failed to get afg sub nodes on ChromeOS devices

[kv2019i]

Multiple sightings have been reported of failure to probe the HDMI codec in i915 at boot, specifically on ChromeOS devices and a recent 5.4/5.5 kernel.

Error logs:
[ 79.756783] snd_hda_codec_hdmi ehdaudio0D2: HDMI: failed to get afg sub nodes
[ 79.756800] snd_hda_codec_hdmi ehdaudio0D2: patch failed -22
[ 79.756862] snd_hda_codec_hdmi ehdaudio0D2: ASoC: failed to probe
component -22
[ 79.756930] bxt_da7219_max98357a glk_da7219_max98357a: ASoC: failed
to instantiate card -22
...
[ 79.769445] bxt_da7219_max98357a: probe of glk_da7219_max98357a failed with error -22

Fixes to #1001 , #1183 and #1374 do not help as e.g. 5.5-rc1 has all the fixes included and the error still happens.

[kv2019i]

FYI @yongzhi1 and @cujomalainey
I've still not been able to reproduce this myself on any device (including a GLK device running ChromeOS, but will continue to try ith this one). As it seems we have to gather sightings and data from a larger set of devices, filing a bug for tracking.

[kv2019i]

@yongzhi1 and @cujomalainey -- I've now reproduced the issue on a GLK Chromebook using a kernel module load-unload stress test.

Flow-wise there is nothing wrong in the sequence, but I did notice a slight difference between legacy-HDA and SOF drivers. If I add an extra GET_PARAMETERS HDA verb to the SOF flow (right after power-up) to imitate what happens with legacy HDA driver, the errors are gone. I'll continue investigating (why this flow change helps and how to turn it into a upstream-acceptable fix).

[kv2019i]

I uploaded a potential bugfix at:
kv2019i@4685d86
https://github.com/kv2019i/linux/tree/topic/cras/l5.4-glk-hdmi-afg-fix

[cujomalainey]

Been trying to bisect this but failures go back pretty far. Your branch by itself is fine.

[kv2019i]

It seems the above fix does not work on all Bobba devices. I posted a 2nd experimental patch to the above branch:
kv2019i@b3c8d21
https://github.com/kv2019i/linux/tree/topic/cras/l5.4-glk-hdmi-afg-fix
... waiting to gather more data whether this helps. Still no signs of this issue in CI or from other reporters.

[cujomalainey]

@kv2019i probe works with that second patch and I get devices 😺

I cherry-picked over the top 2 commits to sof-dev and I have playback again. Thanks so much.

[kv2019i]

@cujomalainey wrote

@kv2019i probe works with that second patch and I get devices 😺

I cherry-picked over the top 2 commits to sof-dev and I have playback again. Thanks so much.

Excellent! 😎 I'll start working out an upstream-worthy patchset. 2nd patch should be easy, but first one may require some more effort to get through.