[PATCH RFC] io_uring: batch rsrc refs put

[Pavel Begunkov <asml.silence@xxxxxxxxx>]

Batch putting down rsrc_node refs. Since struct req_batch is used even
more extensively, can shed some extra cycles.

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

Pretty much a (working) RFC. I wouldn't find any difference with my
setup, but in case you will be striving to close the remaining 5%.

 fs/io_uring.c | 28 ++++++++++++++++++++++++++--
 1 file changed, 26 insertions(+), 2 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index 0fea1342aeef..b7b0d76453ca 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -1930,6 +1930,8 @@ static inline void io_req_complete_post(struct io_kiocb *req, long res,
 
 		io_dismantle_req(req);
 		io_put_task(req->task, 1);
+		if (req->fixed_rsrc_refs)
+			percpu_ref_put(req->fixed_rsrc_refs);
 		list_add(&req->compl.list, &cs->locked_free_list);
 		cs->locked_free_nr++;
 	} else
@@ -2044,8 +2046,6 @@ static void io_dismantle_req(struct io_kiocb *req)
 		kfree(req->async_data);
 	if (req->file)
 		io_put_file(req, req->file, (req->flags & REQ_F_FIXED_FILE));
-	if (req->fixed_rsrc_refs)
-		percpu_ref_put(req->fixed_rsrc_refs);
 	io_req_clean_work(req);
 }
 
@@ -2065,6 +2065,8 @@ static void __io_free_req(struct io_kiocb *req)
 
 	io_dismantle_req(req);
 	io_put_task(req->task, 1);
+	if (req->fixed_rsrc_refs)
+		percpu_ref_put(req->fixed_rsrc_refs);
 
 	kmem_cache_free(req_cachep, req);
 	percpu_ref_put(&ctx->refs);
@@ -2381,6 +2383,9 @@ struct req_batch {
 	struct task_struct	*task;
 	int			task_refs;
 	int			ctx_refs;
+
+	struct percpu_ref	*rsrc_refs;
+	unsigned int		rsrc_refs_nr;
 };
 
 static inline void io_init_req_batch(struct req_batch *rb)
@@ -2388,6 +2393,15 @@ static inline void io_init_req_batch(struct req_batch *rb)
 	rb->task_refs = 0;
 	rb->ctx_refs = 0;
 	rb->task = NULL;
+	rb->rsrc_refs = NULL;
+	rb->rsrc_refs_nr = 0;
+}
+
+static inline void __io_req_batch_flush_rsrc_refs(struct req_batch *rb)
+{
+	/* can get positive ->rsrc_refs_nr with NULL ->rsrc_refs */
+	if (rb->rsrc_refs)
+		percpu_ref_put_many(rb->rsrc_refs, rb->rsrc_refs_nr);
 }
 
 static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
@@ -2397,6 +2411,8 @@ static void io_req_free_batch_finish(struct io_ring_ctx *ctx,
 		io_put_task(rb->task, rb->task_refs);
 	if (rb->ctx_refs)
 		percpu_ref_put_many(&ctx->refs, rb->ctx_refs);
+
+	__io_req_batch_flush_rsrc_refs(rb);
 }
 
 static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
@@ -2413,6 +2429,14 @@ static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req,
 	rb->task_refs++;
 	rb->ctx_refs++;
 
+	if (req->fixed_rsrc_refs != rb->rsrc_refs) {
+		__io_req_batch_flush_rsrc_refs(rb);
+		rb->rsrc_refs = req->fixed_rsrc_refs;
+		rb->rsrc_refs_nr = 0;
+	}
+	/* it's ok to increment for NULL rsrc_refs, we'll handle it */
+	rb->rsrc_refs_nr++;
+
 	io_dismantle_req(req);
 	if (state->free_reqs != ARRAY_SIZE(state->reqs))
 		state->reqs[state->free_reqs++] = req;
-- 
2.24.0