@@ -1036,7 +1036,7 @@ struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
root = root_mem_cgroup;
rcu_read_lock();
-
+restart:
if (reclaim) {
struct mem_cgroup_per_node *mz;
@@ -1063,14 +1063,6 @@ struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
for (;;) {
css = css_next_descendant_pre(css, &root->css);
if (!css) {
- /*
- * Reclaimers share the hierarchy walk, and a
- * new one might jump in right at the end of
- * the hierarchy - make sure they see at least
- * one group and restart from the beginning.
- */
- if (!prev)
- continue;
break;
}
@@ -1093,8 +1085,18 @@ struct mem_cgroup *mem_cgroup_iter(struct mem_cgroup *root,
*/
(void)cmpxchg(&iter->position, pos, memcg);
- if (!memcg)
+ if (!memcg) {
iter->generation++;
+
+ /*
+ * Reclaimers share the hierarchy walk, and a
+ * new one might jump in right at the end of
+ * the hierarchy - make sure they see at least
+ * one group and restart from the beginning.
+ */
+ if (!prev)
+ goto restart;
+ }
}
out_unlock:
The generation number in struct mem_cgroup_reclaim_iter should be incremented on every round-trip. Currently, it is possible for a concurrent reclaimer to jump in at the end of the hierarchy, causing a traversal restart (resetting the iteration position) without incrementing the generation number. By resetting the position without incrementing the generation, it's possible for another ongoing mem_cgroup_iter() thread to walk the tree twice. Move the traversal restart such that the generation number is incremented before the restart. Signed-off-by: Kinsey Ho <kinseyho@google.com> --- mm/memcontrol.c | 22 ++++++++++++---------- 1 file changed, 12 insertions(+), 10 deletions(-)