Re: [PATCH 2/4] mm: memcontrol: re-use global VM event enum

From: Michal Hocko
Date: Tue Apr 11 2017 - 08:30:51 EST


On Mon 10-04-17 10:13:34, Johannes Weiner wrote:
> On Fri, Apr 07, 2017 at 02:47:02PM +0200, Michal Hocko wrote:
> > I do agree that we should share global and memcg specific events constants
> > but I am not sure we want to share all of them. Would it make sense to
> > reorganize the global enum and put those that are shared to the
> > beginning? We wouldn't need the memcg specific translation then.
>
> I'm not sure I follow. Which translation?

Sorry, I should have said s@translation@filtering@ by memcg1_events*

> > Anyway, two comments on the current implementation.
> >
> > On Tue 04-04-17 18:01:46, Johannes Weiner wrote:
> > [...]
> > > +/* Cgroup-specific events, on top of universal VM events */
> > > +enum memcg_event_item {
> > > + MEMCG_LOW = NR_VM_EVENT_ITEMS,
> > > + MEMCG_HIGH,
> > > + MEMCG_MAX,
> > > + MEMCG_OOM,
> > > + MEMCG_NR_EVENTS,
> > > +};
> >
> > The above should mention that each supported global VM event should
> > provide the corresponding translation
> >
> > [...]
> >
> > here...
> > > +/* Universal VM events cgroup1 shows, original sort order */
> > > +unsigned int memcg1_events[] = {
> > > + PGPGIN,
> > > + PGPGOUT,
> > > + PGFAULT,
> > > + PGMAJFAULT,
> > > +};
> > > +
> > > +static const char *const memcg1_event_names[] = {
> > > + "pgpgin",
> > > + "pgpgout",
> > > + "pgfault",
> > > + "pgmajfault",
> > > +};
> >
> > the naming doesn't make it easier to undestand why we need this.
> > global2memcg_event?
>
> This is just to keep the file order consistent. It could have been
> done like memory.stat in cgroup2, where we simply do
>
> seq_printf(s, "pgmajfault %lu\n", stat[PGMAJFAULT]);
>
> but I didn't want to change the v1 code too much. So these two arrays
> are just a sorted list of global VM events shown in v1's memory.stat.

You would still have to know which are the relevant parts of the global
starts that we account for memcg.
--
Michal Hocko
SUSE Labs