On Fri, 7 Oct 2022 at 12:54, Dave Airlie <airlied@xxxxxxxxx> wrote:
>
> On Fri, 7 Oct 2022 at 12:45, Dave Airlie <airlied@xxxxxxxxx> wrote:
> >
> > On Fri, 7 Oct 2022 at 09:45, Linus Torvalds
> > <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> > >
> > > On Thu, Oct 6, 2022 at 1:25 PM Dave Airlie <airlied@xxxxxxxxx> wrote:
> > > >
> > > >
> > > > [ 1234.778760] BUG: kernel NULL pointer dereference, address: 0000000000000088
> > > > [ 1234.778813] RIP: 0010:drm_sched_job_done.isra.0+0xc/0x140 [gpu_sched]
> > >
> > > As far as I can tell, that's the line
> > >
> > > struct drm_gpu_scheduler *sched = s_fence->sched;
> > >
> > > where 's_fence' is NULL. The code is
> > >
> > > 0: 0f 1f 44 00 00 nopl 0x0(%rax,%rax,1)
> > > 5: 41 54 push %r12
> > > 7: 55 push %rbp
> > > 8: 53 push %rbx
> > > 9: 48 89 fb mov %rdi,%rbx
> > > c:* 48 8b af 88 00 00 00 mov 0x88(%rdi),%rbp <-- trapping instruction
> > > 13: f0 ff 8d f0 00 00 00 lock decl 0xf0(%rbp)
> > > 1a: 48 8b 85 80 01 00 00 mov 0x180(%rbp),%rax
> > >
> > > and that next 'lock decl' instruction would have been the
> > >
> > > atomic_dec(&sched->hw_rq_count);
> > >
> > > at the top of drm_sched_job_done().
> > >
> > > Now, as to *why* you'd have a NULL s_fence, it would seem that
> > > drm_sched_job_cleanup() was called with an active job. Looking at that
> > > code, it does
> > >
> > > if (kref_read(&job->s_fence->finished.refcount)) {
> > > /* drm_sched_job_arm() has been called */
> > > dma_fence_put(&job->s_fence->finished);
> > > ...
> > >
> > > but then it does
> > >
> > > job->s_fence = NULL;
> > >
> > > anyway, despite the job still being active. The logic of that kind of
> > > "fake refcount" escapes me. The above looks fundamentally racy, not to
> > > say pointless and wrong (a refcount is a _count_, not a flag, so there
> > > could be multiple references to it, what says that you can just
> > > decrement one of them and say "I'm done").
> > >
> > > Now, _why_ any of that happens, I have no idea. I'm just looking at
> > > the immediate "that pointer is NULL" thing, and reacting to what looks
> > > like a completely bogus refcount pattern.
> > >
> > > But that odd refcount pattern isn't new, so it's presumably some user
> > > on the amd gpu side that changed.
> > >
> > > The problem hasn't happened again for me, but that's not saying a lot,
> > > since it was very random to begin with.
> >
> > I chased down the culprit to a drm sched patch, I'll send you a pull
> > with a revert in it.
> >
> > commit e4dc45b1848bc6bcac31eb1b4ccdd7f6718b3c86
> > Author: Arvind Yadav <Arvind.Yadav@xxxxxxx>
> > Date: Wed Sep 14 22:13:20 2022 +0530
> >
> > drm/sched: Use parent fence instead of finished
> >
> > Using the parent fence instead of the finished fence
> > to get the job status. This change is to avoid GPU
> > scheduler timeout error which can cause GPU reset.
> >
> > Signed-off-by: Arvind Yadav <Arvind.Yadav@xxxxxxx>
> > Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@xxxxxxx>
> > Link: https://patchwork.freedesktop.org/patch/msgid/20220914164321.2156-6-Arvind.Yadav@xxxxxxx
> > Signed-off-by: Christian König <christian.koenig@xxxxxxx>
> >
> > I'll let Arvind and Christian maybe work out what is going wrong there.
>
> I do spy two changes queued for -next that might be relevant, so I
> might try just pulling those instead.
>
> I'll send a PR in next hour once I test it.
Okay sent, let me know if you see any further problems.
Dave.
