From b8fb5b4fdd67f9d18109c5d21d44a8bd4ddb608b Mon Sep 17 00:00:00 2001 From: Pavel Begunkov Date: Tue, 4 Apr 2023 13:39:45 +0100 Subject: io_uring/rsrc: use non-pcpu refcounts for nodes One problem with the current rsrc infra is that often updates will generates lots of rsrc nodes, each carry pcpu refs. That takes quite a lot of memory, especially if there is a stall, and takes lots of CPU cycles. Only pcpu allocations takes >50 of CPU with a naive benchmark updating files in a loop. Replace pcpu refs with normal refcounting. There is already a hot path avoiding atomics / refs, but following patches will further improve it. Signed-off-by: Pavel Begunkov Link: https://lore.kernel.org/r/e9ed8a9457b331a26555ff9443afc64cdaab7247.1680576071.git.asml.silence@gmail.com Signed-off-by: Jens Axboe --- io_uring/rsrc.h | 6 ++++-- 1 file changed, 4 insertions(+), 2 deletions(-) (limited to 'io_uring/rsrc.h') diff --git a/io_uring/rsrc.h b/io_uring/rsrc.h index f27f4975217d..1467b31843bc 100644 --- a/io_uring/rsrc.h +++ b/io_uring/rsrc.h @@ -37,7 +37,7 @@ struct io_rsrc_data { }; struct io_rsrc_node { - struct percpu_ref refs; + refcount_t refs; struct list_head node; struct list_head rsrc_list; struct io_rsrc_data *rsrc_data; @@ -54,6 +54,7 @@ struct io_mapped_ubuf { }; void io_rsrc_put_tw(struct callback_head *cb); +void io_rsrc_node_ref_zero(struct io_rsrc_node *node); void io_rsrc_put_work(struct work_struct *work); void io_rsrc_refs_refill(struct io_ring_ctx *ctx); void io_wait_rsrc_data(struct io_rsrc_data *data); @@ -109,7 +110,8 @@ int io_register_rsrc(struct io_ring_ctx *ctx, void __user *arg, static inline void io_rsrc_put_node(struct io_rsrc_node *node, int nr) { - percpu_ref_put_many(&node->refs, nr); + if (refcount_sub_and_test(nr, &node->refs)) + io_rsrc_node_ref_zero(node); } static inline void io_req_put_rsrc(struct io_kiocb *req) -- cgit v1.2.3