Takashi Iwai wrote:
> At Sun, 06 May 2012 12:32:14 +0400,
> Konstantin Khlebnikov wrote:
>>
>> Konstantin Khlebnikov wrote:
>>> Mainline commit 785f857d1cb0856b612b46a0545b74aa2596e44a
>>> ("ALSA: hda - Set codec to D3 forcibly even if not used")
>>> breaks sound for my notebook (thinkpad x220),
>>> looks like codec cannot wake from power-save state after resume.
>>> After reverting looks like all works fine.
>>>
>>> how to reproduce:
>>> * unplug the power cord (to enable hda power safe mode)
>>> * play something and stop
>>> * wait until the background noise disappears (codec went into power-save mode)
>>> * suspend
>>> * resume
>>> * play something again and hear the silence
>>> --
>>> To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
>>> the body of a message to majordomo@vger.kernel.org
>>> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>>> Please read the FAQ at  http://www.tux.org/lkml/
>>
>> Bump
>
> It's a weekend, so don't expect quick responces from any normal person
> :)

Yeah, sorry. In this country a mess with holidays and weekends.

>
> Which 3.4-rc did you try?  There was a known problem with this codec
> regarding the power-up sequence, but this should have been fixed in
> the very recent version.

Bug still in latest v3.4-rc6.
This is serious degradation since v3.3, so I hope it will be fixed before release.

>
> If the latest Linus tree still shows the problem, try sound git tree
> master or for-next branch.
>     git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound.git

I don't see there anything related to this problem. Bug is still there.
It's 100% reproductible and disappears after reverting that commit.
Looks like driver missed one power-up cycle after system-resume,
after second power-save down/up cycle all work fine.

>
> Then attach g output (run with --no-upload option) before
> and after S3 (i.e. working and non-working states).

see attachments (kernel is v3.4-rc6 with changes from your branch)

start mplayer (work fine)
alsa-info.sh -> alsa-info.before
stop mplayer
suspend
resume
start mplayer (doesn't work)
alsa-info.sh -> alsa-info.after
stop mplayer
wait a second
start mplayer (work fine)

>
>
>> BTW, I don't see any locks around hda power-save code, for example  hda_power_work() reads
>> codec->power_on and codec->power_count without any locks. This seems racy.
>
> Yeah, we may add an extra spinlock.  But it's not too critical for the
> whole operation, so I didn't care much so far.  At most, you'll miss one
> power-save cycle.  Maybe a good time to fix it now for 3.5.

I'm afraid this code can easily miss one power-up cycle and stays in power-save mode.
This can be unrelated to my problem, but from my point of view this is very careless.
For example snd_hda_power_up() can easily race with hda_power_work().

>
>
> thanks,
>
> Takashi