mxc_nand: Make main/spare areas runtime configurable
[safe/jmp/linux-2.6] / kernel / hrtimer.c
index 8fea312..e5d98ce 100644 (file)
 #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>
 
-/**
- * ktime_get - get the monotonic time in ktime_t format
- *
- * returns the time in ktime_t format
- */
-ktime_t ktime_get(void)
-{
-       struct timespec now;
-
-       ktime_get_ts(&now);
-
-       return timespec_to_ktime(now);
-}
-EXPORT_SYMBOL_GPL(ktime_get);
-
-/**
- * ktime_get_real - get the real (wall-) time in ktime_t format
- *
- * returns the time in ktime_t format
- */
-ktime_t ktime_get_real(void)
-{
-       struct timespec now;
-
-       getnstimeofday(&now);
-
-       return timespec_to_ktime(now);
-}
-
-EXPORT_SYMBOL_GPL(ktime_get_real);
+#include <trace/events/timer.h>
 
 /*
  * The timer bases:
@@ -104,31 +77,6 @@ DEFINE_PER_CPU(struct hrtimer_cpu_base, hrtimer_bases) =
        }
 };
 
-/**
- * ktime_get_ts - get the monotonic clock in timespec format
- * @ts:                pointer to timespec variable
- *
- * The function calculates the monotonic clock from the realtime
- * clock and the wall_to_monotonic offset and stores the result
- * in normalized timespec format in the variable pointed to by @ts.
- */
-void ktime_get_ts(struct timespec *ts)
-{
-       struct timespec tomono;
-       unsigned long seq;
-
-       do {
-               seq = read_seqbegin(&xtime_lock);
-               getnstimeofday(ts);
-               tomono = wall_to_monotonic;
-
-       } while (read_seqretry(&xtime_lock, seq));
-
-       set_normalized_timespec(ts, ts->tv_sec + tomono.tv_sec,
-                               ts->tv_nsec + tomono.tv_nsec);
-}
-EXPORT_SYMBOL_GPL(ktime_get_ts);
-
 /*
  * Get the coarse grained time at the softirq based on xtime and
  * wall_to_monotonic.
@@ -189,21 +137,65 @@ struct hrtimer_clock_base *lock_hrtimer_base(const struct hrtimer *timer,
        }
 }
 
+
+/*
+ * Get the preferred target CPU for NOHZ
+ */
+static int hrtimer_get_target(int this_cpu, int pinned)
+{
+#ifdef CONFIG_NO_HZ
+       if (!pinned && get_sysctl_timer_migration() && idle_cpu(this_cpu)) {
+               int preferred_cpu = get_nohz_load_balancer();
+
+               if (preferred_cpu >= 0)
+                       return preferred_cpu;
+       }
+#endif
+       return this_cpu;
+}
+
+/*
+ * With HIGHRES=y 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.
+ *
+ * Called with cpu_base->lock of target cpu held.
+ */
+static int
+hrtimer_check_target(struct hrtimer *timer, struct hrtimer_clock_base *new_base)
+{
+#ifdef CONFIG_HIGH_RES_TIMERS
+       ktime_t expires;
+
+       if (!new_base->cpu_base->hres_active)
+               return 0;
+
+       expires = ktime_sub(hrtimer_get_expires(timer), new_base->offset);
+       return expires.tv64 <= new_base->cpu_base->expires_next.tv64;
+#else
+       return 0;
+#endif
+}
+
 /*
  * Switch the timer base to the current CPU when possible.
  */
 static inline struct hrtimer_clock_base *
-switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base)
+switch_hrtimer_base(struct hrtimer *timer, struct hrtimer_clock_base *base,
+                   int pinned)
 {
        struct hrtimer_clock_base *new_base;
        struct hrtimer_cpu_base *new_cpu_base;
+       int this_cpu = smp_processor_id();
+       int cpu = hrtimer_get_target(this_cpu, pinned);
 
-       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) {
                /*
-                * We are trying to schedule the timer on the local CPU.
+                * We are trying to move timer to new_base.
                 * However we can't change timer's base while it is running,
                 * so we keep it on the same CPU. No hassle vs. reprogramming
                 * the event source in the high resolution case. The softirq
@@ -218,6 +210,14 @@ 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);
+
+               if (cpu != this_cpu && hrtimer_check_target(timer, new_base)) {
+                       cpu = this_cpu;
+                       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;
@@ -235,7 +235,7 @@ 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 */
 
@@ -332,6 +332,8 @@ ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
        return res;
 }
 
+EXPORT_SYMBOL_GPL(ktime_add_safe);
+
 #ifdef CONFIG_DEBUG_OBJECTS_TIMERS
 
 static struct debug_obj_descr hrtimer_debug_descr;
@@ -429,6 +431,7 @@ void hrtimer_init_on_stack(struct hrtimer *timer, clockid_t clock_id,
        debug_object_init_on_stack(timer, &hrtimer_debug_descr);
        __hrtimer_init(timer, clock_id, mode);
 }
+EXPORT_SYMBOL_GPL(hrtimer_init_on_stack);
 
 void destroy_hrtimer_on_stack(struct hrtimer *timer)
 {
@@ -441,6 +444,26 @@ static inline void debug_hrtimer_activate(struct hrtimer *timer) { }
 static inline void debug_hrtimer_deactivate(struct hrtimer *timer) { }
 #endif
 
+static inline void
+debug_init(struct hrtimer *timer, clockid_t clockid,
+          enum hrtimer_mode mode)
+{
+       debug_hrtimer_init(timer);
+       trace_hrtimer_init(timer, clockid, mode);
+}
+
+static inline void debug_activate(struct hrtimer *timer)
+{
+       debug_hrtimer_activate(timer);
+       trace_hrtimer_start(timer);
+}
+
+static inline void debug_deactivate(struct hrtimer *timer)
+{
+       debug_hrtimer_deactivate(timer);
+       trace_hrtimer_cancel(timer);
+}
+
 /* High resolution timer related functions */
 #ifdef CONFIG_HIGH_RES_TIMERS
 
@@ -501,6 +524,13 @@ static void hrtimer_force_reprogram(struct hrtimer_cpu_base *cpu_base)
                        continue;
                timer = rb_entry(base->first, struct hrtimer, node);
                expires = ktime_sub(hrtimer_get_expires(timer), base->offset);
+               /*
+                * clock_was_set() has changed base->offset so the
+                * result might be negative. Fix it up to prevent a
+                * false positive in clockevents_program_event()
+                */
+               if (expires.tv64 < 0)
+                       expires.tv64 = 0;
                if (expires.tv64 < cpu_base->expires_next.tv64)
                        cpu_base->expires_next = expires;
        }
@@ -644,14 +674,20 @@ static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
  * and expiry check is done in the hrtimer_interrupt or in the softirq.
  */
 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
-                                           struct hrtimer_clock_base *base)
+                                           struct hrtimer_clock_base *base,
+                                           int wakeup)
 {
        if (base->cpu_base->hres_active && hrtimer_reprogram(timer, base)) {
-               spin_unlock(&base->cpu_base->lock);
-               raise_softirq_irqoff(HRTIMER_SOFTIRQ);
-               spin_lock(&base->cpu_base->lock);
+               if (wakeup) {
+                       spin_unlock(&base->cpu_base->lock);
+                       raise_softirq_irqoff(HRTIMER_SOFTIRQ);
+                       spin_lock(&base->cpu_base->lock);
+               } else
+                       __raise_softirq_irqoff(HRTIMER_SOFTIRQ);
+
                return 1;
        }
+
        return 0;
 }
 
@@ -696,7 +732,8 @@ static inline int hrtimer_is_hres_enabled(void) { return 0; }
 static inline int hrtimer_switch_to_hres(void) { return 0; }
 static inline void hrtimer_force_reprogram(struct hrtimer_cpu_base *base) { }
 static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
-                                           struct hrtimer_clock_base *base)
+                                           struct hrtimer_clock_base *base,
+                                           int wakeup)
 {
        return 0;
 }
@@ -783,7 +820,7 @@ static int enqueue_hrtimer(struct hrtimer *timer,
        struct hrtimer *entry;
        int leftmost = 1;
 
