Re: [RFC] Fix for conflict of HDMI and SPDIF IEC958 controls

[Jaroslav Kysela <perex@perex.cz>]

Date 15.10.2012 12:31, Takashi Iwai wrote:
> At Mon, 15 Oct 2012 12:21:18 +0200,
> Jaroslav Kysela wrote:
>>
>> Date 12.10.2012 17:18, Takashi Iwai wrote:
>>> Hi,
>>>
>>> there is a long-standing problem in HD-audio regarding IEC958
>>> controls.  When both an SPDIF and an HDMI are created on the same
>>> card (e.g. one from analog codec and one from graphics chip), the
>>> driver assigns the IEC958 controls just with new indices, 0, 1, 2...
>>>
>>> The problem is that there is no way to connect between this index and
>>> the actual PCM device.  Currently, alsa-lib HDA-Intel.conf has a fixed
>>> configuration:
>>>   spdif  -> IEC958 xxx index=0, PCM dev=1
>>>   hdmi,0 -> IEC958 xxx index=0, PCM dev=3
>>>   hdmi,1 -> IEC958 xxx index=1, PCM dev=7
>>>   hdmi,2 -> IEC958 xxx index=2, PCM dev=8
>>>   hdmi,3 -> IEC958 xxx index=3, PCM dev=9
>>>
>>> So obviously spdif and the first hdmi conflict.
>>>
>>> Basically this can be fixed by reassigning each IEC958 control with
>>> the same device number corresponding to the PCM device.  That is, for
>>> SPDIF, assign a control element with device=1, for HDMI,
>>> device=3,7...
>>> However, this obviously breaks the old configuration unless user
>>> upgrades the alsa-lib configuration.  Thus this is no-go.
>>>
>>> Now here is a compromise: the IEC958 control for SPDIF is reassigned
>>> to device=1 *only* when SPDIF and HDMI coexist.  Since the
>>> configuration is anyway broken as is now in such a case, it's no big
>>> deal to fix one side in an incompatible way.  (The reason why SPDIF
>>> is re-assigned is that I guess majority of user require more HDMI than
>>> SPDIF in such a configuration.)
>>>
>>> In addition, we need a fix in alsa-lib.  Also not for breaking the
>>> compatibility with older kernel, we need some fallback check.  I did a
>>> quick hack to alsa-lib conf code and added "skip_rest" option to the
>>> hook element, so that only the first matching element is taken.
>>>
>>> Long story short, I cooked up three patches.  One patch is for kernel,
>>> to add the workaround above, and the two are for alsa-lib, one clean
>>> up and one to introduce the skip_rest option (and application to
>>> HDA-Intel.conf).
>>>
>>> My plan is to merge this to 3.8 tree, so it's no urgent issue.
>>> But it's of course always good to fix something.
>>
>> What about to introduce more cleaner solution - export this information
>> using TLV to user space, so alsa-lib can enumerate (distinguish) the
>> spdif and hdmi devices properly.
> 
> Yes, this is the next step I've planned.  The additional information
> doesn't conflict with this fix.
> 
> The point of this fix is to keep / fix the things working as much as
> possible even without changing user-space.  If a user-space upgrade is
> mandatory, it can be regarded as an obvious regression.
> 
> In short, my patch is:
> - A quick fix
> - Doesn't break the cases without SPDIF/HDMI conflict
> - Fixes HDMI for the conflicting case even without changing user-space
> - Would break the SPDIF output (only when SPDIF/HDMI conflicts)
>   without config update

I understand, but because the config update mostly depends on the whole
alsa-lib package update, then I believe, the proper fix may be
implemented now rather than introducing a config hack option. But the
step-by-step improvement is ok to me, too.

					Jaroslav

-- 
Jaroslav Kysela <perex@perex.cz>
Linux Kernel Sound Maintainer
ALSA Project; Red Hat, Inc.