Re: [PATCH 6.18 000/583] 6.18.51-rc1 review

[Sasha Levin]

> These are causing some audio regressions for Tegra ...
>
> Test failures:	tegra186-p2771-0000: tegra-audio-loopback-testsuite.sh
>                 tegra186-p2771-0000: tegra-audio-mixer-testsuite.sh
> [...]

Dropped all six Tegra "sort register defaults" patches from every branch where
they were queued:

  - "ASoC: tegra: Sort ADMAIF register defaults" -- 6.18
  - "ASoC: tegra: Fix the MIXER enable default value" -- 6.18, 6.12, 6.6, 6.1
  - "ASoC: tegra210_i2s: sort the register default table" -- 6.18, 6.12,
    6.6, 6.1, 5.15, 5.10
  - "ASoC: tegra210_i2s: sort the Tegra264 register default table" -- 6.18
  - "ASoC: tegra: Sort MBDRC register defaults" -- 6.18, 6.12, 6.6, 6.1
  - "ASoC: tegra210_mixer: sort the register default table" -- 6.18, 6.12,
    6.6, 6.1

> My intention was to only backport these for v7.2 because
> audio is broken in that release for Tegra.

Makes sense - and while digging into the I2S case specifically, we found the
root cause: your companion fix 53dd1c1de851 ("ASoC: tegra: Fix the I2S enable
default value") is queued in no stable branch. Once the I2S sort patch makes
TEGRA210_I2S_ENABLE reachable via regcache's bsearch, the wrong recorded
default is left in place with nothing to correct it or re-enable I2S at
runtime_resume.

> I see this is pulling in a bunch of other related patches, but I am not
> sure if these should be pulled in either.

We also acted on that wider concern. We reviewed the whole "sort the register
default table" family (all the non-Tegra codecs in this series: adau1761,
cx2072x, fsl_easrc, max9860, ml26124, pcm512x, pm4125-sdw, the rt* SoundWire
codecs, sgtl5000, sti-sas, tas2552/2764/ 2780/2783-sdw) and are dropping it
from 6.18 and older across the board. These patches accompany regcache commit
4b05ccb17f92 ("regcache: Sort the local copy of an unsorted reg_defaults
array"), which none of those branches have, so sorting the driver-side table is
itself what changes bsearch reachability in regcache_lookup_reg(), the same
mechanism behind the I2S regression above, but with no reported bug to fix on
these branches. Risk with no benefit, so out they go.

Thanks for flagging this.

-- 
Thanks,
Sasha