Re: 2.6.18: hda_intel: azx_get_response timeout, switching to single_cmd mode...

[Takashi Iwai <tiwai@suse.de>]

At Tue, 26 Sep 2006 08:56:11 -0700,
Randy Dunlap wrote:
> 
> On Tue, 26 Sep 2006 17:48:33 +0200 Takashi Iwai wrote:
> 
> > At Tue, 26 Sep 2006 08:37:20 -0700,
> > Randy Dunlap wrote:
> > > 
> > > On Tue, 26 Sep 2006 12:16:33 +0200 Takashi Iwai wrote:
> > > 
> > > > At Mon, 25 Sep 2006 12:58:08 -0700,
> > > > Jeremy Fitzhardinge wrote:
> > > > > 
> > > > > I have a ThinkPad X60 which uses the Intel 82801G HDA audio chip.  This 
> > > > > used to work for me, but lately (sometime during 2.6.18-rcX series) it 
> > > > > stopped working - programs trying to use it tend to just block forever 
> > > > > waiting for /dev/dsp.
> > > > > 
> > > > > The only obvious symptom is:
> > > > > 
> > > > >     hda_intel: azx_get_response timeout, switching to single_cmd mode...
> > > > > 
> > > > > appearing in the kernel log when booting.
> > > > > 
> > > > 
> > > > There is no big change relevant to TP X60 during 2.6.18rc, so I don't
> > > > think it's a regression in the hd-audio driver code.
> > > > 
> > > > > Details attached.  The dmesg output is for the FC6 distro kernel 
> > > > > 2.6.18-1.2689.fc6PAE, but I see the same symptoms with 2.6.18-mm1.
> > > > 
> > > > You must see difference with mm1 (suppose that mm1 already includes
> > > > the latest ALSA patches).  When the CORB/RIRB interrupt gets broken,
> > > > the driver first switches to poling mode, then single_cmd mode as
> > > > fallback.
> > > > 
> > > > Also, try disable_msi=1 option for mm1.  MSI seems broken on some
> > > > systems.
> > > 
> > > is that "pci=nomsi" ?
> > 
> > No, snd-hda-intel driver has a new module option "disable_msi" to
> > disable MSI support on that driver.  As default, it's off, i.e. MSI is
> > enabled if available.  (Well, I feel it's better to rename it
> > enable_msi and set on as default...)
> > 
> > Sorry for unclear text.
> 
> ugh.  We shouldn't have drivers with such options IMO.
> I have seen/used MSI for ethernet, SATA, and audio.

Heh, you're a lucky guy :)

> It either works for all of them or none of them AFAIK.
> 
> Why do you think that it should not just be a global system option/flag?

In the ealier version, we didn't call pci_enable_msi() in the driver
while the newer version does unless you pass disable_msi=1 option.
Thus, this option means to behave like the old driver.  This makes
much easier to find out a regression than a global boot option
affecting all subsytems.


Takashi