Skip to contentSkip to conversation

Fix aplay and alsa-lib write_areas() freezing with 100% cpu usage on stuck slave pcm

RFC PATCH In review Sergey 2 participants 6 messages Jun 13, 2014

lore
Share & tools
lore

Sergey

author
#0 lore raw

cover Fix aplay and alsa-lib write_areas() freezing with 100% cpu usage on stuck slave pcm

These patches originate from aplay -vvv /dev/zero not responding to Ctrl+C when playing to plugged "null", e.g.
  pcm.!default {
    type plug
    slave.pcm { type null }
    slave.format S32_LE
  }

First patch fixes snd_pcm_write_areas() stuck in a loop because of transfer func() always returning 0. That caused application calling it to freeze. In the example above it have frozen aplay. If it was called by firefox, it would freeze firefox.

Is it a good way to fix it? Should snd_pcm_read_areas() be fixed like that too?

Second patch fixes aplay itself, not returning from pcm_write() after SIGINT.

Additionally aplay may restore original interrupt handler, to make second Ctrl+C stop it for sure. In case of other bugs in alsa-lib or third-party slave pcms it seems better than SIGkilling it manually.

Comments and suggestions are welcome.

Regardless of these patches' fate null plugin must be fixed on its own.

Patches (2):
  pcm: break out of snd_pcm_write_areas() if write function returned 0
  aplay: Escape from pcm_write() when in_aborting state

 src/pcm/pcm.c |    2 +-
 aplay/aplay.c |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
Signature · 5 lines
1.7.10.4
_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel
author
#1 lore raw

PATCH 1/2 pcm: break out of snd_pcm_write_areas() if write function returned 0

When pcm is stuck and write func returns 0, snd_pcm_write_areas() loops infinitely freezing the application and not allowing it to handle the case. Can be reproduced with:

  # aplay -vvv /dev/zero
  pcm.!default {
    type plug
    slave.pcm { type null }
    slave.format S32_LE
  }
This patch fixes it by breaking the loop if func returns 0.
Note: pcm_null stuckness must be fixed on its own.
---
 src/pcm/pcm.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
src/pcm/pcm.c +1 -1
index 7e46014..5bb7582 100644
@@ -6816,7 +6816,7 @@ snd_pcm_sframes_t snd_pcm_write_areas(snd_pcm_t *pcm, const snd_pcm_channel_area
 		if (! frames)
 			break;
 		err = func(pcm, areas, offset, frames);
-		if (err < 0)
+		if (err <= 0)
 			break;
 		frames = err;
 		if (state == SND_PCM_STATE_PREPARED) {
Signature · 1 line
1.7.10.4
author
#2 lore raw

PATCH 2/2 aplay: Escape from pcm_write() when in_aborting state

In reply to Sergey

This patch fixes aplay freezing and not responding to Ctrl+C when playback pcm is stuck and returns 0 to write attempts:
Example:

  # aplay -vvv /dev/zero
  pcm.!default {
    type plug
    slave.pcm { type null }
    slave.format S32_LE
  }
---
 aplay/aplay.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
aplay/aplay.c +1 -1
index 77e40a3..a5b9c2a 100644
@@ -1921,7 +1921,7 @@ static ssize_t pcm_write(u_char *data, size_t count)
 		count = chunk_size;
 	}
 	data = remap_data(data, count);
-	while (count > 0) {
+	while (count > 0 && !in_aborting) {
 		if (test_position)
 			do_test_position();
 		check_stdin();
Signature · 1 line
1.7.10.4
#3 lore raw
In reply to Sergey
Quoted 6 lines · Sergey

These patches originate from aplay -vvv /dev/zero not responding to Ctrl+C when playing to plugged "null", e.g.

  pcm.!default {
    type plug
    slave.pcm { type null }
    slave.format S32_LE
Quoted 8 lines

}

First patch fixes snd_pcm_write_areas() stuck in a loop because of transfer func() always returning 0. That caused application calling it to freeze. In the example above it have frozen aplay. If it was called by firefox, it would freeze firefox.

Is it a good way to fix it? Should snd_pcm_read_areas() be fixed like that too?

