Re: [BUG] sched: WARNING in __clear_task_blocked_on()

[John Stultz]

On Wed, Jul 8, 2026 at 7:58 AM Vincent Guittot
<vincent.guittot@xxxxxxxxxx> wrote:
>
> Adding John who has a better view of the Android Common Kernel
> (android17-6.18) and proxy exex support on it
>
> On Wed, 8 Jul 2026 at 15:51, Jeongho Choi <jh1012.choi@xxxxxxxxxxx> wrote:
> >
> > Hello,
> >
> > We are seeing a reproducible kernel panic related to the sched code
> > during probe
> >
> > This issue occurs as a WARNING in the `WARN_ON_ONCE` code of the
> > `__clear_task_blocked_on()` function, which immediately triggers a panic
> > due to the `panic_on_warn` setting.
> >
> > Environment:
> >
> > Architecture: arm64
> > Kernel: Linux 6.18.21
> > Base: Android Common Kernel (android17-6.18)
> > Affected area: include/linux/sched.h
> >
> > The relevant error/panic log is:
> > [    3.107804] [7:       modprobe:  532] WARNING: CPU: 7 PID: 532 at
> > include/linux/sched.h:2259 rwsem_mark_wake+0x3d8/0x408
> >

Yeah. So just for folks who are curious, this is a reported issue with
the android17-6.18-2026-06 branch (which includes the full proxy-exec
series as of v24). A lot of change happened between v24 and v29 when a
number of changes were merged upstream in 7.1 and 7.2, which I've
since mostly sycned back to android17-6.18 tree, but after the
android17-6.18-2026-06 branch was cut.

Its likely this exact case wouldn't be hit with the 7.2 logic or the
current android17-6.18 tree, since we no longer use the PROXY_WAKING
intermediate state.

But the warning that is tripping thing up is due to us trying to wake
a rwsem waiter, and finding its blocked_on is not set to the rwsem we
that it was on the waiter list for.  This is curious, as I've not
reproduced it, so I'm hoping to better understand whats happening,  so
we can ensure the upstream blocked_on state machine doesn't hit
similar unexpected states.

So as of now I don't see any upstream or -stable impact, but I'll
follow up with any needed changes as I learn more.

thanks
-john