On Wed, Jul 08, 2020 at 10:38:44AM -0600, Jens Axboe wrote:
>On 7/8/20 10:33 AM, Matthew Wilcox wrote:
>> On Wed, Jul 08, 2020 at 06:08:12PM +0200, Javier González wrote:
>>>> I just wanted to get clarification there, because to me it sounded like
>>>> you expected Kanchan to do it, and Kanchan assuming it "was sorted". I'd
>>>> consider that a prerequisite for the append series as far as io_uring is
>>>> concerned, hence _someone_ needs to actually do it ;-)
>>
>> I don't know that it's a prerequisite in terms of the patches actually
>> depend on it.  I appreciate you want it first to ensure that we don't bloat
>> the kiocb.
>
>Maybe not for the series, but for the io_uring addition it is.
>
>>> I believe Kanchan meant that now the trade-off we were asking to
>>> clear out is sorted.
>>>
>>> We will send a new version shortly for the current functionality - we
>>> can see what we are missing on when the uring interface is clear.
>>
>> I've started work on a patch series for this.  Mostly just waiting for
>> compilation now ... should be done in the next few hours.
>
>Great!

Jens, Matthew - I'm sorry for creating the confusion. By "looks sorted"
I meant the performance-implications and the room-for-pointer. For the
latter I was thinking to go by your suggestion not to bloat the kiocb, and
use io_kiocb instead.
If we keep, there will be two paths to update that pointer, one using
ki_complete(....,ret2) and another directly - which does not seem good.

On a different note: trimming kiocb by decoupling ki_complete work looks
too good to be done by me :-)