[Intel-gfx] [PATCH 5/6] drm/i915/dp: push eDP caching out into a work queue

[Jesse Barnes]

On Thu, 6 Mar 2014 11:28:13 +0200
Ville Syrjälä <ville.syrjala at linux.intel.com> wrote:

> On Wed, Mar 05, 2014 at 02:48:30PM -0800, Jesse Barnes wrote:
> > It takes awhile to fetch the DPCD and EDID for caching, so take it out
> > of the critical path to improve init time.
> > 
> > Signed-off-by: Jesse Barnes <jbarnes at virtuousgeek.org>
> > ---
> >  drivers/gpu/drm/i915/intel_dp.c | 113 +++++++++++++++++++++++++++++-----------
> >  1 file changed, 82 insertions(+), 31 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/i915/intel_dp.c b/drivers/gpu/drm/i915/intel_dp.c
> > index 738c4e6..763f235 100644
> > --- a/drivers/gpu/drm/i915/intel_dp.c
> > +++ b/drivers/gpu/drm/i915/intel_dp.c
> > @@ -3001,6 +3001,20 @@ intel_dp_handle_test_request(struct intel_dp *intel_dp)
> >  	intel_dp_aux_native_write_1(intel_dp, DP_TEST_RESPONSE, DP_TEST_NAK);
> >  }
> >  
> > +static void intel_flush_edp_cache_work(struct intel_dp *intel_dp)
> > +{
> > +	struct drm_device *dev = intel_dp->attached_connector->base.dev;
> > +
> > +	WARN_ON(!mutex_is_locked(&dev->mode_config.mutex));
> > +
> > +	if (!is_edp(intel_dp))
> > +		return;
> > +
> > +	mutex_unlock(&dev->mode_config.mutex);
> > +	flush_work(&intel_dp->edp_cache_work);
> > +	mutex_lock(&dev->mode_config.mutex);
> 
> This feels like deadlock land to me. If we drop mode_config.mutex
> someone else might grab it and then get blocked on the crtc mutex
> we're already holding, and then we try to re-grab mode_config.mutex...

Yeah I could use unlock_all here instead, or be more careful about
dropping the specific crtc mutex we need.  I did that on the crtc side
of things but obviously missed it here for cases where we'll hold the
crtc lock in this path.

-- 
Jesse Barnes, Intel Open Source Technology Center