This is a note to let you know that I've just added the patch titled
ALSA: hda: Fix unhandled register update during auto-suspend period
to the 5.4-stable tree which can be found at:
http://www.kernel.org/git/?p=linux/kernel/git/stable/stable-queue.git;a=summary
The filename of the patch is:
alsa-hda-fix-unhandled-register-update-during-auto-s.patch
and it can be found in the queue-5.4 subdirectory.
If you, or anyone else, feels it should not be added to the stable tree,
please let <stable@xxxxxxxxxxxxxxx> know about it.
commit 1de04a1e3da486e9b343ead527b40c84bd92d267
Author: Takashi Iwai <tiwai@xxxxxxx>
Date: Thu May 18 13:35:20 2023 +0200
ALSA: hda: Fix unhandled register update during auto-suspend period
[ Upstream commit 81302b1c7c997e8a56c1c2fc63a296ebeb0cd2d0 ]
It's reported that the recording started right after the driver probe
doesn't work properly, and it turned out that this is related with the
codec auto-suspend. Namely, after the probe phase, the usage count
goes zero, and the auto-suspend is programmed, but the codec is kept
still active until the auto-suspend expiration. When an application
(e.g. alsactl) updates the mixer values at this moment, the values are
cached but not actually written. Then, starting arecord thereafter
also results in the silence because of the missing unmute.
The root cause is the handling of "lazy update" mode; when a mixer
value is updated *after* the suspend, it should update only the cache
and exits. At the resume, the cached value is written to the device,
in turn. The problem is that the current code misinterprets the state
of auto-suspend as if it were already suspended.
Although we can add the check of the actual device state after
pm_runtime_get_if_in_use() for catching the missing state, this won't
suffice; the second call of regmap_update_bits_check() will skip
writing the register because the cache has been already updated by the
first call. So we'd need fixes in two different places.
OTOH, a simpler fix is to replace pm_runtime_get_if_in_use() with
pm_runtime_get_if_active() (with ign_usage_count=true). This change
implies that the driver takes the pm refcount if the device is still
in ACTIVE state and continues the processing. A small caveat is that
this will leave the auto-suspend timer. But, since the timer callback
itself checks the device state and aborts gracefully when it's active,
this won't be any substantial problem.
Long story short: we address the missing register-write problem just
by replacing the pm_runtime_*() call in snd_hda_keep_power_up().
Fixes: fc4f000bf8c0 ("ALSA: hda - Fix unexpected resume through regmap code path")
Reported-by: Amadeusz Sławiński <amadeuszx.slawinski@xxxxxxxxxxxxxxx>
Closes: https://lore.kernel.org/r/a7478636-af11-92ab-731c-9b13c582a70d@xxxxxxxxxxxxxxx
Suggested-by: Cezary Rojewski <cezary.rojewski@xxxxxxxxx>
Cc: <stable@xxxxxxxxxxxxxxx>
Link: https://lore.kernel.org/r/20230518113520.15213-1-tiwai@xxxxxxx
Signed-off-by: Takashi Iwai <tiwai@xxxxxxx>
Signed-off-by: Sasha Levin <sashal@xxxxxxxxxx>
diff --git a/sound/hda/hdac_device.c b/sound/hda/hdac_device.c
index 489f996d86bcb..9df0158e89f44 100644
--- a/sound/hda/hdac_device.c
+++ b/sound/hda/hdac_device.c
@@ -607,7 +607,7 @@ EXPORT_SYMBOL_GPL(snd_hdac_power_up_pm);
int snd_hdac_keep_power_up(struct hdac_device *codec)
{
if (!atomic_inc_not_zero(&codec->in_pm)) {
- int ret = pm_runtime_get_if_in_use(&codec->dev);
+ int ret = pm_runtime_get_if_active(&codec->dev, true);
if (!ret)
return -1;
if (ret < 0)
