[igt-dev] ✗ Fi.CI.IGT: failure for lib/igt_fb: A few changes to fix test runs on ADL_P

[Imre Deak]

On Tue, May 04, 2021 at 06:13:11PM +0300, Ville Syrjälä wrote:
> On Tue, May 04, 2021 at 06:00:58PM +0300, Imre Deak wrote:
> > On Fri, Apr 30, 2021 at 12:55:28AM +0000, Patchwork wrote:
> > > == Series Details ==
> > > 
> > > Series: lib/igt_fb: A few changes to fix test runs on ADL_P
> > > URL   : https://patchwork.freedesktop.org/series/89668/
> > > State : failure
> > > 
> > > == Summary ==
> > > 
> > > CI Bug Log - changes from CI_DRM_10028_full -> IGTPW_5771_full
> > > ====================================================
> > > 
> > > Summary
> > > -------
> > > 
> > >   **FAILURE**
> > > 
> > >   Serious unknown changes coming with IGTPW_5771_full absolutely need to be
> > >   verified manually.
> > >   
> > >   If you think the reported changes have nothing to do with the changes
> > >   introduced in IGTPW_5771_full, please notify your bug team to allow them
> > >   to document this new failure mode, which will reduce false positives in CI.
> > > 
> > >   External URL: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_5771/index.html
> > > 
> > > Possible new issues
> > > -------------------
> > > 
> > >   Here are the unknown changes that may have been introduced in IGTPW_5771_full:
> > > 
> > > ### IGT changes ###
> > > 
> > > #### Possible regressions ####
> > > 
> > >   * igt at kms_plane@pixel-format-pipe-a-planes:
> > >     - shard-tglb:         [PASS][1] -> [SKIP][2]
> > >    [1]: https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_10028/shard-tglb3/igt@kms_plane@pixel-format-pipe-a-planes.html
> > >    [2]: https://intel-gfx-ci.01.org/tree/drm-tip/IGTPW_5771/shard-tglb7/igt@kms_plane@pixel-format-pipe-a-planes.html
> > 
> > Test requirement not met in function test_format_plane, file ../tests/kms_plane.c:912:
> > Test requirement: num_unique_crcs(ref_crc[PLANAR_CRC_SET], data->num_colors) > 1
> > 
> > after a bunch of CRC mismatches, so I suppose the CRC generation gets
> > stuck at one point.
> > 
> > It's on the same path this patchset changes, however on TGL the patchset
> > should have no functional change: the stride calculated by
> > __intel_buf_init() and the stride passed to it by
> > igt_fb_create_intel_buf() always match based on my tests. After testing
> > the same CI testlist locally on 2 TGLs, I also rerun the testlist where
> > this problem didn't happen:
> 
> There's definitely something hinky going on with tgl since kms_big_fb
> is also failing atm in ci. But dunno if that's related or not.

That seems to get triggered by kms_psr2_sf followed by kms_big_fb. I
could repro this only (consistently) on TGL/A0 (all TGL shards are like
this) either running kms_psr2_sf/plane-move-sf-dmg-area-1 or
kms_psr2_sf/overlay-primary-update-sf-dmg-area-4 followed by
kms_big_fb/x-tiled-64bpp-rotate-180 .

The same doesn't happen on TGL/B0 with the same panel as used by by TGL
shards, though PSR got disabled on B0 due to some PSR RFB error.

Since this problem seems to happen consistently on A0, affecting later
tests I suggest blacklisting kms_psr2_sf at least on those machines,
until the problem is root caused, or at least to see how the kms_big_fb
results change.

> -- 
> Ville Syrjälä
> Intel