Skip to contentSkip to conversation

Hang in snd_pcm_writei with alsa-pulse plugin

DISCUSSION Matthew Gregan 3 participants 10 messages Nov 1, 2010 – May 6, 2011

lore
Share & tools
lore
#0 lore raw

Hi,

I think I'm seeing a bug in the alsa-pulse plugin where the buffer management ends up corrupt and results in a deadlock waiting for free buffer space. This occurs when resuming from pause using snd_pcm_pause. After resuming, my application tries to write a fixed block of data, expecting snd_pcm_writei to block if the data is larger than the available buffer size (the result of snd_pcm_avail_update).

I originally observed this in the wild in Firefox, which pauses and resumes the sound device whenever network buffering occurs. I'm planning to include the workaround mentioned below in the next Firefox release (Mozilla bug 573924).

What happens is that, after resuming with snd_pcm_pause, a call to snd_pcm_writei never returns. This happens on the write call that would have exceeded the available buffer size, which I would expect to block only until sufficient buffer space became available.

It's possible to get into a similar situation using SND_PCM_NONBLOCK and waiting on the sound device if it returns EAGAIN, except that snd_pcm_writei always returns EAGAIN and snd_pcm_wait returns 1 immediately, resulting in a tight loop in the calling code.

I discovered that I can reliably workaround the problem by ensuring the first writes after resuming from pause are never larger than what snd_pcm_avail_update returns. After writing enough to fill (but not exceed) the available buffer size, the code returns to the fixed buffer size per write strategy and continues as normal.

The problem occurs with the following stack:

#0 __poll (fds=<value optimized out>,
nfds=<value optimized out>, timeout=<value optimized out>) at ../sysdeps/unix/sysv/linux/poll.c:87
#1 snd1_pcm_wait_nocheck (pcm=0x1b9a780, timeout=-1)

    at pcm.c:2367
#2  snd1_pcm_write_areas (pcm=0x1b9a780, 
    areas=0x7fff4ce9b890, offset=<value optimized out>, size=30000, 
    func=0x339ba91d10 <ioplug_priv_transfer_areas>) at pcm.c:6655
#3  snd_pcm_ioplug_writei (pcm=0x1b9a780, 
    buffer=<value optimized out>, size=30000) at pcm_ioplug.c:561
#4  bwrite (pcm=0x1b9a780, towrite=30000) at
    atest2.c:29

#5 main (argc=1, argv=0x7fff4ce9ba68) at atest2.c:86

I'm Fedora 13 x86_64 with all updates from updates-testing. ALSA is 1.0.22-1, PulseAudio is 0.9.21-6, and the kernel is 2.6.34.7-61. I've also tested against the current git versions of alsa-libs and alsa-plugins and can still reproduce the problem.

I've attached a simple test program that reproduces this problem reliably on my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within 2-3 pause/resume cycles. Running the test with "-r" enables the recovery code I mentioned above. It never hangs when tested using the hardware ALSA backend with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

Cheers,
-mjg

Signature · 2 lines
Matthew Gregan                     |/
                                  /|                    [email protected]

#include <assert.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <alsa/asoundlib.h>

#define RATE 48000

snd_pcm_uframes_t setup(snd_pcm_t ** pcm)
{
snd_pcm_uframes_t bsize, psize;

  assert(snd_pcm_open(pcm, "default", SND_PCM_STREAM_PLAYBACK, 0) == 0);
  assert(snd_pcm_set_params(*pcm, SND_PCM_FORMAT_S16_LE,
                            SND_PCM_ACCESS_RW_INTERLEAVED,
                            1, RATE, 1, 250000) == 0);
  assert(snd_pcm_get_params(*pcm, &bsize, &psize) == 0);
  assert(bsize / psize >= 4);

return psize;
}

void bwrite(snd_pcm_t * pcm, snd_pcm_sframes_t towrite) {
int r;
char * garbage;

  garbage = calloc(1, towrite);
  r = snd_pcm_writei(pcm, garbage, towrite);
  free(garbage);
  if (r < 0) {
    fprintf(stderr, "bwrite error recovery\n");
    snd_pcm_recover(pcm, r, 1);

}
}

snd_pcm_sframes_t fill(snd_pcm_t * pcm)
{
snd_pcm_sframes_t avail;

  avail = snd_pcm_avail_update(pcm);
  if (avail < 0) {
    fprintf(stderr, "fill error recovery\n");
    snd_pcm_recover(pcm, avail, 1);
    avail = snd_pcm_avail_update(pcm);
  }
  if (avail) {
    bwrite(pcm, avail);
  }

return avail;
}

int main(int argc, char * argv[])
{
snd_pcm_t * pcm;
snd_pcm_uframes_t psize;
int pause = 0, count = 0, recovery = 0;

  if (argc == 2 && argv[1][0] == '-' && argv[1][1] == 'r') {
    recovery = 1;
  }

/* set up a mono output device at RATE Hz with at least 4 periods */ psize = setup(&pcm);

/* prefill sound buffers and begin playback */
fill(pcm);

  while (++count) {
    if (pause || count % 7 == 0) {
      fprintf(stderr, pause ? "resuming playback\n" : "pausing playback\n");
      pause = !pause;
      snd_pcm_pause(pcm, pause);
      /* if resuming playback and recovery enabled, fill sound buffers */
      if (recovery && !pause) {
        fprintf(stderr, "recovery, wrote %d frames\n", (int) fill(pcm));
      }
    }

    /* if not paused, write a period of sound data */
    if (!pause) {
      snd_pcm_sframes_t avail = snd_pcm_avail_update(pcm);
      bwrite(pcm, psize);
      fprintf(stderr, "playback, wrote %u frames (needed %d)\n", (unsigned int) psize, (int) avail);
    }

    /* sleep for half a period */
    usleep(psize * 1000000 / RATE / 2);

}

return 0;
}

_______________________________________________
Alsa-devel mailing list
[email protected]
http://mailman.alsa-project.org/mailman/listinfo/alsa-devel

#1 lore raw

'Twas brillig, and Matthew Gregan at 01/11/10 01:38 did gyre and gimble:

Quoted 51 lines

Hi,

I think I'm seeing a bug in the alsa-pulse plugin where the buffer management ends up corrupt and results in a deadlock waiting for free buffer space. This occurs when resuming from pause using snd_pcm_pause. After resuming, my application tries to write a fixed block of data, expecting snd_pcm_writei to block if the data is larger than the available buffer size (the result of snd_pcm_avail_update).

I originally observed this in the wild in Firefox, which pauses and resumes the sound device whenever network buffering occurs. I'm planning to include the workaround mentioned below in the next Firefox release (Mozilla bug 573924).

What happens is that, after resuming with snd_pcm_pause, a call to snd_pcm_writei never returns. This happens on the write call that would have exceeded the available buffer size, which I would expect to block only until sufficient buffer space became available.

It's possible to get into a similar situation using SND_PCM_NONBLOCK and waiting on the sound device if it returns EAGAIN, except that snd_pcm_writei always returns EAGAIN and snd_pcm_wait returns 1 immediately, resulting in a tight loop in the calling code.

I discovered that I can reliably workaround the problem by ensuring the first writes after resuming from pause are never larger than what snd_pcm_avail_update returns. After writing enough to fill (but not exceed) the available buffer size, the code returns to the fixed buffer size per write strategy and continues as normal.

The problem occurs with the following stack:

#0 __poll (fds=<value optimized out>,
nfds=<value optimized out>, timeout=<value optimized out>) at ../sysdeps/unix/sysv/linux/poll.c:87
#1 snd1_pcm_wait_nocheck (pcm=0x1b9a780, timeout=-1)

    at pcm.c:2367
#2  snd1_pcm_write_areas (pcm=0x1b9a780, 
    areas=0x7fff4ce9b890, offset=<value optimized out>, size=30000, 
    func=0x339ba91d10 <ioplug_priv_transfer_areas>) at pcm.c:6655
#3  snd_pcm_ioplug_writei (pcm=0x1b9a780, 
    buffer=<value optimized out>, size=30000) at pcm_ioplug.c:561
#4  bwrite (pcm=0x1b9a780, towrite=30000) at
    atest2.c:29

#5 main (argc=1, argv=0x7fff4ce9ba68) at atest2.c:86

I'm Fedora 13 x86_64 with all updates from updates-testing. ALSA is 1.0.22-1, PulseAudio is 0.9.21-6, and the kernel is 2.6.34.7-61. I've also tested against the current git versions of alsa-libs and alsa-plugins and can still reproduce the problem.

