Re: [BUG] Make setup_zone_migrate_reserve() aware of overlappingnodes

[Dave Hansen]

On Wed, 2008-08-20 at 12:08 -0500, Adam Litke wrote:
> I have gotten to the root cause of the hugetlb badness I reported back
> on August 15th.  My system has the following memory topology (note the
> overlapping node):
> 
> 	ïNode 0 Memory: 0x8000000-0x44000000
> 	ïNode 1 Memory: 0x0-0x8000000 0x44000000-0x80000000
> 
> setup_zone_migrate_reserve() scans the address range 0x0-0x8000000
> looking for a pageblock to move onto the MIGRATE_RESERVE list.  Finding
> no candidates, it happily continues the scan into 0x8000000-0x44000000.
> When a pageblock is found, the pages are moved to the MIGRATE_RESERVE
> list on the wrong zone.  Oops.

This eventually gets down into move_freepages() via:

	->setup_zone_migrate_reserve()
	 ->move_freepages_block()
	  ->move_freepages()
right?

It looks like there have been bugs in this area before in
move_freepages().  Should there be a more stringent check in *there*?
Maybe a warning?
> ï
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -2512,6 +2512,10 @@ static void setup_zone_migrate_reserve(struct
> zone *zone)
>                                                         pageblock_order;
>  
>         for (pfn = start_pfn; pfn < end_pfn; pfn += pageblock_nr_pages) {
> +               /* Watch out for overlapping nodes */
> +               if (!early_pfn_in_nid(pfn, zone->node))
> +                       continue;

zone->node doesn't exist on !CONFIG_NUMA. :(

You probably want:

	if (!early_pfn_in_nid(pfn, zone_to_nid(zone)))
		continue;

Are you sure you need the "early_" variant here?  We're not using
early_pfn_valid() right below it.  I guess you could also use:

	if (!page_to_nid(page) != zone_to_nid(zone))
		continue;

-- Dave

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/