Fedora kernel-2.6.17-1.2142_FC4 patched with stable patch-2.6.17.4-vs2.0.2-rc26.diff
[linux-2.6.git] / kernel / posix-timers.c
index 267506e..16c366c 100644 (file)
@@ -35,6 +35,7 @@
 #include <linux/interrupt.h>
 #include <linux/slab.h>
 #include <linux/time.h>
+#include <linux/mutex.h>
 
 #include <asm/uaccess.h>
 #include <asm/semaphore.h>
 #include <linux/wait.h>
 #include <linux/workqueue.h>
 #include <linux/module.h>
-#include <linux/vs_cvirt.h>
 
-#ifndef div_long_long_rem
-#include <asm/div64.h>
-
-#define div_long_long_rem(dividend,divisor,remainder) ({ \
-                      u64 result = dividend;           \
-                      *remainder = do_div(result,divisor); \
-                      result; })
-
-#endif
-#define CLOCK_REALTIME_RES TICK_NSEC  /* In nano seconds. */
-
-static inline u64  mpy_l_X_l_ll(unsigned long mpy1,unsigned long mpy2)
-{
-       return (u64)mpy1 * mpy2;
-}
 /*
  * Management arrays for POSIX timers.  Timers are kept in slab memory
  * Timer ids are allocated by an external routine that keeps track of the
@@ -89,23 +74,6 @@ static kmem_cache_t *posix_timers_cache;
 static struct idr posix_timers_id;
 static DEFINE_SPINLOCK(idr_lock);
 
-/*
- * Just because the timer is not in the timer list does NOT mean it is
- * inactive.  It could be in the "fire" routine getting a new expire time.
- */
-#define TIMER_INACTIVE 1
-
-#ifdef CONFIG_SMP
-# define timer_active(tmr) \
-               ((tmr)->it.real.timer.entry.prev != (void *)TIMER_INACTIVE)
-# define set_timer_inactive(tmr) \
-               do { \
-                       (tmr)->it.real.timer.entry.prev = (void *)TIMER_INACTIVE; \
-               } while (0)
-#else
-# define timer_active(tmr) BARFY       // error to use outside of SMP
-# define set_timer_inactive(tmr) do { } while (0)
-#endif
 /*
  * we assume that the new SIGEV_THREAD_ID shares no bits with the other
  * SIGEV values.  Here we put out an error if this assumption fails.
@@ -166,18 +134,18 @@ static DEFINE_SPINLOCK(idr_lock);
  */
 
 static struct k_clock posix_clocks[MAX_CLOCKS];
+
 /*
- * We only have one real clock that can be set so we need only one abs list,
- * even if we should want to have several clocks with differing resolutions.
+ * These ones are defined below.
  */
-static struct k_clock_abs abs_list = {.list = LIST_HEAD_INIT(abs_list.list),
-                                     .lock = SPIN_LOCK_UNLOCKED};
+static int common_nsleep(const clockid_t, int flags, struct timespec *t,
+                        struct timespec __user *rmtp);
+static void common_timer_get(struct k_itimer *, struct itimerspec *);
+static int common_timer_set(struct k_itimer *, int,
+                           struct itimerspec *, struct itimerspec *);
+static int common_timer_del(struct k_itimer *timer);
 
-static void posix_timer_fn(unsigned long);
-static u64 do_posix_clock_monotonic_gettime_parts(
-       struct timespec *tp, struct timespec *mo);
-int do_posix_clock_monotonic_gettime(struct timespec *tp);
-static int do_posix_clock_monotonic_get(clockid_t, struct timespec *tp);
+static int posix_timer_fn(struct hrtimer *data);
 
 static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags);
 
@@ -202,7 +170,7 @@ static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
  * the function pointer CALL in struct k_clock.
  */
 
-static inline int common_clock_getres(clockid_t which_clock,
+static inline int common_clock_getres(const clockid_t which_clock,
                                      struct timespec *tp)
 {
        tp->tv_sec = 0;
@@ -210,40 +178,31 @@ static inline int common_clock_getres(clockid_t which_clock,
        return 0;
 }
 
-static inline int common_clock_get(clockid_t which_clock, struct timespec *tp)
+/*
+ * Get real time for posix timers
+ */
+static int common_clock_get(clockid_t which_clock, struct timespec *tp)
 {
-       getnstimeofday(tp);
+       ktime_get_real_ts(tp);
        return 0;
 }
 
-static inline int common_clock_set(clockid_t which_clock, struct timespec *tp)
+static inline int common_clock_set(const clockid_t which_clock,
+                                  struct timespec *tp)
 {
        return do_sys_settimeofday(tp, NULL);
 }
 
-static inline int common_timer_create(struct k_itimer *new_timer)
+static int common_timer_create(struct k_itimer *new_timer)
 {
-       INIT_LIST_HEAD(&new_timer->it.real.abs_timer_entry);
-       init_timer(&new_timer->it.real.timer);
-       new_timer->it.real.timer.data = (unsigned long) new_timer;
-       new_timer->it.real.timer.function = posix_timer_fn;
-       set_timer_inactive(new_timer);
+       hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
        return 0;
 }
 
 /*
- * These ones are defined below.
+ * Return nonzero if we know a priori this clockid_t value is bogus.
  */
-static int common_nsleep(clockid_t, int flags, struct timespec *t);
-static void common_timer_get(struct k_itimer *, struct itimerspec *);
-static int common_timer_set(struct k_itimer *, int,
-                           struct itimerspec *, struct itimerspec *);
-static int common_timer_del(struct k_itimer *timer);
-
-/*
- * Return nonzero iff we know a priori this clockid_t value is bogus.
- */
-static inline int invalid_clockid(clockid_t which_clock)
+static inline int invalid_clockid(const clockid_t which_clock)
 {
        if (which_clock < 0)    /* CPU clock, posix_cpu_* will check it */
                return 0;
@@ -251,26 +210,32 @@ static inline int invalid_clockid(clockid_t which_clock)
                return 1;
        if (posix_clocks[which_clock].clock_getres != NULL)
                return 0;
-#ifndef CLOCK_DISPATCH_DIRECT
        if (posix_clocks[which_clock].res != 0)
                return 0;
-#endif
        return 1;
 }
 
+/*
+ * Get monotonic time for posix timers
+ */
+static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp)
+{
+       ktime_get_ts(tp);
+       return 0;
+}
 
 /*
  * Initialize everything, well, just everything in Posix clocks/timers ;)
  */
 static __init int init_posix_timers(void)
 {
-       struct k_clock clock_realtime = {.res = CLOCK_REALTIME_RES,
-                                        .abs_struct = &abs_list
+       struct k_clock clock_realtime = {
+               .clock_getres = hrtimer_get_res,
        };
-       struct k_clock clock_monotonic = {.res = CLOCK_REALTIME_RES,
-               .abs_struct = NULL,
-               .clock_get = do_posix_clock_monotonic_get,
-               .clock_set = do_posix_clock_nosettime
+       struct k_clock clock_monotonic = {
+               .clock_getres = hrtimer_get_res,
+               .clock_get = posix_ktime_get_ts,
+               .clock_set = do_posix_clock_nosettime,
        };
 
        register_posix_clock(CLOCK_REALTIME, &clock_realtime);
@@ -284,117 +249,20 @@ static __init int init_posix_timers(void)
 
