[PATCH] Fixes: Re: intel8x0 has stopped working.

[James Courtier-Dutton <James@superbug.demon.co.uk>]

[-- Attachment #1: Type: text/plain, Size: 2417 bytes --]

James Courtier-Dutton wrote:
> 
> Once thing I have noticed, is that with the alc650, we used to have VRA 
> (alsa 0.9.8), but the 1.0.2 intel8x0 driver ignores the VRA and fixes 
> itself at 48000.
> So, before I never needed the sample rate converters, but as it now 
> fixes the rate at 48000, I need the sample rate converters, and they 
> don't actually work.
> 
> 

I have found out what the problem is with the resamplers not working.
I was using: -
err = snd_pcm_sw_params_set_start_threshold(handle, swparams, 
buffer_size);  <- note buffer_size.

So, the state will stay in SND_PCM_STATE_PREPARED state until the buffer 
is full.
Inside snd_pcm_write_areas() [in alsa-lib] we have a snd_pcm_wait 
statement before we write frames into the buffer.
if (((snd_pcm_uframes_t)avail < pcm->avail_min && size > 
(snd_pcm_uframes_t)avail) || (size >= pcm->xfer_align && 
(snd_pcm_uframes_t)avail < pcm->xfer_align)) {
then it calls snd_pcm_wait()

So, this will call snd_pcm_wait() until avail >= pcm->avail_min.
This will only become true if we are in SND_PCM_STATE_RUNNING state.

below the snd_pcm_wait(), we do try to execute snd_pcm_start() to get us 
into SND_PCM_STATE_RUNNING, but this only gets executed if: -
if (state == SND_PCM_STATE_PREPARED && hw_avail >= (snd_pcm_sframes_t) 
pcm->start_threshold) {

So, we will only reach a running state if the buffer is FULL, but the 
buffer will NEVER become full, because of the previous snd_pcm_wait();
CATCH22!!!

I think a fix to this would be to limit as follows: -
start_threshold <= (buffer_size - avail_min)

I have fixed the issue for now, just by setting the "start_threshold" to 
something other than buffer_size. But I think some work needs to be done 
with alsa-lib to prevent this deadlock situation ever happening in alsa-lib.

 From the application writers point of view, if the buffer is X bytes 
long, and one used snd_pcm_writei() to write X+10 bytes, the buffer 
should become full. Currently, that is not always the case, and alsa-lib 
instead locks in snd_pcm_wait().

I attach a patch that fixes alsa-lib.
All it does it call snd_pcm_start() just before snd_pcm_wait() and due 
to the if statement just above it, will only happen when the buffer is 
going to be filled, therefore ensuring that even if "start_threshold" is 
set too high, snd_pcm_start() will always be called if the buffer is 
full, or about to become full.

Cheers
James

[-- Attachment #2: alsa-lib-lockup-fix.diff --]
[-- Type: text/x-patch, Size: 535 bytes --]

--- src/pcm/pcm.c.old	2004-02-02 23:07:14.355587424 +0000
+++ src/pcm/pcm.c	2004-02-02 23:07:19.871748840 +0000
@@ -6033,6 +6033,11 @@
 		} else if (((snd_pcm_uframes_t)avail < pcm->avail_min && size > (snd_pcm_uframes_t)avail) ||
 		           (size >= pcm->xfer_align && (snd_pcm_uframes_t)avail < pcm->xfer_align)) {
 
+			if (state == SND_PCM_STATE_PREPARED ) {
+				err = snd_pcm_start(pcm);
+				if (err < 0)
+					goto _end;
+                        }
 			if (pcm->mode & SND_PCM_NONBLOCK) {
 				err = -EAGAIN;
 				goto _end;