Quoted 8 lines

I've attached a simple test program that reproduces this problem reliably on my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within 2-3 pause/resume cycles. Running the test with "-r" enables the recovery code I mentioned above. It never hangs when tested using the hardware ALSA backend with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

Reproduced here I presume:
[colin@jimmy pulseaudio (master)]$ ./atest2
playback, wrote 3000 frames (needed 0)
playback, wrote 3000 frames (needed 1480)
playback, wrote 3000 frames (needed 2643)
playback, wrote 3000 frames (needed 3526)
playback, wrote 3000 frames (needed 526)
playback, wrote 3000 frames (needed 1313)
pausing playback
resuming playback
^C
[colin@jimmy pulseaudio (master)]$ ./atest2
playback, wrote 3000 frames (needed 0)
playback, wrote 3000 frames (needed 1584)
playback, wrote 3000 frames (needed 2764)
playback, wrote 3000 frames (needed 3648)
playback, wrote 3000 frames (needed 648)
playback, wrote 3000 frames (needed 1426)
pausing playback
resuming playback
^C

This is with latest PA from stable-queue.

Not sure whether this is "expected" or not, but it's probably worth you posting this on PA devel list too for some further thought if this thread doesn't garner much response here.

Cheers
Col

Signature · 10 lines
Colin Guthrie
gmane(at)colin.guthr.ie
http://colin.guthr.ie/

Day Job:
  Tribalogic Limited [http://www.tribalogic.net/]
Open Source:
  Mageia Contributor [http://www.mageia.org/]
  PulseAudio Hacker [http://www.pulseaudio.org/]
  Trac Hacker [http://trac.edgewall.org/]
#2 lore raw

2010/11/2 Colin Guthrie <[email protected]>

Quoted 96 lines

'Twas brillig, and Matthew Gregan at 01/11/10 01:38 did gyre and gimble:

Quoted 4 lines

Hi,

I think I'm seeing a bug in the alsa-pulse plugin where the buffer management ends up corrupt and results in a deadlock waiting for free

buffer

Quoted 3 lines

space. This occurs when resuming from pause using snd_pcm_pause. After resuming, my application tries to write a fixed block of data, expecting snd_pcm_writei to block if the data is larger than the available buffer

size

Quoted 3 lines

(the result of snd_pcm_avail_update).

I originally observed this in the wild in Firefox, which pauses and

resumes

Quoted 1 line

the sound device whenever network buffering occurs. I'm planning to

include

Quoted 6 lines

the workaround mentioned below in the next Firefox release (Mozilla bug 573924).

What happens is that, after resuming with snd_pcm_pause, a call to snd_pcm_writei never returns. This happens on the write call that would have exceeded the available buffer size, which I would expect to block

only

Quoted 4 lines

until sufficient buffer space became available.

It's possible to get into a similar situation using SND_PCM_NONBLOCK and waiting on the sound device if it returns EAGAIN, except that

snd_pcm_writei

Quoted 1 line

always returns EAGAIN and snd_pcm_wait returns 1 immediately, resulting

in a

Quoted 5 lines

tight loop in the calling code.

I discovered that I can reliably workaround the problem by ensuring the first writes after resuming from pause are never larger than what snd_pcm_avail_update returns. After writing enough to fill (but not

exceed)

Quoted 17 lines

the available buffer size, the code returns to the fixed buffer size per write strategy and continues as normal.

The problem occurs with the following stack:

#0 __poll (fds=<value optimized out>,
nfds=<value optimized out>, timeout=<value optimized out>) at ../sysdeps/unix/sysv/linux/poll.c:87
#1 snd1_pcm_wait_nocheck (pcm=0x1b9a780, timeout=-1)

    at pcm.c:2367
#2  snd1_pcm_write_areas (pcm=0x1b9a780,
    areas=0x7fff4ce9b890, offset=<value optimized out>, size=30000,
    func=0x339ba91d10 <ioplug_priv_transfer_areas>) at pcm.c:6655
#3  snd_pcm_ioplug_writei (pcm=0x1b9a780,
    buffer=<value optimized out>, size=30000) at pcm_ioplug.c:561
#4  bwrite (pcm=0x1b9a780, towrite=30000) at
    atest2.c:29
Quoted 4 lines

#5 main (argc=1, argv=0x7fff4ce9ba68) at atest2.c:86

I'm Fedora 13 x86_64 with all updates from updates-testing. ALSA is 1.0.22-1, PulseAudio is 0.9.21-6, and the kernel is 2.6.34.7-61. I've

also

Quoted 4 lines

tested against the current git versions of alsa-libs and alsa-plugins and can still reproduce the problem.

