Re: [PATCH liburing] test/io_uring_passthrough: add test for vectored fixed-buffers

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

On 5/21/25 11:35, Anuj gupta wrote:
>> LGTM, it's great to have the test, thanks Anuj. FWIW, that's the
>> same way I tested the kernel patch.
>>
>> Somewhat unrelated questions, is there some particular reason why all
>> vectored versions are limited to 1 entry iovec? And why do we even care
>> calling io_uring_prep_read/write*() helpers when non of the rw related
>> fields set are used by passthrough? i.e. iovec passed in the second half
>> of the sqe.
> 
> Thanks, Pavel!
> 
> Regarding the vectored I/O being limited to 1 iovec — yeah, I kept it
> simple initially because the plumbing was easier that way. It’s the same
> in test/read-write.c, where vectored calls also use just one iovec. But
> I agree, for better coverage, it makes sense to test with multiple
> iovecs. I’ll prepare and post a follow-up patch that adds that.

Got it, it's overlooked that rw doesn't pass longer iovecs. It'd be
great to unify the main loops of read-write.c, iopoll.c and nvme
passthrough tests. All of them do the same thing just initialising
sqes in different ways.

> About the use of io_uring_prep_read/write*() helpers — you're right,
> they don’t really add much here since the passthrough command handles
> the fields directly. I’ll work on a cleanup patch to remove those and
> simplify the submission code.

I don't care about the test itself much, but it means there
are lots of unused fields for the nvme commands that are not
checked by the kernel and hence can't be reused in the future.
That's not great

-- 
Pavel Begunkov