Merge branch 'for-4.18' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup

Pull cgroup updates from Tejun Heo:

 - For cpustat, cgroup has a percpu hierarchical stat mechanism which
   propagates up the hierarchy lazily.

   This contains commits to factor out and generalize the mechanism so
   that it can be used for other cgroup stats too.

   The original intention was to update memcg stats to use it but memcg
   went for a different approach, so still the only user is cpustat. The
   factoring out and generalization still make sense and it's likely
   that this can be used for other purposes in the future.

 - cgroup uses kernfs_notify() (which uses fsnotify()) to inform user
   space of certain events. A rate limiting mechanism is added.

 - Other misc changes.

* 'for-4.18' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
  cgroup: css_set_lock should nest inside tasklist_lock
  rdmacg: Convert to use match_string() helper
  cgroup: Make cgroup_rstat_updated() ready for root cgroup usage
  cgroup: Add memory barriers to plug cgroup_rstat_updated() race window
  cgroup: Add cgroup_subsys->css_rstat_flush()
  cgroup: Replace cgroup_rstat_mutex with a spinlock
  cgroup: Factor out and expose cgroup_rstat_*() interface functions
  cgroup: Reorganize kernel/cgroup/rstat.c
  cgroup: Distinguish base resource stat implementation from rstat
  cgroup: Rename stat to rstat
  cgroup: Rename kernel/cgroup/stat.c to kernel/cgroup/rstat.c
  cgroup: Limit event generation frequency
  cgroup: Explicitly remove core interface files

include/linux/cgroup-defs.h

@@ -105,6 +105,8 @@ enum {
 struct cgroup_file {
 	/* do not access any fields from outside cgroup core */
 	struct kernfs_node *kn;
+	unsigned long notified_at;
+	struct timer_list notify_timer;
 };
 
 /*
@@ -128,6 +130,9 @@ struct cgroup_subsys_state {
 	struct list_head sibling;
 	struct list_head children;
 
+	/* flush target list anchored at cgrp->rstat_css_list */
+	struct list_head rstat_css_node;
+
 	/*
 	 * PI: Subsys-unique ID.  0 is unused and root is always 1.  The
 	 * matching css can be looked up using css_from_id().
@@ -256,12 +261,16 @@ struct css_set {
 	struct rcu_head rcu_head;
 };
 
+struct cgroup_base_stat {
+	struct task_cputime cputime;
+};
+
 /*
- * cgroup basic resource usage statistics.  Accounting is done per-cpu in
- * cgroup_cpu_stat which is then lazily propagated up the hierarchy on
- * reads.
+ * rstat - cgroup scalable recursive statistics.  Accounting is done
+ * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the
+ * hierarchy on reads.
  *
- * When a stat gets updated, the cgroup_cpu_stat and its ancestors are
+ * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are
  * linked into the updated tree.  On the following read, propagation only
  * considers and consumes the updated tree.  This makes reading O(the
  * number of descendants which have been active since last read) instead of
@@ -271,20 +280,24 @@ struct css_set {
  * aren't active and stat may be read frequently.  The combination can
  * become very expensive.  By propagating selectively, increasing reading
  * frequency decreases the cost of each read.
+ *
+ * This struct hosts both the fields which implement the above -
+ * updated_children and updated_next - and the fields which track basic
+ * resource statistics on top of it - bsync, bstat and last_bstat.
  */
-struct cgroup_cpu_stat {
+struct cgroup_rstat_cpu {
 	/*
-	 * ->sync protects all the current counters.  These are the only
-	 * fields which get updated in the hot path.
+	 * ->bsync protects ->bstat.  These are the only fields which get
+	 * updated in the hot path.
 	 */
-	struct u64_stats_sync sync;
-	struct task_cputime cputime;
+	struct u64_stats_sync bsync;
+	struct cgroup_base_stat bstat;
 
 	/*
 	 * Snapshots at the last reading.  These are used to calculate the
 	 * deltas to propagate to the global counters.
 	 */
-	struct task_cputime last_cputime;
+	struct cgroup_base_stat last_bstat;
 
 	/*
 	 * Child cgroups with stat updates on this cpu since the last read
@@ -295,18 +308,12 @@ struct cgroup_cpu_stat {
 	 * to the cgroup makes it unnecessary for each per-cpu struct to
 	 * point back to the associated cgroup.
 	 *
-	 * Protected by per-cpu cgroup_cpu_stat_lock.
+	 * Protected by per-cpu cgroup_rstat_cpu_lock.
 	 */
 	struct cgroup *updated_children;	/* terminated by self cgroup */
 	struct cgroup *updated_next;		/* NULL iff not on the list */
 };
 
-struct cgroup_stat {
-	/* per-cpu statistics are collected into the folowing global counters */
-	struct task_cputime cputime;
-	struct prev_cputime prev_cputime;
-};
-
 struct cgroup {
 	/* self css with NULL ->ss, points back to this cgroup */
 	struct cgroup_subsys_state self;
@@ -406,10 +413,14 @@ struct cgroup {
 	 */
 	struct cgroup *dom_cgrp;
 
+	/* per-cpu recursive resource statistics */
+	struct cgroup_rstat_cpu __percpu *rstat_cpu;
+	struct list_head rstat_css_list;
+
 	/* cgroup basic resource statistics */
-	struct cgroup_cpu_stat __percpu *cpu_stat;
-	struct cgroup_stat pending_stat;	/* pending from children */
-	struct cgroup_stat stat;
+	struct cgroup_base_stat pending_bstat;	/* pending from children */
+	struct cgroup_base_stat bstat;
+	struct prev_cputime prev_cputime;	/* for printing out cputime */
 
