Re: [PATCH] mm: vmscan: show zone type in kswapd tracepoints

From: Yafang Shao
Date: Tue Mar 12 2019 - 10:11:28 EST


On Tue, Mar 12, 2019 at 9:38 PM Michal Hocko <mhocko@xxxxxxxxxx> wrote:
>
> On Tue 12-03-19 19:04:43, Yafang Shao wrote:
> > On Mon, Mar 11, 2019 at 4:47 PM Michal Hocko <mhocko@xxxxxxxxxx> wrote:
> > >
> > > On Fri 01-03-19 15:38:54, Yafang Shao wrote:
> > > > If we want to know the zone type, we have to check whether
> > > > CONFIG_ZONE_DMA, CONFIG_ZONE_DMA32 and CONFIG_HIGHMEM are set or not,
> > > > that's not so convenient.
> > > >
> > > > We'd better show the zone type directly.
> > >
> > > I do agree that zone number is quite PITA to process in general but do
> > > we really need this information in the first place? Why do we even care?
> > >
> >
> > Sometimes we want to know this event occurs in which zone, then we can
> > get the information of this zone,
> > for example via /proc/zoneinfo.
> > It could give us more information for debugging.
>
> Could you be more specific please?
>

Honestly speaking, this one hasn't help us fix the real issue yet.

> > > Zones are an MM internal implementation details and the more we export
> > > to the userspace the more we are going to argue about breaking userspace
> > > when touching them. So I would rather not export that information unless
> > > it is terribly useful.
> > >
> >
> > I 'm not sure whether zone type is terribly useful or not, but the
> > 'zid' is useless at all.
> >
> > I don't agree that Zones are MM internal.
> > We can get the zone type in many ways, for example /proc/zoneinfo.
> >
> > If we show this event occurs in which zone, we'd better show the zone type,
> > or we should drop this 'zid'.
>
> Yes, I am suggesting the later. If somebody really needs it then I would
> like to see a _specific_ usecase. Then we can add the proper name.

This 'zid' always seems like a noise currently.
I will send a patch to drop this one.

Thanks
Yafang