Re: [PATCH v2] fs: Add aio iopriority support for block_dev

From: Jens Axboe
Date: Thu May 03 2018 - 17:20:39 EST


On 5/3/18 2:58 PM, Adam Manzanares wrote:
>
>
> On 5/3/18 1:24 PM, Jens Axboe wrote:
>> On 5/3/18 2:15 PM, Adam Manzanares wrote:
>>>
>>>
>>> On 5/3/18 11:33 AM, Matthew Wilcox wrote:
>>>> On Thu, May 03, 2018 at 11:21:14AM -0700, adam.manzanares@xxxxxxx wrote:
>>>>> If we want to avoid bloating struct kiocb, I suggest we turn the private field
>>>>> into a union of the private and ki_ioprio field. It seems like the users of
>>>>> the private field all use it at a point where we can yank the priority from
>>>>> the kiocb before the private field is used. Comments and suggestions welcome.
>>>>
>>>> Or we could just make ki_hint a u8 or u16 ... seems unlikely we'll need
>>>> 32 bits of ki_hint. (currently defined values are 1-5)
>>>
>>> I like the approach of using a u16 for the ki_hint. I'll update and
>>> resubmit.
>>
>> It's intended to be a mask. If you do shrink it for now, then we need some
>> guard code to ensure it can always carry what it needs to.
>>
>
> Got it, I'll add the guard to rw_hint_valid along with a comment about
> being limited by the size of ki_hint in case we get to a situation where
> 16 bits is not enough.

Other way around - the API should not be limited by the fact that some
smaller type was chosen for an internal structure. Hence the guard/check
should not be in rw_hint_valid, but rather around where you assign
ki_hint. If/when we do extend the read/write hints, then we'll
potentially need to bump ki_hint to accommodate it.

--
Jens Axboe