 	/*
 	 * list of pidlists, up to two for each namespace (one for procs, one
@@ -570,6 +581,7 @@ struct cgroup_subsys {
 	void (*css_released)(struct cgroup_subsys_state *css);
 	void (*css_free)(struct cgroup_subsys_state *css);
 	void (*css_reset)(struct cgroup_subsys_state *css);
+	void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu);
 	int (*css_extra_stat_show)(struct seq_file *seq,
 				   struct cgroup_subsys_state *css);
 

include/linux/cgroup.h

@@ -690,11 +690,19 @@ static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id,
 	char *buf, size_t buflen) {}
 #endif /* !CONFIG_CGROUPS */
 
+#ifdef CONFIG_CGROUPS
 /*
- * Basic resource stats.
+ * cgroup scalable recursive statistics.
  */
-#ifdef CONFIG_CGROUPS
+void cgroup_rstat_updated(struct cgroup *cgrp, int cpu);
+void cgroup_rstat_flush(struct cgroup *cgrp);
+void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp);
+void cgroup_rstat_flush_hold(struct cgroup *cgrp);
+void cgroup_rstat_flush_release(void);
 
+/*
+ * Basic resource stats.
+ */
 #ifdef CONFIG_CGROUP_CPUACCT
 void cpuacct_charge(struct task_struct *tsk, u64 cputime);
 void cpuacct_account_field(struct task_struct *tsk, int index, u64 val);

kernel/cgroup/Makefile

 # SPDX-License-Identifier: GPL-2.0
-obj-y := cgroup.o stat.o namespace.o cgroup-v1.o
+obj-y := cgroup.o rstat.o namespace.o cgroup-v1.o
 
 obj-$(CONFIG_CGROUP_FREEZER) += freezer.o
 obj-$(CONFIG_CGROUP_PIDS) += pids.o

kernel/cgroup/cgroup-internal.h

@@ -201,13 +201,12 @@ int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
 int cgroup_task_count(const struct cgroup *cgrp);
 
 /*
- * stat.c
+ * rstat.c
  */
-void cgroup_stat_flush(struct cgroup *cgrp);
-int cgroup_stat_init(struct cgroup *cgrp);
-void cgroup_stat_exit(struct cgroup *cgrp);
-void cgroup_stat_show_cputime(struct seq_file *seq);
-void cgroup_stat_boot(void);
+int cgroup_rstat_init(struct cgroup *cgrp);
+void cgroup_rstat_exit(struct cgroup *cgrp);
+void cgroup_rstat_boot(void);
+void cgroup_base_stat_cputime_show(struct seq_file *seq);
 
 /*
  * namespace.c

kernel/cgroup/cgroup.c

@@ -54,6 +54,7 @@
 #include <linux/proc_ns.h>
 #include <linux/nsproxy.h>
 #include <linux/file.h>
+#include <linux/sched/cputime.h>
 #include <net/sock.h>
 
 #define CREATE_TRACE_POINTS
@@ -61,6 +62,8 @@
 
 #define CGROUP_FILE_NAME_MAX		(MAX_CGROUP_TYPE_NAMELEN +	\
 					 MAX_CFTYPE_NAME + 2)
+/* let's not notify more than 100 times per second */
+#define CGROUP_FILE_NOTIFY_MIN_INTV	DIV_ROUND_UP(HZ, 100)
 
 /*
  * cgroup_mutex is the master lock.  Any modification to cgroup or its
@@ -142,14 +145,14 @@ static struct static_key_true *cgroup_subsys_on_dfl_key[] = {
 };
 #undef SUBSYS
 
-static DEFINE_PER_CPU(struct cgroup_cpu_stat, cgrp_dfl_root_cpu_stat);
+static DEFINE_PER_CPU(struct cgroup_rstat_cpu, cgrp_dfl_root_rstat_cpu);
 
 /*
  * The default hierarchy, reserved for the subsystems that are otherwise
  * unattached - it never has more than a single cgroup, and all tasks are
  * part of that cgroup.
  */
-struct cgroup_root cgrp_dfl_root = { .cgrp.cpu_stat = &cgrp_dfl_root_cpu_stat };
+struct cgroup_root cgrp_dfl_root = { .cgrp.rstat_cpu = &cgrp_dfl_root_rstat_cpu };
 EXPORT_SYMBOL_GPL(cgrp_dfl_root);
 
 /*
@@ -1554,6 +1557,8 @@ static void cgroup_rm_file(struct cgroup *cgrp, const struct cftype *cft)
 		spin_lock_irq(&cgroup_file_kn_lock);
 		cfile->kn = NULL;
 		spin_unlock_irq(&cgroup_file_kn_lock);
+
+		del_timer_sync(&cfile->notify_timer);
 	}
 
 	kernfs_remove_by_name(cgrp->kn, cgroup_file_name(cgrp, cft, name));
@@ -1573,8 +1578,17 @@ static void css_clear_dir(struct cgroup_subsys_state *css)
 
 	css->flags &= ~CSS_VISIBLE;
 
-	list_for_each_entry(cfts, &css->ss->cfts, node)
+	if (!css->ss) {
+		if (cgroup_on_dfl(cgrp))
+			cfts = cgroup_base_files;
+		else
+			cfts = cgroup1_base_files;
+
 		cgroup_addrm_files(css, cgrp, cfts, false);
+	} else {
+		list_for_each_entry(cfts, &css->ss->cfts, node)
+			cgroup_addrm_files(css, cgrp, cfts, false);
+	}
 }
 
 /**
@@ -1598,14 +1612,16 @@ static int css_populate_dir(struct cgroup_subsys_state *css)
 		else
 			cfts = cgroup1_base_files;
 
