[RFC,0/3] Inline sqe_submit

Message

Pavel Begunkov Nov. 5, 2019, 11:04 p.m. UTC

The proposal is to not pass struct sqe_submit as a separate entity,
but always use req->submit instead, so there will be less stuff to
care about. The reasoning begind is code simplification.

Also, I've got steady +1% throughput improvement for nop tests.
Though, it's highly system-dependent, and I wouldn't count on it.

P.S. I'll double check the patches, if the idea is accepted.


Pavel Begunkov (3):
  io_uring: allocate io_kiocb upfront
  io_uring: Use submit info inlined into req
  io_uring: use inlined struct sqe_submit

 fs/io_uring.c | 127 ++++++++++++++++++++++++--------------------------
 1 file changed, 61 insertions(+), 66 deletions(-)

Comments

Jens Axboe Nov. 5, 2019, 11:37 p.m. UTC | #1

On 11/5/19 4:04 PM, Pavel Begunkov wrote:
> The proposal is to not pass struct sqe_submit as a separate entity,
> but always use req->submit instead, so there will be less stuff to
> care about. The reasoning begind is code simplification.
> 
> Also, I've got steady +1% throughput improvement for nop tests.
> Though, it's highly system-dependent, and I wouldn't count on it.
> 
> P.S. I'll double check the patches, if the idea is accepted.

I like the idea (a lot), makes the whole thing easier to follow as well.
Just one comment on patch 3, that needs fixing.

Pavel Begunkov Nov. 5, 2019, 11:45 p.m. UTC | #2

On 06/11/2019 02:37, Jens Axboe wrote:
> On 11/5/19 4:04 PM, Pavel Begunkov wrote:
>> The proposal is to not pass struct sqe_submit as a separate entity,
>> but always use req->submit instead, so there will be less stuff to
>> care about. The reasoning begind is code simplification.
>>
>> Also, I've got steady +1% throughput improvement for nop tests.
>> Though, it's highly system-dependent, and I wouldn't count on it.
>>
>> P.S. I'll double check the patches, if the idea is accepted.
> 
> I like the idea (a lot), makes the whole thing easier to follow as well.

Great, than I'll prepare the patches properly and resend it

> Just one comment on patch 3, that needs fixing.
>

Jens Axboe Nov. 5, 2019, 11:48 p.m. UTC | #3

On 11/5/19 4:45 PM, Pavel Begunkov wrote:
> On 06/11/2019 02:37, Jens Axboe wrote:
>> On 11/5/19 4:04 PM, Pavel Begunkov wrote:
>>> The proposal is to not pass struct sqe_submit as a separate entity,
>>> but always use req->submit instead, so there will be less stuff to
>>> care about. The reasoning begind is code simplification.
>>>
>>> Also, I've got steady +1% throughput improvement for nop tests.
>>> Though, it's highly system-dependent, and I wouldn't count on it.
>>>
>>> P.S. I'll double check the patches, if the idea is accepted.
>>
>> I like the idea (a lot), makes the whole thing easier to follow as well.
> 
> Great, than I'll prepare the patches properly and resend it

Perfect - doesn't look like it'll conflict with the submission path
cleanup (which also looks good), but if it does, just collate them into
a single series.