On Wed, May 21, 2014 at 08:00:22AM -0700, Jesse Barnes wrote:
On Wed, 21 May 2014 08:52:34 +0200 Daniel Vetter daniel@ffwll.ch wrote:
On Tue, May 20, 2014 at 03:25:35PM -0700, Jesse Barnes wrote:
Gets the detect code (which may take awhile) out of the resume path, speeding things up a bit.
Signed-off-by: Jesse Barnes jbarnes@virtuousgeek.org
drivers/gpu/drm/i915/i915_drv.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)
diff --git a/drivers/gpu/drm/i915/i915_drv.c b/drivers/gpu/drm/i915/i915_drv.c index 302495f..571f688 100644 --- a/drivers/gpu/drm/i915/i915_drv.c +++ b/drivers/gpu/drm/i915/i915_drv.c @@ -606,7 +606,7 @@ static int __i915_drm_thaw(struct drm_device *dev, bool restore_gtt_mappings) intel_hpd_init(dev); dev_priv->enable_hotplug_processing = true; /* Config may have changed between suspend and resume */
drm_helper_hpd_irq_event(dev);
async_schedule(drm_helper_hpd_irq_event_async, dev);Does that really help all that much? I've thought the driver core sychnronizes all the async workers again once resume is done. I'm better to schedule this as a fully async work with e.g. a 1s delay, like we do with the rps resume work.
That might be better, I'll check on the synchronization. I thought async_schedule was the new hotness we were supposed to use everywhere...
It's pretty cool for easy async in driver load/resume since it autosyncs. You can even create more async domains if you need finer-grained sync points. But if we know that we can be more lenient than full sync before our driver load/resume completes we need to use classical work queues.
But for stuff like doing paralle modeset restore async domains look like the right thing. -Daniel