-		return cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
-	}
-
-	list_for_each_entry(cfts, &css->ss->cfts, node) {
-		ret = cgroup_addrm_files(css, cgrp, cfts, true);
-		if (ret < 0) {
-			failed_cfts = cfts;
-			goto err;
+		ret = cgroup_addrm_files(&cgrp->self, cgrp, cfts, true);
+		if (ret < 0)
+			return ret;
+	} else {
+		list_for_each_entry(cfts, &css->ss->cfts, node) {
+			ret = cgroup_addrm_files(css, cgrp, cfts, true);
+			if (ret < 0) {
+				failed_cfts = cfts;
+				goto err;
+			}
 		}
 	}
 
@@ -1782,13 +1798,6 @@ static void cgroup_enable_task_cg_lists(void)
 {
 	struct task_struct *p, *g;
 
-	spin_lock_irq(&css_set_lock);
-
-	if (use_task_css_set_links)
-		goto out_unlock;
-
-	use_task_css_set_links = true;
-
 	/*
 	 * We need tasklist_lock because RCU is not safe against
 	 * while_each_thread(). Besides, a forking task that has passed
@@ -1797,6 +1806,13 @@ static void cgroup_enable_task_cg_lists(void)
 	 * tasklist if we walk through it with RCU.
 	 */
 	read_lock(&tasklist_lock);
+	spin_lock_irq(&css_set_lock);
+
+	if (use_task_css_set_links)
+		goto out_unlock;
+
+	use_task_css_set_links = true;
+
 	do_each_thread(g, p) {
 		WARN_ON_ONCE(!list_empty(&p->cg_list) ||
 			     task_css_set(p) != &init_css_set);
@@ -1824,9 +1840,9 @@ static void cgroup_enable_task_cg_lists(void)
 		}
 		spin_unlock(&p->sighand->siglock);
 	} while_each_thread(g, p);
-	read_unlock(&tasklist_lock);
 out_unlock:
 	spin_unlock_irq(&css_set_lock);
+	read_unlock(&tasklist_lock);
 }
 
 static void init_cgroup_housekeeping(struct cgroup *cgrp)
@@ -1844,6 +1860,8 @@ static void init_cgroup_housekeeping(struct cgroup *cgrp)
 	cgrp->dom_cgrp = cgrp;
 	cgrp->max_descendants = INT_MAX;
 	cgrp->max_depth = INT_MAX;
+	INIT_LIST_HEAD(&cgrp->rstat_css_list);
+	prev_cputime_init(&cgrp->prev_cputime);
 
 	for_each_subsys(ss, ssid)
 		INIT_LIST_HEAD(&cgrp->e_csets[ssid]);
@@ -3381,7 +3399,7 @@ static int cpu_stat_show(struct seq_file *seq, void *v)
 	struct cgroup __maybe_unused *cgrp = seq_css(seq)->cgroup;
 	int ret = 0;
 
-	cgroup_stat_show_cputime(seq);
+	cgroup_base_stat_cputime_show(seq);
 #ifdef CONFIG_CGROUP_SCHED
 	ret = cgroup_extra_stat_show(seq, cgrp, cpu_cgrp_id);
 #endif
@@ -3521,6 +3539,12 @@ static int cgroup_kn_set_ugid(struct kernfs_node *kn)
 	return kernfs_setattr(kn, &iattr);
 }
 
+static void cgroup_file_notify_timer(struct timer_list *timer)
+{
+	cgroup_file_notify(container_of(timer, struct cgroup_file,
+					notify_timer));
+}
+
 static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
 			   struct cftype *cft)
 {
@@ -3547,6 +3571,8 @@ static int cgroup_add_file(struct cgroup_subsys_state *css, struct cgroup *cgrp,
 	if (cft->file_offset) {
 		struct cgroup_file *cfile = (void *)css + cft->file_offset;
 
+		timer_setup(&cfile->notify_timer, cgroup_file_notify_timer, 0);
+
 		spin_lock_irq(&cgroup_file_kn_lock);
 		cfile->kn = kn;
 		spin_unlock_irq(&cgroup_file_kn_lock);
@@ -3796,8 +3822,17 @@ void cgroup_file_notify(struct cgroup_file *cfile)
 	unsigned long flags;
 
 	spin_lock_irqsave(&cgroup_file_kn_lock, flags);
-	if (cfile->kn)
-		kernfs_notify(cfile->kn);
+	if (cfile->kn) {
+		unsigned long last = cfile->notified_at;
+		unsigned long next = last + CGROUP_FILE_NOTIFY_MIN_INTV;
+
+		if (time_in_range(jiffies, last, next)) {
+			timer_reduce(&cfile->notify_timer, next);
+		} else {
+			kernfs_notify(cfile->kn);
+			cfile->notified_at = jiffies;
+		}
+	}
 	spin_unlock_irqrestore(&cgroup_file_kn_lock, flags);
 }
 
@@ -4560,7 +4595,7 @@ static void css_free_rwork_fn(struct work_struct *work)
 			cgroup_put(cgroup_parent(cgrp));
 			kernfs_put(cgrp->kn);
 			if (cgroup_on_dfl(cgrp))
-				cgroup_stat_exit(cgrp);
+				cgroup_rstat_exit(cgrp);
 			kfree(cgrp);
 		} else {
 			/*
@@ -4587,6 +4622,11 @@ static void css_release_work_fn(struct work_struct *work)
 
 	if (ss) {
 		/* css release path */
+		if (!list_empty(&css->rstat_css_node)) {
+			cgroup_rstat_flush(cgrp);
+			list_del_rcu(&css->rstat_css_node);
+		}
+
 		cgroup_idr_replace(&ss->css_idr, NULL, css->id);
 		if (ss->css_released)
 			ss->css_released(css);
@@ -4597,7 +4637,7 @@ static void css_release_work_fn(struct work_struct *work)
 		trace_cgroup_release(cgrp);
 
 		if (cgroup_on_dfl(cgrp))
-			cgroup_stat_flush(cgrp);
+			cgroup_rstat_flush(cgrp);
 
 		for (tcgrp = cgroup_parent(cgrp); tcgrp;
 		     tcgrp = cgroup_parent(tcgrp))
@@ -4648,6 +4688,7 @@ static void init_and_link_css(struct cgroup_subsys_state *css,
 	css->id = -1;
 	INIT_LIST_HEAD(&css->sibling);
 	INIT_LIST_HEAD(&css->children);
+	INIT_LIST_HEAD(&css->rstat_css_node);
 	css->serial_nr = css_serial_nr_next++;
 	atomic_set(&css->online_cnt, 0);
 
