* [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues
@ 2026-11-16 9:51 Tommaso Merciai
2026-11-16 9:57 ` netdev-bot+sinfo
2026-11-19 21:52 ` netdev-bot+sashiko
0 siblings, 2 replies; 3+ messages in thread
From: Tommaso Merciai @ 2026-11-16 9:51 UTC (permalink / raw)
To: Maxime Chevallier, Andrew Lunn, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni, Maxime Coquelin, Alexandre Torgue,
Giuseppe Cavallaro
Cc: netdev, linux-stm32, linux-arm-kernel, linux-kernel,
linux-renesas-soc, biju.das.jz, tomm.merciai, dinghui,
dinghui1111, Tommaso Merciai
stmmac_suspend() stops the Tx queues in netif_device_detach() and
disables the Tx NAPI afterwards in stmmac_disable_all_queues(). A Tx
completion in flight can wake a queue from stmmac_tx_clean() in between,
so a later stmmac_xmit() writes the Tx tail pointer after the platform
suspend callback has gated the clocks.
Call netif_tx_disable() once NAPI is disabled, as done for bcmgenet in
commit 7e87508b5c4d ("net: bcmgenet: stop Tx NAPI before disabling the
queues").
SError Interrupt on CPU2, code 0x00000000be000011 -- SError
stmmac_tx_timer_arm+0x20/0xa4 (P)
stmmac_xmit+0xd4c/0x1650
dev_hard_start_xmit+0x8c/0x104
sch_direct_xmit+0x8c/0x35c
__dev_queue_xmit+0x808/0xf2c
Fixes: 47dd7a540b8a ("net: add support for STMicroelectronics Ethernet controllers.")
Signed-off-by: Tommaso Merciai <tommaso.merciai.xr@bp.renesas.com>
---
drivers/net/ethernet/stmicro/stmmac/stmmac_main.c | 3 +++
1 file changed, 3 insertions(+)
diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
index 550a0bd647e1..7dac41b05cfa 100644
--- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
+++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
@@ -8525,6 +8525,9 @@ int stmmac_suspend(struct device *dev)
stmmac_disable_all_queues(priv);
+ /* NAPI may have woken a TX queue after netif_device_detach() */
+ netif_tx_disable(ndev);
+
for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
hrtimer_cancel(&priv->dma_conf.tx_queue[chan].txtimer);
---
base-commit: aa40afa831ebfe45e40a23721711059f2ac584d4
change-id: 20261116-tm-stmmac-eth-s2idle-07df298cf01f
prerequisite-message-id: <cover.1794646293.git.dinghui@lixiang.com>
prerequisite-patch-id: ab0e628f44dfcf5418f268d42521ee9cbb65f167
prerequisite-patch-id: 2eba4552886f22cc596278c6dae42bc85b5771c4
prerequisite-patch-id: 523947458d15820f92f68f7c33766ed5142502c9
prerequisite-patch-id: 0aac69a82a0325455096278b2cf4a9bae70862d6
Best regards,
--
Tommaso Merciai <tommaso.merciai.xr@bp.renesas.com>
^ permalink raw reply related [flat|nested] 3+ messages in thread
* Re: [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues
2026-11-16 9:51 [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues Tommaso Merciai
@ 2026-11-16 9:57 ` netdev-bot+sinfo
2026-11-19 21:52 ` netdev-bot+sashiko
1 sibling, 0 replies; 3+ messages in thread
From: netdev-bot+sinfo @ 2026-11-16 9:57 UTC (permalink / raw)
To: Tommaso Merciai
Cc: Maxime Chevallier, Andrew Lunn, David S. Miller, Eric Dumazet,
Jakub Kicinski, Paolo Abeni, Maxime Coquelin, Alexandre Torgue,
Giuseppe Cavallaro, netdev, linux-stm32, linux-arm-kernel,
linux-kernel, linux-renesas-soc, biju.das.jz, tomm.merciai,
dinghui, dinghui1111
Hi!
This is an automated message. This series looks like a fix, but its
commit messages seem to be missing some information:
- What hardware the change was tested on. For driver fixes please
mention the device (and if relevant firmware version) used for
testing, or say that the change was not tested on real hardware.