 __initcall(init_posix_timers);
 
-static void tstojiffie(struct timespec *tp, int res, u64 *jiff)
-{
-       long sec = tp->tv_sec;
-       long nsec = tp->tv_nsec + res - 1;
-
-       if (nsec > NSEC_PER_SEC) {
-               sec++;
-               nsec -= NSEC_PER_SEC;
-       }
-
-       /*
-        * The scaling constants are defined in <linux/time.h>
-        * The difference between there and here is that we do the
-        * res rounding and compute a 64-bit result (well so does that
-        * but it then throws away the high bits).
-        */
-       *jiff =  (mpy_l_X_l_ll(sec, SEC_CONVERSION) +
-                 (mpy_l_X_l_ll(nsec, NSEC_CONVERSION) >> 
-                  (NSEC_JIFFIE_SC - SEC_JIFFIE_SC))) >> SEC_JIFFIE_SC;
-}
-
-/*
- * This function adjusts the timer as needed as a result of the clock
- * being set.  It should only be called for absolute timers, and then
- * under the abs_list lock.  It computes the time difference and sets
- * the new jiffies value in the timer.  It also updates the timers
- * reference wall_to_monotonic value.  It is complicated by the fact
- * that tstojiffies() only handles positive times and it needs to work
- * with both positive and negative times.  Also, for negative offsets,
- * we need to defeat the res round up.
- *
- * Return is true if there is a new time, else false.
- */
-static long add_clockset_delta(struct k_itimer *timr,
-                              struct timespec *new_wall_to)
-{
-       struct timespec delta;
-       int sign = 0;
-       u64 exp;
-
-       set_normalized_timespec(&delta,
-                               new_wall_to->tv_sec -
-                               timr->it.real.wall_to_prev.tv_sec,
-                               new_wall_to->tv_nsec -
-                               timr->it.real.wall_to_prev.tv_nsec);
-       if (likely(!(delta.tv_sec | delta.tv_nsec)))
-               return 0;
-       if (delta.tv_sec < 0) {
-               set_normalized_timespec(&delta,
-                                       -delta.tv_sec,
-                                       1 - delta.tv_nsec -
-                                       posix_clocks[timr->it_clock].res);
-               sign++;
-       }
-       tstojiffie(&delta, posix_clocks[timr->it_clock].res, &exp);
-       timr->it.real.wall_to_prev = *new_wall_to;
-       timr->it.real.timer.expires += (sign ? -exp : exp);
-       return 1;
-}
-
-static void remove_from_abslist(struct k_itimer *timr)
-{
-       if (!list_empty(&timr->it.real.abs_timer_entry)) {
-               spin_lock(&abs_list.lock);
-               list_del_init(&timr->it.real.abs_timer_entry);
-               spin_unlock(&abs_list.lock);
-       }
-}
-
 static void schedule_next_timer(struct k_itimer *timr)
 {
-       struct timespec new_wall_to;
-       struct now_struct now;
-       unsigned long seq;
+       struct hrtimer *timer = &timr->it.real.timer;
 
-       /*
-        * Set up the timer for the next interval (if there is one).
-        * Note: this code uses the abs_timer_lock to protect
-        * it.real.wall_to_prev and must hold it until exp is set, not exactly
-        * obvious...
-
-        * This function is used for CLOCK_REALTIME* and
-        * CLOCK_MONOTONIC* timers.  If we ever want to handle other
-        * CLOCKs, the calling code (do_schedule_next_timer) would need
-        * to pull the "clock" info from the timer and dispatch the
-        * "other" CLOCKs "next timer" code (which, I suppose should
-        * also be added to the k_clock structure).
-        */
-       if (!timr->it.real.incr)
+       if (timr->it.real.interval.tv64 == 0)
                return;
 
-       do {
-               seq = read_seqbegin(&xtime_lock);
-               new_wall_to =   wall_to_monotonic;
-               posix_get_now(&now);
-       } while (read_seqretry(&xtime_lock, seq));
-
-       if (!list_empty(&timr->it.real.abs_timer_entry)) {
-               spin_lock(&abs_list.lock);
-               add_clockset_delta(timr, &new_wall_to);
+       timr->it_overrun += hrtimer_forward(timer, timer->base->get_time(),
+                                           timr->it.real.interval);
 
-               posix_bump_timer(timr, now);
-
-               spin_unlock(&abs_list.lock);
-       } else {
-               posix_bump_timer(timr, now);
-       }
        timr->it_overrun_last = timr->it_overrun;
        timr->it_overrun = -1;
        ++timr->it_requeue_pending;
-       add_timer(&timr->it.real.timer);
+       hrtimer_restart(timer);
 }
 
 /*
@@ -415,15 +283,15 @@ void do_schedule_next_timer(struct siginfo *info)
 
        timr = lock_timer(info->si_tid, &flags);
 
-       if (!timr || timr->it_requeue_pending != info->si_sys_private)
-               goto exit;
+       if (timr && timr->it_requeue_pending == info->si_sys_private) {
+               if (timr->it_clock < 0)
+                       posix_cpu_timer_schedule(timr);
+               else
+                       schedule_next_timer(timr);
+
+               info->si_overrun = timr->it_overrun_last;
+       }
 
-       if (timr->it_clock < 0) /* CPU clock */
-               posix_cpu_timer_schedule(timr);
-       else
-               schedule_next_timer(timr);
-       info->si_overrun = timr->it_overrun_last;
-exit:
        if (timr)
                unlock_timer(timr, flags);
 }
@@ -432,35 +300,31 @@ int posix_timer_event(struct k_itimer *timr,int si_private)
 {
        memset(&timr->sigq->info, 0, sizeof(siginfo_t));
        timr->sigq->info.si_sys_private = si_private;
-       /*
-        * Send signal to the process that owns this timer.
-
-        * This code assumes that all the possible abs_lists share the
-        * same lock (there is only one list at this time). If this is
-        * not the case, the CLOCK info would need to be used to find
-        * the proper abs list lock.
-        */
+       /* Send signal to the process that owns this timer.*/
 
        timr->sigq->info.si_signo = timr->it_sigev_signo;
        timr->sigq->info.si_errno = 0;
        timr->sigq->info.si_code = SI_TIMER;
        timr->sigq->info.si_tid = timr->it_id;
        timr->sigq->info.si_value = timr->it_sigev_value;
+
        if (timr->it_sigev_notify & SIGEV_THREAD_ID) {
-               if (unlikely(timr->it_process->flags & PF_EXITING)) {
-                       timr->it_sigev_notify = SIGEV_SIGNAL;
-                       put_task_struct(timr->it_process);
-                       timr->it_process = timr->it_process->group_leader;
-                       goto group;
-               }
-               return send_sigqueue(timr->it_sigev_signo, timr->sigq,
-                       timr->it_process);
-       }
-       else {
-       group:
-               return send_group_sigqueue(timr->it_sigev_signo, timr->sigq,
-                       timr->it_process);
+               struct task_struct *leader;
+               int ret;
+
+               ret = send_sigqueue(timr->it_sigev_signo, timr->sigq,
+                                   timr->it_process);
+               if (likely(ret >= 0))
+                       return ret;
+
+               timr->it_sigev_notify = SIGEV_SIGNAL;
+               leader = timr->it_process->group_leader;
+               put_task_struct(timr->it_process);
+               timr->it_process = leader;
        }
+
+       return send_group_sigqueue(timr->it_sigev_signo, timr->sigq,
+                                  timr->it_process);
 }
 EXPORT_SYMBOL_GPL(posix_timer_event);
 
@@ -471,67 +335,40 @@ EXPORT_SYMBOL_GPL(posix_timer_event);
 
  * This code is for CLOCK_REALTIME* and CLOCK_MONOTONIC* timers.
  */
-static void posix_timer_fn(unsigned long __data)
+static int posix_timer_fn(struct hrtimer *timer)
 {
-       struct k_itimer *timr = (struct k_itimer *) __data;
+       struct k_itimer *timr;
        unsigned long flags;
-       unsigned long seq;
-       struct timespec delta, new_wall_to;
-       u64 exp = 0;
-       int do_notify = 1;
+       int si_private = 0;
+       int ret = HRTIMER_NORESTART;
 
+       timr = container_of(timer, struct k_itimer, it.real.timer);
        spin_lock_irqsave(&timr->it_lock, flags);
-       set_timer_inactive(timr);
-       if (!list_empty(&timr->it.real.abs_timer_entry)) {
-               spin_lock(&abs_list.lock);
-               do {
-                       seq = read_seqbegin(&xtime_lock);
-                       new_wall_to =   wall_to_monotonic;
-               } while (read_seqretry(&xtime_lock, seq));
-               set_normalized_timespec(&delta,
-                                       new_wall_to.tv_sec -
-                                       timr->it.real.wall_to_prev.tv_sec,
-                                       new_wall_to.tv_nsec -
-                                       timr->it.real.wall_to_prev.tv_nsec);
-               if (likely((delta.tv_sec | delta.tv_nsec ) == 0)) {
-                       /* do nothing, timer is on time */
-               } else if (delta.tv_sec < 0) {
-                       /* do nothing, timer is already late */
-               } else {
-                       /* timer is early due to a clock set */
-                       tstojiffie(&delta,
-                                  posix_clocks[timr->it_clock].res,
-                                  &exp);
-                       timr->it.real.wall_to_prev = new_wall_to;
-                       timr->it.real.timer.expires += exp;
-                       add_timer(&timr->it.real.timer);
-                       do_notify = 0;
-               }
-               spin_unlock(&abs_list.lock);
 
-       }
-       if (do_notify)  {
-               int si_private=0;
+       if (timr->it.real.interval.tv64 != 0)
+               si_private = ++timr->it_requeue_pending;
 
-               if (timr->it.real.incr)
-                       si_private = ++timr->it_requeue_pending;
-               else {
-                       remove_from_abslist(timr);
+       if (posix_timer_event(timr, si_private)) {
+               /*
+                * signal was not sent because of sig_ignor
+                * we will not get a call back to restart it AND
+                * it should be restarted.
+                */
+               if (timr->it.real.interval.tv64 != 0) {
+                       timr->it_overrun +=
+                               hrtimer_forward(timer,
+                                               timer->base->softirq_time,
+                                               timr->it.real.interval);
+                       ret = HRTIMER_RESTART;
+                       ++timr->it_requeue_pending;
                }
-
-               if (posix_timer_event(timr, si_private))
-                       /*
-                        * signal was not sent because of sig_ignor
-                        * we will not get a call back to restart it AND
-                        * it should be restarted.
-                        */
-                       schedule_next_timer(timr);
        }
-       unlock_timer(timr, flags); /* hold thru abs lock to keep irq off */
-}
 
+       unlock_timer(timr, flags);
+       return ret;
+}
 
-static inline struct task_struct * good_sigevent(sigevent_t * event)
+static struct task_struct * good_sigevent(sigevent_t * event)
 {
        struct task_struct *rtn = current->group_leader;
 
@@ -548,7 +385,7 @@ static inline struct task_struct * good_sigevent(sigevent_t * event)
        return rtn;
 }
 
-void register_posix_clock(clockid_t clock_id, struct k_clock *new_clock)
+void register_posix_clock(const clockid_t clock_id, struct k_clock *new_clock)
 {
        if ((unsigned) clock_id >= MAX_CLOCKS) {
                printk("POSIX clock register failed for clock_id %d\n",
@@ -594,7 +431,7 @@ static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
 /* Create a POSIX.1b interval timer. */
 
 asmlinkage long
-sys_timer_create(clockid_t which_clock,
+sys_timer_create(const clockid_t which_clock,
                 struct sigevent __user *timer_event_spec,
                 timer_t __user * created_timer_id)
 {
@@ -620,8 +457,7 @@ sys_timer_create(clockid_t which_clock,
                goto out;
        }
        spin_lock_irq(&idr_lock);
-       error = idr_get_new(&posix_timers_id,
-                           (void *) new_timer,
+       error = idr_get_new(&posix_timers_id, (void *) new_timer,
                            &new_timer_id);
        spin_unlock_irq(&idr_lock);
        if (error == -EAGAIN)
@@ -721,27 +557,6 @@ out:
        return error;
 }
 
-/*
- * good_timespec
- *
- * This function checks the elements of a timespec structure.
- *
- * Arguments:
- * ts       : Pointer to the timespec structure to check
- *
- * Return value:
- * If a NULL pointer was passed in, or the tv_nsec field was less than 0
- * or greater than NSEC_PER_SEC, or the tv_sec field was less than 0,
- * this function returns 0. Otherwise it returns 1.
- */
-static int good_timespec(const struct timespec *ts)
-{
-       if ((!ts) || (ts->tv_sec < 0) ||
-                       ((unsigned) ts->tv_nsec >= NSEC_PER_SEC))
-               return 0;
-       return 1;
-}
-
 /*
  * Locking issues: We need to protect the result of the id look up until
  * we get the timer locked down so it is not deleted under us.  The
@@ -794,39 +609,42 @@ static struct k_itimer * lock_timer(timer_t timer_id, unsigned long *flags)
 static void
 common_timer_get(struct k_itimer *timr, struct itimerspec *cur_setting)
 {
-       unsigned long expires;
-       struct now_struct now;
-
-       do
-               expires = timr->it.real.timer.expires;
-       while ((volatile long) (timr->it.real.timer.expires) != expires);
-
-       posix_get_now(&now);
-
-       if (expires &&
-           ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) &&
-           !timr->it.real.incr &&
-           posix_time_before(&timr->it.real.timer, &now))
-               timr->it.real.timer.expires = expires = 0;
-       if (expires) {
-               if (timr->it_requeue_pending & REQUEUE_PENDING ||
-                   (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
-                       posix_bump_timer(timr, now);
-                       expires = timr->it.real.timer.expires;
-               }
-               else
-                       if (!timer_pending(&timr->it.real.timer))
-                               expires = 0;
-               if (expires)
-                       expires -= now.jiffies;
-       }
-       jiffies_to_timespec(expires, &cur_setting->it_value);
-       jiffies_to_timespec(timr->it.real.incr, &cur_setting->it_interval);
+       ktime_t now, remaining, iv;
+       struct hrtimer *timer = &timr->it.real.timer;
 
-       if (cur_setting->it_value.tv_sec < 0) {
-               cur_setting->it_value.tv_nsec = 1;
-               cur_setting->it_value.tv_sec = 0;
-       }
+       memset(cur_setting, 0, sizeof(struct itimerspec));
+
+       iv = timr->it.real.interval;
+
+       /* interval timer ? */
+       if (iv.tv64)
+               cur_setting->it_interval = ktime_to_timespec(iv);
+       else if (!hrtimer_active(timer) &&
+                (timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
+               return;
+
+       now = timer->base->get_time();
+
+       /*
+        * When a requeue is pending or this is a SIGEV_NONE
+        * timer move the expiry time forward by intervals, so
+        * expiry is > now.
+        */
+       if (iv.tv64 && (timr->it_requeue_pending & REQUEUE_PENDING ||
+           (timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE))
+               timr->it_overrun += hrtimer_forward(timer, now, iv);
+
+       remaining = ktime_sub(timer->expires, now);
+       /* Return 0 only, when the timer is expired and not pending */
+       if (remaining.tv64 <= 0) {
+               /*
+                * A single shot SIGEV_NONE timer must return 0, when
+                * it is expired !
+                */
+               if ((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE)
+                       cur_setting->it_value.tv_nsec = 1;
+       } else
+               cur_setting->it_value = ktime_to_timespec(remaining);
 }
 
 /* Get the time remaining on a POSIX.1b interval timer. */
@@ -850,6 +668,7 @@ sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting)
 
        return 0;
 }
+
 /*
  * Get the number of overruns of a POSIX.1b interval timer.  This is to
  * be the overrun of the timer last delivered.  At the same time we are
@@ -859,7 +678,6 @@ sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting)
  * the call back to do_schedule_next_timer().  So all we need to do is
  * to pick up the frozen overrun.
  */
-
 asmlinkage long
 sys_timer_getoverrun(timer_t timer_id)
 {
@@ -876,166 +694,55 @@ sys_timer_getoverrun(timer_t timer_id)
 
        return overrun;
 }
-/*
- * Adjust for absolute time
- *
- * If absolute time is given and it is not CLOCK_MONOTONIC, we need to
- * adjust for the offset between the timer clock (CLOCK_MONOTONIC) and
- * what ever clock he is using.
- *
- * If it is relative time, we need to add the current (CLOCK_MONOTONIC)
- * time to it to get the proper time for the timer.
- */
-static int adjust_abs_time(struct k_clock *clock, struct timespec *tp, 
-                          int abs, u64 *exp, struct timespec *wall_to)
-{
-       struct timespec now;
-       struct timespec oc = *tp;
-       u64 jiffies_64_f;
-       int rtn =0;
-
-       if (abs) {
-               /*
-                * The mask pick up the 4 basic clocks 
-                */
-               if (!((clock - &posix_clocks[0]) & ~CLOCKS_MASK)) {
-                       jiffies_64_f = do_posix_clock_monotonic_gettime_parts(
-                               &now,  wall_to);
-                       /*
-                        * If we are doing a MONOTONIC clock
-                        */
-                       if((clock - &posix_clocks[0]) & CLOCKS_MONO){
-                               now.tv_sec += wall_to->tv_sec;
-                               now.tv_nsec += wall_to->tv_nsec;
-                       }
-               } else {
-                       /*
-                        * Not one of the basic clocks
-                        */
-                       clock->clock_get(clock - posix_clocks, &now);
-                       jiffies_64_f = get_jiffies_64();
-               }
-               /*
-                * Take away now to get delta
-                */
-               oc.tv_sec -= now.tv_sec;
-               oc.tv_nsec -= now.tv_nsec;
-               /*
-                * Normalize...
-                */
-               while ((oc.tv_nsec - NSEC_PER_SEC) >= 0) {
-                       oc.tv_nsec -= NSEC_PER_SEC;
-                       oc.tv_sec++;
-               }
-               while ((oc.tv_nsec) < 0) {
-                       oc.tv_nsec += NSEC_PER_SEC;
-                       oc.tv_sec--;
-               }
-       }else{
-               jiffies_64_f = get_jiffies_64();
-       }
-       /*
-        * Check if the requested time is prior to now (if so set now)
-        */
-       if (oc.tv_sec < 0)
-               oc.tv_sec = oc.tv_nsec = 0;
-
-       if (oc.tv_sec | oc.tv_nsec)
-               set_normalized_timespec(&oc, oc.tv_sec,
-                                       oc.tv_nsec + clock->res);
-       tstojiffie(&oc, clock->res, exp);
-
-       /*
-        * Check if the requested time is more than the timer code
-        * can handle (if so we error out but return the value too).
-        */
-       if (*exp > ((u64)MAX_JIFFY_OFFSET))
-                       /*
-                        * This is a considered response, not exactly in
-                        * line with the standard (in fact it is silent on
-                        * possible overflows).  We assume such a large 
-                        * value is ALMOST always a programming error and
-                        * try not to compound it by setting a really dumb
-                        * value.
-                        */
-                       rtn = -EINVAL;
-       /*
-        * return the actual jiffies expire time, full 64 bits
-        */
-       *exp += jiffies_64_f;
-       return rtn;
-}
 
 /* Set a POSIX.1b interval timer. */
 /* timr->it_lock is taken. */
-static inline int
+static int
 common_timer_set(struct k_itimer *timr, int flags,
                 struct itimerspec *new_setting, struct itimerspec *old_setting)
 {
-       struct k_clock *clock = &posix_clocks[timr->it_clock];
-       u64 expire_64;
+       struct hrtimer *timer = &timr->it.real.timer;
+       enum hrtimer_mode mode;
 
        if (old_setting)
                common_timer_get(timr, old_setting);
 
