KVM: SVM: remove now obsolete FIXME comment
[safe/jmp/linux-2.6] / kernel / hrtimer.c
index de93a81..f78777a 100644 (file)
@@ -59,6 +59,7 @@ ktime_t ktime_get(void)
 
        return timespec_to_ktime(now);
 }
+EXPORT_SYMBOL_GPL(ktime_get);
 
 /**
  * ktime_get_real - get the real (wall-) time in ktime_t format
@@ -135,20 +136,17 @@ EXPORT_SYMBOL_GPL(ktime_get_ts);
 static void hrtimer_get_softirq_time(struct hrtimer_cpu_base *base)
 {
        ktime_t xtim, tomono;
-       struct timespec xts;
+       struct timespec xts, tom;
        unsigned long seq;
 
        do {
                seq = read_seqbegin(&xtime_lock);
-#ifdef CONFIG_NO_HZ
-               getnstimeofday(&xts);
-#else
-               xts = xtime;
-#endif
+               xts = current_kernel_time();
+               tom = wall_to_monotonic;
        } while (read_seqretry(&xtime_lock, seq));
 
        xtim = timespec_to_ktime(xts);
-       tomono = timespec_to_ktime(wall_to_monotonic);
+       tomono = timespec_to_ktime(tom);
        base->clock_base[CLOCK_REALTIME].softirq_time = xtim;
        base->clock_base[CLOCK_MONOTONIC].softirq_time =
                ktime_add(xtim, tomono);
@@ -277,12 +275,38 @@ ktime_t ktime_add_ns(const ktime_t kt, u64 nsec)
 
        return ktime_add(kt, tmp);
 }
+
+EXPORT_SYMBOL_GPL(ktime_add_ns);
+
+/**
+ * ktime_sub_ns - Subtract a scalar nanoseconds value from a ktime_t variable
+ * @kt:                minuend
+ * @nsec:      the scalar nsec value to subtract
+ *
+ * Returns the subtraction of @nsec from @kt in ktime_t format
+ */
+ktime_t ktime_sub_ns(const ktime_t kt, u64 nsec)
+{
+       ktime_t tmp;
+
+       if (likely(nsec < NSEC_PER_SEC)) {
+               tmp.tv64 = nsec;
+       } else {
+               unsigned long rem = do_div(nsec, NSEC_PER_SEC);
+
+               tmp = ktime_set((long)nsec, rem);
+       }
+
+       return ktime_sub(kt, tmp);
+}
+
+EXPORT_SYMBOL_GPL(ktime_sub_ns);
 # endif /* !CONFIG_KTIME_SCALAR */
 
 /*
  * Divide a ktime value by a nanosecond value
  */
-unsigned long ktime_divns(const ktime_t kt, s64 div)
+u64 ktime_divns(const ktime_t kt, s64 div)
 {
        u64 dclc, inc, dns;
        int sft = 0;
@@ -297,10 +321,43 @@ unsigned long ktime_divns(const ktime_t kt, s64 div)
        dclc >>= sft;
        do_div(dclc, (unsigned long) div);
 
-       return (unsigned long) dclc;
+       return dclc;
 }
 #endif /* BITS_PER_LONG >= 64 */
 
+/*
+ * Add two ktime values and do a safety check for overflow:
+ */
+ktime_t ktime_add_safe(const ktime_t lhs, const ktime_t rhs)
+{
+       ktime_t res = ktime_add(lhs, rhs);
+
+       /*
+        * We use KTIME_SEC_MAX here, the maximum timeout which we can
+        * return to user space in a timespec:
+        */
+       if (res.tv64 < 0 || res.tv64 < lhs.tv64 || res.tv64 < rhs.tv64)
+               res = ktime_set(KTIME_SEC_MAX, 0);
+
+       return res;
+}
+
+/*
+ * Check, whether the timer is on the callback pending list
+ */
+static inline int hrtimer_cb_pending(const struct hrtimer *timer)
+{
+       return timer->state & HRTIMER_STATE_PENDING;
+}
+
+/*
+ * Remove a timer from the callback pending list
+ */
+static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
+{
+       list_del_init(&timer->cb_entry);
+}
+
 /* High resolution timer related functions */
 #ifdef CONFIG_HIGH_RES_TIMERS
 
