I think 3.2 and 3.4 need to backport 43d77867a4f3 ("slub: refactoring unfreeze_partials()")
On 3.4 kernel, we found there are tens of thousands of free task_struct slabs
in the partial list of node 0 when they should have been discarded, which is
essentially a huge memory leak.
Looking into the code, seems the cause is, we check n->nr_partial but the page
might be in n2 where n != n2, and then the page is added to n2->partial though
n2->partial > s->min_partial.
static void unfreeze_partials(struct kmem_cache *s)
{
...
while ((page = c->partial)) {
...
c->partial = page->next;
...
if (!new.inuse && (!n || n->nr_partial > s->min_partial))
m = M_FREE;
else {
struct kmem_cache_node *n2 = get_node(s,
page_to_nid(page));
m = M_PARTIAL;
if (n != n2) {
...
n = n2;
...
}
...
}
...
}
--
To unsubscribe from this list: send the line "unsubscribe stable" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html