I've attached a simple test program that reproduces this problem reliably

on

Quoted 3 lines

my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within

2-3

Quoted 1 line

pause/resume cycles. Running the test with "-r" enables the recovery

code I

Quoted 1 line

mentioned above. It never hangs when tested using the hardware ALSA

backend

Quoted 2 lines

with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

Reproduced here I presume:
[colin@jimmy pulseaudio (master)]$ ./atest2
playback, wrote 3000 frames (needed 0)
playback, wrote 3000 frames (needed 1480)
playback, wrote 3000 frames (needed 2643)
playback, wrote 3000 frames (needed 3526)
playback, wrote 3000 frames (needed 526)
playback, wrote 3000 frames (needed 1313)
pausing playback
resuming playback
^C
[colin@jimmy pulseaudio (master)]$ ./atest2
playback, wrote 3000 frames (needed 0)
playback, wrote 3000 frames (needed 1584)
playback, wrote 3000 frames (needed 2764)
playback, wrote 3000 frames (needed 3648)
playback, wrote 3000 frames (needed 648)
playback, wrote 3000 frames (needed 1426)
pausing playback
resuming playback
^C

Quoted 7 lines

This is with latest PA from stable-queue.

Not sure whether this is "expected" or not, but it's probably worth you posting this on PA devel list too for some further thought if this thread doesn't garner much response here.

No much response because there are bugs in atest2.c

/* prefill sound buffers and begin playback */
fill(pcm);

  while (++count) {

The program had filled the buffer but the output does not indicate those write

I can confirm that the program seem hang after a few pause/unpause when using alsa-pulse plugin

However it assert when using hw device

  assert(bsize / psize >= 4);
#3 lore raw
Quoted 9 lines · Raymond Yau

No much response because there are bugs in atest2.c

/* prefill sound buffers and begin playback */
fill(pcm);

  while (++count) {

The program had filled the buffer but the output does not indicate those write

This doesn't affect the result of the testcase. count is only used to pause and resume less frequently than every iteration of the write loop.

Quoted 6 lines

I can confirm that the program seem hang after a few pause/unpause when using alsa-pulse plugin

However it assert when using hw device

  assert(bsize / psize >= 4);

The assert is present because I've only tested on systems where this assertion holds true. It's likely that the loop timing would need to be changed to work correctly in other cases.

Thanks for confirming that you can reproduce the problem. I've since discovered that it's possible to produce the same problem with the PulseAudio API directly, so I'll take this up on pulseaudio-discuss@.

Cheers,
-mjg

Signature · 2 lines
Matthew Gregan                     |/
                                  /|                    [email protected]
#4 lore raw

2010/11/2 Matthew Gregan <[email protected]>

Quoted 26 lines
Quoted 9 lines · Raymond Yau

No much response because there are bugs in atest2.c

/* prefill sound buffers and begin playback */
fill(pcm);

  while (++count) {

The program had filled the buffer but the output does not indicate those write

This doesn't affect the result of the testcase. count is only used to pause
and resume less frequently than every iteration of the write loop.

Quoted 6 lines

I can confirm that the program seem hang after a few pause/unpause when using alsa-pulse plugin

However it assert when using hw device

  assert(bsize / psize >= 4);

The assert is present because I've only tested on systems where this assertion holds true. It's likely that the loop timing would need to be changed to work correctly in other cases.

Quoted 7 lines

Thanks for confirming that you can reproduce the problem. I've since discovered that it's possible to produce the same problem with the PulseAudio API directly, so I'll take this up on pulseaudio-discuss@.

Cheers,
-mjg

The interesting thing is it did not hang when I increase the sleep time to one period

    /* sleep for half a period */

+ usleep(psize * 1000000 / RATE );
- usleep(psize * 1000000 / RATE /2 );

#6 lore raw
In reply to Matthew Gregan

2010/11/1 Matthew Gregan <[email protected]>

Quoted 16 lines

Hi,

I've attached a simple test program that reproduces this problem reliably on
my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within 2-3 pause/resume cycles. Running the test with "-r" enables the recovery code I
mentioned above. It never hangs when tested using the hardware ALSA backend
with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

when increase the sleep time from half period to one period, the hang occur after around 50 pause/resume cycles.

did PA server stop sending data to the sound card immediately when your program pause ?