@@ -385,16 +442,27 @@ static int hrtimer_reprogram(struct hrtimer *timer,
        ktime_t expires = ktime_sub(timer->expires, base->offset);
        int res;
 
+       WARN_ON_ONCE(timer->expires.tv64 < 0);
+
        /*
         * When the callback is running, we do not reprogram the clock event
         * device. The timer callback is either running on a different CPU or
-        * the callback is executed in the hrtimer_interupt context. The
+        * the callback is executed in the hrtimer_interrupt context. The
         * reprogramming is handled either by the softirq, which called the
         * callback or at the end of the hrtimer_interrupt.
         */
        if (hrtimer_callback_running(timer))
                return 0;
 
+       /*
+        * CLOCK_REALTIME timer might be requested with an absolute
+        * expiry time which is less than base->offset. Nothing wrong
+        * about that, just avoid to call into the tick code, which
+        * has now objections against negative expiry values.
+        */
+       if (expires.tv64 < 0)
+               return -ETIME;
+
        if (expires.tv64 >= expires_next->tv64)
                return 0;
 
@@ -458,19 +526,15 @@ void clock_was_set(void)
 }
 
 /*
- * Check, whether the timer is on the callback pending list
+ * During resume we might have to reprogram the high resolution timer
+ * interrupt (on the local CPU):
  */
-static inline int hrtimer_cb_pending(const struct hrtimer *timer)
+void hres_timers_resume(void)
 {
-       return timer->state & HRTIMER_STATE_PENDING;
-}
+       WARN_ON_ONCE(num_online_cpus() > 1);
 
-/*
- * Remove a timer from the callback pending list
- */
-static inline void hrtimer_remove_cb_pending(struct hrtimer *timer)
-{
-       list_del_init(&timer->cb_entry);
+       /* Retrigger the CPU local events: */
+       retrigger_next_event(NULL);
 }
 
 /*
@@ -480,7 +544,6 @@ static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
 {
        base->expires_next.tv64 = KTIME_MAX;
        base->hres_active = 0;
-       INIT_LIST_HEAD(&base->cb_pending);
 }
 
 /*
@@ -488,7 +551,6 @@ static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base)
  */
 static inline void hrtimer_init_timer_hres(struct hrtimer *timer)
 {
-       INIT_LIST_HEAD(&timer->cb_entry);
 }
 
 /*
@@ -540,19 +602,22 @@ static inline int hrtimer_enqueue_reprogram(struct hrtimer *timer,
 /*
  * Switch to high resolution mode
  */
-static void hrtimer_switch_to_hres(void)
+static int hrtimer_switch_to_hres(void)
 {
-       struct hrtimer_cpu_base *base = &__get_cpu_var(hrtimer_bases);
+       int cpu = smp_processor_id();
+       struct hrtimer_cpu_base *base = &per_cpu(hrtimer_bases, cpu);
        unsigned long flags;
 
        if (base->hres_active)
-               return;
+               return 1;
 
        local_irq_save(flags);
 
        if (tick_init_highres()) {
                local_irq_restore(flags);
-               return;
+               printk(KERN_WARNING "Could not switch to high resolution "
+                                   "mode on CPU %d\n", cpu);
+               return 0;
        }
        base->hres_active = 1;
        base->clock_base[CLOCK_REALTIME].resolution = KTIME_HIGH_RES;
@@ -563,25 +628,29 @@ static void hrtimer_switch_to_hres(void)
        /* "Retrigger" the interrupt to get things going */
        retrigger_next_event(NULL);
        local_irq_restore(flags);
-       printk(KERN_INFO "Switched to high resolution mode on CPU %d\n",
+       printk(KERN_DEBUG "Switched to high resolution mode on CPU %d\n",
               smp_processor_id());
+       return 1;
 }
 