Please do not repost the series just to address the above. Instead,
reply to this email with the missing information, so that reviewers
can take it into account.
*IMPORTANT*: If you do repost (for other reasons) - please include
the relevant info in the cover letter or commit messages.
--
The evaluation is done by an LLM so it may be wrong, if you think
that is the case please reply and explain.
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues
2026-11-16 9:51 [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues Tommaso Merciai
2026-11-16 9:57 ` netdev-bot+sinfo
@ 2026-11-19 21:52 ` netdev-bot+sashiko
1 sibling, 0 replies; 3+ messages in thread
From: netdev-bot+sashiko @ 2026-11-19 21:52 UTC (permalink / raw)
To: tommaso.merciai.xr
Cc: maxime.chevallier, andrew+netdev, davem, edumazet, kuba, pabeni,
mcoquelin.stm32, alexandre.torgue, peppe.cavallaro, netdev,
linux-stm32, linux-arm-kernel, linux-kernel, linux-renesas-soc,
biju.das.jz, tomm.merciai, dinghui, dinghui1111
Thank you for your contribution! Sashiko AI review found 2 potential
issue(s) to consider.
Critical: 0 · High: 2 (1) · Medium: 0 · Low: 0
If you believe none of the issues are legitimate, please reply to each
issue of Medium severity or higher and explain why the AI is wrong.
If any of them are legitimate and you plan to respin, please let patchwork
know by including "pw-bot: cr" as a separate line at the end of your reply
(one such reply per series is enough).
- [High] The fix only fences the ndo_start_xmit producer.
- [High] The patch assumes NAPI is the last thing that can wake a TX
queue once stmmac_disable_all_queues() returns.
Pre-existing issues:
- [High] This problem wasn't introduced by this patch, but
stmmac_xdp_release() quiesces TX in the order the patch's new comment…
--
net: stmmac: stop Tx NAPI before disabling the queues
This commit adds a netif_tx_disable() call to stmmac_suspend(), right
after stmmac_disable_all_queues(). A TX queue woken by stmmac_tx_clean()
between netif_device_detach() and napi_disable() is then stopped again
before the platform suspend callback gates the clocks. It follows an
earlier bcmgenet fix.
> diff --git a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> index 25894771e9e9f..41d62f3b7ce51 100644
> --- a/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> +++ b/drivers/net/ethernet/stmicro/stmmac/stmmac_main.c
> @@ -8412,6 +8412,9 @@ int stmmac_suspend(struct device *dev)
>
> stmmac_disable_all_queues(priv);
>
> + /* NAPI may have woken a TX queue after netif_device_detach() */
> + netif_tx_disable(ndev);
> +
[Severity: High]
Does this also cover the XDP transmit path? stmmac_xdp_xmit() is the
ndo_xdp_xmit callback, and its only gate is the STMMAC_DOWN bit:
stmmac_xdp_xmit() {
if (unlikely(test_bit(STMMAC_DOWN, &priv->state)))
return -ENETDOWN;
...
__netif_tx_lock(nq, cpu);
...
if (flags & XDP_XMIT_FLUSH) {
stmmac_flush_tx_descriptors(priv, queue);
stmmac_tx_timer_arm(priv, queue);
}
__netif_tx_unlock(nq);
...
}
stmmac_suspend() never sets STMMAC_DOWN. stmmac_xdp_xmit() also doesn't
check netif_xmit_stopped() or netif_device_present().
So netif_tx_disable() only waits for an XDP transmit that already holds
the txq lock. It doesn't stop new ones from starting.
The core redirect path doesn't block this either. xdp_ok_fwd_dev() only
checks IFF_UP and the MTU, and netif_device_detach() leaves IFF_UP set.
Once an XDP program is attached, stmmac_xdp_set_prog() advertises the
device as a redirect target through xdp_features_set_redirect_target().
Suppose another interface redirects into this one through a devmap, for
example a NIC that suspends later in dpm order, or a veth. Could this
sequence then run after stmmac_stop_all_dma() and priv->plat->suspend()?
bq_xmit_all()
stmmac_xdp_xmit()
stmmac_xdp_xmit_xdpf()
stmmac_enable_dma_transmission() <- MMIO
stmmac_flush_tx_descriptors()
stmmac_set_queue_tx_tail_ptr() <- MMIO
stmmac_tx_timer_arm() <- re-arms cancelled txtimer
That looks like the same SError the commit message describes: a tail
pointer write after the clocks are gated.
