[Intel-gfx] [CI 10/21] drm/i915: Mark up all locked waiters

[chris at chris-wilson.co.uk]

On Thu, Sep 22, 2016 at 04:05:51PM -0300, Paulo Zanoni wrote:
> Em Qua, 2016-09-14 às 09:40 +0100, chris at chris-wilson.co.uk escreveu:
> > On Tue, Sep 13, 2016 at 03:21:48PM +0300, Mika Kuoppala wrote:
> > > 
> > > Mika Kuoppala <mika.kuoppala at linux.intel.com> writes:
> > > > 
> > > > "Zanoni, Paulo R" <paulo.r.zanoni at intel.com> writes:
> > > > > 
> > > > > > 
> > > > > > +#if IS_ENABLED(CONFIG_LOCKDEP)
> > > > > > +	GEM_BUG_ON(!!lockdep_is_held(&req->i915-
> > > > > > >drm.struct_mutex)
> > > > > > !=
> > > > > > +		   !!(flags & I915_WAIT_LOCKED));
> > > > > > +#endif
> > > > > 
> > > > > I'm hitting this on my SKL. It usually happens right after the
> > > > > login,
> > > > > when I click the terminal icon on the toolbar in order to open
> > > > > it (on
> > > > > Cinnamon). When it doesn't happen at that time, I just open a
> > > > > browser
> > > > > and browse for like 30 seconds until it happens.
> > > > > 
> > > > > I did manual reverts of this series up to this patch and
> > > > > obviously the
> > > > > problem goes away (no more GEM_BUG_ON to hit). I didn't try to
> > > > > do a
> > > > > real bisect to check if this is the first bad commit or if it's
> > > > > something later on the series. It could even be an older
> > > > > problem that
> > > > > just got exposed by the new BUG(). I'm available to test
> > > > > patches, just
> > > > > send them to me.
> > > > 
> > > > No patches yet but there is a comment on 
> > > > intel_prepare_plane_fb() that says that it must be called with
> > > > mutex
> > > > held.
> > > > 
> > > > Quite likely the new GEM_BUG_ON catches this not being true.
> > > > 
> > > 
> > > As Chris pointed out in irc, its about the reverse. Waiting
> > > with mutex when you shouldn't.
> > 
> > Fwiw, the fix is now on the list.
> 
> Can you please tell us which patch it is? I see we still have this
> problem.

https://patchwork.freedesktop.org/series/12703/

Note that the problem is only shown when you enable the BUG_ON. The
impact otherwise is very low.
-Chris

-- 
Chris Wilson, Intel Open Source Technology Centre