Re: [PATCH] mm: exclude memory less nodes from zone_reclaim

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

On Thu, 20 Feb 2014 10:51:44 +0100 Michal Hocko <mhocko@suse.cz> 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, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>