Re: [PATCH V2 2/2]: powerpc/hotplug/mm: Fix hot-add memory node assoc

[Balbir Singh <bsingharora@gmail.com>]

On Thu, 25 May 2017 12:37:40 -0500
Michael Bringmann <mwb@linux.vnet.ibm.com> wrote:

> Removing or adding memory via the PowerPC hotplug interface shows
> anomalies in the association between memory and nodes.  The code
> was updated to ensure that all nodes found at boot are still available
> to subsequent DLPAR hotplug-memory operations, even if they are not
> needed at boot time.
> 
> Signed-off-by: Michael Bringmann <mwb@linux.vnet.ibm.com>
> ---
> Changes in V2:
>   -- Simplify patches to ensure more nodes in possible map, removing
>      code from PowerPC numa.c that constrained possible map to size
>      of online map.
> ---
>  arch/powerpc/mm/numa.c |    7 -------
>  1 file changed, 7 deletions(-)
> 
> diff --git a/arch/powerpc/mm/numa.c b/arch/powerpc/mm/numa.c
> index 15c2dd5..18f3038 100644
> --- a/arch/powerpc/mm/numa.c
> +++ b/arch/powerpc/mm/numa.c
> @@ -907,13 +907,6 @@ void __init initmem_init(void)
>  
>  	memblock_dump_all();
>  
> -	/*
> -	 * Reduce the possible NUMA nodes to the online NUMA nodes,
> -	 * since we do not support node hotplug. This ensures that  we
> -	 * lower the maximum NUMA node ID to what is actually present.
> -	 */
> -	nodes_and(node_possible_map, node_possible_map, node_online_map);
> -

There is an overhead with turning this off if you have too many cgroups
with the memory controller. I think this fix was added for a pathological
test case. On my system I see 84 cgroups with 1 node, so the probable
overhead is 84*255*sizeof(struct mem_cgroup_tree_per_node).

I tried some patches to reduce the overhead, but those need more overhauling
and rework.

Balbir Singh.