timer: Reduce timer migration overhead if disabled

Eric reported that the timer_migration sysctl is not really nice
performance wise as it needs to check at every timer insertion whether
the feature is enabled or not. Further the check does not live in the
timer code, so we have an extra function call which checks an extra
cache line to figure out that it is disabled.

We can do better and store that information in the per cpu (hr)timer
bases. I pondered to use a static key, but that's a nightmare to
update from the nohz code and the timer base cache line is hot anyway
when we select a timer base.

The old logic enabled the timer migration unconditionally if
CONFIG_NO_HZ was set even if nohz was disabled on the kernel command
line.

With this modification, we start off with migration disabled. The user
visible sysctl is still set to enabled. If the kernel switches to NOHZ
migration is enabled, if the user did not disable it via the sysctl
prior to the switch. If nohz=off is on the kernel command line,
migration stays disabled no matter what.

Before:
  47.76%  hog       [.] main
  14.84%  [kernel]  [k] _raw_spin_lock_irqsave
   9.55%  [kernel]  [k] _raw_spin_unlock_irqrestore
   6.71%  [kernel]  [k] mod_timer
   6.24%  [kernel]  [k] lock_timer_base.isra.38
   3.76%  [kernel]  [k] detach_if_pending
   3.71%  [kernel]  [k] del_timer
   2.50%  [kernel]  [k] internal_add_timer
   1.51%  [kernel]  [k] get_nohz_timer_target
   1.28%  [kernel]  [k] __internal_add_timer
   0.78%  [kernel]  [k] timerfn
   0.48%  [kernel]  [k] wake_up_nohz_cpu

After:
  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
Reported-by: Eric Dumazet <edumazet@google.com>
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: 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/20150526224512.127050787@linutronix.deSigned-off-by: Thomas Gleixner <tglx@linutronix.de>

include/linux/hrtimer.h

@@ -163,6 +163,7 @@ enum  hrtimer_base_type {
  * @cpu:		cpu number
  * @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
  * @expires_next:	absolute time of the next event which was scheduled
  *			via clock_set_next_event()
  * @next_timer:		Pointer to the first expiring timer
@@ -186,6 +187,7 @@ struct hrtimer_cpu_base {
 	unsigned int			cpu;
 	unsigned int			active_bases;
 	unsigned int			clock_was_set_seq;
+	bool				migration_enabled;
 #ifdef CONFIG_HIGH_RES_TIMERS
 	unsigned int			in_hrtirq	: 1,
 					hres_active	: 1,

include/linux/sched.h

@@ -335,14 +335,10 @@ extern int runqueue_is_locked(int cpu);
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
 extern void nohz_balance_enter_idle(int cpu);
 extern void set_cpu_sd_state_idle(void);
-extern int get_nohz_timer_target(int pinned);
+extern int get_nohz_timer_target(void);
 #else
 static inline void nohz_balance_enter_idle(int cpu) { }
 static inline void set_cpu_sd_state_idle(void) { }
-static inline int get_nohz_timer_target(int pinned)
-{
-	return smp_processor_id();
-}
 #endif
 
 /*

include/linux/sched/sysctl.h

@@ -57,24 +57,12 @@ extern unsigned int sysctl_numa_balancing_scan_size;
 extern unsigned int sysctl_sched_migration_cost;
 extern unsigned int sysctl_sched_nr_migrate;
 extern unsigned int sysctl_sched_time_avg;
-extern unsigned int sysctl_timer_migration;
 extern unsigned int sysctl_sched_shares_window;
 
 int sched_proc_update_handler(struct ctl_table *table, int write,
 		void __user *buffer, size_t *length,
 		loff_t *ppos);
 #endif
-#ifdef CONFIG_SCHED_DEBUG
-static inline unsigned int get_sysctl_timer_migration(void)
-{
-	return sysctl_timer_migration;
-}
-#else
-static inline unsigned int get_sysctl_timer_migration(void)
-{
-	return 1;
-}
-#endif
 
