timers: Logic to move non pinned timers

* Arun R Bharadwaj <arun@linux.vnet.ibm.com> [2009-04-16 12:11:36]:

This patch migrates all non pinned timers and hrtimers to the current
idle load balancer, from all the idle CPUs. Timers firing on busy CPUs
are not migrated.

While migrating hrtimers, care should be taken to check if migrating
a hrtimer would result in a latency or not. So we compare the expiry of the
hrtimer with the next timer interrupt on the target cpu and migrate the
hrtimer only if it expires *after* the next interrupt on the target cpu.
So, added a clockevents_get_next_event() helper function to return the
next_event on the target cpu's clock_event_device.

[ tglx: cleanups and simplifications ]

Signed-off-by: Arun R Bharadwaj <arun@linux.vnet.ibm.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>

6 files changed, 101 insertions, 5 deletions

diff --git a/include/linux/clockchips.h b/include/linux/clockchips.h
index 3a1dbba4d3ae..20a100fe2b4f 100644
--- a/include/linux/clockchips.h
+++ b/include/linux/clockchips.h
@@ -143,3 +143,12 @@ extern void clockevents_notify(unsigned long reason, void *arg);
 #endif
 
 #endif
+
+#ifdef CONFIG_GENERIC_CLOCKEVENTS
+extern ktime_t clockevents_get_next_event(int cpu);
+#else
+static inline ktime_t clockevents_get_next_event(int cpu)
+{
+        return (ktime_t) { .tv64 = KTIME_MAX };
+}
+#endif
diff --git a/include/linux/sched.h b/include/linux/sched.h
index 618504010400..311dec123974 100644
--- a/include/linux/sched.h
+++ b/include/linux/sched.h
@@ -257,6 +257,7 @@ extern void task_rq_unlock_wait(struct task_struct *p);
 extern cpumask_var_t nohz_cpu_mask;
 #if defined(CONFIG_SMP) && defined(CONFIG_NO_HZ)
 extern int select_nohz_load_balancer(int cpu);
+extern int get_nohz_load_balancer(void);
 #else
 static inline int select_nohz_load_balancer(int cpu)
 {
@@ -1772,6 +1773,17 @@ int sched_nr_latency_handler(struct ctl_table *table, int write,
                 struct file *file, 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
 extern unsigned int sysctl_sched_rt_period;
 extern int sysctl_sched_rt_runtime;
 
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index c71bcd549241..b675a67c9ac3 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -43,6 +43,8 @@
 #include <linux/seq_file.h>
 #include <linux/err.h>
 #include <linux/debugobjects.h>
+#include <linux/sched.h>
+#include <linux/timer.h>
 
 #include <asm/uaccess.h>
 
@@ -198,8 +200,19 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
 {
         struct hrtimer_clock_base *new_base;
         struct hrtimer_cpu_base *new_cpu_base;
+        int cpu, preferred_cpu = -1;
+
+        cpu = smp_processor_id();
+#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
+        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
+                preferred_cpu = get_nohz_load_balancer();
+                if (preferred_cpu >= 0)
+                        cpu = preferred_cpu;
+        }
+#endif
 
-        new_cpu_base = &__get_cpu_var(hrtimer_bases);
+again:
+        new_cpu_base = &per_cpu(hrtimer_bases, cpu);
         new_base = &new_cpu_base->clock_base[base->index];
 
         if (base != new_base) {
@@ -219,6 +232,40 @@ switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
                 timer->base = NULL;
                 spin_unlock(&base->cpu_base->lock);
                 spin_lock(&new_base->cpu_base->lock);
+
+                /* Optimized away for NOHZ=n SMP=n */
+                if (cpu == preferred_cpu) {
+                        /* Calculate clock monotonic expiry time */
+#ifdef CONFIG_HIGH_RES_TIMERS
+                        ktime_t expires = ktime_sub(hrtimer_get_expires(timer),
+                                                        new_base->offset);
+#else
+                        ktime_t expires = hrtimer_get_expires(timer);
+#endif
+
+                        /*
+                         * Get the next event on target cpu from the
+                         * clock events layer.
+                         * This covers the highres=off nohz=on case as well.
+                         */
+                        ktime_t next = clockevents_get_next_event(cpu);
+
+                        ktime_t delta = ktime_sub(expires, next);
+
+                        /*
+                         * We do not migrate the timer when it is expiring
+                         * before the next event on the target cpu because
+                         * we cannot reprogram the target cpu hardware and
+                         * we would cause it to fire late.
+                         */
+                        if (delta.tv64 < 0) {
+                                cpu = smp_processor_id();
+                                spin_unlock(&new_base->cpu_base->lock);
+                                spin_lock(&base->cpu_base->lock);
+                                timer->base = base;
+                                goto again;
+                        }
+                }
                 timer->base = new_base;
         }
         return new_base;
@@ -236,7 +283,7 @@ lock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
         return base;
 }
 
-# define switch_hrtimer_base(t, b)      (b)
+# define switch_hrtimer_base(t, b, p)   (b)
 
 #endif  /* !CONFIG_SMP */
 
diff --git a/kernel/sched.c b/kernel/sched.c
index 7f1dd56af863..9fe3774a0fd3 100644
--- a/kernel/sched.c
+++ b/kernel/sched.c
@@ -4244,6 +4244,11 @@ static struct {
         .load_balancer = ATOMIC_INIT(-1),
 };
 
+int get_nohz_load_balancer(void)
+{
+        return atomic_read(&nohz.load_balancer);
+}
+
 /*
  * This routine will try to nominate the ilb (idle load balancing)
  * owner among the cpus whose ticks are stopped. ilb owner will do the idle
diff --git a/kernel/time/clockevents.c b/kernel/time/clockevents.c
index d13be216a790..ab20ded013bd 100644
--- a/kernel/time/clockevents.c
+++ b/kernel/time/clockevents.c
@@ -18,6 +18,7 @@
 #include <linux/notifier.h>
 #include <linux/smp.h>
 #include <linux/sysdev.h>
+#include <linux/tick.h>
 
 /* The registered clock event devices */
 static LIST_HEAD(clockevent_devices);
@@ -251,4 +252,15 @@ void clockevents_notify(unsigned long reason, void *arg)
         spin_unlock(&clockevents_lock);
 }
 EXPORT_SYMBOL_GPL(clockevents_notify);
+
+ktime_t clockevents_get_next_event(int cpu)
+{
+        struct tick_device *td;
+        struct clock_event_device *dev;
+
+        td = &per_cpu(tick_cpu_device, cpu);
+        dev = td->evtdev;
+
+        return dev->next_event;
+}
 #endif
diff --git a/kernel/timer.c b/kernel/timer.c
index 3424dfd11d50..3f841db5edf9 100644
--- a/kernel/timer.c
+++ b/kernel/timer.c
@@ -37,6 +37,7 @@
 #include <linux/delay.h>
 #include <linux/tick.h>
 #include <linux/kallsyms.h>
+#include <linux/sched.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
@@ -609,9 +610,7 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
 {
         struct tvec_base *base, *new_base;
         unsigned long flags;
-        int ret;
-
-        ret = 0;
+        int ret = 0 , cpu;
 
         timer_stats_timer_set_start_info(timer);
         BUG_ON(!timer->function);
@@ -630,6 +629,18 @@ __mod_timer(struct timer_list *timer, unsigned long expires,
 
         new_base = __get_cpu_var(tvec_bases);
 
+        cpu = smp_processor_id();
+
+#if defined(CONFIG_NO_HZ) && defined(CONFIG_SMP)
+        if (!pinned && get_sysctl_timer_migration() && idle_cpu(cpu)) {
+                int preferred_cpu = get_nohz_load_balancer();
+
+                if (preferred_cpu >= 0)
+                        cpu = preferred_cpu;
+        }
+#endif
+        new_base = per_cpu(tvec_bases, cpu);
+
         if (base != new_base) {
                 /*
                  * We are trying to schedule the timer on the local CPU.