io_uring: optimise io_prep_linked_timeout()

Linked timeout handling during issuing is heavy, it adds extra
instructions and forces to save the next linked timeout before
io_issue_sqe().

Follwing the same reasoning as in refcounting patches, a request can't
be freed by the time it returns from io_issue_sqe(), so now we don't
need to do io_prep_linked_timeout() in advance, and it can be delayed to
colder paths optimising the generic path.

Also, it should also save quite a lot for requests with linked timeouts
and completed inline on timeout spinlocking + hrtimer_start() +
hrtimer_try_to_cancel() and so on.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Link: https://lore.kernel.org/r/19bfc9a0d26c5c5f1e359f7650afe807ca8ef879.1628981736.git.asml.silence@gmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

fs/io_uring.c

@@ -1306,8 +1306,16 @@ static void io_req_track_inflight(struct io_kiocb *req)
 	}
 }
 
+static inline void io_unprep_linked_timeout(struct io_kiocb *req)
+{
+	req->flags &= ~REQ_F_LINK_TIMEOUT;
+}
+
 static struct io_kiocb *__io_prep_linked_timeout(struct io_kiocb *req)
 {
+	if (WARN_ON_ONCE(!req->link))
+		return NULL;
+
 	req->flags &= ~REQ_F_ARM_LTIMEOUT;
 	req->flags |= REQ_F_LINK_TIMEOUT;
 
@@ -1932,6 +1940,7 @@ static bool io_disarm_next(struct io_kiocb *req)
 	if (req->flags & REQ_F_ARM_LTIMEOUT) {
 		struct io_kiocb *link = req->link;
 
+		req->flags &= ~REQ_F_ARM_LTIMEOUT;
 		if (link && link->opcode == IORING_OP_LINK_TIMEOUT) {
 			io_remove_next_linked(req);
 			io_cqring_fill_event(link->ctx, link->user_data,
@@ -6485,7 +6494,7 @@ static void io_queue_linked_timeout(struct io_kiocb *req)
 static void __io_queue_sqe(struct io_kiocb *req)
 	__must_hold(&req->ctx->uring_lock)
 {
-	struct io_kiocb *linked_timeout = io_prep_linked_timeout(req);
+	struct io_kiocb *linked_timeout;
 	int ret;
 
 issue_sqe:
@@ -6503,10 +6512,19 @@ static void __io_queue_sqe(struct io_kiocb *req)
 			state->compl_reqs[state->compl_nr++] = req;
 			if (state->compl_nr == ARRAY_SIZE(state->compl_reqs))
 				io_submit_flush_completions(ctx);
+			return;
 		}
+
+		linked_timeout = io_prep_linked_timeout(req);
+		if (linked_timeout)
+			io_queue_linked_timeout(linked_timeout);
 	} else if (ret == -EAGAIN && !(req->flags & REQ_F_NOWAIT)) {
+		linked_timeout = io_prep_linked_timeout(req);
+
 		switch (io_arm_poll_handler(req)) {
 		case IO_APOLL_READY:
+			if (linked_timeout)
+				io_unprep_linked_timeout(req);
 			goto issue_sqe;
 		case IO_APOLL_ABORTED:
 			/*
@@ -6516,11 +6534,12 @@ static void __io_queue_sqe(struct io_kiocb *req)
 			io_queue_async_work(req);
 			break;
 		}
+
+		if (linked_timeout)
+			io_queue_linked_timeout(linked_timeout);
 	} else {
 		io_req_complete_failed(req, ret);
 	}
-	if (linked_timeout)
-		io_queue_linked_timeout(linked_timeout);
 }
 
 static inline void io_queue_sqe(struct io_kiocb *req)