Re: [Jack-Devel] Jack 1.9.7 on ARM crashes when killing a client

On Mon, Feb 14, 2011 at 10:31:14AM +0100, Stéphane Letz wrote:
> 
> Le 11 févr. 2011 à 17:34, Stéphane Letz a écrit :
> 
> > 
> > Le 11 févr. 2011 à 17:19, Valerio Pilo a écrit :
> > 
> >> In data venerdì 11 febbraio 2011 17:07:18, hai scritto:
> >>>> I just tried Jack 0.118 on the ARM box - it works without a problem, using this cmdline:
> >>>>> jackd -v -r -t 2000 -d alsa -d chrouter -D -r 44100 -S -p 2048 -n 4
> >>> 
> >>> Any log of jack1 when application is killed ?
> >> 
> >> You can find it attached.
> > 
> > OK. It seem that JACK1 ALSA backend does not trigger the same error (JACK2 ALSA backend was printing "ALSA: could not complete playback"), but there is no similar error message in JACK1 log.
> > 
> > So... back to square one. Either :
> > 
> > - there is a real issue in this ALSA backend on ARM
> > 
> > - or we consider again this "ALSA: could not complete playback" ALSA backend error as a "recoverable" one at JackAudioDriver level, and we revert part of  r 4079  commit, so that the ALSA Xrun cause a "skipped" cycle, but the server continue to run.
> > 
> > What do other jack hackers think?
> > 
> > Stéphane 
> > 
> 
> Since I cannot reproduce the original issue here on my Linux machine, I cannot go much further. To summarize:
> 
> - when JACK2 in synchronous mode, the server does : blocking ALSA read (actually, wait + read), graph activation, wait for graph to complete, write ALSA. 
> 
> - when a client is killed, it may fail to correctly complete a cycle, before the server detect the client has been removed, and cleanup the graph.  The server has a timeout to decide when the graph has really failed before trying to write back in ALSA. Since the server is late, ALSA backend issue a  "ALSA: could not complete playback" error in alsa_driver_write. This is considered as a "non recoverable error" and the backend wrapping thread is just stopped.
> 
> - how does JACK1 behave is a similar situation? I guess there is a "graph late" problem also? Is there a timeout to decide when the graph has really failed? why is the "ALSA: could not complete playback" error not triggered in this precise case?

jack1 will be late too.
but jack1 does not try to write to the driver when the graph is late.
the next alsa wait will detect the xrun and recover.
so the next cycle, things should be fine again.


> 
> Thanks.
> 
> Stéphane 
> 

-- 
torben Hohn

1297691732.17940_0.ltw:2,a <20110214135512.GB3055 at siel dot b>