[PATCH v2 2/2] timers/nohz: Update nohz load in remote tick

[Scott Wood]

From: "Peter Zijlstra (Intel)" <peterz@xxxxxxxxxxxxx>

The way loadavg is tracked during nohz only pays attention to the load
upon entering nohz.  This can be particularly noticeable if full nohz is
entered while non-idle, and then the cpu goes idle and stays that way for
a long time.

Use the remote tick to ensure that full nohz cpus report their deltas
within a reasonable time.

Signed-off-by: Peter Zijlstra (Intel) <peterz@xxxxxxxxxxxxx>
[swood: added changelog and removed recheck of stopped tick]
Signed-off-by: Scott Wood <swood@xxxxxxxxxx>
---
 include/linux/sched/nohz.h |  2 ++
 kernel/sched/core.c        |  4 +++-
 kernel/sched/loadavg.c     | 33 +++++++++++++++++++++++----------
 3 files changed, 28 insertions(+), 11 deletions(-)

diff --git a/include/linux/sched/nohz.h b/include/linux/sched/nohz.h
index 1abe91ff6e4a..6d67e9a5af6b 100644
--- a/include/linux/sched/nohz.h
+++ b/include/linux/sched/nohz.h
@@ -15,9 +15,11 @@ static inline void nohz_balance_enter_idle(int cpu) { }
 
 #ifdef CONFIG_NO_HZ_COMMON
 void calc_load_nohz_start(void);
+void calc_load_nohz_remote(struct rq *rq);
 void calc_load_nohz_stop(void);
 #else
 static inline void calc_load_nohz_start(void) { }
+static inline void calc_load_nohz_remote(struct rq *rq) { }
 static inline void calc_load_nohz_stop(void) { }
 #endif /* CONFIG_NO_HZ_COMMON */
 
diff --git a/kernel/sched/core.c b/kernel/sched/core.c
index dfb8ea801700..2e4a505e48af 100644
--- a/kernel/sched/core.c
+++ b/kernel/sched/core.c
@@ -3676,6 +3676,7 @@ static void sched_tick_remote(struct work_struct *work)
 	if (cpu_is_offline(cpu))
 		goto out_unlock;
 
+	curr = rq->curr;
 	update_rq_clock(rq);
 
 	if (!is_idle_task(curr)) {
@@ -3688,10 +3689,11 @@ static void sched_tick_remote(struct work_struct *work)
 	}
 	curr->sched_class->task_tick(rq, curr, 0);
 
+	calc_load_nohz_remote(rq);
 out_unlock:
 	rq_unlock_irq(rq, &rf);
-
 out_requeue:
+
 	/*
 	 * Run the remote tick once per second (1Hz). This arbitrary
 	 * frequency is large enough to avoid overload but short enough
diff --git a/kernel/sched/loadavg.c b/kernel/sched/loadavg.c
index 28a516575c18..de22da666ac7 100644
--- a/kernel/sched/loadavg.c
+++ b/kernel/sched/loadavg.c
@@ -231,16 +231,11 @@ static inline int calc_load_read_idx(void)
 	return calc_load_idx & 1;
 }
 
-void calc_load_nohz_start(void)
+static void calc_load_nohz_fold(struct rq *rq)
 {
-	struct rq *this_rq = this_rq();
 	long delta;
 
-	/*
-	 * We're going into NO_HZ mode, if there's any pending delta, fold it
-	 * into the pending NO_HZ delta.
-	 */
-	delta = calc_load_fold_active(this_rq, 0);
+	delta = calc_load_fold_active(rq, 0);
 	if (delta) {
 		int idx = calc_load_write_idx();
 
@@ -248,6 +243,24 @@ void calc_load_nohz_start(void)
 	}
 }
 
+void calc_load_nohz_start(void)
+{
+	/*
+	 * We're going into NO_HZ mode, if there's any pending delta, fold it
+	 * into the pending NO_HZ delta.
+	 */
+	calc_load_nohz_fold(this_rq());
+}
+
+/*
+ * Keep track of the load for NOHZ_FULL, must be called between
+ * calc_load_nohz_{start,stop}().
+ */
+void calc_load_nohz_remote(struct rq *rq)
+{
+	calc_load_nohz_fold(rq);
+}
+
 void calc_load_nohz_stop(void)
 {
 	struct rq *this_rq = this_rq();
@@ -268,7 +281,7 @@ void calc_load_nohz_stop(void)
 		this_rq->calc_load_update += LOAD_FREQ;
 }
 
-static long calc_load_nohz_fold(void)
+static long calc_load_nohz_read(void)
 {
 	int idx = calc_load_read_idx();
 	long delta = 0;
@@ -323,7 +336,7 @@ static void calc_global_nohz(void)
 }
 #else /* !CONFIG_NO_HZ_COMMON */
 
-static inline long calc_load_nohz_fold(void) { return 0; }
+static inline long calc_load_nohz_read(void) { return 0; }
 static inline void calc_global_nohz(void) { }
 
 #endif /* CONFIG_NO_HZ_COMMON */
@@ -346,7 +359,7 @@ void calc_global_load(unsigned long ticks)
 	/*
 	 * Fold the 'old' NO_HZ-delta to include all NO_HZ CPUs.
 	 */
-	delta = calc_load_nohz_fold();
+	delta = calc_load_nohz_read();
 	if (delta)
 		atomic_long_add(delta, &calc_load_tasks);
 
-- 
1.8.3.1