Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
From:   Roman Gushchin <guroan@gmail.com>
To:     linux-mm@kvack.org, kernel-team@fb.com
Cc:     linux-kernel@vger.kernel.org, Tejun Heo <tj@kernel.org>,
        Rik van Riel <riel@surriel.com>,
        Johannes Weiner <hannes@cmpxchg.org>,
        Michal Hocko <mhocko@kernel.org>, Roman Gushchin <guro@fb.com>
Subject: [PATCH v3 4/6] mm: release per-node memcg percpu data prematurely
Date:   Wed, 13 Mar 2019 11:39:51 -0700
Message-Id: <20190313183953.17854-5-guro@fb.com>
In-Reply-To: <20190313183953.17854-1-guro@fb.com>
References: <20190313183953.17854-1-guro@fb.com>
MIME-Version: 1.0
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Similar to memcg-level statistics, per-node data isn't expected
to be hot after cgroup removal. Switching over to atomics and
prematurely releasing percpu data helps to reduce the memory
footprint of dying cgroups.

Signed-off-by: Roman Gushchin <guro@fb.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
---
 include/linux/memcontrol.h |  1 +
 mm/memcontrol.c            | 24 +++++++++++++++++++++++-
 2 files changed, 24 insertions(+), 1 deletion(-)

diff --git a/include/linux/memcontrol.h b/include/linux/memcontrol.h
index 569337514230..f296693d102b 100644
--- a/include/linux/memcontrol.h
+++ b/include/linux/memcontrol.h
@@ -127,6 +127,7 @@ struct mem_cgroup_per_node {
 	struct lruvec		lruvec;
 
 	struct lruvec_stat __rcu /* __percpu */ *lruvec_stat_cpu;
+	struct lruvec_stat __percpu *lruvec_stat_cpu_offlined;
 	atomic_long_t		lruvec_stat[NR_VM_NODE_STAT_ITEMS];
 
 	unsigned long		lru_zone_size[MAX_NR_ZONES][NR_LRU_LISTS];
diff --git a/mm/memcontrol.c b/mm/memcontrol.c
index efd5bc131a38..1b5fe826d6d0 100644
--- a/mm/memcontrol.c
+++ b/mm/memcontrol.c
@@ -4460,7 +4460,7 @@ static void free_mem_cgroup_per_node_info(struct mem_cgroup *memcg, int node)
 	if (!pn)
 		return;
 
-	free_percpu(pn->lruvec_stat_cpu);
+	WARN_ON_ONCE(pn->lruvec_stat_cpu != NULL);
 	kfree(pn);
 }
 
@@ -4616,7 +4616,17 @@ static int mem_cgroup_css_online(struct cgroup_subsys_state *css)
 static void percpu_rcu_free(struct rcu_head *rcu)
 {
 	struct mem_cgroup *memcg = container_of(rcu, struct mem_cgroup, rcu);
+	int node;
+
+	for_each_node(node) {
+		struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
 
+		if (!pn)
+			continue;
+
+		free_percpu(pn->lruvec_stat_cpu_offlined);
+		WARN_ON_ONCE(pn->lruvec_stat_cpu != NULL);
+	}
 	free_percpu(memcg->vmstats_percpu_offlined);
 	WARN_ON_ONCE(memcg->vmstats_percpu);
 
@@ -4625,6 +4635,18 @@ static void percpu_rcu_free(struct rcu_head *rcu)
 
 static void mem_cgroup_offline_percpu(struct mem_cgroup *memcg)
 {
+	int node;
+
+	for_each_node(node) {
+		struct mem_cgroup_per_node *pn = memcg->nodeinfo[node];
+
+		if (!pn)
+			continue;
+
+		pn->lruvec_stat_cpu_offlined = (struct lruvec_stat __percpu *)
+			rcu_dereference(pn->lruvec_stat_cpu);
+		rcu_assign_pointer(pn->lruvec_stat_cpu, NULL);
+	}
 	memcg->vmstats_percpu_offlined = (struct memcg_vmstats_percpu __percpu*)
 		rcu_dereference(memcg->vmstats_percpu);
 	rcu_assign_pointer(memcg->vmstats_percpu, NULL);
-- 
2.20.1