Commit a4c51bde13ee78405827287ef24f80b1b40d45d7
Committed by
Greg Kroah-Hartman
1 parent
5c0f0c017c
Exists in
ti-linux-3.14.y
and in
2 other branches
mm: exclude memoryless nodes from zone_reclaim
commit 70ef57e6c22c3323dce179b7d0d433c479266612 upstream. We had a report about strange OOM killer strikes on a PPC machine although there was a lot of swap free and a tons of anonymous memory which could be swapped out. In the end it turned out that the OOM was a side effect of zone reclaim which wasn't unmapping and swapping out and so the system was pushed to the OOM. Although this sounds like a bug somewhere in the kswapd vs. zone reclaim vs. direct reclaim interaction numactl on the said hardware suggests that the zone reclaim should not have been set in the first place: node 0 cpus: 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 node 0 size: 0 MB node 0 free: 0 MB node 2 cpus: node 2 size: 7168 MB node 2 free: 6019 MB node distances: node 0 2 0: 10 40 2: 40 10 So all the CPUs are associated with Node0 which doesn't have any memory while Node2 contains all the available memory. Node distances cause an automatic zone_reclaim_mode enabling. Zone reclaim is intended to keep the allocations local but this doesn't make any sense on the memoryless nodes. So let's exclude such nodes for init_zone_allows_reclaim which evaluates zone reclaim behavior and suitable reclaim_nodes. Signed-off-by: Michal Hocko <mhocko@suse.cz> Acked-by: David Rientjes <rientjes@google.com> Acked-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com> Tested-by: Nishanth Aravamudan <nacc@linux.vnet.ibm.com> Acked-by: Mel Gorman <mgorman@suse.de> Signed-off-by: Andrew Morton <akpm@linux-foundation.org> Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org> Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Showing 1 changed file with 3 additions and 2 deletions Side-by-side Diff
mm/page_alloc.c
... | ... | @@ -1869,7 +1869,7 @@ |
1869 | 1869 | { |
1870 | 1870 | int i; |
1871 | 1871 | |
1872 | - for_each_online_node(i) | |
1872 | + for_each_node_state(i, N_MEMORY) | |
1873 | 1873 | if (node_distance(nid, i) <= RECLAIM_DISTANCE) |
1874 | 1874 | node_set(i, NODE_DATA(nid)->reclaim_nodes); |
1875 | 1875 | else |
... | ... | @@ -4933,7 +4933,8 @@ |
4933 | 4933 | |
4934 | 4934 | pgdat->node_id = nid; |
4935 | 4935 | pgdat->node_start_pfn = node_start_pfn; |
4936 | - init_zone_allows_reclaim(nid); | |
4936 | + if (node_state(nid, N_MEMORY)) | |
4937 | + init_zone_allows_reclaim(nid); | |
4937 | 4938 | #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP |
4938 | 4939 | get_pfn_range_for_nid(nid, &start_pfn, &end_pfn); |
4939 | 4940 | #endif |