sched: rt group scheduling

Extend group scheduling to also cover the realtime classes. It uses the time
limiting introduced by the previous patch to allow multiple realtime groups.

The hard time limit is required to keep behaviour deterministic.

The algorithms used make the realtime scheduler O(tg), linear scaling wrt the
number of task groups. This is the worst case behaviour I can't seem to get out
of, the avg. case of the algorithms can be improved, I focused on correctness
and worst case.

[ akpm@linux-foundation.org: move side-effects out of BUG_ON(). ]

Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

5 files changed, 549 insertions, 206 deletions

diff --git a/include/linux/init_task.h b/include/linux/init_task.h
index ee65d87..796019b 100644
--- a/include/linux/init_task.h
+++ b/include/linux/init_task.h
@@ -130,12 +130,13 @@ extern struct group_info init_groups;
 	.normal_prio	= MAX_PRIO-20,					\
 	.policy		= SCHED_NORMAL,					\
 	.cpus_allowed	= CPU_MASK_ALL,					\
-	.nr_cpus_allowed = NR_CPUS,					\
 	.mm		= NULL,						\
 	.active_mm	= &init_mm,					\
 	.rt		= {						\
 		.run_list	= LIST_HEAD_INIT(tsk.rt.run_list),	\
-		.time_slice	= HZ, },				\
+		.time_slice	= HZ, 					\
+		.nr_cpus_allowed = NR_CPUS,				\
+	},								\
 	.ioprio		= 0,						\
 	.tasks		= LIST_HEAD_INIT(tsk.tasks),			\
 	.ptrace_children= LIST_HEAD_INIT(tsk.ptrace_children),		\
diff --git a/include/linux/sched.h b/include/linux/sched.h
index d5ea144..04eecbf 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -934,6 +934,15 @@ struct sched_rt_entity {
 	struct list_head run_list;
 	unsigned int time_slice;
 	unsigned long timeout;
+	int nr_cpus_allowed;
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	struct sched_rt_entity	*parent;
+	/* rq on which this entity is (to be) queued: */
+	struct rt_rq		*rt_rq;
+	/* rq "owned" by this entity/group: */
+	struct rt_rq		*my_q;
+#endif
 };
 
 struct task_struct {
@@ -978,7 +987,6 @@ struct task_struct {
 
 	unsigned int policy;
 	cpumask_t cpus_allowed;
-	int nr_cpus_allowed;
 
 #ifdef CONFIG_PREEMPT_RCU
 	int rcu_read_lock_nesting;
diff --git a/kernel/fork.c b/kernel/fork.c
index 9f8ef32..0c969f4 100644
--- a/kernel/fork.c
+++ b/kernel/fork.c
@@ -1246,7 +1246,7 @@ static struct task_struct *copy_process(unsigned long clone_flags,
 	 * parent's CPU). This avoids alot of nasty races.
 	 */
 	p->cpus_allowed = current->cpus_allowed;
-	p->nr_cpus_allowed = current->nr_cpus_allowed;
+	p->rt.nr_cpus_allowed = current->rt.nr_cpus_allowed;
 	if (unlikely(!cpu_isset(task_cpu(p), p->cpus_allowed) ||
 			!cpu_online(task_cpu(p))))
 		set_task_cpu(p, smp_processor_id());
diff --git a/kernel/sched.c b/kernel/sched.c
index e9a7bee..5ea2c53 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -161,6 +161,8 @@ struct rt_prio_array {
 
 struct cfs_rq;
 
+static LIST_HEAD(task_groups);
+
 /* task group related information */
 struct task_group {
 #ifdef CONFIG_FAIR_CGROUP_SCHED
@@ -171,6 +173,11 @@ struct task_group {
 	/* runqueue "owned" by this group on each cpu */
 	struct cfs_rq **cfs_rq;
 
+	struct sched_rt_entity **rt_se;
+	struct rt_rq **rt_rq;
+
+	unsigned int rt_ratio;
+
 	/*
 	 * shares assigned to a task group governs how much of cpu bandwidth
 	 * is allocated to the group. The more shares a group has, the more is
@@ -208,6 +215,7 @@ struct task_group {
 	unsigned long shares;
 
 	struct rcu_head rcu;
+	struct list_head list;
 };
 
 /* Default task group's sched entity on each cpu */
@@ -215,9 +223,15 @@ static DEFINE_PER_CPU(struct sched_entity, init_sched_entity);
 /* Default task group's cfs_rq on each cpu */
 static DEFINE_PER_CPU(struct cfs_rq, init_cfs_rq) ____cacheline_aligned_in_smp;
 
+static DEFINE_PER_CPU(struct sched_rt_entity, init_sched_rt_entity);
+static DEFINE_PER_CPU(struct rt_rq, init_rt_rq) ____cacheline_aligned_in_smp;
+
 static struct sched_entity *init_sched_entity_p[NR_CPUS];
 static struct cfs_rq *init_cfs_rq_p[NR_CPUS];
 
+static struct sched_rt_entity *init_sched_rt_entity_p[NR_CPUS];
+static struct rt_rq *init_rt_rq_p[NR_CPUS];
+
 /* task_group_mutex serializes add/remove of task groups and also changes to
  * a task group's cpu shares.
  */
@@ -240,6 +254,9 @@ static void set_se_shares(struct sched_entity *se, unsigned long shares);
 struct task_group init_task_group = {
 	.se	= init_sched_entity_p,
 	.cfs_rq = init_cfs_rq_p,
+
+	.rt_se	= init_sched_rt_entity_p,
+	.rt_rq	= init_rt_rq_p,
 };
 
 #ifdef CONFIG_FAIR_USER_SCHED
@@ -269,10 +286,13 @@ static inline struct task_group *task_group(struct task_struct *p)
 }
 
