On Mon, Mar 23, 2015 at 06:45:41PM +0100, Stefan Wahren wrote:

> > So, this isn't actually quite a busy wait because we do a schedule()
> > rather than a cpu_relax() but still it could devolve into that - 20ms
> > seems a long time to burn doing that. If we're expecting this to finish
> > very quickly can we do an initial busy wait then fall back to something
> > with an actual sleep or soemthing?

> We will need to keep the initial usleep_range(). So do you think of the
> following:

> 1. usleep_range
> 2. busy_wait for BM_POWER_STS_DC_OK
> 3. msleep (for the timeout case)

That's what I'm suggesting - do the initial busy wait for a reasonably
short time period and then fall back to something that sleeps.