On Tue, 2011-09-27 at 06:52 +0800, Andrew Morton wrote:

On Mon, 26 Sep 2011 15:23:20 +0200
Michal Hocko [off-list ref] wrote:

quoted

On Mon 26-09-11 16:11:52, Shaohua Li wrote:

quoted

I saw DMA zone is always unreclaimable in my system. 
zone->all_unreclaimable isn't initialized till a page from the zone is
freed. This isn't a big problem normally, but a little confused, so
fix here.

The value is initialized when a node is allocated. setup_node_data uses
alloc_remap which memsets the whole structure or memblock allocation
which is initialized to 0 as well AFAIK and memory hotplug uses
arch_alloc_nodedata which is kzalloc.

setup_node_data() does memset(NODE_DATA(nid), 0, sizeof(pg_data_t)) just
to be sure.

However, Shaohua reports that "DMA zone is always unreclaimable in my system",
and presumably this patch fixed it.  So we don't know what's going on?



Presumably all the other "zone->foo = 0" assignments in free_area_init_core()
are unneeded.

Looks I didn't run my test correctly, sorry. I just check it, and this
is a vmscan bug, I'll work out a new patch.

--
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/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>