diff options
author | Michael Chan <mchan@broadcom.com> | 2005-08-26 00:36:58 +0200 |
---|---|---|
committer | David S. Miller <davem@sunset.davemloft.net> | 2005-08-30 01:10:38 +0200 |
commit | e89bbf1049aac3625fdafe3657ed8d7d5373d351 (patch) | |
tree | 15e0fc01d553ce5bc38de546d80f5f2e3736b0bf /drivers/net/bnx2.c | |
parent | [BNX2]: speedup serdes linkup (diff) | |
download | linux-e89bbf1049aac3625fdafe3657ed8d7d5373d351.tar.xz linux-e89bbf1049aac3625fdafe3657ed8d7d5373d351.zip |
[BNX2]: remove atomics in tx
Remove atomic operations in the fast tx path. Expensive atomic
operations were used to keep track of the number of available tx
descriptors. The new code uses the difference between the consumer
and producer index to determine the number of free tx descriptors.
As suggested by Jeff Garzik, the name of the inline function is
changed to all lower case.
Signed-off-by: Michael Chan <mchan@broadcom.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Diffstat (limited to 'drivers/net/bnx2.c')
-rw-r--r-- | drivers/net/bnx2.c | 35 |
1 files changed, 17 insertions, 18 deletions
diff --git a/drivers/net/bnx2.c b/drivers/net/bnx2.c index 015ff7906601..da903b3ebfb0 100644 --- a/drivers/net/bnx2.c +++ b/drivers/net/bnx2.c @@ -107,6 +107,15 @@ static struct flash_spec flash_table[] = MODULE_DEVICE_TABLE(pci, bnx2_pci_tbl); +static inline u32 bnx2_tx_avail(struct bnx2 *bp) +{ + u32 diff = TX_RING_IDX(bp->tx_prod) - TX_RING_IDX(bp->tx_cons); + + if (diff > MAX_TX_DESC_CNT) + diff = (diff & MAX_TX_DESC_CNT) - 1; + return (bp->tx_ring_size - diff); +} + static u32 bnx2_reg_rd_ind(struct bnx2 *bp, u32 offset) { @@ -1338,22 +1347,19 @@ bnx2_tx_int(struct bnx2 *bp) } } - atomic_add(tx_free_bd, &bp->tx_avail_bd); + bp->tx_cons = sw_cons; if (unlikely(netif_queue_stopped(bp->dev))) { unsigned long flags; spin_lock_irqsave(&bp->tx_lock, flags); if ((netif_queue_stopped(bp->dev)) && - (atomic_read(&bp->tx_avail_bd) > MAX_SKB_FRAGS)) { + (bnx2_tx_avail(bp) > MAX_SKB_FRAGS)) { netif_wake_queue(bp->dev); } spin_unlock_irqrestore(&bp->tx_lock, flags); } - - bp->tx_cons = sw_cons; - } static inline void @@ -2971,7 +2977,6 @@ bnx2_init_tx_ring(struct bnx2 *bp) bp->tx_prod = 0; bp->tx_cons = 0; bp->tx_prod_bseq = 0; - atomic_set(&bp->tx_avail_bd, bp->tx_ring_size); val = BNX2_L2CTX_TYPE_TYPE_L2; val |= BNX2_L2CTX_TYPE_SIZE_L2; @@ -4057,9 +4062,7 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev) u16 prod, ring_prod; int i; - if (unlikely(atomic_read(&bp->tx_avail_bd) < - (skb_shinfo(skb)->nr_frags + 1))) { - + if (unlikely(bnx2_tx_avail(bp) < (skb_shinfo(skb)->nr_frags + 1))) { netif_stop_queue(dev); printk(KERN_ERR PFX "%s: BUG! Tx ring full when queue awake!\n", dev->name); @@ -4156,8 +4159,6 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev) prod = NEXT_TX_BD(prod); bp->tx_prod_bseq += skb->len; - atomic_sub(last_frag + 1, &bp->tx_avail_bd); - REG_WR16(bp, MB_TX_CID_ADDR + BNX2_L2CTX_TX_HOST_BIDX, prod); REG_WR(bp, MB_TX_CID_ADDR + BNX2_L2CTX_TX_HOST_BSEQ, bp->tx_prod_bseq); @@ -4166,16 +4167,14 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev) bp->tx_prod = prod; dev->trans_start = jiffies; - if (unlikely(atomic_read(&bp->tx_avail_bd) <= MAX_SKB_FRAGS)) { + if (unlikely(bnx2_tx_avail(bp) <= MAX_SKB_FRAGS)) { unsigned long flags; spin_lock_irqsave(&bp->tx_lock, flags); - if (atomic_read(&bp->tx_avail_bd) <= MAX_SKB_FRAGS) { - netif_stop_queue(dev); - - if (atomic_read(&bp->tx_avail_bd) > MAX_SKB_FRAGS) - netif_wake_queue(dev); - } + netif_stop_queue(dev); + + if (bnx2_tx_avail(bp) > MAX_SKB_FRAGS) + netif_wake_queue(dev); spin_unlock_irqrestore(&bp->tx_lock, flags); } |