Re: [PATCH v2 0/9] Add support for vectored registered buffers

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

On 3/6/25 22:59, Jens Axboe wrote:
> On 3/6/25 5:10 AM, Jens Axboe wrote:
>> On 3/6/25 5:01 AM, Jens Axboe wrote:
>>>
>>> On Tue, 04 Mar 2025 15:40:21 +0000, Pavel Begunkov wrote:
>>>> Add registered buffer support for vectored io_uring operations. That
>>>> allows to pass an iovec, all entries of which must belong to and
>>>> point into the same registered buffer specified by sqe->buf_index.
>>>>
>>>> The series covers zerocopy sendmsg and reads / writes. Reads and
>>>> writes are implemented as new opcodes, while zerocopy sendmsg
>>>> reuses IORING_RECVSEND_FIXED_BUF for the api.
>>>>
>>>> [...]
>>>
>>> Applied, thanks!
>>>
>>> [1/9] io_uring: introduce struct iou_vec
>>>        commit: 32fd3277b4ae0f5e6f3a306b464f9b031e2408a8
>>> [2/9] io_uring: add infra for importing vectored reg buffers
>>>        commit: 1a3339cbca2225dbcdc1f4da2b25ab83da818f1d
>>> [3/9] io_uring/rw: implement vectored registered rw
>>>        commit: 7965e1cd6199cf9c87fa02e904cbc50c45c7310f
>>> [4/9] io_uring/rw: defer reg buf vec import
>>>        commit: 5f0a1f815dad9490db822013a2f1feba3371f4d1
>>> [5/9] io_uring/net: combine msghdr copy
>>>        commit: bc007e0aea60926b75b6a459ad8cf7ac357fb290
>>> [6/9] io_uring/net: pull vec alloc out of msghdr import
>>>        commit: 8ff671f394f97e31bc6c1acec9ebbdb108177df9
>>> [7/9] io_uring/net: convert to struct iou_vec
>>>        commit: 57b309177530bf99e59da21d1b1888ac4024072a
>>> [8/9] io_uring/net: implement vectored reg bufs for zctx
>>>        commit: 6836bdad87cb83e96df0702d02d264224b0ffd2d
>>> [9/9] io_uring: cap cached iovec/bvec size
>>>        commit: 0be2ba0a44e3670ac3f9eecd674341d77767288d
>>
>> Note: the vectored fixed read/write opcodes got renumbered, as they
>> didn't sit on top of the epoll wait patches. Just a heads up, in terms
>> of the liburing side.
>>
>> I'll get the basic epoll wait bits pushed up to liburing as well.
> 
> And one more note: this breaks 32-bit compiles due to a bad assumption
> on iovec vs bio_vec sizing, so I've dropped it for now. Hopefully we can
> get a v3 into the 6.15 branch.

I saw that, at least the build check did the job. I'm inclined
to the option in my reply to 2/9. It's generic in a sense that
it can be unconditional, might be a good idea to kill the
if and have extra memory consumption for now until improved.

-- 
Pavel Begunkov