Re: [PATCH] mm: memcontrol: drop superfluous entry in the per-memcg stats array

[Johannes Weiner <hannes-druUgvl0LCNAfugRpC6u6w@public.gmane.org>]

On Tue, Jan 26, 2016 at 01:30:24PM -0800, Andrew Morton wrote:
> On Tue, 26 Jan 2016 15:55:29 -0500 Johannes Weiner <hannes-druUgvl0LCNAfugRpC6u6w@public.gmane.org> wrote:
> 
> > MEM_CGROUP_STAT_NSTATS is just a delimiter for cgroup1 statistics, not
> > an actual array entry. Reuse it for the first cgroup2 stat entry, like
> > in the event array.
> > 
> > ...
> >
> > --- a/include/linux/memcontrol.h
> > +++ b/include/linux/memcontrol.h
> > @@ -51,7 +51,7 @@ enum mem_cgroup_stat_index {
> >  	MEM_CGROUP_STAT_SWAP,		/* # of pages, swapped out */
> >  	MEM_CGROUP_STAT_NSTATS,
> >  	/* default hierarchy stats */
> > -	MEMCG_SOCK,
> > +	MEMCG_SOCK = MEM_CGROUP_STAT_NSTATS,
> >  	MEMCG_NR_STAT,
> >  };
> 
> The code looks a bit odd.  How come mem_cgroup_stat_names[] ends with
> "swap"?  Should MEMCG_SOCK be in there at all?

It's cgroup1 vs. cgroup2 statistics. I'm using the same array in order
to use the original statistics infrastructure. It's a little weird, it
will be much cleaner once everything is converted to percpu_counter.

> And the naming is a bit sad.  "MEM_CGROUP_STAT_FILE_MAPPED" maps to
> "mapped_file", not "file_mapped".

MEM_CGROUP_STAT_FILE_MAPPED is named after NR_FILE_MAPPED because
they're both accounted from the same sites. Who knows why the
user-visible stat was then called mapped_file... :/

And in cgroup2 it's called file_mapped! At least there it'll be
consistent.