[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <CAAywjhR4znr9fsAdBKmYAwcyP8JgoesLkuS8p9D0goJBFFePWg@mail.gmail.com>
Date: Wed, 21 May 2025 10:28:20 -0700
From: Samiullah Khawaja <skhawaja@...gle.com>
To: Wei Wang <weiwan@...gle.com>
Cc: Jakub Kicinski <kuba@...nel.org>, "David S . Miller" <davem@...emloft.net>,
Eric Dumazet <edumazet@...gle.com>, Paolo Abeni <pabeni@...hat.com>, almasrymina@...gle.com,
willemb@...gle.com, jdamato@...tly.com, mkarsten@...terloo.ca,
netdev@...r.kernel.org
Subject: Re: [PATCH net-next v2] net: stop napi kthreads when THREADED napi is disabled
On Wed, May 21, 2025 at 9:41 AM Wei Wang <weiwan@...gle.com> wrote:
>
> On Tue, May 20, 2025 at 7:09 PM Jakub Kicinski <kuba@...nel.org> wrote:
> >
> > On Mon, 19 May 2025 22:43:25 +0000 Samiullah Khawaja wrote:
> > > -/* Called with irq disabled */
> > > -static inline void ____napi_schedule(struct softnet_data *sd,
> > > - struct napi_struct *napi)
> > > +static inline bool ____try_napi_schedule_threaded(struct softnet_data *sd,
> > > + struct napi_struct *napi)
> > > {
> > > struct task_struct *thread;
> > > + unsigned long new, val;
> > >
> > > - lockdep_assert_irqs_disabled();
> > > + do {
> > > + val = READ_ONCE(napi->state);
> > > +
> > > + if (!(val & NAPIF_STATE_THREADED))
> > > + return false;
> >
> > Do we really need to complicate the fastpath to make the slowpath easy?
> >
> > Plus I'm not sure it works.
> >
> > CPU 0 (IRQ) CPU 1 (NAPI thr) CPU 2 (config)
> > if (test_bit(NAPI_STATE_SCHED_THREADED))
> > ...
> >
> > ____napi_schedule()
> > cmpxchg(...)
> > wake_up_process(thread);
> > clear_bit(NAPI_STATE_THREADED)
> > kthread_stop(thread)
> >
> > if (kthread_should_stop())
> > exit
> >
> > Right?
+1
If the kthread checks whether it was not SCHED before it died then
this should not occur.
> >
> Hmm right... I think the main issue is that while dev_set_threaded()
> clears STATE_THREADED, SCHED_THREADED could already be set meaning the
> napi is already scheduled. And napi_thread_wait() does not really
> check STATE_THREADED...
>
> > I think the shutting down thread should do this:
> >
> > while (true) {
> > state = READ_ONCE()
> >
> > // safe to clear if thread owns the NAPI,
> > // or NAPI is completely idle
> > if (state & SCHED_THREADED || !(state & SCHED)) {
This might suffer from the problem you highlighted earlier,
CPU 0 (IRQ) CPU 1 (NAPI thr) CPU 2 (config)
____napi_schedule()
if (test_bit(NAPI_STATE_THREADED))
if (thread) {
kthread_stop()
if (state & SCHED_THREADED || !(state & SCHED)) {
state &= ~THREADED;
if (try_cmp_xchg())
break
set_bit(NAPI_STATE_SCHED_THREADED)
wake_up_process(thread);
This would happen without the try_cmp_xchg logic that I added in my
patch in the __napi_schedule (in the fast path). __napi_schedule would
have to make sure that the kthread is not stopping while it is trying
to do SCHED. This is similar to the logic we have in
napi_schedule_prep that handles the STATE_DISABLE, STATE_SCHED and
STATE_MISSED scenarios. Also if it falls back to normal softirq, it
needs to make sure that the kthread is not polling at the same time.
> I think we should make sure SCHED_THREADED is cleared as well, or
> otherwise the thread is in the middle of calling napi_threaded_poll()
> and we can't just disable the thread?
+1
We need to make sure that any scheduled polling should be completed.
>
> > state &= ~THREADED;
>
> STATE_THREADED to be exact. Right?
>
> > } else {
> > msleep(1);
> > continue;
> > }
> >
> > if (try_cmpxchg())
> > break;
> > }
> >
> > But that's just an idea, it could also be wrong... :S
Powered by blists - more mailing lists