[Intel-gfx] [PATCH v2 2/2] Test case for drm_vblank_cleanup refcount validation patch

[Daniel Vetter]

On Wed, Nov 01, 2017 at 09:48:28AM +0530, PrasannaKumar Muralidharan wrote:
> Hi Daniel,
> 
> On 31 October 2017 at 21:57, Daniel Vetter <daniel at ffwll.ch> wrote:
> > On Tue, Oct 31, 2017 at 08:37:21PM +0530, PrasannaKumar Muralidharan wrote:
> >> My patch is supposed to catch problem with drivers. It warns when
> >> vblank refcount is non-zero in drm_vblank_cleanup call. From CI log it
> >> can be seen that warning being triggered. I feel that my patch is
> >> exposing existing issue.
> >>
> >> If I misinterpreted something please let me know.
> >
> > This is probably what's happening, but I can't merge a patch that breaks
> > CI. So we need to track down that issue before merging.
> 
> I understand. Being new to DRM subsystem I don't know if I can
> contribute in finding the issue. I would be able to help if I could
> get some guidance.

If you have an intel laptop anywhere at hand that you could use to
reproduce the issue, that would be a good start.

Then go through the setup for the kms/drm validation suite:

https://dri.freedesktop.org/docs/drm/gpu/drm-uapi.html#testing-and-validation

That should allow you to locally reproduce the issue (it seems to affect
many machines, so I'd assume it doesn't matter which one you have).

Once you can reproduce it should be doable to figure out where we leak
this reference (but usually it's a bit of hard work).

Btw I discussed your patch a bit on irc, a first step would be to also
print the refcount when it's leaked, not just warn about it. Knowing how
many references are leaked is sometimes a good hint about what's going on.
-Daniel
-- 
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch