Re: [PATCH v2 4/9] mm: Scrub memory from idle loop

["Jan Beulich" <JBeulich@suse.com>]

>>> On 03.04.17 at 18:50, <boris.ostrovsky@oracle.com> wrote:
> --- a/xen/common/page_alloc.c
> +++ b/xen/common/page_alloc.c
> @@ -1043,16 +1043,44 @@ static struct page_info *merge_chunks(struct page_info *pg, unsigned int node,
>      return pg;
>  }
>  
> -static void scrub_free_pages(unsigned int node)
> +static nodemask_t node_scrubbing;
> +static unsigned int node_to_scrub(bool_t get_node)

Blank line between these two please.

> +{
> +    nodeid_t node = cpu_to_node(smp_processor_id()), local_node;
> +
> +    if ( node == NUMA_NO_NODE )
> +        node = 0;
> +    local_node = node;
> +
> +    /*
> +     * Check local node fist and then then see if there are any memory-only
> +     * nodes that may need scrubbing
> +     */
> +    while ( 1 )
> +    {
> +        if ( node_need_scrub[node] &&
> +             (!node_test_and_set(node, node_scrubbing) || !get_node) )
> +                return node;
> +        do {
> +            node = cycle_node(node, node_online_map);
> +            if ( node == local_node )
> +                return NUMA_NO_NODE;
> +        } while ( !cpumask_empty(&node_to_cpumask(node)) );

I think this needs logic to prefer closer nodes over more distant
ones, or else the latency of coming back out of the idle loop may
grow needlessly high.

> +    }
> +}
> +
> +bool_t scrub_free_pages()

Plain bool please and missing void.

>  {
>      struct page_info *pg;
>      unsigned int i, zone;
> -    int order;
> +    int order, cpu = smp_processor_id();

Both unsigned int.

Jan


_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
https://lists.xen.org/xen-devel