[PATCH 12/13] drm/scheduler: rework entity flush, kill and fini

[Rob Clark]

On Thu, Nov 17, 2022 at 7:12 AM Dmitry Osipenko
<dmitry.osipenko at collabora.com> wrote:
>
> On 11/17/22 18:09, Christian König wrote:
> > Am 17.11.22 um 15:41 schrieb Dmitry Osipenko:
> >> [SNIP]
> >>> drm_sched_entity_flush() should be called from the flush callback from
> >>> the file_operations structure of panfrost. See amdgpu_flush() and
> >>> amdgpu_ctx_mgr_entity_flush(). This makes sure that we wait for all
> >>> entities of the process/file descriptor to be flushed out.
> >>>
> >>> drm_sched_entity_fini() must be called before you free the memory the
> >>> entity structure or otherwise we would run into an use after free.
> >> Right, drm_sched_entity_destroy() invokes these two functions and
> >> Panfrost uses drm_sched_entity_destroy().
> >
> > Than I have no idea what's going wrong here.
> >
> > The scheduler should trivially finish with the entity and call
> > complete(&entity->entity_idle) in it's main loop. No idea why this
> > doesn't happen. Can you investigate?
>
> I'll take a closer look. Hoped you may have a quick idea of what's wrong :)
>

As Jonathan mentioned, the same thing is happening on msm.  I can
reproduce this by adding an assert in mesa (in this case, triggered
after 100 draws) and running an app under gdb.  After the assert is
hit, if I try to exit mesa, it hangs.

The problem is that we somehow call drm_sched_entity_kill() twice.
The first time completes, but now the entity_idle completion is no
longer done, so the second call hangs forever.

BR,
-R