[PATCH] memcg: fix behavior of shard anon pages at task_move (WasRe: [PATCH v3 2/2] memcg: avoid THP split in task migration

[KAMEZAWA Hiroyuki]

On Fri, 9 Mar 2012 15:01:09 +0900
Daisuke Nishimura <nishimura@xxxxxxxxxxxxxxxxx> wrote:

> On Fri, 9 Mar 2012 12:24:48 +0900
> KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> wrote:
> > > I'd rather delete than add code here!
> > > 
> > 
> > As a user, for Fujitsu, I believe it's insane to move task between cgroups.
> > So, I have no benefit from this code, at all.
> > Ok, maybe I'm not a stakeholder,here.
> > 
> I agree that moving tasks between cgroup is not a sane operation,
> users won't do it so frequently, but I cannot prevent that.
> That's why I implemented this feature.
> 
> > If users say all shared pages should be moved, ok, let's move.
> > But change of behavior should be documented and implemented in an independet
> > patch. CC'ed Nishimura-san, he implemetned this, a real user.
> > 
> To be honest, shared anon is not my concern. My concern is 
> shared memory(that's why, mapcount is not checked as for file pages.
> I assume all processes sharing the same shared memory will be moved together).
> So, it's all right for me to change the behavior for shared anon(or leave
> it as it is).
> 

Thank you for comment. Then, here is a patch.

Other opinions ?

==