Re: [PATCH 08/31] mm, vmscan: simplify the logic deciding whether kswapd sleeps

From: Mel Gorman
Date: Wed Jul 06 2016 - 04:31:31 EST


On Wed, Jul 06, 2016 at 09:30:54AM +0900, Minchan Kim wrote:
> On Tue, Jul 05, 2016 at 11:26:39AM +0100, Mel Gorman wrote:
>
> <snip>
>
> > > > @@ -3418,10 +3426,10 @@ void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
> > > > if (!cpuset_zone_allowed(zone, GFP_KERNEL | __GFP_HARDWALL))
> > > > return;
> > > > pgdat = zone->zone_pgdat;
> > > > - if (pgdat->kswapd_max_order < order) {
> > > > - pgdat->kswapd_max_order = order;
> > > > - pgdat->classzone_idx = min(pgdat->classzone_idx, classzone_idx);
> > > > - }
> > > > + if (pgdat->kswapd_classzone_idx == -1)
> > > > + pgdat->kswapd_classzone_idx = classzone_idx;
> > >
> > > It's tricky. Couldn't we change kswapd_classzone_idx to integer type
> > > and remove if above if condition?
> > >
> >
> > It's tricky and not necessarily better overall. It's perfectly possible
> > to be woken up for zone index 0 so it's changing -1 to another magic
> > value.
>
> I don't get it. What is a problem with this?
>

It becomes difficult to tell the difference between "no wakeup and init to
zone 0" and "wakeup and reclaim for zone 0". At least that's the problem
I ran into when I tried before settling on -1.

--
Mel Gorman
SUSE Labs