io_uring: optimize local tw add ctx pinning

We currently pin the ctx for io_req_local_work_add() with
percpu_ref_get/put, which implies two rcu_read_lock/unlock pairs and some
extra overhead on top in the fast path. Replace it with a pure rcu read
and let io_ring_exit_work() synchronise against it.
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/cbdfcb6b232627f30e9e50ef91f13c4f05910247.1680782017.git.asml.silence@gmail.comSigned-off-by: Jens Axboe <axboe@kernel.dk>

io_uring/io_uring.c

@@ -1332,9 +1332,9 @@ void __io_req_task_work_add(struct io_kiocb *req, bool allow_local)
 	struct io_ring_ctx *ctx = req->ctx;
 
 	if (allow_local && ctx->flags & IORING_SETUP_DEFER_TASKRUN) {
-		percpu_ref_get(&ctx->refs);
+		rcu_read_lock();
 		io_req_local_work_add(req);
-		percpu_ref_put(&ctx->refs);
+		rcu_read_unlock();
 		return;
 	}
 
@@ -3052,6 +3052,10 @@ static __cold void io_ring_exit_work(struct work_struct *work)
 	spin_lock(&ctx->completion_lock);
 	spin_unlock(&ctx->completion_lock);
 
+	/* pairs with RCU read section in io_req_local_work_add() */
+	if (ctx->flags & IORING_SETUP_DEFER_TASKRUN)
+		synchronize_rcu();
+
 	io_ring_ctx_free(ctx);
 }