Re: [PATCH] Add file based RSS accounting for memory resourcecontroller (v2)

From: KAMEZAWA Hiroyuki
Date: Thu Apr 16 2009 - 22:05:38 EST


On Fri, 17 Apr 2009 07:10:42 +0530
Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote:

> * KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> [2009-04-17 09:14:59]:
>
> > On Thu, 16 Apr 2009 17:33:16 +0530
> > Balbir Singh <balbir@xxxxxxxxxxxxxxxxxx> wrote:
> >
> > > * KAMEZAWA Hiroyuki <kamezawa.hiroyu@xxxxxxxxxxxxxx> [2009-04-16 17:15:35]:
> > >
> > > >
> > > > > Sorry, some troubles found. Ignore above Ack. 3points now.
> > > > >
> > > > > 1. get_cpu should be after (*)
> > > > > ==mem_cgroup_update_mapped_file_stat()
> > > > > + int cpu = get_cpu();
> > > > > +
> > > > > + if (!page_is_file_cache(page))
> > > > > + return;
> > > > > +
> > > > > + if (unlikely(!mm))
> > > > > + mm = &init_mm;
> > > > > +
> > > > > + mem = try_get_mem_cgroup_from_mm(mm);
> > > > > + if (!mem)
> > > > > + return;
> > > > > + ----------------------------------------(*)
> > > > > + stat = &mem->stat;
> > > > > + cpustat = &stat->cpustat[cpu];
> > > > > +
> > > > > + __mem_cgroup_stat_add_safe(cpustat, MEM_CGROUP_STAT_MAPPED_FILE, val);
> > > > > + put_cpu();
> > > > > +}
> > > > > ==
> > >
> > > Yes or I should have a goto
> > >
> > > > >
> > > > > 2. In above, "mem" shouldn't be got from "mm"....please get "mem" from page_cgroup.
> > > > > (Because it's file cache, pc->mem_cgroup is not NULL always.)
> > >
> > > Hmmm.. not sure I understand this part. Are you suggesting that mm can
> > > be NULL?
> > No.
> >
> > > I added the check for !mm as a safety check. Since this
> > > routine is only called from rmap context, mm is not NULL, hence mem
> > > should not be NULL. Did you find a race between mm->owner assignment
> > > and lookup via mm->owner?
> > >
> > No.
> >
> > page_cgroup->mem_cgroup != try_get_mem_cgroup_from_mm(mm); in many many cases.
> >
> > For example, libc and /bin/*** is tend to be loaded into default cgroup at boot but
> > used by many cgroups. But mapcount of page caches for /bin/*** is 0 if not running.
> >
> > Then, File_Mapped can be greater than Cached easily if you use mm->owner.
> >
> > I can't estimate RSS in *my* cgroup if File_Mapped includes pages which is under
> > other cgroups. It's meaningless.
> > Especially, when Cached==0 but File_Mapped > 0, I think "oh, the kernel leaks somehing..hmm..."
> >
> > By useing page_cgroup->mem_cgroup, we can avoid above mess.
>
> Yes, I see your point. I wanted mapped_file to show up in the cgroup
> that mapped the page. But this works for me as well, but that means
> we'll nest the page cgroup lock under the PTE lock.

Don't worry. we do that nest at ANON's uncharge(), already.

About cost:

IIUC, the number of "mapcount 0->1/1->0" of file caches are much smaller than
that of o Anon. And there will be not very much cache pingpong.

If you use PCG_MAPPED flag in page_cgroup (as my patch), you can use
not-atomic version of set/clear when update is only under lock_page_cgroup().
If you find better way, plz use it. But we can't avoid some kind of atomic ops
for correct accounting, I think.

Thanks,
-Kame


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/