 /* Change a task's cfs_rq and parent entity if it moves across CPUs/groups */
-static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu)
+static inline void set_task_rq(struct task_struct *p, unsigned int cpu)
 {
 	p->se.cfs_rq = task_group(p)->cfs_rq[cpu];
 	p->se.parent = task_group(p)->se[cpu];
+
+	p->rt.rt_rq  = task_group(p)->rt_rq[cpu];
+	p->rt.parent = task_group(p)->rt_se[cpu];
 }
 
 static inline void lock_task_group_list(void)
@@ -297,7 +317,7 @@ static inline void unlock_doms_cur(void)
 
 #else
 
-static inline void set_task_cfs_rq(struct task_struct *p, unsigned int cpu) { }
+static inline void set_task_rq(struct task_struct *p, unsigned int cpu) { }
 static inline void lock_task_group_list(void) { }
 static inline void unlock_task_group_list(void) { }
 static inline void lock_doms_cur(void) { }
@@ -343,13 +363,22 @@ struct cfs_rq {
 struct rt_rq {
 	struct rt_prio_array active;
 	unsigned long rt_nr_running;
+#if defined CONFIG_SMP || defined CONFIG_FAIR_GROUP_SCHED
+	int highest_prio; /* highest queued rt task prio */
+#endif
 #ifdef CONFIG_SMP
 	unsigned long rt_nr_migratory;
-	int highest_prio; /* highest queued rt task prio */
 	int overloaded;
 #endif
+	int rt_throttled;
 	u64 rt_time;
-	u64 rt_throttled;
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	struct rq *rq;
+	struct list_head leaf_rt_rq_list;
+	struct task_group *tg;
+	struct sched_rt_entity *rt_se;
+#endif
 };
 
 #ifdef CONFIG_SMP
@@ -411,12 +440,14 @@ struct rq {
 	u64 nr_switches;
 
 	struct cfs_rq cfs;
+	struct rt_rq rt;
+	u64 rt_period_expire;
+
 #ifdef CONFIG_FAIR_GROUP_SCHED
 	/* list of leaf cfs_rq on this cpu: */
 	struct list_head leaf_cfs_rq_list;
+	struct list_head leaf_rt_rq_list;
 #endif
-	struct rt_rq rt;
-	u64 rt_period_expire;
 
 	/*
 	 * This is part of a global counter where only the total sum
@@ -613,9 +644,9 @@ const_debug unsigned int sysctl_sched_rt_period = 1000;
 
 /*
  * ratio of time -rt tasks may consume.
- * default: 100%
+ * default: 95%
  */
-const_debug unsigned int sysctl_sched_rt_ratio = SCHED_RT_FRAC;
+const_debug unsigned int sysctl_sched_rt_ratio = 62259;
 
 /*
  * For kernel-internal use: high-speed (but slightly incorrect) per-cpu
@@ -1337,7 +1368,7 @@ unsigned long weighted_cpuload(const int cpu)
 
 static inline void __set_task_cpu(struct task_struct *p, unsigned int cpu)
 {
-	set_task_cfs_rq(p, cpu);
+	set_task_rq(p, cpu);
 #ifdef CONFIG_SMP
 	/*
 	 * After ->cpu is set up to a new value, task_rq_lock(p, ...) can be
@@ -5281,7 +5312,7 @@ int set_cpus_allowed(struct task_struct *p, cpumask_t new_mask)
 		p->sched_class->set_cpus_allowed(p, &new_mask);
 	else {
 		p->cpus_allowed = new_mask;
-		p->nr_cpus_allowed = cpus_weight(new_mask);
+		p->rt.nr_cpus_allowed = cpus_weight(new_mask);
 	}
 
 	/* Can the task run on the task's current CPU? If so, we're done */
@@ -7079,8 +7110,50 @@ static void init_rt_rq(struct rt_rq *rt_rq, struct rq *rq)
 
 	rt_rq->rt_time = 0;
 	rt_rq->rt_throttled = 0;
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	rt_rq->rq = rq;
+#endif
 }
 
+#ifdef CONFIG_FAIR_GROUP_SCHED
+static void init_tg_cfs_entry(struct rq *rq, struct task_group *tg,
+		struct cfs_rq *cfs_rq, struct sched_entity *se,
+		int cpu, int add)
+{
+	tg->cfs_rq[cpu] = cfs_rq;
+	init_cfs_rq(cfs_rq, rq);
+	cfs_rq->tg = tg;
+	if (add)
+		list_add(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
+
+	tg->se[cpu] = se;
+	se->cfs_rq = &rq->cfs;
+	se->my_q = cfs_rq;
+	se->load.weight = tg->shares;
+	se->load.inv_weight = div64_64(1ULL<<32, se->load.weight);
+	se->parent = NULL;
+}
+
+static void init_tg_rt_entry(struct rq *rq, struct task_group *tg,
+		struct rt_rq *rt_rq, struct sched_rt_entity *rt_se,
+		int cpu, int add)
+{
+	tg->rt_rq[cpu] = rt_rq;
+	init_rt_rq(rt_rq, rq);
+	rt_rq->tg = tg;
+	rt_rq->rt_se = rt_se;
+	if (add)
+		list_add(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
+
+	tg->rt_se[cpu] = rt_se;
+	rt_se->rt_rq = &rq->rt;
+	rt_se->my_q = rt_rq;
+	rt_se->parent = NULL;
+	INIT_LIST_HEAD(&rt_se->run_list);
+}
+#endif
+
 void __init sched_init(void)
 {
 	int highest_cpu = 0;
@@ -7090,6 +7163,10 @@ void __init sched_init(void)
 	init_defrootdomain();
 #endif
 
