[RFC PATCH 5/5] sched: Update blocked load from NEWIDLE

[Peter Zijlstra]

Suggested-by: Brendan Jackman <brendan.jackman@xxxxxxx>
Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
---
 kernel/sched/core.c  |    1 +
 kernel/sched/fair.c  |   47 +++++++++++++++++++++++++++++++++++++++++------
 kernel/sched/sched.h |    1 +
 3 files changed, 43 insertions(+), 6 deletions(-)

--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -6002,6 +6002,7 @@ void __init sched_init(void)
 		rq_attach_root(rq, &def_root_domain);
 #ifdef CONFIG_NO_HZ_COMMON
 		rq->last_load_update_tick = jiffies;
+		rq->last_blocked_load_update_tick = jiffies;
 		atomic_set(&rq->nohz_flags, 0);
 #endif
 #ifdef CONFIG_NO_HZ_FULL
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -5379,6 +5379,14 @@ decay_load_missed(unsigned long load, un
 	}
 	return load;
 }
+
+static struct {
+	cpumask_var_t idle_cpus_mask;
+	atomic_t nr_cpus;
+	unsigned long next_balance;     /* in jiffy units */
+	unsigned long next_stats;
+} nohz ____cacheline_aligned;
+
 #endif /* CONFIG_NO_HZ_COMMON */
 
 /**
@@ -6942,6 +6950,7 @@ enum fbq_type { regular, remote, all };
 #define LBF_NEED_BREAK	0x02
 #define LBF_DST_PINNED  0x04
 #define LBF_SOME_PINNED	0x08
+#define LBF_NOHZ_STATS	0x10
 
 struct lb_env {
 	struct sched_domain	*sd;
@@ -7380,6 +7389,10 @@ static void update_blocked_averages(int
 		if (cfs_rq_is_decayed(cfs_rq))
 			list_del_leaf_cfs_rq(cfs_rq);
 	}
+
+#ifdef CONFIG_NO_HZ_COMMON
+	rq->last_blocked_load_update_tick = jiffies;
+#endif
 	rq_unlock_irqrestore(rq, &rf);
 }
 
@@ -7439,6 +7452,9 @@ static inline void update_blocked_averag
 	rq_lock_irqsave(rq, &rf);
 	update_rq_clock(rq);
 	update_cfs_rq_load_avg(cfs_rq_clock_task(cfs_rq), cfs_rq);
+#ifdef CONFIG_NO_HZ_COMMON
+	rq->last_blocked_load_update_tick = jiffies;
+#endif
 	rq_unlock_irqrestore(rq, &rf);
 }
 
@@ -7773,6 +7789,19 @@ group_type group_classify(struct sched_g
 	return group_other;
 }
 
+static void update_nohz_stats(struct rq *rq)
+{
+	unsigned int cpu = rq->cpu;
+
+	if (!cpumask_test_cpu(cpu, nohz.idle_cpus_mask))
+		return;
+
+	if (!time_after(jiffies, rq->last_blocked_load_update_tick))
+		return;
+
+	update_blocked_averages(cpu);
+}
+
 /**
  * update_sg_lb_stats - Update sched_group's statistics for load balancing.
  * @env: The load balancing environment.
@@ -7795,6 +7824,9 @@ static inline void update_sg_lb_stats(st
 	for_each_cpu_and(i, sched_group_span(group), env->cpus) {
 		struct rq *rq = cpu_rq(i);
 
+		if (env->flags & LBF_NOHZ_STATS)
+			update_nohz_stats(rq);
+
 		/* Bias balancing toward cpus of our domain */
 		if (local_group)
 			load = target_load(i, load_idx);
@@ -7950,6 +7982,15 @@ static inline void update_sd_lb_stats(st
 	if (child && child->flags & SD_PREFER_SIBLING)
 		prefer_sibling = 1;
 
+#ifdef CONFIG_NO_HZ_COMMON
+	if (env->idle == CPU_NEWLY_IDLE) {
+		env->flags |= LBF_NOHZ_STATS;
+
+		if (cpumask_subset(nohz.idle_cpus_mask, sched_domain_span(env->sd)))
+			nohz.next_stats = jiffies + msecs_to_jiffies(LOAD_AVG_PERIOD);
+	}
+#endif
+
 	load_idx = get_sd_load_idx(env->sd, env->idle);
 
 	do {
@@ -8968,12 +9009,6 @@ static inline int on_null_domain(struct
  *   needed, they will kick the idle load balancer, which then does idle
  *   load balancing for all the idle CPUs.
  */
-static struct {
-	cpumask_var_t idle_cpus_mask;
-	atomic_t nr_cpus;
-	unsigned long next_balance;     /* in jiffy units */
-	unsigned long next_stats;
-} nohz ____cacheline_aligned;
 
 static inline int find_new_ilb(void)
 {
--- a/kernel/sched/sched.h
+++ b/kernel/sched/sched.h
@@ -722,6 +722,7 @@ struct rq {
 #ifdef CONFIG_NO_HZ_COMMON
 #ifdef CONFIG_SMP
 	unsigned long last_load_update_tick;
+	unsigned long last_blocked_load_update_tick;
 #endif /* CONFIG_SMP */
 	atomic_t nohz_flags;
 #endif /* CONFIG_NO_HZ_COMMON */