diff mbox series

net: Handle napi_schedule() calls from non-interrupt

Message ID 20250221173009.21742-1-frederic@kernel.org (mailing list archive)
State New
Delegated to: Netdev Maintainers
Headers show
Series net: Handle napi_schedule() calls from non-interrupt | expand

Commit Message

Frederic Weisbecker Feb. 21, 2025, 5:30 p.m. UTC
napi_schedule() is expected to be called either:

* From an interrupt, where raised softirqs are handled on IRQ exit

* From a softirq disabled section, where raised softirqs are handled on
  the next call to local_bh_enable().

* From a softirq handler, where raised softirqs are handled on the next
  round in do_softirq(), or further deferred to a dedicated kthread.

Other bare tasks context may end up ignoring the raised NET_RX vector
until the next random softirq handling opportunity, which may not
happen before a while if the CPU goes idle afterwards with the tick
stopped.

Such "misuses" have been detected on several places thanks to messages
of the kind:

	"NOHZ tick-stop error: local softirq work is pending, handler #08!!!"

Chasing each and every misuse can be a long journey given the amount of
existing callers. Fixing them can also prove challenging if the caller
may be called from different kind of context.

Therefore fix this from napi_schedule() itself with waking up ksoftirqd
when softirqs are raised from task contexts.

Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
Closes: 354a2690-9bbf-4ccb-8769-fa94707a9340@molgen.mpg.de
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
---
 net/core/dev.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

Comments

Joe Damato Feb. 21, 2025, 5:59 p.m. UTC | #1
On Fri, Feb 21, 2025 at 06:30:09PM +0100, Frederic Weisbecker wrote:
> napi_schedule() is expected to be called either:
> 
> * From an interrupt, where raised softirqs are handled on IRQ exit
> 
> * From a softirq disabled section, where raised softirqs are handled on
>   the next call to local_bh_enable().
> 
> * From a softirq handler, where raised softirqs are handled on the next
>   round in do_softirq(), or further deferred to a dedicated kthread.
> 
> Other bare tasks context may end up ignoring the raised NET_RX vector
> until the next random softirq handling opportunity, which may not
> happen before a while if the CPU goes idle afterwards with the tick
> stopped.
> 
> Such "misuses" have been detected on several places thanks to messages
> of the kind:
> 
> 	"NOHZ tick-stop error: local softirq work is pending, handler #08!!!"

Might be helpful to include the stack trace of the offender you did
find which led to this change?

> Chasing each and every misuse can be a long journey given the amount of
> existing callers. Fixing them can also prove challenging if the caller
> may be called from different kind of context.

Any way to estimate how many misuses there are with coccinelle or
similar to get a grasp on the scope?

Based on the scope of the problem it might be better to fix the
known offenders and add a WARN_ON_ONCE or something instead of the
proposed change? Not sure, but having more information might help
make that determination.

> Therefore fix this from napi_schedule() itself with waking up ksoftirqd
> when softirqs are raised from task contexts.
> 
> Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
> Closes: 354a2690-9bbf-4ccb-8769-fa94707a9340@molgen.mpg.de

AFAIU, Closes tags should point to URLs not message IDs.

If this is a fix, the subject line should be:
   [PATCH net]

And there should be a Fixes tag referencing the SHA which caused the
issue and the patch should CC stable.

See:

https://www.kernel.org/doc/html/v6.13/process/maintainer-netdev.html#netdev-faq
Frederic Weisbecker Feb. 21, 2025, 10:12 p.m. UTC | #2
Le Fri, Feb 21, 2025 at 12:59:26PM -0500, Joe Damato a écrit :
> On Fri, Feb 21, 2025 at 06:30:09PM +0100, Frederic Weisbecker wrote:
> > napi_schedule() is expected to be called either:
> > 
> > * From an interrupt, where raised softirqs are handled on IRQ exit
> > 
> > * From a softirq disabled section, where raised softirqs are handled on
> >   the next call to local_bh_enable().
> > 
> > * From a softirq handler, where raised softirqs are handled on the next
> >   round in do_softirq(), or further deferred to a dedicated kthread.
> > 
> > Other bare tasks context may end up ignoring the raised NET_RX vector
> > until the next random softirq handling opportunity, which may not
> > happen before a while if the CPU goes idle afterwards with the tick
> > stopped.
> > 
> > Such "misuses" have been detected on several places thanks to messages
> > of the kind:
> > 
> > 	"NOHZ tick-stop error: local softirq work is pending, handler #08!!!"
> 
> Might be helpful to include the stack trace of the offender you did
> find which led to this change?

