On Fri, Jul 29, 2011 at 04:50:43PM +0800, Mel Gorman wrote:
> On Fri, Jul 29, 2011 at 08:35:25AM +0800, Shaohua Li wrote:
> > On Thu, 2011-07-28 at 18:56 +0800, Mel Gorman wrote:
> > > On Thu, Jul 28, 2011 at 04:13:01PM +0800, Shaohua Li wrote:
> > > > correctly clear ZONE_CONGESTED. If a zone watermark is ok, we
> > > > should clear ZONE_CONGESTED regardless if this is a high order
> > > > allocation, because pages can be reclaimed in other tasks but
> > > > ZONE_CONGESTED is only cleared in kswapd.
> > > >
> > >
> > > What problem does this solve?
> > >
> > > As it is, for high order allocations it takes the following steps
> > >
> > > If reclaiming at high order {
> > > for each zone {
> > > if all_unreclaimable
> > > skip
> > > if watermark is not met
> > > order = 0
> > > loop again
> > >
> > > /* watermark is met */
> > > clear congested
> > > }
> > > }
> > >
> > > If high orders are failing, kswapd balances for order-0 where there
> > > is already a cleaning of ZONE_CONGESTED if the zone was shrunk and
> > > became balanced. I see the case for hunk 1 of the patch because now
> > > it'll clear ZONE_CONGESTED for zones that are already balanced which
> > > might have a noticable effect on wait_iff_congested. Is this what
> > > you see? Even if it is, it does not explain hunk 2 of the patch.
> > I first looked at the hunk 2 place and thought we don't clear
> > ZONE_CONGESTED there. I then figured out we need do the same thing for
> > the hunk 1. But you are correct, with hunk 1, hunk 2 isn't required.
> > updated patch.
> >
> >
> >
> > correctly clear ZONE_CONGESTED. If a zone watermark is ok, we
> > should clear ZONE_CONGESTED because pages can be reclaimed in
> > other tasks but ZONE_CONGESTED is only cleared in kswapd.
> >
> > Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx>
>
> It would be nice if the changelog was expanded a bit to explain
> why the patch is necessary. You say this is to "correctly clear
> ZONE_CONGESTED" but do not explain why the current code is wrong
> or what the user-visible impact is. For example, even cutting and
> pasting bits of the discussion like the following would have been
> an improvement.
>
> ==== CUT HERE ===
> kswapd is responsible for clearing ZONE_CONGESTED after it balances
> a zone. Unfortunately, if ZONE_CONGESTED was set during a high-order
> allocation, it is possible that kswapd misses clearing it.
>
> At the end of balance_pgdat(), kswapd uses the following logic;
>
> If reclaiming at high order {
> for each zone {
> if all_unreclaimable
> skip
> if watermark is not met
> order = 0
> loop again
>
> /* watermark is met */
> clear congested
> }
> }
>
> i.e. it clears ZONE_CONGESTED if it the zone is balanced. if not,
> it restarts balancing at order-0. However, if the higher zones are
> balanced for order-0, kswapd will miss clearing ZONE_CONGESTED
> as that only happens after a zone is shrunk. This can mean that
> wait_iff_congested() stalls unnecessarily. This patch makes kswapd
> clear ZONE_CONGESTED during its initial highmem->dma scan for zones
> that are already balanced.
>
> ==== CUT HERE ====
>
> This makes review a lot easier and will be helpful in the future if
> someone uses git blame.
>
> Whether you update the changelog or not;
>
> Acked-by: Mel Gorman <mgorman@xxxxxxx>
Ok, updated the changelog.
ZONE_CONGESTED is only cleared in kswapd, but pages can be freed in any task.
It's possible ZONE_CONGESTED isn't cleared in some cases:
1. the zone is already balanced just entering balance_pgdat() for order-0 because
concurrent tasks free memory. In this case, later check will skip the zone as
it's balanced so the flag isn't cleared.
2. high order balance fallbacks to order-0. quote from Mel:
At the end of balance_pgdat(), kswapd uses the following logic;
If reclaiming at high order {
for each zone {
if all_unreclaimable
skip
if watermark is not met
order = 0
loop again
/* watermark is met */
clear congested
}
}
i.e. it clears ZONE_CONGESTED if it the zone is balanced. if not,
it restarts balancing at order-0. However, if the higher zones are
balanced for order-0, kswapd will miss clearing ZONE_CONGESTED
as that only happens after a zone is shrunk.
This can mean that wait_iff_congested() stalls unnecessarily. This patch
makes kswapd clear ZONE_CONGESTED during its initial highmem->dma scan
for zones that are already balanced.
Signed-off-by: Shaohua Li <shaohua.li@xxxxxxxxx>
Acked-by: Mel Gorman <mgorman@xxxxxxx>
Reviewed-by: Minchan Kim <minchan.kim@xxxxxxxxx>
---
mm/vmscan.c | 3 +++
1 file changed, 3 insertions(+)
Index: linux/mm/vmscan.c
===================================================================
--- linux.orig/mm/vmscan.c 2011-07-29 08:24:10.000000000 +0800
+++ linux/mm/vmscan.c 2011-07-29 08:26:29.000000000 +0800
@@ -2494,6 +2494,9 @@ loop_again:
high_wmark_pages(zone), 0, 0)) {
end_zone = i;
break;
+ } else {
+ /* If balanced, clear the congested flag */
+ zone_clear_flag(zone, ZONE_CONGESTED);
}
}
if (i < 0)
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxx. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>