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	list_add(&init_task_group.list, &task_groups);
+#endif
+
 	for_each_possible_cpu(i) {
 		struct rq *rq;
 
@@ -7099,30 +7176,20 @@ void __init sched_init(void)
 		rq->nr_running = 0;
 		rq->clock = 1;
 		init_cfs_rq(&rq->cfs, rq);
+		init_rt_rq(&rq->rt, rq);
 #ifdef CONFIG_FAIR_GROUP_SCHED
-		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
-		{
-			struct cfs_rq *cfs_rq = &per_cpu(init_cfs_rq, i);
-			struct sched_entity *se =
-					 &per_cpu(init_sched_entity, i);
-
-			init_cfs_rq_p[i] = cfs_rq;
-			init_cfs_rq(cfs_rq, rq);
-			cfs_rq->tg = &init_task_group;
-			list_add(&cfs_rq->leaf_cfs_rq_list,
-							 &rq->leaf_cfs_rq_list);
-
-			init_sched_entity_p[i] = se;
-			se->cfs_rq = &rq->cfs;
-			se->my_q = cfs_rq;
-			se->load.weight = init_task_group_load;
-			se->load.inv_weight =
-				 div64_64(1ULL<<32, init_task_group_load);
-			se->parent = NULL;
-		}
 		init_task_group.shares = init_task_group_load;
+		INIT_LIST_HEAD(&rq->leaf_cfs_rq_list);
+		init_tg_cfs_entry(rq, &init_task_group,
+				&per_cpu(init_cfs_rq, i),
+				&per_cpu(init_sched_entity, i), i, 1);
+
+		init_task_group.rt_ratio = sysctl_sched_rt_ratio; /* XXX */
+		INIT_LIST_HEAD(&rq->leaf_rt_rq_list);
+		init_tg_rt_entry(rq, &init_task_group,
+				&per_cpu(init_rt_rq, i),
+				&per_cpu(init_sched_rt_entity, i), i, 1);
 #endif
-		init_rt_rq(&rq->rt, rq);
 		rq->rt_period_expire = 0;
 
 		for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
@@ -7460,12 +7527,36 @@ static int load_balance_monitor(void *unused)
 }
 #endif	/* CONFIG_SMP */
 
+static void free_sched_group(struct task_group *tg)
+{
+	int i;
+
+	for_each_possible_cpu(i) {
+		if (tg->cfs_rq)
+			kfree(tg->cfs_rq[i]);
+		if (tg->se)
+			kfree(tg->se[i]);
+		if (tg->rt_rq)
+			kfree(tg->rt_rq[i]);
+		if (tg->rt_se)
+			kfree(tg->rt_se[i]);
+	}
+
+	kfree(tg->cfs_rq);
+	kfree(tg->se);
+	kfree(tg->rt_rq);
+	kfree(tg->rt_se);
+	kfree(tg);
+}
+
 /* allocate runqueue etc for a new task group */
 struct task_group *sched_create_group(void)
 {
 	struct task_group *tg;
 	struct cfs_rq *cfs_rq;
 	struct sched_entity *se;
+	struct rt_rq *rt_rq;
+	struct sched_rt_entity *rt_se;
 	struct rq *rq;
 	int i;
 
@@ -7479,100 +7570,89 @@ struct task_group *sched_create_group(void)
 	tg->se = kzalloc(sizeof(se) * NR_CPUS, GFP_KERNEL);
 	if (!tg->se)
 		goto err;
+	tg->rt_rq = kzalloc(sizeof(rt_rq) * NR_CPUS, GFP_KERNEL);
+	if (!tg->rt_rq)
+		goto err;
+	tg->rt_se = kzalloc(sizeof(rt_se) * NR_CPUS, GFP_KERNEL);
+	if (!tg->rt_se)
+		goto err;
+
+	tg->shares = NICE_0_LOAD;
+	tg->rt_ratio = 0; /* XXX */
 
 	for_each_possible_cpu(i) {
 		rq = cpu_rq(i);
 
-		cfs_rq = kmalloc_node(sizeof(struct cfs_rq), GFP_KERNEL,
-							 cpu_to_node(i));
+		cfs_rq = kmalloc_node(sizeof(struct cfs_rq),
+				GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
 		if (!cfs_rq)
 			goto err;
 
-		se = kmalloc_node(sizeof(struct sched_entity), GFP_KERNEL,
-							cpu_to_node(i));
+		se = kmalloc_node(sizeof(struct sched_entity),
+				GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
 		if (!se)
 			goto err;
 
-		memset(cfs_rq, 0, sizeof(struct cfs_rq));
-		memset(se, 0, sizeof(struct sched_entity));
+		rt_rq = kmalloc_node(sizeof(struct rt_rq),
+				GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
+		if (!rt_rq)
+			goto err;
 
-		tg->cfs_rq[i] = cfs_rq;
-		init_cfs_rq(cfs_rq, rq);
-		cfs_rq->tg = tg;
+		rt_se = kmalloc_node(sizeof(struct sched_rt_entity),
+				GFP_KERNEL|__GFP_ZERO, cpu_to_node(i));
+		if (!rt_se)
+			goto err;
 
-		tg->se[i] = se;
-		se->cfs_rq = &rq->cfs;
-		se->my_q = cfs_rq;
-		se->load.weight = NICE_0_LOAD;
-		se->load.inv_weight = div64_64(1ULL<<32, NICE_0_LOAD);
-		se->parent = NULL;
+		init_tg_cfs_entry(rq, tg, cfs_rq, se, i, 0);
+		init_tg_rt_entry(rq, tg, rt_rq, rt_se, i, 0);
 	}
 
