Re: [PATCH 1/2] io_uring: split alloc and add of overflow

[Caleb Sander Mateos <csander@purestorage.com>]

On Fri, May 16, 2025 at 9:15 AM Jens Axboe <axboe@kernel.dk> wrote:
>
> Add a new helper, io_alloc_ocqe(), that simply allocates and fills an
> overflow entry. Then it can get done outside of the locking section,
> and hence use more appropriate gfp_t allocation flags rather than always
> default to GFP_ATOMIC.
>
> Suggested-by: Pavel Begunkov <asml.silence@gmail.com>
> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> ---
>  io_uring/io_uring.c | 75 +++++++++++++++++++++++++++++----------------
>  1 file changed, 48 insertions(+), 27 deletions(-)
>
> diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
> index 9a9b8d35349b..2519fab303c4 100644
> --- a/io_uring/io_uring.c
> +++ b/io_uring/io_uring.c
> @@ -718,20 +718,11 @@ static __cold void io_uring_drop_tctx_refs(struct task_struct *task)
>         }
>  }
>
> -static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
> -                                    s32 res, u32 cflags, u64 extra1, u64 extra2)
> +static bool io_cqring_add_overflow(struct io_ring_ctx *ctx,
> +                                  struct io_overflow_cqe *ocqe)
>  {
> -       struct io_overflow_cqe *ocqe;
> -       size_t ocq_size = sizeof(struct io_overflow_cqe);
> -       bool is_cqe32 = (ctx->flags & IORING_SETUP_CQE32);
> -
>         lockdep_assert_held(&ctx->completion_lock);
>
> -       if (is_cqe32)
> -               ocq_size += sizeof(struct io_uring_cqe);
> -
> -       ocqe = kmalloc(ocq_size, GFP_ATOMIC | __GFP_ACCOUNT);
> -       trace_io_uring_cqe_overflow(ctx, user_data, res, cflags, ocqe);
>         if (!ocqe) {
>                 struct io_rings *r = ctx->rings;
>
> @@ -749,22 +740,44 @@ static bool io_cqring_event_overflow(struct io_ring_ctx *ctx, u64 user_data,
>                 atomic_or(IORING_SQ_CQ_OVERFLOW, &ctx->rings->sq_flags);
>
>         }
> -       ocqe->cqe.user_data = user_data;
> -       ocqe->cqe.res = res;
> -       ocqe->cqe.flags = cflags;
> -       if (is_cqe32) {
> -               ocqe->cqe.big_cqe[0] = extra1;
> -               ocqe->cqe.big_cqe[1] = extra2;
> -       }
>         list_add_tail(&ocqe->list, &ctx->cq_overflow_list);
>         return true;
>  }
>
> -static void io_req_cqe_overflow(struct io_kiocb *req)
> +static struct io_overflow_cqe *io_alloc_ocqe(struct io_ring_ctx *ctx,
> +                                            u64 user_data, s32 res, u32 cflags,
> +                                            u64 extra1, u64 extra2, gfp_t gfp)
> +{
> +       struct io_overflow_cqe *ocqe;
> +       size_t ocq_size = sizeof(struct io_overflow_cqe);
> +       bool is_cqe32 = (ctx->flags & IORING_SETUP_CQE32);
> +
> +       if (is_cqe32)
> +               ocq_size += sizeof(struct io_uring_cqe);
> +
> +       ocqe = kmalloc(ocq_size, gfp | __GFP_ACCOUNT);
> +       trace_io_uring_cqe_overflow(ctx, user_data, res, cflags, ocqe);
> +       if (ocqe) {
> +               ocqe->cqe.user_data = user_data;
> +               ocqe->cqe.res = res;
> +               ocqe->cqe.flags = cflags;
> +               if (is_cqe32) {
> +                       ocqe->cqe.big_cqe[0] = extra1;
> +                       ocqe->cqe.big_cqe[1] = extra2;
> +               }
> +       }
> +       return ocqe;
> +}
> +
> +static void io_req_cqe_overflow(struct io_kiocb *req, gfp_t gfp)
>  {
> -       io_cqring_event_overflow(req->ctx, req->cqe.user_data,
> -                               req->cqe.res, req->cqe.flags,
> -                               req->big_cqe.extra1, req->big_cqe.extra2);
> +       struct io_ring_ctx *ctx = req->ctx;
> +       struct io_overflow_cqe *ocqe;
> +
> +       ocqe = io_alloc_ocqe(ctx, req->cqe.user_data, req->cqe.res,
> +                            req->cqe.flags, req->big_cqe.extra1,
> +                            req->big_cqe.extra2, gfp);
> +       io_cqring_add_overflow(ctx, ocqe);
>         memset(&req->big_cqe, 0, sizeof(req->big_cqe));
>  }
>
> @@ -832,8 +845,12 @@ bool io_post_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags
>
>         io_cq_lock(ctx);
>         filled = io_fill_cqe_aux(ctx, user_data, res, cflags);
> -       if (!filled)
> -               filled = io_cqring_event_overflow(ctx, user_data, res, cflags, 0, 0);
> +       if (unlikely(!filled)) {
> +               struct io_overflow_cqe *ocqe;
> +
> +               ocqe = io_alloc_ocqe(ctx, user_data, res, cflags, 0, 0, GFP_ATOMIC);
> +               filled = io_cqring_add_overflow(ctx, ocqe);
> +       }
>         io_cq_unlock_post(ctx);
>         return filled;
>  }
> @@ -848,8 +865,11 @@ void io_add_aux_cqe(struct io_ring_ctx *ctx, u64 user_data, s32 res, u32 cflags)
>         lockdep_assert(ctx->lockless_cq);
>
>         if (!io_fill_cqe_aux(ctx, user_data, res, cflags)) {
> +               struct io_overflow_cqe *ocqe;
> +
> +               ocqe = io_alloc_ocqe(ctx, user_data, res, cflags, 0, 0, GFP_KERNEL);
>                 spin_lock(&ctx->completion_lock);
> -               io_cqring_event_overflow(ctx, user_data, res, cflags, 0, 0);
> +               io_cqring_add_overflow(ctx, ocqe);
>                 spin_unlock(&ctx->completion_lock);
>         }
>         ctx->submit_state.cq_flush = true;
> @@ -1442,10 +1462,11 @@ void __io_submit_flush_completions(struct io_ring_ctx *ctx)
>                     unlikely(!io_fill_cqe_req(ctx, req))) {
>                         if (ctx->lockless_cq) {
>                                 spin_lock(&ctx->completion_lock);
> -                               io_req_cqe_overflow(req);
> +                               io_req_cqe_overflow(req, GFP_ATOMIC);
>                                 spin_unlock(&ctx->completion_lock);
>                         } else {
> -                               io_req_cqe_overflow(req);
> +                               gfp_t gfp = ctx->lockless_cq ? GFP_KERNEL : GFP_ATOMIC;

This is in the else case of an if (ctx->lockless_cq). Isn't
ctx->lockless_cq known to be false?

Best,
Caleb

> +                               io_req_cqe_overflow(req, gfp);
>                         }
>                 }
>         }
> --
> 2.49.0
>
>