@@ -4656,6 +4697,9 @@ static void init_and_link_css(struct cgroup_subsys_state *css,
 		css_get(css->parent);
 	}
 
+	if (cgroup_on_dfl(cgrp) && ss->css_rstat_flush)
+		list_add_rcu(&css->rstat_css_node, &cgrp->rstat_css_list);
+
 	BUG_ON(cgroup_css(cgrp, ss));
 }
 
@@ -4757,6 +4801,7 @@ static struct cgroup_subsys_state *css_create(struct cgroup *cgrp,
 err_list_del:
 	list_del_rcu(&css->sibling);
 err_free_css:
+	list_del_rcu(&css->rstat_css_node);
 	INIT_RCU_WORK(&css->destroy_rwork, css_free_rwork_fn);
 	queue_rcu_work(cgroup_destroy_wq, &css->destroy_rwork);
 	return ERR_PTR(err);
@@ -4785,7 +4830,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent)
 		goto out_free_cgrp;
 
 	if (cgroup_on_dfl(parent)) {
-		ret = cgroup_stat_init(cgrp);
+		ret = cgroup_rstat_init(cgrp);
 		if (ret)
 			goto out_cancel_ref;
 	}
@@ -4850,7 +4895,7 @@ static struct cgroup *cgroup_create(struct cgroup *parent)
 	cgroup_idr_remove(&root->cgroup_idr, cgrp->id);
 out_stat_exit:
 	if (cgroup_on_dfl(parent))
-		cgroup_stat_exit(cgrp);
+		cgroup_rstat_exit(cgrp);
 out_cancel_ref:
 	percpu_ref_exit(&cgrp->self.refcnt);
 out_free_cgrp:
@@ -5090,10 +5135,8 @@ static int cgroup_destroy_locked(struct cgroup *cgrp)
 	for_each_css(css, ssid, cgrp)
 		kill_css(css);
 
-	/*
-	 * Remove @cgrp directory along with the base files.  @cgrp has an
-	 * extra ref on its kn.
-	 */
+	/* clear and remove @cgrp dir, @cgrp has an extra ref on its kn */
+	css_clear_dir(&cgrp->self);
 	kernfs_remove(cgrp->kn);
 
 	if (parent && cgroup_is_threaded(cgrp))
@@ -5245,7 +5288,7 @@ int __init cgroup_init(void)
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup_base_files));
 	BUG_ON(cgroup_init_cftypes(NULL, cgroup1_base_files));
 
-	cgroup_stat_boot();
+	cgroup_rstat_boot();
 
 	/*
 	 * The latency of the synchronize_sched() is too high for cgroups,

kernel/cgroup/rdma.c

@@ -362,35 +362,32 @@ EXPORT_SYMBOL(rdmacg_unregister_device);
 static int parse_resource(char *c, int *intval)
 {
 	substring_t argstr;
-	const char **table = &rdmacg_resource_names[0];
 	char *name, *value = c;
 	size_t len;
-	int ret, i = 0;
+	int ret, i;
 
 	name = strsep(&value, "=");
 	if (!name || !value)
 		return -EINVAL;
 
-	len = strlen(value);
+	i = match_string(rdmacg_resource_names, RDMACG_RESOURCE_MAX, name);
+	if (i < 0)
+		return i;
 
-	for (i = 0; i < RDMACG_RESOURCE_MAX; i++) {
-		if (strcmp(table[i], name))
-			continue;
+	len = strlen(value);
 
-		argstr.from = value;
-		argstr.to = value + len;
+	argstr.from = value;
+	argstr.to = value + len;
 
-		ret = match_int(&argstr, intval);
-		if (ret >= 0) {
-			if (*intval < 0)
-				break;
-			return i;
-		}
-		if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
-			*intval = S32_MAX;
-			return i;
-		}
-		break;
+	ret = match_int(&argstr, intval);
+	if (ret >= 0) {
+		if (*intval < 0)
+			return -EINVAL;
+		return i;
+	}
+	if (strncmp(value, RDMACG_MAX_STR, len) == 0) {
+		*intval = S32_MAX;
+		return i;
 	}
 	return -EINVAL;
 }

kernel/cgroup/stat.c → kernel/cgroup/rstat.c

@@ -2,38 +2,47 @@
 
 #include <linux/sched/cputime.h>
 
-static DEFINE_MUTEX(cgroup_stat_mutex);
-static DEFINE_PER_CPU(raw_spinlock_t, cgroup_cpu_stat_lock);
+static DEFINE_SPINLOCK(cgroup_rstat_lock);
+static DEFINE_PER_CPU(raw_spinlock_t, cgroup_rstat_cpu_lock);
 
-static struct cgroup_cpu_stat *cgroup_cpu_stat(struct cgroup *cgrp, int cpu)
+static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu);
+
+static struct cgroup_rstat_cpu *cgroup_rstat_cpu(struct cgroup *cgrp, int cpu)
 {
-	return per_cpu_ptr(cgrp->cpu_stat, cpu);
+	return per_cpu_ptr(cgrp->rstat_cpu, cpu);
 }
 