 /*
  *  control realtime throttling:

include/linux/timer.h

@@ -238,6 +238,15 @@ extern void run_local_timers(void);
 struct hrtimer;
 extern enum hrtimer_restart it_real_fn(struct hrtimer *);
 
+#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
+#include <linux/sysctl.h>
+
+extern unsigned int sysctl_timer_migration;
+int timer_migration_handler(struct ctl_table *table, int write,
+			    void __user *buffer, size_t *lenp,
+			    loff_t *ppos);
+#endif
+
 unsigned long __round_jiffies(unsigned long j, int cpu);
 unsigned long __round_jiffies_relative(unsigned long j, int cpu);
 unsigned long round_jiffies(unsigned long j);

kernel/rcu/tree_plugin.h

@@ -1432,8 +1432,6 @@ module_param(rcu_idle_gp_delay, int, 0644);
 static int rcu_idle_lazy_gp_delay = RCU_IDLE_LAZY_GP_DELAY;
 module_param(rcu_idle_lazy_gp_delay, int, 0644);
 
-extern int tick_nohz_active;
-
 /*
  * Try to advance callbacks for all flavors of RCU on the current CPU, but
  * only if it has been awhile since the last time we did so.  Afterwards,

kernel/sched/core.c

@@ -572,13 +572,12 @@ void resched_cpu(int cpu)
  * selecting an idle cpu will add more delays to the timers than intended
  * (as that cpu's timer base may not be uptodate wrt jiffies etc).
  */
-int get_nohz_timer_target(int pinned)
+int get_nohz_timer_target(void)
 {
-	int cpu = smp_processor_id();
-	int i;
+	int i, cpu = smp_processor_id();
 	struct sched_domain *sd;
 
-	if (pinned || !get_sysctl_timer_migration() || !idle_cpu(cpu))
+	if (!idle_cpu(cpu))
 		return cpu;
 
 	rcu_read_lock();
@@ -7050,8 +7049,6 @@ void __init sched_init_smp(void)
 }
 #endif /* CONFIG_SMP */
 
-const_debug unsigned int sysctl_timer_migration = 1;
-
 int in_sched_functions(unsigned long addr)
 {
 	return in_lock_functions(addr) ||

kernel/sysctl.c

@@ -349,15 +349,6 @@ static struct ctl_table kern_table[] = {
 		.mode		= 0644,
 		.proc_handler	= proc_dointvec,
 	},
-	{
-		.procname	= "timer_migration",
-		.data		= &sysctl_timer_migration,
-		.maxlen		= sizeof(unsigned int),
-		.mode		= 0644,
-		.proc_handler	= proc_dointvec_minmax,
-		.extra1		= &zero,
-		.extra2		= &one,
-	},
 #endif /* CONFIG_SMP */
 #ifdef CONFIG_NUMA_BALANCING
 	{
@@ -1132,6 +1123,15 @@ static struct ctl_table kern_table[] = {
 		.extra1		= &zero,
 		.extra2		= &one,
 	},
+#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
+	{
+		.procname	= "timer_migration",
+		.data		= &sysctl_timer_migration,
+		.maxlen		= sizeof(unsigned int),
+		.mode		= 0644,
+		.proc_handler	= timer_migration_handler,
+	},
+#endif
 	{ }
 };
 

kernel/time/hrtimer.c

@@ -177,6 +177,24 @@ hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
 #endif
 }
 
+#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
+static inline
+struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
+					 int pinned)
+{
+	if (pinned || !base->migration_enabled)
+		return this_cpu_ptr(&hrtimer_bases);
+	return &per_cpu(hrtimer_bases, get_nohz_timer_target());
+}
+#else
+static inline
+struct hrtimer_cpu_base *get_target_base(struct hrtimer_cpu_base *base,
+					 int pinned)
+{
+	return this_cpu_ptr(&hrtimer_bases);
+}
+#endif
+
 /*
  * Switch the timer base to the current CPU when possible.
  */
@@ -184,14 +202,13 @@ static inline struct hrtimer_clock_base *
 switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 		    int pinned)
 {
+	struct hrtimer_cpu_base *new_cpu_base, *this_base;
 	struct hrtimer_clock_base *new_base;
-	struct hrtimer_cpu_base *new_cpu_base;
-	int this_cpu = smp_processor_id();
-	int cpu = get_nohz_timer_target(pinned);
 	int basenum = base->index;
 
+	this_base = this_cpu_ptr(&hrtimer_bases);
+	new_cpu_base = get_target_base(this_base, pinned);
 again:
-	new_cpu_base = &per_cpu(hrtimer_bases, cpu);
 	new_base = &new_cpu_base->clock_base[basenum];
 
 	if (base != new_base) {
@@ -212,17 +229,19 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 		raw_spin_unlock(&base->cpu_base->lock);
 		raw_spin_lock(&new_base->cpu_base->lock);
 
-		if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
-			cpu = this_cpu;
+		if (new_cpu_base != this_base &&
+		    hrtimer_check_target(timer, new_base)) {
 			raw_spin_unlock(&new_base->cpu_base->lock);
 			raw_spin_lock(&base->cpu_base->lock);
+			new_cpu_base = this_base;
 			timer->base = base;
 			goto again;
 		}
 		timer->base = new_base;
 	} else {
-		if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
-			cpu = this_cpu;
+		if (new_cpu_base != this_base &&
+		    hrtimer_check_target(timer, new_base)) {
+			new_cpu_base = this_base;
 			goto again;
 		}
 	}

kernel/time/tick-internal.h

@@ -149,4 +149,18 @@ extern void tick_nohz_init(void);
 static inline void tick_nohz_init(void) { }
 #endif
 
+#ifdef CONFIG_NO_HZ_COMMON
+extern unsigned long tick_nohz_active;
+#else
+#define tick_nohz_active (0)
+#endif
+
+#if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ_COMMON)
+extern void timers_update_migration(void);
+#else
+static inline void timers_update_migration(void) { }
+#endif
+
+DECLARE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases);
+
 extern u64 get_next_timer_interrupt(unsigned long basej, u64 basem);

kernel/time/tick-sched.c

@@ -399,7 +399,7 @@ void __init tick_nohz_init(void)
  * NO HZ enabled ?
  */
 static int tick_nohz_enabled __read_mostly  = 1;
-int tick_nohz_active  __read_mostly;
+unsigned long tick_nohz_active  __read_mostly;
 /*
  * Enable / Disable tickless mode
  */
@@ -956,6 +956,16 @@ static void tick_nohz_handler(struct clock_event_device *dev)
 	tick_program_event(hrtimer_get_expires(&ts->sched_timer), 1);
 }
 
+static inline void tick_nohz_activate(struct tick_sched *ts, int mode)
+{
+	if (!tick_nohz_enabled)
+		return;
+	ts->nohz_mode = mode;
+	/* One update is enough */
+	if (!test_and_set_bit(0, &tick_nohz_active))
+		timers_update_migration();
+}
+
 /**
  * tick_nohz_switch_to_nohz - switch to nohz mode
  */
@@ -970,9 +980,6 @@ static void tick_nohz_switch_to_nohz(void)
 	if (tick_switch_to_oneshot(tick_nohz_handler))
 		return;
 
-	tick_nohz_active = 1;
-	ts->nohz_mode = NOHZ_MODE_LOWRES;
-
 	/*
 	 * Recycle the hrtimer in ts, so we can share the
 	 * hrtimer_forward with the highres code.
@@ -984,6 +991,7 @@ static void tick_nohz_switch_to_nohz(void)
 	hrtimer_forward_now(&ts->sched_timer, tick_period);
 	hrtimer_set_expires(&ts->sched_timer, next);
 	tick_program_event(next, 1);
+	tick_nohz_activate(ts, NOHZ_MODE_LOWRES);
 }
 
 /*
@@ -1035,6 +1043,7 @@ static inline void tick_nohz_irq_enter(void)
 
 static inline void tick_nohz_switch_to_nohz(void) { }
 static inline void tick_nohz_irq_enter(void) { }
+static inline void tick_nohz_activate(struct tick_sched *ts, int mode) { }
 
 #endif /* CONFIG_NO_HZ_COMMON */
 
@@ -1117,13 +1126,7 @@ void tick_setup_sched_timer(void)
 
 	hrtimer_forward(&ts->sched_timer, now, tick_period);
 	hrtimer_start_expires(&ts->sched_timer, HRTIMER_MODE_ABS_PINNED);
-
-#ifdef CONFIG_NO_HZ_COMMON
-	if (tick_nohz_enabled) {
-		ts->nohz_mode = NOHZ_MODE_HIGHRES;
-		tick_nohz_active = 1;
-	}
-#endif
+	tick_nohz_activate(ts, NOHZ_MODE_HIGHRES);
 }
 #endif /* HIGH_RES_TIMERS */
 

kernel/time/timer.c

@@ -85,6 +85,7 @@ struct tvec_base {
 	unsigned long active_timers;
 	unsigned long all_timers;
 	int cpu;
+	bool migration_enabled;
 	struct tvec_root tv1;
 	struct tvec tv2;
 	struct tvec tv3;
@@ -95,6 +96,54 @@ struct tvec_base {
 
 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)
+{
+	bool on = sysctl_timer_migration && tick_nohz_active;
+	unsigned int cpu;
+
+	/* Avoid the loop, if nothing to update */
+	if (this_cpu_read(tvec_bases.migration_enabled) == on)
+		return;
+
+	for_each_possible_cpu(cpu) {
+		per_cpu(tvec_bases.migration_enabled, cpu) = on;
+		per_cpu(hrtimer_bases.migration_enabled, cpu) = on;
+	}
+}
+
+int timer_migration_handler(struct ctl_table *table, int write,
+			    void __user *buffer, size_t *lenp,
+			    loff_t *ppos)
+{
+	static DEFINE_MUTEX(mutex);
+	int ret;
+
+	mutex_lock(&mutex);
+	ret = proc_dointvec(table, write, buffer, lenp, ppos);
+	if (!ret && write)
+		timers_update_migration();
+	mutex_unlock(&mutex);
+	return ret;
+}
+
+static inline struct tvec_base *get_target_base(struct tvec_base *base,
+						int pinned)
+{
+	if (pinned || !base->migration_enabled)
+		return this_cpu_ptr(&tvec_bases);
+	return per_cpu_ptr(&tvec_bases, get_nohz_timer_target());
+}
+#else
+static inline struct tvec_base *get_target_base(struct tvec_base *base,
+						int pinned)
+{
+	return this_cpu_ptr(&tvec_bases);
+}
+#endif
+
 static unsigned long round_jiffies_common(unsigned long j, int cpu,
 		bool force_up)
 {
@@ -716,11 +765,11 @@ static struct tvec_base *lock_timer_base(struct timer_list *timer,
 
 static inline int
 __mod_timer(struct timer_list *timer, unsigned long expires,
-						bool pending_only, int pinned)
+	    bool pending_only, int pinned)
 {
 	struct tvec_base *base, *new_base;
 	unsigned long flags;
-	int ret = 0 , cpu;
+	int ret = 0;
 
 	timer_stats_timer_set_start_info(timer);
 	BUG_ON(!timer->function);
@@ -733,8 +782,7 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
 
 	debug_activate(timer, expires);
 
-	cpu = get_nohz_timer_target(pinned);
-	new_base = per_cpu_ptr(&tvec_bases, cpu);
+	new_base = get_target_base(base, pinned);
 
 	if (base != new_base) {
 		/*
@@ -751,7 +799,8 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
 			spin_unlock(&base->lock);
 			base = new_base;
 			spin_lock(&base->lock);
-			timer->flags = (timer->flags & ~TIMER_BASEMASK) | cpu;
+			timer->flags &= ~TIMER_BASEMASK;
+			timer->flags |= base->cpu;
 		}
 	}
 

kernel/time/timer_list.c

@@ -29,8 +29,6 @@ struct timer_list_iter {
 
 typedef void (*print_fn_t)(struct seq_file *m, unsigned int *classes);
 
-DECLARE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases);
-
 /*
  * This allows printing both to /proc/timer_list and
  * to the console (on SysRq-Q):