[Intel-gfx] ✗ Fi.CI.IGT: failure for drm/i915: Program SKL+ watermarks/ddb more carefully (rev7)

[Ville Syrjälä]

On Thu, Nov 15, 2018 at 04:23:46PM +0200, Ville Syrjälä wrote:
> On Thu, Nov 15, 2018 at 05:21:46AM -0000, Patchwork wrote:
> > == Series Details ==
> > 
> > Series: drm/i915: Program SKL+ watermarks/ddb more carefully (rev7)
> > URL   : https://patchwork.freedesktop.org/series/51878/
> > State : failure
> > 
> > == Summary ==
> > 
> > = CI Bug Log - changes from CI_DRM_5140_full -> Patchwork_10827_full =
> > 
> > == Summary - FAILURE ==
> > 
> >   Serious unknown changes coming with Patchwork_10827_full absolutely need to be
> >   verified manually.
> >   
> >   If you think the reported changes have nothing to do with the changes
> >   introduced in Patchwork_10827_full, please notify your bug team to allow them
> >   to document this new failure mode, which will reduce false positives in CI.
> > 
> >   
> > 
> > == Possible new issues ==
> > 
> >   Here are the unknown changes that may have been introduced in Patchwork_10827_full:
> > 
> >   === IGT changes ===
> > 
> >     ==== Possible regressions ====
> > 
> >     igt at prime_vgem@basic-fence-flip:
> >       shard-apl:          PASS -> DMESG-WARN
> >       shard-kbl:          PASS -> DMESG-WARN
> 
> 
> <3> [72.612353] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 0 (expected e=0 b=0 l=0, got e=1 b=3 l=1)
> <3> [72.612661] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 1 (expected e=0 b=0 l=0, got e=1 b=4 l=1)
> <3> [72.612784] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 2 (expected e=0 b=0 l=0, got e=1 b=4 l=1)
> <3> [72.612905] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 3 (expected e=0 b=0 l=0, got e=1 b=6 l=2)
> <3> [72.613026] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 4 (expected e=0 b=0 l=0, got e=1 b=6 l=2)
> <3> [72.613147] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 5 (expected e=0 b=0 l=0, got e=1 b=6 l=2)
> <3> [72.613267] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 6 (expected e=0 b=0 l=0, got e=1 b=6 l=2)
> <3> [72.613388] [drm:verify_wm_state [i915]] *ERROR* mismatch in WM pipe A cursor level 7 (expected e=0 b=0 l=0, got e=1 b=6 l=2)
> 
> I suspect this would be caused by the cursor becoming fully clipped 
> but still logically enabled (fb != NULL) during the  previous test
> (kms_chv_cursor_fail). And then when it comes time to turn off the
> pipe we won't reprogram the cursor because its visibility didn't
> change and thus we won't clear its watermarks either. I think I'm
> going to write a targeted test for that just to make sure my
> analysis is correct.

Actually that can't be it I think. There was no complaint about a
non-zero DDB allocation left behind, so somehow we've got non-zero
watermarks with no DDB allocated. That doesn't quite make sense.

-- 
Ville Syrjälä
Intel