Locking in the clk API

[u.kleine-koenig@pengutronix.de (Uwe Kleine-König)]

Hello Russell,

On Tue, Jan 11, 2011 at 10:47:09AM +0000, Russell King - ARM Linux wrote:
> On Tue, Jan 11, 2011 at 11:39:29AM +0100, Uwe Kleine-K?nig wrote:
> > A quick look into Digi's BSP (digiEL-5.0) shows they implemented
> > something I suggested earlier here:
> >
> > [...]
> > 
> > 
> > I think the idea is nice.  At least it allows with a single lock to
> > implement both, sleeping and atomic clks without the need to mark the
> > atomicity in a global flag.
> 
> It doesn't.  clk_enable() here can still end up trying to sleep when
> it's called from IRQ context - the code doesn't solve that.  All it
> means is that the intermediate code doesn't care whether clk->endisable
> ends up sleeping or not.
Obviousley you're right and your last sentence is all I intended to
claim.

> What it does do is return -EBUSY if there are two concurrent attempts
> to enable the same clock.  How many drivers today deal sanely with
> such an error from clk_enable(), and how many would just fail their
> probe() call on such an occurance?
Yes, that's the ugly part.

Best regards
Uwe

-- 
Pengutronix e.K.                           | Uwe Kleine-K?nig            |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |