We limit how much request we batch on the completion path, use a fixed size array instead of lists for completion batching. That also allows to split io_submit_flush_completions() into 2 steps: the first is filling CQEs, and the second actually frees requests.
There are plenty of benefits: - list head tossing is expensive + removes LIST_INIT in state prep - doesn't do extra unlock/lock to put a linked request - filling CQEs first gives better latency - will be used to handle list entry aliasing and add batch free there Signed-off-by: Pavel Begunkov <[email protected]> --- fs/io_uring.c | 33 +++++++++++++-------------------- 1 file changed, 13 insertions(+), 20 deletions(-) diff --git a/fs/io_uring.c b/fs/io_uring.c index 609c7da044d7..3277a06e2fb6 100644 --- a/fs/io_uring.c +++ b/fs/io_uring.c @@ -680,11 +680,12 @@ struct io_defer_entry { }; #define IO_IOPOLL_BATCH 8 +#define IO_COMPL_BATCH 32 struct io_comp_state { - unsigned int nr; - struct list_head list; struct io_ring_ctx *ctx; + unsigned int nr; + struct io_kiocb *reqs[IO_COMPL_BATCH]; }; struct io_submit_state { @@ -1794,28 +1795,21 @@ static void io_req_free_batch(struct req_batch *rb, struct io_kiocb *req) static void io_submit_flush_completions(struct io_comp_state *cs) { + struct io_kiocb *req; struct io_ring_ctx *ctx = cs->ctx; + int i, nr = cs->nr; spin_lock_irq(&ctx->completion_lock); - while (!list_empty(&cs->list)) { - struct io_kiocb *req; - - req = list_first_entry(&cs->list, struct io_kiocb, compl.list); - list_del(&req->compl.list); + for (i = 0; i < nr; ++i) { + req = cs->reqs[i]; __io_cqring_fill_event(req, req->result, req->compl.cflags); - if (!(req->flags & REQ_F_LINK_HEAD)) { - req->flags |= REQ_F_COMP_LOCKED; - io_put_req(req); - } else { - spin_unlock_irq(&ctx->completion_lock); - io_put_req(req); - spin_lock_irq(&ctx->completion_lock); - } } io_commit_cqring(ctx); spin_unlock_irq(&ctx->completion_lock); - io_cqring_ev_posted(ctx); + + for (i = 0; i < nr; ++i) + io_put_req(cs->reqs[i]); cs->nr = 0; } @@ -1829,8 +1823,8 @@ static void __io_req_complete(struct io_kiocb *req, long res, unsigned cflags, io_clean_op(req); req->result = res; req->compl.cflags = cflags; - list_add_tail(&req->compl.list, &cs->list); - if (++cs->nr >= 32) + cs->reqs[cs->nr++] = req; + if (cs->nr == IO_COMPL_BATCH) io_submit_flush_completions(cs); } } @@ -6118,7 +6112,7 @@ static int io_submit_sqe(struct io_kiocb *req, const struct io_uring_sqe *sqe, */ static void io_submit_state_end(struct io_submit_state *state) { - if (!list_empty(&state->comp.list)) + if (state->comp.nr) io_submit_flush_completions(&state->comp); blk_finish_plug(&state->plug); io_state_file_put(state); @@ -6137,7 +6131,6 @@ static void io_submit_state_start(struct io_submit_state *state, state->plug.nowait = true; #endif state->comp.nr = 0; - INIT_LIST_HEAD(&state->comp.list); state->comp.ctx = ctx; state->free_reqs = 0; state->file = NULL; -- 2.24.0
