Re: [PATCH 07/11] io_uring: use atomic_t for refcounts

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

 



[context preserved for additional CCs]

On Tue, Dec 10, 2019 at 4:57 PM Jens Axboe <axboe@xxxxxxxxx> wrote:
> Recently had a regression that turned out to be because
> CONFIG_REFCOUNT_FULL was set.

I assume "regression" here refers to a performance regression? Do you
have more concrete numbers on this? Is one of the refcounting calls
particularly problematic compared to the others?

I really don't like it when raw atomic_t is used for refcounting
purposes - not only because that gets rid of the overflow checks, but
also because it is less clear semantically.

> Our ref count usage is really simple,

In my opinion, for a refcount to qualify as "really simple", it must
be possible to annotate each relevant struct member and local variable
with the (fixed) bias it carries when alive and non-NULL. This
refcount is more complicated than that.

> so let's just use atomic_t and get rid of the dependency on the full
> reference count checking being enabled or disabled.
>
> Signed-off-by: Jens Axboe <axboe@xxxxxxxxx>
> ---
>  fs/io_uring.c | 22 +++++++++++-----------
>  1 file changed, 11 insertions(+), 11 deletions(-)
>
> diff --git a/fs/io_uring.c b/fs/io_uring.c
> index 9a596b819334..05419a152b32 100644
> --- a/fs/io_uring.c
> +++ b/fs/io_uring.c
> @@ -360,7 +360,7 @@ struct io_kiocb {
>         };
>         struct list_head        link_list;
>         unsigned int            flags;
> -       refcount_t              refs;
> +       atomic_t                refs;
>  #define REQ_F_NOWAIT           1       /* must not punt to workers */
>  #define REQ_F_IOPOLL_COMPLETED 2       /* polled IO has completed */
>  #define REQ_F_FIXED_FILE       4       /* ctx owns file */
> @@ -770,7 +770,7 @@ static void io_cqring_fill_event(struct io_kiocb *req, long res)
>                 WRITE_ONCE(ctx->rings->cq_overflow,
>                                 atomic_inc_return(&ctx->cached_cq_overflow));
>         } else {
> -               refcount_inc(&req->refs);
> +               atomic_inc(&req->refs);
>                 req->result = res;
>                 list_add_tail(&req->list, &ctx->cq_overflow_list);
>         }
> @@ -852,7 +852,7 @@ static struct io_kiocb *io_get_req(struct io_ring_ctx *ctx,
>         req->ctx = ctx;
>         req->flags = 0;
>         /* one is dropped after submission, the other at completion */
> -       refcount_set(&req->refs, 2);
> +       atomic_set(&req->refs, 2);
>         req->result = 0;
>         INIT_IO_WORK(&req->work, io_wq_submit_work);
>         return req;
> @@ -1035,13 +1035,13 @@ static void io_put_req_find_next(struct io_kiocb *req, struct io_kiocb **nxtptr)
>  {
>         io_req_find_next(req, nxtptr);
>
> -       if (refcount_dec_and_test(&req->refs))
> +       if (atomic_dec_and_test(&req->refs))
>                 __io_free_req(req);
>  }
>
>  static void io_put_req(struct io_kiocb *req)
>  {
> -       if (refcount_dec_and_test(&req->refs))
> +       if (atomic_dec_and_test(&req->refs))
>                 io_free_req(req);
>  }
>
> @@ -1052,14 +1052,14 @@ static void io_put_req(struct io_kiocb *req)
>  static void __io_double_put_req(struct io_kiocb *req)
>  {
>         /* drop both submit and complete references */
> -       if (refcount_sub_and_test(2, &req->refs))
> +       if (atomic_sub_and_test(2, &req->refs))
>                 __io_free_req(req);
>  }
>
>  static void io_double_put_req(struct io_kiocb *req)
>  {
>         /* drop both submit and complete references */
> -       if (refcount_sub_and_test(2, &req->refs))
> +       if (atomic_sub_and_test(2, &req->refs))
>                 io_free_req(req);
>  }
>
> @@ -1108,7 +1108,7 @@ static void io_iopoll_complete(struct io_ring_ctx *ctx, unsigned int *nr_events,
>                 io_cqring_fill_event(req, req->result);
>                 (*nr_events)++;
>
> -               if (refcount_dec_and_test(&req->refs)) {
> +               if (atomic_dec_and_test(&req->refs)) {
>                         /* If we're not using fixed files, we have to pair the
>                          * completion part with the file put. Use regular
>                          * completions for those, only batch free for fixed
> @@ -3169,7 +3169,7 @@ static enum hrtimer_restart io_link_timeout_fn(struct hrtimer *timer)
>         if (!list_empty(&req->link_list)) {
>                 prev = list_entry(req->link_list.prev, struct io_kiocb,
>                                   link_list);
> -               if (refcount_inc_not_zero(&prev->refs)) {
> +               if (atomic_inc_not_zero(&prev->refs)) {
>                         list_del_init(&req->link_list);
>                         prev->flags &= ~REQ_F_LINK_TIMEOUT;
>                 } else
> @@ -4237,7 +4237,7 @@ static void io_get_work(struct io_wq_work *work)
>  {
>         struct io_kiocb *req = container_of(work, struct io_kiocb, work);
>
> -       refcount_inc(&req->refs);
> +       atomic_inc(&req->refs);
>  }
>
>  static int io_sq_offload_start(struct io_ring_ctx *ctx,
> @@ -4722,7 +4722,7 @@ static void io_uring_cancel_files(struct io_ring_ctx *ctx,
>                         if (req->work.files != files)
>                                 continue;
>                         /* req is being completed, ignore */
> -                       if (!refcount_inc_not_zero(&req->refs))
> +                       if (!atomic_inc_not_zero(&req->refs))
>                                 continue;
>                         cancel_req = req;
>                         break;
> --
> 2.24.0
>



[Index of Archives]     [Linux Samsung SoC]     [Linux Rockchip SoC]     [Linux Actions SoC]     [Linux for Synopsys ARC Processors]     [Linux NFS]     [Linux NILFS]     [Linux USB Devel]     [Video for Linux]     [Linux Audio Users]     [Yosemite News]     [Linux Kernel]     [Linux SCSI]


  Powered by Linux