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>
This commit is contained in:
Peter Zijlstra 2008-04-19 19:44:57 +02:00 committed by Ingo Molnar
parent 57d3da2911
commit d0b27fa778
6 changed files with 320 additions and 88 deletions

View File

@ -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

View File

@ -115,6 +115,11 @@ unsigned long long __attribute__((weak)) sched_clock(void)
*/
#define DEF_TIMESLICE (100 * HZ / 1000)
/*
* single value that denotes runtime == period, ie unlimited time.
*/
#define RUNTIME_INF ((u64)~0ULL)
#ifdef CONFIG_SMP
/*
* Divide a load by a sched group cpu_power : (load / sg->__cpu_power)
@ -156,6 +161,80 @@ struct rt_prio_array {
struct list_head queue[MAX_RT_PRIO];
};
struct rt_bandwidth {
ktime_t rt_period;
u64 rt_runtime;
struct hrtimer rt_period_timer;
};
static struct rt_bandwidth def_rt_bandwidth;
static int do_sched_rt_period_timer(struct rt_bandwidth *rt_b, int overrun);
static enum hrtimer_restart sched_rt_period_timer(struct hrtimer *timer)
{
struct rt_bandwidth *rt_b =
container_of(timer, struct rt_bandwidth, rt_period_timer);
ktime_t now;
int overrun;
int idle = 0;
for (;;) {
now = hrtimer_cb_get_time(timer);
overrun = hrtimer_forward(timer, now, rt_b->rt_period);
if (!overrun)
break;
idle = do_sched_rt_period_timer(rt_b, overrun);
}
return idle ? HRTIMER_NORESTART : HRTIMER_RESTART;
}
static
void init_rt_bandwidth(struct rt_bandwidth *rt_b, u64 period, u64 runtime)
{
rt_b->rt_period = ns_to_ktime(period);
rt_b->rt_runtime = runtime;
hrtimer_init(&rt_b->rt_period_timer,
CLOCK_MONOTONIC, HRTIMER_MODE_REL);
rt_b->rt_period_timer.function = sched_rt_period_timer;
rt_b->rt_period_timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
}
static void start_rt_bandwidth(struct rt_bandwidth *rt_b)
{
ktime_t now;
if (rt_b->rt_runtime == RUNTIME_INF)
return;
if (hrtimer_active(&rt_b->rt_period_timer))
return;
spin_lock(&rt_b->rt_runtime_lock);
for (;;) {
if (hrtimer_active(&rt_b->rt_period_timer))
break;
now = hrtimer_cb_get_time(&rt_b->rt_period_timer);
hrtimer_forward(&rt_b->rt_period_timer, now, rt_b->rt_period);
hrtimer_start(&rt_b->rt_period_timer,
rt_b->rt_period_timer.expires,
HRTIMER_MODE_ABS);
}
spin_unlock(&rt_b->rt_runtime_lock);
}
#ifdef CONFIG_RT_GROUP_SCHED
static void destroy_rt_bandwidth(struct rt_bandwidth *rt_b)
{
hrtimer_cancel(&rt_b->rt_period_timer);
}
#endif
#ifdef CONFIG_GROUP_SCHED
#include <linux/cgroup.h>
@ -182,7 +261,7 @@ struct task_group {
struct sched_rt_entity **rt_se;
struct rt_rq **rt_rq;
u64 rt_runtime;
struct rt_bandwidth rt_bandwidth;
#endif
struct rcu_head rcu;
@ -407,8 +486,6 @@ struct rq {
struct cfs_rq cfs;
struct rt_rq rt;
u64 rt_period_expire;
int rt_throttled;
#ifdef CONFIG_FAIR_GROUP_SCHED
/* list of leaf cfs_rq on this cpu: */
@ -592,23 +669,6 @@ static void update_rq_clock(struct rq *rq)
#define task_rq(p) cpu_rq(task_cpu(p))
#define cpu_curr(cpu) (cpu_rq(cpu)->curr)
unsigned long rt_needs_cpu(int cpu)
{
struct rq *rq = cpu_rq(cpu);
u64 delta;
if (!rq->rt_throttled)
return 0;
if (rq->clock > rq->rt_period_expire)
return 1;
delta = rq->rt_period_expire - rq->clock;
do_div(delta, NSEC_PER_SEC / HZ);
return (unsigned long)delta;
}
/*
* Tunables that become constants when CONFIG_SCHED_DEBUG is off:
*/
@ -664,10 +724,18 @@ static __read_mostly int scheduler_running;
*/
int sysctl_sched_rt_runtime = 950000;
/*
* single value that denotes runtime == period, ie unlimited time.
*/
#define RUNTIME_INF ((u64)~0ULL)
static inline u64 global_rt_period(void)
{
return (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
}
static inline u64 global_rt_runtime(void)
{
if (sysctl_sched_rt_period < 0)
return RUNTIME_INF;
return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
}
static const unsigned long long time_sync_thresh = 100000;
@ -3854,7 +3922,6 @@ void scheduler_tick(void)
update_last_tick_seen(rq);
update_cpu_load(rq);
curr->sched_class->task_tick(rq, curr, 0);
update_sched_rt_period(rq);
spin_unlock(&rq->lock);
#ifdef CONFIG_SMP
@ -4689,7 +4756,7 @@ int sched_setscheduler(struct task_struct *p, int policy,
* Do not allow realtime tasks into groups that have no runtime
* assigned.
*/
if (rt_policy(policy) && task_group(p)->rt_runtime == 0)
if (rt_policy(policy) && task_group(p)->rt_bandwidth.rt_runtime == 0)
return -EPERM;
#endif
@ -7288,6 +7355,14 @@ void __init sched_init(void)
init_defrootdomain();
#endif
init_rt_bandwidth(&def_rt_bandwidth,
global_rt_period(), global_rt_runtime());
#ifdef CONFIG_RT_GROUP_SCHED
init_rt_bandwidth(&init_task_group.rt_bandwidth,
global_rt_period(), global_rt_runtime());
#endif
#ifdef CONFIG_GROUP_SCHED
list_add(&init_task_group.list, &task_groups);
#endif
@ -7312,15 +7387,11 @@ void __init sched_init(void)
#endif
#ifdef CONFIG_RT_GROUP_SCHED
init_task_group.rt_runtime =
sysctl_sched_rt_runtime * NSEC_PER_USEC;
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
rq->rt_period_expire = 0;
rq->rt_throttled = 0;
for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
rq->cpu_load[j] = 0;
@ -7506,8 +7577,6 @@ void set_curr_task(int cpu, struct task_struct *p)
#endif
#ifdef CONFIG_GROUP_SCHED
#ifdef CONFIG_FAIR_GROUP_SCHED
static void free_fair_sched_group(struct task_group *tg)
{
@ -7596,6 +7665,8 @@ static void free_rt_sched_group(struct task_group *tg)
{
int i;
destroy_rt_bandwidth(&tg->rt_bandwidth);
for_each_possible_cpu(i) {
if (tg->rt_rq)
kfree(tg->rt_rq[i]);
@ -7621,7 +7692,8 @@ static int alloc_rt_sched_group(struct task_group *tg)
if (!tg->rt_se)
goto err;
tg->rt_runtime = 0;
init_rt_bandwidth(&tg->rt_bandwidth,
ktime_to_ns(def_rt_bandwidth.rt_period), 0);
for_each_possible_cpu(i) {
rq = cpu_rq(i);
@ -7674,6 +7746,7 @@ static inline void unregister_rt_sched_group(struct task_group *tg, int cpu)
}
#endif
#ifdef CONFIG_GROUP_SCHED
static void free_sched_group(struct task_group *tg)
{
free_fair_sched_group(tg);
@ -7775,6 +7848,7 @@ void sched_move_task(struct task_struct *tsk)
task_rq_unlock(rq, &flags);
}
#endif
#ifdef CONFIG_FAIR_GROUP_SCHED
static void set_se_shares(struct sched_entity *se, unsigned long shares)
@ -7871,16 +7945,15 @@ static int __rt_schedulable(struct task_group *tg, u64 period, u64 runtime)
struct task_group *tgi;
unsigned long total = 0;
unsigned long global_ratio =
to_ratio(sysctl_sched_rt_period,
sysctl_sched_rt_runtime < 0 ?
RUNTIME_INF : sysctl_sched_rt_runtime);
to_ratio(global_rt_period(), global_rt_runtime());
rcu_read_lock();
list_for_each_entry_rcu(tgi, &task_groups, list) {
if (tgi == tg)
continue;
total += to_ratio(period, tgi->rt_runtime);
total += to_ratio(ktime_to_ns(tgi->rt_bandwidth.rt_period),
tgi->rt_bandwidth.rt_runtime);
}
rcu_read_unlock();
@ -7898,16 +7971,11 @@ static inline int tg_has_rt_tasks(struct task_group *tg)
return 0;
}
int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
static int tg_set_bandwidth(struct task_group *tg,
u64 rt_period, u64 rt_runtime)
{
u64 rt_runtime, rt_period;
int err = 0;
rt_period = (u64)sysctl_sched_rt_period * NSEC_PER_USEC;
rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
if (rt_runtime_us == -1)
rt_runtime = RUNTIME_INF;
mutex_lock(&rt_constraints_mutex);
read_lock(&tasklist_lock);
if (rt_runtime_us == 0 && tg_has_rt_tasks(tg)) {
@ -7918,7 +7986,8 @@ int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
err = -EINVAL;
goto unlock;
}
tg->rt_runtime = rt_runtime;
tg->rt_bandwidth.rt_period = ns_to_ktime(rt_period);
tg->rt_bandwidth.rt_runtime = rt_runtime;
unlock:
read_unlock(&tasklist_lock);
mutex_unlock(&rt_constraints_mutex);
@ -7926,19 +7995,96 @@ int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
return err;
}
int sched_group_set_rt_runtime(struct task_group *tg, long rt_runtime_us)
{
u64 rt_runtime, rt_period;
rt_period = ktime_to_ns(tg->rt_bandwidth.rt_period);
rt_runtime = (u64)rt_runtime_us * NSEC_PER_USEC;
if (rt_runtime_us < 0)
rt_runtime = RUNTIME_INF;
return tg_set_bandwidth(tg, rt_period, rt_runtime);
}
long sched_group_rt_runtime(struct task_group *tg)
{
u64 rt_runtime_us;
if (tg->rt_runtime == RUNTIME_INF)
if (tg->rt_bandwidth.rt_runtime == RUNTIME_INF)
return -1;
rt_runtime_us = tg->rt_runtime;
rt_runtime_us = tg->rt_bandwidth.rt_runtime;
do_div(rt_runtime_us, NSEC_PER_USEC);
return rt_runtime_us;
}
int sched_group_set_rt_period(struct task_group *tg, long rt_period_us)
{
u64 rt_runtime, rt_period;
rt_period = (u64)rt_period_us * NSEC_PER_USEC;
rt_runtime = tg->rt_bandwidth.rt_runtime;
return tg_set_bandwidth(tg, rt_period, rt_runtime);
}
long sched_group_rt_period(struct task_group *tg)
{
u64 rt_period_us;
rt_period_us = ktime_to_ns(tg->rt_bandwidth.rt_period);
do_div(rt_period_us, NSEC_PER_USEC);
return rt_period_us;
}
static int sched_rt_global_constraints(void)
{
int ret = 0;
mutex_lock(&rt_constraints_mutex);
if (!__rt_schedulable(NULL, 1, 0))
ret = -EINVAL;
mutex_unlock(&rt_constraints_mutex);
return ret;
}
#else
static int sched_rt_global_constraints(void)
{
return 0;
}
#endif
#endif /* CONFIG_GROUP_SCHED */
int sched_rt_handler(struct ctl_table *table, int write,
struct file *filp, void __user *buffer, size_t *lenp,
loff_t *ppos)
{
int ret;
int old_period, old_runtime;
static DEFINE_MUTEX(mutex);
mutex_lock(&mutex);
old_period = sysctl_sched_rt_period;
old_runtime = sysctl_sched_rt_runtime;
ret = proc_dointvec(table, write, filp, buffer, lenp, ppos);
if (!ret && write) {
ret = sched_rt_global_constraints();
if (ret) {
sysctl_sched_rt_period = old_period;
sysctl_sched_rt_runtime = old_runtime;
} else {
def_rt_bandwidth.rt_runtime = global_rt_runtime();
def_rt_bandwidth.rt_period =
ns_to_ktime(global_rt_period());
}
}
mutex_unlock(&mutex);
return ret;
}
#ifdef CONFIG_CGROUP_SCHED
@ -7988,7 +8134,7 @@ cpu_cgroup_can_attach(struct cgroup_subsys *ss, struct cgroup *cgrp,
{
#ifdef CONFIG_RT_GROUP_SCHED
/* Don't accept realtime tasks when there is no way for them to run */
if (rt_task(tsk) && cgroup_tg(cgrp)->rt_runtime == 0)
if (rt_task(tsk) && cgroup_tg(cgrp)->rt_bandwidth.rt_runtime == 0)
return -EINVAL;
#else
/* We don't support RT-tasks being in separate groups */
@ -8066,6 +8212,17 @@ static ssize_t cpu_rt_runtime_read(struct cgroup *cgrp, struct cftype *cft,
return simple_read_from_buffer(buf, nbytes, ppos, tmp, len);
}
static int cpu_rt_period_write_uint(struct cgroup *cgrp, struct cftype *cftype,
u64 rt_period_us)
{
return sched_group_set_rt_period(cgroup_tg(cgrp), rt_period_us);
}
static u64 cpu_rt_period_read_uint(struct cgroup *cgrp, struct cftype *cft)
{
return sched_group_rt_period(cgroup_tg(cgrp));
}
#endif
static struct cftype cpu_files[] = {
@ -8082,6 +8239,11 @@ static struct cftype cpu_files[] = {
.read = cpu_rt_runtime_read,
.write = cpu_rt_runtime_write,
},
{
.name = "rt_period_us",
.read_uint = cpu_rt_period_read_uint,
.write_uint = cpu_rt_period_write_uint,
},
#endif
};

View File

@ -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
struct rt_rq *sched_rt_period_rt_rq(struct rt_bandwidth *rt_b, int cpu)
{
return container_of(rt_b, struct task_group, rt_bandwidth)->rt_rq[cpu];
}
#else
static inline u64 sched_rt_runtime(struct rt_rq *rt_rq)
{
if (sysctl_sched_rt_runtime == -1)
return RUNTIME_INF;
return (u64)sysctl_sched_rt_runtime * NSEC_PER_USEC;
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
}

View File

@ -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,

View File

@ -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;
/*

View File

@ -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
};