Re: [RFC PATCH] mm/vmscan: fix high cpu usage of kswapd if there

[hejianet]

Hi Michal

On 22/02/2017 7:41 PM, Michal Hocko wrote:
> On Wed 22-02-17 17:04:48, Jia He wrote:
> > When I try to dynamically allocate the hugepages more than system total
> > free memory:
> > e.g. echo 4000 >/proc/sys/vm/nr_hugepages
>
> I assume that the command has terminated with less huge pages allocated
> than requested but
Yes, at last the allocated hugepages are less than 4000
HugePages_Total:    1864
HugePages_Free:     1864
HugePages_Rsvd:        0
HugePages_Surp:        0
Hugepagesize:      16384 kB

In the bad case, although kswapd takes 100% cpu, the number of
HugePages_Total is not increase at all.

> > Node 3, zone      DMA
> [...]
> >   pages free     2951
> >         min      2821
> >         low      3526
> >         high     4231
>
> it left the zone below high watermark with
>
> >    node_scanned  0
> >         spanned  245760
> >         present  245760
> >         managed  245388
> >       nr_free_pages 2951
> >       nr_zone_inactive_anon 0
> >       nr_zone_active_anon 0
> >       nr_zone_inactive_file 0
> >       nr_zone_active_file 0
>
> no pages reclaimable, so kswapd will not go to sleep. It would be quite
> easy and comfortable to call it a misconfiguration but it seems that
> it might be quite easy to hit with NUMA machines which have large
> differences in the node sizes. I guess it makes sense to back off
> the kswapd rather than burning CPU without any way to make forward
> progress.
agree.
> [...]
>
> > diff --git a/mm/vmscan.c b/mm/vmscan.c
> > index 532a2a7..a05e3ab 100644
> > --- a/mm/vmscan.c
> > +++ b/mm/vmscan.c
> > @@ -3139,7 +3139,8 @@ static bool prepare_kswapd_sleep(pg_data_t *pgdat, int order, int classzone_idx)
> >  		if (!managed_zone(zone))
> >  			continue;
> >
> > -		if (!zone_balanced(zone, order, classzone_idx))
> > +		if (!zone_balanced(zone, order, classzone_idx)
> > +			&& zone_reclaimable_pages(zone))
> >  			return false;
>
> OK, this makes some sense, although zone_reclaimable_pages doesn't count
> SLAB reclaimable pages. So we might go to sleep with a reclaimable slab
> still around. This is not really easy to address because the reclaimable
> slab doesn't really imply that those pages will be reclaimed...
Yes, even in the bad case, when kswapd takes all the cpu, the reclaimable
pages are not decreased
> >  	}
> >
> > @@ -3502,6 +3503,7 @@ void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
> >  {
> >  	pg_data_t *pgdat;
> >  	int z;
> > +	int node_has_relaimable_pages = 0;
> >
> >  	if (!managed_zone(zone))
> >  		return;
> > @@ -3522,8 +3524,15 @@ void wakeup_kswapd(struct zone *zone, int order, enum zone_type classzone_idx)
> >
> >  		if (zone_balanced(zone, order, classzone_idx))
> >  			return;
> > +
> > +		if (!zone_reclaimable_pages(zone))
> > +			node_has_relaimable_pages = 1;
>
> What, this doesn't make any sense? Did you mean if (zone_reclaimable_pages)?
I mean, if any one zone has reclaimable pages, then this zone's *node* has
reclaimable pages. Thus, the kswapN for this node should be waken up.
e.g. node 1 has 2 zones.
zone A has no reclaimable pages but zone B has.
Thus node 1 has reclaimable pages, and kswapd1 will be waken up.
I use node_has_relaimable_pages in the loop to check all the zones' reclaimable
pages number. So I prefer the name node_has_relaimable_pages instead of
zone_has_relaimable_pages

Did I understand it correctly? Thanks

B.R.
Jia
> >  	}
> >
> > +	/* Dont wake kswapd if no reclaimable pages */
> > +	if (!node_has_relaimable_pages)
> > +		return;
> > +
> >  	trace_mm_vmscan_wakeup_kswapd(pgdat->node_id, zone_idx(zone), order);
> >  	wake_up_interruptible(&pgdat->kswapd_wait);
> >  }
> > --
> > 1.8.5.6
> >
> > --
> > To unsubscribe, send a message with 'unsubscribe linux-mm' in
> > the body to majordomo@xxxxxxxxxx  For more info on Linux MM,
> > see: http://www.linux-mm.org/ .
> > Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>