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

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

On 5/16/25 17:08, Jens Axboe 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>

I didn't suggest that. If anything, it complicates CQE posting
helpers when we should be moving in the opposite direction.

> 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)

...
>   	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;

if (!ctx->lockless_cq)
	gfp_t gfp = ctx->lockless_cq ? GFP_KERNEL : GFP_ATOMIC;

-- 
Pavel Begunkov