hrtimer: add support for starting timers on remote CPUs

This depends on the ability to send the special 'pull_timers' IPI
even with IRQs disabled.

3 files changed, 89 insertions, 0 deletions

diff --git a/include/linux/hrtimer.h b/include/linux/hrtimer.h
index 7a9398e197..4f45c83151 100644
--- a/include/linux/hrtimer.h
+++ b/include/linux/hrtimer.h
@@ -200,6 +200,19 @@ struct hrtimer_cpu_base {
         struct list_head                cb_pending;
         unsigned long                   nr_events;
 #endif
+        struct list_head                to_pull;
+};
+
+#define HRTIMER_START_ON_INACTIVE       0
+#define HRTIMER_START_ON_QUEUED         1
+#define HRTIMER_START_ON_PROCESSED      2
+
+struct hrtimer_start_on_info {
+        struct list_head        list;
+        struct hrtimer*         timer;
+        ktime_t                 time;
+        enum hrtimer_mode       mode;
+        atomic_t                state;
 };
 
 #ifdef CONFIG_HIGH_RES_TIMERS
@@ -262,6 +275,9 @@ extern void hrtimer_init(struct hrtimer *timer, clockid_t which_clock,
 /* Basic timer operations: */
 extern int hrtimer_start(struct hrtimer *timer, ktime_t tim,
                          const enum hrtimer_mode mode);
+extern int hrtimer_start_on(int cpu, struct hrtimer_start_on_info* info,
+                            struct hrtimer *timer, ktime_t time,
+                            const enum hrtimer_mode mode);
 extern int hrtimer_cancel(struct hrtimer *timer);
 extern int hrtimer_try_to_cancel(struct hrtimer *timer);
 
diff --git a/include/linux/smp.h b/include/linux/smp.h
index c25e66bcec..84df090ce4 100644
--- a/include/linux/smp.h
+++ b/include/linux/smp.h
@@ -35,6 +35,12 @@ extern void smp_send_reschedule(int cpu);
 
 
 /*
+ * sends a 'pull timers' event to another CPU:
+ */
+extern void smp_send_pull_timers(int cpu);
+
+
+/*
  * Prepare machine for booting other CPUs.
  */
 extern void smp_prepare_cpus(unsigned int max_cpus);
diff --git a/kernel/hrtimer.c b/kernel/hrtimer.c
index f994bb8065..c47baab445 100644
--- a/kernel/hrtimer.c
+++ b/kernel/hrtimer.c
@@ -813,6 +813,73 @@ remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
         return 0;
 }
 
+void hrtimer_pull(void)
+{
+        struct hrtimer_cpu_base* base = &__get_cpu_var(hrtimer_bases);
+        struct hrtimer_start_on_info* info;
+        struct list_head *pos, *safe, list;
+
+        spin_lock(&base->lock);
+        list_replace_init(&base->to_pull, &list);
+        spin_unlock(&base->lock);
+
+        list_for_each_safe(pos, safe, &list) {
+                info = list_entry(pos, struct hrtimer_start_on_info, list);
+                TRACE("pulled timer 0x%x\n", info->timer);
+                list_del(pos);
+                hrtimer_start(info->timer, info->time, info->mode);
+                mb();
+                atomic_set(&info->state, HRTIMER_START_ON_PROCESSED);
+        }
+}
+
+int hrtimer_start_on(int cpu, struct hrtimer_start_on_info* info,
+                     struct hrtimer *timer, ktime_t time,
+                     const enum hrtimer_mode mode)
+{
+        unsigned long flags;
+        struct hrtimer_cpu_base* base;
+        int in_use = 0, was_empty;
+
+        /* serialize access to info through the timer base */
+        lock_hrtimer_base(timer, &flags);
+
+        in_use = atomic_read(&info->state) != HRTIMER_START_ON_INACTIVE;
+        if (!in_use) {
+                INIT_LIST_HEAD(&info->list);
+                info->timer = timer;
+                info->time  = time;
+                info->mode  = mode;
+                /* mark as in use */
+                atomic_set(&info->state, HRTIMER_START_ON_QUEUED);
+        }
+
+        unlock_hrtimer_base(timer, &flags);
+
+        if (!in_use) {
+                /* initiate pull  */
+                preempt_disable();
+                if (cpu == smp_processor_id()) {
+                        /* start timer locally */
+                        hrtimer_start(info->timer, info->time, info->mode);
+                        atomic_set(&info->state, HRTIMER_START_ON_PROCESSED);
+                } else {
+                        base = &per_cpu(hrtimer_bases, cpu);
+                        spin_lock_irqsave(&base->lock, flags);
+                        was_empty = list_empty(&base->to_pull);
+                        list_add(&info->list, &base->to_pull);
+                        spin_unlock_irqrestore(&base->lock, flags);
+                        if (was_empty)
+                                /* only send IPI if other no else 
+                                 * has done so already
+                                 */
+                                smp_send_pull_timers(cpu);
+                }
+                preempt_enable();
+        }
+        return in_use;
+}
+
 /**
  * hrtimer_start - (re)start an relative timer on the current CPU
  * @timer:      the timer to be added