util/queue: fix a race condition in the fence code

A tempting alternative fix would be adding a lock/unlock pair in
util_queue_fence_is_signalled. However, that wouldn't actually
improve anything in the semantics of util_queue_fence_is_signalled,
while making that test much more heavy-weight. So this lock/unlock
pair in util_queue_fence_destroy for "flushing out" other threads
that may still be in util_queue_fence_signal looks like the better
fix.

v2: rephrase the comment

Cc: [email protected]
Reviewed-by: Marek Olšák <[email protected]>
Reviewed-by: Gustaw Smolarczyk <[email protected]>

1 files changed, 13 insertions, 0 deletions

diff --git a/src/util/u_queue.c b/src/util/u_queue.c
index 449da7dc9ab..3b05110e9f8 100644
--- a/src/util/u_queue.c
+++ b/src/util/u_queue.c
@@ -120,6 +120,19 @@ void
 util_queue_fence_destroy(struct util_queue_fence *fence)
 {
    assert(fence->signalled);
+
+   /* Ensure that another thread is not in the middle of
+    * util_queue_fence_signal (having set the fence to signalled but still
+    * holding the fence mutex).
+    *
+    * A common contract between threads is that as soon as a fence is signalled
+    * by thread A, thread B is allowed to destroy it. Since
+    * util_queue_fence_is_signalled does not lock the fence mutex (for
+    * performance reasons), we must do so here.
+    */
+   mtx_lock(&fence->mutex);
+   mtx_unlock(&fence->mutex);
+
    cnd_destroy(&fence->cond);
    mtx_destroy(&fence->mutex);
 }