Nope: The whole culprit was in the null plugin - the stream was not started, thus new samples were not delivered.

I tried to fix this in:

http://git.alsa-project.org/?p=alsa-lib.git;a=commit;h=9c3086fb749cd5545be3f15cdde1592390006258

Quoted 1 line

Second patch fixes aplay itself, not returning from pcm_write() after SIGINT.

Thanks, looks good, but I added the in_aborting check to other functions, too:

http://git.alsa-project.org/?p=alsa-utils.git;a=commit;h=c06dbf077459923066c96556678835cd1f55123a

Quoted 16 lines

Additionally aplay may restore original interrupt handler, to make second Ctrl+C stop it for sure. In case of other bugs in alsa-lib or third-party slave pcms it seems better than SIGkilling it manually.

Comments and suggestions are welcome.

Regardless of these patches' fate null plugin must be fixed on its own.

Patches (2):
pcm: break out of snd_pcm_write_areas() if write function returned 0
aplay: Escape from pcm_write() when in_aborting state

 src/pcm/pcm.c |    2 +-
 aplay/aplay.c |    2 +-
 2 files changed, 2 insertions(+), 2 deletions(-)
Signature · 3 lines
Jaroslav Kysela <[email protected]>
Linux Kernel Sound Maintainer
ALSA Project; Red Hat, Inc.
author
#4 lore raw
Quoted 12 lines · Fri, 13 Jun 2014 Jaroslav Kysela wrote:
Quoted 6 lines

First patch fixes snd_pcm_write_areas() stuck in a loop because of transfer func() always returning 0. That caused application calling it to freeze. In the example above it have frozen aplay. If it was called by firefox, it would freeze firefox.

Is it a good way to fix it? Should snd_pcm_read_areas() be fixed like that too?

Nope: The whole culprit was in the null plugin - the stream was not started, thus new samples were not delivered.

I tried to fix this in:

Quoted 8 lines

http://git.alsa-project.org/?p=alsa-lib.git;a=commit;h=9c3086fb749cd5545be3f15cdde1592390006258

Quoted 1 line

Second patch fixes aplay itself, not returning from pcm_write() after SIGINT.

Thanks, looks good, but I added the in_aborting check to other functions, too:

http://git.alsa-project.org/?p=alsa-utils.git;a=commit;h=c06dbf077459923066c96556678835cd1f55123a

It works great. Thank you!

PS: What do you think about that small aplay patch?:
http://mailman.alsa-project.org/pipermail/alsa-devel/2014-June/077665.html
aplay -vvv output looks weird without it.

Signature · 3 lines
  Sergey

fprintf(stderr, "%-20.*s %i%%\n", perc[0]/5, "####################", perc[0]);
#5 lore raw
Quoted 21 lines · Sergey
Quoted 19 lines · Fri, 13 Jun 2014 Jaroslav Kysela wrote:
Quoted 6 lines

First patch fixes snd_pcm_write_areas() stuck in a loop because of transfer func() always returning 0. That caused application calling it to freeze. In the example above it have frozen aplay. If it was called by firefox, it would freeze firefox.

Is it a good way to fix it? Should snd_pcm_read_areas() be fixed like that too?

Nope: The whole culprit was in the null plugin - the stream was not started, thus new samples were not delivered.

I tried to fix this in:

http://git.alsa-project.org/?p=alsa-lib.git;a=commit;h=9c3086fb749cd5545be3f15cdde1592390006258

Quoted 1 line

Second patch fixes aplay itself, not returning from pcm_write() after SIGINT.

Thanks, looks good, but I added the in_aborting check to other functions, too:

Quoted 7 lines
Quoted 1 line

http://git.alsa-project.org/?p=alsa-utils.git;a=commit;h=c06dbf077459923066c96556678835cd1f55123a

It works great. Thank you!

PS: What do you think about that small aplay patch?:
http://mailman.alsa-project.org/pipermail/alsa-devel/2014-June/077665.html
aplay -vvv output looks weird without it.

Thanks. Applied now.

Jaroslav

Signature · 3 lines
Jaroslav Kysela <[email protected]>
Linux Kernel Sound Maintainer
ALSA Project; Red Hat, Inc.