memcg: Add static key for memcg priority

hulk inclusion
category: feature
bugzilla: https://gitee.com/openeuler/kernel/issues/I3ZN3O
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: NLiu Shixin <liushixin2@huawei.com>
Signed-off-by: NZheng Zengkai <zhengzengkai@huawei.com>

include/linux/memcontrol.h

@@ -342,9 +342,16 @@ struct mem_cgroup {
 };
 
 #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

@@ -2825,6 +2825,17 @@ 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		= SYSCTL_ZERO,
+		.extra2		= SYSCTL_ONE,
+	},
 #endif
 	{
 		.procname	= "lowmem_reserve_ratio",

mm/memcontrol.c

@@ -3957,10 +3957,16 @@ 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);
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return;
+
 	if (!parent)
 		return;
 
@@ -3971,6 +3977,9 @@ static void memcg_qos_init(struct mem_cgroup *memcg)
 static s64 memcg_qos_read(struct cgroup_subsys_state *css,
 				      struct cftype *cft)
 {
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return 0;
+
 	return mem_cgroup_from_css(css)->memcg_priority;
 }
 
@@ -3979,6 +3988,9 @@ static int memcg_qos_write(struct cgroup_subsys_state *css,
 {
 	struct mem_cgroup *memcg = mem_cgroup_from_css(css);
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return -EACCES;
+
 	if (val >= 0)
 		memcg->memcg_priority = 0;
 	else
@@ -4022,6 +4034,8 @@ 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)
@@ -4058,6 +4072,9 @@ void memcg_print_bad_task(void *arg, int ret)
 {
 	struct oom_control *oc = arg;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return;
+
 	if (!ret && oc->chosen) {
 		struct mem_cgroup *memcg;
 
@@ -4067,6 +4084,41 @@ 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;
+
+	rcu_read_lock();
+	css_for_each_descendant_pre(css, &root_mem_cgroup->css) {
+		iter = mem_cgroup_from_css(css);
+		iter->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

@@ -316,6 +316,8 @@ static bool oom_next_task(struct task_struct *task, struct oom_control *oc,
 	struct mem_cgroup *cur_memcg;
 	struct mem_cgroup *oc_memcg;
 
+	if (!static_branch_likely(&memcg_qos_stat_key))
+		return !points || points < oc->chosen_points;
 
 	if (!points)
 		return true;
@@ -341,10 +343,7 @@ 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