On Tue, Aug 30, 2016 at 07:55:08PM +0530, Srikar Dronamraju wrote:

quoted

quoted

This patch seems to hurt FA_DUMP functionality. This behaviour is not
seen on v4.7 but only after this patch.

So when a kernel on a multinode machine with memblock_reserve() such
that most of the nodes have zero available memory, kswapd seems to be
consuming 100% of the time.

Why is FA_DUMP specifically the trigger? If the nodes have zero available
memory then is the zone_populated() check failing when FA_DUMP is enabled? If
so, that would both allow kswapd to wake and stay awake.

The trigger is memblock_reserve() for the complete node memory.  And
this is exactly what FA_DUMP does.  Here again the node has memory but
its all reserved so there is no free memory in the node.

Did you mean populated_zone() when you said zone_populated or have I
mistaken? populated_zone() does return 1 since it checks for
zone->present_pages.

Yes, I meant populated_zone(). Using present pages may have hidden
a long-lived corner case as it was unexpected that an entire node
would be reserved. The old code happened to survive *probably* because
pgdat_reclaimable would look false and kswapd checks for pgdat being
balanced would happen to do the right thing in this case.

Can you check if something like this works?

diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
index d572b78b65e1..cf64a5456cf6 100644
--- a/include/linux/mmzone.h
+++ b/include/linux/mmzone.h

@@ -830,7 +830,7 @@ unsigned long __init node_memmap_size_bytes(int, unsigned long, unsigned long);
 
 static inline int populated_zone(struct zone *zone)
 {
-	return (!!zone->present_pages);
+	return (!!zone->managed_pages);
 }
 
 extern int movable_zone;

-- 
Mel Gorman
SUSE Labs