Re: [PATCHv6] io_uring: add support for IORING_SETUP_SQE_MIXED

[Jens Axboe <axboe@kernel.dk>]

On 10/22/25 7:53 AM, Keith Busch wrote:
> On Wed, Oct 22, 2025 at 07:34:22AM -0600, Jens Axboe wrote:
>> On 10/22/25 6:00 AM, Nathan Chancellor wrote:
>>> On Tue, Oct 21, 2025 at 04:02:28PM -0600, Jens Axboe wrote:
>>>>
>>>> On Thu, 16 Oct 2025 11:09:38 -0700, Keith Busch wrote:
>>>>> Normal rings support 64b SQEs for posting submissions, while certain
>>>>> features require the ring to be configured with IORING_SETUP_SQE128, as
>>>>> they need to convey more information per submission. This, in turn,
>>>>> makes ALL the SQEs be 128b in size. This is somewhat wasteful and
>>>>> inefficient, particularly when only certain SQEs need to be of the
>>>>> bigger variant.
>>>>>
>>>>> [...]
>>>>
>>>> Applied, thanks!
>>>>
>>>> [1/1] io_uring: add support for IORING_SETUP_SQE_MIXED
>>>>       commit: 31dc41afdef21f264364288a30013b538c46152e
>>>
>>> This needs a pretty obvious fix up as clang points out:
>>>
>>>   io_uring/fdinfo.c:103:22: error: variable 'sqe' is uninitialized when used here [-Werror,-Wuninitialized]
>>>     103 |                 opcode = READ_ONCE(sqe->opcode);
>>>         |                                    ^~~
>>>
>>> I would have sent a formal patch but since it is at the top, I figured
>>> it would get squashed anyways.
>>
>> Indeed - I'll fold this in. Keith, can you add an fdinfo test case for
>> mixed as well?
> 
> Thanks, and will do. Any quick hints on how to ensure fdinfo finds some
> entries to report? I assume I can just put some entries on the queue,
> but don't call io_uring_enter.

Yep that should do it. You probably want to do something similar to
__io_uring_flush_sq() since it needs to be visible on the kernel side,
by having the SQ tail written too. io_uring_submit() does this, but
you'd want to do it without calling io_uring_enter() to submit the
entries.

And then probably do that with various states of mixed sized SQEs in the
SQ ring and verify it looks correct, too.

-- 
Jens Axboe