 /**
- * cgroup_cpu_stat_updated - keep track of updated cpu_stat
+ * cgroup_rstat_updated - keep track of updated rstat_cpu
  * @cgrp: target cgroup
- * @cpu: cpu on which cpu_stat was updated
+ * @cpu: cpu on which rstat_cpu was updated
  *
- * @cgrp's cpu_stat on @cpu was updated.  Put it on the parent's matching
- * cpu_stat->updated_children list.  See the comment on top of
- * cgroup_cpu_stat definition for details.
+ * @cgrp's rstat_cpu on @cpu was updated.  Put it on the parent's matching
+ * rstat_cpu->updated_children list.  See the comment on top of
+ * cgroup_rstat_cpu definition for details.
  */
-static void cgroup_cpu_stat_updated(struct cgroup *cgrp, int cpu)
+void cgroup_rstat_updated(struct cgroup *cgrp, int cpu)
 {
-	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_cpu_stat_lock, cpu);
+	raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu);
 	struct cgroup *parent;
 	unsigned long flags;
 
+	/* nothing to do for root */
+	if (!cgroup_parent(cgrp))
+		return;
+
+	/*
+	 * Paired with the one in cgroup_rstat_cpu_pop_upated().  Either we
+	 * see NULL updated_next or they see our updated stat.
+	 */
+	smp_mb();
+
 	/*
-	 * Speculative already-on-list test.  This may race leading to
-	 * temporary inaccuracies, which is fine.
-	 *
 	 * Because @parent's updated_children is terminated with @parent
 	 * instead of NULL, we can tell whether @cgrp is on the list by
 	 * testing the next pointer for NULL.
 	 */
-	if (cgroup_cpu_stat(cgrp, cpu)->updated_next)
+	if (cgroup_rstat_cpu(cgrp, cpu)->updated_next)
 		return;
 
 	raw_spin_lock_irqsave(cpu_lock, flags);
@@ -41,42 +50,43 @@ static void cgroup_cpu_stat_updated(struct cgroup *cgrp, int cpu)
 	/* put @cgrp and all ancestors on the corresponding updated lists */
 	for (parent = cgroup_parent(cgrp); parent;
 	     cgrp = parent, parent = cgroup_parent(cgrp)) {
-		struct cgroup_cpu_stat *cstat = cgroup_cpu_stat(cgrp, cpu);
-		struct cgroup_cpu_stat *pcstat = cgroup_cpu_stat(parent, cpu);
+		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
+		struct cgroup_rstat_cpu *prstatc = cgroup_rstat_cpu(parent, cpu);
 
 		/*
 		 * Both additions and removals are bottom-up.  If a cgroup
 		 * is already in the tree, all ancestors are.
 		 */
-		if (cstat->updated_next)
+		if (rstatc->updated_next)
 			break;
 
-		cstat->updated_next = pcstat->updated_children;
-		pcstat->updated_children = cgrp;
+		rstatc->updated_next = prstatc->updated_children;
+		prstatc->updated_children = cgrp;
 	}
 
 	raw_spin_unlock_irqrestore(cpu_lock, flags);
 }
+EXPORT_SYMBOL_GPL(cgroup_rstat_updated);
 
 /**
- * cgroup_cpu_stat_pop_updated - iterate and dismantle cpu_stat updated tree
+ * cgroup_rstat_cpu_pop_updated - iterate and dismantle rstat_cpu updated tree
  * @pos: current position
  * @root: root of the tree to traversal
  * @cpu: target cpu
  *
- * Walks the udpated cpu_stat tree on @cpu from @root.  %NULL @pos starts
+ * Walks the udpated rstat_cpu tree on @cpu from @root.  %NULL @pos starts
  * the traversal and %NULL return indicates the end.  During traversal,
  * each returned cgroup is unlinked from the tree.  Must be called with the
- * matching cgroup_cpu_stat_lock held.
+ * matching cgroup_rstat_cpu_lock held.
  *
  * The only ordering guarantee is that, for a parent and a child pair
  * covered by a given traversal, if a child is visited, its parent is
  * guaranteed to be visited afterwards.
  */
-static struct cgroup *cgroup_cpu_stat_pop_updated(struct cgroup *pos,
-						  struct cgroup *root, int cpu)
+static struct cgroup *cgroup_rstat_cpu_pop_updated(struct cgroup *pos,
+						   struct cgroup *root, int cpu)
 {
-	struct cgroup_cpu_stat *cstat;
+	struct cgroup_rstat_cpu *rstatc;
 	struct cgroup *parent;
 
 	if (pos == root)
@@ -93,10 +103,10 @@ static struct cgroup *cgroup_cpu_stat_pop_updated(struct cgroup *pos,
 
 	/* walk down to the first leaf */
 	while (true) {
-		cstat = cgroup_cpu_stat(pos, cpu);
-		if (cstat->updated_children == pos)
+		rstatc = cgroup_rstat_cpu(pos, cpu);
+		if (rstatc->updated_children == pos)
 			break;
-		pos = cstat->updated_children;
+		pos = rstatc->updated_children;
 	}
 
 	/*
@@ -106,157 +116,283 @@ static struct cgroup *cgroup_cpu_stat_pop_updated(struct cgroup *pos,
 	 * child in most cases. The only exception is @root.
 	 */
 	parent = cgroup_parent(pos);
-	if (parent && cstat->updated_next) {
-		struct cgroup_cpu_stat *pcstat = cgroup_cpu_stat(parent, cpu);
-		struct cgroup_cpu_stat *ncstat;
+	if (parent && rstatc->updated_next) {
+		struct cgroup_rstat_cpu *prstatc = cgroup_rstat_cpu(parent, cpu);
+		struct cgroup_rstat_cpu *nrstatc;
 		struct cgroup **nextp;
 
-		nextp = &pcstat->updated_children;
+		nextp = &prstatc->updated_children;
 		while (true) {
-			ncstat = cgroup_cpu_stat(*nextp, cpu);
+			nrstatc = cgroup_rstat_cpu(*nextp, cpu);
 			if (*nextp == pos)
 				break;
 
