race condition in snd_pcm_drain() sometimes hangs userspace processes

I'm seeing the same problem. (confirming)

bt of the hanged state of esd, stracing esd appears to remove the problem:
#0 0xffffe410 in ?? ()
#1 0xbfff9068 in ?? ()
#2 0xffffffff in ?? ()
#3 0x00000001 in ?? ()
#4 0xb7f1a4b5 in poll () from /lib/tls/i686/cmov/libc.so.6
#5 0xb7dd3ca1 in snd_pcm_wait_nocheck () from /usr/lib/libasound.so.2
#6 0xb7e09a2d in _snd_pcm_dmix_open () from /usr/lib/libasound.so.2
#7 0xb7dd1a9c in snd_pcm_drain () from /usr/lib/libasound.so.2
#8 0xb7dee0ea in _snd_pcm_rate_open () from /usr/lib/libasound.so.2
#9 0xb7dd1a9c in snd_pcm_drain () from /usr/lib/libasound.so.2
#10 0xb7fd12f9 in esd_audio_flush () from /usr/lib/libesd.so.0
#11 0x0804bdd0 in clean_exit ()
#12 0x0804a927 in clean_exit ()
#13 0xb7e72e8a in __libc_start_main () from /lib/tls/i686/cmov/libc.so.6
#14 0x08049751 in ?? ()
(gdb) cont
Continuing.

Oh, and it's not about daemonizing or not. I fired up 'esd -nobeep' in a
terminal and killed gnome-panel which then freezes and offers a quick way to debug.

do you use libesd0 or libesd-alsa0?

I've got libesd-alsa0 installed

Same here, libesd-alsa0 installed.

I can reproduce it on my iBook.

For the records:
 * Hang occurs with both hoary's and breezy's esound (daemon+libs), on both
Hoary and Breezy kernel.
 * Hang does not occur with libesd0, only with libesd-alsa0.
 * esd itself works perfectly (esdplay etc.)
 * It only seems to affect gnome applications.

This is not a regression of the newer 0.2.36 version. This is most probably a
bug that appears in conjunction with a newer Gnome library that is responsible
for sound events, however, we did not isolate that library yet.

As a temporary workaround, either use libesd0 or switch to polypaudio.

(In reply to comment #0)

> i also noticed that running esd from the cmdline never exits (process keeps
> foreground), and assume this is the cause of the login hang.

No, esd does not daemonize on its own, this is not the cause.

 esound (0.2.36-1ubuntu2) breezy; urgency=low
 .
   * audio_alsa09.c: Disable snd_pcm_drain() call for now since it sometimes
     triggers a race condition in the ALSA kernel driver which causes the
     esd process to hang. This is not a proper bug fix, but at least makes all
     those broken Colony installations out there work again. (Ubuntu #12276)

The workaround is uploaded, but it's not a final fix. Readjusting the bug
details for the real cause.

please stop reassigning bugs at the wrong place

Submitted upstream:

    https://bugtrack.alsa-project.org/alsa-bug/view.php?id=1306

Upstream says that a bug like this was fixed in ALSA 1.0.9a.
In which versions of ALSA has the race condition been found?

$ cat /proc/asound/version
Advanced Linux Sound Architecture Driver Version 1.0.9 (Sun May 29 07:31:02 2005
UTC).

I see that the kernel has version 1.0.9. The bug was (probably) fixed in 1.0.9a.

/usr/share/doc/alsa-base/changelog.gz:
Changelog between 1.0.9 and 1.0.9a releases
  + PCM Midlevel
    - Fix races between PCM drain and other ops

It seems that the ALSA drivers in the kernel need to be updated.

Fix in our current alsa.