io_uring: make cached_cq_overflow non atomic_t
authorPavel Begunkov <asml.silence@gmail.com>
Sun, 18 Oct 2020 09:17:40 +0000 (10:17 +0100)
committerJens Axboe <axboe@kernel.dk>
Mon, 19 Oct 2020 19:29:29 +0000 (13:29 -0600)
ctx->cached_cq_overflow is changed only under completion_lock. Convert
it from atomic_t to just int, and mark all places when it's read without
lock with READ_ONCE, which guarantees atomicity (relaxed ordering).

Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
fs/io_uring.c

index dce73e4..620a11f 100644 (file)
@@ -277,7 +277,7 @@ struct io_ring_ctx {
                unsigned                sq_mask;
                unsigned                sq_thread_idle;
                unsigned                cached_sq_dropped;
-               atomic_t                cached_cq_overflow;
+               unsigned                cached_cq_overflow;
                unsigned long           sq_check_overflow;
 
                struct list_head        defer_list;
@@ -1179,7 +1179,7 @@ static bool req_need_defer(struct io_kiocb *req, u32 seq)
                struct io_ring_ctx *ctx = req->ctx;
 
                return seq != ctx->cached_cq_tail
-                               + atomic_read(&ctx->cached_cq_overflow);
+                               + READ_ONCE(ctx->cached_cq_overflow);
        }
 
        return false;
@@ -1624,8 +1624,9 @@ static bool io_cqring_overflow_flush(struct io_ring_ctx *ctx, bool force,
                        WRITE_ONCE(cqe->res, req->result);
                        WRITE_ONCE(cqe->flags, req->compl.cflags);
                } else {
+                       ctx->cached_cq_overflow++;
                        WRITE_ONCE(ctx->rings->cq_overflow,
-                               atomic_inc_return(&ctx->cached_cq_overflow));
+                                  ctx->cached_cq_overflow);
                }
        }
 
@@ -1667,8 +1668,8 @@ static void __io_cqring_fill_event(struct io_kiocb *req, long res, long cflags)
                 * then we cannot store the request for later flushing, we need
                 * to drop it on the floor.
                 */
-               WRITE_ONCE(ctx->rings->cq_overflow,
-                               atomic_inc_return(&ctx->cached_cq_overflow));
+               ctx->cached_cq_overflow++;
+               WRITE_ONCE(ctx->rings->cq_overflow, ctx->cached_cq_overflow);
        } else {
                if (list_empty(&ctx->cq_overflow_list)) {
                        set_bit(0, &ctx->sq_check_overflow);