 #else
 
 static inline int hrtimer_hres_active(void) { return 0; }
 static inline int hrtimer_is_hres_enabled(void) { return 0; }
-static inline void hrtimer_switch_to_hres(void) { }
+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)
 {
        return 0;
 }
-static inline int hrtimer_cb_pending(struct hrtimer *timer) { return 0; }
-static inline void hrtimer_remove_cb_pending(struct hrtimer *timer) { }
 static inline void hrtimer_init_hres(struct hrtimer_cpu_base *base) { }
 static inline void hrtimer_init_timer_hres(struct hrtimer *timer) { }
+static inline int hrtimer_reprogram(struct hrtimer *timer,
+                                   struct hrtimer_clock_base *base)
+{
+       return 0;
+}
 
 #endif /* CONFIG_HIGH_RES_TIMERS */
 
@@ -598,7 +667,7 @@ void __timer_stats_hrtimer_set_start_info(struct hrtimer *timer, void *addr)
 #endif
 
 /*
- * Counterpart to lock_timer_base above:
+ * Counterpart to lock_hrtimer_base above:
  */
 static inline
 void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
@@ -615,10 +684,9 @@ void unlock_hrtimer_base(const struct hrtimer *timer, unsigned long *flags)
  * Forward the timer expiry so it will expire in the future.
  * Returns the number of overruns.
  */
-unsigned long
-hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
+u64 hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
 {
-       unsigned long orun = 1;
+       u64 orun = 1;
        ktime_t delta;
 
        delta = ktime_sub(now, timer->expires);
@@ -642,10 +710,11 @@ hrtimer_forward(struct hrtimer *timer, ktime_t now, ktime_t interval)
                 */
                orun++;
        }
-       timer->expires = ktime_add(timer->expires, interval);
+       timer->expires = ktime_add_safe(timer->expires, interval);
 
        return orun;
 }
+EXPORT_SYMBOL_GPL(hrtimer_forward);
 
 /*
  * enqueue_hrtimer - internal function to (re)start a timer
@@ -659,6 +728,7 @@ static void enqueue_hrtimer(struct hrtimer *timer,
        struct rb_node **link = &base->active.rb_node;
        struct rb_node *parent = NULL;
        struct hrtimer *entry;
+       int leftmost = 1;
 
        /*
         * Find the right place in the rbtree:
@@ -670,18 +740,19 @@ static void enqueue_hrtimer(struct hrtimer *timer,
                 * We dont care about collisions. Nodes with
                 * the same expiry time stay together.
                 */
-               if (timer->expires.tv64 < entry->expires.tv64)
+               if (timer->expires.tv64 < entry->expires.tv64) {
                        link = &(*link)->rb_left;
-               else
+               } else {
                        link = &(*link)->rb_right;
+                       leftmost = 0;
+               }
        }
 
        /*
         * Insert the timer to the rbtree and check whether it
         * replaces the first pending timer
         */
-       if (!base->first || timer->expires.tv64 <
-           rb_entry(base->first, struct hrtimer, node)->expires.tv64) {
+       if (leftmost) {
                /*
                 * Reprogram the clock event device. When the timer is already
                 * expired hrtimer_enqueue_reprogram has either called the
@@ -790,7 +861,7 @@ hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
        new_base = switch_hrtimer_base(timer, base);
 
        if (mode == HRTIMER_MODE_REL) {
-               tim = ktime_add(tim, new_base->get_time());
+               tim = ktime_add_safe(tim, new_base->get_time());
                /*
                 * CONFIG_TIME_LOW_RES is a temporary way for architectures
                 * to signal that they simply return xtime in
@@ -799,14 +870,19 @@ hrtimer_start(struct hrtimer *timer, ktime_t tim, const enum hrtimer_mode mode)
                 * timeouts. This will go away with the GTOD framework.
                 */
 #ifdef CONFIG_TIME_LOW_RES
-               tim = ktime_add(tim, base->resolution);
+               tim = ktime_add_safe(tim, base->resolution);
 #endif
        }
        timer->expires = tim;
 
        timer_stats_hrtimer_set_start_info(timer);
 
-       enqueue_hrtimer(timer, new_base, base == new_base);
+       /*
+        * Only allow reprogramming if the new base is on this CPU.
+        * (it might still be on another CPU if the timer was pending)
+        */
+       enqueue_hrtimer(timer, new_base,
+                       new_base->cpu_base == &__get_cpu_var(hrtimer_bases));
 
        unlock_hrtimer_base(timer, &flags);
 
@@ -939,6 +1015,7 @@ 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
@@ -968,6 +1045,85 @@ int hrtimer_get_res(const clockid_t which_clock, struct timespec *tp)
 }
 EXPORT_SYMBOL_GPL(hrtimer_get_res);
 
+static void run_hrtimer_pending(struct hrtimer_cpu_base *cpu_base)
+{
+       spin_lock_irq(&cpu_base->lock);
+
+       while (!list_empty(&cpu_base->cb_pending)) {
+               enum hrtimer_restart (*fn)(struct hrtimer *);
+               struct hrtimer *timer;
+               int restart;
+
+               timer = list_entry(cpu_base->cb_pending.next,
+                                  struct hrtimer, cb_entry);
+
+               timer_stats_account_hrtimer(timer);
+
+               fn = timer->function;
+               __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
+               spin_unlock_irq(&cpu_base->lock);
+
+               restart = fn(timer);
+
+               spin_lock_irq(&cpu_base->lock);
+
+               timer->state &= ~HRTIMER_STATE_CALLBACK;
+               if (restart == HRTIMER_RESTART) {
+                       BUG_ON(hrtimer_active(timer));
+                       /*
+                        * Enqueue the timer, allow reprogramming of the event
+                        * device
+                        */
+                       enqueue_hrtimer(timer, timer->base, 1);
+               } else if (hrtimer_active(timer)) {
+                       /*
+                        * If the timer was rearmed on another CPU, reprogram
+                        * the event device.
+                        */
+                       if (timer->base->first == &timer->node)
+                               hrtimer_reprogram(timer, timer->base);
+               }
+       }
+       spin_unlock_irq(&cpu_base->lock);
+}
+
+static void __run_hrtimer(struct hrtimer *timer)
+{
+       struct hrtimer_clock_base *base = timer->base;
+       struct hrtimer_cpu_base *cpu_base = base->cpu_base;
+       enum hrtimer_restart (*fn)(struct hrtimer *);
+       int restart;
+
+       __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
+       timer_stats_account_hrtimer(timer);
+
+       fn = timer->function;
+       if (timer->cb_mode == HRTIMER_CB_IRQSAFE_NO_SOFTIRQ) {
+               /*
+                * Used for scheduler timers, avoid lock inversion with
+                * rq->lock and tasklist_lock.
+                *
+                * These timers are required to deal with enqueue expiry
+                * themselves and are not allowed to migrate.
+                */
+               spin_unlock(&cpu_base->lock);
+               restart = fn(timer);
+               spin_lock(&cpu_base->lock);
+       } else
+               restart = fn(timer);
+
+       /*
+        * Note: We clear the CALLBACK bit after enqueue_hrtimer to avoid
+        * reprogramming of the event hardware. This happens at the end of this
+        * function anyway.
+        */
+       if (restart != HRTIMER_NORESTART) {
+               BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
+               enqueue_hrtimer(timer, base, 0);
+       }
+       timer->state &= ~HRTIMER_STATE_CALLBACK;
+}
+
 #ifdef CONFIG_HIGH_RES_TIMERS
 
 /*
@@ -1025,21 +1181,7 @@ void hrtimer_interrupt(struct clock_event_device *dev)
                                continue;
                        }
 
-                       __remove_hrtimer(timer, base,
-                                        HRTIMER_STATE_CALLBACK, 0);
-                       timer_stats_account_hrtimer(timer);
-
-                       /*
-                        * Note: We clear the CALLBACK bit after
-                        * enqueue_hrtimer to avoid reprogramming of
-                        * the event hardware. This happens at the end
-                        * of this function anyway.
-                        */
-                       if (timer->function(timer) != HRTIMER_NORESTART) {
-                               BUG_ON(timer->state != HRTIMER_STATE_CALLBACK);
-                               enqueue_hrtimer(timer, base, 0);
-                       }
-                       timer->state &= ~HRTIMER_STATE_CALLBACK;
+                       __run_hrtimer(timer);
                }
                spin_unlock(&cpu_base->lock);
                base++;
@@ -1060,106 +1202,21 @@ void hrtimer_interrupt(struct clock_event_device *dev)
 
 static void run_hrtimer_softirq(struct softirq_action *h)
 {
-       struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
-
-       spin_lock_irq(&cpu_base->lock);
-
-       while (!list_empty(&cpu_base->cb_pending)) {
-               enum hrtimer_restart (*fn)(struct hrtimer *);
-               struct hrtimer *timer;
-               int restart;
-
-               timer = list_entry(cpu_base->cb_pending.next,
-                                  struct hrtimer, cb_entry);
-
-               timer_stats_account_hrtimer(timer);
-
-               fn = timer->function;
-               __remove_hrtimer(timer, timer->base, HRTIMER_STATE_CALLBACK, 0);
-               spin_unlock_irq(&cpu_base->lock);
-
-               restart = fn(timer);
-
-               spin_lock_irq(&cpu_base->lock);
-
-               timer->state &= ~HRTIMER_STATE_CALLBACK;
-               if (restart == HRTIMER_RESTART) {
-                       BUG_ON(hrtimer_active(timer));
-                       /*
-                        * Enqueue the timer, allow reprogramming of the event
-                        * device
-                        */
-                       enqueue_hrtimer(timer, timer->base, 1);
-               } else if (hrtimer_active(timer)) {
-                       /*
-                        * If the timer was rearmed on another CPU, reprogram
-                        * the event device.
-                        */
-                       if (timer->base->first == &timer->node)
-                               hrtimer_reprogram(timer, timer->base);
-               }
-       }
-       spin_unlock_irq(&cpu_base->lock);
+       run_hrtimer_pending(&__get_cpu_var(hrtimer_bases));
 }
 
 #endif /* CONFIG_HIGH_RES_TIMERS */
 
 /*
- * Expire the per base hrtimer-queue:
- */
-static inline void run_hrtimer_queue(struct hrtimer_cpu_base *cpu_base,
-                                    int index)
-{
-       struct rb_node *node;
-       struct hrtimer_clock_base *base = &cpu_base->clock_base[index];
-
-       if (!base->first)
-               return;
-
-       if (base->get_softirq_time)
-               base->softirq_time = base->get_softirq_time();
-
-       spin_lock_irq(&cpu_base->lock);
-
-       while ((node = base->first)) {
-               struct hrtimer *timer;
-               enum hrtimer_restart (*fn)(struct hrtimer *);
-               int restart;
-
-               timer = rb_entry(node, struct hrtimer, node);
-               if (base->softirq_time.tv64 <= timer->expires.tv64)
-                       break;
-
-               timer_stats_account_hrtimer(timer);
-
-               fn = timer->function;
-               __remove_hrtimer(timer, base, HRTIMER_STATE_CALLBACK, 0);
-               spin_unlock_irq(&cpu_base->lock);
-
-               restart = fn(timer);
-
-               spin_lock_irq(&cpu_base->lock);
-
-               timer->state &= ~HRTIMER_STATE_CALLBACK;
-               if (restart != HRTIMER_NORESTART) {
-                       BUG_ON(hrtimer_active(timer));
-                       enqueue_hrtimer(timer, base, 0);
-               }
-       }
-       spin_unlock_irq(&cpu_base->lock);
-}
-
-/*
  * Called from timer softirq every jiffy, expire hrtimers:
  *
  * For HRT its the fall back code to run the softirq in the timer
  * softirq context in case the hrtimer initialization failed or has
  * not been done yet.
  */
-void hrtimer_run_queues(void)
+void hrtimer_run_pending(void)
 {
        struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
-       int i;
 
        if (hrtimer_hres_active())
                return;
@@ -1175,10 +1232,56 @@ void hrtimer_run_queues(void)
        if (tick_check_oneshot_change(!hrtimer_is_hres_enabled()))
                hrtimer_switch_to_hres();
 
-       hrtimer_get_softirq_time(cpu_base);
+       run_hrtimer_pending(cpu_base);
+}
 
