[Intel-gfx] [PATCH] drm/i915: Fix ilk watermarks calculation when primary plane is disabled

[Josh Boyer]

On Thu, May 21, 2015 at 3:21 AM, Ander Conselvan De Oliveira
<conselvan2 at gmail.com> wrote:
> On Wed, 2015-05-20 at 16:53 +0300, Jani Nikula wrote:
>> On Wed, 20 May 2015, Ander Conselvan de Oliveira <ander.conselvan.de.oliveira at intel.com> wrote:
>> > On Fedora 21 or 22, when the transition from the X server to the wayland
>> > compositor is done, the CRTC with the login screen is left active with a
>> > disabled fb. A cursor ioctl after the transition causes the watermarks
>> > to be updated, but due to the logic in intel_crtc_active() checking for
>> > the primary plane fb, the update considers all planes to be disabled,
>> > untimately setting the wrong watermark values and causing screen
>> > flicker. Since the crtc is active, a modeset done later is skipped and
>> > replaced with a flip, which doesn't update the watermarks.
>> >
>> > This regression was introduced somewhere between v3.16 and v3.17.
>> > Another issue prevented me from doing a full bisect, but the issue was
>> > introduced in one of the following skipped commits:
>> >
>> >     commit 7707e6535f43328e05e4729ac96eee864b90e8a4
>> >     Author: Rob Clark <robdclark at gmail.com>
>> >     Date:   Thu Jul 17 23:30:04 2014 -0400
>> >
>> >         drm/i915: use helpers
>> >
>> >     commit ca5a1b9ba0fb5291b555a23b76dbe5f6c30bfd7a
>> >     Merge: c7dbc6c 3488229
>> >     Author: Dave Airlie <airlied at redhat.com>
>> >     Date:   Wed Jul 9 10:38:42 2014 +1000
>> >
>> >         Merge tag 'drm-intel-next-2014-06-20' of git://anongit.freedesktop.org/drm-intel into drm-next
>> >
>> >     commit c51f71679042a5f388d9580ffbede14c897f1e86
>> >     Merge: b957f45 7b3c29f
>> >     Author: Dave Airlie <airlied at redhat.com>
>> >     Date:   Sat Jul 19 16:43:41 2014 +1000
>> >
>> >         Merge tag 'drm-intel-next-2014-07-11' of git://anongit.freedesktop.org/drm-intel into drm-next
>> >
>> > This patch is a simplified version of the following commits:
>> >
>> >     commit 3dd512fbda0d87d1c3fb44bf878b262baee98fb6
>> >     Author: Matt Roper <matthew.d.roper at intel.com>
>> >     Date:   Fri Feb 27 10:12:00 2015 -0800
>> >
>> >         drm/i915: Kill intel_crtc->cursor_{width, height} (v2)
>> >
>> >     commit 54da691deb123c045259ebf4f5c67381244f58f1
>> >     Author: Thomas Gummerer <t.gummerer at gmail.com>
>> >     Date:   Thu May 14 09:16:39 2015 +0200
>> >
>> >         drm/i915: fix screen flickering
>>
>> This is expected to land in v4.1-rc5, i.e. these are all upstream
>> commits. And I assume none of them apply to stable kernels directly.
>
> A lot of cherry-pick and some amending would be necessary, because of
> the changes going on for the atomic conversion. Hence the approach here
> was to capture just the important changes into this small patch.
>
>> >
>> >     commit 3ef00284e6a48f7deb0784ccca0478ebb7d4bcfc
>> >     Author: Matt Roper <matthew.d.roper at intel.com>
>> >     Date:   Mon Mar 9 10:19:24 2015 -0700
>> >
>> >         drm/i915: Use crtc->state->active in ilk/skl watermark calculations (v3)
>> >
>> > Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=90508
>> > Bugzilla: https://bugzilla.redhat.com/show_bug.cgi?id=1218688
>> > Cc: stable at vger.kernel.org
>>
>> I (and most likely the stable team) would like to know which stable
>> kernel versions this is targeting.
>
> This is for 4.0.
>
>> Do you have Tested-bys against the stable versions you're targeting?
>
> Only my own testing so far. I guess Ray Strode's doesn't count since he
> tested on top of Fedora's kernel?

Fedora's F22 kernel is based on 4.0.4 right now.  We only have two
minor patches on top of i915 that we carry.  One converts a warning to
a debug, the other disables the verbose state checks by default.
Neither patch would impact this functionality.  I think Ray's testing
counts, but it's up to upstream I guess.

josh