But cannot reproduce your problem using the "pause" function of aplay

which sound card are you using and did you change the pulseaudlo daemon.conf ?

#7 lore raw
Quoted 5 lines · Raymond Yau

when increase the sleep time from half period to one period, the hang occur after around 50 pause/resume cycles.

did PA server stop sending data to the sound card immediately when your program pause ?

Yes. As far as I can tell, at least.

Quoted 4 lines

But cannot reproduce your problem using the "pause" function of aplay

which sound card are you using and did you change the pulseaudlo daemon.conf ?

I'm using the default daemon.conf shipped with Fedora 13, which contains nothing but comments and newlines. The sound card is: Intel Corporation 5 Series/3400 Series Chipset High Definition Audio (rev 06) in a Lenovo W510. I've seen this problem on other machines (including in VMWare and VirtualBox based VMs) too, but I don't know what sound hardware they were using.

Cheers,
-mjg

Signature · 2 lines
Matthew Gregan                     |/
                                  /|                    [email protected]
#5 lore raw
In reply to Matthew Gregan

2010/11/2 Matthew Gregan <[email protected]>

Quoted 10 lines

At 2010-11-02T10:43:54+0800, Raymond Yau wrote:

Thanks for confirming that you can reproduce the problem. I've since discovered that it's possible to produce the same problem with the PulseAudio API directly, so I'll take this up on pulseaudio-discuss@.

Cheers,
-mjg

A buggy PA client can easily crash the PA server

http://0pointer.de/lennart/projects/pulseaudio/doxygen/sample.html

PulseAudio supports any sample rate between 1 Hz and 192000 Hz

http://thread.gmane.org/gmane.comp.audio.pulseaudio.general/7759

Just change the rate of your program ptest.c from 48000 Hz to 1 Hz , this will abort the PA server

E: sink-input.c: Assertion 'tchunk.length > 0' failed at pulsecore/sink-input.c:551, function pa_sink_input_peek(). Aborting.

Seem that PA server did not check whether the info send by the PA client is really supported by the server

#8 lore raw
In reply to Matthew Gregan

2010/11/1 Matthew Gregan <[email protected]>

Quoted 17 lines

Hi,

I think I'm seeing a bug in the alsa-pulse plugin where the buffer management ends up corrupt and results in a deadlock waiting for free buffer
space. This occurs when resuming from pause using snd_pcm_pause. After resuming, my application tries to write a fixed block of data, expecting snd_pcm_writei to block if the data is larger than the available buffer size
(the result of snd_pcm_avail_update).

I originally observed this in the wild in Firefox, which pauses and resumes the sound device whenever network buffering occurs. I'm planning to include
the workaround mentioned below in the next Firefox release (Mozilla bug 573924).

Quoted 5 lines

What happens is that, after resuming with snd_pcm_pause, a call to snd_pcm_writei never returns. This happens on the write call that would have exceeded the available buffer size, which I would expect to block only until sufficient buffer space became available.

It depend on whether pulse_write wait until all requested frames are played or just put to the playback ring buffer

especailly when your program write audio more than the available free buffer space

http://www.alsa-project.org/alsa-doc/alsa-lib/group___p_c_m.htm

If the blocking behaviour is selected and it is running, then routine waits until all requested frames are played or put to the playback ring buffer. The returned number of frames can be less only if a signal or underrun occurred.

If the non-blocking behaviour is selected, then routine doesn't wait at all.

Quoted 54 lines

It's possible to get into a similar situation using SND_PCM_NONBLOCK and waiting on the sound device if it returns EAGAIN, except that snd_pcm_writei
always returns EAGAIN and snd_pcm_wait returns 1 immediately, resulting in a
tight loop in the calling code.

I discovered that I can reliably workaround the problem by ensuring the first writes after resuming from pause are never larger than what snd_pcm_avail_update returns. After writing enough to fill (but not exceed)
the available buffer size, the code returns to the fixed buffer size per write strategy and continues as normal.