Is a driver-level state needed here that every TX entry point checks?
One option is to set STMMAC_DOWN (or a dedicated suspended bit) before
stmmac_stop_all_dma() and clear it on resume. Another is to check
netif_device_present() in stmmac_xdp_xmit().
[Severity: High]
The new comment assumes that once stmmac_disable_all_queues() returns,
nothing but NAPI can wake a TX queue. Is that true? stmmac_tx_err() also
wakes the queue, from hardirq context:
stmmac_tx_err() {
netif_tx_stop_queue(netdev_get_tx_queue(priv->dev, chan));
stmmac_stop_tx_dma(priv, chan);
...
stmmac_start_tx_dma(priv, chan);
priv->xstats.tx_errors++;
netif_tx_wake_queue(netdev_get_tx_queue(priv->dev, chan));
}
It is called through stmmac_interrupt()->stmmac_dma_interrupt() and
through stmmac_msi_intr_tx(). Both handlers return early only on
STMMAC_DOWN, which stmmac_suspend() never sets.
The channel IRQs also stay enabled until
stmmac_stop_all_dma()->stmmac_deinit_chan() masks them.
Say a tx_hard_error status is handled after this netif_tx_disable().
TX DMA is restarted and the txq is woken. Can the qdisc then call
stmmac_xmit() after priv->plat->suspend() has gated the clocks?
On dwmac4, stmmac_stop_all_dma() may open this window by itself.
DMA_CHAN_INTR_ABNORMAL enables RPS, so stopping RX DMA can raise AIS.
dwmac4_dma_interrupt() then decodes TPS, which is set when
stmmac_stop_tx_dma() clears ST, as tx_hard_error:
dwmac4_dma_interrupt() {
...
if (unlikely(intr_status & DMA_CHAN_STATUS_TPS)) {
x->tx_process_stopped_irq++;
ret = tx_hard_error;
}
...
}
A fatal bus error during the suspend window would take the same path.
Should the DMA interrupts be masked, or a down state set, before the
queues are disabled here?
> for (chan = 0; chan < priv->plat->tx_queues_to_use; chan++)
> hrtimer_cancel(&priv->dma_conf.tx_queue[chan].txtimer);
>
[Severity: High]
This isn't a bug introduced by this patch, but stmmac_xdp_release() in
the same file still uses the ordering this patch fixes in
stmmac_suspend():
stmmac_xdp_release() {
/* Ensure tx function is not running */
netif_tx_disable(dev);
/* Disable NAPI process */
stmmac_disable_all_queues(priv);
...
stmmac_free_irq(dev, REQ_IRQ_ERR_ALL, 0);
...
stmmac_stop_all_dma(priv);
...
free_dma_desc_resources(priv, &priv->dma_conf);
...
}
A tx NAPI poll that runs before napi_disable() completes can reach
stmmac_tx_clean(). There it finds the queue stopped and
stmmac_tx_avail() > STMMAC_TX_THRESH, so it calls netif_tx_wake_queue(),
which also reschedules the qdisc.
stmmac_xmit() can then run, holding only the txq lock, while
stmmac_xdp_release() carries on into free_dma_desc_resources(). Could
stmmac_xmit() write descriptors and tx_skbuff entries into ring memory
that is being freed?
This can be hit by attaching or detaching an XDP program on a running
interface while TX traffic is flowing, through
stmmac_xdp_set_prog()->stmmac_xdp_release().
__stmmac_release(), and stmmac_suspend() after this patch, both disable
NAPI before calling netif_tx_disable(). Should stmmac_xdp_release() use
the same order?
--
Sashiko AI review · https://netdev-ai.bots.linux.dev/sashiko/#/patchset/20261116-tm-stmmac-eth-s2idle-v1-1-bbe165b743b5%40bp.renesas.com
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2026-11-19 21:52 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-16 9:51 [PATCH net] net: stmmac: stop Tx NAPI before disabling the queues Tommaso Merciai
2026-11-16 9:57 ` netdev-bot+sinfo
2026-11-19 21:52 ` netdev-bot+sashiko
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox