[RFC PATCH 2/3] sched: introduce compute capacity for CPUs, groupsand domains

From: Chris Redpath
Date: Tue Apr 16 2013 - 11:26:26 EST

Next message: Chris Redpath: "[RFC PATCH 1/3] ARM: (Experimental) Provide Estimated CPU Capacitymeasure"
Previous message: Chris Redpath: "[RFC PATCH 0/3] Per-Task Load Tracking in the presence of DVFS"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]

Using the per-cpu compute capacity exported from topology
when CONFIG_ARCH_SCALE_INVARIANT_CPU_CAPACITY is active, place this
information alongside cpu_power in the scheduler and combine for the
various aggregating entities.

Change-Id: I4984c335bcdc128680e7459b3f86bb05e04593cc
---
include/linux/sched.h | 7 +++++
include/trace/events/sched.h | 24 +++++++++++++++
kernel/sched/core.c | 2 ++
kernel/sched/debug.c | 3 ++
kernel/sched/fair.c | 69 ++++++++++++++++++++++++++++++++++++++----
kernel/sched/sched.h | 4 +++
6 files changed, 103 insertions(+), 6 deletions(-)

diff --git a/include/linux/sched.h b/include/linux/sched.h
index 7c64f30..f2ee59a 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -863,6 +863,13 @@ struct sched_group_power {
unsigned int power, power_orig;
unsigned long next_update;
/*
+ * Compute capacity of this group, where each CPU has a compute
+ * capacity expressed as a value [0..SCHED_POWER_SCALE] against
+ * the most powerful CPU in the system of capacity SCHED_POWER_SCALE.
+ */
+ unsigned int compute_capacity;
+ unsigned int max_compute_capacity;
+ /*
* Number of busy cpus in this group.
*/
atomic_t nr_busy_cpus;
diff --git a/include/trace/events/sched.h b/include/trace/events/sched.h
index 8932919..45e27bc 100644
--- a/include/trace/events/sched.h
+++ b/include/trace/events/sched.h
@@ -985,6 +985,30 @@ TRACE_EVENT(sched_fsi,
);

/*
+ * Extra debug trace points
+ */
+TRACE_EVENT(sched_upd_cap,
+
+ TP_PROTO(int dst_cpu, unsigned long curr, unsigned long max ),
+
+ TP_ARGS(dst_cpu, curr, max ),
+
+ TP_STRUCT__entry(
+ __field(int, dst_cpu)
+ __field(unsigned long, curr)
+ __field(unsigned long, max)
+ ),
+
+ TP_fast_assign(
+ __entry->dst_cpu = dst_cpu;
+ __entry->curr = curr;
+ __entry->max = max;
+ ),
+
+ TP_printk("cpu=%d curr=%lu max=%lu",
+ __entry->dst_cpu, __entry->curr, __entry->max)
+);
+/*
* Tracepoint for showing priority inheritance modifying a tasks
* priority.
*/
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index ec7406d..e535222 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6940,6 +6940,8 @@ void __init sched_init(void)
rq->sd = NULL;
rq->rd = NULL;
rq->cpu_power = SCHED_POWER_SCALE;
+ rq->curr_compute_capacity = SCHED_POWER_SCALE;
+ rq->max_compute_capacity = SCHED_POWER_SCALE;
rq->post_schedule = 0;
rq->active_balance = 0;
rq->next_balance = jiffies;
diff --git a/kernel/sched/debug.c b/kernel/sched/debug.c
index b9d54d0..9102bb4 100644
--- a/kernel/sched/debug.c
+++ b/kernel/sched/debug.c
@@ -290,6 +290,9 @@ do { \
#define PN(x) \
SEQ_printf(m, " .%-30s: %Ld.%06ld\n", #x, SPLIT_NS(rq->x))

