sched: rt-group: synchonised bandwidth period

Various SMP balancing algorithms require that the bandwidth period
run in sync.

Possible improvements are moving the rt_bandwidth thing into root_domain
and keeping a span per rt_bandwidth which marks throttled cpus.
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

include/linux/sched.h

@@ -1563,6 +1563,10 @@ int sched_nr_latency_handler(struct ctl_table *table, int write,
 extern unsigned int sysctl_sched_rt_period;
 extern int sysctl_sched_rt_runtime;
 
+int sched_rt_handler(struct ctl_table *table, int write,
+		struct file *filp, void __user *buffer, size_t *lenp,
+		loff_t *ppos);
+
 extern unsigned int sysctl_sched_compat_yield;
 
 #ifdef CONFIG_RT_MUTEXES
@@ -2052,6 +2056,9 @@ extern unsigned long sched_group_shares(struct task_group *tg);
 extern int sched_group_set_rt_runtime(struct task_group *tg,
 				      long rt_runtime_us);
 extern long sched_group_rt_runtime(struct task_group *tg);
+extern int sched_group_set_rt_period(struct task_group *tg,
+				      long rt_period_us);
+extern long sched_group_rt_period(struct task_group *tg);
 #endif
 #endif
 

kernel/sched_rt.c

@@ -62,7 +62,7 @@ static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
 	if (!rt_rq->tg)
 		return RUNTIME_INF;
 
-	return rt_rq->tg->rt_runtime;
+	return rt_rq->tg->rt_bandwidth.rt_runtime;
 }
 
 #define for_each_leaf_rt_rq(rt_rq, rq) \
@@ -127,14 +127,29 @@ static int rt_se_boosted(struct sched_rt_entity *rt_se)
 	return p->prio != p->normal_prio;
 }
 
+#ifdef CONFIG_SMP
+static inline cpumask_t sched_rt_period_mask(void)
+{
+	return cpu_rq(smp_processor_id())->rd->span;
+}
 #else
+static inline cpumask_t sched_rt_period_mask(void)
+{
+	return cpu_online_map;
+}
+#endif
 
-static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
+static inline
+struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
 {
-	if (sysctl_sched_rt_runtime == -1)
-		return RUNTIME_INF;
+	return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
+}
+
+#else
 
-	return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
+static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
+{
+	return def_rt_bandwidth.rt_runtime;
 }
 
 #define for_each_leaf_rt_rq(rt_rq, rq) \
@@ -173,8 +188,55 @@ static inline int rt_rq_throttled(struct rt_rq *rt_rq)
 {
 	return rt_rq->rt_throttled;
 }
+
+static inline cpumask_t sched_rt_period_mask(void)
+{
+	return cpu_online_map;
+}
+
+static inline
+struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
+{
+	return &cpu_rq(cpu)->rt;
+}
+
 #endif
 
+static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun)
+{
+	int i, idle = 1;
+	cpumask_t span;
+
+	if (rt_b->rt_runtime == RUNTIME_INF)
+		return 1;
+
+	span = sched_rt_period_mask();
+	for_each_cpu_mask(i, span) {
+		int enqueue = 0;
+		struct rt_rq *rt_rq = sched_rt_period_rt_rq(rt_b, i);
+		struct rq *rq = rq_of_rt_rq(rt_rq);
+
+		spin_lock(&rq->lock);
+		if (rt_rq->rt_time) {
+			u64 runtime = rt_b->rt_runtime;
+
+			rt_rq->rt_time -= min(rt_rq->rt_time, overrun*runtime);
+			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
+				rt_rq->rt_throttled = 0;
+				enqueue = 1;
+			}
+			if (rt_rq->rt_time || rt_rq->rt_nr_running)
+				idle = 0;
+		}
+
+		if (enqueue)
+			sched_rt_rq_enqueue(rt_rq);
+		spin_unlock(&rq->lock);
+	}
+
+	return idle;
+}
+
 static inline int rt_se_prio(struct sched_rt_entity *rt_se)
 {
 #ifdef CONFIG_RT_GROUP_SCHED
@@ -198,11 +260,7 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 		return rt_rq_throttled(rt_rq);
 
 	if (rt_rq->rt_time > runtime) {
-		struct rq *rq = rq_of_rt_rq(rt_rq);
-
-		rq->rt_throttled = 1;
 		rt_rq->rt_throttled = 1;
-
 		if (rt_rq_throttled(rt_rq)) {
 			sched_rt_rq_dequeue(rt_rq);
 			return 1;
@@ -212,29 +270,6 @@ static int sched_rt_runtime_exceeded(struct rt_rq *rt_rq)
 	return 0;
 }
 
