string: factorize skip_spaces and export it to be generally available
[safe/jmp/linux-2.6] / kernel / timer.c
index f3d35d4..5db5a8d 100644 (file)
@@ -37,6 +37,8 @@
 #include <linux/delay.h>
 #include <linux/tick.h>
 #include <linux/kallsyms.h>
+#include <linux/perf_event.h>
+#include <linux/sched.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
@@ -44,6 +46,9 @@
 #include <asm/timex.h>
 #include <asm/io.h>
 
+#define CREATE_TRACE_POINTS
+#include <trace/events/timer.h>
+
 u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
 
 EXPORT_SYMBOL(jiffies_64);
@@ -70,6 +75,7 @@ struct tvec_base {
        spinlock_t lock;
        struct timer_list *running_timer;
        unsigned long timer_jiffies;
+       unsigned long next_timer;
        struct tvec_root tv1;
        struct tvec tv2;
        struct tvec tv3;
@@ -112,27 +118,8 @@ timer_set_base(struct timer_list *timer, struct tvec_base *new_base)
                                      tbase_get_deferrable(timer->base));
 }
 
-/**
- * __round_jiffies - function to round jiffies to a full second
- * @j: the time in (absolute) jiffies that should be rounded
- * @cpu: the processor number on which the timeout will happen
- *
- * __round_jiffies() rounds an absolute time in the future (in jiffies)
- * up or down to (approximately) full seconds. This is useful for timers
- * for which the exact time they fire does not matter too much, as long as
- * they fire approximately every X seconds.
- *
- * By rounding these timers to whole seconds, all such timers will fire
- * at the same time, rather than at various times spread out. The goal
- * of this is to have the CPU wake up less, which saves power.
- *
- * The exact rounding is skewed for each processor to avoid all
- * processors firing at the exact same time, which could lead
- * to lock contention or spurious cache line bouncing.
- *
- * The return value is the rounded version of the @j parameter.
- */
-unsigned long __round_jiffies(unsigned long j, int cpu)
+static unsigned long round_jiffies_common(unsigned long j, int cpu,
+               bool force_up)
 {
        int rem;
        unsigned long original = j;
@@ -154,8 +141,9 @@ unsigned long __round_jiffies(unsigned long j, int cpu)
         * due to delays of the timer irq, long irq off times etc etc) then
         * we should round down to the whole second, not up. Use 1/4th second
         * as cutoff for this rounding as an extreme upper bound for this.
+        * But never round down if @force_up is set.
         */
-       if (rem < HZ/4) /* round down */
+       if (rem < HZ/4 && !force_up) /* round down */
                j = j - rem;
        else /* round up */
                j = j - rem + HZ;
@@ -167,6 +155,31 @@ unsigned long __round_jiffies(unsigned long j, int cpu)
                return original;
        return j;
 }
+
+/**
+ * __round_jiffies - function to round jiffies to a full second
+ * @j: the time in (absolute) jiffies that should be rounded
+ * @cpu: the processor number on which the timeout will happen
+ *
+ * __round_jiffies() rounds an absolute time in the future (in jiffies)
+ * up or down to (approximately) full seconds. This is useful for timers
+ * for which the exact time they fire does not matter too much, as long as
+ * they fire approximately every X seconds.
+ *
+ * By rounding these timers to whole seconds, all such timers will fire
+ * at the same time, rather than at various times spread out. The goal
+ * of this is to have the CPU wake up less, which saves power.
+ *
+ * The exact rounding is skewed for each processor to avoid all
+ * processors firing at the exact same time, which could lead
+ * to lock contention or spurious cache line bouncing.
+ *
+ * The return value is the rounded version of the @j parameter.
+ */
+unsigned long __round_jiffies(unsigned long j, int cpu)
+{
+       return round_jiffies_common(j, cpu, false);
+}
 EXPORT_SYMBOL_GPL(__round_jiffies);
 
 /**
@@ -191,13 +204,10 @@ EXPORT_SYMBOL_GPL(__round_jiffies);
  */
 unsigned long __round_jiffies_relative(unsigned long j, int cpu)
 {
-       /*
-        * In theory the following code can skip a jiffy in case jiffies
-        * increments right between the addition and the later subtraction.
-        * However since the entire point of this function is to use approximate
-        * timeouts, it's entirely ok to not handle that.
-        */
-       return  __round_jiffies(j + jiffies, cpu) - jiffies;
+       unsigned long j0 = jiffies;
+
+       /* Use j0 because jiffies might change while we run */
+       return round_jiffies_common(j + j0, cpu, false) - j0;
 }
 EXPORT_SYMBOL_GPL(__round_jiffies_relative);
 
@@ -218,7 +228,7 @@ EXPORT_SYMBOL_GPL(__round_jiffies_relative);
  */
 unsigned long round_jiffies(unsigned long j)
 {
-       return __round_jiffies(j, raw_smp_processor_id());
+       return round_jiffies_common(j, raw_smp_processor_id(), false);
 }
 EXPORT_SYMBOL_GPL(round_jiffies);
 
@@ -243,6 +253,71 @@ unsigned long round_jiffies_relative(unsigned long j)
 }
 EXPORT_SYMBOL_GPL(round_jiffies_relative);
 
+/**
+ * __round_jiffies_up - function to round jiffies up to a full second
+ * @j: the time in (absolute) jiffies that should be rounded
+ * @cpu: the processor number on which the timeout will happen
+ *
+ * This is the same as __round_jiffies() except that it will never
+ * round down.  This is useful for timeouts for which the exact time
+ * of firing does not matter too much, as long as they don't fire too
+ * early.
+ */
+unsigned long __round_jiffies_up(unsigned long j, int cpu)
+{
+       return round_jiffies_common(j, cpu, true);
+}
+EXPORT_SYMBOL_GPL(__round_jiffies_up);
+
+/**
+ * __round_jiffies_up_relative - function to round jiffies up to a full second
+ * @j: the time in (relative) jiffies that should be rounded
+ * @cpu: the processor number on which the timeout will happen
+ *
+ * This is the same as __round_jiffies_relative() except that it will never
+ * round down.  This is useful for timeouts for which the exact time
+ * of firing does not matter too much, as long as they don't fire too
+ * early.
+ */
+unsigned long __round_jiffies_up_relative(unsigned long j, int cpu)
+{
+       unsigned long j0 = jiffies;
+
+       /* Use j0 because jiffies might change while we run */
+       return round_jiffies_common(j + j0, cpu, true) - j0;
+}
+EXPORT_SYMBOL_GPL(__round_jiffies_up_relative);
+
+/**
+ * round_jiffies_up - function to round jiffies up to a full second
+ * @j: the time in (absolute) jiffies that should be rounded
+ *
+ * This is the same as round_jiffies() except that it will never
+ * round down.  This is useful for timeouts for which the exact time
+ * of firing does not matter too much, as long as they don't fire too
+ * early.
+ */
+unsigned long round_jiffies_up(unsigned long j)
+{
+       return round_jiffies_common(j, raw_smp_processor_id(), true);
+}
+EXPORT_SYMBOL_GPL(round_jiffies_up);
+
+/**
+ * round_jiffies_up_relative - function to round jiffies up to a full second
+ * @j: the time in (relative) jiffies that should be rounded
+ *
+ * This is the same as round_jiffies_relative() except that it will never
+ * round down.  This is useful for timeouts for which the exact time
+ * of firing does not matter too much, as long as they don't fire too
+ * early.
+ */
+unsigned long round_jiffies_up_relative(unsigned long j)
+{
+       return __round_jiffies_up_relative(j, raw_smp_processor_id());
+}
+EXPORT_SYMBOL_GPL(round_jiffies_up_relative);
+
 
 static inline void set_running_timer(struct tvec_base *base,
                                        struct timer_list *timer)
@@ -309,6 +384,8 @@ static void timer_stats_account_timer(struct timer_list *timer)
 {
        unsigned int flag = 0;
 
+       if (likely(!timer->start_site))
+               return;
        if (unlikely(tbase_get_deferrable(timer->base)))
                flag |= TIMER_STATS_FLAG_DEFERRABLE;
 
@@ -320,14 +397,155 @@ static void timer_stats_account_timer(struct timer_list *timer)
 static void timer_stats_account_timer(struct timer_list *timer) {}
 #endif
 
-/**
- * init_timer - initialize a timer.
- * @timer: the timer to be initialized
- *
- * init_timer() must be done to a timer prior calling *any* of the
- * other timer functions.
+#ifdef CONFIG_DEBUG_OBJECTS_TIMERS
+
+static struct debug_obj_descr timer_debug_descr;
+
+/*
+ * fixup_init is called when:
+ * - an active object is initialized
+ */
+static int timer_fixup_init(void *addr, enum debug_obj_state state)
+{
+       struct timer_list *timer = addr;
+
+       switch (state) {
+       case ODEBUG_STATE_ACTIVE:
+               del_timer_sync(timer);
+               debug_object_init(timer, &timer_debug_descr);
+               return 1;
+       default:
+               return 0;
+       }
+}
+
+/*
+ * fixup_activate is called when:
+ * - an active object is activated
+ * - an unknown object is activated (might be a statically initialized object)
  */
-void init_timer(struct timer_list *timer)
+static int timer_fixup_activate(void *addr, enum debug_obj_state state)
+{
+       struct timer_list *timer = addr;
+
+       switch (state) {
+
+       case ODEBUG_STATE_NOTAVAILABLE:
+               /*
+                * This is not really a fixup. The timer was
+                * statically initialized. We just make sure that it
+                * is tracked in the object tracker.
+                */
+               if (timer->entry.next == NULL &&
+                   timer->entry.prev == TIMER_ENTRY_STATIC) {
+                       debug_object_init(timer, &timer_debug_descr);
+                       debug_object_activate(timer, &timer_debug_descr);
+                       return 0;
+               } else {
+                       WARN_ON_ONCE(1);
+               }
+               return 0;
+
+       case ODEBUG_STATE_ACTIVE:
+               WARN_ON(1);
+
+       default:
+               return 0;
+       }
+}
+
+/*
+ * fixup_free is called when:
+ * - an active object is freed
+ */
+static int timer_fixup_free(void *addr, enum debug_obj_state state)
+{
+       struct timer_list *timer = addr;
+
+       switch (state) {
+       case ODEBUG_STATE_ACTIVE:
+               del_timer_sync(timer);
+               debug_object_free(timer, &timer_debug_descr);
+               return 1;
+       default:
+               return 0;
+       }
+}
+
+static struct debug_obj_descr timer_debug_descr = {
+       .name           = "timer_list",
+       .fixup_init     = timer_fixup_init,
+       .fixup_activate = timer_fixup_activate,
+       .fixup_free     = timer_fixup_free,
+};
+
+static inline void debug_timer_init(struct timer_list *timer)
+{
+       debug_object_init(timer, &timer_debug_descr);
+}
+
+static inline void debug_timer_activate(struct timer_list *timer)
+{
+       debug_object_activate(timer, &timer_debug_descr);
+}
+
+static inline void debug_timer_deactivate(struct timer_list *timer)
+{
+       debug_object_deactivate(timer, &timer_debug_descr);
+}
+
+static inline void debug_timer_free(struct timer_list *timer)
+{
+       debug_object_free(timer, &timer_debug_descr);
+}
+
+static void __init_timer(struct timer_list *timer,
+                        const char *name,
+                        struct lock_class_key *key);
+
+void init_timer_on_stack_key(struct timer_list *timer,
+                            const char *name,
+                            struct lock_class_key *key)
+{
+       debug_object_init_on_stack(timer, &timer_debug_descr);
+       __init_timer(timer, name, key);
+}
+EXPORT_SYMBOL_GPL(init_timer_on_stack_key);
+
+void destroy_timer_on_stack(struct timer_list *timer)
+{
+       debug_object_free(timer, &timer_debug_descr);
+}
+EXPORT_SYMBOL_GPL(destroy_timer_on_stack);
+
+#else
+static inline void debug_timer_init(struct timer_list *timer) { }
+static inline void debug_timer_activate(struct timer_list *timer) { }
+static inline void debug_timer_deactivate(struct timer_list *timer) { }
+#endif
+
+static inline void debug_init(struct timer_list *timer)
+{
+       debug_timer_init(timer);
+       trace_timer_init(timer);
+}
+
+static inline void
+debug_activate(struct timer_list *timer, unsigned long expires)
+{
+       debug_timer_activate(timer);
+       trace_timer_start(timer, expires);
+}
+
+static inline void debug_deactivate(struct timer_list *timer)
+{
+       debug_timer_deactivate(timer);
+       trace_timer_cancel(timer);
+}
+
+static void __init_timer(struct timer_list *timer,
+                        const char *name,
+                        struct lock_class_key *key)
 {
        timer->entry.next = NULL;
        timer->base = __raw_get_cpu_var(tvec_bases);
@@ -336,21 +554,44 @@ void init_timer(struct timer_list *timer)
        timer->start_pid = -1;
        memset(timer->start_comm, 0, TASK_COMM_LEN);
 #endif
+       lockdep_init_map(&timer->lockdep_map, name, key, 0);
+}
+
+/**
+ * init_timer_key - initialize a timer
+ * @timer: the timer to be initialized
+ * @name: name of the timer
+ * @key: lockdep class key of the fake lock used for tracking timer
+ *       sync lock dependencies
+ *
+ * init_timer_key() must be done to a timer prior calling *any* of the
+ * other timer functions.
+ */
+void init_timer_key(struct timer_list *timer,
+                   const char *name,
+                   struct lock_class_key *key)
+{
+       debug_init(timer);
+       __init_timer(timer, name, key);
 }
-EXPORT_SYMBOL(init_timer);
+EXPORT_SYMBOL(init_timer_key);
 
-void init_timer_deferrable(struct timer_list *timer)
+void init_timer_deferrable_key(struct timer_list *timer,
+                              const char *name,
+                              struct lock_class_key *key)
 {
-       init_timer(timer);
+       init_timer_key(timer, name, key);
        timer_set_deferrable(timer);
 }
-EXPORT_SYMBOL(init_timer_deferrable);
+EXPORT_SYMBOL(init_timer_deferrable_key);
 
 static inline void detach_timer(struct timer_list *timer,
                                int clear_pending)
 {
        struct list_head *entry = &timer->entry;
 
+       debug_deactivate(timer);
+
        __list_del(entry->prev, entry->next);
        if (clear_pending)
                entry->next = NULL;
@@ -389,11 +630,13 @@ static struct tvec_base *lock_timer_base(struct timer_list *timer,
        }
 }
 
-int __mod_timer(struct timer_list *timer, unsigned long expires)
+static inline int
+__mod_timer(struct timer_list *timer, unsigned long expires,
+                                               bool pending_only, int pinned)
 {
        struct tvec_base *base, *new_base;
        unsigned long flags;
-       int ret = 0;
+       int ret = 0 , cpu;
 
        timer_stats_timer_set_start_info(timer);
        BUG_ON(!timer->function);
@@ -402,11 +645,31 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
 
        if (timer_pending(timer)) {
                detach_timer(timer, 0);
+               if (timer->expires == base->next_timer &&
+                   !tbase_get_deferrable(timer->base))
+                       base->next_timer = base->timer_jiffies;
                ret = 1;
+       } else {
+               if (pending_only)
+                       goto out_unlock;
        }
 
+       debug_activate(timer, 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.
@@ -426,42 +689,32 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
        }
 
        timer->expires = expires;
+       if (time_before(timer->expires, base->next_timer) &&
+           !tbase_get_deferrable(timer->base))
+               base->next_timer = timer->expires;
        internal_add_timer(base, timer);
+
+out_unlock:
        spin_unlock_irqrestore(&base->lock, flags);
 
        return ret;
 }
 
-EXPORT_SYMBOL(__mod_timer);
-
 /**
- * add_timer_on - start a timer on a particular CPU
- * @timer: the timer to be added
- * @cpu: the CPU to start it on
+ * mod_timer_pending - modify a pending timer's timeout
+ * @timer: the pending timer to be modified
+ * @expires: new timeout in jiffies
  *
- * This is not very scalable on SMP. Double adds are not possible.
+ * mod_timer_pending() is the same for pending timers as mod_timer(),
+ * but will not re-activate and modify already deleted timers.
+ *
+ * It is useful for unserialized use of timers.
  */
-void add_timer_on(struct timer_list *timer, int cpu)
+int mod_timer_pending(struct timer_list *timer, unsigned long expires)
 {
-       struct tvec_base *base = per_cpu(tvec_bases, cpu);
-       unsigned long flags;
-
-       timer_stats_timer_set_start_info(timer);
-       BUG_ON(timer_pending(timer) || !timer->function);
-       spin_lock_irqsave(&base->lock, flags);
-       timer_set_base(timer, base);
-       internal_add_timer(base, timer);
-       /*
-        * Check whether the other CPU is idle and needs to be
-        * triggered to reevaluate the timer wheel when nohz is
-        * active. We are protected against the other CPU fiddling
-        * with the timer by holding the timer base lock. This also
-        * makes sure that a CPU on the way to idle can not evaluate
-        * the timer wheel.
-        */
-       wake_up_idle_cpu(cpu);
-       spin_unlock_irqrestore(&base->lock, flags);
+       return __mod_timer(timer, expires, true, TIMER_NOT_PINNED);
 }
+EXPORT_SYMBOL(mod_timer_pending);
 
 /**
  * mod_timer - modify a timer's timeout
@@ -485,21 +738,94 @@ void add_timer_on(struct timer_list *timer, int cpu)
  */
 int mod_timer(struct timer_list *timer, unsigned long expires)
 {
-       BUG_ON(!timer->function);
-
-       timer_stats_timer_set_start_info(timer);
        /*
         * This is a common optimization triggered by the
         * networking code - if the timer is re-modified
         * to be the same thing then just return:
         */
+       if (timer_pending(timer) && timer->expires == expires)
+               return 1;
+
+       return __mod_timer(timer, expires, false, TIMER_NOT_PINNED);
+}
+EXPORT_SYMBOL(mod_timer);
+
+/**
+ * mod_timer_pinned - modify a timer's timeout
+ * @timer: the timer to be modified
+ * @expires: new timeout in jiffies
+ *
+ * mod_timer_pinned() is a way to update the expire field of an
+ * active timer (if the timer is inactive it will be activated)
+ * and not allow the timer to be migrated to a different CPU.
+ *
+ * mod_timer_pinned(timer, expires) is equivalent to:
+ *
+ *     del_timer(timer); timer->expires = expires; add_timer(timer);
+ */
+int mod_timer_pinned(struct timer_list *timer, unsigned long expires)
+{
        if (timer->expires == expires && timer_pending(timer))
                return 1;
 
-       return __mod_timer(timer, expires);
+       return __mod_timer(timer, expires, false, TIMER_PINNED);
 }
+EXPORT_SYMBOL(mod_timer_pinned);
 
-EXPORT_SYMBOL(mod_timer);
+/**
+ * add_timer - start a timer
+ * @timer: the timer to be added
+ *
+ * The kernel will do a ->function(->data) callback from the
+ * timer interrupt at the ->expires point in the future. The
+ * current time is 'jiffies'.
+ *
+ * The timer's ->expires, ->function (and if the handler uses it, ->data)
+ * fields must be set prior calling this function.
+ *
+ * Timers with an ->expires field in the past will be executed in the next
+ * timer tick.
+ */
+void add_timer(struct timer_list *timer)
+{
+       BUG_ON(timer_pending(timer));
+       mod_timer(timer, timer->expires);
+}
+EXPORT_SYMBOL(add_timer);
+
+/**
+ * add_timer_on - start a timer on a particular CPU
+ * @timer: the timer to be added
+ * @cpu: the CPU to start it on
+ *
+ * This is not very scalable on SMP. Double adds are not possible.
+ */
+void add_timer_on(struct timer_list *timer, int cpu)
+{
+       struct tvec_base *base = per_cpu(tvec_bases, cpu);
+       unsigned long flags;
+
+       timer_stats_timer_set_start_info(timer);
+       BUG_ON(timer_pending(timer) || !timer->function);
+       spin_lock_irqsave(&base->lock, flags);
+       timer_set_base(timer, base);
+       debug_activate(timer, timer->expires);
+       if (time_before(timer->expires, base->next_timer) &&
+           !tbase_get_deferrable(timer->base))
+               base->next_timer = timer->expires;
+       internal_add_timer(base, timer);
+       /*
+        * Check whether the other CPU is idle and needs to be
+        * triggered to reevaluate the timer wheel when nohz is
+        * active. We are protected against the other CPU fiddling
+        * with the timer by holding the timer base lock. This also
+        * makes sure that a CPU on the way to idle can not evaluate
+        * the timer wheel.
+        */
+       wake_up_idle_cpu(cpu);
+       spin_unlock_irqrestore(&base->lock, flags);
+}
+EXPORT_SYMBOL_GPL(add_timer_on);
 
 /**
  * del_timer - deactive a timer.
@@ -523,6 +849,9 @@ int del_timer(struct timer_list *timer)
                base = lock_timer_base(timer, &flags);
                if (timer_pending(timer)) {
                        detach_timer(timer, 1);
+                       if (timer->expires == base->next_timer &&
+                           !tbase_get_deferrable(timer->base))
+                               base->next_timer = base->timer_jiffies;
                        ret = 1;
                }
                spin_unlock_irqrestore(&base->lock, flags);
@@ -530,7 +859,6 @@ int del_timer(struct timer_list *timer)
 
        return ret;
 }
-
 EXPORT_SYMBOL(del_timer);
 
 #ifdef CONFIG_SMP
@@ -557,6 +885,9 @@ int try_to_del_timer_sync(struct timer_list *timer)
        ret = 0;
        if (timer_pending(timer)) {
                detach_timer(timer, 1);
+               if (timer->expires == base->next_timer &&
+                   !tbase_get_deferrable(timer->base))
+                       base->next_timer = base->timer_jiffies;
                ret = 1;
        }
 out:
@@ -564,7 +895,6 @@ out:
 
        return ret;
 }
-
 EXPORT_SYMBOL(try_to_del_timer_sync);
 
 /**
@@ -586,6 +916,15 @@ EXPORT_SYMBOL(try_to_del_timer_sync);
  */
 int del_timer_sync(struct timer_list *timer)
 {
+#ifdef CONFIG_LOCKDEP
+       unsigned long flags;
+
+       local_irq_save(flags);
+       lock_map_acquire(&timer->lockdep_map);
+       lock_map_release(&timer->lockdep_map);
+       local_irq_restore(flags);
+#endif
+
        for (;;) {
                int ret = try_to_del_timer_sync(timer);
                if (ret >= 0)
@@ -593,7 +932,6 @@ int del_timer_sync(struct timer_list *timer)
                cpu_relax();
        }
 }
-
 EXPORT_SYMBOL(del_timer_sync);
 #endif
 
@@ -658,10 +996,38 @@ static inline void __run_timers(struct tvec_base *base)
 
                        set_running_timer(base, timer);
                        detach_timer(timer, 1);
+
                        spin_unlock_irq(&base->lock);
                        {
                                int preempt_count = preempt_count();
+
+#ifdef CONFIG_LOCKDEP
+                               /*
+                                * It is permissible to free the timer from
+                                * inside the function that is called from
+                                * it, this we need to take into account for
+                                * lockdep too. To avoid bogus "held lock
+                                * freed" warnings as well as problems when
+                                * looking into timer->lockdep_map, make a
+                                * copy and use that here.
+                                */
+                               struct lockdep_map lockdep_map =
+                                       timer->lockdep_map;
+#endif
+                               /*
+                                * Couple the lock chain with the lock chain at
+                                * del_timer_sync() by acquiring the lock_map
+                                * around the fn() call here and in
+                                * del_timer_sync().
+                                */
+                               lock_map_acquire(&lockdep_map);
+
+                               trace_timer_expire_entry(timer);
                                fn(data);
+                               trace_timer_expire_exit(timer);
+
+                               lock_map_release(&lockdep_map);
+
                                if (preempt_count != preempt_count()) {
                                        printk(KERN_ERR "huh, entered %p "
                                               "with preempt_count %08x, exited"
@@ -678,11 +1044,11 @@ static inline void __run_timers(struct tvec_base *base)
        spin_unlock_irq(&base->lock);
 }
 
-#if defined(CONFIG_NO_IDLE_HZ) || defined(CONFIG_NO_HZ)
+#ifdef CONFIG_NO_HZ
 /*
  * Find out when the next timer event is due to happen. This
- * is used on S/390 to stop all activity when a cpus is idle.
- * This functions needs to be called disabled.
+ * is used on S/390 to stop all activity when a CPU is idle.
+ * This function needs to be called with interrupts disabled.
  */
 static unsigned long __next_timer_interrupt(struct tvec_base *base)
 {
@@ -727,6 +1093,9 @@ cascade:
                index = slot = timer_jiffies & TVN_MASK;
                do {
                        list_for_each_entry(nte, varp->vec + slot, entry) {
+                               if (tbase_get_deferrable(nte->base))
+                                       continue;
+
                                found = 1;
                                if (time_before(nte->expires, expires))
                                        expires = nte->expires;
@@ -805,7 +1174,9 @@ unsigned long get_next_timer_interrupt(unsigned long now)
        unsigned long expires;
 
        spin_lock(&base->lock);
-       expires = __next_timer_interrupt(base);
+       if (time_before_eq(base->next_timer, base->timer_jiffies))
+               base->next_timer = __next_timer_interrupt(base);
+       expires = base->next_timer;
        spin_unlock(&base->lock);
 
        if (time_before_eq(expires, now))
@@ -813,29 +1184,6 @@ unsigned long get_next_timer_interrupt(unsigned long now)
 
        return cmp_next_hrtimer_event(now, expires);
 }
-
-#ifdef CONFIG_NO_IDLE_HZ
-unsigned long next_timer_interrupt(void)
-{
-       return get_next_timer_interrupt(jiffies);
-}
-#endif
-
-#endif
-
-#ifndef CONFIG_VIRT_CPU_ACCOUNTING
-void account_process_tick(struct task_struct *p, int user_tick)
-{
-       cputime_t one_jiffy = jiffies_to_cputime(1);
-
-       if (user_tick) {
-               account_user_time(p, one_jiffy);
-               account_user_time_scaled(p, cputime_to_scaled(one_jiffy));
-       } else {
-               account_system_time(p, HARDIRQ_OFFSET, one_jiffy);
-               account_system_time_scaled(p, cputime_to_scaled(one_jiffy));
-       }
-}
 #endif
 
 /*
@@ -850,60 +1198,21 @@ void update_process_times(int user_tick)
        /* Note: this timer irq context must be accounted for as well. */
        account_process_tick(p, user_tick);
        run_local_timers();
-       if (rcu_pending(cpu))
-               rcu_check_callbacks(cpu, user_tick);
+       rcu_check_callbacks(cpu, user_tick);
+       printk_tick();
        scheduler_tick();
        run_posix_cpu_timers(p);
 }
 
