On Thu, 20 Feb 2014 10:51:44 +0100 Michal Hocko <[email protected]> wrote:
> 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 doesn't unmap and swapp 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
> 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 memory less nodes. So let's exclude such nodes
> for init_zone_allows_reclaim which evaluates zone reclaim behavior and
> suitable reclaim_nodes.
>
> ...
>
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -1855,7 +1855,7 @@ static void __paginginit init_zone_allows_reclaim(int
> nid)
> {
> int i;
>
> - for_each_online_node(i)
> + for_each_node_state(i, N_MEMORY)
> if (node_distance(nid, i) <= RECLAIM_DISTANCE)
> node_set(i, NODE_DATA(nid)->reclaim_nodes);
> else
> @@ -4901,7 +4901,8 @@ void __paginginit free_area_init_node(int nid, unsigned
> long *zones_size,
>
> pgdat->node_id = nid;
> pgdat->node_start_pfn = node_start_pfn;
> - init_zone_allows_reclaim(nid);
> + if (node_state(nid, N_MEMORY))
> + init_zone_allows_reclaim(nid);
> #ifdef CONFIG_HAVE_MEMBLOCK_NODE_MAP
> get_pfn_range_for_nid(nid, &start_pfn, &end_pfn);
> #endif
What happens if someone later hot-adds some memory to that node?
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
