Re: [PATCH] Kernel crash

[Takashi Iwai <tiwai@suse.de>]

At Wed, 15 Sep 2004 19:50:08 +0200,
Giuliano Pochini wrote:
> 
> On Wed, 15 Sep 2004 12:24:03 +0200
> Takashi Iwai <tiwai@suse.de> wrote:
> 
> > I believe the second part is not needed.  DRAINING in capture implies
> > that the DMA is already stopped (like XRUN).
> 
> snd_pcm_capture_drain() says:
> 
>         case SNDRV_PCM_STATE_RUNNING:
>                 if (snd_pcm_update_hw_ptr(substream) >= 0) {
>                         snd_pcm_stop(substream,
>                                      snd_pcm_capture_avail(runtime) > 0 ?
>                                      SNDRV_PCM_STATE_DRAINING : SNDRV_PCM_STATE_SETUP);
>                         break;
>                 }
>                 /* Fall through */
>         case SNDRV_PCM_STATE_XRUN:
>                _xrun_recovery:
>                 snd_pcm_change_state(substream,
>                                      snd_pcm_capture_avail(runtime) > 0 ?
>                                      SNDRV_PCM_STATE_DRAINING : SNDRV_PCM_STATE_SETUP);
>                 break;
> 
> It may not be stopped when snd_pcm_update_hw_ptr()<0, if I'm not missing
> something.

snd_pcm_update_hw_ptr() returns a negative value only when xrun
happens or the drain is finished.  In both cases, snd_pcm_stop() is
called internally before returning from this function.


> > If the second part really fixes the problem, the bug is in somewhere
> > else.
> 
> It fixes it and yes, there is another problem: snd_pcm_change_state()
> overrides the state machine.
> 
> snd_pcm_playback_drain() (when state=RUNNING) calls
> snd_pcm_change_state(DRAINING). When substreams are linked, it changes the
> state of all linked substreams *without* passing through
> snd_pcm_capture_drain(DRAINING).
> 
> I checked all calls to snd_pcm_change_state() and the only problematic one
> is at line 1288 inside snd_pcm_capture_drain(). The problem exists only if
> capture and playback substreams are linked because a capture substream in
> DRAINING state implies it's stopped, while a playback substream doesn't. My
> patch is not the proper fix, but it workarounds the bug by fixing that
> asymmetry. I make snd_pcm_capture_drop() and snd_pcm_playback_drop() behave
> the same wrt DRAINING substreams.

I think it'd be better to add a new field to substream to indicate the
DMA running status so that snd_pcm_stop() can be called safely
regardless of the current status.

The situation right now is a mess, the DMA running state depending on
the stream direction and the state.


Takashi


-------------------------------------------------------
This SF.Net email is sponsored by: thawte's Crypto Challenge Vl
Crack the code and win a Sony DCRHC40 MiniDV Digital Handycam
Camcorder. More prizes in the weekly Lunch Hour Challenge.
Sign up NOW http://ad.doubleclick.net/clk;10740251;10262165;m