Re: [REGRESSION][BISECTED] Audio volume issues since 4178d78cd7a8

[Takashi Iwai]

On Fri, 04 Oct 2024 11:30:22 +0200,
Takashi Iwai wrote:
> 
> On Fri, 04 Oct 2024 11:25:32 +0200,
> Christoffer Sandberg wrote:
> > 
> > 
> > 
> > On 4.10.2024 10:25, Takashi Iwai wrote:
> > > On Fri, 04 Oct 2024 10:18:10 +0200,
> > > Christoffer Sandberg wrote:
> > >> 
> > >> 
> > >> 
> > >> On 2.10.2024 23:28, Jerry Luo wrote:
> > >> > On 10/2/24 10:00 AM, Takashi Iwai wrote:
> > >> >> On Wed, 02 Oct 2024 10:21:22 +0200,
> > >> >> Christoffer Sandberg wrote:
> > >> >>>
> > >> >>>
> > >> >>> On 30.9.2024 09:44, Takashi Iwai wrote:
> > >> >>>> On Mon, 23 Sep 2024 21:37:42 +0200,
> > >> >>>> Jerry Luo wrote:
> > >> >>>>>
> > >> >>>>> Hi Takashi,
> > >> >>>>>
> > >> >>>>> On Mon, 16 Sep 2024 19:22:05 +0200,
> > >> >>>>>
> > >> >>>>> Takashi Iwai wrote:
> > >> >>>>>
> > >> >>>>>      Could you give alsa-info.sh output from both working and
> > >> >>>>> non-working
> > >> >>>>>      cases?  Run the script with --no-upload option and attach the
> > >> >>>>> outputs.
> > >> >>>>>
> > >> >>>>>      thanks,
> > >> >>>>>
> > >> >>>>>      Takashi
> > >> >>>>>
> > >> >>>>> Issue now reappear, output from alsa-info.sh are attached. If they
> > >> >>>>> are still
> > >> >>>>> needed.
> > >> >>>> Thanks.  The obvious difference seems to be the assignment of two
> > >> >>>> DACs
> > >> >>>> 0x10 and 0x11 for headphone and speaker outputs.
> > >> >>>>
> > >> >>>> Christoffer, how are those on your machines?
> > >> >>> I attached alsa-info from the Sirius Gen2 device.
> > >> >>>
> > >> >>> Comparing the working/nonworking of Jerry, yeah, the assignment of
> > >> >>> 0x10 and 0x11 looks switched around. I don't see what difference this
> > >> >>> would make. Also, node 0x22 has "bass speaker" controls in the
> > >> >>> non-working version.
> > >> >>>
> > >> >>> Comparing the Sirius Gen2 alsa-info with Jerrys, to me it looks like
> > >> >>> the non-working version corresponds to our working version.
> > >> >>>
> > >> >>> I would expect the non-working version to happen all the time though
> > >> >>> with regards to the "bass speaker" controls. Why would this only
> > >> >>> happen sometimes?
> > >> >> Thanks!  The assignment of DACs depend on the pins and topology, so it
> > >> >> can be a bit sensitive.
> > >> >>
> > >> >> Now looking more closely at both outputs, I wonder how the commit
> > >> >> breaks pang14.  Maybe it has a PCI SSID 2782:12c5 (or 12c3) while the
> > >> >> codec SSID is 2782:12b3?  If so, the patch below should fix.
> > >> 
> > >> Interesting, you're right, PCI SSID c3/c5 and codec SSID c3/c5 for the
> > >> Siriuses.
> > >> 
> > >> I had a look around. In patch_realtek there are some cases where codec
> > >> SSID match is needed as well. Would it be better/safer to directly do
> > >> this immediately or keep it as an exception where it breaks or for
> > >> known sensitive models/brands?
> > > 
> > > It needs a better handling, yes.  OTOH, the driver became gigantic and
> > > it's very tough to change the basic matching stuff.  That is, we can't
> > > flip from PCI SSID to codec SSID out of sudden, as it'll break
> > > certainly many other systems.
> > > 
> > > What I have in mind is to add an extra flag to the matching table to
> > > indicate the codec SSID matching, something like:
> > > 
> > > --- a/sound/pci/hda/hda_local.h
> > > +++ b/sound/pci/hda/hda_local.h
> > > @@ -282,6 +282,7 @@ struct hda_fixup {
> > >  	int type;
> > >  	bool chained:1;		/* call the chained fixup(s) after this */
> > >  	bool chained_before:1;	/* call the chained fixup(s) before this */
> > > +	bool match_codec_ssid:1; /* match with codec SSID instead of
> > > PCI SSID */
> > >  	int chain_id;
> > >  	union {
> > >  		const struct hda_pintbl *pins;
> > > 
> > 
> > Sounds reasonable to me. It would mean either-or then though.
> 
> Ah yes, it matches primarily with codec SSID instead of fallback.
> I guess if this flag is set, we shouldn't check PCI SSID, though.
> 
> > > Although this will help in this case, many of existing code do check
> > > codec ID in addition to PCI SSID, and this flag won't help for them as
> > > is.
> > 
> > For the fixups where codec SSID is already known for all cases it
> > would be possible to represent current logic with matching both SSIDs.
> > 
> > Otherwise I can not judge whether matching both PCI SSID and codec
> > SSID at the same time would be needed.
> > 
> > In any case, your approach would reduce code size if this is a
> > recurring thing.
> 
> OK, I'll try to prepare a cleanup patch later.

It turned out that my original idea above wasn't applicable, as we
want to filter out rather in the quirk table, not the target.
So I ended up with another patch and submitted shortly ago:
  https://lore.kernel.org/20241008120233.7154-1-tiwai@xxxxxxx


Takashi