Bug 103989

above is followed by this softdog:
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3409/fi-elk-e7500/igt@kms_frontbuffer_tracking@basic.html

From IRC:
"<mupuf> CI_DRM_3411 breaks fi-elk-e7500
<mupuf> *ERROR* mismatch in has_audio (expected yes, found no)
<mupuf> vsyrjala: is it from you?
<mupuf> does not look like it
<vsyrjala> caused by hdmi->dp display change on elk
* siro__1 (~siro__@ip-95-222-73-162.hsi15.unitymediagroup.de) has joined
<vsyrjala> fixes already on the list though ;)
<mupuf> marta_: Fun for you tomorrow ^
<mupuf> vsyrjala: thx :)
<vsyrjala> https://patchwork.freedesktop.org/series/34638/"

Note this issue started when DP -> HDMI cable was replaced with DP -> DP.

commit 20ff39fa4312dfaee8d1314a208e6a5a3ee51cbc
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Wed Nov 29 18:43:01 2017 +0200

    drm/i915: Disable DP audio for g4x

fixes the has_audio issue. If the underruns persist we need a new bug for them.

(In reply to Ville Syrjala from comment #4)
> commit 20ff39fa4312dfaee8d1314a208e6a5a3ee51cbc
> Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
> Date:   Wed Nov 29 18:43:01 2017 +0200
> 
>     drm/i915: Disable DP audio for g4x
> 
> fixes the has_audio issue. If the underruns persist we need a new bug for
> them.

Audio issue is wrong analyze from Martin, the issue has always started with FIFO underruns.

The pattern with FIFO underruns and then softdog on:
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3446/fi-elk-e7500/igt@kms_frontbuffer_tracking@basic.html
still persist. Please read the information on what the bug is actually about before stating that the issue is resolved.

https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3411/fi-elk-e7500/igt@debugfs_test@read_all_entries.html

is actually the first occurrence.

Here is new version of the same issue:
FIFO underrun:
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3460/fi-elk-e7500/igt@debugfs_test@read_all_entries.html
...
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3460/fi-elk-e7500/igt@gem_exec_suspend@basic-s4-devices.html

The GPU can't be re-set:
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3460/fi-elk-e7500/igt@gem_ringfill@basic-default-hang.html

then the usual FIFO underruns

and this time softdog incomplete on:
https://intel-gfx-ci.01.org/tree/drm-tip/CI_DRM_3460/fi-elk-e7500/igt@kms_pipe_crc_basic@hang-read-crc-pipe-a.html

NOTE, The fi-elk-e7500 Hardware was changed: DP monitor back to DP-HDMI dongle and monitor. Now the issue can't be reproduced. 

From IRC:
<marta_malmoe> danvet, tsa, I don't know I think there is something wrong with the igt@debugfs_test@read_all_entries where the elk thing starts. Also the SNB issue I linked above flip-flops incomplete on that test.
<danvet> would indicate a nasty bug somewhere in one of the debugfs files we're using
<danvet> might need to improve the tracing of the testcase to better understand where it fails
<ickle> it's the always the crc debugfs iirc
<danvet> the old or new ones?
<danvet> and I thought krisman had some patches for those
<danvet> or mlankhorst
<ickle> were not the old ones removed? I thought mlankhost already excised them

Last occurred 2 days, 20 hours ago according to cibuglogger

Note if bug #105225 is really holding up, then we should remove the flip_done filter from cibuglog for this one here.

A CI Bug Log filter associated to this bug has been updated:

{- fi-elk-e7500: igt@debugfs_test@read_all_entries - INFO: Timed out: reading sysfs entry -}
{+  fi-hsw-peppy fi-elk-e7500: igt@debugfs_test@read_all_entries - INFO: Timed out: reading sysfs entry +}

New failures caught by the filter:

* https://intel-gfx-ci.01.org/tree/drm-tip/drmtip_202/fi-hsw-peppy/igt@debugfs_test@read_all_entries_display_on.html

A CI Bug Log filter associated to this bug has been updated:

{- fi-elk-e7500: All tests - *ERROR* CPU pipe A FIFO underrun -}
{+ fi-elk-e7500: All tests - *ERROR* CPU pipe A FIFO underrun +}

New failures caught by the filter:

  * https://intel-gfx-ci.01.org/tree/drm-tip/drmtip_288/fi-elk-e7500/igt@kms_cursor_crc@pipe-b-cursor-64x64-sliding.html

All of CI links in this report are dead now, but my understanding is that when this bug was first reported the pattern was that watermark underruns were seen and then the machine would hang shortly thereafter, leaving tests incomplete.  Digging through the more recent CI results for this defect, it looks like the cibuglog filters may be a bit overzealous now.  There have been some kernel panics resulting in incomplete runs recently, but as far as I can see, those panics are caused by the snd_hda driver (a NULL pointer dereference) which would be more appropriately attached to https://bugzilla.kernel.org/show_bug.cgi?id=204565 (CI issue #1670).

There are also still very occasional watermark underruns at the very end of cursor tests when the fb console is restored (last instance seen was ~1 week ago), but those are non-fatal now and the system continues to run properly after the underrun is detected.  Watermark underruns by themselves can lead to brief flickering/corruption, but given how rarely they're showing up now (and that they're only showing up when the fbcon is restored) it's likely that they'd be completely unnoticed in general usage.  Dropping the priority/severity of this ticket down to medium.

-- GitLab Migration Automatic Message --

This bug has been migrated to freedesktop.org's GitLab instance and has been closed from further activity.

You can subscribe and participate further through the new bug through this link to our GitLab instance: https://gitlab.freedesktop.org/drm/intel/issues/66.