[Intel-gfx] [PATCH 04/37] drm/i915: rework locking for intel_dpio|sbi_read|write

Jesse Barnes jbarnes at virtuousgeek.org
Wed Dec 12 23:05:52 CET 2012


On Wed, 12 Dec 2012 23:00:34 +0100
Daniel Vetter <daniel at ffwll.ch> wrote:

> On Wed, Dec 12, 2012 at 12:54:47PM -0800, Jesse Barnes wrote:
> > On Wed, 12 Dec 2012 14:06:44 +0100
> > Daniel Vetter <daniel.vetter at ffwll.ch> wrote:
> > 
> > > Spinning for up to 200 us with interrupts locked out is not good. So
> > > let's just spin (and even that seems to be excessive).
> > > 
> > > And we don't call these functions from interrupt context, so this is
> > > not required. Besides that doing anything in interrupt contexts which
> > > might take a few hundred us is a no-go. So just convert the entire
> > > thing to a mutex. Also move the mutex-grabbing out of the read/write
> > > functions (add a WARN_ON(!is_locked)) instead) since all callers are
> > > nicely grouped together.
> > > 
> > > Finally the real motivation for this change: Dont grab the modeset
> > > mutex in the dpio debugfs file, we don't need that consistency. And
> > > correctness of the dpio interface is ensured with the dpio_lock.
> > > 
> > > Signed-off-by: Daniel Vetter <daniel.vetter at ffwll.ch>
> > 
> > Looks fine, I guess you could convert the wait_for_atomic_us to the
> > non-atomic variant now that you have a mutex.  Either way:
> 
> I like that _us purely to document the correct timeout ...

Yeah we'd need a non-atomic _us variant to clean it up.  No biggie.

-- 
Jesse Barnes, Intel Open Source Technology Center



More information about the Intel-gfx mailing list