Silly microoptimisation just because we can, along with explanation.

-ck

---
 kernel/sched/bfs.c |    9 +++++----
 1 file changed, 5 insertions(+), 4 deletions(-)

Index: linux-4.7-ck4/kernel/sched/bfs.c
===================================================================
--- linux-4.7-ck4.orig/kernel/sched/bfs.c	2016-09-13 17:21:53.538220126 +1000
+++ linux-4.7-ck4/kernel/sched/bfs.c	2016-09-13 17:21:53.536220171 +1000
@@ -1009,8 +1009,9 @@ static int effective_prio(struct task_st
 }
 
 /*
- * Update the load average for feeding into cpu frequency governors. Use a rolling
- * average with ~ time constant of 32ms
+ * Update the load average for feeding into cpu frequency governors. Use a
+ * rough estimate of a rolling average with ~ time constant of 32ms.
+ * 80/128 ~ 0.63. * 80 / 32768 / 128 == * 5 / 262144
  */
 static void update_load_avg(struct rq *rq)
 {
@@ -1019,10 +1020,10 @@ static void update_load_avg(struct rq *r
 		unsigned long us_interval = (rq->clock - rq->load_update) >> 10;
 		long load;
 
-		load = rq->load_avg - (rq->load_avg * us_interval * 80 / 32768 / 128);
+		load = rq->load_avg - (rq->load_avg * us_interval * 5 / 262144);
 		if (unlikely(load < 0))
 			load = 0;
-		load += rq->soft_affined * rq_load_avg(rq) * us_interval * 80 / 32768 / 128;
+		load += rq->soft_affined * rq_load_avg(rq) * us_interval * 5 / 262144;
 		rq->load_avg = load;
 	}
 	rq->load_update = rq->clock;