The problem occurs with the following stack:

#0 __poll (fds=<value optimized out>,
nfds=<value optimized out>, timeout=<value optimized out>) at ../sysdeps/unix/sysv/linux/poll.c:87
#1 snd1_pcm_wait_nocheck (pcm=0x1b9a780, timeout=-1)
at pcm.c:2367
#2 snd1_pcm_write_areas (pcm=0x1b9a780,
areas=0x7fff4ce9b890, offset=<value optimized out>, size=30000, func=0x339ba91d10 <ioplug_priv_transfer_areas>) at pcm.c:6655
#3 snd_pcm_ioplug_writei (pcm=0x1b9a780,
buffer=<value optimized out>, size=30000) at pcm_ioplug.c:561
#4 bwrite (pcm=0x1b9a780, towrite=30000) at
atest2.c:29
#5 main (argc=1, argv=0x7fff4ce9ba68) at atest2.c:86

I'm Fedora 13 x86_64 with all updates from updates-testing. ALSA is 1.0.22-1, PulseAudio is 0.9.21-6, and the kernel is 2.6.34.7-61. I've also tested against the current git versions of alsa-libs and alsa-plugins and can still reproduce the problem.

I've attached a simple test program that reproduces this problem reliably on
my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within 2-3 pause/resume cycles. Running the test with "-r" enables the recovery code I
mentioned above. It never hangs when tested using the hardware ALSA backend
with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

Cheers,
-mjg
--
Matthew Gregan

#9 lore raw
In reply to Matthew Gregan

2010/11/1 Matthew Gregan <[email protected]>

Quoted 69 lines

Hi,

I think I'm seeing a bug in the alsa-pulse plugin where the buffer management ends up corrupt and results in a deadlock waiting for free buffer
space. This occurs when resuming from pause using snd_pcm_pause. After resuming, my application tries to write a fixed block of data, expecting snd_pcm_writei to block if the data is larger than the available buffer size
(the result of snd_pcm_avail_update).

I originally observed this in the wild in Firefox, which pauses and resumes the sound device whenever network buffering occurs. I'm planning to include
the workaround mentioned below in the next Firefox release (Mozilla bug 573924).

What happens is that, after resuming with snd_pcm_pause, a call to snd_pcm_writei never returns. This happens on the write call that would have exceeded the available buffer size, which I would expect to block only until sufficient buffer space became available.

It's possible to get into a similar situation using SND_PCM_NONBLOCK and waiting on the sound device if it returns EAGAIN, except that snd_pcm_writei
always returns EAGAIN and snd_pcm_wait returns 1 immediately, resulting in a
tight loop in the calling code.

I discovered that I can reliably workaround the problem by ensuring the first writes after resuming from pause are never larger than what snd_pcm_avail_update returns. After writing enough to fill (but not exceed)
the available buffer size, the code returns to the fixed buffer size per write strategy and continues as normal.

The problem occurs with the following stack:

#0 __poll (fds=<value optimized out>,
nfds=<value optimized out>, timeout=<value optimized out>) at ../sysdeps/unix/sysv/linux/poll.c:87
#1 snd1_pcm_wait_nocheck (pcm=0x1b9a780, timeout=-1)
at pcm.c:2367
#2 snd1_pcm_write_areas (pcm=0x1b9a780,
areas=0x7fff4ce9b890, offset=<value optimized out>, size=30000, func=0x339ba91d10 <ioplug_priv_transfer_areas>) at pcm.c:6655
#3 snd_pcm_ioplug_writei (pcm=0x1b9a780,
buffer=<value optimized out>, size=30000) at pcm_ioplug.c:561
#4 bwrite (pcm=0x1b9a780, towrite=30000) at
atest2.c:29
#5 main (argc=1, argv=0x7fff4ce9ba68) at atest2.c:86

I'm Fedora 13 x86_64 with all updates from updates-testing. ALSA is 1.0.22-1, PulseAudio is 0.9.21-6, and the kernel is 2.6.34.7-61. I've also tested against the current git versions of alsa-libs and alsa-plugins and can still reproduce the problem.

I've attached a simple test program that reproduces this problem reliably on
my machine. It writes a period sized buffer in a loop, waiting half a period until the next attempt. Every few iterations, it pauses the sound device for half a period and then resumes it. It usually hangs within 2-3 pause/resume cycles. Running the test with "-r" enables the recovery code I
mentioned above. It never hangs when tested using the hardware ALSA backend
with alsa-pulse disabled, but my sound hardware doesn't seem to support snd_pcm_pause.

