BSW KBL SKL: igt@kms_hdmi_inject@inject-audio - fail/dmesg-fail - snd_hda_codec_hdmi hdaudioC0D2: HDMI: pin NID 0x5 not registered

On intel-CI we get <4> [119.636421] snd_hda_codec_hdmi hdaudioC0D2: HDMI: pin NID 0x5 not registered
This bug is duplicate of IGT gitlab issues (3 or 152)

Making this to track for core-for-ci

added BAT_BUG feature: display/DP feature: display/audio platform: BSW/CHT platform: KBL platform: SKL labels

changed title from BSW SKL: igt@kms_hdmi_inject@inject-audio - fail/dmesg-fail - snd_hda_codec_hdmi hdaudioC0D2: HDMI: pin NID 0x5 not registered to BSW KBL SKL: igt@kms_hdmi_inject@inject-audio - fail/dmesg-fail - snd_hda_codec_hdmi hdaudioC0D2: HDMI: pin NID 0x5 not registered

Should be fixed by https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_127305v1/index.html?testfilter=audio but until done in alsa upstrteam, get this for core-for-ci.

added core-for-CI label

Let's keep this open to track real fix from Kai / Alsa. These now at https://cgit.freedesktop.org/drm-intel/log/?h=topic/core-for-CI

mentioned in issue igt-gpu-tools#3

mentioned in commit vigneshraman/linux@ef721ae9

mentioned in commit vigneshraman/linux@acd95797

mentioned in commit vigneshraman/linux@c6d91fe1

fi-kbl-guc is still showing the symptoms. Was it overlooked or is it some other issue?

Overlooked. It might be that this kbl-guc needs still some other ID. Kai is looking that.

@jani.saarinen @vsyrjala Patch sent https://lists.freedesktop.org/archives/intel-gfx/2023-December/341005.html

So the core issue in all these machines is that on these systems, the display audio codec (as programmed by BIOS), reports only one of the three pins as "externally connected". The audio driver will by default ignore these pins, assuming the connectors are not exposed in the system (not connected anywhere). But when we inspect the systems in CI that trigger the problem, in each, i915 is finding all three connectors, so in the end audio and display drivers are out of sync. When i915 reports a connection on a connector audio think is not connected, we get an error and the inject-audio check fails. What the patch does is to force all three pins to be exposed.

On modern platforms we only register the connectors the VBT claims as being present. So I guess there is some kind of mismatch what the video vs. audio side of the BIOS thinks. But the VBT is notoriously unreliable so not sure we can say anything definitive about what connectors are actually present without examining the board physically.

And one huge mystery is why on eg. fi-kbl-guc sometimes works and sometimes doesn't. I though I even saw some cases where the behaviour changed mid test run, but can't be 100% sure if that was really the case or if I just imagined it.

Thanks Kai. Link to patch https://patchwork.freedesktop.org/series/127515/

Now BAT results have https://intel-gfx-ci.01.org/tree/drm-tip/Patchwork_127515v1/index.html?

Possible fixes

• igt@kms_hdmi_inject@inject-audio:
  • fi-kbl-guc: FAIL (IGT#3) -> PASS

Is this now right?

Looks good now. The series was merged to upstream sound tree as well. I submitted the merged version to intel-gfx and CI looks still ok: https://patchwork.freedesktop.org/series/127706/

@kaivehmanenintel when this lands on upstream kernel. Next rc1 ? Or what we should do here @jani to get these alinged. Revert current ones and merge the propoer ones? Or do you diff at all?

@jani.saarinen It seems these went in even faster and patches are already in v6.7-rc6 (and cc'ed to stable).

That is good then.

mentioned in commit FireBurn/linux@924f5ca2

Update: https://lore.kernel.org/r/87le9rvfwf.fsf@intel.com

The final one dropped from core-for-CI, closing.

closed

Thanks @jani