-static void update_sched_rt_period(struct rq *rq)
-{
-	struct rt_rq *rt_rq;
-	u64 period;
-
-	while (rq->clock > rq->rt_period_expire) {
-		period = (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
-		rq->rt_period_expire += period;
-
-		for_each_leaf_rt_rq(rt_rq, rq) {
-			u64 runtime = sched_rt_runtime(rt_rq);
-
-			rt_rq->rt_time -= min(rt_rq->rt_time, runtime);
-			if (rt_rq->rt_throttled && rt_rq->rt_time < runtime) {
-				rt_rq->rt_throttled = 0;
-				sched_rt_rq_enqueue(rt_rq);
-			}
-		}
-
-		rq->rt_throttled = 0;
-	}
-}
-
 /*
  * Update the current task's runtime statistics. Skip current tasks that
  * are not in our scheduling class.
@@ -284,6 +319,11 @@ void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
 #ifdef CONFIG_RT_GROUP_SCHED
 	if (rt_se_boosted(rt_se))
 		rt_rq->rt_nr_boosted++;
+
+	if (rt_rq->tg)
+		start_rt_bandwidth(&rt_rq->tg->rt_bandwidth);
+#else
+	start_rt_bandwidth(&def_rt_bandwidth);
 #endif
 }
 

kernel/sysctl.c

@@ -307,7 +307,7 @@ static struct ctl_table kern_table[] = {
 		.data		= &sysctl_sched_rt_period,
 		.maxlen		= sizeof(unsigned int),
 		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
+		.proc_handler	= &sched_rt_handler,
 	},
 	{
 		.ctl_name	= CTL_UNNUMBERED,
@@ -315,7 +315,7 @@ static struct ctl_table kern_table[] = {
 		.data		= &sysctl_sched_rt_runtime,
 		.maxlen		= sizeof(int),
 		.mode		= 0644,
-		.proc_handler	= &proc_dointvec,
+		.proc_handler	= &sched_rt_handler,
 	},
 	{
 		.ctl_name	= CTL_UNNUMBERED,

kernel/time/tick-sched.c

@@ -191,7 +191,6 @@ u64 get_cpu_idle_time_us(int cpu, u64 *last_update_time)
 void tick_nohz_stop_sched_tick(void)
 {
 	unsigned long seq, last_jiffies, next_jiffies, delta_jiffies, flags;
-	unsigned long rt_jiffies;
 	struct tick_sched *ts;
 	ktime_t last_update, expires, now;
 	struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
@@ -243,10 +242,6 @@ void tick_nohz_stop_sched_tick(void)
 	next_jiffies = get_next_timer_interrupt(last_jiffies);
 	delta_jiffies = next_jiffies - last_jiffies;
 
-	rt_jiffies = rt_needs_cpu(cpu);
-	if (rt_jiffies && rt_jiffies < delta_jiffies)
-		delta_jiffies = rt_jiffies;
-
 	if (rcu_needs_cpu(cpu))
 		delta_jiffies = 1;
 	/*

kernel/user.c

@@ -193,6 +193,33 @@ static ssize_t cpu_rt_runtime_store(struct kobject *kobj,
 
 static struct kobj_attribute cpu_rt_runtime_attr =
 	__ATTR(cpu_rt_runtime, 0644, cpu_rt_runtime_show, cpu_rt_runtime_store);
+
+static ssize_t cpu_rt_period_show(struct kobject *kobj,
+				   struct kobj_attribute *attr,
+				   char *buf)
+{
+	struct user_struct *up = container_of(kobj, struct user_struct, kobj);
+
+	return sprintf(buf, "%lu\n", sched_group_rt_period(up->tg));
+}
+
+static ssize_t cpu_rt_period_store(struct kobject *kobj,
+				    struct kobj_attribute *attr,
+				    const char *buf, size_t size)
+{
+	struct user_struct *up = container_of(kobj, struct user_struct, kobj);
+	unsigned long rt_period;
+	int rc;
+
+	sscanf(buf, "%lu", &rt_period);
+
+	rc = sched_group_set_rt_period(up->tg, rt_period);
+
+	return (rc ? rc : size);
+}
+
+static struct kobj_attribute cpu_rt_period_attr =
+	__ATTR(cpu_rt_period, 0644, cpu_rt_period_show, cpu_rt_period_store);
 #endif
 
 /* default attributes per uid directory */
@@ -202,6 +229,7 @@ static struct attribute *uids_attributes[] = {
 #endif
 #ifdef CONFIG_RT_GROUP_SCHED
 	&cpu_rt_runtime_attr.attr,
+	&cpu_rt_period_attr.attr,
 #endif
 	NULL
 };