Hi,
While I'm playing with memory controller of 2.6.23-rc4-mm1, I met following.
==
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# echo $$ > /opt/mem_control/group_1/tasks
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# cat
/opt/mem_control/group_1/memory.limit
32768
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# cat
/opt/mem_control/group_1/memory.usage
286
// Memory is limited to 512 GiB. try "dd" 1GiB (page size is 16KB)
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# dd if=/dev/zero of=/tmp/tmpfile bs=1024
count=1048576
Killed
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# ls
Killed
//above are caused by OOM.
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# cat
/opt/mem_control/group_1/memory.usage
32763
[EMAIL PROTECTED] test-2.6.23-rc4-mm1]# cat
/opt/mem_control/group_1/memory.limit
32768
// fully filled by page cache. no reclaim run.
==
The reason this happens is because I used kernelcore= boot option, i.e
ZONE_MOVABLE. Seems try_to_free_mem_container_pages() ignores ZONE_MOVABLE.
Quick fix is attached, but Mel's one-zonelist-pernode patch may change this.
I'll continue to watch.
Thanks,
-Kame
==
Now, there is ZONE_MOVABLE...
page cache and user pages are allocated from gfp_zone(GFP_HIGHUSER_MOVABLE)
Signed-off-by: KAMEZAWA Hiroyuki <[EMAIL PROTECTED]>
---
mm/vmscan.c | 9 ++-------
1 file changed, 2 insertions(+), 7 deletions(-)
Index: linux-2.6.23-rc4-mm1.bak/mm/vmscan.c
===================================================================
--- linux-2.6.23-rc4-mm1.bak.orig/mm/vmscan.c
+++ linux-2.6.23-rc4-mm1.bak/mm/vmscan.c
@@ -1351,12 +1351,6 @@ unsigned long try_to_free_pages(struct z
#ifdef CONFIG_CONTAINER_MEM_CONT
-#ifdef CONFIG_HIGHMEM
-#define ZONE_USERPAGES ZONE_HIGHMEM
-#else
-#define ZONE_USERPAGES ZONE_NORMAL
-#endif
-
unsigned long try_to_free_mem_container_pages(struct mem_container *mem_cont)
{
struct scan_control sc = {
@@ -1371,9 +1365,10 @@ unsigned long try_to_free_mem_container_
};
int node;
struct zone **zones;
+ int target_zone = gfp_zone(GFP_HIGHUSER_MOVABLE);
for_each_online_node(node) {
- zones = NODE_DATA(node)->node_zonelists[ZONE_USERPAGES].zones;
+ zones = NODE_DATA(node)->node_zonelists[target_zone].zones;
if (do_try_to_free_pages(zones, sc.gfp_mask, &sc))
return 1;
}
_______________________________________________
Containers mailing list
[EMAIL PROTECTED]
https://lists.linux-foundation.org/mailman/listinfo/containers
_______________________________________________
Devel mailing list
Devel@openvz.org
https://openvz.org/mailman/listinfo/devel
