On Wed 25-01-12 15:07:47, Ying Han wrote:

On Tue, Jan 24, 2012 at 12:43 AM, Michal Hocko [off-list ref] wrote:

quoted

On Mon 23-01-12 14:05:33, Ying Han wrote:

[...]

quoted

quoted

Just want to make sure I understand it, even we make the lock
per-memcg, there is still a false sharing of pc within one memcg.

Yes that is true. I have missed that we might fault in several pages at
once but this would happen only during task move, right? And that is not
a hot path anyway. Or?

I was thinking of page-statistics update which is hot path. If the
moving task and non-moving task share the same per-memcg lock, any
page-statistic update from the non-moving task will be blocked? Sorry
If i missed something here :)

OK, I got your point, finally. I guess there is a plan to reduce the
effect by array of locks.

-- 
Michal Hocko
SUSE Labs
SUSE LINUX s.r.o.
Lihovarska 1060/12
190 00 Praha 9    
Czech Republic