Re: [alsa-devel] Timeout issues in wait_for_avail function

[Takashi Iwai]

On Thu, 31 May 2018 09:43:10 +0200,
Zengtao (B) wrote:
> 
> Hi Takashi:
> 
> Thank you for your reply.
> 
> >-----Original Message-----
> >From: Takashi Iwai [mailto:tiwai@xxxxxxx]
> >Sent: Sunday, May 13, 2018 6:40 PM
> >To: Zengtao (B) <prime.zeng@xxxxxxxxxxxxx>
> >Cc: perex@xxxxxxxx; tiwai@xxxxxxxx; alsa-devel@xxxxxxxxxxxxxxxx;
> >linux-kernel@xxxxxxxxxxxxxxx
> >Subject: Re: [alsa-devel] Timeout issues in wait_for_avail function
> >
> >On Mon, 07 May 2018 12:49:34 +0200,
> >Zengtao (B) wrote:
> >>
> >> Hi perex and tiwai:
> >>
> >> I have met a timeout case when capture audio from snd-usb-audio
> >> device, when the host call the pcm_read and get into the wait_for_avail
> >function.
> >> The following happends
> >> 1.  No available data for capture(maybe because of the late response
> >> audio data by the uac device)
> >
> >Hrm, in the case of capture, the data must be available.
> >If it's not the case, something is wrong.
> >
> >> 2.  The current thread falls into sleep state and no one wakes up it.
> >> 3.  The current thread will sleep 10s(schedule_timeout(1000)) and then
> >wakeup.
> >>
> >> I have two question about the wait_for_avail:
> >> 1. The timeout value too long, is it a reasonable value?
> >> 2. Is there any mechanism to wake up the thread if there is data from the
> >hw.
> >
> >The scenario above shouldn't happen, so no need for discussion.
> >Rather we should check why it's woken up even though no data is available.
> >
> 
> It really happens on my platform, and anyway 10 seconds timeout seems
> not a reasonable value.

The timeout isn't about guaranteeing anything.  It's merely a
last-resort safeguard.  So it *has to* be long, per definition.

That is, if you hit this timeout, it means that something is already
wrong.  In that case, fix the real cause instead of tweaking the
timeout.


Takashi

> And I don't there is any guarantee that there must be avail
> data when we reach wait_for_avail, in fact, inside the wait_for_avail, there is branch
> when no data is available.
> 
> >You can check the tracepoints to see the action of PCM stream, and confirm
> >whether it's really no data, or it's just lost by some reason (or looks as if so).
> >
> I trace the data flow, we reach the wait_for_avail function before the usb snd data arrives, 
> so we wait until 10 seconds timeout. 
> 
> >
> >Takashi
> 
> 
> Regards
> Zengtao