Netdev List
 help / color / mirror / Atom feed
From: Florian Fainelli <f.fainelli@gmail.com>
To: netdev@vger.kernel.org
Cc: Florian Fainelli <f.fainelli@gmail.com>,
	Andrew Lunn <andrew+netdev@lunn.ch>,
	"David S. Miller" <davem@davemloft.net>,
	Eric Dumazet <edumazet@kernel.org>,
	Jakub Kicinski <kuba@kernel.org>, Paolo Abeni <pabeni@redhat.com>,
	Heiner Kallweit <hkallweit1@gmail.com>,
	Russell King <linux@armlinux.org.uk>,
	Jijie Shao <shaojijie@huawei.com>,
	Simon Horman <horms@kernel.org>,
	Runyu Xiao <runyu.xiao@seu.edu.cn>,
	David Yang <mmyangfl@gmail.com>,
	Herbert Valerio Riedel <hvr@gnu.org>,
	Danny Kukawka <danny.kukawka@bisect.de>,
	Mugunthan V N <mugunthanvnm@ti.com>,
	Sebastian Hesselbarth <sebastian.hesselbarth@gmail.com>,
	linux-kernel@vger.kernel.org (open list),
	Manuel Lauss <manuel.lauss@gmail.com>
Subject: [PATCH net v4 05/10] net: au1000: serialize TX completion with packet submission
Date: Fri, 20 Nov 2026 16:48:04 -0800	[thread overview]
Message-ID: <20261121004809.1690280-6-f.fainelli@gmail.com> (raw)
In-Reply-To: <20261121004809.1690280-1-f.fainelli@gmail.com>

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@gmail.com>
---
 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


  parent reply	other threads:[~2026-11-21  0:48 UTC|newest]

Thread overview: 12+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2026-11-21  0:47 [PATCH net v4 00/10] net: au1000: Bug fixes Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 01/10] net: au1000: recycle RX descriptors on allocation failure Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 02/10] net: au1000: clear all-multicast mode when restoring hash filtering Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 03/10] net: au1000: disconnect the PHY before freeing the MDIO bus Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 04/10] net: au1000: report MDIO transaction timeouts Florian Fainelli
2026-11-21  0:48 ` Florian Fainelli [this message]
2026-11-21  0:48 ` [PATCH net v4 06/10] net: au1000: restore receive filtering before enabling RX Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 07/10] net: au1000: restore RX buffer addresses when restarting the MAC Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 08/10] net: phy: smsc: wait for LAN83C183 power-down recovery Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 09/10] net: au1000: keep the MAC stopped until the PHY link is ready Florian Fainelli
2026-11-21  0:48 ` [PATCH net v4 10/10] net: au1000: reject frames larger than the TX bounce buffer Florian Fainelli
2026-11-21  0:53 ` [PATCH net v4 00/10] net: au1000: Bug fixes netdev-bot+sinfo

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=20261121004809.1690280-6-f.fainelli@gmail.com \
    --to=f.fainelli@gmail.com \
    --cc=andrew+netdev@lunn.ch \
    --cc=danny.kukawka@bisect.de \
    --cc=davem@davemloft.net \
    --cc=edumazet@kernel.org \
    --cc=hkallweit1@gmail.com \
    --cc=horms@kernel.org \
    --cc=hvr@gnu.org \
    --cc=kuba@kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=linux@armlinux.org.uk \
    --cc=manuel.lauss@gmail.com \
    --cc=mmyangfl@gmail.com \
    --cc=mugunthanvnm@ti.com \
    --cc=netdev@vger.kernel.org \
    --cc=pabeni@redhat.com \
    --cc=runyu.xiao@seu.edu.cn \
    --cc=sebastian.hesselbarth@gmail.com \
    --cc=shaojijie@huawei.com \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox