drm/i915/selftests: Flush all active callbacks

Flushing the outer i915_active is not enough, as we need the barrier to
be applied across all the active dma_fence callbacks. So we must
serialise with each outstanding fence.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=112096
References: f79520bb33 ("drm/i915/selftests: Synchronize checking active status with retirement")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Andi Shyti <andi.shyti@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20191101181022.25633-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gt/selftest_engine_heartbeat.c

@@ -53,9 +53,7 @@ static struct pulse *pulse_create(void)
 
 static void pulse_unlock_wait(struct pulse *p)
 {
-	mutex_lock(&p->active.mutex);
-	mutex_unlock(&p->active.mutex);
-	flush_work(&p->active.work);
+	i915_active_unlock_wait(&p->active);
 }
 
 static int __live_idle_pulse(struct intel_engine_cs *engine,

drivers/gpu/drm/i915/i915_active.h

@@ -215,5 +215,6 @@ void i915_active_acquire_barrier(struct i915_active *ref);
 void i915_request_add_active_barriers(struct i915_request *rq);
 
 void i915_active_print(struct i915_active *ref, struct drm_printer *m);
+void i915_active_unlock_wait(struct i915_active *ref);
 
 #endif /* _I915_ACTIVE_H_ */

drivers/gpu/drm/i915/selftests/i915_active.c

@@ -250,3 +250,36 @@ void i915_active_print(struct i915_active *ref, struct drm_printer *m)
 		i915_active_release(ref);
 	}
 }
+
+static void spin_unlock_wait(spinlock_t *lock)
+{
+	spin_lock_irq(lock);
+	spin_unlock_irq(lock);
+}
+
+void i915_active_unlock_wait(struct i915_active *ref)
+{
+	if (i915_active_acquire_if_busy(ref)) {
+		struct active_node *it, *n;
+
+		rcu_read_lock();
+		rbtree_postorder_for_each_entry_safe(it, n, &ref->tree, node) {
+			struct dma_fence *f;
+
+			/* Wait for all active callbacks */
+			f = rcu_dereference(it->base.fence);
+			if (f)
+				spin_unlock_wait(f->lock);
+		}
+		rcu_read_unlock();
+
+		i915_active_release(ref);
+	}
+
+	/* And wait for the retire callback */
+	mutex_lock(&ref->mutex);
+	mutex_unlock(&ref->mutex);
+
+	/* ... which may have been on a thread instead */
+	flush_work(&ref->work);
+}