 			WARN_ON_ONCE(*nextp == parent);
-			nextp = &ncstat->updated_next;
+			nextp = &nrstatc->updated_next;
 		}
 
-		*nextp = cstat->updated_next;
-		cstat->updated_next = NULL;
+		*nextp = rstatc->updated_next;
+		rstatc->updated_next = NULL;
+
+		/*
+		 * Paired with the one in cgroup_rstat_cpu_updated().
+		 * Either they see NULL updated_next or we see their
+		 * updated stat.
+		 */
+		smp_mb();
 	}
 
 	return pos;
 }
 
-static void cgroup_stat_accumulate(struct cgroup_stat *dst_stat,
-				   struct cgroup_stat *src_stat)
+/* see cgroup_rstat_flush() */
+static void cgroup_rstat_flush_locked(struct cgroup *cgrp, bool may_sleep)
+	__releases(&cgroup_rstat_lock) __acquires(&cgroup_rstat_lock)
 {
-	dst_stat->cputime.utime += src_stat->cputime.utime;
-	dst_stat->cputime.stime += src_stat->cputime.stime;
-	dst_stat->cputime.sum_exec_runtime += src_stat->cputime.sum_exec_runtime;
+	int cpu;
+
+	lockdep_assert_held(&cgroup_rstat_lock);
+
+	for_each_possible_cpu(cpu) {
+		raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_rstat_cpu_lock,
+						       cpu);
+		struct cgroup *pos = NULL;
+
+		raw_spin_lock(cpu_lock);
+		while ((pos = cgroup_rstat_cpu_pop_updated(pos, cgrp, cpu))) {
+			struct cgroup_subsys_state *css;
+
+			cgroup_base_stat_flush(pos, cpu);
+
+			rcu_read_lock();
+			list_for_each_entry_rcu(css, &pos->rstat_css_list,
+						rstat_css_node)
+				css->ss->css_rstat_flush(css, cpu);
+			rcu_read_unlock();
+		}
+		raw_spin_unlock(cpu_lock);
+
+		/* if @may_sleep, play nice and yield if necessary */
+		if (may_sleep && (need_resched() ||
+				  spin_needbreak(&cgroup_rstat_lock))) {
+			spin_unlock_irq(&cgroup_rstat_lock);
+			if (!cond_resched())
+				cpu_relax();
+			spin_lock_irq(&cgroup_rstat_lock);
+		}
+	}
 }
 
-static void cgroup_cpu_stat_flush_one(struct cgroup *cgrp, int cpu)
+/**
+ * cgroup_rstat_flush - flush stats in @cgrp's subtree
+ * @cgrp: target cgroup
+ *
+ * Collect all per-cpu stats in @cgrp's subtree into the global counters
+ * and propagate them upwards.  After this function returns, all cgroups in
+ * the subtree have up-to-date ->stat.
+ *
+ * This also gets all cgroups in the subtree including @cgrp off the
+ * ->updated_children lists.
+ *
+ * This function may block.
+ */
+void cgroup_rstat_flush(struct cgroup *cgrp)
 {
-	struct cgroup *parent = cgroup_parent(cgrp);
-	struct cgroup_cpu_stat *cstat = cgroup_cpu_stat(cgrp, cpu);
-	struct task_cputime *last_cputime = &cstat->last_cputime;
-	struct task_cputime cputime;
-	struct cgroup_stat delta;
-	unsigned seq;
+	might_sleep();
 
-	lockdep_assert_held(&cgroup_stat_mutex);
+	spin_lock_irq(&cgroup_rstat_lock);
+	cgroup_rstat_flush_locked(cgrp, true);
+	spin_unlock_irq(&cgroup_rstat_lock);
+}
 
-	/* fetch the current per-cpu values */
-	do {
-		seq = __u64_stats_fetch_begin(&cstat->sync);
-		cputime = cstat->cputime;
-	} while (__u64_stats_fetch_retry(&cstat->sync, seq));
+/**
+ * cgroup_rstat_flush_irqsafe - irqsafe version of cgroup_rstat_flush()
+ * @cgrp: target cgroup
+ *
+ * This function can be called from any context.
+ */
+void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp)
+{
+	unsigned long flags;
 
-	/* accumulate the deltas to propgate */
-	delta.cputime.utime = cputime.utime - last_cputime->utime;
-	delta.cputime.stime = cputime.stime - last_cputime->stime;
-	delta.cputime.sum_exec_runtime = cputime.sum_exec_runtime -
-					 last_cputime->sum_exec_runtime;
-	*last_cputime = cputime;
+	spin_lock_irqsave(&cgroup_rstat_lock, flags);
+	cgroup_rstat_flush_locked(cgrp, false);
+	spin_unlock_irqrestore(&cgroup_rstat_lock, flags);
+}
 
-	/* transfer the pending stat into delta */
-	cgroup_stat_accumulate(&delta, &cgrp->pending_stat);
-	memset(&cgrp->pending_stat, 0, sizeof(cgrp->pending_stat));
+/**
+ * cgroup_rstat_flush_begin - flush stats in @cgrp's subtree and hold
+ * @cgrp: target cgroup
+ *
+ * Flush stats in @cgrp's subtree and prevent further flushes.  Must be
+ * paired with cgroup_rstat_flush_release().
+ *
+ * This function may block.
+ */
+void cgroup_rstat_flush_hold(struct cgroup *cgrp)
+	__acquires(&cgroup_rstat_lock)
+{
+	might_sleep();
+	spin_lock_irq(&cgroup_rstat_lock);
+	cgroup_rstat_flush_locked(cgrp, true);
+}
 
