[alsa-devel] [PATCH] ALSA: hda - Force polling mode on CFL for fixing codec communication

[Takashi Iwai]

On Thu, 22 Mar 2018 08:19:51 +0100,
Hui Wang wrote:
> 
> On 2018年03月21日 21:09, Hui Wang wrote:
> > On 2018年03月21日 17:48, Takashi Iwai wrote:
> >> On Wed, 21 Mar 2018 10:39:35 +0100,
> >> Ughreja, Rakesh A wrote:
> >>>
> >>>
> >>>> -----Original Message-----
> >>>> From: alsa-devel-bounces at alsa-project.org
> >>>> [mailto:alsa-devel-bounces at alsa-
> >>>> project.org] On Behalf Of Takashi Iwai
> >>>> Sent: Wednesday, March 21, 2018 2:45 PM
> >>>> To: alsa-devel at alsa-project.org
> >>>> Cc: Hui Wang <hui.wang at canonical.com>
> >>>> Subject: [alsa-devel] [PATCH] ALSA: hda - Force polling mode on
> >>>> CFL for fixing
> >>>> codec communication
> >>>>
> >>>> We've observed too long probe time with Coffee Lake machines, and the
> >>>> likely cause is some communication problem between the HD-audio
> >>>> controller and the codec chips.  While the controller expects an IRQ
> >>>> wakeup for each codec response, it seems sometimes missing, and it
> >>>> takes one second for the controller driver to time out and read the
> >>>> response in the polling mode.
> >>>>
> >>> I think I faced similar problem while doing the HDA ASoC work.
> >>>
> >>> I don't have any way to reproduce and test this. Can we increase
> >>> the delay after the link is powered ON. The present delay is 100usec,
> >>> can we make it 521 uSec ? Possibly the codec is not ready after
> >>> the link is powered on and so it's not responding.
> >> I'm afraid that this wouldn't help since the issue appears repeatedly
> >> at later points.
> >>
> >> But still it's worth to try.  Hui, could you investigate?
> >
> > OK,  I will test it tomorrow.
> >
> > Regards,
> > Hui.
> >
> I disabled the polling_mode and changed udelay(100) to
> udelay(521)/mdelay(521),  the problem still can be reproduced.
> 
> Looks like only enabling polling_mode can fix this problem. extending
> delay after intel_ml_lctl_set_power(chip, 1); does not help fix this
> problem.

Thanks, that's what I expected.  Will include the tentative fix with
polling mode in the next pull request.


Takashi