[PATCH 1/2] sched: rt throttling vs no_hz

From: Peter Zijlstra
Date: Thu Jan 03 2008 - 14:14:31 EST


We need to teach no_hz about the rt throttling because its tick driven.

Signed-off-by: Peter Zijlstra <a.p.zijlstra@xxxxxxxxx>
---
include/linux/sched.h | 2 ++
kernel/sched.c | 23 ++++++++++++++++++++++-
kernel/sched_rt.c | 30 ++++++++++++++++--------------
kernel/time/tick-sched.c | 5 +++++
4 files changed, 45 insertions(+), 15 deletions(-)

Index: linux-2.6/include/linux/sched.h
===================================================================
--- linux-2.6.orig/include/linux/sched.h
+++ linux-2.6/include/linux/sched.h
@@ -230,6 +230,8 @@ static inline int select_nohz_load_balan
}
#endif

+extern unsigned long rt_needs_cpu(int cpu);
+
/*
* Only dump TASK_* tasks. (0 for all tasks)
*/
Index: linux-2.6/kernel/sched.c
===================================================================
--- linux-2.6.orig/kernel/sched.c
+++ linux-2.6/kernel/sched.c
@@ -442,6 +442,7 @@ 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: */
@@ -594,6 +595,23 @@ static void update_rq_clock(struct rq *r
#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:
*/
@@ -7099,9 +7117,11 @@ static void init_rt_rq(struct rt_rq *rt_
/* delimiter for bitsearch: */
__set_bit(MAX_RT_PRIO, array->bitmap);

+#if defined CONFIG_SMP || defined CONFIG_FAIR_GROUP_SCHED
+ rt_rq->highest_prio = MAX_RT_PRIO;
+#endif
#ifdef CONFIG_SMP
rt_rq->rt_nr_migratory = 0;
- rt_rq->highest_prio = MAX_RT_PRIO;
rt_rq->overloaded = 0;
#endif

@@ -7186,6 +7206,7 @@ void __init sched_init(void)
list_add(&init_task_group.list, &task_groups);
#endif
rq->rt_period_expire = 0;
+ rq->rt_throttled = 0;

for (j = 0; j < CPU_LOAD_IDX_MAX; j++)
rq->cpu_load[j] = 0;
Index: linux-2.6/kernel/sched_rt.c
===================================================================
--- linux-2.6.orig/kernel/sched_rt.c
+++ linux-2.6/kernel/sched_rt.c
@@ -175,7 +175,11 @@ static int sched_rt_ratio_exceeded(struc
ratio = (period * rt_ratio) >> SCHED_RT_FRAC_SHIFT;

if (rt_rq->rt_time > ratio) {
+ struct rq *rq = rq_of_rt_rq(rt_rq);
+
+ rq->rt_throttled = 1;
rt_rq->rt_throttled = 1;
+
sched_rt_ratio_dequeue(rt_rq);
return 1;
}
@@ -183,18 +187,6 @@ static int sched_rt_ratio_exceeded(struc
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)
{
struct rt_rq *rt_rq;
@@ -204,8 +196,18 @@ static void update_sched_rt_period(struc
period = (u64)sysctl_sched_rt_period * NSEC_PER_MSEC;
rq->rt_period_expire += period;

- for_each_leaf_rt_rq(rt_rq, rq)
- __update_sched_rt_period(rt_rq, period);
+ for_each_leaf_rt_rq(rt_rq, rq) {
+ 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);
+ }
+ }
+
+ rq->rt_throttled = 0;
}
}

Index: linux-2.6/kernel/time/tick-sched.c
===================================================================
--- linux-2.6.orig/kernel/time/tick-sched.c
+++ linux-2.6/kernel/time/tick-sched.c
@@ -153,6 +153,7 @@ void tick_nohz_update_jiffies(void)
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, delta;
struct clock_event_device *dev = __get_cpu_var(tick_cpu_device).evtdev;
@@ -216,6 +217,10 @@ 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;
/*

--

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/