[alsa-devel] Per board ucm files on x86?

[Pierre-Louis Bossart]

On 12/11/2017 07:40 AM, Takashi Iwai wrote:
> On Mon, 11 Dec 2017 13:30:35 +0100,
> Hans de Goede wrote:
>> Hi All,
>>
>> This weekend I've created a modified ucm config based on:
>>
>> https://github.com/plbossart/UCM/tree/master/chtrt5645
>>
>> For a board which has a single speaker connected to the
>> left channel (standard mono speaker setup) and a stereo
>> headphone jack with working jack detection.
>>
>> I've been unable to come up with a ucm file which allows
>> selecting between a "Stereo Speaker + Headphone" vs
>> "Mono Speaker + Headphone" output profile.
>>
>> https://github.com/plbossart/UCM/tree/master/byt-rt5640
>> has "Mono Speaker", "Stereo Speaker" and "Headphone"
>> profiles but does not auto-switch between Headphone
>> and speaker based on jack detection, I've been unable to
>> allow selecting either stereo or mono speaker while
>> keeping auto-switching to/from the headphones.
>>
>> But thinking more about this I don't think that having
>> "Stereo Speaker + Headphone" and "Mono Speaker + Headphone"
>> profiles is the answer. Profiles make sense on machines with
>> a bunch of outputs where we don't no what the user is going
>> to plug in, but in this case the stereo vs mono speaker
>> distinction is a clear property of the device, which we
>> should IMHO autodetect based on the device-model.
>>
>> So I think we need a way to have different ucm files per board,
>> so instead of loading /usr/share/ucm/chtrt5645/*.conf on my
>> device, alsa should try to load /usr/share/ucm/chtrt5645-<boardname>/*.conf
>>
>> Specifically I'm thinking about using udev + hwdb (dmi string)
>> matching to set an ALSA_UCM_NAME udev property.
>>
>> If the consensus is that this is a good idea I can take a shot
>> at writing patches for this (in my spare time mostly), the
>> downside of this approach is it would cause a dependency on
>> libudev for the alsa ucm code.
> You don't need to patch, I guess.  The recent code should set a string
> generated from DMI as the longname of the card, and alsa-lib UCM
> parser prefers the longname to the driver name field.  That is,
> /usr/share/ucm/$LONGNAME/$LONGNAME.conf would be read at first, then
> /usr/share/ucm/$DRIVER/$DRIVER.conf is used as fallback.
>
The long name is really awful at times, I've seen PulseAudio try to load 
UCM profiles of boards with over 40 characters with multiple repeats.
Also from a maintainability perspective we need the include mechanism 
and I don't think it was ever merged - or what I saw wasn't very useful.