[alsa-devel] [PATCH] [ML403-AC97CR] Fix capture/periodic overrun bug.
Hi all,
this is a bug fix for my ALSA driver I posted in August. Since then I hadn't the time to look at it, but some days ago I took some time and found the reason why there are so many overruns while capturing. Furthermore, this patch removes the one "heavy" printk for capture debugging.
From: Joachim Foerster JOFT@gmx.de
We have to do fairly accurate counting of the minimal periods, instead of being lazy and just setting the number to zero as soon as one period elapses.
Signed-off-by: Joachim Foerster JOFT@gmx.de --- sound/drivers/ml403-ac97cr.c | 6 ++---- sound/drivers/pcm-indirect2.c | 20 ++------------------ 2 files changed, 4 insertions(+), 22 deletions(-)
diff --git a/sound/drivers/ml403-ac97cr.c b/sound/drivers/ml403-ac97cr.c index 2636249..1b16a21 100644 --- a/sound/drivers/ml403-ac97cr.c +++ b/sound/drivers/ml403-ac97cr.c @@ -28,11 +28,9 @@ * accesses to a minimum, because after a variable amount of accesses, the AC97 * controller doesn't raise the register access finished bit anymore ... * - * - Capture support works - basically, but after ~30s (with rates > ~20kHz) - * ALSA stops reading captured samples from the intermediate buffer and - * therefore a overrun happens - ATM I don't know what's wrong. - * * - Playback support seems to be pretty stable - no issues here. + * - Capture support "works" now, too. Overruns don't happen any longer so often. + * But there might still be some ... */
#include <sound/driver.h> diff --git a/sound/drivers/pcm-indirect2.c b/sound/drivers/pcm-indirect2.c index a36215a..ef74a38 100644 --- a/sound/drivers/pcm-indirect2.c +++ b/sound/drivers/pcm-indirect2.c @@ -403,7 +403,7 @@ snd_pcm_indirect2_playback_interrupt(struct snd_pcm_substream *substream, rec->min_multiple); rec->mul_elapsed++; #endif - rec->min_periods = 0; + rec->min_periods = (rec->min_periods % rec->min_multiple); snd_pcm_period_elapsed(substream); } } @@ -568,24 +568,8 @@ snd_pcm_indirect2_capture_interrupt(struct snd_pcm_substream *substream, rec->mul_elapsed_real += (rec->min_periods / rec->min_multiple); rec->mul_elapsed++; - - if (!(rec->mul_elapsed % 4)) { - struct snd_pcm_runtime *runtime = substream->runtime; - unsigned int appl_ptr = - frames_to_bytes(runtime, - (unsigned int)runtime->control-> - appl_ptr) % rec->sw_buffer_size; - int diff = rec->sw_data - appl_ptr; - if (diff < 0) - diff += rec->sw_buffer_size; - snd_printk(KERN_DEBUG - "STAT: mul_elapsed: %d, sw_data: %u, " - "appl_ptr (bytes): %u, diff: %d\n", - rec->mul_elapsed, rec->sw_data, appl_ptr, - diff); - } #endif - rec->min_periods = 0; + rec->min_periods = (rec->min_periods % rec->min_multiple); snd_pcm_period_elapsed(substream); } }
At Fri, 02 Nov 2007 18:52:19 +0100, Joachim Foerster wrote:
Hi all,
this is a bug fix for my ALSA driver I posted in August. Since then I hadn't the time to look at it, but some days ago I took some time and found the reason why there are so many overruns while capturing. Furthermore, this patch removes the one "heavy" printk for capture debugging.
From: Joachim Foerster JOFT@gmx.de
We have to do fairly accurate counting of the minimal periods, instead of being lazy and just setting the number to zero as soon as one period elapses.
Signed-off-by: Joachim Foerster JOFT@gmx.de
Oh, it seems that your patch was lost in the patch queue before ALSA 1.0.15 release. Sorry, I applied the last one now to ALSA HG tree together with this one.
The patch to arch/ppc/syslib/virtex_device.c isn't still applied (because it's outside the ALSA HG tree). For applying this to alsa git tree, Jaroslav would need to apply manually.
Jaroslav, could you add it please?
... Or, feel free to apply that patch to ppc tree separately. It shouldn't break anything, at least :)
Thanks,
Takashi
On 11/5/07, Takashi Iwai tiwai@suse.de wrote:
Oh, it seems that your patch was lost in the patch queue before ALSA 1.0.15 release. Sorry, I applied the last one now to ALSA HG tree together with this one.
The patch to arch/ppc/syslib/virtex_device.c isn't still applied (because it's outside the ALSA HG tree). For applying this to alsa git tree, Jaroslav would need to apply manually.
Jaroslav, could you add it please?
Please don't send this change in through the alsa git tree. arch/ppc is semi frozen and changes should really go in via paulus' tree.
I'll take another look at it and see if I can get it in.
Cheers, g.
At Mon, 5 Nov 2007 08:59:10 -0700, Grant Likely wrote:
On 11/5/07, Takashi Iwai tiwai@suse.de wrote:
Oh, it seems that your patch was lost in the patch queue before ALSA 1.0.15 release. Sorry, I applied the last one now to ALSA HG tree together with this one.
The patch to arch/ppc/syslib/virtex_device.c isn't still applied (because it's outside the ALSA HG tree). For applying this to alsa git tree, Jaroslav would need to apply manually.
Jaroslav, could you add it please?
Please don't send this change in through the alsa git tree. arch/ppc is semi frozen and changes should really go in via paulus' tree.
I'll take another look at it and see if I can get it in.
OK, thanks for information. I'll be happy to pass it to you :)
Takashi
participants (3)
-
Grant Likely
-
Joachim Foerster
-
Takashi Iwai