[Intel-gfx] i915 issues with 4.6

[Ville Syrjälä]

On Tue, May 17, 2016 at 07:09:58PM -0500, Pierre-Louis Bossart wrote:
> 
> 
> On 05/17/2016 01:16 PM, Ville Syrjälä wrote:
> > On Tue, May 17, 2016 at 01:00:13PM -0500, Pierre-Louis Bossart wrote:
> >> Hi,
> >> I was porting the HDMI audio patch to 4.6 as a service to several folks
> >> who need them before we finally upstream the reworked code, and I notice
> >> that the UI would freeze within 2s to 5mn, with all sorts of GPU hangs
> >> and issues reported in dmesg and /sys/calls/drm/card0/error
> >> Things work fine with the latest drm-intel-nightly branch from last night
> >> Known issue or should I file a bug?
> > One GPU hang generally looks like any other from the symptom POV.
> > Without the error state it's pretty much impossible to speculate.
> > Just filing a new bug is usually the best policy.
> >
> > If it's indeed a kernel regression a bisect would also be helpful.
> > Or in this case perhaps a reverse bisect since you say nightly is
> > fine, so we might want to know what fixed it.
> 
> bug filed at https://bugs.freedesktop.org/show_bug.cgi?id=95467
> 
> V4.5 works fine
> v4.6 is broken in less than 20s when moving windows around (can still 
> get to virtual console)
> drm-intel-nightly works fine
> 
> git bisect says this commit fixes the issue:
> 
> commit a5e485a95c9c4cdd93b4c6dc53eee3bd1e50de11
> Author: Ville Syrjälä <ville.syrjala at linux.intel.com <mailto:ville.syrjala at linux.intel.com>>
> Date:   Wed Apr 13 21:19:51 2016 +0300
> 
>      drm/i915: Clear VLV_IER around irq processing

Argh. So somehow the broken commit [1] got into 4.6. We'll need to get
it reverted.

[1] 9dbaab56ac09 ("drm/i915: Exit cherryview_irq_handler() after one pass")

-- 
Ville Syrjälä
Intel OTC