linked streams question

linked streams question

[Eliot Blennerhassett]

Greetings,

I'm adding linked stream support to my driver (asihpi), and having some 
trouble with the end of playback.

What happens is that the slave streams never transition to SETUP state at the 
end of playback.
(Eg playing 4 channel wav file to 2 linked stereo streams with aplay).

The master stream reaches snd_pcm_drain_done() which then causes 
trigger(stop),  resulting in all linked streams being stopped.
But only the master state is set to SETUP

However, now the slave streams are stopped,  period_elapsed() is not called 
for a final time to (I assme) cause them to transition to the SETUP state.
Eventually I get "playback drain error (DMA or IRQ trouble?)\n"

My workaround is to set the state of each substream to SETUP in the 
trigger-stop operator, but no other driver needs to do this so I think it is 
probably the wrong solution - maybe misses some other cleanup code?  

Any suggestions on the proper way to handle this?

regards

Eliot

Re: linked streams question

[Takashi Iwai]

At Thu, 05 Jul 2007 19:53:36 +1200,
Eliot Blennerhassett wrote:
> 
> Greetings,
> 
> I'm adding linked stream support to my driver (asihpi), and having some 
> trouble with the end of playback.
> 
> What happens is that the slave streams never transition to SETUP state at the 
> end of playback.
> (Eg playing 4 channel wav file to 2 linked stereo streams with aplay).
> 
> The master stream reaches snd_pcm_drain_done() which then causes 
> trigger(stop),  resulting in all linked streams being stopped.
> But only the master state is set to SETUP
> 
> However, now the slave streams are stopped,  period_elapsed() is not called 
> for a final time to (I assme) cause them to transition to the SETUP state.
> Eventually I get "playback drain error (DMA or IRQ trouble?)\n"
> 
> My workaround is to set the state of each substream to SETUP in the 
> trigger-stop operator, but no other driver needs to do this so I think it is 
> probably the wrong solution - maybe misses some other cleanup code?  

snd_pcm_drain() has a loop to wait until the all substreams are
drained.  That's why snd_pcm_drain_done() stops only one substream,
not all linked substreams.

But, yes, there seems to be a problem with hardware-linked substreams
that stops the multiple substreams.  This mess comes from the fact
that we handle both soft-linked and hard-linked substreams as if they
are identical.  We'd need a deeper clean up, I'm afraid.


Takashi