-	tg->shares = NICE_0_LOAD;
-
 	lock_task_group_list();
 	for_each_possible_cpu(i) {
 		rq = cpu_rq(i);
 		cfs_rq = tg->cfs_rq[i];
 		list_add_rcu(&cfs_rq->leaf_cfs_rq_list, &rq->leaf_cfs_rq_list);
+		rt_rq = tg->rt_rq[i];
+		list_add_rcu(&rt_rq->leaf_rt_rq_list, &rq->leaf_rt_rq_list);
 	}
+	list_add_rcu(&tg->list, &task_groups);
 	unlock_task_group_list();
 
 	return tg;
 
 err:
-	for_each_possible_cpu(i) {
-		if (tg->cfs_rq)
-			kfree(tg->cfs_rq[i]);
-		if (tg->se)
-			kfree(tg->se[i]);
-	}
-	kfree(tg->cfs_rq);
-	kfree(tg->se);
-	kfree(tg);
-
+	free_sched_group(tg);
 	return ERR_PTR(-ENOMEM);
 }
 
 /* rcu callback to free various structures associated with a task group */
-static void free_sched_group(struct rcu_head *rhp)
+static void free_sched_group_rcu(struct rcu_head *rhp)
 {
-	struct task_group *tg = container_of(rhp, struct task_group, rcu);
-	struct cfs_rq *cfs_rq;
-	struct sched_entity *se;
-	int i;
-
 	/* now it should be safe to free those cfs_rqs */
-	for_each_possible_cpu(i) {
-		cfs_rq = tg->cfs_rq[i];
-		kfree(cfs_rq);
-
-		se = tg->se[i];
-		kfree(se);
-	}
-
-	kfree(tg->cfs_rq);
-	kfree(tg->se);
-	kfree(tg);
+	free_sched_group(container_of(rhp, struct task_group, rcu));
 }
 
 /* Destroy runqueue etc associated with a task group */
 void sched_destroy_group(struct task_group *tg)
 {
 	struct cfs_rq *cfs_rq = NULL;
+	struct rt_rq *rt_rq = NULL;
 	int i;
 
 	lock_task_group_list();
 	for_each_possible_cpu(i) {
 		cfs_rq = tg->cfs_rq[i];
 		list_del_rcu(&cfs_rq->leaf_cfs_rq_list);
+		rt_rq = tg->rt_rq[i];
+		list_del_rcu(&rt_rq->leaf_rt_rq_list);
 	}
+	list_del_rcu(&tg->list);
 	unlock_task_group_list();
 
 	BUG_ON(!cfs_rq);
 
 	/* wait for possible concurrent references to cfs_rqs complete */
-	call_rcu(&tg->rcu, free_sched_group);
+	call_rcu(&tg->rcu, free_sched_group_rcu);
 }
 
 /* change task's runqueue when it moves between groups.
@@ -7588,11 +7668,6 @@ void sched_move_task(struct task_struct *tsk)
 
 	rq = task_rq_lock(tsk, &flags);
 
-	if (tsk->sched_class != &fair_sched_class) {
-		set_task_cfs_rq(tsk, task_cpu(tsk));
-		goto done;
-	}
-
 	update_rq_clock(rq);
 
 	running = task_current(rq, tsk);
@@ -7604,7 +7679,7 @@ void sched_move_task(struct task_struct *tsk)
 			tsk->sched_class->put_prev_task(rq, tsk);
 	}
 
-	set_task_cfs_rq(tsk, task_cpu(tsk));
+	set_task_rq(tsk, task_cpu(tsk));
 
 	if (on_rq) {
 		if (unlikely(running))
@@ -7612,7 +7687,6 @@ void sched_move_task(struct task_struct *tsk)
 		enqueue_task(rq, tsk, 0);
 	}
 
-done:
 	task_rq_unlock(rq, &flags);
 }
 
@@ -7697,6 +7771,31 @@ unsigned long sched_group_shares(struct task_group *tg)
 	return tg->shares;
 }
 
+/*
+ * Ensure the total rt_ratio <= sysctl_sched_rt_ratio
+ */
+int sched_group_set_rt_ratio(struct task_group *tg, unsigned long rt_ratio)
+{
+	struct task_group *tgi;
+	unsigned long total = 0;
+
+	rcu_read_lock();
+	list_for_each_entry_rcu(tgi, &task_groups, list)
+		total += tgi->rt_ratio;
+	rcu_read_unlock();
+
+	if (total + rt_ratio - tg->rt_ratio > sysctl_sched_rt_ratio)
+		return -EINVAL;
+
+	tg->rt_ratio = rt_ratio;
+	return 0;
+}
+
+unsigned long sched_group_rt_ratio(struct task_group *tg)
+{
+	return tg->rt_ratio;
+}
+
 #endif	/* CONFIG_FAIR_GROUP_SCHED */
 
 #ifdef CONFIG_FAIR_CGROUP_SCHED
@@ -7772,12 +7871,30 @@ static u64 cpu_shares_read_uint(struct cgroup *cgrp, struct cftype *cft)
 	return (u64) tg->shares;
 }
 
+static int cpu_rt_ratio_write_uint(struct cgroup *cgrp, struct cftype *cftype,
+		u64 rt_ratio_val)
+{
+	return sched_group_set_rt_ratio(cgroup_tg(cgrp), rt_ratio_val);
+}
+
+static u64 cpu_rt_ratio_read_uint(struct cgroup *cgrp, struct cftype *cft)
+{
+	struct task_group *tg = cgroup_tg(cgrp);
+
+	return (u64) tg->rt_ratio;
+}
+
 static struct cftype cpu_files[] = {
 	{
 		.name = "shares",
 		.read_uint = cpu_shares_read_uint,
 		.write_uint = cpu_shares_write_uint,
 	},
+	{
+		.name = "rt_ratio",
+		.read_uint = cpu_rt_ratio_read_uint,
+		.write_uint = cpu_rt_ratio_write_uint,
+	},
 };
 
 static int cpu_cgroup_populate(struct cgroup_subsys *ss, struct cgroup *cont)
diff --git a/kernel/sched_rt.c b/kernel/sched_rt.c
index fd10d96..1178257 100644
--- a/kernel/sched_rt.c
+++ b/kernel/sched_rt.c
@@ -45,47 +45,167 @@ static void update_rt_migration(struct rq *rq)
 }
 #endif /* CONFIG_SMP */
 
-static int sched_rt_ratio_exceeded(struct rq *rq, struct rt_rq *rt_rq)
+static inline struct task_struct *rt_task_of(struct sched_rt_entity *rt_se)
 {
+	return container_of(rt_se, struct task_struct, rt);
+}
+
+static inline int on_rt_rq(struct sched_rt_entity *rt_se)
+{
+	return !list_empty(&rt_se->run_list);
+}
+
+#ifdef CONFIG_FAIR_GROUP_SCHED
+
+static inline unsigned int sched_rt_ratio(struct rt_rq *rt_rq)
+{
+	if (!rt_rq->tg)
+		return SCHED_RT_FRAC;
+
+	return rt_rq->tg->rt_ratio;
+}
+
+#define for_each_leaf_rt_rq(rt_rq, rq) \
+	list_for_each_entry(rt_rq, &rq->leaf_rt_rq_list, leaf_rt_rq_list)
+
+static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
+{
+	return rt_rq->rq;
+}
+
+static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
+{
+	return rt_se->rt_rq;
+}
+
+#define for_each_sched_rt_entity(rt_se) \
+	for (; rt_se; rt_se = rt_se->parent)
+
+static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
+{
+	return rt_se->my_q;
+}
+
+static void enqueue_rt_entity(struct sched_rt_entity *rt_se);
+static void dequeue_rt_entity(struct sched_rt_entity *rt_se);
+
+static void sched_rt_ratio_enqueue(struct rt_rq *rt_rq)
+{
+	struct sched_rt_entity *rt_se = rt_rq->rt_se;
+
+	if (rt_se && !on_rt_rq(rt_se) && rt_rq->rt_nr_running) {
+		enqueue_rt_entity(rt_se);
+		resched_task(rq_of_rt_rq(rt_rq)->curr);
+	}
+}
+
+static void sched_rt_ratio_dequeue(struct rt_rq *rt_rq)
+{
+	struct sched_rt_entity *rt_se = rt_rq->rt_se;
+
+	if (rt_se && on_rt_rq(rt_se))
+		dequeue_rt_entity(rt_se);
+}
+
+#else
+
+static inline unsigned int sched_rt_ratio(struct rt_rq *rt_rq)
+{
+	return sysctl_sched_rt_ratio;
+}
+
+#define for_each_leaf_rt_rq(rt_rq, rq) \
+	for (rt_rq = &rq->rt; rt_rq; rt_rq = NULL)
+
+static inline struct rq *rq_of_rt_rq(struct rt_rq *rt_rq)
+{
+	return container_of(rt_rq, struct rq, rt);
+}
+
+static inline struct rt_rq *rt_rq_of_se(struct sched_rt_entity *rt_se)
+{
+	struct task_struct *p = rt_task_of(rt_se);
+	struct rq *rq = task_rq(p);
+
+	return &rq->rt;
+}
+
+#define for_each_sched_rt_entity(rt_se) \
+	for (; rt_se; rt_se = NULL)
+
+static inline struct rt_rq *group_rt_rq(struct sched_rt_entity *rt_se)
+{
+	return NULL;
+}
+
+static inline void sched_rt_ratio_enqueue(struct rt_rq *rt_rq)
+{
+}
+
+static inline void sched_rt_ratio_dequeue(struct rt_rq *rt_rq)
+{
+}
+
+#endif
+
+static inline int rt_se_prio(struct sched_rt_entity *rt_se)
+{
+#ifdef CONFIG_FAIR_GROUP_SCHED
+	struct rt_rq *rt_rq = group_rt_rq(rt_se);
+
+	if (rt_rq)
+		return rt_rq->highest_prio;
+#endif
+
+	return rt_task_of(rt_se)->prio;
+}
+
+static int sched_rt_ratio_exceeded(struct rt_rq *rt_rq)
+{
+	unsigned int rt_ratio = sched_rt_ratio(rt_rq);
 	u64 period, ratio;
 
-	if (sysctl_sched_rt_ratio == SCHED_RT_FRAC)
+	if (rt_ratio == SCHED_RT_FRAC)
 		return 0;
 
 	if (rt_rq->rt_throttled)
 		return 1;
 
 	period = (u64)sysctl_sched_rt_period * NSEC_PER_MSEC;
-	ratio = (period * sysctl_sched_rt_ratio) >> SCHED_RT_FRAC_SHIFT;
+	ratio = (period * rt_ratio) >> SCHED_RT_FRAC_SHIFT;
 
 	if (rt_rq->rt_time > ratio) {
-		rt_rq->rt_throttled = rq->clock + period - rt_rq->rt_time;
+		rt_rq->rt_throttled = 1;
+		sched_rt_ratio_dequeue(rt_rq);
 		return 1;
 	}
 
 	return 0;
 }
 
+static void __update_sched_rt_period(struct rt_rq *rt_rq, u64 period)
+{
+	unsigned long rt_ratio = sched_rt_ratio(rt_rq);
+	u64 ratio = (period * rt_ratio) >> SCHED_RT_FRAC_SHIFT;
+
+	rt_rq->rt_time -= min(rt_rq->rt_time, ratio);
+	if (rt_rq->rt_throttled) {
+		rt_rq->rt_throttled = 0;
+		sched_rt_ratio_enqueue(rt_rq);
+	}
+}
+
 static void update_sched_rt_period(struct rq *rq)
 {
-	while (rq->clock > rq->rt_period_expire) {
-		u64 period, ratio;
+	struct rt_rq *rt_rq;
+	u64 period;
 
+	while (rq->clock > rq->rt_period_expire) {
 		period = (u64)sysctl_sched_rt_period * NSEC_PER_MSEC;
-		ratio = (period * sysctl_sched_rt_ratio) >> SCHED_RT_FRAC_SHIFT;
-
-		rq->rt.rt_time -= min(rq->rt.rt_time, ratio);
 		rq->rt_period_expire += period;
-	}
 
-	/*
-	 * When the rt throttle is expired, let them rip.
-	 * (XXX: use hrtick when available)
-	 */
-	if (rq->rt.rt_throttled && rq->clock > rq->rt.rt_throttled) {
-		rq->rt.rt_throttled = 0;
-		if (!sched_rt_ratio_exceeded(rq, &rq->rt))
-			resched_task(rq->curr);
+		for_each_leaf_rt_rq(rt_rq, rq)
+			__update_sched_rt_period(rt_rq, period);
 	}
 }
 
@@ -96,6 +216,8 @@ static void update_sched_rt_period(struct rq *rq)
 static void update_curr_rt(struct rq *rq)
 {
 	struct task_struct *curr = rq->curr;
+	struct sched_rt_entity *rt_se = &curr->rt;
+	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
 	u64 delta_exec;
 
 	if (!task_has_rt_policy(curr))
@@ -111,95 +233,184 @@ static void update_curr_rt(struct rq *rq)
 	curr->se.exec_start = rq->clock;
 	cpuacct_charge(curr, delta_exec);
 
-	rq->rt.rt_time += delta_exec;
-	update_sched_rt_period(rq);
-	if (sched_rt_ratio_exceeded(rq, &rq->rt))
+	rt_rq->rt_time += delta_exec;
+	/*
+	 * might make it a tad more accurate:
+	 *
+	 * update_sched_rt_period(rq);
+	 */
+	if (sched_rt_ratio_exceeded(rt_rq))
 		resched_task(curr);
 }
 
-static inline void inc_rt_tasks(struct task_struct *p, struct rq *rq)
+static inline
+void inc_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
 {
-	WARN_ON(!rt_task(p));
-	rq->rt.rt_nr_running++;
+	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
+	rt_rq->rt_nr_running++;
+#if defined CONFIG_SMP || defined CONFIG_FAIR_GROUP_SCHED
+	if (rt_se_prio(rt_se) < rt_rq->highest_prio)
+		rt_rq->highest_prio = rt_se_prio(rt_se);
+#endif
 #ifdef CONFIG_SMP
-	if (p->prio < rq->rt.highest_prio)
-		rq->rt.highest_prio = p->prio;
-	if (p->nr_cpus_allowed > 1)
+	if (rt_se->nr_cpus_allowed > 1) {
+		struct rq *rq = rq_of_rt_rq(rt_rq);
 		rq->rt.rt_nr_migratory++;
+	}
 
-	update_rt_migration(rq);
-#endif /* CONFIG_SMP */
+	update_rt_migration(rq_of_rt_rq(rt_rq));
+#endif
 }
 
-static inline void dec_rt_tasks(struct task_struct *p, struct rq *rq)
+static inline
+void dec_rt_tasks(struct sched_rt_entity *rt_se, struct rt_rq *rt_rq)
 {
-	WARN_ON(!rt_task(p));
-	WARN_ON(!rq->rt.rt_nr_running);
-	rq->rt.rt_nr_running--;
-#ifdef CONFIG_SMP
-	if (rq->rt.rt_nr_running) {
+	WARN_ON(!rt_prio(rt_se_prio(rt_se)));
+	WARN_ON(!rt_rq->rt_nr_running);
+	rt_rq->rt_nr_running--;
+#if defined CONFIG_SMP || defined CONFIG_FAIR_GROUP_SCHED
+	if (rt_rq->rt_nr_running) {
 		struct rt_prio_array *array;
 
-		WARN_ON(p->prio < rq->rt.highest_prio);
-		if (p->prio == rq->rt.highest_prio) {
+		WARN_ON(rt_se_prio(rt_se) < rt_rq->highest_prio);
+		if (rt_se_prio(rt_se) == rt_rq->highest_prio) {
 			/* recalculate */
-			array = &rq->rt.active;
-			rq->rt.highest_prio =
+			array = &rt_rq->active;
+			rt_rq->highest_prio =
 				sched_find_first_bit(array->bitmap);
 		} /* otherwise leave rq->highest prio alone */
 	} else
-		rq->rt.highest_prio = MAX_RT_PRIO;
-	if (p->nr_cpus_allowed > 1)
+		rt_rq->highest_prio = MAX_RT_PRIO;
+#endif
+#ifdef CONFIG_SMP
+	if (rt_se->nr_cpus_allowed > 1) {
+		struct rq *rq = rq_of_rt_rq(rt_rq);
 		rq->rt.rt_nr_migratory--;
+	}
 
-	update_rt_migration(rq);
+	update_rt_migration(rq_of_rt_rq(rt_rq));
 #endif /* CONFIG_SMP */
 }
 
-static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
+static void enqueue_rt_entity(struct sched_rt_entity *rt_se)
 {
-	struct rt_prio_array *array = &rq->rt.active;
+	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
+	struct rt_prio_array *array = &rt_rq->active;
+	struct rt_rq *group_rq = group_rt_rq(rt_se);
 
-	list_add_tail(&p->rt.run_list, array->queue + p->prio);
-	__set_bit(p->prio, array->bitmap);
-	inc_cpu_load(rq, p->se.load.weight);
+	if (group_rq && group_rq->rt_throttled)
+		return;
 
-	inc_rt_tasks(p, rq);
+	list_add_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se));
+	__set_bit(rt_se_prio(rt_se), array->bitmap);
 
-	if (wakeup)
-		p->rt.timeout = 0;
+	inc_rt_tasks(rt_se, rt_rq);
+}
+
+static void dequeue_rt_entity(struct sched_rt_entity *rt_se)
+{
+	struct rt_rq *rt_rq = rt_rq_of_se(rt_se);
+	struct rt_prio_array *array = &rt_rq->active;
+
+	list_del_init(&rt_se->run_list);
+	if (list_empty(array->queue + rt_se_prio(rt_se)))
+		__clear_bit(rt_se_prio(rt_se), array->bitmap);
+
+	dec_rt_tasks(rt_se, rt_rq);
+}
+
+/*
+ * Because the prio of an upper entry depends on the lower
+ * entries, we must remove entries top - down.
+ *
+ * XXX: O(1/2 h^2) because we can only walk up, not down the chain.
+ *      doesn't matter much for now, as h=2 for GROUP_SCHED.
+ */
+static void dequeue_rt_stack(struct task_struct *p)
+{
+	struct sched_rt_entity *rt_se, *top_se;
+
+	/*
+	 * dequeue all, top - down.
+	 */
+	do {
+		rt_se = &p->rt;
+		top_se = NULL;
+		for_each_sched_rt_entity(rt_se) {
+			if (on_rt_rq(rt_se))
+				top_se = rt_se;
+		}
+		if (top_se)
+			dequeue_rt_entity(top_se);
+	} while (top_se);
 }
 
 /*
  * Adding/removing a task to/from a priority array:
  */
+static void enqueue_task_rt(struct rq *rq, struct task_struct *p, int wakeup)
+{
+	struct sched_rt_entity *rt_se = &p->rt;
+
+	if (wakeup)
+		rt_se->timeout = 0;
+
+	dequeue_rt_stack(p);
+
+	/*
+	 * enqueue everybody, bottom - up.
+	 */
+	for_each_sched_rt_entity(rt_se)
+		enqueue_rt_entity(rt_se);
+
+	inc_cpu_load(rq, p->se.load.weight);
+}
+
 static void dequeue_task_rt(struct rq *rq, struct task_struct *p, int sleep)
 {
-	struct rt_prio_array *array = &rq->rt.active;
+	struct sched_rt_entity *rt_se = &p->rt;
+	struct rt_rq *rt_rq;
 
 	update_curr_rt(rq);
 
-	list_del(&p->rt.run_list);
-	if (list_empty(array->queue + p->prio))
-		__clear_bit(p->prio, array->bitmap);
-	dec_cpu_load(rq, p->se.load.weight);
+	dequeue_rt_stack(p);
+
+	/*
+	 * re-enqueue all non-empty rt_rq entities.
+	 */
+	for_each_sched_rt_entity(rt_se) {
+		rt_rq = group_rt_rq(rt_se);
+		if (rt_rq && rt_rq->rt_nr_running)
+			enqueue_rt_entity(rt_se);
+	}
 
-	dec_rt_tasks(p, rq);
+	dec_cpu_load(rq, p->se.load.weight);
 }
 
 /*
  * Put task to the end of the run list without the overhead of dequeue
  * followed by enqueue.
  */
+static
+void requeue_rt_entity(struct rt_rq *rt_rq, struct sched_rt_entity *rt_se)
+{
+	struct rt_prio_array *array = &rt_rq->active;
+
+	list_move_tail(&rt_se->run_list, array->queue + rt_se_prio(rt_se));
+}
+
 static void requeue_task_rt(struct rq *rq, struct task_struct *p)
 {
-	struct rt_prio_array *array = &rq->rt.active;
+	struct sched_rt_entity *rt_se = &p->rt;
+	struct rt_rq *rt_rq;
 
-	list_move_tail(&p->rt.run_list, array->queue + p->prio);
+	for_each_sched_rt_entity(rt_se) {
+		rt_rq = rt_rq_of_se(rt_se);
+		requeue_rt_entity(rt_rq, rt_se);
+	}
 }
 
-static void
-yield_task_rt(struct rq *rq)
+static void yield_task_rt(struct rq *rq)
 {
 	requeue_task_rt(rq, rq->curr);
 }
@@ -229,7 +440,7 @@ static int select_task_rq_rt(struct task_struct *p, int sync)
 	 * cold cache anyway.
 	 */
 	if (unlikely(rt_task(rq->curr)) &&
-	    (p->nr_cpus_allowed > 1)) {
+	    (p->rt.nr_cpus_allowed > 1)) {
 		int cpu = find_lowest_rq(p);
 
 		return (cpu == -1) ? task_cpu(p) : cpu;
@@ -252,27 +463,51 @@ static void check_preempt_curr_rt(struct rq *rq, struct task_struct *p)
 		resched_task(rq->curr);
 }
 
-static struct task_struct *pick_next_task_rt(struct rq *rq)
+static struct sched_rt_entity *pick_next_rt_entity(struct rq *rq,
+						   struct rt_rq *rt_rq)
 {
-	struct rt_prio_array *array = &rq->rt.active;
-	struct task_struct *next;
+	struct rt_prio_array *array = &rt_rq->active;
+	struct sched_rt_entity *next = NULL;
 	struct list_head *queue;
-	struct rt_rq *rt_rq = &rq->rt;
 	int idx;
 
-	if (sched_rt_ratio_exceeded(rq, rt_rq))
-		return NULL;
+	if (sched_rt_ratio_exceeded(rt_rq))
+		goto out;
 
 	idx = sched_find_first_bit(array->bitmap);
-	if (idx >= MAX_RT_PRIO)
-		return NULL;
+	BUG_ON(idx >= MAX_RT_PRIO);
 
 	queue = array->queue + idx;
-	next = list_entry(queue->next, struct task_struct, rt.run_list);
+	next = list_entry(queue->next, struct sched_rt_entity, run_list);
+ out:
+	return next;
+}
 
-	next->se.exec_start = rq->clock;
+static struct task_struct *pick_next_task_rt(struct rq *rq)
+{
+	struct sched_rt_entity *rt_se;
+	struct task_struct *p;
+	struct rt_rq *rt_rq;
 
-	return next;
+ retry:
+	rt_rq = &rq->rt;
+
+	if (unlikely(!rt_rq->rt_nr_running))
+		return NULL;
+
+	if (sched_rt_ratio_exceeded(rt_rq))
+		return NULL;
+
+	do {
+		rt_se = pick_next_rt_entity(rq, rt_rq);
+		if (unlikely(!rt_se))
+			goto retry;
+		rt_rq = group_rt_rq(rt_se);
+	} while (rt_rq);
+
+	p = rt_task_of(rt_se);
+	p->se.exec_start = rq->clock;
+	return p;
 }
 
 static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
@@ -282,6 +517,7 @@ static void put_prev_task_rt(struct rq *rq, struct task_struct *p)
 }
 
 #ifdef CONFIG_SMP
+
 /* Only try algorithms three times */
 #define RT_MAX_TRIES 3
 
@@ -292,7 +528,7 @@ static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
 {
 	if (!task_running(rq, p) &&
 	    (cpu < 0 || cpu_isset(cpu, p->cpus_allowed)) &&
-	    (p->nr_cpus_allowed > 1))
+	    (p->rt.nr_cpus_allowed > 1))
 		return 1;
 	return 0;
 }
@@ -300,52 +536,33 @@ static int pick_rt_task(struct rq *rq, struct task_struct *p, int cpu)
 /* Return the second highest RT task, NULL otherwise */
 static struct task_struct *pick_next_highest_task_rt(struct rq *rq, int cpu)
 {
-	struct rt_prio_array *array = &rq->rt.active;
-	struct task_struct *next;
-	struct list_head *queue;
+	struct task_struct *next = NULL;
+	struct sched_rt_entity *rt_se;
+	struct rt_prio_array *array;
+	struct rt_rq *rt_rq;
 	int idx;
 
-	if (likely(rq->rt.rt_nr_running < 2))
-		return NULL;
-
-	idx = sched_find_first_bit(array->bitmap);
-	if (unlikely(idx >= MAX_RT_PRIO)) {
-		WARN_ON(1); /* rt_nr_running is bad */
-		return NULL;
-	}
-
-	queue = array->queue + idx;
-	BUG_ON(list_empty(queue));
-
-	next = list_entry(queue->next, struct task_struct, rt.run_list);
-	if (unlikely(pick_rt_task(rq, next, cpu)))
-		goto out;
-
-	if (queue->next->next != queue) {
-		/* same prio task */
-		next = list_entry(queue->next->next, struct task_struct,
-				  rt.run_list);
-		if (pick_rt_task(rq, next, cpu))
-			goto out;
-	}
-
- retry:
-	/* slower, but more flexible */
-	idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
-	if (unlikely(idx >= MAX_RT_PRIO))
-		return NULL;
-
-	queue = array->queue + idx;
-	BUG_ON(list_empty(queue));
-
-	list_for_each_entry(next, queue, rt.run_list) {
-		if (pick_rt_task(rq, next, cpu))
-			goto out;
+	for_each_leaf_rt_rq(rt_rq, rq) {
+		array = &rt_rq->active;
+		idx = sched_find_first_bit(array->bitmap);
+ next_idx:
+		if (idx >= MAX_RT_PRIO)
+			continue;
+		if (next && next->prio < idx)
+			continue;
+		list_for_each_entry(rt_se, array->queue + idx, run_list) {
+			struct task_struct *p = rt_task_of(rt_se);
+			if (pick_rt_task(rq, p, cpu)) {
+				next = p;
+				break;
+			}
+		}
+		if (!next) {
+			idx = find_next_bit(array->bitmap, MAX_RT_PRIO, idx+1);
+			goto next_idx;
+		}
 	}
 
-	goto retry;
-
- out:
 	return next;
 }
 
@@ -774,12 +991,12 @@ static void set_cpus_allowed_rt(struct task_struct *p, cpumask_t *new_mask)
 	 * Update the migration status of the RQ if we have an RT task
 	 * which is running AND changing its weight value.
 	 */
-	if (p->se.on_rq && (weight != p->nr_cpus_allowed)) {
+	if (p->se.on_rq && (weight != p->rt.nr_cpus_allowed)) {
 		struct rq *rq = task_rq(p);
 
-		if ((p->nr_cpus_allowed <= 1) && (weight > 1)) {
+		if ((p->rt.nr_cpus_allowed <= 1) && (weight > 1)) {
 			rq->rt.rt_nr_migratory++;
-		} else if ((p->nr_cpus_allowed > 1) && (weight <= 1)) {
+		} else if ((p->rt.nr_cpus_allowed > 1) && (weight <= 1)) {
 			BUG_ON(!rq->rt.rt_nr_migratory);
 			rq->rt.rt_nr_migratory--;
 		}
@@ -788,7 +1005,7 @@ static void set_cpus_allowed_rt(struct task_struct *p, cpumask_t *new_mask)
 	}
 
 	p->cpus_allowed    = *new_mask;
-	p->nr_cpus_allowed = weight;
+	p->rt.nr_cpus_allowed = weight;
 }
 
 /* Assumes rq->lock is held */