[alsa-devel] Timer instability

[Takashi Iwai]

At Wed, 25 Feb 2009 16:24:18 +0000,
Clive Messer wrote:
> 
> On Wednesday 25 Feb 2009 15:51:33 you wrote:
> > At Wed, 25 Feb 2009 15:11:44 +0000,
> >
> > Clive Messer wrote:
> > > 36192	36191	100068	3	0	0	1	2
> > > 36195	36194	100090	2	0	0	1	2
> > > 36198	36198	100113	1	0	0	1	2
> > > 36207	36206	63854	2816	1600	0	1	3
> > > 36212	36211	63854	2815	1601	4	1	3
> > > 36216	36216	63854	2814	1602	4	1	3
> > > 36221	36220	63854	2813	1603	4	1	3
> > > 36225	36225	101224	4460	-44	4	1	3
> >
> > This is pretty weird.  Is this behavior always reproducible?
> 
> Nope. I get different results every time I run it.
> 
> > What shows with the patch below to alsa-lib?
> 
> Here's one (attached) I just did with the new patch that shows the crazy high 
> numbers again.
> hwsync: 2080(0), 6496(0)
> 70758	70758	47142	0	4416	0	0	3
> hwsync: 1472(0), 6496(0)
> 70763	70762	104573379136040	4971973988617026976	-4971973988617022560	4	1	3

The problem is that hwptr goes backward here.  But it didn't happen in
this hwsync call but sometime before that.

Could you try the patch below instead?  This will track all sync_ptr calls.


Takashi

---
diff --git a/src/pcm/pcm_hw.c b/src/pcm/pcm_hw.c
index e9ce092..ad8d4a5 100644
--- a/src/pcm/pcm_hw.c
+++ b/src/pcm/pcm_hw.c
@@ -131,6 +131,10 @@ struct timespec snd_pcm_hw_fast_tstamp(snd_pcm_t *pcm)
 static int sync_ptr1(snd_pcm_hw_t *hw, unsigned int flags)
 {
 	int err;
+	long old_hwptr, new_hwptr;
+	long old_applptr, new_applptr;
+	old_hwptr = *pcm->hw.ptr;
+	old_applptr = *pcm->appl.ptr;
 	hw->sync_ptr->flags = flags;
 	err = ioctl((hw)->fd, SNDRV_PCM_IOCTL_SYNC_PTR, (hw)->sync_ptr);
 	if (err < 0) {
@@ -138,6 +142,11 @@ static int sync_ptr1(snd_pcm_hw_t *hw, unsigned int flags)
 		SYSMSG("SNDRV_PCM_IOCTL_SYNC_PTR failed");
 		return err;
 	}
+	new_hwptr = *pcm->hw.ptr;
+	new_applptr = *pcm->appl.ptr;
+	printf("sync_ptr1: %ld(%ld), %ld(%ld)\n",
+	       new_hwptr, new_hwptr - old_hwptr,
+	       new_applptr, new_applptr - old_applptr);
 	return 0;
 }