Re: [BUG/REGRESSION] Screen flickering

[Matt Roper <matthew.d.roper@intel.com>]

On Tue, May 12, 2015 at 09:54:07PM +0200, Thomas Gummerer wrote:
> 
> Hi,
> 
> I noticed that on my machine the screen starts to flicker after I
> suspend and resume my machine, on the main laptop display if an external
> display is attached with kernel v4.1-rc1.  I tracked the regression down
> to commit c9f038a1a592 ("drm/i915: Don't assume primary & cursor are
> always on for wm calculation (v4)"), and sent a patch that fixes that
> behavior at [1] about two weeks ago, although I'm not sure it's the
> right thing to do, as I'm not very familiar with the code.  The same bug
> still exists in vv4.1-rc3.
> 
> Jan Niehusmann confirmed the behavior, but there has been no further
> discussion on the topic.  I also forgot to cc the people that were
> involved in the patch that caused the regression (sorry about that).
> 
> Is there anything else that I can do to help fixing this issue?
> 
> Thanks,
> Thomas
> 
> [1] http://lists.freedesktop.org/archives/intel-gfx/2015-April/065494.html

Sorry, I missed your patch when you first sent it.  That type of fix
looks like an okay workaround while we do a more in-depth rework of the
watermark system.  However I think your patch could cause a crash if we
disable the primary plane via the universal plane interface; if we do
that, p->pri.bytes_per_pixel is set to 0, but since we're now pretending
the primary plane is always enabled, ilk_wm_fbc() can eventually get
called and use that 0 in the denominator of a division operation.

If you just squash the following change into your patch, I think it should be
safe:

        diff --git a/drivers/gpu/drm/i915/intel_pm.c b/drivers/gpu/drm/i915/intel_pm.c
        index eb97cbe..99fa8ee 100644
        --- a/drivers/gpu/drm/i915/intel_pm.c
        +++ b/drivers/gpu/drm/i915/intel_pm.c
        @@ -2076,7 +2076,7 @@ static void ilk_compute_wm_parameters(struct drm_crtc *crtc,
                        p->pri.bytes_per_pixel =
                                crtc->primary->state->fb->bits_per_pixel / 8;
                else
        -               p->pri.bytes_per_pixel = 0;
        +               p->pri.bytes_per_pixel = 4;
         
                p->cur.bytes_per_pixel = 4;
                /*


Matt

-- 
Matt Roper
Graphics Software Engineer
IoTG Platform Enabling & Development
Intel Corporation
(916) 356-2795