+ P(cpu_power);
+ P(curr_compute_capacity);
+ P(max_compute_capacity);
P(nr_running);
SEQ_printf(m, " .%-30s: %lu\n", "load",
rq->load.weight);
diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index d9af9c1..f6bbe1e 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -1267,6 +1267,27 @@ static u32 __compute_runnable_contrib(u64 n)
return contrib + runnable_avg_yN_sum[n];
}

+#ifdef CONFIG_ARCH_SCALE_INVARIANT_CPU_CAPACITY
+#define SCHED_ARCH_SCALE_POWER_SHIFT 10
+#endif
+static inline unsigned long compute_capacity_of(int cpu)
+{
+ return cpu_rq(cpu)->curr_compute_capacity;
+}
+
+static inline unsigned long max_compute_capacity_of(int cpu)
+{
+ return cpu_rq(cpu)->max_compute_capacity;
+}
+
+static inline void update_cpu_capacity(int cpu)
+{
+ int tmp_capacity = arch_get_cpu_capacity(cpu);
+ int tmp_max_capacity = arch_get_max_cpu_capacity(cpu);
+ trace_sched_upd_cap(cpu, tmp_capacity, tmp_max_capacity);
+ cpu_rq(cpu)->max_compute_capacity = tmp_max_capacity;
+ cpu_rq(cpu)->curr_compute_capacity = tmp_capacity;
+}
/*
* We can represent the historical contribution to runnable average as the
* coefficients of a geometric series. To do this we sub-divide our runnable
@@ -4360,6 +4381,8 @@ struct sd_lb_stats {
unsigned long total_load; /* Total load of all groups in sd */
unsigned long total_pwr; /* Total power of all groups in sd */
unsigned long avg_load; /* Average load across all groups in sd */
+ unsigned long total_cap; /* Total current compute capacity of all groups in sd */
+ unsigned long total_maxcap; /* Total max compute capacity of all groups in sd */

/** Statistics of this group */
unsigned long this_load;
@@ -4388,7 +4411,9 @@ struct sg_lb_stats {
unsigned long group_load; /* Total load over the CPUs of the group */
unsigned long sum_nr_running; /* Nr tasks running in the group */
unsigned long sum_weighted_load; /* Weighted load of group's tasks */
- unsigned long group_capacity;
+ unsigned long group_compute_capacity; /* current compute capacity of the group */
+ unsigned long group_max_compute_capacity; /* maximum compute capacity of the group */
+ unsigned long group_capacity; /* Nr tasks this group can handle before considered overloaded */
unsigned long idle_cpus;
unsigned long group_weight;
int group_imb; /* Is there an imbalance in the group ? */
@@ -4430,6 +4455,23 @@ unsigned long __weak arch_scale_freq_power(struct sched_domain *sd, int cpu)
{
return default_scale_freq_power(sd, cpu);
}
+unsigned long __weak arch_cpu_capacity(int cpu)
+{
+ return SCHED_POWER_SCALE;
+}
+unsigned long __weak arch_max_cpu_capacity(int cpu)
+{
+ return SCHED_POWER_SCALE;
+}
+
+unsigned long __weak arch_get_cpu_capacity(int cpu)
+{
+ return SCHED_POWER_SCALE;
+}
+unsigned long __weak arch_get_max_cpu_capacity(int cpu)
+{
+ return SCHED_POWER_SCALE;
+}

unsigned long default_scale_smt_power(struct sched_domain *sd, int cpu)
{
@@ -4506,6 +4548,7 @@ static void update_cpu_power(struct sched_domain *sd, int cpu)
power = 1;

cpu_rq(cpu)->cpu_power = power;
+ update_cpu_capacity(cpu);
sdg->sgp->power = power;
}

@@ -4514,6 +4557,7 @@ void update_group_power(struct sched_domain *sd, int cpu)
struct sched_domain *child = sd->child;
struct sched_group *group, *sdg = sd->groups;
unsigned long power;
+ unsigned long compute_capacity, max_compute_capacity;
unsigned long interval;

interval = msecs_to_jiffies(sd->balance_interval);
@@ -4526,6 +4570,8 @@ void update_group_power(struct sched_domain *sd, int cpu)
}

power = 0;
+ compute_capacity = 0;
+ max_compute_capacity = 0;

if (child->flags & SD_OVERLAP) {
/*
@@ -4533,8 +4579,11 @@ void update_group_power(struct sched_domain *sd, int cpu)
* span the current group.
*/

- for_each_cpu(cpu, sched_group_cpus(sdg))
+ for_each_cpu(cpu, sched_group_cpus(sdg)) {
power += power_of(cpu);
+ compute_capacity += compute_capacity_of(cpu);
+ max_compute_capacity += max_compute_capacity_of(cpu);
+ }
} else {
/*
* !SD_OVERLAP domains can assume that child groups
@@ -4544,11 +4593,15 @@ void update_group_power(struct sched_domain *sd, int cpu)
group = child->groups;
do {
power += group->sgp->power;
+ compute_capacity += group->sgp->compute_capacity;
+ max_compute_capacity += group->sgp->max_compute_capacity;
group = group->next;
} while (group != child->groups);
}

sdg->sgp->power_orig = sdg->sgp->power = power;
+ sdg->sgp->compute_capacity = compute_capacity;
+ sdg->sgp->max_compute_capacity = max_compute_capacity;
}

/*
@@ -4639,6 +4692,8 @@ static inline void update_sg_lb_stats(struct lb_env *env,
sgs->group_load += load;
sgs->sum_nr_running += nr_running;
sgs->sum_weighted_load += weighted_cpuload(i);
+ sgs->group_compute_capacity += compute_capacity_of(i);
+ sgs->group_max_compute_capacity += max_compute_capacity_of(i);
if (idle_cpu(i))
sgs->idle_cpus++;
}
@@ -4774,6 +4829,8 @@ static inline void update_sd_lb_stats(struct lb_env *env,

sds->total_load += sgs.group_load;
sds->total_pwr += sg->sgp->power;
+ sds->total_cap += sg->sgp->compute_capacity;
+ sds->total_maxcap += sg->sgp->compute_capacity;

/*
* In case the child domain prefers tasks go to siblings
@@ -5122,12 +5179,12 @@ static struct rq *find_busiest_queue(struct lb_env *env,

for_each_cpu(i, sched_group_cpus(group)) {
unsigned long power = power_of(i);
- unsigned long capacity = DIV_ROUND_CLOSEST(power,
+ unsigned long task_capacity = DIV_ROUND_CLOSEST(power,
SCHED_POWER_SCALE);
unsigned long wl;

- if (!capacity)
- capacity = fix_small_capacity(env->sd, group);
+ if (!task_capacity)
+ task_capacity = fix_small_capacity(env->sd, group);

if (!cpumask_test_cpu(i, env->cpus))
continue;
@@ -5151,7 +5208,7 @@ static struct rq *find_busiest_queue(struct lb_env *env,
* When comparing with imbalance, use weighted_cpuload()
* which is not scaled with the cpu power.
*/
- if (capacity && rq->nr_running == 1 && wl > env->imbalance)
+ if (task_capacity && rq->nr_running == 1 && wl > env->imbalance)
continue;

/*
diff --git a/kernel/sched/sched.h b/kernel/sched/sched.h
index 6f8976b..0946f40 100644
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -412,6 +412,10 @@ struct rq {

unsigned long cpu_power;

+ /* CPU compute capacity estimation */
+ unsigned long max_compute_capacity;
+ unsigned long curr_compute_capacity;
+
unsigned char idle_balance;
/* For active balancing */
int post_schedule;
--
1.7.9.5

--
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/

Next message: Chris Redpath: "[RFC PATCH 1/3] ARM: (Experimental) Provide Estimated CPU Capacitymeasure"
Previous message: Chris Redpath: "[RFC PATCH 0/3] Per-Task Load Tracking in the presence of DVFS"
Messages sorted by: [ date ] [ thread ] [ subject ] [ author ]