Re: [PATCH V5 5/8] memcg: add per cgroup writeback pages accounting

From: Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org>
To: Sha Zhengju <handai.szj-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>
Cc: "linux-fsdevel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org"
	<linux-fsdevel-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>,
	"linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org"
	<linux-mm-Bw31MaZKKs3YtjvyW6yDsg@public.gmane.org>,
	Cgroups <cgroups-u79uwXL29TY76Z2rM5mHXA@public.gmane.org>,
	KAMEZAWA Hiroyuki
	<kamezawa.hiroyu-+CUm20s59erQFUHtdCDX3A@public.gmane.org>,
	Glauber Costa <glommer-Re5JQEeQqe8AvxtiuMwx3w@public.gmane.org>,
	Greg Thelen <gthelen-hpIqsD4AKlfQT0dZR+AlfA@public.gmane.org>,
	Wu Fengguang
	<fengguang.wu-ral2JQCrhuEAvxtiuMwx3w@public.gmane.org>,
	Andrew Morton
	<akpm-de/tnXTf+JLsfHDXvbKv3WD2FQJk+8+b@public.gmane.org>,
	Sha Zhengju <handai.szj-3b8fjiQLQpfQT0dZR+AlfA@public.gmane.org>
Subject: Re: [PATCH V5 5/8] memcg: add per cgroup writeback pages accounting
Date: Sun, 4 Aug 2013 12:08:55 +0200	[thread overview]
Message-ID: <20130804100855.GD24005@dhcp22.suse.cz> (raw)
In-Reply-To: <CAFj3OHV-VCKJfe6bv4UMvv+uj4LELDXsieRZFJD06Yrdyy=XxA-JsoAwUIsXosN+BqQ9rBEUg@public.gmane.org>

On Sat 03-08-13 17:25:01, Sha Zhengju wrote:
> On Thu, Aug 1, 2013 at 10:53 PM, Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org> wrote:
> > On Thu 01-08-13 19:54:11, Sha Zhengju wrote:
> >> From: Sha Zhengju <handai.szj-3b8fjiQLQpfQT0dZR+AlfA@public.gmane.org>
> >>
> >> Similar to dirty page, we add per cgroup writeback pages accounting. The lock
> >> rule still is:
> >>         mem_cgroup_begin_update_page_stat()
> >>         modify page WRITEBACK stat
> >>         mem_cgroup_update_page_stat()
> >>         mem_cgroup_end_update_page_stat()
> >>
> >> There're two writeback interfaces to modify: test_{clear/set}_page_writeback().
> >> Lock order:
> >>       --> memcg->move_lock
> >>         --> mapping->tree_lock
> >>
> >> Signed-off-by: Sha Zhengju <handai.szj-3b8fjiQLQpfQT0dZR+AlfA@public.gmane.org>
> >
> > Looks good to me. Maybe I would suggest moving this patch up the stack
> > so that it might get merged earlier as it is simpler than dirty pages
> > accounting. Unless you insist on having the full series merged at once.
> 
> I think the following three patches can be merged earlier:
>       1/8 memcg: remove MEMCG_NR_FILE_MAPPED
>       3/8 memcg: check for proper lock held in mem_cgroup_update_page_stat
>       5/8 memcg: add per cgroup writeback pages accounting
> 
> Do I need to resent them again for you or they're enough?

This is a question for Andrew. I would go with them as they are.

> One more word, since dirty accounting is essential to future memcg
> dirty page throttling and it is not an optional feature now, I suspect
> whether we can merge the following two as well and leave the overhead
> optimization a separate series.  :p

I wouldn't hurry it. We need numbers for serious testing to see the
overhead. It is still just a small step towards dirty throttling.

>       4/5 memcg: add per cgroup dirty pages accounting
>       8/8 memcg: Document cgroup dirty/writeback memory statistics
> 
> The 2/8 ceph one still need more improvement, I'll separate it next version.
> 
> >
> > Acked-by: Michal Hocko <mhocko-AlSwsSmVLrQ@public.gmane.org>
> 
> Thank you.
[...]
-- 
Michal Hocko
SUSE Labs