> Il giorno 20 set 2019, alle ore 15:05, Jens Axboe <axboe@xxxxxxxxx> ha scritto:
>
> On 9/20/19 12:58 AM, Paolo Valente wrote:
>>
>>
>>> Il giorno 18 set 2019, alle ore 18:19, Paolo Valente <paolo.valente@xxxxxxxxxx> ha scritto:
>>>
>>>
>>>
>>>> Il giorno 18 set 2019, alle ore 17:19, Tejun Heo <tj@xxxxxxxxxx> ha scritto:
>>>>
>>>> Hello,
>>>>
>>>> On Wed, Sep 18, 2019 at 07:18:50AM +0200, Paolo Valente wrote:
>>>>> A solution that both fulfills userspace request and doesn't break
>>>>> anything for hypothetical users of the current interface already made
>>>>> it to mainline, and Linus liked it too. It is:
>>>>
>>>> Linus didn't like it. The implementation was a bit nasty. That was
>>>> why it became a subject in the first place.
>>>>
>>>>> 19e9da9e86c4 ("block, bfq: add weight symlink to the bfq.weight cgroup parameter")
>>>>>
>>>>> But it was then reverted on Tejun's request to do exactly what we
>>>>> don't want do any longer now:
>>>>> cf8929885de3 ("cgroup/bfq: revert bfq.weight symlink change")
>>>>
>>>> Note that the interface was wrong at the time too.
>>>>
>>>>> So, Jens, Tejun, can we please just revert that revert?
>>>>
>>>> I think presenting both io.weight and io.bfq.weight interfaces are
>>>> probably the best course of action at this point but why does it have
>>>> to be a symlink? What's wrong with just creating another file with
>>>> the same backing function?
>>>>
>>>
>>> I think a symlink would be much clearer for users, given the confusion
>>> already caused by two names for the same parameter. But let's hear
>>> others' opinion too.
>>>
>>
>> Jens, could you express your opinion on this? Any solution you and
>> Tejun agree on is ok for me. Also this new (fourth) possible
>> implementation of this fix, provided that then it is definitely ok for
>> both of you.
>
> Retaining both interfaces is arguably the right solution.
So you also are voting for BFQ to create two files, instead of having a
symlink, aren't you? I just want to be certain before submitting one
more solution.
Looking forward to your confirmation,
Paolo
> It would be
> nice if we didn't have to, but the first bfq variant was incompatible
> with the in-kernel one, so we'll always have that out in the wild.
> Adding everything to stable doesn't work, as we still have existing
> kernels out there with the interface. In fact, in some ways that's
> worse, as you definitely don't want interfaces to change between two
> stable kernels.
>
> I know it's not ideal, and some better initial planning would have
> made it better, but we have to deal with the situation as it stands
> now.
>
> --
> Jens Axboe