Quoted 1 line

using your test program and follow the instruction in

http://colin.guthr.ie/2010/09/compiling-and-running-pulseaudio-from-git/

It seem that PA server lost some of the audio at rewind when it resume from cork and uncork, if you enable DEBUG_TIMING, in alsa-sink.c PA server seem still writing

Are there any statistics about how much data received from the client and how much data PA server wrute to the sound card

: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 351936 bytes.
D: alsa-sink.c: before: 87984
D: alsa-sink.c: after: 87984
D: alsa-sink.c: Rewound 351936 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 2337
D: sink-input.c: Have to rewind 351936 bytes on render memblockq.
D: source.c: Processing rewind...
D: core-subscribe.c: Dropped redundant event due to change event.
D: reserve-wrap.c: Device lock status of reserve-monitor-wrapper@Audio1changed: not busy
D: protocol-dbus.c: Interface org.PulseAudio.Core1.Stream added for object /org/pulseaudio/core1/playback_stream0
D: protocol-native.c: Requesting rewind due to end of underrun.
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10032 bytes.
D: alsa-sink.c: before: 2508
D: alsa-sink.c: after: 2508
D: alsa-sink.c: Rewound 10032 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 2259
D: sink-input.c: Have to rewind 10032 bytes on render memblockq.
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10704 bytes.
D: alsa-sink.c: before: 2676
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: alsa-sink.c: after: 2676
D: alsa-sink.c: Rewound 10704 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 2012
D: source.c: Processing rewind...
D: protocol-native.c: Requesting rewind due to end of underrun.
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10672 bytes.
D: alsa-sink.c: before: 2668
D: alsa-sink.c: after: 2668
D: alsa-sink.c: Rewound 10672 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 2012
D: sink-input.c: Have to rewind 10672 bytes on render memblockq.
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: alsa-sink.c: Limited to 10672 bytes.
D: alsa-sink.c: before: 2668
D: alsa-sink.c: after: 2668
D: alsa-sink.c: Rewound 10672 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1720
D: source.c: Processing rewind...
D: protocol-native.c: Requesting rewind due to end of underrun.
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10672 bytes.
D: alsa-sink.c: before: 2668
D: alsa-sink.c: after: 2668
D: alsa-sink.c: Rewound 10672 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1690
D: sink-input.c: Have to rewind 10672 bytes on render memblockq.
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: protocol-native.c: Requesting rewind due to end of underrun.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: alsa-sink.c: Limited to 10704 bytes.
D: alsa-sink.c: before: 2676
D: alsa-sink.c: after: 2676
D: alsa-sink.c: Rewound 10704 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1457
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: alsa-sink.c: Limited to 10672 bytes.
D: alsa-sink.c: before: 2668
D: alsa-sink.c: after: 2668
D: alsa-sink.c: Rewound 10672 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1369
D: source.c: Processing rewind...
D: protocol-native.c: Requesting rewind due to end of underrun.
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10672 bytes.
D: alsa-sink.c: before: 2668
D: alsa-sink.c: after: 2668
D: alsa-sink.c: Rewound 10672 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1372
D: sink-input.c: Have to rewind 10672 bytes on render memblockq.
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: protocol-native.c: Requesting rewind due to end of underrun.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: alsa-sink.c: Limited to 10704 bytes.
D: alsa-sink.c: before: 2676
D: alsa-sink.c: after: 2676
D: alsa-sink.c: Rewound 10704 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1352
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes idle, timeout in 5 seconds.
D: protocol-native.c: Requesting rewind due to end of underrun.
D: sink-input.c: Requesting rewind due to uncorking
D: alsa-sink.c: Requested to rewind 352768 bytes.
D: module-suspend-on-idle.c: Sink alsa_output.1.analog-stereo becomes busy.
D: alsa-sink.c: Limited to 10704 bytes.
D: alsa-sink.c: before: 2676
D: alsa-sink.c: after: 2676
D: alsa-sink.c: Rewound 10704 bytes.
D: sink.c: Processing rewind...
D: sink.c: latency = 1409
D: source.c: Processing rewind...
D: sink-input.c: Requesting rewind due to corking