-	/* propagate delta into the global stat and the parent's pending */
-	cgroup_stat_accumulate(&cgrp->stat, &delta);
-	if (parent)
-		cgroup_stat_accumulate(&parent->pending_stat, &delta);
+/**
+ * cgroup_rstat_flush_release - release cgroup_rstat_flush_hold()
+ */
+void cgroup_rstat_flush_release(void)
+	__releases(&cgroup_rstat_lock)
+{
+	spin_unlock_irq(&cgroup_rstat_lock);
 }
 
-/* see cgroup_stat_flush() */
-static void cgroup_stat_flush_locked(struct cgroup *cgrp)
+int cgroup_rstat_init(struct cgroup *cgrp)
 {
 	int cpu;
 
-	lockdep_assert_held(&cgroup_stat_mutex);
+	/* the root cgrp has rstat_cpu preallocated */
+	if (!cgrp->rstat_cpu) {
+		cgrp->rstat_cpu = alloc_percpu(struct cgroup_rstat_cpu);
+		if (!cgrp->rstat_cpu)
+			return -ENOMEM;
+	}
 
+	/* ->updated_children list is self terminated */
 	for_each_possible_cpu(cpu) {
-		raw_spinlock_t *cpu_lock = per_cpu_ptr(&cgroup_cpu_stat_lock, cpu);
-		struct cgroup *pos = NULL;
+		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
 
-		raw_spin_lock_irq(cpu_lock);
-		while ((pos = cgroup_cpu_stat_pop_updated(pos, cgrp, cpu)))
-			cgroup_cpu_stat_flush_one(pos, cpu);
-		raw_spin_unlock_irq(cpu_lock);
+		rstatc->updated_children = cgrp;
+		u64_stats_init(&rstatc->bsync);
 	}
+
+	return 0;
 }
 
-/**
- * cgroup_stat_flush - flush stats in @cgrp's subtree
- * @cgrp: target cgroup
- *
- * Collect all per-cpu stats in @cgrp's subtree into the global counters
- * and propagate them upwards.  After this function returns, all cgroups in
- * the subtree have up-to-date ->stat.
- *
- * This also gets all cgroups in the subtree including @cgrp off the
- * ->updated_children lists.
+void cgroup_rstat_exit(struct cgroup *cgrp)
+{
+	int cpu;
+
+	cgroup_rstat_flush(cgrp);
+
+	/* sanity check */
+	for_each_possible_cpu(cpu) {
+		struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
+
+		if (WARN_ON_ONCE(rstatc->updated_children != cgrp) ||
+		    WARN_ON_ONCE(rstatc->updated_next))
+			return;
+	}
+
+	free_percpu(cgrp->rstat_cpu);
+	cgrp->rstat_cpu = NULL;
+}
+
+void __init cgroup_rstat_boot(void)
+{
+	int cpu;
+
+	for_each_possible_cpu(cpu)
+		raw_spin_lock_init(per_cpu_ptr(&cgroup_rstat_cpu_lock, cpu));
+
+	BUG_ON(cgroup_rstat_init(&cgrp_dfl_root.cgrp));
+}
+
+/*
+ * Functions for cgroup basic resource statistics implemented on top of
+ * rstat.
  */
-void cgroup_stat_flush(struct cgroup *cgrp)
+static void cgroup_base_stat_accumulate(struct cgroup_base_stat *dst_bstat,
+					struct cgroup_base_stat *src_bstat)
 {
-	mutex_lock(&cgroup_stat_mutex);
-	cgroup_stat_flush_locked(cgrp);
-	mutex_unlock(&cgroup_stat_mutex);
+	dst_bstat->cputime.utime += src_bstat->cputime.utime;
+	dst_bstat->cputime.stime += src_bstat->cputime.stime;
+	dst_bstat->cputime.sum_exec_runtime += src_bstat->cputime.sum_exec_runtime;
 }
 
-static struct cgroup_cpu_stat *cgroup_cpu_stat_account_begin(struct cgroup *cgrp)
+static void cgroup_base_stat_flush(struct cgroup *cgrp, int cpu)
 {
-	struct cgroup_cpu_stat *cstat;
+	struct cgroup *parent = cgroup_parent(cgrp);
+	struct cgroup_rstat_cpu *rstatc = cgroup_rstat_cpu(cgrp, cpu);
+	struct task_cputime *last_cputime = &rstatc->last_bstat.cputime;
+	struct task_cputime cputime;
+	struct cgroup_base_stat delta;
+	unsigned seq;
+
+	/* fetch the current per-cpu values */
+	do {
+		seq = __u64_stats_fetch_begin(&rstatc->bsync);
+		cputime = rstatc->bstat.cputime;
+	} while (__u64_stats_fetch_retry(&rstatc->bsync, seq));
 
-	cstat = get_cpu_ptr(cgrp->cpu_stat);
-	u64_stats_update_begin(&cstat->sync);
-	return cstat;
+	/* calculate the delta to propgate */
+	delta.cputime.utime = cputime.utime - last_cputime->utime;
+	delta.cputime.stime = cputime.stime - last_cputime->stime;
+	delta.cputime.sum_exec_runtime = cputime.sum_exec_runtime -
+					 last_cputime->sum_exec_runtime;
+	*last_cputime = cputime;
+
+	/* transfer the pending stat into delta */
+	cgroup_base_stat_accumulate(&delta, &cgrp->pending_bstat);
+	memset(&cgrp->pending_bstat, 0, sizeof(cgrp->pending_bstat));
+
+	/* propagate delta into the global stat and the parent's pending */
+	cgroup_base_stat_accumulate(&cgrp->bstat, &delta);
+	if (parent)
+		cgroup_base_stat_accumulate(&parent->pending_bstat, &delta);
 }
 