 /*
- * Nr of active tasks - counted in fixed-point numbers
- */
-static unsigned long count_active_tasks(void)
-{
-       return nr_active() * FIXED_1;
-}
-
-/*
- * Hmm.. Changed this, as the GNU make sources (load.c) seems to
- * imply that avenrun[] is the standard name for this kind of thing.
- * Nothing else seems to be standardized: the fractional size etc
- * all seem to differ on different machines.
- *
- * Requires xtime_lock to access.
- */
-unsigned long avenrun[3];
-
-EXPORT_SYMBOL(avenrun);
-
-/*
- * calc_load - given tick count, update the avenrun load estimates.
- * This is called while holding a write_lock on xtime_lock.
- */
-static inline void calc_load(unsigned long ticks)
-{
-       unsigned long active_tasks; /* fixed-point */
-       static int count = LOAD_FREQ;
-
-       count -= ticks;
-       if (unlikely(count < 0)) {
-               active_tasks = count_active_tasks();
-               do {
-                       CALC_LOAD(avenrun[0], EXP_1, active_tasks);
-                       CALC_LOAD(avenrun[1], EXP_5, active_tasks);
-                       CALC_LOAD(avenrun[2], EXP_15, active_tasks);
-                       count += LOAD_FREQ;
-               } while (count < 0);
-       }
-}
-
-/*
  * This function runs timers and the timer-tq in bottom half context.
  */
 static void run_timer_softirq(struct softirq_action *h)
 {
        struct tvec_base *base = __get_cpu_var(tvec_bases);
 
+       perf_event_do_pending();
+
        hrtimer_run_pending();
 
        if (time_after_eq(jiffies, base->timer_jiffies))
@@ -921,16 +1230,6 @@ void run_local_timers(void)
 }
 
 /*
- * Called by the timer interrupt. xtime_lock must already be taken
- * by the timer IRQ!
- */
-static inline void update_times(unsigned long ticks)
-{
-       update_wall_time();
-       calc_load(ticks);
-}
-
-/*
  * The 64-bit jiffies value is not atomic - you MUST NOT read it
  * without sampling the sequence number in xtime_lock.
  * jiffies is defined in the linker script...
@@ -939,7 +1238,8 @@ static inline void update_times(unsigned long ticks)
 void do_timer(unsigned long ticks)
 {
        jiffies_64 += ticks;
-       update_times(ticks);
+       update_wall_time();
+       calc_global_load();
 }
 
 #ifdef __ARCH_WANT_SYS_ALARM
@@ -948,7 +1248,7 @@ void do_timer(unsigned long ticks)
  * For backwards compatibility?  This can be done in libc so Alpha
  * and all newer ports shouldn't need it.
  */
-asmlinkage unsigned long sys_alarm(unsigned int seconds)
+SYSCALL_DEFINE1(alarm, unsigned int, seconds)
 {
        return alarm_setitimer(seconds);
 }
@@ -971,7 +1271,7 @@ asmlinkage unsigned long sys_alarm(unsigned int seconds)
  *
  * This is SMP safe as current->tgid does not change.
  */
-asmlinkage long sys_getpid(void)
+SYSCALL_DEFINE0(getpid)
 {
        return task_tgid_vnr(current);
 }
@@ -982,7 +1282,7 @@ asmlinkage long sys_getpid(void)
  * value of ->real_parent under rcu_read_lock(), see
  * release_task()->call_rcu(delayed_put_task_struct).
  */
-asmlinkage long sys_getppid(void)
+SYSCALL_DEFINE0(getppid)
 {
        int pid;
 
@@ -993,28 +1293,28 @@ asmlinkage long sys_getppid(void)
        return pid;
 }
 
-asmlinkage long sys_getuid(void)
+SYSCALL_DEFINE0(getuid)
 {
        /* Only we change this so SMP safe */
-       return current->uid;
+       return current_uid();
 }
 
-asmlinkage long sys_geteuid(void)
+SYSCALL_DEFINE0(geteuid)
 {
        /* Only we change this so SMP safe */
-       return current->euid;
+       return current_euid();
 }
 
-asmlinkage long sys_getgid(void)
+SYSCALL_DEFINE0(getgid)
 {
        /* Only we change this so SMP safe */
-       return current->gid;
+       return current_gid();
 }
 
-asmlinkage long sys_getegid(void)
+SYSCALL_DEFINE0(getegid)
 {
        /* Only we change this so SMP safe */
-       return  current->egid;
+       return  current_egid();
 }
 
 #endif
@@ -1086,11 +1386,14 @@ signed long __sched schedule_timeout(signed long timeout)
 
        expire = timeout + jiffies;
 
-       setup_timer(&timer, process_timeout, (unsigned long)current);
-       __mod_timer(&timer, expire);
+       setup_timer_on_stack(&timer, process_timeout, (unsigned long)current);
+       __mod_timer(&timer, expire, false, TIMER_NOT_PINNED);
        schedule();
        del_singleshot_timer_sync(&timer);
 
