[alsa-devel] Codecs without auto-parser

[David Henningsson]

On 2011-02-03 14:16, Takashi Iwai wrote:
> At Wed, 02 Feb 2011 16:16:21 +0100,
> David Henningsson wrote:
>>
>> On 2011-02-01 22:34, Takashi Iwai wrote:
>>> At Tue, 01 Feb 2011 08:37:25 +0100,
>>> David Henningsson wrote:
>>>>
>>>> On 2011-01-28 13:10, Takashi Iwai wrote:
>>>>> At Fri, 28 Jan 2011 12:51:30 +0100,
>>>>> David Henningsson wrote:
>>>>>>
>>>>>> On 2011-01-28 09:01, Takashi Iwai wrote:
>>>>>>> And I hope that we should go further a bit for now -- more clean up of
>>>>>>> the cxt5066 code either checking BIOS pins or hp/mic/spk pre-defined
>>>>>>> pins.  Currently, the code is fairly messy (partly because of olpc
>>>>>>> support), and now is a good chance to improve it a bit more.
>>>>>>
>>>>>> Maybe so. I'm not sure I can commit to doing that work right now, as I
>>>>>> have other work commitments waiting for me to take care of them.
>>>>>
>>>>> OK.
>>>>
>>>> Hmm. I got a few things cleared up yesterday, and so I might have some
>>>> time to actually do this.
>>>
>>> Great.
>>>
>>>> However, it also seems like I have to fix an
>>>> AD1984A machine, and that driver is in equally bad shape (as in: no
>>>> auto-parser, just a list of models).
>>>
>>> Yeah, I know...
>>>
>>>> I could add another model, but
>>>> suppose I'd do this the right way, how would I do it?
>>>>
>>>> I guess there are three options to start off with:
>>>>
>>>> 1) ad1988 seems to have an auto-parser. There are a lot of hardcoded
>>>> nids in there, but I guess I could copy-and-paste some into a new
>>>> auto-parser for ad1984a.
>>>
>>> Well... AD1988 and AD1984A aren't so similar, IIRC.
>>> AD1984A is more straightforward hardware implementation while AD1988
>>> can have more connections.  Also AD1988 auto-parser isn't in the best
>>> form in comparison with other codecs.
>>>
>>> That being said, if AD1984A can fit with AD1988's auto-parser (i.e.
>>> my memory was too vague), it's fine.  We can adapt it.
>>>
>>> If this is hard, one more quirk model won't be too bad for AD1984A
>>> since this is a pretty simple codec, and there won't be so more devices
>>> with this old chip.
>>
>> Ok. I had one more look at AD1988's auto-parser and I don't think it's
>> good to start off that one.
>>
>>>> 2) It seems to me like the most competent candidate for auto-parser
>>>> currently is the cx_auto stuff - compared to other auto-parsers the
>>>> hardcoded nids are fewer. Perhaps this one could be extended to also do
>>>> auto-parsing for codecs from other vendors? Or at least some convenience
>>>> functions moved to hda_codec.c in order to remove duplication between
>>>> vendors?
>>>
>>> Yes, I can loudly tell people "I have a dream, unify all codec parsers!" :)
>>> But, I'm afraid that a big hammer doesn't work.  The topology is fairly
>>> different between codecs, and resolving all makes the parser complex.
>>> So, some hints would be needed to simplify the logic, after all.
>>>
>>>> 3) There is also the generic parser. It seems to be the one most capable
>>>> of handling randomly complex graphs, but is lacking automute/jack-detect
>>>> support, and maybe other functions as well (?).
>>>
>>> The generic parser is the very first parser, and it never worked rightly.
>>> This should be really replaced with a more clever one.
>>
>> Hrmm...am I the only one thinking that your answer to 2) and 3) are in
>> direct contradiction? :-)
>
> The reality is bitter, you know :)
>
>> I'm looking at fill_cx_auto_dacs. That doesn't look like Conexant
>> specific stuff at all. cx_auto_hp_automute looks quite generic as well,
>> except for a pointer deref in the beginning.
>> cx_auto_build_output_controls could be merged with
>> xxxx_create_multi_out_ctls present in the Realtek and Analog parsers.
>> And so on.
>>
>> Shouldn't we move and rename fill_cx_auto_dacs to hda_codec.c and use it
>> for all codecs, instead of hardcoding dac nids everywhere? At least in
>> general? (There might be exceptions, where there is a dac you don't want
>> to use.)
>
> Well, Conexant auto-parser is a fairly recent one I wrote, so it's
> more or less generic.  But there are still a few glitches that
> Conexant parser doesn't handle properly, IIRC.
>
> We can give it a try with different codecs using hda-emu to see what
> doesn't fit.
>
>> Anyway, you wanted some work done to Conexant 5066. Could you clarify
>> that a little? You likely want to call snd_hda_parse_pin_def_config -
>> but are you expecting yet another auto-parser after that? Or just to
>> choose model based on the result? And should we remove all the model
>> quirks (except possibly for OLPC) afterwards?
>
> My initial idea was to adapt the current Conexant parser for 5066.
> If it doesn't work by some reason, adjust the code, then replaces the
> model quirks with the pin-config presets.

Ok. While waiting for your answer I continued thinking and writing a 
little code - here's how I would see a generic version of auto_parse_output:

1) Make a breadth-first-search, starting at each (output) pin, to find 
what DACs are reachable from each pin.

2) Start assigning DACs to pins, start off with line-outs (similar to 
what cx_auto does today). If HP's and Speakers can't have their own 
DACs, they'll have to share with their line-out counterpart (usually Front).

3) Investigate path from pin to DAC and add mute switches as close to 
the pin:s as possible.

4) Add volume controls on all capable widgets that are a part of the 
path. For 3 and 4 the naming must be decided based on what paths they 
are a part of.

(On top of that, add a quirk system that works with snd_hda_codec_read 
so that any command can be overridden to returned a fixed value instead. 
That quirk system could then be used to not only override pin configs 
but also remove paths so the above algorithm does not discover them, and 
other things.)

What do you think? Maybe I should give up these thoughts for now and 
just go with your suggestion...but then I would still have to fix that 
model for AD1984a...

-- 
David Henningsson, Canonical Ltd.
http://launchpad.net/~diwic