Re: [PATCH V2 2/2] change shrinker API by passing shrink_control struct

[KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>]

> Hmm, got Nick's email wrong.
> 
> --Ying

Ping.
Can you please explain current status? When I can see your answer?



> 
> On Tue, Apr 26, 2011 at 6:15 PM, Ying Han <yinghan@google.com> wrote:
>> On Tue, Apr 26, 2011 at 5:47 PM, KOSAKI Motohiro
>> <kosaki.motohiro@jp.fujitsu.com> wrote:
>>>> > > \x1a{
>>>> > > \x1a \x1a \x1a struct xfs_mount *mp;
>>>> > > \x1a \x1a \x1a struct xfs_perag *pag;
>>>> > > \x1a \x1a \x1a xfs_agnumber_t \x1aag;
>>>> > > \x1a \x1a \x1a int \x1a \x1a \x1a \x1a \x1a \x1a reclaimable;
>>>> > > + \x1a \x1a int nr_to_scan = sc->nr_slab_to_reclaim;
>>>> > > + \x1a \x1a gfp_t gfp_mask = sc->gfp_mask;
>>>> >
>>>> > And, this very near meaning field .nr_scanned and .nr_slab_to_reclaim
>>>> > poped up new question.
>>>> > Why don't we pass more clever slab shrinker target? Why do we need pass
>>>> > similar two argument?
>>>> >
>>>>
>>>> I renamed the nr_slab_to_reclaim and nr_scanned in shrink struct.
>>>
>>> Oh no. that's not naming issue. example, Nick's previous similar patch pass
>>> zone-total-pages and how-much-scanned-pages. (ie shrink_slab don't calculate
>>> current magical target scanning objects anymore)
>>> \x1a \x1a \x1a \x1aie, \x1a"4 * \x1amax_pass \x1a* (scanned / nr- lru_pages-in-zones)"
>>>
>>> Instead, individual shrink_slab callback calculate this one.
>>> see git://git.kernel.org/pub/scm/linux/kernel/git/npiggin/linux-npiggin.git
>>>
>>> I'm curious why you change the design from another guy's previous very similar effort and
>>> We have to be convinced which is better.
>>
>> Thank you for the pointer. My patch is intended to consolidate all
>> existing parameters passed from reclaim code
>> to the shrinker.
>>
>> Talked w/ Nick and Andrew from last LSF, \x1awe agree that this patch
>> will be useful for other extensions later which allows us easily
>> adding extensions to the shrinkers without shrinker files. Nick and I
>> talked about the effort later to pass the nodemask down to the
>> shrinker. He is cc-ed in the thread. Another thing I would like to
>> repost is to add the reclaim priority down to the shrinker, which we
>> won't throw tons of page caches pages by reclaiming one inode slab
>> object.
>>
>> --Ying