On Wed, Jun 15, 2016 at 04:04:46PM +0800, Boqun Feng wrote:

On Tue, Jun 14, 2016 at 06:48:04PM -0400, Waiman Long wrote:

quoted

@@ -198,7 +198,7 @@ void osq_unlock(struct optimistic_spin_queue *lock)
 	 * Second most likely case.
 	 */
 	node = this_cpu_ptr(&osq_node);
-	next = xchg(&node->next, NULL);
+	next = xchg_release(&node->next, NULL);
 	if (next) {
 		WRITE_ONCE(next->locked, 1);

So we still use WRITE_ONCE() rather than smp_store_release() here?

Though, IIUC, This is fine for all the archs but ARM64, because there
will always be a xchg_release()/xchg() before the WRITE_ONCE(), which
carries a necessary barrier to upgrade WRITE_ONCE() to a RELEASE.

Not sure. On PPC for example, you'll use lwsync() but will that not
attach to the store to &node->next instead?

Still leaving that store and the WRITE_ONCE() unordered.

Also I don't see the control dependency between xchg-load and WRITE_ONCE
helping anything to order the two stores.


So yeah, subtle if not broken, definitely needs more explanation.