[PATCH v3 5/6] io_uring: persistent req bulk allocation cache

[Pavel Begunkov <asml.silence@gmail.com>]

Save bulk allocated requests across io_uring_enter(), so lower QD also
could benefit from that. This is not much of an optimisation, and for
current cache sizes would probably affect only offloaded ~QD=1.

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
---
 fs/io_uring.c | 28 ++++++++++++++++++++++------
 1 file changed, 22 insertions(+), 6 deletions(-)

diff --git a/fs/io_uring.c b/fs/io_uring.c
index cbe639caa096..66742d5772fa 100644
--- a/fs/io_uring.c
+++ b/fs/io_uring.c
@@ -845,6 +845,23 @@ static struct io_ring_ctx *io_ring_ctx_alloc(struct io_uring_params *p)
 	return NULL;
 }
 
+static void io_init_submit_state(struct io_ring_ctx *ctx)
+{
+	struct io_submit_state *state = &ctx->submit_state;
+
+	state->free_reqs = 0;
+	state->cur_req = 0;
+}
+
+static void io_clear_submit_state(struct io_ring_ctx *ctx)
+{
+	struct io_submit_state *state = &ctx->submit_state;
+
+	if (state->free_reqs)
+		kmem_cache_free_bulk(req_cachep, state->free_reqs,
+					&state->reqs[state->cur_req]);
+}
+
 static inline bool __req_need_defer(struct io_kiocb *req)
 {
 	struct io_ring_ctx *ctx = req->ctx;
@@ -1171,10 +1188,9 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx)
 	struct io_submit_state *state = &ctx->submit_state;
 
 	if (!state->free_reqs) {
-		size_t sz;
+		size_t sz = ARRAY_SIZE(state->reqs);
 		int ret;
 
-		sz = min_t(size_t, state->ios_left, ARRAY_SIZE(state->reqs));
 		ret = kmem_cache_alloc_bulk(req_cachep, gfp, sz, state->reqs);
 
 		/*
@@ -4835,9 +4851,6 @@ static void io_submit_end(struct io_ring_ctx *ctx)
 	struct io_submit_state *state = &ctx->submit_state;
 
 	io_file_put(state);
-	if (state->free_reqs)
-		kmem_cache_free_bulk(req_cachep, state->free_reqs,
-					&state->reqs[state->cur_req]);
 	if (state->link)
 		io_queue_link_head(state->link);
 }
@@ -4850,7 +4863,6 @@ static void io_submit_start(struct io_ring_ctx *ctx, unsigned int max_ios,
 {
 	struct io_submit_state *state = &ctx->submit_state;
 
-	state->free_reqs = 0;
 	state->file = NULL;
 	state->ios_left = max_ios;
 
@@ -6247,6 +6259,8 @@ static void io_ring_ctx_free(struct io_ring_ctx *ctx)
 	if (ctx->sqo_mm)
 		mmdrop(ctx->sqo_mm);
 
+	io_clear_submit_state(ctx);
+
 	io_iopoll_reap_events(ctx);
 	io_sqe_buffer_unregister(ctx);
 	io_sqe_files_unregister(ctx);
@@ -6767,6 +6781,8 @@ static int io_uring_create(unsigned entries, struct io_uring_params *p)
 	if (ret)
 		goto err;
 
+	io_init_submit_state(ctx);
+
 	ret = io_sq_offload_start(ctx, p);
 	if (ret)
 		goto err;
-- 
2.24.0