Re: [Jack-Devel] A couple of related issues in jackd on ARM

Le 8 janv. 2013 à 20:13, MONTANARO Luciano (MM) a écrit :

> Hi all again,
> I am having new problems with jackd2, after our embedded platform switched to the 3.1 kernel.
> Actually, I think they were present with our previous kernel as well, but they have become systematic while once they were sporadic.
> 
> I think I have found the problematic code, but I need some insight from someone that knows the code.
> The first problem I am seeing is a very high number of xrun notifications, tens of them a second.
> They is due to alsa_driver_wait() returning with 0 as its frame number, which normally means an xrun has occurred, but in this case what seems to be happening is that jack_wait returns "normally", but with a capture_avail of 0 and a playback_avail of 512 on 9 calls out of 10. The tenth call I get both capture_avail and playback_avail set to 512. The wait_status is set to 0, as in "everything normal", I guess, and so I tried to modify JackAlsaDriver to skip the notification in this case, and things seem to be working fine, no noticeable glitches in the audio playback at all.

Cannot answer here: what ALSA gurus think of the possible fix?

> 
> The second problem I was referring to is that this high number of notifications is triggering a problem in the JackRequest code:
> The CheckRes() macro checks for errors from the Read/Write calls, but these calls may silently fail: they can return without writing or reading, and this is accounted as success, but then Read() may return uninitialized data or Write is not retried when it should.
> 
> There is code like this:
> 
>         CheckRes(JackRequest::Write(trans));
>         CheckRes(trans->Write(&fName, sizeof(fName)));
>         CheckRes(trans->Write(&fProtocol, sizeof(int)));
>         CheckRes(trans->Write(&fOptions, sizeof(int)));
>         CheckRes(trans->Write(&fUUID, sizeof(int)));
> 
> Where I think after each Write there is potential for a needed retry, which is not done at all. Am I missing something?

CheckRes was supposed to fail in case of error?

Any log of the problem to show?

Stéphane 

1357674332.25723_0.ltw:2,a <48F7C588-39F8-4983-A384-5EAB28DE3F1C at grame dot fr>