prevent kswapd from freeing excessive amounts of lowmem

The current VM can get itself into trouble fairly easily on systems with a small ZONE_HIGHMEM, which is common on i686 computers with 1GB of memory.

On one side, page_alloc() will allocate down to zone->pages_low, while on the other side, kswapd() and balance_pgdat() will try to free memory from every zone, until every zone has more free pages than zone->pages_high.

Highmem can be filled up to zone->pages_low with page tables, ramfs, vmalloc allocations and other unswappable things quite easily and without many bad side effects, since we still have a huge ZONE_NORMAL to do future allocations from.

However, as long as the number of free pages in the highmem zone is below zone->pages_high, kswapd will continue swapping things out from ZONE_NORMAL, too!

Sami Farin managed to get his system into a stage where kswapd had freed about 700MB of low memory and was still "going strong".

The attached patch will make kswapd stop paging out data from zones when there is more than enough memory free.

We do go above
zone->pages_high in order to keep pressure between zones equal in normal circumstances, but the patch should prevent the kind of excesses that made Sami's computer totally unusable.

Please merge this into -mm.

Signed-off-by Rik van Riel <[email protected]>

[linux-2.6-excessive-pageout.patch text/x-patch (715B)]

mm/vmscan.c renamed from mm/vmscan.c.excessive +0 -1

@@ -1371,7 +1371,13 @@ loop_again:
 			temp_priority[i] = priority;
 			sc.nr_scanned = 0;
 			note_zone_scanning_priority(zone, priority);
-			nr_reclaimed += shrink_zone(priority, zone, &sc);

I guess for a very small upper zone and a very large lower zone this could still put the scan balancing out of whack, fixable by a smarter version of "8*zone->pages_high" but it doesn't seem very likely that this will affect things much.

Why doesn't direct reclaim need similar treatment?

Andrew Morton wrote:

Quoted 5 lines

I guess for a very small upper zone and a very large lower zone this could still put the scan balancing out of whack, fixable by a smarter version of "8*zone->pages_high" but it doesn't seem very likely that this will affect things much.

Quoted 1 line

Why doesn't direct reclaim need similar treatment?

Because we only go into the direct reclaim path once every zone is at or below zone->pages_low, and the direct reclaim path will exit once we have freed more than swap_cluster_max pages.

Quoted 2 lines · On Thu, 06 Sep 2007 12:38:13 -0400 Rik van Riel <[email protected]> wrote:

Andrew Morton wrote:

(What happened to the other stuff I said?)

Quoted 5 lines

Quoted 5 lines

I guess for a very small upper zone and a very large lower zone this could still put the scan balancing out of whack, fixable by a smarter version of "8*zone->pages_high" but it doesn't seem very likely that this will affect things much.

Quoted 6 lines

Quoted 1 line

Why doesn't direct reclaim need similar treatment?

Because we only go into the direct reclaim path once every zone is at or below zone->pages_low, and the direct reclaim path will exit once we have freed more than swap_cluster_max pages.

hm. Now I need to remember why direct-reclaim does that :(

The current VM can get itself into trouble fairly easily on systems
with a small ZONE_HIGHMEM, which is common on i686 computers with
1GB of memory.

On one side, page_alloc() will allocate down to zone->pages_low,
while on the other side, kswapd() and balance_pgdat() will try
to free memory from every zone, until every zone has more free
pages than zone->pages_high.

Highmem can be filled up to zone->pages_low with page tables,
ramfs, vmalloc allocations and other unswappable things quite
easily and without many bad side effects, since we still have
a huge ZONE_NORMAL to do future allocations from.

However, as long as the number of free pages in the highmem
zone is below zone->pages_high, kswapd will continue swapping
things out from ZONE_NORMAL, too!

Sami Farin managed to get his system into a stage where kswapd
had freed about 700MB of low memory and was still "going strong".

The attached patch will make kswapd stop paging out data from
zones when there is more than enough memory free. We do go above
zone->pages_high in order to keep pressure between zones equal
in normal circumstances, but the patch should prevent the kind
of excesses that made Sami's computer totally unusable.

Please merge this into -mm.

Signed-off-by Rik van Riel <[email protected]>

Hi!

Quoted 30 lines

The current VM can get itself into trouble fairly easily on systems with a small ZONE_HIGHMEM, which is common on i686 computers with 1GB of memory.

On one side, page_alloc() will allocate down to zone->pages_low, while on the other side, kswapd() and balance_pgdat() will try to free memory from every zone, until every zone has more free pages than zone->pages_high.

Highmem can be filled up to zone->pages_low with page tables, ramfs, vmalloc allocations and other unswappable things quite easily and without many bad side effects, since we still have a huge ZONE_NORMAL to do future allocations from.

However, as long as the number of free pages in the highmem zone is below zone->pages_high, kswapd will continue swapping things out from ZONE_NORMAL, too!

Sami Farin managed to get his system into a stage where kswapd had freed about 700MB of low memory and was still "going strong".

The attached patch will make kswapd stop paging out data from zones when there is more than enough memory free. We do go above zone->pages_high in order to keep pressure between zones equal in normal circumstances, but the patch should prevent the kind of excesses that made Sami's computer totally unusable.

Please merge this into -mm.

Signed-off-by Rik van Riel <[email protected]>

mm/vmscan.c loop_again: · line 1377

Quoted 11 lines

mm/vmscan.c renamed from mm/vmscan.c.excessive +4 -1

@@ -1371,7 +1371,13 @@ loop_again:
 			temp_priority[i] = priority;
 			sc.nr_scanned = 0;
 			note_zone_scanning_priority(zone, priority);
-			nr_reclaimed += shrink_zone(priority, zone, &sc);
+			/*
+			 * We put equal pressure on every zone, unless one
+			 * zone has way too many pages free already.
+			 */

That does not seem right. Having empty HIGHMEM and full LOWMEM would be very bad, right? We may stop freeing when there's enough LOWMEM free, but not if there's only HIGHMEM free.