On Tue, Oct 15, 2019 at 05:04:44PM +0300, Konstantin Khlebnikov wrote:
> On 15/10/2019 16.53, Johannes Weiner wrote:
> > On Tue, Oct 15, 2019 at 11:09:59AM +0300, Konstantin Khlebnikov wrote:
> > > Mapped, dirty and writeback pages are also counted in per-lruvec stats.
> > > These counters needs update when page is moved between cgroups.
> > >
> > > Fixes: 00f3ca2c2d66 ("mm: memcontrol: per-lruvec stats infrastructure")
> > > Signed-off-by: Konstantin Khlebnikov <khlebnikov@xxxxxxxxxxxxxx>
> >
> > Acked-by: Johannes Weiner <hannes@xxxxxxxxxxx>
> >
> > Please mention in the changelog that currently is nobody *consuming*
> > the lruvec versions of these counters and that there is no
> > user-visible effect. Thanks
> >
>
> Maybe just kill all these per-lruvec counters?
> I see only one user which have no alternative data source: WORKINGSET_ACTIVATE.
>
> This will save some memory: 32 * sizeof(long) * nr_nodes * nr_cpus bytes
This is backwards, see my reply to Michal, as well as the patches at
https://lore.kernel.org/linux-mm/20190603210746.15800-1-hannes@xxxxxxxxxxx/
We're not using the lruvec counters in all places where we should.
