drm/i915: Emit dma-fence (and execlists submit) first from signaler

When introduced, I thought that reducing client latency from the
signaler was the priority. Since its inception the signaler has become
responsible for keeping the execlists full, via the dma-fence. As this
is very important to minimise overall execution time, signal the
dma-fence first and then signal any waiting clients.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170124110009.28947-8-chris@chris-wilson.co.uk

diff --git a/drivers/gpu/drm/i915/intel_breadcrumbs.c b/drivers/gpu/drm/i915/intel_breadcrumbs.c
index 6b24f2544b6b681d8df21de0410294376f7275c0..9fd002bcebb692e4028a596598fd313360d52ea2 100644 (file)
--- a/drivers/gpu/drm/i915/intel_breadcrumbs.c
+++ b/drivers/gpu/drm/i915/intel_breadcrumbs.c
@@ -459,16 +459,16 @@ static int intel_breadcrumbs_signaler(void *arg)
                 */
                request = READ_ONCE(b->first_signal);
                if (signal_complete(request)) {
+                       local_bh_disable();
+                       dma_fence_signal(&request->fence);
+                       local_bh_enable(); /* kick start the tasklets */
+
                        /* Wake up all other completed waiters and select the
                         * next bottom-half for the next user interrupt.
                         */
                        intel_engine_remove_wait(engine,
                                                 &request->signaling.wait);
 
-                       local_bh_disable();
-                       dma_fence_signal(&request->fence);
-                       local_bh_enable(); /* kick start the tasklets */
-
                        /* Find the next oldest signal. Note that as we have
                         * not been holding the lock, another client may
                         * have installed an even older signal than the one