Commit e89bbf10 authored by Michael Chan's avatar Michael Chan Committed by David S. Miller

[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: default avatarMichael Chan <mchan@broadcom.com>
Signed-off-by: default avatarDavid S. Miller <davem@davemloft.net>
parent cd339a0e
...@@ -107,6 +107,15 @@ static struct flash_spec flash_table[] = ...@@ -107,6 +107,15 @@ static struct flash_spec flash_table[] =
MODULE_DEVICE_TABLE(pci, bnx2_pci_tbl); 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 static u32
bnx2_reg_rd_ind(struct bnx2 *bp, u32 offset) bnx2_reg_rd_ind(struct bnx2 *bp, u32 offset)
{ {
...@@ -1338,22 +1347,19 @@ bnx2_tx_int(struct bnx2 *bp) ...@@ -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))) { if (unlikely(netif_queue_stopped(bp->dev))) {
unsigned long flags; unsigned long flags;
spin_lock_irqsave(&bp->tx_lock, flags); spin_lock_irqsave(&bp->tx_lock, flags);
if ((netif_queue_stopped(bp->dev)) && 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); netif_wake_queue(bp->dev);
} }
spin_unlock_irqrestore(&bp->tx_lock, flags); spin_unlock_irqrestore(&bp->tx_lock, flags);
} }
bp->tx_cons = sw_cons;
} }
static inline void static inline void
...@@ -2971,7 +2977,6 @@ bnx2_init_tx_ring(struct bnx2 *bp) ...@@ -2971,7 +2977,6 @@ bnx2_init_tx_ring(struct bnx2 *bp)
bp->tx_prod = 0; bp->tx_prod = 0;
bp->tx_cons = 0; bp->tx_cons = 0;
bp->tx_prod_bseq = 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_TYPE_L2;
val |= BNX2_L2CTX_TYPE_SIZE_L2; val |= BNX2_L2CTX_TYPE_SIZE_L2;
...@@ -4057,9 +4062,7 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev) ...@@ -4057,9 +4062,7 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
u16 prod, ring_prod; u16 prod, ring_prod;
int i; int i;
if (unlikely(atomic_read(&bp->tx_avail_bd) < if (unlikely(bnx2_tx_avail(bp) < (skb_shinfo(skb)->nr_frags + 1))) {
(skb_shinfo(skb)->nr_frags + 1))) {
netif_stop_queue(dev); netif_stop_queue(dev);
printk(KERN_ERR PFX "%s: BUG! Tx ring full when queue awake!\n", printk(KERN_ERR PFX "%s: BUG! Tx ring full when queue awake!\n",
dev->name); dev->name);
...@@ -4156,8 +4159,6 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev) ...@@ -4156,8 +4159,6 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
prod = NEXT_TX_BD(prod); prod = NEXT_TX_BD(prod);
bp->tx_prod_bseq += skb->len; 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_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); 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) ...@@ -4166,16 +4167,14 @@ bnx2_start_xmit(struct sk_buff *skb, struct net_device *dev)
bp->tx_prod = prod; bp->tx_prod = prod;
dev->trans_start = jiffies; 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; unsigned long flags;
spin_lock_irqsave(&bp->tx_lock, flags); spin_lock_irqsave(&bp->tx_lock, flags);
if (atomic_read(&bp->tx_avail_bd) <= MAX_SKB_FRAGS) {
netif_stop_queue(dev); netif_stop_queue(dev);
if (atomic_read(&bp->tx_avail_bd) > MAX_SKB_FRAGS) if (bnx2_tx_avail(bp) > MAX_SKB_FRAGS)
netif_wake_queue(dev); netif_wake_queue(dev);
}
spin_unlock_irqrestore(&bp->tx_lock, flags); spin_unlock_irqrestore(&bp->tx_lock, flags);
} }
......
...@@ -3841,12 +3841,12 @@ struct bnx2 { ...@@ -3841,12 +3841,12 @@ struct bnx2 {
struct status_block *status_blk; struct status_block *status_blk;
u32 last_status_idx; u32 last_status_idx;
atomic_t tx_avail_bd;
struct tx_bd *tx_desc_ring; struct tx_bd *tx_desc_ring;
struct sw_bd *tx_buf_ring; struct sw_bd *tx_buf_ring;
u32 tx_prod_bseq; u32 tx_prod_bseq;
u16 tx_prod; u16 tx_prod;
u16 tx_cons; u16 tx_cons;
int tx_ring_size;
#ifdef BCM_VLAN #ifdef BCM_VLAN
struct vlan_group *vlgrp; struct vlan_group *vlgrp;
...@@ -3929,7 +3929,6 @@ struct bnx2 { ...@@ -3929,7 +3929,6 @@ struct bnx2 {
u16 fw_wr_seq; u16 fw_wr_seq;
u16 fw_drv_pulse_wr_seq; u16 fw_drv_pulse_wr_seq;
int tx_ring_size;
dma_addr_t tx_desc_mapping; dma_addr_t tx_desc_mapping;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment