[WAN]: Added missing netif_dormant_off() to generic HDLC
[safe/jmp/linux-2.6] / kernel / timer.c
index 1433d87..05809c2 100644 (file)
@@ -33,6 +33,7 @@
 #include <linux/posix-timers.h>
 #include <linux/cpu.h>
 #include <linux/syscalls.h>
+#include <linux/delay.h>
 
 #include <asm/uaccess.h>
 #include <asm/unistd.h>
@@ -46,10 +47,13 @@ static void time_interpolator_update(long delta_nsec);
 #define time_interpolator_update(x)
 #endif
 
+u64 jiffies_64 __cacheline_aligned_in_smp = INITIAL_JIFFIES;
+
+EXPORT_SYMBOL(jiffies_64);
+
 /*
  * per-CPU timer vector definitions:
  */
-
 #define TVN_BITS (CONFIG_BASE_SMALL ? 4 : 6)
 #define TVR_BITS (CONFIG_BASE_SMALL ? 6 : 8)
 #define TVN_SIZE (1 << TVN_BITS)
@@ -57,11 +61,6 @@ static void time_interpolator_update(long delta_nsec);
 #define TVN_MASK (TVN_SIZE - 1)
 #define TVR_MASK (TVR_SIZE - 1)
 
-struct timer_base_s {
-       spinlock_t lock;
-       struct timer_list *running_timer;
-};
-
 typedef struct tvec_s {
        struct list_head vec[TVN_SIZE];
 } tvec_t;
@@ -71,7 +70,8 @@ typedef struct tvec_root_s {
 } tvec_root_t;
 
 struct tvec_t_base_s {
-       struct timer_base_s t_base;
+       spinlock_t lock;
+       struct timer_list *running_timer;
        unsigned long timer_jiffies;
        tvec_root_t tv1;
        tvec_t tv2;
@@ -81,40 +81,19 @@ struct tvec_t_base_s {
 } ____cacheline_aligned_in_smp;
 
 typedef struct tvec_t_base_s tvec_base_t;
-static DEFINE_PER_CPU(tvec_base_t, tvec_bases);
+
+tvec_base_t boot_tvec_bases;
+EXPORT_SYMBOL(boot_tvec_bases);
+static DEFINE_PER_CPU(tvec_base_t *, tvec_bases) = { &boot_tvec_bases };
 
 static inline void set_running_timer(tvec_base_t *base,
                                        struct timer_list *timer)
 {
 #ifdef CONFIG_SMP
-       base->t_base.running_timer = timer;
+       base->running_timer = timer;
 #endif
 }
 
-static void check_timer_failed(struct timer_list *timer)
-{
-       static int whine_count;
-       if (whine_count < 16) {
-               whine_count++;
-               printk("Uninitialised timer!\n");
-               printk("This is just a warning.  Your computer is OK\n");
-               printk("function=0x%p, data=0x%lx\n",
-                       timer->function, timer->data);
-               dump_stack();
-       }
-       /*
-        * Now fix it up
-        */
-       timer->magic = TIMER_MAGIC;
-}
-
-static inline void check_timer(struct timer_list *timer)
-{
-       if (timer->magic != TIMER_MAGIC)
-               check_timer_failed(timer);
-}
-
-
 static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
 {
        unsigned long expires = timer->expires;
@@ -157,15 +136,6 @@ static void internal_add_timer(tvec_base_t *base, struct timer_list *timer)
        list_add_tail(&timer->entry, vec);
 }
 
-typedef struct timer_base_s timer_base_t;
-/*
- * Used by TIMER_INITIALIZER, we can't use per_cpu(tvec_bases)
- * at compile time, and we need timer->base to lock the timer.
- */
-timer_base_t __init_timer_base
-       ____cacheline_aligned_in_smp = { .lock = SPIN_LOCK_UNLOCKED };
-EXPORT_SYMBOL(__init_timer_base);
-
 /***
  * init_timer - initialize a timer.
  * @timer: the timer to be initialized
@@ -176,8 +146,7 @@ EXPORT_SYMBOL(__init_timer_base);
 void fastcall init_timer(struct timer_list *timer)
 {
        timer->entry.next = NULL;
-       timer->base = &per_cpu(tvec_bases, raw_smp_processor_id()).t_base;
-       timer->magic = TIMER_MAGIC;
+       timer->base = __raw_get_cpu_var(tvec_bases);
 }
 EXPORT_SYMBOL(init_timer);
 
@@ -193,7 +162,7 @@ static inline void detach_timer(struct timer_list *timer,
 }
 
 /*
- * We are using hashed locking: holding per_cpu(tvec_bases).t_base.lock
+ * We are using hashed locking: holding per_cpu(tvec_bases).lock
  * means that all timers which are tied to this base via timer->base are
  * locked, and the base itself is locked too.
  *
@@ -204,10 +173,10 @@ static inline void detach_timer(struct timer_list *timer,
  * possible to set timer->base = NULL and drop the lock: the timer remains
  * locked.
  */
-static timer_base_t *lock_timer_base(struct timer_list *timer,
+static tvec_base_t *lock_timer_base(struct timer_list *timer,
                                        unsigned long *flags)
 {
-       timer_base_t *base;
+       tvec_base_t *base;
 
        for (;;) {
                base = timer->base;
@@ -224,13 +193,11 @@ static timer_base_t *lock_timer_base(struct timer_list *timer,
 
 int __mod_timer(struct timer_list *timer, unsigned long expires)
 {
-       timer_base_t *base;
-       tvec_base_t *new_base;
+       tvec_base_t *base, *new_base;
        unsigned long flags;
        int ret = 0;
 
        BUG_ON(!timer->function);
-       check_timer(timer);
 
        base = lock_timer_base(timer, &flags);
 
@@ -239,9 +206,9 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
                ret = 1;
        }
 
-       new_base = &__get_cpu_var(tvec_bases);
+       new_base = __get_cpu_var(tvec_bases);
 
-       if (base != &new_base->t_base) {
+       if (base != new_base) {
                /*
                 * We are trying to schedule the timer on the local CPU.
                 * However we can't change timer's base while it is running,
@@ -249,21 +216,19 @@ int __mod_timer(struct timer_list *timer, unsigned long expires)
                 * handler yet has not finished. This also guarantees that
                 * the timer is serialized wrt itself.
                 */
-               if (unlikely(base->running_timer == timer)) {
-                       /* The timer remains on a former base */
-                       new_base = container_of(base, tvec_base_t, t_base);
-               } else {
+               if (likely(base->running_timer != timer)) {
                        /* See the comment in lock_timer_base() */
                        timer->base = NULL;
                        spin_unlock(&base->lock);
-                       spin_lock(&new_base->t_base.lock);
-                       timer->base = &new_base->t_base;
+                       base = new_base;
+                       spin_lock(&base->lock);
+                       timer->base = base;
                }
        }
 
        timer->expires = expires;
-       internal_add_timer(new_base, timer);
-       spin_unlock_irqrestore(&new_base->t_base.lock, flags);
+       internal_add_timer(base, timer);
+       spin_unlock_irqrestore(&base->lock, flags);
 
        return ret;
 }
@@ -279,17 +244,14 @@ EXPORT_SYMBOL(__mod_timer);
  */
 void add_timer_on(struct timer_list *timer, int cpu)
 {
-       tvec_base_t *base = &per_cpu(tvec_bases, cpu);
+       tvec_base_t *base = per_cpu(tvec_bases, cpu);
        unsigned long flags;
 
        BUG_ON(timer_pending(timer) || !timer->function);
-
-       check_timer(timer);
-
-       spin_lock_irqsave(&base->t_base.lock, flags);
-       timer->base = &base->t_base;
+       spin_lock_irqsave(&base->lock, flags);
+       timer->base = base;
        internal_add_timer(base, timer);
-       spin_unlock_irqrestore(&base->t_base.lock, flags);
+       spin_unlock_irqrestore(&base->lock, flags);
 }
 
 
@@ -316,8 +278,6 @@ int mod_timer(struct timer_list *timer, unsigned long expires)
 {
        BUG_ON(!timer->function);
 
-       check_timer(timer);
-
        /*
         * This is a common optimization triggered by the
         * networking code - if the timer is re-modified
@@ -344,12 +304,10 @@ EXPORT_SYMBOL(mod_timer);
  */
 int del_timer(struct timer_list *timer)
 {
-       timer_base_t *base;
+       tvec_base_t *base;
        unsigned long flags;
        int ret = 0;
 
-       check_timer(timer);
-
        if (timer_pending(timer)) {
                base = lock_timer_base(timer, &flags);
                if (timer_pending(timer)) {
@@ -373,7 +331,7 @@ EXPORT_SYMBOL(del_timer);
  */
 int try_to_del_timer_sync(struct timer_list *timer)
 {
-       timer_base_t *base;
+       tvec_base_t *base;
        unsigned long flags;
        int ret = -1;
 
@@ -412,12 +370,11 @@ out:
  */
 int del_timer_sync(struct timer_list *timer)
 {
-       check_timer(timer);
-
        for (;;) {
                int ret = try_to_del_timer_sync(timer);
                if (ret >= 0)
                        return ret;
+               cpu_relax();
        }
 }
 
@@ -427,23 +384,19 @@ EXPORT_SYMBOL(del_timer_sync);
 static int cascade(tvec_base_t *base, tvec_t *tv, int index)
 {
        /* cascade all the timers from tv up one level */
-       struct list_head *head, *curr;
+       struct timer_list *timer, *tmp;
+       struct list_head tv_list;
+
+       list_replace_init(tv->vec + index, &tv_list);
 
-       head = tv->vec + index;
-       curr = head->next;
        /*
-        * We are removing _all_ timers from the list, so we don't  have to
-        * detach them individually, just clear the list afterwards.
+        * We are removing _all_ timers from the list, so we
+        * don't have to detach them individually.
         */
-       while (curr != head) {
-               struct timer_list *tmp;
-
-               tmp = list_entry(curr, struct timer_list, entry);
-               BUG_ON(tmp->base != &base->t_base);
-               curr = curr->next;
-               internal_add_timer(base, tmp);
+       list_for_each_entry_safe(timer, tmp, &tv_list, entry) {
+               BUG_ON(timer->base != base);
+               internal_add_timer(base, timer);
        }
-       INIT_LIST_HEAD(head);
 
        return index;
 }
@@ -461,12 +414,12 @@ static inline void __run_timers(tvec_base_t *base)
 {
        struct timer_list *timer;
 
-       spin_lock_irq(&base->t_base.lock);
+       spin_lock_irq(&base->lock);
        while (time_after_eq(jiffies, base->timer_jiffies)) {
-               struct list_head work_list = LIST_HEAD_INIT(work_list);
+               struct list_head work_list;
                struct list_head *head = &work_list;
                int index = base->timer_jiffies & TVR_MASK;
+
                /*
                 * Cascade timers:
                 */
@@ -475,8 +428,8 @@ static inline void __run_timers(tvec_base_t *base)
                                (!cascade(base, &base->tv3, INDEX(1))) &&
                                        !cascade(base, &base->tv4, INDEX(2)))
                        cascade(base, &base->tv5, INDEX(3));
-               ++base->timer_jiffies; 
-               list_splice_init(base->tv1.vec + index, &work_list);
+               ++base->timer_jiffies;
+               list_replace_init(base->tv1.vec + index, &work_list);
                while (!list_empty(head)) {
                        void (*fn)(unsigned long);
                        unsigned long data;
@@ -487,7 +440,7 @@ static inline void __run_timers(tvec_base_t *base)
 
                        set_running_timer(base, timer);
                        detach_timer(timer, 1);
-                       spin_unlock_irq(&base->t_base.lock);
+                       spin_unlock_irq(&base->lock);
                        {
                                int preempt_count = preempt_count();
                                fn(data);
@@ -500,11 +453,11 @@ static inline void __run_timers(tvec_base_t *base)
                                        BUG();
                                }
                        }
-                       spin_lock_irq(&base->t_base.lock);
+                       spin_lock_irq(&base->lock);
                }
        }
        set_running_timer(base, NULL);
-       spin_unlock_irq(&base->t_base.lock);
+       spin_unlock_irq(&base->lock);
 }
 
 #ifdef CONFIG_NO_IDLE_HZ
@@ -519,13 +472,25 @@ unsigned long next_timer_interrupt(void)
        struct list_head *list;
        struct timer_list *nte;
        unsigned long expires;
+       unsigned long hr_expires = MAX_JIFFY_OFFSET;
+       ktime_t hr_delta;
        tvec_t *varray[4];
        int i, j;
 
-       base = &__get_cpu_var(tvec_bases);
-       spin_lock(&base->t_base.lock);
+       hr_delta = hrtimer_get_next_event();
+       if (hr_delta.tv64 != KTIME_MAX) {
+               struct timespec tsdelta;
+               tsdelta = ktime_to_timespec(hr_delta);
+               hr_expires = timespec_to_jiffies(&tsdelta);
+               if (hr_expires < 3)
+                       return hr_expires + jiffies;
+       }
+       hr_expires += jiffies;
+
+       base = __get_cpu_var(tvec_bases);
+       spin_lock(&base->lock);
        expires = base->timer_jiffies + (LONG_MAX >> 1);
-       list = 0;
+       list = NULL;
 
        /* Look for timer events in tv1. */
        j = base->timer_jiffies & TVR_MASK;
@@ -571,7 +536,27 @@ found:
                                expires = nte->expires;
                }
        }
-       spin_unlock(&base->t_base.lock);
+       spin_unlock(&base->lock);
+
+       /*
+        * It can happen that other CPUs service timer IRQs and increment
+        * jiffies, but we have not yet got a local timer tick to process
+        * the timer wheels.  In that case, the expiry time can be before
+        * jiffies, but since the high-resolution timer here is relative to
+        * jiffies, the default expression when high-resolution timers are
+        * not active,
+        *
+        *   time_before(MAX_JIFFY_OFFSET + jiffies, expires)
+        *
+        * would falsely evaluate to true.  If that is the case, just
+        * return jiffies so that we can immediately fire the local timer
+        */
+       if (time_before(expires, jiffies))
+               return jiffies;
+
+       if (time_before(hr_expires, expires))
+               return hr_expires;
+
        return expires;
 }
 #endif
@@ -613,7 +598,6 @@ long time_tolerance = MAXFREQ;              /* frequency tolerance (ppm)    */
 long time_precision = 1;               /* clock precision (us)         */
 long time_maxerror = NTP_PHASE_LIMIT;  /* maximum error (us)           */
 long time_esterror = NTP_PHASE_LIMIT;  /* estimated error (us)         */
-static long time_phase;                        /* phase offset (scaled us)     */
 long time_freq = (((NSEC_PER_SEC + HZ/2) % HZ - HZ/2) << SHIFT_USEC) / NSEC_PER_USEC;
                                        /* frequency offset (scaled ppm)*/
 static long time_adj;                  /* tick adjust (scaled 1 / HZ)  */
@@ -632,178 +616,145 @@ long time_next_adjust;
  */
 static void second_overflow(void)
 {
-    long ltemp;
-
-    /* Bump the maxerror field */
-    time_maxerror += time_tolerance >> SHIFT_USEC;
-    if ( time_maxerror > NTP_PHASE_LIMIT ) {
-       time_maxerror = NTP_PHASE_LIMIT;
-       time_status |= STA_UNSYNC;
-    }
-
-    /*
-     * Leap second processing. If in leap-insert state at
-     * the end of the day, the system clock is set back one
-     * second; if in leap-delete state, the system clock is
-     * set ahead one second. The microtime() routine or
-     * external clock driver will insure that reported time
-     * is always monotonic. The ugly divides should be
-     * replaced.
-     */
-    switch (time_state) {
-
-    case TIME_OK:
-       if (time_status & STA_INS)
-           time_state = TIME_INS;
-       else if (time_status & STA_DEL)
-           time_state = TIME_DEL;
-       break;
-
-    case TIME_INS:
-       if (xtime.tv_sec % 86400 == 0) {
-           xtime.tv_sec--;
-           wall_to_monotonic.tv_sec++;
-           /* The timer interpolator will make time change gradually instead
-            * of an immediate jump by one second.
-            */
-           time_interpolator_update(-NSEC_PER_SEC);
-           time_state = TIME_OOP;
-           clock_was_set();
-           printk(KERN_NOTICE "Clock: inserting leap second 23:59:60 UTC\n");
+       long ltemp;
+
+       /* Bump the maxerror field */
+       time_maxerror += time_tolerance >> SHIFT_USEC;
+       if (time_maxerror > NTP_PHASE_LIMIT) {
+               time_maxerror = NTP_PHASE_LIMIT;
+               time_status |= STA_UNSYNC;
        }
-       break;
-
-    case TIME_DEL:
-       if ((xtime.tv_sec + 1) % 86400 == 0) {
-           xtime.tv_sec++;
-           wall_to_monotonic.tv_sec--;
-           /* Use of time interpolator for a gradual change of time */
-           time_interpolator_update(NSEC_PER_SEC);
-           time_state = TIME_WAIT;
-           clock_was_set();
-           printk(KERN_NOTICE "Clock: deleting leap second 23:59:59 UTC\n");
+
+       /*
+        * Leap second processing. If in leap-insert state at the end of the
+        * day, the system clock is set back one second; if in leap-delete
+        * state, the system clock is set ahead one second. The microtime()
+        * routine or external clock driver will insure that reported time is
+        * always monotonic. The ugly divides should be replaced.
+        */
+       switch (time_state) {
+       case TIME_OK:
+               if (time_status & STA_INS)
+                       time_state = TIME_INS;
+               else if (time_status & STA_DEL)
+                       time_state = TIME_DEL;
+               break;
+       case TIME_INS:
+               if (xtime.tv_sec % 86400 == 0) {
+                       xtime.tv_sec--;
+                       wall_to_monotonic.tv_sec++;
+                       /*
+                        * The timer interpolator will make time change
+                        * gradually instead of an immediate jump by one second
+                        */
+                       time_interpolator_update(-NSEC_PER_SEC);
+                       time_state = TIME_OOP;
+                       clock_was_set();
+                       printk(KERN_NOTICE "Clock: inserting leap second "
+                                       "23:59:60 UTC\n");
+               }
+               break;
+       case TIME_DEL:
+               if ((xtime.tv_sec + 1) % 86400 == 0) {
+                       xtime.tv_sec++;
+                       wall_to_monotonic.tv_sec--;
+                       /*
+                        * Use of time interpolator for a gradual change of
+                        * time
+                        */
+                       time_interpolator_update(NSEC_PER_SEC);
+                       time_state = TIME_WAIT;
+                       clock_was_set();
+                       printk(KERN_NOTICE "Clock: deleting leap second "
+                                       "23:59:59 UTC\n");
+               }
+               break;
+       case TIME_OOP:
+               time_state = TIME_WAIT;
+               break;
+       case TIME_WAIT:
+               if (!(time_status & (STA_INS | STA_DEL)))
+               time_state = TIME_OK;
        }
-       break;
-
-    case TIME_OOP:
-       time_state = TIME_WAIT;
-       break;
-
-    case TIME_WAIT:
-       if (!(time_status & (STA_INS | STA_DEL)))
-           time_state = TIME_OK;
-    }
-
-    /*
-     * Compute the phase adjustment for the next second. In
-     * PLL mode, the offset is reduced by a fixed factor
-     * times the time constant. In FLL mode the offset is
-     * used directly. In either mode, the maximum phase
-     * adjustment for each second is clamped so as to spread
-     * the adjustment over not more than the number of
-     * seconds between updates.
-     */
-    if (time_offset < 0) {
-       ltemp = -time_offset;
-       if (!(time_status & STA_FLL))
-           ltemp >>= SHIFT_KG + time_constant;
-       if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
-           ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
-       time_offset += ltemp;
-       time_adj = -ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
-    } else {
+
+       /*
+        * Compute the phase adjustment for the next second. In PLL mode, the
+        * offset is reduced by a fixed factor times the time constant. In FLL
+        * mode the offset is used directly. In either mode, the maximum phase
+        * adjustment for each second is clamped so as to spread the adjustment
+        * over not more than the number of seconds between updates.
+        */
        ltemp = time_offset;
        if (!(time_status & STA_FLL))
-           ltemp >>= SHIFT_KG + time_constant;
-       if (ltemp > (MAXPHASE / MINSEC) << SHIFT_UPDATE)
-           ltemp = (MAXPHASE / MINSEC) << SHIFT_UPDATE;
+               ltemp = shift_right(ltemp, SHIFT_KG + time_constant);
+       ltemp = min(ltemp, (MAXPHASE / MINSEC) << SHIFT_UPDATE);
+       ltemp = max(ltemp, -(MAXPHASE / MINSEC) << SHIFT_UPDATE);
        time_offset -= ltemp;
        time_adj = ltemp << (SHIFT_SCALE - SHIFT_HZ - SHIFT_UPDATE);
-    }
-
-    /*
-     * Compute the frequency estimate and additional phase
-     * adjustment due to frequency error for the next
-     * second. When the PPS signal is engaged, gnaw on the
-     * watchdog counter and update the frequency computed by
-     * the pll and the PPS signal.
-     */
-    pps_valid++;
-    if (pps_valid == PPS_VALID) {      /* PPS signal lost */
-       pps_jitter = MAXTIME;
-       pps_stabil = MAXFREQ;
-       time_status &= ~(STA_PPSSIGNAL | STA_PPSJITTER |
-                        STA_PPSWANDER | STA_PPSERROR);
-    }
-    ltemp = time_freq + pps_freq;
-    if (ltemp < 0)
-       time_adj -= -ltemp >>
-           (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
-    else
-       time_adj += ltemp >>
-           (SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE);
+
+       /*
+        * Compute the frequency estimate and additional phase adjustment due
+        * to frequency error for the next second.
+        */
+       ltemp = time_freq;
+       time_adj += shift_right(ltemp,(SHIFT_USEC + SHIFT_HZ - SHIFT_SCALE));
 
 #if HZ == 100
-    /* Compensate for (HZ==100) != (1 << SHIFT_HZ).
-     * Add 25% and 3.125% to get 128.125; => only 0.125% error (p. 14)
-     */
-    if (time_adj < 0)
-       time_adj -= (-time_adj >> 2) + (-time_adj >> 5);
-    else
-       time_adj += (time_adj >> 2) + (time_adj >> 5);
+       /*
+        * Compensate for (HZ==100) != (1 << SHIFT_HZ).  Add 25% and 3.125% to
+        * get 128.125; => only 0.125% error (p. 14)
+        */
+       time_adj += shift_right(time_adj, 2) + shift_right(time_adj, 5);
+#endif
+#if HZ == 250
+       /*
+        * Compensate for (HZ==250) != (1 << SHIFT_HZ).  Add 1.5625% and
+        * 0.78125% to get 255.85938; => only 0.05% error (p. 14)
+        */
+       time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
 #endif
 #if HZ == 1000
-    /* Compensate for (HZ==1000) != (1 << SHIFT_HZ).
-     * Add 1.5625% and 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
-     */
-    if (time_adj < 0)
-       time_adj -= (-time_adj >> 6) + (-time_adj >> 7);
-    else
-       time_adj += (time_adj >> 6) + (time_adj >> 7);
+       /*
+        * Compensate for (HZ==1000) != (1 << SHIFT_HZ).  Add 1.5625% and
+        * 0.78125% to get 1023.4375; => only 0.05% error (p. 14)
+        */
+       time_adj += shift_right(time_adj, 6) + shift_right(time_adj, 7);
 #endif
 }
 
-/* in the NTP reference this is called "hardclock()" */
-static void update_wall_time_one_tick(void)
-{
-       long time_adjust_step, delta_nsec;
-
-       if ( (time_adjust_step = time_adjust) != 0 ) {
-           /* We are doing an adjtime thing. 
-            *
-            * Prepare time_adjust_step to be within bounds.
-            * Note that a positive time_adjust means we want the clock
-            * to run faster.
-            *
-            * Limit the amount of the step to be in the range
-            * -tickadj .. +tickadj
-            */
-            if (time_adjust > tickadj)
-               time_adjust_step = tickadj;
-            else if (time_adjust < -tickadj)
-               time_adjust_step = -tickadj;
-
-           /* Reduce by this step the amount of time left  */
-           time_adjust -= time_adjust_step;
-       }
-       delta_nsec = tick_nsec + time_adjust_step * 1000;
-       /*
-        * Advance the phase, once it gets to one microsecond, then
-        * advance the tick more.
-        */
-       time_phase += time_adj;
-       if (time_phase <= -FINENSEC) {
-               long ltemp = -time_phase >> (SHIFT_SCALE - 10);
-               time_phase += ltemp << (SHIFT_SCALE - 10);
-               delta_nsec -= ltemp;
-       }
-       else if (time_phase >= FINENSEC) {
-               long ltemp = time_phase >> (SHIFT_SCALE - 10);
-               time_phase -= ltemp << (SHIFT_SCALE - 10);
-               delta_nsec += ltemp;
+/*
+ * Returns how many microseconds we need to add to xtime this tick
+ * in doing an adjustment requested with adjtime.
+ */
+static long adjtime_adjustment(void)
+{
+       long time_adjust_step;
+
+       time_adjust_step = time_adjust;
+       if (time_adjust_step) {
+               /*
+                * We are doing an adjtime thing.  Prepare time_adjust_step to
+                * be within bounds.  Note that a positive time_adjust means we
+                * want the clock to run faster.
+                *
+                * Limit the amount of the step to be in the range
+                * -tickadj .. +tickadj
+                */
+               time_adjust_step = min(time_adjust_step, (long)tickadj);
+               time_adjust_step = max(time_adjust_step, (long)-tickadj);
        }
-       xtime.tv_nsec += delta_nsec;
-       time_interpolator_update(delta_nsec);
+       return time_adjust_step;
+}
+
+/* in the NTP reference this is called "hardclock()" */
+static void update_ntp_one_tick(void)
+{
+       long time_adjust_step;
+
+       time_adjust_step = adjtime_adjustment();
+       if (time_adjust_step)
+               /* Reduce by this step the amount of time left  */
+               time_adjust -= time_adjust_step;
 
        /* Changes by adjtime() do not take effect till next tick. */
        if (time_next_adjust != 0) {
@@ -813,23 +764,407 @@ static void update_wall_time_one_tick(void)
 }
 
 /*
- * Using a loop looks inefficient, but "ticks" is
- * usually just one (we shouldn't be losing ticks,
- * we're doing this this way mainly for interrupt
- * latency reasons, not because we think we'll
- * have lots of lost timer ticks
+ * Return how long ticks are at the moment, that is, how much time
+ * update_wall_time_one_tick will add to xtime next time we call it
+ * (assuming no calls to do_adjtimex in the meantime).
+ * The return value is in fixed-point nanoseconds shifted by the
+ * specified number of bits to the right of the binary point.
+ * This function has no side-effects.
  */
-static void update_wall_time(unsigned long ticks)
+u64 current_tick_length(void)
 {
+       long delta_nsec;
+       u64 ret;
+
+       /* calculate the finest interval NTP will allow.
+        *    ie: nanosecond value shifted by (SHIFT_SCALE - 10)
+        */
+       delta_nsec = tick_nsec + adjtime_adjustment() * 1000;
+       ret = (u64)delta_nsec << TICK_LENGTH_SHIFT;
+       ret += (s64)time_adj << (TICK_LENGTH_SHIFT - (SHIFT_SCALE - 10));
+
+       return ret;
+}
+
+/* XXX - all of this timekeeping code should be later moved to time.c */
+#include <linux/clocksource.h>
+static struct clocksource *clock; /* pointer to current clocksource */
+
+#ifdef CONFIG_GENERIC_TIME
+/**
+ * __get_nsec_offset - Returns nanoseconds since last call to periodic_hook
+ *
+ * private function, must hold xtime_lock lock when being
+ * called. Returns the number of nanoseconds since the
+ * last call to update_wall_time() (adjusted by NTP scaling)
+ */
+static inline s64 __get_nsec_offset(void)
+{
+       cycle_t cycle_now, cycle_delta;
+       s64 ns_offset;
+
+       /* read clocksource: */
+       cycle_now = clocksource_read(clock);
+
+       /* calculate the delta since the last update_wall_time: */
+       cycle_delta = (cycle_now - clock->cycle_last) & clock->mask;
+
+       /* convert to nanoseconds: */
+       ns_offset = cyc2ns(clock, cycle_delta);
+
+       return ns_offset;
+}
+
+/**
+ * __get_realtime_clock_ts - Returns the time of day in a timespec
+ * @ts:                pointer to the timespec to be set
+ *
+ * Returns the time of day in a timespec. Used by
+ * do_gettimeofday() and get_realtime_clock_ts().
+ */
+static inline void __get_realtime_clock_ts(struct timespec *ts)
+{
+       unsigned long seq;
+       s64 nsecs;
+
        do {
-               ticks--;
-               update_wall_time_one_tick();
-               if (xtime.tv_nsec >= 1000000000) {
-                       xtime.tv_nsec -= 1000000000;
+               seq = read_seqbegin(&xtime_lock);
+
+               *ts = xtime;
+               nsecs = __get_nsec_offset();
+
+       } while (read_seqretry(&xtime_lock, seq));
+
+       timespec_add_ns(ts, nsecs);
+}
+
+/**
+ * getnstimeofday - Returns the time of day in a timespec
+ * @ts:                pointer to the timespec to be set
+ *
+ * Returns the time of day in a timespec.
+ */
+void getnstimeofday(struct timespec *ts)
+{
+       __get_realtime_clock_ts(ts);
+}
+
+EXPORT_SYMBOL(getnstimeofday);
+
+/**
+ * do_gettimeofday - Returns the time of day in a timeval
+ * @tv:                pointer to the timeval to be set
+ *
+ * NOTE: Users should be converted to using get_realtime_clock_ts()
+ */
+void do_gettimeofday(struct timeval *tv)
+{
+       struct timespec now;
+
+       __get_realtime_clock_ts(&now);
+       tv->tv_sec = now.tv_sec;
+       tv->tv_usec = now.tv_nsec/1000;
+}
+
+EXPORT_SYMBOL(do_gettimeofday);
+/**
+ * do_settimeofday - Sets the time of day
+ * @tv:                pointer to the timespec variable containing the new time
+ *
+ * Sets the time of day to the new time and update NTP and notify hrtimers
+ */
+int do_settimeofday(struct timespec *tv)
+{
+       unsigned long flags;
+       time_t wtm_sec, sec = tv->tv_sec;
+       long wtm_nsec, nsec = tv->tv_nsec;
+
+       if ((unsigned long)tv->tv_nsec >= NSEC_PER_SEC)
+               return -EINVAL;
+
+       write_seqlock_irqsave(&xtime_lock, flags);
+
+       nsec -= __get_nsec_offset();
+
+       wtm_sec  = wall_to_monotonic.tv_sec + (xtime.tv_sec - sec);
+       wtm_nsec = wall_to_monotonic.tv_nsec + (xtime.tv_nsec - nsec);
+
+       set_normalized_timespec(&xtime, sec, nsec);
+       set_normalized_timespec(&wall_to_monotonic, wtm_sec, wtm_nsec);
+
+       clock->error = 0;
+       ntp_clear();
+
+       write_sequnlock_irqrestore(&xtime_lock, flags);
+
+       /* signal hrtimers about time change */
+       clock_was_set();
+
+       return 0;
+}
+
+EXPORT_SYMBOL(do_settimeofday);
+
+/**
+ * change_clocksource - Swaps clocksources if a new one is available
+ *
+ * Accumulates current time interval and initializes new clocksource
+ */
+static int change_clocksource(void)
+{
+       struct clocksource *new;
+       cycle_t now;
+       u64 nsec;
+       new = clocksource_get_next();
+       if (clock != new) {
+               now = clocksource_read(new);
+               nsec =  __get_nsec_offset();
+               timespec_add_ns(&xtime, nsec);
+
+               clock = new;
+               clock->cycle_last = now;
+               printk(KERN_INFO "Time: %s clocksource has been installed.\n",
+                                       clock->name);
+               return 1;
+       } else if (clock->update_callback) {
+               return clock->update_callback();
+       }
+       return 0;
+}
+#else
+#define change_clocksource() (0)
+#endif
+
+/**
+ * timeofday_is_continuous - check to see if timekeeping is free running
+ */
+int timekeeping_is_continuous(void)
+{
+       unsigned long seq;
+       int ret;
+
+       do {
+               seq = read_seqbegin(&xtime_lock);
+
+               ret = clock->is_continuous;
+
+       } while (read_seqretry(&xtime_lock, seq));
+
+       return ret;
+}
+
+/*
+ * timekeeping_init - Initializes the clocksource and common timekeeping values
+ */
+void __init timekeeping_init(void)
+{
+       unsigned long flags;
+
+       write_seqlock_irqsave(&xtime_lock, flags);
+       clock = clocksource_get_next();
+       clocksource_calculate_interval(clock, tick_nsec);
+       clock->cycle_last = clocksource_read(clock);
+       ntp_clear();
+       write_sequnlock_irqrestore(&xtime_lock, flags);
+}
+
+
+static int timekeeping_suspended;
+/*
+ * timekeeping_resume - Resumes the generic timekeeping subsystem.
+ * @dev:       unused
+ *
+ * This is for the generic clocksource timekeeping.
+ * xtime/wall_to_monotonic/jiffies/wall_jiffies/etc are
+ * still managed by arch specific suspend/resume code.
+ */
+static int timekeeping_resume(struct sys_device *dev)
+{
+       unsigned long flags;
+
+       write_seqlock_irqsave(&xtime_lock, flags);
+       /* restart the last cycle value */
+       clock->cycle_last = clocksource_read(clock);
+       clock->error = 0;
+       timekeeping_suspended = 0;
+       write_sequnlock_irqrestore(&xtime_lock, flags);
+       return 0;
+}
+
+static int timekeeping_suspend(struct sys_device *dev, pm_message_t state)
+{
+       unsigned long flags;
+
+       write_seqlock_irqsave(&xtime_lock, flags);
+       timekeeping_suspended = 1;
+       write_sequnlock_irqrestore(&xtime_lock, flags);
+       return 0;
+}
+
+/* sysfs resume/suspend bits for timekeeping */
+static struct sysdev_class timekeeping_sysclass = {
+       .resume         = timekeeping_resume,
+       .suspend        = timekeeping_suspend,
+       set_kset_name("timekeeping"),
+};
+
+static struct sys_device device_timer = {
+       .id             = 0,
+       .cls            = &timekeeping_sysclass,
+};
+
+static int __init timekeeping_init_device(void)
+{
+       int error = sysdev_class_register(&timekeeping_sysclass);
+       if (!error)
+               error = sysdev_register(&device_timer);
+       return error;
+}
+
+device_initcall(timekeeping_init_device);
+
+/*
+ * If the error is already larger, we look ahead even further
+ * to compensate for late or lost adjustments.
+ */
+static __always_inline int clocksource_bigadjust(s64 error, s64 *interval, s64 *offset)
+{
+       s64 tick_error, i;
+       u32 look_ahead, adj;
+       s32 error2, mult;
+
+       /*
+        * Use the current error value to determine how much to look ahead.
+        * The larger the error the slower we adjust for it to avoid problems
+        * with losing too many ticks, otherwise we would overadjust and
+        * produce an even larger error.  The smaller the adjustment the
+        * faster we try to adjust for it, as lost ticks can do less harm
+        * here.  This is tuned so that an error of about 1 msec is adusted
+        * within about 1 sec (or 2^20 nsec in 2^SHIFT_HZ ticks).
+        */
+       error2 = clock->error >> (TICK_LENGTH_SHIFT + 22 - 2 * SHIFT_HZ);
+       error2 = abs(error2);
+       for (look_ahead = 0; error2 > 0; look_ahead++)
+               error2 >>= 2;
+
+       /*
+        * Now calculate the error in (1 << look_ahead) ticks, but first
+        * remove the single look ahead already included in the error.
+        */
+       tick_error = current_tick_length() >> (TICK_LENGTH_SHIFT - clock->shift + 1);
+       tick_error -= clock->xtime_interval >> 1;
+       error = ((error - tick_error) >> look_ahead) + tick_error;
+
+       /* Finally calculate the adjustment shift value.  */
+       i = *interval;
+       mult = 1;
+       if (error < 0) {
+               error = -error;
+               *interval = -*interval;
+               *offset = -*offset;
+               mult = -1;
+       }
+       for (adj = 0; error > i; adj++)
+               error >>= 1;
+
+       *interval <<= adj;
+       *offset <<= adj;
+       return mult << adj;
+}
+
+/*
+ * Adjust the multiplier to reduce the error value,
+ * this is optimized for the most common adjustments of -1,0,1,
+ * for other values we can do a bit more work.
+ */
+static void clocksource_adjust(struct clocksource *clock, s64 offset)
+{
+       s64 error, interval = clock->cycle_interval;
+       int adj;
+
+       error = clock->error >> (TICK_LENGTH_SHIFT - clock->shift - 1);
+       if (error > interval) {
+               error >>= 2;
+               if (likely(error <= interval))
+                       adj = 1;
+               else
+                       adj = clocksource_bigadjust(error, &interval, &offset);
+       } else if (error < -interval) {
+               error >>= 2;
+               if (likely(error >= -interval)) {
+                       adj = -1;
+                       interval = -interval;
+                       offset = -offset;
+               } else
+                       adj = clocksource_bigadjust(error, &interval, &offset);
+       } else
+               return;
+
+       clock->mult += adj;
+       clock->xtime_interval += interval;
+       clock->xtime_nsec -= offset;
+       clock->error -= (interval - offset) << (TICK_LENGTH_SHIFT - clock->shift);
+}
+
+/*
+ * update_wall_time - Uses the current clocksource to increment the wall time
+ *
+ * Called from the timer interrupt, must hold a write on xtime_lock.
+ */
+static void update_wall_time(void)
+{
+       cycle_t offset;
+
+       /* Make sure we're fully resumed: */
+       if (unlikely(timekeeping_suspended))
+               return;
+
+#ifdef CONFIG_GENERIC_TIME
+       offset = (clocksource_read(clock) - clock->cycle_last) & clock->mask;
+#else
+       offset = clock->cycle_interval;
+#endif
+       clock->xtime_nsec += (s64)xtime.tv_nsec << clock->shift;
+
+       /* normally this loop will run just once, however in the
+        * case of lost or late ticks, it will accumulate correctly.
+        */
+       while (offset >= clock->cycle_interval) {
+               /* accumulate one interval */
+               clock->xtime_nsec += clock->xtime_interval;
+               clock->cycle_last += clock->cycle_interval;
+               offset -= clock->cycle_interval;
+
+               if (clock->xtime_nsec >= (u64)NSEC_PER_SEC << clock->shift) {
+                       clock->xtime_nsec -= (u64)NSEC_PER_SEC << clock->shift;
                        xtime.tv_sec++;
                        second_overflow();
                }
-       } while (ticks);
+
+               /* interpolator bits */
+               time_interpolator_update(clock->xtime_interval
+                                               >> clock->shift);
+               /* increment the NTP state machine */
+               update_ntp_one_tick();
+
+               /* accumulate error between NTP and clock interval */
+               clock->error += current_tick_length();
+               clock->error -= clock->xtime_interval << (TICK_LENGTH_SHIFT - clock->shift);
+       }
+
+       /* correct the clock when NTP error is too big */
+       clocksource_adjust(clock, offset);
+
+       /* store full nanoseconds into xtime */
+       xtime.tv_nsec = (s64)clock->xtime_nsec >> clock->shift;
+       clock->xtime_nsec -= (s64)xtime.tv_nsec << clock->shift;
+
+       /* check to see if there is a new clocksource to use */
+       if (change_clocksource()) {
+               clock->error = 0;
+               clock->xtime_nsec = 0;
+               clocksource_calculate_interval(clock, tick_nsec);
+       }
 }
 
 /*
@@ -858,7 +1193,7 @@ void update_process_times(int user_tick)
  */
 static unsigned long count_active_tasks(void)
 {
-       return (nr_running() + nr_uninterruptible()) * FIXED_1;
+       return nr_active() * FIXED_1;
 }
 
 /*
@@ -900,7 +1235,7 @@ unsigned long wall_jiffies = INITIAL_JIFFIES;
  * playing with xtime and avenrun.
  */
 #ifndef ARCH_HAVE_XTIME_LOCK
-seqlock_t xtime_lock __cacheline_aligned_in_smp = SEQLOCK_UNLOCKED;
+__cacheline_aligned_in_smp DEFINE_SEQLOCK(xtime_lock);
 
 EXPORT_SYMBOL(xtime_lock);
 #endif
@@ -910,8 +1245,9 @@ EXPORT_SYMBOL(xtime_lock);
  */
 static void run_timer_softirq(struct softirq_action *h)
 {
-       tvec_base_t *base = &__get_cpu_var(tvec_bases);
+       tvec_base_t *base = __get_cpu_var(tvec_bases);
 
+       hrtimer_run_queues();
        if (time_after_eq(jiffies, base->timer_jiffies))
                __run_timers(base);
 }
@@ -922,6 +1258,7 @@ static void run_timer_softirq(struct softirq_action *h)
 void run_local_timers(void)
 {
        raise_softirq(TIMER_SOFTIRQ);
+       softlockup_tick();
 }
 
 /*
@@ -933,10 +1270,8 @@ static inline void update_times(void)
        unsigned long ticks;
 
        ticks = jiffies - wall_jiffies;
-       if (ticks) {
-               wall_jiffies += ticks;
-               update_wall_time(ticks);
-       }
+       wall_jiffies += ticks;
+       update_wall_time();
        calc_load(ticks);
 }
   
@@ -949,8 +1284,9 @@ static inline void update_times(void)
 void do_timer(struct pt_regs *regs)
 {
        jiffies_64++;
+       /* prevent loading jiffies before storing new jiffies_64 value. */
+       barrier();
        update_times();
-       softlockup_tick(regs);
 }
 
 #ifdef __ARCH_WANT_SYS_ALARM
@@ -961,19 +1297,7 @@ void do_timer(struct pt_regs *regs)
  */
 asmlinkage unsigned long sys_alarm(unsigned int seconds)
 {
-       struct itimerval it_new, it_old;
-       unsigned int oldalarm;
-
-       it_new.it_interval.tv_sec = it_new.it_interval.tv_usec = 0;
-       it_new.it_value.tv_sec = seconds;
-       it_new.it_value.tv_usec = 0;
-       do_setitimer(ITIMER_REAL, &it_new, &it_old);
-       oldalarm = it_old.it_value.tv_sec;
-       /* ehhh.. We can't return 0 if we have an alarm pending.. */
-       /* And we'd better return too much than too little anyway */
-       if ((!oldalarm && it_old.it_value.tv_usec) || it_old.it_value.tv_usec >= 500000)
-               oldalarm++;
-       return oldalarm;
+       return alarm_setitimer(seconds);
 }
 
 #endif
@@ -1071,7 +1395,7 @@ asmlinkage long sys_getegid(void)
 
 static void process_timeout(unsigned long __data)
 {
-       wake_up_process((task_t *)__data);
+       wake_up_process((struct task_struct *)__data);
 }
 
 /**
@@ -1128,8 +1452,8 @@ fastcall signed long __sched schedule_timeout(signed long timeout)
                if (timeout < 0)
                {
                        printk(KERN_ERR "schedule_timeout: wrong timeout "
-                              "value %lx from %p\n", timeout,
-                              __builtin_return_address(0));
+                               "value %lx from %p\n", timeout,
+                               __builtin_return_address(0));
                        current->state = TASK_RUNNING;
                        goto out;
                }
@@ -1137,12 +1461,8 @@ fastcall signed long __sched schedule_timeout(signed long timeout)
 
        expire = timeout + jiffies;
 
-       init_timer(&timer);
-       timer.expires = expire;
-       timer.data = (unsigned long) current;
-       timer.function = process_timeout;
-
-       add_timer(&timer);
+       setup_timer(&timer, process_timeout, (unsigned long)current);
+       __mod_timer(&timer, expire);
        schedule();
        del_singleshot_timer_sync(&timer);
 
@@ -1151,71 +1471,30 @@ fastcall signed long __sched schedule_timeout(signed long timeout)
  out:
        return timeout < 0 ? 0 : timeout;
 }
-
 EXPORT_SYMBOL(schedule_timeout);
 
-/* Thread ID - the internal kernel "pid" */
-asmlinkage long sys_gettid(void)
+/*
+ * We can use __set_current_state() here because schedule_timeout() calls
+ * schedule() unconditionally.
+ */
+signed long __sched schedule_timeout_interruptible(signed long timeout)
 {
-       return current->pid;
+       __set_current_state(TASK_INTERRUPTIBLE);
+       return schedule_timeout(timeout);
 }
+EXPORT_SYMBOL(schedule_timeout_interruptible);
 
-static long __sched nanosleep_restart(struct restart_block *restart)
+signed long __sched schedule_timeout_uninterruptible(signed long timeout)
 {
-       unsigned long expire = restart->arg0, now = jiffies;
-       struct timespec __user *rmtp = (struct timespec __user *) restart->arg1;
-       long ret;
-
-       /* Did it expire while we handled signals? */
-       if (!time_after(expire, now))
-               return 0;
-
-       current->state = TASK_INTERRUPTIBLE;
-       expire = schedule_timeout(expire - now);
-
-       ret = 0;
-       if (expire) {
-               struct timespec t;
-               jiffies_to_timespec(expire, &t);
-
-               ret = -ERESTART_RESTARTBLOCK;
-               if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
-                       ret = -EFAULT;
-               /* The 'restart' block is already filled in */
-       }
-       return ret;
+       __set_current_state(TASK_UNINTERRUPTIBLE);
+       return schedule_timeout(timeout);
 }
+EXPORT_SYMBOL(schedule_timeout_uninterruptible);
 
-asmlinkage long sys_nanosleep(struct timespec __user *rqtp, struct timespec __user *rmtp)
+/* Thread ID - the internal kernel "pid" */
+asmlinkage long sys_gettid(void)
 {
-       struct timespec t;
-       unsigned long expire;
-       long ret;
-
-       if (copy_from_user(&t, rqtp, sizeof(t)))
-               return -EFAULT;
-
-       if ((t.tv_nsec >= 1000000000L) || (t.tv_nsec < 0) || (t.tv_sec < 0))
-               return -EINVAL;
-
-       expire = timespec_to_jiffies(&t) + (t.tv_sec || t.tv_nsec);
-       current->state = TASK_INTERRUPTIBLE;
-       expire = schedule_timeout(expire);
-
-       ret = 0;
-       if (expire) {
-               struct restart_block *restart;
-               jiffies_to_timespec(expire, &t);
-               if (rmtp && copy_to_user(rmtp, &t, sizeof(t)))
-                       return -EFAULT;
-
-               restart = &current_thread_info()->restart_block;
-               restart->fn = nanosleep_restart;
-               restart->arg0 = jiffies + expire;
-               restart->arg1 = (unsigned long) rmtp;
-               ret = -ERESTART_RESTARTBLOCK;
-       }
-       return ret;
+       return current->pid;
 }
 
 /*
@@ -1307,13 +1586,50 @@ asmlinkage long sys_sysinfo(struct sysinfo __user *info)
        return 0;
 }
 
-static void __devinit init_timers_cpu(int cpu)
+/*
+ * lockdep: we want to track each per-CPU base as a separate lock-class,
+ * but timer-bases are kmalloc()-ed, so we need to attach separate
+ * keys to them:
+ */
+static struct lock_class_key base_lock_keys[NR_CPUS];
+
+static int __devinit init_timers_cpu(int cpu)
 {
        int j;
        tvec_base_t *base;
+       static char __devinitdata tvec_base_done[NR_CPUS];
+
+       if (!tvec_base_done[cpu]) {
+               static char boot_done;
+
+               if (boot_done) {
+                       /*
+                        * The APs use this path later in boot
+                        */
+                       base = kmalloc_node(sizeof(*base), GFP_KERNEL,
+                                               cpu_to_node(cpu));
+                       if (!base)
+                               return -ENOMEM;
+                       memset(base, 0, sizeof(*base));
+                       per_cpu(tvec_bases, cpu) = base;
+               } else {
+                       /*
+                        * This is for the boot CPU - we use compile-time
+                        * static initialisation because per-cpu memory isn't
+                        * ready yet and because the memory allocators are not
+                        * initialised either.
+                        */
+                       boot_done = 1;
+                       base = &boot_tvec_bases;
+               }
+               tvec_base_done[cpu] = 1;
+       } else {
+               base = per_cpu(tvec_bases, cpu);
+       }
+
+       spin_lock_init(&base->lock);
+       lockdep_set_class(&base->lock, base_lock_keys + cpu);
 
-       base = &per_cpu(tvec_bases, cpu);
-       spin_lock_init(&base->t_base.lock);
        for (j = 0; j < TVN_SIZE; j++) {
                INIT_LIST_HEAD(base->tv5.vec + j);
                INIT_LIST_HEAD(base->tv4.vec + j);
@@ -1324,6 +1640,7 @@ static void __devinit init_timers_cpu(int cpu)
                INIT_LIST_HEAD(base->tv1.vec + j);
 
        base->timer_jiffies = jiffies;
+       return 0;
 }
 
 #ifdef CONFIG_HOTPLUG_CPU
@@ -1334,7 +1651,7 @@ static void migrate_timer_list(tvec_base_t *new_base, struct list_head *head)
        while (!list_empty(head)) {
                timer = list_entry(head->next, struct timer_list, entry);
                detach_timer(timer, 0);
-               timer->base = &new_base->t_base;
+               timer->base = new_base;
                internal_add_timer(new_base, timer);
        }
 }
@@ -1346,15 +1663,15 @@ static void __devinit migrate_timers(int cpu)
        int i;
 
        BUG_ON(cpu_online(cpu));
-       old_base = &per_cpu(tvec_bases, cpu);
-       new_base = &get_cpu_var(tvec_bases);
+       old_base = per_cpu(tvec_bases, cpu);
+       new_base = get_cpu_var(tvec_bases);
 
        local_irq_disable();
-       spin_lock(&new_base->t_base.lock);
-       spin_lock(&old_base->t_base.lock);
+       spin_lock(&new_base->lock);
+       spin_lock(&old_base->lock);
+
+       BUG_ON(old_base->running_timer);
 
-       if (old_base->t_base.running_timer)
-               BUG();
        for (i = 0; i < TVR_SIZE; i++)
                migrate_timer_list(new_base, old_base->tv1.vec + i);
        for (i = 0; i < TVN_SIZE; i++) {
@@ -1364,20 +1681,21 @@ static void __devinit migrate_timers(int cpu)
                migrate_timer_list(new_base, old_base->tv5.vec + i);
        }
 
-       spin_unlock(&old_base->t_base.lock);
-       spin_unlock(&new_base->t_base.lock);
+       spin_unlock(&old_base->lock);
+       spin_unlock(&new_base->lock);
        local_irq_enable();
        put_cpu_var(tvec_bases);
 }
 #endif /* CONFIG_HOTPLUG_CPU */
 
-static int __devinit timer_cpu_notify(struct notifier_block *self, 
+static int __devinit timer_cpu_notify(struct notifier_block *self,
                                unsigned long action, void *hcpu)
 {
        long cpu = (long)hcpu;
        switch(action) {
        case CPU_UP_PREPARE:
-               init_timers_cpu(cpu);
+               if (init_timers_cpu(cpu) < 0)
+                       return NOTIFY_BAD;
                break;
 #ifdef CONFIG_HOTPLUG_CPU
        case CPU_DEAD:
@@ -1405,8 +1723,8 @@ void __init init_timers(void)
 
 #ifdef CONFIG_TIME_INTERPOLATION
 
-struct time_interpolator *time_interpolator;
-static struct time_interpolator *time_interpolator_list;
+struct time_interpolator *time_interpolator __read_mostly;
+static struct time_interpolator *time_interpolator_list __read_mostly;
 static DEFINE_SPINLOCK(time_interpolator_lock);
 
 static inline u64 time_interpolator_get_cycles(unsigned int src)
@@ -1420,16 +1738,16 @@ static inline u64 time_interpolator_get_cycles(unsigned int src)
                        return x();
 
                case TIME_SOURCE_MMIO64 :
-                       return readq((void __iomem *) time_interpolator->addr);
+                       return readq_relaxed((void __iomem *)time_interpolator->addr);
 
                case TIME_SOURCE_MMIO32 :
-                       return readl((void __iomem *) time_interpolator->addr);
+                       return readl_relaxed((void __iomem *)time_interpolator->addr);
 
                default: return get_cycles();
        }
 }
 
-static inline u64 time_interpolator_get_counter(void)
+static inline u64 time_interpolator_get_counter(int writelock)
 {
        unsigned int src = time_interpolator->source;
 
@@ -1443,6 +1761,15 @@ static inline u64 time_interpolator_get_counter(void)
                        now = time_interpolator_get_cycles(src);
                        if (lcycle && time_after(lcycle, now))
                                return lcycle;
+
+                       /* When holding the xtime write lock, there's no need
+                        * to add the overhead of the cmpxchg.  Readers are
+                        * force to retry until the write lock is released.
+                        */
+                       if (writelock) {
+                               time_interpolator->last_cycle = now;
+                               return now;
+                       }
                        /* Keep track of the last timer value returned. The use of cmpxchg here
                         * will cause contention in an SMP environment.
                         */
@@ -1456,7 +1783,7 @@ static inline u64 time_interpolator_get_counter(void)
 void time_interpolator_reset(void)
 {
        time_interpolator->offset = 0;
-       time_interpolator->last_counter = time_interpolator_get_counter();
+       time_interpolator->last_counter = time_interpolator_get_counter(1);
 }
 
 #define GET_TI_NSECS(count,i) (((((count) - i->last_counter) & (i)->mask) * (i)->nsec_per_cyc) >> (i)->shift)
@@ -1468,7 +1795,7 @@ unsigned long time_interpolator_get_offset(void)
                return 0;
 
        return time_interpolator->offset +
-               GET_TI_NSECS(time_interpolator_get_counter(), time_interpolator);
+               GET_TI_NSECS(time_interpolator_get_counter(0), time_interpolator);
 }
 
 #define INTERPOLATOR_ADJUST 65536
@@ -1483,16 +1810,18 @@ static void time_interpolator_update(long delta_nsec)
        if (!time_interpolator)
                return;
 
-       /* The interpolator compensates for late ticks by accumulating
-         * the late time in time_interpolator->offset. A tick earlier than
-        * expected will lead to a reset of the offset and a corresponding
-        * jump of the clock forward. Again this only works if the
-        * interpolator clock is running slightly slower than the regular clock
-        * and the tuning logic insures that.
-         */
+       /*
+        * The interpolator compensates for late ticks by accumulating the late
+        * time in time_interpolator->offset. A tick earlier than expected will
+        * lead to a reset of the offset and a corresponding jump of the clock
+        * forward. Again this only works if the interpolator clock is running
+        * slightly slower than the regular clock and the tuning logic insures
+        * that.
+        */
 
-       counter = time_interpolator_get_counter();
-       offset = time_interpolator->offset + GET_TI_NSECS(counter, time_interpolator);
+       counter = time_interpolator_get_counter(1);
+       offset = time_interpolator->offset +
+                       GET_TI_NSECS(counter, time_interpolator);
 
        if (delta_nsec < 0 || (unsigned long) delta_nsec < offset)
                time_interpolator->offset = offset - delta_nsec;
@@ -1509,7 +1838,7 @@ static void time_interpolator_update(long delta_nsec)
         */
        if (jiffies % INTERPOLATOR_ADJUST == 0)
        {
-               if (time_interpolator->skips == 0 && time_interpolator->offset > TICK_NSEC)
+               if (time_interpolator->skips == 0 && time_interpolator->offset > tick_nsec)
                        time_interpolator->nsec_per_cyc--;
                if (time_interpolator->ns_skipped > INTERPOLATOR_MAX_SKIP && time_interpolator->offset == 0)
                        time_interpolator->nsec_per_cyc++;
@@ -1533,8 +1862,7 @@ register_time_interpolator(struct time_interpolator *ti)
        unsigned long flags;
 
        /* Sanity check */
-       if (ti->frequency == 0 || ti->mask == 0)
-               BUG();
+       BUG_ON(ti->frequency == 0 || ti->mask == 0);
 
        ti->nsec_per_cyc = ((u64)NSEC_PER_SEC << ti->shift) / ti->frequency;
        spin_lock(&time_interpolator_lock);
@@ -1589,10 +1917,8 @@ void msleep(unsigned int msecs)
 {
        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
 
-       while (timeout) {
-               set_current_state(TASK_UNINTERRUPTIBLE);
-               timeout = schedule_timeout(timeout);
-       }
+       while (timeout)
+               timeout = schedule_timeout_uninterruptible(timeout);
 }
 
 EXPORT_SYMBOL(msleep);
@@ -1605,10 +1931,8 @@ unsigned long msleep_interruptible(unsigned int msecs)
 {
        unsigned long timeout = msecs_to_jiffies(msecs) + 1;
 
-       while (timeout && !signal_pending(current)) {
-               set_current_state(TASK_INTERRUPTIBLE);
-               timeout = schedule_timeout(timeout);
-       }
+       while (timeout && !signal_pending(current))
+               timeout = schedule_timeout_interruptible(timeout);
        return jiffies_to_msecs(timeout);
 }