[PATCH 6/7 v3] mm/memcontrol: optimize stock usage for cgroup v2

[Joshua Hahn <joshua.hahnjy@gmail.com>]

In cgroup v2, it is unlikely for memcg charges to happen directly on
non-leaf cgroups. There are a few exceptions, such as processes that
have yet to be migrated to children, and tasks that are reparented on
memcg destruction, that charge to the parent.

Because it is rare for parent cgroups to receive direct charges, stock
that remains in them are wasted memory.

Drain parent stocks when the first child is created to return those
pages for other memcgs to use.

This optimization is not for cgroup v1, where tasks can be attached to
any cgroup in the hierarchy, meaning stock can be consumed & refilled
for non-leaf cgroups as well.

Signed-off-by: Joshua Hahn <joshua.hahnjy@gmail.com>
---
 mm/memcontrol.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)

diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index f20c9b829224a..64b82f1782720 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -4280,6 +4280,21 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
 	 */
 	xa_store(&mem_cgroup_private_ids, memcg->id.id, memcg, GFP_KERNEL);
 
+	/*
+	 * It is unlikely for non-leaf memcgs to get direct charges on v2.
+	 * Drain the parent's stock if we are the first child.
+	 */
+	if (cgroup_subsys_on_dfl(memory_cgrp_subsys)) {
+		struct mem_cgroup *parent = parent_mem_cgroup(memcg);
+		int cpu;
+
+		if (parent && !mem_cgroup_is_root(parent) &&
+			      !css_has_online_children(&parent->css)) {
+			for_each_online_cpu(cpu)
+				work_on_cpu(cpu, drain_stock_on_cpu, parent);
+		}
+	}
+
 	return 0;
 free_objcg:
 	for_each_node(nid) {
-- 
2.53.0-Meta