Re: [PATCH -mmotm 0/8] memcg: recharge at task move

From: KAMEZAWA Hiroyuki <kamezawa.hiroyu@jp.fujitsu.com>
To: Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp>
Cc: linux-mm <linux-mm@kvack.org>,
	Andrew Morton <akpm@linux-foundation.org>,
	Balbir Singh <balbir@linux.vnet.ibm.com>,
	Li Zefan <lizf@cn.fujitsu.com>, Paul Menage <menage@google.com>
Subject: Re: [PATCH -mmotm 0/8] memcg: recharge at task move
Date: Mon, 9 Nov 2009 14:16:09 +0900	[thread overview]
Message-ID: <20091109141609.331eee77.kamezawa.hiroyu@jp.fujitsu.com> (raw)
In-Reply-To: <20091109104446.b2d9ef66.nishimura@mxp.nes.nec.co.jp>

On Mon, 9 Nov 2009 10:44:46 +0900
Daisuke Nishimura <nishimura@mxp.nes.nec.co.jp> wrote: 
> > Seems much nicer but I have some nitpicks as already commented.
> > 
> > For [8/8], mm->swap_usage counter may be a help for making it faster.
> > Concern is how it's shared but will not be very big error.
> > 
> will change as I mentioned in another mail.
> 
> I'll repost 3 and 4 as cleanup(I think they are ready for inclusion),
> and post removal-of-memcg_tasklist as a separate patch.
> 
> I'll postpone the body of this feature(waiting for your percpu change
> and per-process swap counter at least).
> 
Thanks. I agree 1-4 are ready for merge. But, please get ack by Paul.
or Li zefan. Paul, Li, how do you think about [1/8] ?

> > > TODO:
> > > - add support for file cache, shmem/tmpfs, and shared(mapcount > 1) pages.
> > > - implement madvise(2) to let users decide the target vma for recharge.
> > > 
> > 
> > About this, I think "force_move_shared_account" flag is enough, I think.
> > But we have to clarify "mmap()ed but not on page table" entries are not
> > moved....
> > 
> You mean swap entries of shmem/tmpfs, do you ? I agree they are hard to handle..
> 
Yes and No. Including mmaped file.

> 
> My concern is:
> 
> - I want to add support for private file caches, shmes/tmpfs pages(including swaps of them),
>   and "shared" pages by some means in future, and let an admin or a middle-ware
>   decide how to handle them.
Yes, I agree that middle ware may want that.

> - Once this feature has been merged(at .33?), I don't want to change the behavior
>   when a user set "recharge_at_immigrate=1".
>   So, I'll "extend" the meaning of "recharge_at_immigrate" or add a new flag file
>   to support other type of charges.

Yes, I think making recharge_at_immigrate as "bitmask" makes sense.
So, showing "recharge_at_immigrate" is a bitmask _now_ both in code and documentation
in clear way will be a help for future.

Thanks,
-Kame

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>