[alsa-devel] [PATCH 05/10] ASoC: twl6040: One workqueue should be enough

[Péter Ujfalusi]

On Monday 26 September 2011 22:29:15 Mark Brown wrote:
> On Mon, Sep 26, 2011 at 08:20:11PM +0300, Ujfalusi, Peter wrote:
> > For the ramp code execution we should avoid any delay at all costs,
> > since it can have audible side effects.
> 
> If you're that sensitive to latency does a workqueue offer sufficient
> guarantees?

It seams to be pretty good based on my tests.

> > In my past experience the system wq can cause latency, since we usually
> > have slow devices on it (GPS, proximity sensor, magnetometer, etc).
> > Reading from those usually takes considerable amount of time, and
> > we can end up with delayed execution of our ramp code.
> 
> How long are these work items taking?  I'm wondering if the other work
> items are taking sufficiently long to be actively disruptive if they
> should be isolated from other things rather than the other way around.
> Of course it could also be a large buildup of work...

It could be any random driver (we can have random components on a device).
This is just precaution to protect the audio from nasty things.

--
Péter