[PATCH net v4 05/10] net: au1000: serialize TX completion with packet submission

From: Florian Fainelli

Date: Fri Nov 20 2026 - 19:48:55 EST


TX completion runs in hard IRQ context while ndo_start_xmit updates the
same descriptors and ring state without locking. An interrupt between
netif_stop_queue() and setting tx_full can miss the wakeup and leave the
queue stopped until the watchdog runs. Use the existing IRQ-safe device
lock in both paths to serialize reclamation and queue state changes.
Keep the packet copy and padding outside this lock so uncached DMA
buffer writes do not extend the IRQ-disabled critical section.

When submission reclaims a completed descriptor at tx_head, advance
tx_tail and clear TX_T_DONE before dropping the lock so the interrupt
handler cannot reclaim it twice. The TX queue lock reserves the idle
descriptor while its buffer is filled; completion stops at a descriptor
without TX_T_DONE and never accesses its payload. Reacquire the device
lock to publish the length, DMA ownership and tx_head together. Consume
the copied skb after releasing the lock with dev_consume_skb_any(),
which also handles netpoll entry with hardware interrupts disabled.

Fixes: 1da177e4c3f4 ("Linux-2.6.12-rc2")
Signed-off-by: Florian Fainelli <f.fainelli@xxxxxxxxx>
---
drivers/net/ethernet/amd/au1000_eth.c | 22 ++++++++++++++++++----
1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/drivers/net/ethernet/amd/au1000_eth.c b/drivers/net/ethernet/amd/au1000_eth.c
index 1935b1d4b173..60631f0dce18 100644
--- a/drivers/net/ethernet/amd/au1000_eth.c
+++ b/drivers/net/ethernet/amd/au1000_eth.c
@@ -872,7 +872,9 @@ static void au1000_tx_ack(struct net_device *dev)
{
struct au1000_private *aup = netdev_priv(dev);
struct tx_dma *ptxd;
+ unsigned long flags;

+ spin_lock_irqsave(&aup->lock, flags);
ptxd = aup->tx_dma_ring[aup->tx_tail];

while (ptxd->buff_stat & TX_T_DONE) {
@@ -889,6 +891,8 @@ static void au1000_tx_ack(struct net_device *dev)
netif_wake_queue(dev);
}
}
+
+ spin_unlock_irqrestore(&aup->lock, flags);
}

/*
@@ -972,21 +976,27 @@ static netdev_tx_t au1000_tx(struct sk_buff *skb, struct net_device *dev)
u32 buff_stat;
struct db_dest *pDB;
int i;
+ unsigned long flags;

netif_dbg(aup, tx_queued, dev, "tx: aup %x len=%d, data=%p, head %d\n",
(unsigned)aup, skb->len,
skb->data, aup->tx_head);

+ spin_lock_irqsave(&aup->lock, flags);
ptxd = aup->tx_dma_ring[aup->tx_head];
buff_stat = ptxd->buff_stat;
if (buff_stat & TX_DMA_ENABLE) {
/* We've wrapped around and the transmitter is still busy */
netif_stop_queue(dev);
aup->tx_full = 1;
+ spin_unlock_irqrestore(&aup->lock, flags);
return NETDEV_TX_BUSY;
} else if (buff_stat & TX_T_DONE) {
au1000_update_tx_stats(dev, ptxd->status);
+ ptxd->buff_stat = buff_stat & ~TX_T_DONE;
ptxd->len = 0;
+ wmb(); /* drain writebuffer */
+ aup->tx_tail = (aup->tx_tail + 1) & (NUM_TX_DMA - 1);
}

if (aup->tx_full) {
@@ -995,22 +1005,26 @@ static netdev_tx_t au1000_tx(struct sk_buff *skb, struct net_device *dev)
}

pDB = aup->tx_db_inuse[aup->tx_head];
+ /* The TX queue lock reserves this idle descriptor until submission. */
+ spin_unlock_irqrestore(&aup->lock, flags);
+
skb_copy_from_linear_data(skb, (void *)pDB->vaddr, skb->len);
if (skb->len < ETH_ZLEN) {
for (i = skb->len; i < ETH_ZLEN; i++)
((char *)pDB->vaddr)[i] = 0;
+ }

- ptxd->len = ETH_ZLEN;
- } else
- ptxd->len = skb->len;
+ spin_lock_irqsave(&aup->lock, flags);
+ ptxd->len = max_t(unsigned int, skb->len, ETH_ZLEN);

ps->tx_packets++;
ps->tx_bytes += ptxd->len;

ptxd->buff_stat = lower_32_bits(pDB->dma_addr) | TX_DMA_ENABLE;
wmb(); /* drain writebuffer */
- dev_kfree_skb(skb);
aup->tx_head = (aup->tx_head + 1) & (NUM_TX_DMA - 1);
+ spin_unlock_irqrestore(&aup->lock, flags);
+ dev_consume_skb_any(skb);
return NETDEV_TX_OK;
}

--
2.53.0