-static void cgroup_cpu_stat_account_end(struct cgroup *cgrp,
-					struct cgroup_cpu_stat *cstat)
+static struct cgroup_rstat_cpu *
+cgroup_base_stat_cputime_account_begin(struct cgroup *cgrp)
 {
-	u64_stats_update_end(&cstat->sync);
-	cgroup_cpu_stat_updated(cgrp, smp_processor_id());
-	put_cpu_ptr(cstat);
+	struct cgroup_rstat_cpu *rstatc;
+
+	rstatc = get_cpu_ptr(cgrp->rstat_cpu);
+	u64_stats_update_begin(&rstatc->bsync);
+	return rstatc;
+}
+
+static void cgroup_base_stat_cputime_account_end(struct cgroup *cgrp,
+						 struct cgroup_rstat_cpu *rstatc)
+{
+	u64_stats_update_end(&rstatc->bsync);
+	cgroup_rstat_updated(cgrp, smp_processor_id());
+	put_cpu_ptr(rstatc);
 }
 
 void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec)
 {
-	struct cgroup_cpu_stat *cstat;
+	struct cgroup_rstat_cpu *rstatc;
 
-	cstat = cgroup_cpu_stat_account_begin(cgrp);
-	cstat->cputime.sum_exec_runtime += delta_exec;
-	cgroup_cpu_stat_account_end(cgrp, cstat);
+	rstatc = cgroup_base_stat_cputime_account_begin(cgrp);
+	rstatc->bstat.cputime.sum_exec_runtime += delta_exec;
+	cgroup_base_stat_cputime_account_end(cgrp, rstatc);
 }
 
 void __cgroup_account_cputime_field(struct cgroup *cgrp,
 				    enum cpu_usage_stat index, u64 delta_exec)
 {
-	struct cgroup_cpu_stat *cstat;
+	struct cgroup_rstat_cpu *rstatc;
 
-	cstat = cgroup_cpu_stat_account_begin(cgrp);
+	rstatc = cgroup_base_stat_cputime_account_begin(cgrp);
 
 	switch (index) {
 	case CPUTIME_USER:
 	case CPUTIME_NICE:
-		cstat->cputime.utime += delta_exec;
+		rstatc->bstat.cputime.utime += delta_exec;
 		break;
 	case CPUTIME_SYSTEM:
 	case CPUTIME_IRQ:
 	case CPUTIME_SOFTIRQ:
-		cstat->cputime.stime += delta_exec;
+		rstatc->bstat.cputime.stime += delta_exec;
 		break;
 	default:
 		break;
 	}
 
-	cgroup_cpu_stat_account_end(cgrp, cstat);
+	cgroup_base_stat_cputime_account_end(cgrp, rstatc);
 }
 
-void cgroup_stat_show_cputime(struct seq_file *seq)
+void cgroup_base_stat_cputime_show(struct seq_file *seq)
 {
 	struct cgroup *cgrp = seq_css(seq)->cgroup;
 	u64 usage, utime, stime;
@@ -264,15 +400,10 @@ void cgroup_stat_show_cputime(struct seq_file *seq)
 	if (!cgroup_parent(cgrp))
 		return;
 
-	mutex_lock(&cgroup_stat_mutex);
-
-	cgroup_stat_flush_locked(cgrp);
-
-	usage = cgrp->stat.cputime.sum_exec_runtime;
-	cputime_adjust(&cgrp->stat.cputime, &cgrp->stat.prev_cputime,
-		       &utime, &stime);
-
-	mutex_unlock(&cgroup_stat_mutex);
+	cgroup_rstat_flush_hold(cgrp);
+	usage = cgrp->bstat.cputime.sum_exec_runtime;
+	cputime_adjust(&cgrp->bstat.cputime, &cgrp->prev_cputime, &utime, &stime);
+	cgroup_rstat_flush_release();
 
 	do_div(usage, NSEC_PER_USEC);
 	do_div(utime, NSEC_PER_USEC);
@@ -283,56 +414,3 @@ void cgroup_stat_show_cputime(struct seq_file *seq)
 		   "system_usec %llu\n",
 		   usage, utime, stime);
 }
-
-int cgroup_stat_init(struct cgroup *cgrp)
-{
-	int cpu;
-
-	/* the root cgrp has cpu_stat preallocated */
-	if (!cgrp->cpu_stat) {
-		cgrp->cpu_stat = alloc_percpu(struct cgroup_cpu_stat);
-		if (!cgrp->cpu_stat)
-			return -ENOMEM;
-	}
-
-	/* ->updated_children list is self terminated */
-	for_each_possible_cpu(cpu) {
-		struct cgroup_cpu_stat *cstat = cgroup_cpu_stat(cgrp, cpu);
-
-		cstat->updated_children = cgrp;
-		u64_stats_init(&cstat->sync);
-	}
-
-	prev_cputime_init(&cgrp->stat.prev_cputime);
-
-	return 0;
-}
-
-void cgroup_stat_exit(struct cgroup *cgrp)
-{
-	int cpu;
-
-	cgroup_stat_flush(cgrp);
-
-	/* sanity check */
-	for_each_possible_cpu(cpu) {
-		struct cgroup_cpu_stat *cstat = cgroup_cpu_stat(cgrp, cpu);
-
-		if (WARN_ON_ONCE(cstat->updated_children != cgrp) ||
-		    WARN_ON_ONCE(cstat->updated_next))
-			return;
-	}
-
-	free_percpu(cgrp->cpu_stat);
-	cgrp->cpu_stat = NULL;
-}
-
-void __init cgroup_stat_boot(void)
-{
-	int cpu;
-
-	for_each_possible_cpu(cpu)
-		raw_spin_lock_init(per_cpu_ptr(&cgroup_cpu_stat_lock, cpu));
-
-	BUG_ON(cgroup_stat_init(&cgrp_dfl_root.cgrp));
-}