io_uring: optimise wq_has_sleeper's smb_mb()

Inside __io_cq_unlock_post(), we unlock a spinlock and then are trying
to wake a CQ waitqueue, for that

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>

io_uring/io_uring.c

@@ -527,6 +527,12 @@ void __io_commit_cqring_flush(struct io_ring_ctx *ctx)
 		io_eventfd_signal(ctx);
 }
 
+#ifdef __x86_64__
+#define smp_mb__after_spin_unlock()	do {} while (0)
+#else
+#define smp_mb__after_spin_unlock()	smp_mb()
+#endif
+
 static inline void __io_cq_unlock_post(struct io_ring_ctx *ctx)
 	__releases(ctx->completion_lock)
 {
@@ -535,7 +541,8 @@ static inline void __io_cq_unlock_post(struct io_ring_ctx *ctx)
 
 	/* former io_cqring_ev_posted */
 	io_commit_cqring_flush(ctx);
-	if (wq_has_sleeper(&ctx->cq_wait))
+	smp_mb__after_spin_unlock();
+	if (waitqueue_active(&ctx->cq_wait))
 		io_cqring_wake(ctx);
 }