There are several of them. Here is one example:

	__raise_softirq_irqoff
	__napi_schedule
	rtl8152_runtime_resume.isra.0
	rtl8152_resume
	usb_resume_interface.isra.0
	usb_resume_both
	__rpm_callback
	rpm_callback
	rpm_resume
	__pm_runtime_resume
	usb_autoresume_device
	usb_remote_wakeup
	hub_event
	process_one_work
	worker_thread
	kthread
	ret_from_fork
	ret_from_fork_asm

There is also drivers/net/usb/r8152.c::rtl_work_func_t

And also netdevsim:
https://lore.kernel.org/netdev/20250219-netdevsim-v3-1-811e2b8abc4c@debian.org/

And probably others...

> 
> > Chasing each and every misuse can be a long journey given the amount of
> > existing callers. Fixing them can also prove challenging if the caller
> > may be called from different kind of context.
> 
> Any way to estimate how many misuses there are with coccinelle or
> similar to get a grasp on the scope?

I don't think Coccinelle can find them all. The best it can do is to find direct
calls to napi_schedule() from a workqueue or kthread handler.

I proposed a runtime detection here:

  https://lore.kernel.org/lkml/20250212174329.53793-2-frederic@kernel.org/

But I plan to actually introduce a more generic detection in
__raise_softirq_irqsoff() itself instead.
 
> Based on the scope of the problem it might be better to fix the
> known offenders and add a WARN_ON_ONCE or something instead of the
> proposed change? Not sure, but having more information might help
> make that determination.

Well, based on the fix proposal I see here:
https://lore.kernel.org/netdev/20250219-netdevsim-v3-1-811e2b8abc4c@debian.org/

I think that fixing this on the caller level can be very error prone
and involve nasty workarounds.

Oh you just made me look at the past:

  019edd01d174 ("ath10k: sdio: Add missing BH locking around napi_schdule()")
  330068589389 ("idpf: disable local BH when scheduling napi for marker packets")
  e3d5d70cb483 ("net: lan78xx: fix "softirq work is pending" error")
  e55c27ed9ccf ("mt76: mt7615: add missing bh-disable around rx napi schedule")
  c0182aa98570 ("mt76: mt7915: add missing bh-disable around tx napi enable/schedule")
  970be1dff26d ("mt76: disable BH around napi_schedule() calls")
  019edd01d174 ("ath10k: sdio: Add missing BH locking around napi_schdule()")
  30bfec4fec59 ("can: rx-offload: can_rx_offload_threaded_irq_finish(): add new  function to be called from threaded interrupt")
  e63052a5dd3c ("mlx5e: add add missing BH locking around napi_schdule()")
  83a0c6e58901 ("i40e: Invoke softirqs after napi_reschedule")
  bd4ce941c8d5 ("mlx4: Invoke softirqs after napi_reschedule")
  8cf699ec849f ("mlx4: do not call napi_schedule() without care")
  ec13ee80145c ("virtio_net: invoke softirqs after __napi_schedule")

I think this just shows how successful it has been to leave the responsibility to the
caller so far.

And also note that these issues are reported for years sometimes firsthand to us
in the timer subsystem because this is the place where we detect entering in idle
with softirqs pending.

> 
> > Therefore fix this from napi_schedule() itself with waking up ksoftirqd
> > when softirqs are raised from task contexts.
> > 
> > Reported-by: Paul Menzel <pmenzel@molgen.mpg.de>
> > Closes: 354a2690-9bbf-4ccb-8769-fa94707a9340@molgen.mpg.de
> 
> AFAIU, Closes tags should point to URLs not message IDs.

Good point!

> 
> If this is a fix, the subject line should be:
>    [PATCH net]

Ok.

> 
> And there should be a Fixes tag referencing the SHA which caused the
> issue and the patch should CC stable.

At least since bea3348eef27 ("[NET]: Make NAPI polling independent of struct
net_device objects."). It's hard for me to be sure it's not older.


> 
> See:
> 
> https://www.kernel.org/doc/html/v6.13/process/maintainer-netdev.html#netdev-faq

Thanks.
diff mbox series

Patch

diff --git a/net/core/dev.c b/net/core/dev.c
index c0021cbd28fc..2419cc558a64 100644
--- a/net/core/dev.c
+++ b/net/core/dev.c
@@ -4692,7 +4692,7 @@  static inline void ____napi_schedule(struct softnet_data *sd,
 	 * we have to raise NET_RX_SOFTIRQ.
 	 */
 	if (!sd->in_net_rx_action)
-		__raise_softirq_irqoff(NET_RX_SOFTIRQ);
+		raise_softirq_irqoff(NET_RX_SOFTIRQ);
 }
 
 #ifdef CONFIG_RPS