[PATCH 12/13] dma-buf: Refactor reservation_object_add_shared_fence()

Chris Wilson chris at chris-wilson.co.uk
Sun Feb 4 12:47:08 UTC 2018


Adding a shared fence to a reservation_object is currently split into
two handlers, one to insert the fence into the existing array and the
other to replace the existing array with a new larger array. The first
step in both of these routines involves scanning the existing array to
decide if it can prune any of the existing fences.

As both routines perform essentially the same loop, we can combine them
into one routine. During the first scan over the existing array, we
search for a slot with which we can reuse for the new fence (discarding
the previous). If we find no available slot to reuse, and the array is
already at its max size, then we know that we need to switch to the
larger array, and that no existing fence needs to be discarded, they can
all be copied over.

add/remove: 0/0 grow/shrink: 0/3 up/down: 0/-627 (-627)
Function                                     old     new   delta
__warned                                    2352    2350      -2
reservation_object_reserve_shared            196     185     -11
reservation_object_add_shared_fence         1272     658    -614

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Christian K├Ânig <christian.koenig at amd.com>
Cc: Alex Deucher <alexander.deucher at amd.com>
---
 drivers/dma-buf/reservation.c | 178 ++++++++++++++----------------------------
 1 file changed, 60 insertions(+), 118 deletions(-)

diff --git a/drivers/dma-buf/reservation.c b/drivers/dma-buf/reservation.c
index 314eb1071cce..de7b6e709a68 100644
--- a/drivers/dma-buf/reservation.c
+++ b/drivers/dma-buf/reservation.c
@@ -76,8 +76,6 @@ int reservation_object_reserve_shared(struct reservation_object *obj)
 	if (old && old->shared_max) {
 		if (old->shared_count < old->shared_max) {
 			/* perform an in-place update */
-			kfree(obj->staged);
-			obj->staged = NULL;
 			return 0;
 		} else
 			max = old->shared_max * 2;
@@ -95,146 +93,90 @@ int reservation_object_reserve_shared(struct reservation_object *obj)
 
 	obj->staged = fobj;
 	fobj->shared_max = max;
+	fobj->shared_count = 0;
 	return 0;
 }
 EXPORT_SYMBOL(reservation_object_reserve_shared);
 
-static void
-reservation_object_add_shared_inplace(struct reservation_object *obj,
-				      struct reservation_object_list *fobj,
-				      struct dma_fence *fence)
+/**
+ * reservation_object_add_shared_fence - Add a fence to a shared slot
+ * @obj: the reservation object
+ * @fence: the shared fence to add
+ *
+ * Add a fence to a shared slot, obj->lock must be held, and
+ * reservation_object_reserve_shared() has been called.
+ */
+void reservation_object_add_shared_fence(struct reservation_object *obj,
+					 struct dma_fence *fence)
 {
-	struct dma_fence *signaled = NULL;
-	u32 i, signaled_idx;
+	struct reservation_object_list *old, *fobj;
+	struct dma_fence *replace = NULL;
+	u32 max = 0, i = 0;
 
 	dma_fence_get(fence);
 
-	preempt_disable();
-	write_seqcount_begin(&obj->seq);
-
-	for (i = 0; i < fobj->shared_count; ++i) {
-		struct dma_fence *old_fence;
-
-		old_fence = rcu_dereference_protected(fobj->shared[i],
-						reservation_object_held(obj));
-
-		if (old_fence->context == fence->context) {
-			/* memory barrier is added by write_seqcount_begin */
-			RCU_INIT_POINTER(fobj->shared[i], fence);
-			write_seqcount_end(&obj->seq);
-			preempt_enable();
-
-			dma_fence_put(old_fence);
-			return;
-		}
-
-		if (!signaled && dma_fence_is_signaled(old_fence)) {
-			signaled = old_fence;
-			signaled_idx = i;
-		}
-	}
-
 	/*
-	 * memory barrier is added by write_seqcount_begin,
-	 * fobj->shared_count is protected by this lock too
+	 * Find an existing slot we can overwrite with the new fence.
+	 *
+	 * It is assumed that fences are added sequentially, that a second
+	 * fence from the same context as an existing fence is strictly later
+	 * than the existing fence. Therefore any new fence on an existing
+	 * context can simply replace that existing fence.
+	 *
+	 * Likewise, if a fence is already signaled it is no longer required
+	 * as part of the reservation_object and can be replaced by the new
+	 * fence. This allows us to gradually prune the shared fence array
+	 * and prevent its excessive growth.
 	 */
-	if (signaled) {
-		RCU_INIT_POINTER(fobj->shared[signaled_idx], fence);
-	} else {
-		RCU_INIT_POINTER(fobj->shared[fobj->shared_count], fence);
-		fobj->shared_count++;
-	}
-
-	write_seqcount_end(&obj->seq);
-	preempt_enable();
+	old = reservation_object_get_list(obj);
+	if (old) {
+		u32 ctx = fence->context;
 