-       for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
-               run_hrtimer_queue(cpu_base, i);
+/*
+ * Called from hardirq context every jiffy
+ */
+void hrtimer_run_queues(void)
+{
+       struct rb_node *node;
+       struct hrtimer_cpu_base *cpu_base = &__get_cpu_var(hrtimer_bases);
+       struct hrtimer_clock_base *base;
+       int index, gettime = 1;
+
+       if (hrtimer_hres_active())
+               return;
+
+       for (index = 0; index < HRTIMER_MAX_CLOCK_BASES; index++) {
+               base = &cpu_base->clock_base[index];
+
+               if (!base->first)
+                       continue;
+
+               if (base->get_softirq_time)
+                       base->softirq_time = base->get_softirq_time();
+               else if (gettime) {
+                       hrtimer_get_softirq_time(cpu_base);
+                       gettime = 0;
+               }
+
+               spin_lock(&cpu_base->lock);
+
+               while ((node = base->first)) {
+                       struct hrtimer *timer;
+
+                       timer = rb_entry(node, struct hrtimer, node);
+                       if (base->softirq_time.tv64 <= timer->expires.tv64)
+                               break;
+
+                       if (timer->cb_mode == HRTIMER_CB_SOFTIRQ) {
+                               __remove_hrtimer(timer, base,
+                                       HRTIMER_STATE_PENDING, 0);
+                               list_add_tail(&timer->cb_entry,
+                                       &base->cpu_base->cb_pending);
+                               continue;
+                       }
+
+                       __run_hrtimer(timer);
+               }
+               spin_unlock(&cpu_base->lock);
+       }
 }
 
 /*
@@ -1202,7 +1305,7 @@ void hrtimer_init_sleeper(struct hrtimer_sleeper *sl, struct task_struct *task)
        sl->timer.function = hrtimer_wakeup;
        sl->task = task;
 #ifdef CONFIG_HIGH_RES_TIMERS
-       sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_RESTART;
+       sl->timer.cb_mode = HRTIMER_CB_IRQSAFE_NO_SOFTIRQ;
 #endif
 }
 
@@ -1213,6 +1316,8 @@ static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mod
        do {
                set_current_state(TASK_INTERRUPTIBLE);
                hrtimer_start(&t->timer, t->timer.expires, mode);
+               if (!hrtimer_active(&t->timer))
+                       t->task = NULL;
 
                if (likely(t->task))
                        schedule();
@@ -1222,36 +1327,45 @@ static int __sched do_nanosleep(struct hrtimer_sleeper *t, enum hrtimer_mode mod
 
        } while (t->task && !signal_pending(current));
 
+       __set_current_state(TASK_RUNNING);
+
        return t->task == NULL;
 }
 
+static int update_rmtp(struct hrtimer *timer, struct timespec __user *rmtp)
+{
+       struct timespec rmt;
+       ktime_t rem;
+
+       rem = ktime_sub(timer->expires, timer->base->get_time());
+       if (rem.tv64 <= 0)
+               return 0;
+       rmt = ktime_to_timespec(rem);
+
+       if (copy_to_user(rmtp, &rmt, sizeof(*rmtp)))
+               return -EFAULT;
+
+       return 1;
+}
+
 long __sched hrtimer_nanosleep_restart(struct restart_block *restart)
 {
        struct hrtimer_sleeper t;
-       struct timespec __user *rmtp;
-       struct timespec tu;
-       ktime_t time;
-
-       restart->fn = do_no_restart_syscall;
+       struct timespec __user  *rmtp;
 
-       hrtimer_init(&t.timer, restart->arg0, HRTIMER_MODE_ABS);
-       t.timer.expires.tv64 = ((u64)restart->arg3 << 32) | (u64) restart->arg2;
+       hrtimer_init(&t.timer, restart->nanosleep.index, HRTIMER_MODE_ABS);
+       t.timer.expires.tv64 = restart->nanosleep.expires;
 
        if (do_nanosleep(&t, HRTIMER_MODE_ABS))
                return 0;
 
-       rmtp = (struct timespec __user *) restart->arg1;
+       rmtp = restart->nanosleep.rmtp;
        if (rmtp) {
-               time = ktime_sub(t.timer.expires, t.timer.base->get_time());
-               if (time.tv64 <= 0)
-                       return 0;
-               tu = ktime_to_timespec(time);
-               if (copy_to_user(rmtp, &tu, sizeof(tu)))
-                       return -EFAULT;
+               int ret = update_rmtp(&t.timer, rmtp);
+               if (ret <= 0)
+                       return ret;
        }
 
