drm/i915: Track number of pending freed objects

During igt, we frequently call into the driver to reset both HW and
driver state (idling the device, waiting for it to become idle and
freeing off old objects) to ensure that we start each test/subtest/pass
from known state. This process incurs an RCU barrier or two to ensure
that any such pending frees are indeed flushed before we return.
However, unconditionally waiting on the RCU barrier adds needless delay
to many callers, which adds up to several seconds when repeated thousands
of times. We can skip the rcu_barrier() if by tracking how many outstanding
frees we have, we know there are none.

The same path is used along suspend, where we may be able to save the
unconditional RCU barrier.

To put it into perspective with a completely meaningless
microbenchmark, igt/gem_sync/idle is improved from 50ms to 30us on bdw.

v2: Remove the extra synchronize_rcu() inside i915_drop_caches_set()
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20180219220631.25001-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/i915_debugfs.c

@@ -4079,10 +4079,8 @@ i915_drop_caches_set(void *data, u64 val)
 	if (val & DROP_IDLE)
 		drain_delayed_work(&dev_priv->gt.idle_work);
 
-	if (val & DROP_FREED) {
-		synchronize_rcu();
+	if (val & DROP_FREED)
 		i915_gem_drain_freed_objects(dev_priv);
-	}
 
 	return ret;
 }

drivers/gpu/drm/i915/i915_drv.h

@@ -1095,6 +1095,11 @@ struct i915_gem_mm {
 	struct llist_head free_list;
 	struct work_struct free_work;
 	spinlock_t free_lock;
+	/**
+	 * Count of objects pending destructions. Used to skip needlessly
+	 * waiting on an RCU barrier if no objects are waiting to be freed.
+	 */
+	atomic_t free_count;
 
 	/**
 	 * Small stash of WC pages
@@ -3134,6 +3139,9 @@ void i915_gem_free_object(struct drm_gem_object *obj);
 
 static inline void i915_gem_drain_freed_objects(struct drm_i915_private *i915)
 {
+	if (!atomic_read(&i915->mm.free_count))
+		return;
+
 	/* A single pass should suffice to release all the freed objects (along
 	 * most call paths) , but be a little more paranoid in that freeing
 	 * the objects does take a little amount of time, during which the rcu

drivers/gpu/drm/i915/i915_gem.c

@@ -4758,6 +4758,9 @@ static void __i915_gem_free_objects(struct drm_i915_private *i915,
 		kfree(obj->bit_17);
 		i915_gem_object_free(obj);
 
+		GEM_BUG_ON(!atomic_read(&i915->mm.free_count));
+		atomic_dec(&i915->mm.free_count);
+
 		if (on)
 			cond_resched();
 	}
@@ -4846,6 +4849,7 @@ void i915_gem_free_object(struct drm_gem_object *gem_obj)
 	 * i915_gem_busy_ioctl(). For the corresponding synchronized
 	 * lookup see i915_gem_object_lookup_rcu().
 	 */
+	atomic_inc(&to_i915(obj->base.dev)->mm.free_count);
 	call_rcu(&obj->rcu, __i915_gem_free_object_rcu);
 }
 
@@ -5546,7 +5550,8 @@ i915_gem_load_init(struct drm_i915_private *dev_priv)
 void i915_gem_load_cleanup(struct drm_i915_private *dev_priv)
 {
 	i915_gem_drain_freed_objects(dev_priv);
-	WARN_ON(!llist_empty(&dev_priv->mm.free_list));
+	GEM_BUG_ON(!llist_empty(&dev_priv->mm.free_list));
+	GEM_BUG_ON(atomic_read(&dev_priv->mm.free_count));
 	WARN_ON(dev_priv->mm.object_count);
 
 	mutex_lock(&dev_priv->drm.struct_mutex);