-	dma_fence_put(signaled);
-}
+		for (; i < old->shared_count; ++i) {
+			struct dma_fence *check;
 
-static void
-reservation_object_add_shared_replace(struct reservation_object *obj,
-				      struct reservation_object_list *old,
-				      struct reservation_object_list *fobj,
-				      struct dma_fence *fence)
-{
-	unsigned i, j, k;
+			check = rcu_dereference_protected(old->shared[i],
+							  reservation_object_held(obj));
 
-	dma_fence_get(fence);
+			if (check->context == ctx ||
+			    dma_fence_is_signaled(check)) {
+				replace = check;
+				break;
+			}
+		}
 
-	if (!old) {
-		RCU_INIT_POINTER(fobj->shared[0], fence);
-		fobj->shared_count = 1;
-		goto done;
+		fobj = old;
+		max = old->shared_max;
 	}
 
-	/*
-	 * no need to bump fence refcounts, rcu_read access
-	 * requires the use of kref_get_unless_zero, and the
-	 * references from the old struct are carried over to
-	 * the new.
-	 */
-	for (i = 0, j = 0, k = fobj->shared_max; i < old->shared_count; ++i) {
-		struct dma_fence *check;
-
-		check = rcu_dereference_protected(old->shared[i],
-						reservation_object_held(obj));
+	if (i >= max) {
+		fobj = obj->staged;
+		obj->staged = NULL;
 
-		if (check->context == fence->context ||
-		    dma_fence_is_signaled(check))
-			RCU_INIT_POINTER(fobj->shared[--k], check);
-		else
-			RCU_INIT_POINTER(fobj->shared[j++], check);
+		if (old) {
+			/* Borrow the old fence references */
+			memcpy(fobj->shared, old->shared,
+			       old->shared_count * sizeof(*old->shared));
+			fobj->shared_count = old->shared_count;
+		}
 	}
-	fobj->shared_count = j;
-	RCU_INIT_POINTER(fobj->shared[fobj->shared_count], fence);
-	fobj->shared_count++;
 
-done:
 	preempt_disable();
 	write_seqcount_begin(&obj->seq);
-	/*
-	 * RCU_INIT_POINTER can be used here,
-	 * seqcount provides the necessary barriers
-	 */
-	RCU_INIT_POINTER(obj->fence, fobj);
-	write_seqcount_end(&obj->seq);
-	preempt_enable();
-
-	if (!old)
-		return;
-
-	/* Drop the references to the signaled fences */
-	for (i = k; i < fobj->shared_max; ++i) {
-		struct dma_fence *f;
 
-		f = rcu_dereference_protected(fobj->shared[i],
-					      reservation_object_held(obj));
-		dma_fence_put(f);
-	}
-	kfree_rcu(old, rcu);
-}
-
-/**
- * reservation_object_add_shared_fence - Add a fence to a shared slot
- * @obj: the reservation object
- * @fence: the shared fence to add
- *
- * Add a fence to a shared slot, obj->lock must be held, and
- * reservation_object_reserve_shared() has been called.
- */
-void reservation_object_add_shared_fence(struct reservation_object *obj,
-					 struct dma_fence *fence)
-{
-	struct reservation_object_list *old, *fobj = obj->staged;
+	RCU_INIT_POINTER(fobj->shared[i], fence);
+	if (!replace)
+		fobj->shared_count++;
+	if (fobj != old)
+		RCU_INIT_POINTER(obj->fence, fobj);
 
-	old = reservation_object_get_list(obj);
-	obj->staged = NULL;
+	/* write_seqcount_end() provides the necessary mb for the RCU writes */
+	write_seqcount_end(&obj->seq);
+	preempt_enable();
 
-	if (!fobj) {
-		BUG_ON(old->shared_count >= old->shared_max);
-		reservation_object_add_shared_inplace(obj, old, fence);
-	} else
-		reservation_object_add_shared_replace(obj, old, fobj, fence);
+	dma_fence_put(replace);
+	if (old && fobj != old)
+		kfree_rcu(old, rcu);
 }
 EXPORT_SYMBOL(reservation_object_add_shared_fence);
 
-- 
2.15.1



More information about the Intel-gfx-trybot mailing list