Alsa deadlock fix (was Re: 2.6-test8 : alsa hangs my box)

Alsa deadlock fix (was Re: 2.6-test8 : alsa hangs my box)

[Kevin Puetz]

I am seeing this hang also, on a different card, whenever a native alsa and an 
OSS-emulation app collide. No nVidia stuff this time, and not snd_intel8x0 
either, this is w/ a via82xx motherboard's integrated sound. From what little 
one can tell from the sysrq-T backtrace, it looks like a classic deadlock 
(both tasks stick in down(), probably on different mutexes). 

It also looks like it has alreay been fixed (in alsa 0.9.7b) - I think the 
kernel's alsa is approximately equivalent to 0.9.6 these days? So, it looks 
like either some work to port that patch or a wholesale merge of alsa 0.9.8 
would be good before 2.6.0... probably the former given Linus's 
strongly-worded comments with -test9 :-) 

But a deadlock that results in two unkillable state 'D' processes whenever you 
an OSS and an alsa-native app access the soundcard at the same time, which 
happens on both VIA and Intel integrated sound, seems like a hang which might 
meet his criteria of problems which "causes lockups or just basic 
nonworkingness: and this happens on hardware that normal people are expected 
to have". So I figured it's worth asking if someone who already knows what 
the fix consisted of could push it toward Linus ...

Re: Alsa deadlock fix

[Måns Rullgård]

Kevin Puetz <puetzk@puetzk.org> writes:

> But a deadlock that results in two unkillable state 'D' processes
> whenever you an OSS and an alsa-native app access the soundcard at
> the same time, which happens on both VIA and Intel integrated sound,
> seems like a hang which might meet his criteria of problems which
> "causes lockups or just basic nonworkingness: and this happens on
> hardware that normal people are expected to have". So I figured it's
> worth asking if someone who already knows what the fix consisted of
> could push it toward Linus ...

It's fixed in -test9.

-- 
Måns Rullgård
mru@kth.se

Re: Alsa deadlock fix

[Kevin Puetz]

Måns Rullgård wrote:

> Kevin Puetz <puetzk@puetzk.org> writes:
> 
>> But a deadlock that results in two unkillable state 'D' processes
>> whenever you an OSS and an alsa-native app access the soundcard at
>> the same time, which happens on both VIA and Intel integrated sound,
>> seems like a hang which might meet his criteria of problems which
>> "causes lockups or just basic nonworkingness: and this happens on
>> hardware that normal people are expected to have". So I figured it's
>> worth asking if someone who already knows what the fix consisted of
>> could push it toward Linus ...
> 
> It's fixed in -test9.
> 
Thanks, sorry about the bogus report then. I looked at the -test8 -> -test9
changes before I sent it, but missed it since it was from akpm (not
jaroslav) and didn't mention alsa in the changelog message. Thus it missed
both categories I thought to search for in the applied changes :-P