timer: Minimize nohz off overhead

If nohz is disabled on the kernel command line the [hr]timer code
still calls wake_up_nohz_cpu() and tick_nohz_full_cpu(), a pretty
pointless exercise. Cache nohz_active in [hr]timer per cpu bases and
avoid the overhead.

Before:
  48.10%  hog       [.] main
  15.25%  [kernel]  [k] _raw_spin_lock_irqsave
   9.76%  [kernel]  [k] _raw_spin_unlock_irqrestore
   6.50%  [kernel]  [k] mod_timer
   6.44%  [kernel]  [k] lock_timer_base.isra.38
   3.87%  [kernel]  [k] detach_if_pending
   3.80%  [kernel]  [k] del_timer
   2.67%  [kernel]  [k] internal_add_timer
   1.33%  [kernel]  [k] __internal_add_timer
   0.73%  [kernel]  [k] timerfn
   0.54%  [kernel]  [k] wake_up_nohz_cpu

After:
  48.73%  hog       [.] main
  15.36%  [kernel]  [k] _raw_spin_lock_irqsave
   9.77%  [kernel]  [k] _raw_spin_unlock_irqrestore
   6.61%  [kernel]  [k] lock_timer_base.isra.38
   6.42%  [kernel]  [k] mod_timer
   3.90%  [kernel]  [k] detach_if_pending
   3.76%  [kernel]  [k] del_timer
   2.41%  [kernel]  [k] internal_add_timer
   1.39%  [kernel]  [k] __internal_add_timer
   0.76%  [kernel]  [k] timerfn

We probably should have a cached value for nohz full in the per cpu
bases as well to avoid the cpumask check. The base cache line is hot
already, the cpumask not necessarily.

Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Paul McKenney <paulmck@linux.vnet.ibm.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Eric Dumazet <edumazet@google.com>
Cc: Viresh Kumar <viresh.kumar@linaro.org>
Cc: John Stultz <john.stultz@linaro.org>
Cc: Joonwoo Park <joonwoop@codeaurora.org>
Cc: Wenbo Wang <wenbo.wang@memblaze.com>
Link: http://lkml.kernel.org/r/20150526224512.207378134@linutronix.de
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

5 files changed, 19 insertions, 8 deletions

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 6955102..76dd4f0 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -164,6 +164,7 @@ enum  hrtimer_base_type {
  * @active_bases:	Bitfield to mark bases with active timers
  * @clock_was_set_seq:	Sequence counter of clock was set events
  * @migration_enabled:	The migration of hrtimers to other cpus is enabled
+ * @nohz_active:	The nohz functionality is enabled
  * @expires_next:	absolute time of the next event which was scheduled
  *			via clock_set_next_event()
  * @next_timer:		Pointer to the first expiring timer
@@ -188,6 +189,7 @@ struct hrtimer_cpu_base {
 	unsigned int			active_bases;
 	unsigned int			clock_was_set_seq;
 	bool				migration_enabled;
+	bool				nohz_active;
 #ifdef CONFIG_HIGH_RES_TIMERS
 	unsigned int			in_hrtirq	: 1,
 					hres_active	: 1,
diff --git a/kernel/time/hrtimer.c b/kernel/time/hrtimer.c
index 6115f4d..db5c950 100644
--- a/kernel/time/hrtimer.c
+++ b/kernel/time/hrtimer.c
@@ -994,7 +994,8 @@ void hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
 		 * Kick to reschedule the next tick to handle the new timer
 		 * on dynticks target.
 		 */
-		wake_up_nohz_cpu(new_base->cpu_base->cpu);
+		if (new_base->cpu_base->nohz_active)
+			wake_up_nohz_cpu(new_base->cpu_base->cpu);
 	} else {
 		hrtimer_reprogram(timer, new_base);
 	}
diff --git a/kernel/time/tick-internal.h b/kernel/time/tick-internal.h
index 2edde84..966a5a6 100644
--- a/kernel/time/tick-internal.h
+++ b/kernel/time/tick-internal.h
@@ -156,9 +156,9 @@ extern unsigned long tick_nohz_active;
 #endif
 
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
-extern void timers_update_migration(void);
+extern void timers_update_migration(bool update_nohz);
 #else
-static inline void timers_update_migration(void) { }
+static inline void timers_update_migration(bool update_nohz) { }
 #endif
 
 DECLARE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases);
diff --git a/kernel/time/tick-sched.c b/kernel/time/tick-sched.c
index b1cb016..c792429 100644
--- a/kernel/time/tick-sched.c
+++ b/kernel/time/tick-sched.c
@@ -963,7 +963,7 @@ static inline void tick_nohz_activate(struct tick_sched *ts, int mode)
 	ts->nohz_mode = mode;
 	/* One update is enough */
 	if (!test_and_set_bit(0, &tick_nohz_active))
-		timers_update_migration();
+		timers_update_migration(true);
 }
 
 /**
diff --git a/kernel/time/timer.c b/kernel/time/timer.c
index 343142e..520499d 100644
--- a/kernel/time/timer.c
+++ b/kernel/time/timer.c
@@ -86,6 +86,7 @@ struct tvec_base {
 	unsigned long all_timers;
 	int cpu;
 	bool migration_enabled;
+	bool nohz_active;
 	struct tvec_root tv1;
 	struct tvec tv2;
 	struct tvec tv3;
@@ -99,7 +100,7 @@ static DEFINE_PER_CPU(struct tvec_base, tvec_bases);
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 unsigned int sysctl_timer_migration = 1;
 
-void timers_update_migration(void)
+void timers_update_migration(bool update_nohz)
 {
 	bool on = sysctl_timer_migration && tick_nohz_active;
 	unsigned int cpu;
@@ -111,6 +112,10 @@ void timers_update_migration(void)
 	for_each_possible_cpu(cpu) {
 		per_cpu(tvec_bases.migration_enabled, cpu) = on;
 		per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
+		if (!update_nohz)
+			continue;
+		per_cpu(tvec_bases.nohz_active, cpu) = true;
+		per_cpu(hrtimer_bases.nohz_active, cpu) = true;
 	}
 }
 
@@ -124,7 +129,7 @@ int timer_migration_handler(struct ctl_table *table, int write,
 	mutex_lock(&mutex);
 	ret = proc_dointvec(table, write, buffer, lenp, ppos);
 	if (!ret && write)
-		timers_update_migration();
+		timers_update_migration(false);
 	mutex_unlock(&mutex);
 	return ret;
 }
@@ -436,8 +441,11 @@ static void internal_add_timer(struct tvec_base *base, struct timer_list *timer)
 	 * require special care against races with idle_cpu(), lets deal
 	 * with that later.
 	 */
-	if (!(timer->flags & TIMER_DEFERRABLE) || tick_nohz_full_cpu(base->cpu))
-		wake_up_nohz_cpu(base->cpu);
+	if (base->nohz_active) {
+		if (!(timer->flags & TIMER_DEFERRABLE) ||
+		    tick_nohz_full_cpu(base->cpu))
+			wake_up_nohz_cpu(base->cpu);
+	}
 }
 
 #ifdef CONFIG_TIMER_STATS