IMHO generally xruns cannot be avoided, it's a standard situation. The difference is that restarting the stream avoids the channel swap on VC4 HDMI, whereas standard xrun handling as recommended by alsa does not avoid the channel swap.My guess is a sample is being lost in the dma to hdmi audio fifo. That will result in a channel swap from that point onwards.
I'll need to check from logging whether when the underrun messages appear whether something unusual happens.
I'm suspecting it may trigger a dma channel abort, and restart which perhaps is where the sample gets lost.
Example of the recommended xrun handling is https://github.com/tiwai/alsa-lib/blob/ ... #L226-L246 - at xrun the driver calls snd_pcm_stop(SND_PCM_STATE_XRUN) (e.g. https://github.com/torvalds/linux/blob/ ... .c#L63-L66 ) , which returns SND_PCM_STATE_XRUN to the client, which subsequenty calls snd_pcm_prepare() of the driver. This sequence should avoid the channel swap too (but it does not).
Statistics: Posted by phofman — Tue Jan 30, 2024 6:03 pm
