memory barrier is in bnx2x_tx_int(), we have to keep
@@ -3180,7 +3178,7 @@ netdev_tx_t bnx2x_start_xmit(struct sk_buff *skb, struct net_device *dev)
                smp_mb();
 
                fp->eth_q_stats.driver_xoff++;
-               if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 3)
+               if (bnx2x_tx_avail(bp, txdata) >= MAX_SKB_FRAGS + 4)
                        netif_tx_wake_queue(txq);
        }
        txdata->tx_pkt++;