1 files changed, 43 insertions, 10 deletions

diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index 39eb6011bc38..75844a8f9aeb 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -2517,25 +2517,32 @@ static void __update_cpu_load(struct rq *this_rq, unsigned long this_load,
         sched_avg_update(this_rq);
 }
 
+#ifdef CONFIG_NO_HZ
+/*
+ * There is no sane way to deal with nohz on smp when using jiffies because the
+ * cpu doing the jiffies update might drift wrt the cpu doing the jiffy reading
+ * causing off-by-one errors in observed deltas; {0,2} instead of {1,1}.
+ *
+ * Therefore we cannot use the delta approach from the regular tick since that
+ * would seriously skew the load calculation. However we'll make do for those
+ * updates happening while idle (nohz_idle_balance) or coming out of idle
+ * (tick_nohz_idle_exit).
+ *
+ * This means we might still be one tick off for nohz periods.
+ */
+
 /*
  * Called from nohz_idle_balance() to update the load ratings before doing the
  * idle balance.
  */
 void update_idle_cpu_load(struct rq *this_rq)
 {
-        unsigned long curr_jiffies = jiffies;
+        unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
         unsigned long load = this_rq->load.weight;
         unsigned long pending_updates;
 
         /*
-         * Bloody broken means of dealing with nohz, but better than nothing..
-         * jiffies is updated by one cpu, another cpu can drift wrt the jiffy
-         * update and see 0 difference the one time and 2 the next, even though
-         * we ticked at roughtly the same rate.
-         *
-         * Hence we only use this from nohz_idle_balance() and skip this
-         * nonsense when called from the scheduler_tick() since that's
-         * guaranteed a stable rate.
+         * bail if there's load or we're actually up-to-date.
          */
         if (load || curr_jiffies == this_rq->last_load_update_tick)
                 return;
@@ -2547,12 +2554,38 @@ void update_idle_cpu_load(struct rq *this_rq)
 }
 
 /*
+ * Called from tick_nohz_idle_exit() -- try and fix up the ticks we missed.
+ */
+void update_cpu_load_nohz(void)
+{
+        struct rq *this_rq = this_rq();
+        unsigned long curr_jiffies = ACCESS_ONCE(jiffies);
+        unsigned long pending_updates;
+
+        if (curr_jiffies == this_rq->last_load_update_tick)
+                return;
+
+        raw_spin_lock(&this_rq->lock);
+        pending_updates = curr_jiffies - this_rq->last_load_update_tick;
+        if (pending_updates) {
+                this_rq->last_load_update_tick = curr_jiffies;
+                /*
+                 * We were idle, this means load 0, the current load might be
+                 * !0 due to remote wakeups and the sort.
+                 */
+                __update_cpu_load(this_rq, 0, pending_updates);
+        }
+        raw_spin_unlock(&this_rq->lock);
+}
+#endif /* CONFIG_NO_HZ */
+
+/*
  * Called from scheduler_tick()
  */
 static void update_cpu_load_active(struct rq *this_rq)
 {
         /*
-         * See the mess in update_idle_cpu_load().
+         * See the mess around update_idle_cpu_load() / update_cpu_load_nohz().
          */
         this_rq->last_load_update_tick = jiffies;
         __update_cpu_load(this_rq, this_rq->load.weight, 1);