Debugging dmix?

[Mark Hills]

Recommendations for debugging dmix?

For a long time I have a problem where audio output regresses to glitches, 
only solved by closing all applications using audio and reopening.

Especially common when Chromium is used: 3-4 times a day, frustrating.

I have a very explicit dmix configuration (see below); no PulseAudio. 
Reproducable on latest alsa-lib (fb48ad9e; later than v1.2.2)

I can rule out _some_ aspects of hardware/drivers (Layla3G); eg. when the 
fault occurs, other soundcard channels are fine.

Any way to understand what dmix is doing? Clients? Control process?

Here's a recording taken from the analogue output (L+R chanel get zero'd):

  http://www.pogo.org.uk/~mark/tmp/dmix-glitch.png 
  http://www.pogo.org.uk/~mark/tmp/dmix-glitch.flac

Some process overwriting the ring buffer? Not high enough thread priority?

Seems like it might be more prevalant with more applications using sound. 
Maybe why Chromium makes it worse; several threads/sandboxes, perhaps.

dmix has all the trappings of something fiendishly clever and concise, but 
quite opaque. I'd need time to study the code (it has few comments, a lot 
of assumed knowledge) to understand its IPC. Can anyone give a head start?

Specifically, how does the final mix get written to the output? eg. Thread 
priorities, if that process goes away etc.

Thanks

---

pcm.layla_multi {
        type route

        slave.pcm {
                type dmix
                ipc_key 833282
                ipc_key_add_uid true
                slave {
                        pcm "hw:Layla3G,0,0"
                        period_size 256
                        buffer_size 32768
                        channels 4
                        rate 48000
                        format S24_3LE
                }
        }

        ttable {
                # Headphones

                0.0 1
                1.1 1

                # Speakers

                0.2 1
                1.3 1
        }
}

pcm.!default {
        type plug
        slave.pcm layla_multi
}

-- 
Mark