snd-aoa: g5 tas codec problems

Johannes Berg johannes at
Fri Jul 7 18:03:34 EST 2006

On Fri, 2006-07-07 at 17:47 +1000, Benjamin Herrenschmidt wrote:
> Ok, so now I have it working on the G5 :)

Great :)
Powerbook woke itself an hour ago, no idea why, sorry for confusing you
on irc.

>  - Maybe it's me or maybe it's just too complicated, but I haven't quite
> grasped the whole interaction between the soundbus,i2sbus,fabric and
> codecs... especially initialisation ordering. It would be nice if we
> could spend some time going through that and simplifying :) It leads to
> at least one of the problems

Yeah, well... I sorta know.

>  - The patch fixes a couple of nits related to having the modules
> built-in: soundbus must really be a subsys_initcall() so it's
> initialized before anybody else, and I've put the soundbus/ dir before
> the codecs in the link order because the TAS is unhappy if loaded before
> i2s (see below)


>  - The TAS is a nasty beast. It needs the i2s clocks enabled or it goes
> bunk... That's the problem with the G5. I've added a clock notifier and
> reset it completely when the clocks come back, that's what fixes the G5
> sound (looks like it loses state somewhat when not clocked). It would be
> nice to streamline/cleanup some of the TAS handling, maybe with register
> shadows like darwin or just with proper state variables to re-consitute
> the whole thing and have a "fast mode" load since we need to do it so
> often


>  - Because of the above, starting to play a sound 1- takes some time to
> init things and 2- clacks (setting the mutes on TAS before losing clocks
> isn't enough, I think the fact that it goes bonk makes it parasite the
> analog outputs). We need to also mute the amps around that. I haven't
> quite figured how to do that from i2sbus though :) Also, we should try
> (if not already the case) to cache our clock/i2s state so that
> subsequent prepare() don't try to change things that are already ok.
> That way we avoid having to blast the codec each time. But right now,
> the important thing is to add mutes. People with external amplifiers
> will really not like those clacs...

Right. The way I did that with the onyx is that I told the GPIOs to mute
all amps from the clock switch callback.

I see you added DRC too, thanks :) I'll take a closer look after
breakfast ;)

-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 828 bytes
Desc: This is a digitally signed message part
URL: <>

More information about the Linuxppc-dev mailing list