Re: [PATCHv5 2/4] Add support IORING_SETUP_SQE_MIXED

[Jens Axboe <axboe@kernel.dk>]

(Adding right list)

On 10/21/25 3:33 PM, Keith Busch wrote:
> +/*
> + * Return a 128B sqe to fill. Applications must later call io_uring_submit()
> + * when it's ready to tell the kernel about it. The caller may call this
> + * function multiple times before calling io_uring_submit().
> + *
> + * Returns a vacant 128B sqe, or NULL if we're full. If the current tail is the
> + * last entry in the ring, this function will insert a nop + skip complete such
> + * that the 128b entry wraps back to the beginning of the queue for a
> + * contiguous big sq entry. It's up to the caller to use a 128b opcode in order
> + * for the kernel to know how to advance its sq head pointer.
> + */
> +IOURINGINLINE struct io_uring_sqe *io_uring_get_sqe128_mixed(struct io_uring *ring)
> +	LIBURING_NOEXCEPT
> +{

I would probably just name this io_uring_get_sqe128() and have it work
for both MIXED and SQE128. That would make for a cleaner API for the
application.

> +	struct io_uring_sq *sq = &ring->sq;
> +	unsigned head = io_uring_load_sq_head(ring), tail = sq->sqe_tail;
> +	struct io_uring_sqe *sqe;
> +
> +	if (!(ring->flags & IORING_SETUP_SQE_MIXED))
> +		return NULL;
> +
> +	if (((tail + 1) & sq->ring_mask) == 0) {
> +		if ((tail + 2) - head >= sq->ring_entries)
> +			return NULL;
> +
> +		sqe = _io_uring_get_sqe(ring);
> +		io_uring_prep_nop(sqe);
> +		sqe->flags |= IOSQE_CQE_SKIP_SUCCESS;
> +		tail = sq->sqe_tail;
> +	} else if ((tail + 1) - head >= sq->ring_entries) {
> +		return NULL;
> +	}
> +
> +	sqe = &sq->sqes[tail & sq->ring_mask];
> +	sq->sqe_tail = tail + 2;
> +	io_uring_initialize_sqe(sqe);
> +
> +	return sqe;
> +}

Spurious newline before turn, just group them.

-- 
Jens Axboe