drm/i915: Push i915_sw_fence_wait into the nonblocking atomic commit

Blocking in a worker is ok, that's what the unbound_wq is for. And it
unifies the paths between the blocking and nonblocking commit, giving
me just one path where I have to implement the deadlock avoidance
trickery in the next patch.

I first tried to implement the following patch without this rework, but
force-completing i915_sw_fence creates some serious challenges around
properly cleaning things up. So wasn't a feasible short-term approach.
Another approach would be to simple keep track of all pending atomic
commit work items and manually queue them from the reset code. With the
caveat that double-queue in case we race with the i915_sw_fence must be
avoided. Given all that, taking the cost of a double schedule in atomic
for the short-term fix is the best approach, but can be changed in the future of course.

v2: Amend commit message (Chris).

v3: Add comment explaining why we do nothing in the sw_fence complete
callback (Michel).

Reviewed-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com> (v2)
Cc: Michel Thierry <michel.thierry@intel.com>
Reviewed-by: Michel Thierry <michel.thierry@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20170808080828.23650-2-daniel.vetter@ffwll.ch

drivers/gpu/drm/i915/intel_display.c

@@ -12311,6 +12311,8 @@ static void intel_atomic_commit_tail(struct drm_atomic_state *state)
 	unsigned crtc_vblank_mask = 0;
 	int i;
 
+	i915_sw_fence_wait(&intel_state->commit_ready);
+
 	drm_atomic_helper_wait_for_dependencies(state);
 
 	if (intel_state->modeset)
@@ -12476,10 +12478,8 @@ intel_atomic_commit_ready(struct i915_sw_fence *fence,
 
 	switch (notify) {
 	case FENCE_COMPLETE:
-		if (state->base.commit_work.func)
-			queue_work(system_unbound_wq, &state->base.commit_work);
+		/* we do blocking waits in the worker, nothing to do here */
 		break;
-
 	case FENCE_FREE:
 		{
 			struct intel_atomic_helper *helper =
@@ -12581,14 +12581,14 @@ static int intel_atomic_commit(struct drm_device *dev,
 	}
 
 	drm_atomic_state_get(state);
-	INIT_WORK(&state->commit_work,
-		  nonblock ? intel_atomic_commit_work : NULL);
+	INIT_WORK(&state->commit_work, intel_atomic_commit_work);
 
 	i915_sw_fence_commit(&intel_state->commit_ready);
-	if (!nonblock) {
-		i915_sw_fence_wait(&intel_state->commit_ready);
+	if (nonblock)
+		queue_work(system_unbound_wq, &state->commit_work);
+	else
 		intel_atomic_commit_tail(state);
-	}
+
 
 	return 0;
 }