Re: [syzbot] [wireless?] WARNING: net/mac80211/tx.c:LINE at __ieee80211_beacon_get, CPU: syz.NUM.NUM/NUM

[Aleksandr Nogikh]

On Tue, Jun 17, 2025 at 11:50 AM Johannes Berg
<johannes@xxxxxxxxxxxxxxxx> wrote:
>
> On Tue, 2025-06-17 at 11:48 +0200, Aleksandr Nogikh wrote:
> > On Tue, Jun 17, 2025 at 11:43 AM Johannes Berg
> > <johannes@xxxxxxxxxxxxxxxx> wrote:
> > >
> > > On Tue, 2025-06-17 at 11:34 +0200, Aleksandr Nogikh wrote:
> > > > #syz dup: WARNING in __ieee80211_beacon_get
> > > >
> > >
> > > Not just this one :)
> > >
> > > https://lore.kernel.org/linux-wireless/20250617104902.146e10919be1.I85f352ca4a2dce6f556e5ff45ceaa5f3769cb5ce@changeid/
> > >
> >
> > Ah, interesting :)
> >
> > FWIW, in this particular case, syzbot sent the duplicate report
> > because the WARNING format has somewhat changed in the latest
> > linux-next. So before we updated syzbot's parsing rules, it had
> > managed to re-report quite a few duplicates.
>
> Right, I had noticed that, but then I looked and the old counter is
> already at well over 100k so I decided to finally look at it again ;-)

Nice! So re-reporting things is not always that bad ;)
Thanks for looking into the problem!

>
> This is a really long-standing problem that we discussed a few times in
> the past I think, and basically the system is loaded enough that the
> hwsim hrtimer can fire on time and pull the beacon, but the workqueues
> are overloaded and cannot do the necessary work within the ~100ms beacon
> interval ...
>
> Should be rare in practice, but a WARN_ON() that doesn't say anything
> about what's going on doesn't help anyway.

That sounds totally reasonable indeed.

>
> johannes