Re: [PATCH] Revert "ALSA: hda: call runtime_allow() for all hda controllers"

[Takashi Iwai <tiwai@suse.de>]

On Mon, 03 Aug 2020 19:00:30 +0200,
Pierre-Louis Bossart wrote:
> 
> 
> 
> On 8/3/20 11:36 AM, Takashi Iwai wrote:
> > On Mon, 03 Aug 2020 17:27:12 +0200,
> > Pierre-Louis Bossart wrote:
> >>
> >>
> >>
> >> On 8/3/20 1:46 AM, Hui Wang wrote:
> >>> This reverts commit 9a6418487b56 ("ALSA: hda: call runtime_allow()
> >>> for all hda controllers").
> >>>
> >>> The reverted patch already introduced some regressions on some
> >>> machines:
> >>>    - on gemini-lake machines, the error of "azx_get_response timeout"
> >>>      happens in the hda driver.
> >>>    - on the machines with alc662 codec, the audio jack detection doesn't
> >>>      work anymore.
> >>>
> >>> BugLink: https://bugzilla.kernel.org/show_bug.cgi?id=208511
> >>> Cc: <stable@vger.kernel.org>
> >>> Signed-off-by: Hui Wang <hui.wang@canonical.com>
> >>> ---
> >>>    sound/pci/hda/hda_intel.c | 1 -
> >>>    1 file changed, 1 deletion(-)
> >>>
> >>> diff --git a/sound/pci/hda/hda_intel.c b/sound/pci/hda/hda_intel.c
> >>> index e699873c8293..e34a4d5d047c 100644
> >>> --- a/sound/pci/hda/hda_intel.c
> >>> +++ b/sound/pci/hda/hda_intel.c
> >>> @@ -2352,7 +2352,6 @@ static int azx_probe_continue(struct azx *chip)
> >>>      	if (azx_has_pm_runtime(chip)) {
> >>>    		pm_runtime_use_autosuspend(&pci->dev);
> >>> -		pm_runtime_allow(&pci->dev);
> >>>    		pm_runtime_put_autosuspend(&pci->dev);
> >>>    	}
> >>
> >> Do I get this right that this permanently disables pm_runtime on all
> >> Intel HDaudio controllers?
> >
> > It just drops the unconditional enablement of runtime PM.
> > It can be enabled via sysfs, and that's the old default (let admin
> > enabling it via udev or whatever).
> 
> Sorry I am confused now.
> Kai seemed to suggest in the Bugzilla comments that this would be
> temporary, until these problems with i915 and ALC662 get fixed?

Right, that's the plan.  This patch revert to the old state before the
forced-all-enable call we've taken in 5.7.  On 5.7 and onwards, all
HD-audio controllers are enforced to use the runtime PM.  Before that
version, the runtime PM was enabled *as default* only for limited
devices (typically the ones bound with GPU); for other devices, the
runtime PM is manually enabled from user-space via sysfs (and many
distros enable them in anyway).

The forced enablement was merged with a hope that now all HD-audio
controllers behave nicely, but it turned out to cause a regression, so
it was reverted.  Once when we find out the real cause, we can flip
the flag again.


Takashi