memcg: Add static key for memcg priority

hulk inclusion
category: feature
bugzilla: 51827
CVE: NA

--------------------------------------

This patch adds a default-false static key to disable memcg priority
feature. If you want to enable it by writing 1:

echo 1 > /proc/sys/vm/memcg_qos_enable
Signed-off-by: NJing Xiangfeng <jingxiangfeng@huawei.com>
Reviewed-by: NKefeng Wang <wangkefeng.wang@huawei.com>
Signed-off-by: NYang Yingliang <yangyingliang@huawei.com>

include/linux/memcontrol.h

@@ -331,9 +331,16 @@ struct mem_cgroup_extension {
 	container_of(cgroup, struct mem_cgroup_extension, memcg)
 
 #ifdef CONFIG_MEMCG_QOS
+#define ENABLE_MEMCG_QOS	1
+#define DISABLE_MEMCG_QOS	0
+extern int sysctl_memcg_qos_stat;
+DECLARE_STATIC_KEY_FALSE(memcg_qos_stat_key);
+
 bool memcg_low_priority_scan_tasks(int (*)(struct task_struct *, void *),
 				   void *);
 void memcg_print_bad_task(void *arg, int ret);
+extern int sysctl_memcg_qos_handler(struct ctl_table *table,
+		int write, void __user *buffer, size_t *length, loff_t *ppos);
 #endif
 
 /*

kernel/sysctl.c

@@ -1449,6 +1449,17 @@ static struct ctl_table vm_table[] = {
 		.mode		= 0644,
 		.proc_handler	= hugetlb_overcommit_handler,
 	},
+#endif
+#ifdef CONFIG_MEMCG_QOS
+	{
+		.procname	= "memcg_qos_enable",
+		.data		= &sysctl_memcg_qos_stat,
+		.maxlen		= sizeof(int),
+		.mode		= 0644,
+		.proc_handler	= sysctl_memcg_qos_handler,
+		.extra1		= &zero,
+		.extra2		= &one,
+	},
 #endif
 	{
 		.procname	= "lowmem_reserve_ratio",

mm/memcontrol.c

@@ -3404,12 +3404,18 @@ static int mem_cgroup_move_charge_write(struct cgroup_subsys_state *css,
 #endif
 
 #ifdef CONFIG_MEMCG_QOS
+int sysctl_memcg_qos_stat = DISABLE_MEMCG_QOS;
+DEFINE_STATIC_KEY_FALSE(memcg_qos_stat_key);
+
 static void memcg_qos_init(struct mem_cgroup *memcg)
 {
 	struct mem_cgroup *parent = parent_mem_cgroup(memcg);
 	struct mem_cgroup_extension *memcg_ext;
 	struct mem_cgroup_extension *parent_ext;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return;
+
 	if (!parent)
 		return;
 
@@ -3426,6 +3432,9 @@ static s64 memcg_qos_read(struct cgroup_subsys_state *css,
 	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
 	struct mem_cgroup_extension *memcg_ext;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return 0;
+
 	memcg_ext = to_memcg_ext(memcg);
 
 	return memcg_ext->memcg_priority;
@@ -3437,6 +3446,9 @@ static int memcg_qos_write(struct cgroup_subsys_state *css,
 	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
 	struct mem_cgroup_extension *memcg_ext;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return -EACCES;
+
 	memcg_ext = to_memcg_ext(memcg);
 
 	if (val >= 0)
@@ -3484,6 +3496,8 @@ bool memcg_low_priority_scan_tasks(int (*fn)(struct task_struct *, void *),
 	int ret = 0;
 	bool retry = true;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return false;
 retry:
 	max = memcg_find_max_usage(last);
 	if (!max)
@@ -3522,6 +3536,9 @@ void memcg_print_bad_task(void *arg, int ret)
 	struct mem_cgroup *memcg;
 	struct mem_cgroup_extension *memcg_ext;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return;
+
 	if (!ret && oc->chosen) {
 		memcg = mem_cgroup_from_task(oc->chosen);
 		memcg_ext = to_memcg_ext(memcg);
@@ -3530,6 +3547,45 @@ void memcg_print_bad_task(void *arg, int ret)
 				oc->chosen->pid, oc->chosen->comm);
 	}
 }
+
+static void memcg_qos_reset(void)
+{
+	struct mem_cgroup *iter;
+	struct cgroup_subsys_state *css;
+	struct mem_cgroup_extension *memcg_ext;
+
+	rcu_read_lock();
+	css_for_each_descendant_pre(css, &root_mem_cgroup->css) {
+		iter = mem_cgroup_from_css(css);
+		memcg_ext = to_memcg_ext(iter);
+
+		if (memcg_ext->memcg_priority)
+			memcg_ext->memcg_priority = 0;
+	}
+	rcu_read_unlock();
+}
+
+int sysctl_memcg_qos_handler(struct ctl_table *table, int write,
+		void __user *buffer, size_t *length, loff_t *ppos)
+{
+	int ret;
+
+	ret = proc_dointvec_minmax(table, write, buffer, length, ppos);
+	if (ret)
+		return ret;
+	if (write) {
+		if (sysctl_memcg_qos_stat == ENABLE_MEMCG_QOS) {
+			static_branch_enable(&memcg_qos_stat_key);
+			pr_info("enable memcg priority.\n");
+		} else {
+			static_branch_disable(&memcg_qos_stat_key);
+			memcg_qos_reset();
+			pr_info("disable memcg priority.\n");
+		}
+	}
+
+	return ret;
+}
 #endif
 
 #ifdef CONFIG_NUMA

mm/oom_kill.c

@@ -330,6 +330,8 @@ static bool oom_next_task(struct task_struct *task, struct oom_control *oc,
 	struct mem_cgroup_extension *cur_ext;
 	struct mem_cgroup_extension *oc_ext;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return !points || points < oc->chosen_points;
 
 	if (!points)
 		return true;
@@ -357,10 +359,7 @@ static bool oom_next_task(struct task_struct *task, struct oom_control *oc,
 static inline bool oom_next_task(struct task_struct *task,
 				struct oom_control *oc, unsigned long points)
 {
-	if (!points || points < oc->chosen_points)
-		return true;
-
-	return false;
+	return !points || points < oc->chosen_points;
 }
 #endif