Re: [Lhms-devel] Re: 2.6.14-mm2

[Dave Hansen <haveblue@us.ibm.com>]

On Wed, 2005-11-16 at 21:56 +0900, KAMEZAWA Hiroyuki wrote:
> Index: linux-2.6.14-mm2/mm/page_alloc.c
> ===================================================================
> --- linux-2.6.14-mm2.orig/mm/page_alloc.c
> +++ linux-2.6.14-mm2/mm/page_alloc.c
> @@ -2054,11 +2054,11 @@ static void __init free_area_init_core(s
>   		zone->nr_active = 0;
>   		zone->nr_inactive = 0;
>   		atomic_set(&zone->reclaim_in_progress, 0);
> +		init_currently_empty_zone(zone, zone_start_pfn, size);
>   		if (!size)
>   			continue;
> 
>   		zonetable_add(zone, nid, j, zone_start_pfn, size);
> -		init_currently_empty_zone(zone, zone_start_pfn, size);
>   		zone_start_pfn += size;
>   	}
>   }

Can you explain in a little bit more detail why this matters, and
exactly how it fixes your problem.  I'm not sure it's correct.

"init_currently_empty_zone" could more properly be called something like
"init currently empty zone to now have memory".  There's no reason to
call it, unless you have an empty zone *AND* you some memory to put in
it now.  If you call it with a size of 0, the things like memmap_init
inside of it don't make any sense.

Also, if you're doing hot-adds of _new_ zones at runtime, you need to do
something fancy with the zonelist locking that I never got around to
because nobody needs it yet.  See something along these lines:

http://www.sr71.net/patches/2.6.14/2.6.14-rc2-git8-mhp1/broken-out/E2-for-debugging-handle-add-to-empty-zone.patch

-- Dave