Re: OOM-killer invoked but why ?

[Andrew Morton <akpm@linux-foundation.org>]

On Thu, 31 Jan 2008 13:53:05 +0100 Claude Frantz <r31dmaeu@pc0312b.rz.unibw-muenchen.de> wrote:

> Hello !
> 
> I'm faced to a problem where the OOM-killer is invoked but I cannot find
> the reason why. The machine is rather powerfull, the load is very moderate,
> the disk swap space is nearly unused. The only strange observation which
> appears to me is the slow but progressive decreasing of kbbuffers during
> many hours.
> 
> Can you help me to diagnose the problem and to find a good solution ?
> 
> ...
>
> Jan 28 03:50:49 toaster kernel: 177466 pages slab
> Jan 28 03:50:49 toaster kernel: 1915 pages pagetables
> Jan 28 03:50:49 toaster kernel: Out of memory: kill process 10859 (amavisd) score 36218 or a child
> Jan 28 03:50:49 toaster kernel: Killed process 19146 (amavisd)

slab.  Maybe you've been bitten by the quicklist leak.  If you're able to
patch your kernel then please try this fix:

commit 96990a4ae979df9e235d01097d6175759331e88c
Author: Christoph Lameter <clameter@sgi.com>
Date:   Mon Jan 14 00:55:14 2008 -0800

    quicklists: Only consider memory that can be used with GFP_KERNEL
    
    Quicklists calculates the size of the quicklists based on the number of
    free pages.  This must be the number of free pages that can be allocated
    with GFP_KERNEL.  node_page_state() includes the pages in ZONE_HIGHMEM and
    ZONE_MOVABLE which may lead the quicklists to become too large causing OOM.
    
    Signed-off-by: Christoph Lameter <clameter@sgi.com>
    Tested-by: Dhaval Giani <dhaval@linux.vnet.ibm.com>
    Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
    Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

diff --git a/mm/quicklist.c b/mm/quicklist.c
index ae8189c..3f703f7 100644
--- a/mm/quicklist.c
+++ b/mm/quicklist.c
@@ -26,9 +26,17 @@ DEFINE_PER_CPU(struct quicklist, quicklist)[CONFIG_NR_QUICK];
 static unsigned long max_pages(unsigned long min_pages)
 {
 	unsigned long node_free_pages, max;
+	struct zone *zones = NODE_DATA(numa_node_id())->node_zones;
+
+	node_free_pages =
+#ifdef CONFIG_ZONE_DMA
+		zone_page_state(&zones[ZONE_DMA], NR_FREE_PAGES) +
+#endif
+#ifdef CONFIG_ZONE_DMA32
+		zone_page_state(&zones[ZONE_DMA32], NR_FREE_PAGES) +
+#endif
+		zone_page_state(&zones[ZONE_NORMAL], NR_FREE_PAGES);
 
-	node_free_pages = node_page_state(numa_node_id(),
-			NR_FREE_PAGES);
 	max = node_free_pages / FRACTION_OF_NODE_MEM;
 	return max(max, min_pages);
 }


I note that this didn't have the stable@kernel.org cc.  Christoph, did we
deliberately decide not to backport?