-       restart->fn = hrtimer_nanosleep_restart;
-
        /* The other values in restart are already filled in */
        return -ERESTART_RESTARTBLOCK;
 }
@@ -1261,8 +1375,6 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
 {
        struct restart_block *restart;
        struct hrtimer_sleeper t;
-       struct timespec tu;
-       ktime_t rem;
 
        hrtimer_init(&t.timer, clockid, mode);
        t.timer.expires = timespec_to_ktime(*rqtp);
@@ -1274,20 +1386,16 @@ long hrtimer_nanosleep(struct timespec *rqtp, struct timespec __user *rmtp,
                return -ERESTARTNOHAND;
 
        if (rmtp) {
-               rem = ktime_sub(t.timer.expires, t.timer.base->get_time());
-               if (rem.tv64 <= 0)
-                       return 0;
-               tu = ktime_to_timespec(rem);
-               if (copy_to_user(rmtp, &tu, sizeof(tu)))
-                       return -EFAULT;
+               int ret = update_rmtp(&t.timer, rmtp);
+               if (ret <= 0)
+                       return ret;
        }
 
        restart = &current_thread_info()->restart_block;
        restart->fn = hrtimer_nanosleep_restart;
-       restart->arg0 = (unsigned long) t.timer.base->index;
-       restart->arg1 = (unsigned long) rmtp;
-       restart->arg2 = t.timer.expires.tv64 & 0xFFFFFFFF;
-       restart->arg3 = t.timer.expires.tv64 >> 32;
+       restart->nanosleep.index = t.timer.base->index;
+       restart->nanosleep.rmtp = rmtp;
+       restart->nanosleep.expires = t.timer.expires.tv64;
 
        return -ERESTART_RESTARTBLOCK;
 }
@@ -1309,17 +1417,17 @@ sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
 /*
  * Functions related to boot-time initialization:
  */
-static void __devinit init_hrtimers_cpu(int cpu)
+static void __cpuinit init_hrtimers_cpu(int cpu)
 {
        struct hrtimer_cpu_base *cpu_base = &per_cpu(hrtimer_bases, cpu);
        int i;
 
        spin_lock_init(&cpu_base->lock);
-       lockdep_set_class(&cpu_base->lock, &cpu_base->lock_key);
 
        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++)
                cpu_base->clock_base[i].cpu_base = cpu_base;
 
+       INIT_LIST_HEAD(&cpu_base->cb_pending);
        hrtimer_init_hres(cpu_base);
 }
 
@@ -1355,16 +1463,16 @@ static void migrate_hrtimers(int cpu)
        tick_cancel_sched_timer(cpu);
 
        local_irq_disable();
-       double_spin_lock(&new_base->lock, &old_base->lock,
-                        smp_processor_id() < cpu);
+       spin_lock(&new_base->lock);
+       spin_lock_nested(&old_base->lock, SINGLE_DEPTH_NESTING);
 
        for (i = 0; i < HRTIMER_MAX_CLOCK_BASES; i++) {
                migrate_hrtimer_list(&old_base->clock_base[i],
                                     &new_base->clock_base[i]);
        }
 
-       double_spin_unlock(&new_base->lock, &old_base->lock,
-                          smp_processor_id() < cpu);
+       spin_unlock(&old_base->lock);
+       spin_unlock(&new_base->lock);
        local_irq_enable();
        put_cpu_var(hrtimer_bases);
 }
@@ -1373,16 +1481,18 @@ static void migrate_hrtimers(int cpu)
 static int __cpuinit hrtimer_cpu_notify(struct notifier_block *self,
                                        unsigned long action, void *hcpu)
 {
-       long cpu = (long)hcpu;
+       unsigned int cpu = (long)hcpu;
 
        switch (action) {
 
        case CPU_UP_PREPARE:
+       case CPU_UP_PREPARE_FROZEN:
                init_hrtimers_cpu(cpu);
                break;
 
 #ifdef CONFIG_HOTPLUG_CPU
        case CPU_DEAD:
+       case CPU_DEAD_FROZEN:
                clockevents_notify(CLOCK_EVT_NOTIFY_CPU_DEAD, &cpu);
                migrate_hrtimers(cpu);
                break;