req->sequence is used only for deferred (i.e. DRAIN) requests, but is a burden of every request even if isn't used. Remove req->sequence from io_kiocb together with its initialisation in io_init_req(). In place of that keep the field in io_defer_entry, and calculate it in io_req_defer()'s slow path. Signed-off-by: Pavel Begunkov <asml.silence@xxxxxxxxx> --- fs/io_uring.c | 40 ++++++++++++++++++++++++++-------------- 1 file changed, 26 insertions(+), 14 deletions(-) diff --git a/fs/io_uring.c b/fs/io_uring.c index 93e8192983e1..db7f86b6da09 100644 --- a/fs/io_uring.c +++ b/fs/io_uring.c @@ -639,6 +639,7 @@ struct io_kiocb { u8 iopoll_completed; u16 buf_index; + u32 result; struct io_ring_ctx *ctx; unsigned int flags; @@ -646,8 +647,6 @@ struct io_kiocb { struct task_struct *task; unsigned long fsize; u64 user_data; - u32 result; - u32 sequence; struct list_head link_list; @@ -681,6 +680,7 @@ struct io_kiocb { struct io_defer_entry { struct list_head list; struct io_kiocb *req; + u32 seq; }; #define IO_IOPOLL_BATCH 8 @@ -1088,13 +1088,13 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p) return NULL; } -static inline bool req_need_defer(struct io_kiocb *req) +static bool req_need_defer(struct io_kiocb *req, u32 seq) { if (unlikely(req->flags & REQ_F_IO_DRAIN)) { struct io_ring_ctx *ctx = req->ctx; - return req->sequence != ctx->cached_cq_tail - + atomic_read(&ctx->cached_cq_overflow); + return seq != ctx->cached_cq_tail + + atomic_read(&ctx->cached_cq_overflow); } return false; @@ -1237,7 +1237,7 @@ static void __io_queue_deferred(struct io_ring_ctx *ctx) struct io_defer_entry *de = list_first_entry(&ctx->defer_list, struct io_defer_entry, list); - if (req_need_defer(de->req)) + if (req_need_defer(de->req, de->seq)) break; list_del_init(&de->list); /* punt-init is done before queueing for defer */ @@ -5400,14 +5400,30 @@ static int io_req_defer_prep(struct io_kiocb *req, return ret; } +static int io_get_sequence(struct io_kiocb *req) +{ + struct io_kiocb *pos; + struct io_ring_ctx *ctx = req->ctx; + u32 nr_reqs = 1; + + if (req->flags & REQ_F_LINK_HEAD) + list_for_each_entry(pos, &req->link_list, link_list) + nr_reqs++; + + return ctx->cached_sq_head - ctx->cached_sq_dropped - nr_reqs; +} + + static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) { struct io_ring_ctx *ctx = req->ctx; struct io_defer_entry *de; int ret; + u32 seq; /* Still need defer if there is pending req in defer list. */ - if (!req_need_defer(req) && list_empty_careful(&ctx->defer_list)) + if (likely(list_empty_careful(&ctx->defer_list) && + !(req->flags & REQ_F_IO_DRAIN))) return 0; if (!req->io) { @@ -5422,8 +5438,9 @@ static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) if (!de) return -ENOMEM; + seq = io_get_sequence(req); spin_lock_irq(&ctx->completion_lock); - if (!req_need_defer(req) && list_empty(&ctx->defer_list)) { + if (!req_need_defer(req, seq) && list_empty(&ctx->defer_list)) { spin_unlock_irq(&ctx->completion_lock); kfree(de); return 0; @@ -5431,6 +5448,7 @@ static int io_req_defer(struct io_kiocb *req, const struct io_uring_sqe *sqe) trace_io_uring_defer(ctx, req, req->user_data); de->req = req; + de->seq = seq; list_add_tail(&de->list, &ctx->defer_list); spin_unlock_irq(&ctx->completion_lock); return -EIOCBQUEUED; @@ -6207,12 +6225,6 @@ static int io_init_req(struct io_ring_ctx *ctx, struct io_kiocb *req, unsigned int sqe_flags; int id; - /* - * All io need record the previous position, if LINK vs DARIN, - * it can be used to mark the position of the first IO in the - * link list. - */ - req->sequence = ctx->cached_sq_head - ctx->cached_sq_dropped; req->opcode = READ_ONCE(sqe->opcode); req->user_data = READ_ONCE(sqe->user_data); req->io = NULL; -- 2.24.0