[alsa-devel] [RFC] ALSA: Reduce delay for a blocking pcm_drain.

[Clemens Ladisch]

Dylan Reid wrote:
> This patch addresses two issues related to calling snd_pcm_drain.
>
> If no_period_wakeup is set, then snd_pcm_drain would wait
> MAX_SCHEDULE_TIMEOUT jiffies, without a wakeup pending this will leave
> the calling task blocked indefinitely.

no_period_wakeup is used by applications that do _not_ want to use any
of ALSA's blocking functions (and thus want to avoid the interrupt
overhead) and use their own timers instead.  This also applies to
snd_pcm_drain.

Is there any actual application that tries to use snd_pcm_drain together
with no_period_wakeup?

> Also if the stream is running with period wakeups but with a long
> period, the delay could be seconds.  If only a small part of the
> buffer is being used, this is unnecessary.

Wakeup at period boundaries is part of the ALSA API.  (This is all what
periods are for.)

> Instead wait for the remaining samples to play out, plus one
> millisecond.

The device's clock and the Linux system clock might have larger
differences.  The only reliable synchronization source is the period
interrupt.


Regards,
Clemens