-       debug_hrtimer_activate(timer);
+       debug_activate(timer);
 
        /*
         * Find the right place in the rbtree:
@@ -869,7 +906,7 @@ remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
                 * reprogramming happens in the interrupt handler. This is a
                 * rare case and less expensive than a smp call.
                 */
-               debug_hrtimer_deactivate(timer);
+               debug_deactivate(timer);
                timer_stats_hrtimer_clear_start_info(timer);
                reprogram = base->cpu_base == &__get_cpu_var(hrtimer_bases);
                __remove_hrtimer(timer, base, HRTIMER_STATE_INACTIVE,
@@ -879,20 +916,9 @@ remove_hrtimer(struct hrtimer *timer, struct hrtimer_clock_base *base)
        return 0;
 }
 
-/**
- * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
- * @timer:     the timer to be added
- * @tim:       expiry time
- * @delta_ns:  "slack" range for the timer
- * @mode:      expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
- *
- * Returns:
- *  0 on success
- *  1 when the timer was active
- */
-int
-hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_ns,
-                       const enum hrtimer_mode mode)
+int __hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
+               unsigned long delta_ns, const enum hrtimer_mode mode,
+               int wakeup)
 {
        struct hrtimer_clock_base *base, *new_base;
        unsigned long flags;
@@ -904,9 +930,9 @@ hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_n
        ret = remove_hrtimer(timer, base);
 
        /* Switch the timer base, if necessary: */
-       new_base = switch_hrtimer_base(timer, base);
+       new_base = switch_hrtimer_base(timer, base, mode & HRTIMER_MODE_PINNED);
 
-       if (mode == HRTIMER_MODE_REL) {
+       if (mode & HRTIMER_MODE_REL) {
                tim = ktime_add_safe(tim, new_base->get_time());
                /*
                 * CONFIG_TIME_LOW_RES is a temporary way for architectures
@@ -933,12 +959,29 @@ hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim, unsigned long delta_n
         * XXX send_remote_softirq() ?
         */
        if (leftmost && new_base->cpu_base == &__get_cpu_var(hrtimer_bases))
-               hrtimer_enqueue_reprogram(timer, new_base);
+               hrtimer_enqueue_reprogram(timer, new_base, wakeup);
 
        unlock_hrtimer_base(timer, &flags);
 
        return ret;
 }
+
+/**
+ * hrtimer_start_range_ns - (re)start an hrtimer on the current CPU
+ * @timer:     the timer to be added
+ * @tim:       expiry time
+ * @delta_ns:  "slack" range for the timer
+ * @mode:      expiry mode: absolute (HRTIMER_ABS) or relative (HRTIMER_REL)
+ *
+ * Returns:
+ *  0 on success
+ *  1 when the timer was active
+ */
+int hrtimer_start_range_ns(struct hrtimer *timer, ktime_t tim,
+               unsigned long delta_ns, const enum hrtimer_mode mode)
+{
+       return __hrtimer_start_range_ns(timer, tim, delta_ns, mode, 1);
+}
 EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
 
 /**
@@ -954,7 +997,7 @@ EXPORT_SYMBOL_GPL(hrtimer_start_range_ns);
 int
 hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
 {
-       return hrtimer_start_range_ns(timer, tim, 0, mode);
+       return __hrtimer_start_range_ns(timer, tim, 0, mode, 1);
 }
 EXPORT_SYMBOL_GPL(hrtimer_start);
 
@@ -1078,7 +1121,6 @@ static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
                clock_id = CLOCK_MONOTONIC;
 
        timer->base = &cpu_base->clock_base[clock_id];
-       INIT_LIST_HEAD(&timer->cb_entry);
        hrtimer_init_timer_hres(timer);
 
 #ifdef CONFIG_TIMER_STATS
@@ -1097,7 +1139,7 @@ static void __hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
 void hrtimer_init(struct hrtimer *timer, clockid_t clock_id,
                  enum hrtimer_mode mode)
 {
-       debug_hrtimer_init(timer);
+       debug_init(timer, clock_id, mode);
        __hrtimer_init(timer, clock_id, mode);
 }
 EXPORT_SYMBOL_GPL(hrtimer_init);
@@ -1121,7 +1163,7 @@ int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
 }
 EXPORT_SYMBOL_GPL(hrtimer_get_res);
 
-static void __run_hrtimer(struct hrtimer *timer)
+static void __run_hrtimer(struct hrtimer *timer, ktime_t *now)
 {
        struct hrtimer_clock_base *base = timer->base;
        struct hrtimer_cpu_base *cpu_base = base->cpu_base;
@@ -1130,7 +1172,7 @@ static void __run_hrtimer(struct hrtimer *timer)
 
        WARN_ON(!irqs_disabled());
 
-       debug_hrtimer_deactivate(timer);
+       debug_deactivate(timer);
        __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
        timer_stats_account_hrtimer(timer);
        fn = timer->function;
@@ -1141,7 +1183,9 @@ static void __run_hrtimer(struct hrtimer *timer)
         * the timer base.
         */
        spin_unlock(&cpu_base->lock);
+       trace_hrtimer_expire_entry(timer, now);
        restart = fn(timer);
+       trace_hrtimer_expire_exit(timer);
        spin_lock(&cpu_base->lock);
 
        /*
@@ -1206,14 +1250,22 @@ void hrtimer_interrupt(struct clock_event_device *dev)
 
        expires_next.tv64 = KTIME_MAX;
 
+       spin_lock(&cpu_base->lock);
+       /*
+        * We set expires_next to KTIME_MAX here with cpu_base->lock
+        * held to prevent that a timer is enqueued in our queue via
+        * the migration code. This does not affect enqueueing of
+        * timers which run their callback and need to be requeued on
+        * this CPU.
+        */
+       cpu_base->expires_next.tv64 = KTIME_MAX;
+
        base = cpu_base->clock_base;
 
        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
                ktime_t basenow;
                struct rb_node *node;
 
-               spin_lock(&cpu_base->lock);
-
                basenow = ktime_add(now, base->offset);
 
                while ((node = base->first)) {
@@ -1244,13 +1296,17 @@ void hrtimer_interrupt(struct clock_event_device *dev)
                                break;
                        }
 
-                       __run_hrtimer(timer);
+                       __run_hrtimer(timer, &basenow);
                }
-               spin_unlock(&cpu_base->lock);
                base++;
        }
 
+       /*
+        * Store the new expiry value so the migration code can verify
+        * against it.
+        */
        cpu_base->expires_next = expires_next;
+       spin_unlock(&cpu_base->lock);
 
        /* Reprogramming necessary ? */
        if (expires_next.tv64 != KTIME_MAX) {
@@ -1362,7 +1418,7 @@ void hrtimer_run_queues(void)
                                        hrtimer_get_expires_tv64(timer))
                                break;
 
-                       __run_hrtimer(timer);
+                       __run_hrtimer(timer, &base->softirq_time);
                }
                spin_unlock(&cpu_base->lock);
        }
@@ -1389,6 +1445,7 @@ void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
        sl->timer.function = hrtimer_wakeup;
        sl->task = task;
 }
+EXPORT_SYMBOL_GPL(hrtimer_init_sleeper);
 
 static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mode)
 {
@@ -1538,7 +1595,7 @@ static void migrate_hrtimer_list(struct hrtimer_clock_base *old_base,
        while ((node = rb_first(&old_base->active))) {
                timer = rb_entry(node, struct hrtimer, node);
                BUG_ON(hrtimer_callback_running(timer));
-               debug_hrtimer_deactivate(timer);
+               debug_deactivate(timer);
 
                /*
                 * Mark it as STATE_MIGRATE not INACTIVE otherwise the
@@ -1608,6 +1665,10 @@ static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
                break;
 
 #ifdef CONFIG_HOTPLUG_CPU
+       case CPU_DYING:
+       case CPU_DYING_FROZEN:
+               clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DYING, &scpu);
+               break;
        case CPU_DEAD:
        case CPU_DEAD_FROZEN:
        {