+       /* Remove the timer from the object tracker */
+       destroy_timer_on_stack(&timer);
+
        timeout = expire - jiffies;
 
  out:
@@ -1124,7 +1427,7 @@ signed long __sched schedule_timeout_uninterruptible(signed long timeout)
 EXPORT_SYMBOL(schedule_timeout_uninterruptible);
 
 /* Thread ID - the internal kernel "pid" */
-asmlinkage long sys_gettid(void)
+SYSCALL_DEFINE0(gettid)
 {
        return task_pid_vnr(current);
 }
@@ -1137,37 +1440,17 @@ int do_sysinfo(struct sysinfo *info)
 {
        unsigned long mem_total, sav_total;
        unsigned int mem_unit, bitcount;
-       unsigned long seq;
+       struct timespec tp;
 
        memset(info, 0, sizeof(struct sysinfo));
 
-       do {
-               struct timespec tp;
-               seq = read_seqbegin(&xtime_lock);
+       ktime_get_ts(&tp);
+       monotonic_to_bootbased(&tp);
+       info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
 
-               /*
-                * This is annoying.  The below is the same thing
-                * posix_get_clock_monotonic() does, but it wants to
-                * take the lock which we want to cover the loads stuff
-                * too.
-                */
-
-               getnstimeofday(&tp);
-               tp.tv_sec += wall_to_monotonic.tv_sec;
-               tp.tv_nsec += wall_to_monotonic.tv_nsec;
-               monotonic_to_bootbased(&tp);
-               if (tp.tv_nsec - NSEC_PER_SEC >= 0) {
-                       tp.tv_nsec = tp.tv_nsec - NSEC_PER_SEC;
-                       tp.tv_sec++;
-               }
-               info->uptime = tp.tv_sec + (tp.tv_nsec ? 1 : 0);
+       get_avenrun(info->loads, 0, SI_LOAD_SHIFT - FSHIFT);
 
-               info->loads[0] = avenrun[0] << (SI_LOAD_SHIFT - FSHIFT);
-               info->loads[1] = avenrun[1] << (SI_LOAD_SHIFT - FSHIFT);
-               info->loads[2] = avenrun[2] << (SI_LOAD_SHIFT - FSHIFT);
-
-               info->procs = nr_threads;
-       } while (read_seqretry(&xtime_lock, seq));
+       info->procs = nr_threads;
 
        si_meminfo(info);
        si_swapinfo(info);
@@ -1216,7 +1499,7 @@ out:
        return 0;
 }
 
-asmlinkage long sys_sysinfo(struct sysinfo __user *info)
+SYSCALL_DEFINE1(sysinfo, struct sysinfo __user *, info)
 {
        struct sysinfo val;
 
@@ -1281,6 +1564,7 @@ static int __cpuinit init_timers_cpu(int cpu)
                INIT_LIST_HEAD(base->tv1.vec + j);
 
        base->timer_jiffies = jiffies;
+       base->next_timer = base->timer_jiffies;
        return 0;
 }
 
@@ -1293,6 +1577,9 @@ static void migrate_timer_list(struct tvec_base *new_base, struct list_head *hea
                timer = list_first_entry(head, struct timer_list, entry);
                detach_timer(timer, 0);
                timer_set_base(timer, new_base);
+               if (time_before(timer->expires, new_base->next_timer) &&
+                   !tbase_get_deferrable(timer->base))
+                       new_base->next_timer = timer->expires;
                internal_add_timer(new_base, timer);
        }
 }
@@ -1306,9 +1593,11 @@ static void __cpuinit migrate_timers(int cpu)
        BUG_ON(cpu_online(cpu));
        old_base = per_cpu(tvec_bases, cpu);
        new_base = get_cpu_var(tvec_bases);
-
-       local_irq_disable();
-       spin_lock(&new_base->lock);
+       /*
+        * The caller is globally serialized and nobody else
+        * takes two locks at once, deadlock is not possible.
+        */
+       spin_lock_irq(&new_base->lock);
        spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
 
        BUG_ON(old_base->running_timer);
@@ -1323,8 +1612,7 @@ static void __cpuinit migrate_timers(int cpu)
        }
 
        spin_unlock(&old_base->lock);
-       spin_unlock(&new_base->lock);
-       local_irq_enable();
+       spin_unlock_irq(&new_base->lock);
        put_cpu_var(tvec_bases);
 }
 #endif /* CONFIG_HOTPLUG_CPU */
@@ -1365,7 +1653,7 @@ void __init init_timers(void)
 
        BUG_ON(err == NOTIFY_BAD);
        register_cpu_notifier(&timers_nb);
-       open_softirq(TIMER_SOFTIRQ, run_timer_softirq, NULL);
+       open_softirq(TIMER_SOFTIRQ, run_timer_softirq);
 }
 
 /**