        /* disable the timer */
-       timr->it.real.incr = 0;
+       timr->it.real.interval.tv64 = 0;
        /*
         * careful here.  If smp we could be in the "fire" routine which will
         * be spinning as we hold the lock.  But this is ONLY an SMP issue.
         */
-#ifdef CONFIG_SMP
-       if (timer_active(timr) && !del_timer(&timr->it.real.timer))
-               /*
-                * It can only be active if on an other cpu.  Since
-                * we have cleared the interval stuff above, it should
-                * clear once we release the spin lock.  Of course once
-                * we do that anything could happen, including the
-                * complete melt down of the timer.  So return with
-                * a "retry" exit status.
-                */
+       if (hrtimer_try_to_cancel(timer) < 0)
                return TIMER_RETRY;
 
-       set_timer_inactive(timr);
-#else
-       del_timer(&timr->it.real.timer);
-#endif
-       remove_from_abslist(timr);
-
        timr->it_requeue_pending = (timr->it_requeue_pending + 2) & 
                ~REQUEUE_PENDING;
        timr->it_overrun_last = 0;
-       timr->it_overrun = -1;
-       /*
-        *switch off the timer when it_value is zero
-        */
-       if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec) {
-               timr->it.real.timer.expires = 0;
+
+       /* switch off the timer when it_value is zero */
+       if (!new_setting->it_value.tv_sec && !new_setting->it_value.tv_nsec)
                return 0;
-       }
 
-       if (adjust_abs_time(clock,
-                           &new_setting->it_value, flags & TIMER_ABSTIME, 
-                           &expire_64, &(timr->it.real.wall_to_prev))) {
-               return -EINVAL;
-       }
-       timr->it.real.timer.expires = (unsigned long)expire_64;
-       tstojiffie(&new_setting->it_interval, clock->res, &expire_64);
-       timr->it.real.incr = (unsigned long)expire_64;
+       mode = flags & TIMER_ABSTIME ? HRTIMER_ABS : HRTIMER_REL;
+       hrtimer_init(&timr->it.real.timer, timr->it_clock, mode);
+       timr->it.real.timer.function = posix_timer_fn;
 
-       /*
-        * We do not even queue SIGEV_NONE timers!  But we do put them
-        * in the abs list so we can do that right.
-        */
-       if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE))
-               add_timer(&timr->it.real.timer);
-
-       if (flags & TIMER_ABSTIME && clock->abs_struct) {
-               spin_lock(&clock->abs_struct->lock);
-               list_add_tail(&(timr->it.real.abs_timer_entry),
-                             &(clock->abs_struct->list));
-               spin_unlock(&clock->abs_struct->lock);
+       timer->expires = timespec_to_ktime(new_setting->it_value);
+
+       /* Convert interval */
+       timr->it.real.interval = timespec_to_ktime(new_setting->it_interval);
+
+       /* SIGEV_NONE timers are not queued ! See common_timer_get */
+       if (((timr->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE)) {
+               /* Setup correct expiry time for relative timers */
+               if (mode == HRTIMER_REL)
+                       timer->expires = ktime_add(timer->expires,
+                                                  timer->base->get_time());
+               return 0;
        }
+
+       hrtimer_start(timer, timer->expires, mode);
        return 0;
 }
 
@@ -1057,8 +764,8 @@ sys_timer_settime(timer_t timer_id, int flags,
        if (copy_from_user(&new_spec, new_setting, sizeof (new_spec)))
                return -EFAULT;
 
-       if ((!good_timespec(&new_spec.it_interval)) ||
-           (!good_timespec(&new_spec.it_value)))
+       if (!timespec_valid(&new_spec.it_interval) ||
+           !timespec_valid(&new_spec.it_value))
                return -EINVAL;
 retry:
        timr = lock_timer(timer_id, &flag);
@@ -1074,8 +781,8 @@ retry:
                goto retry;
        }
 
-       if (old_setting && !error && copy_to_user(old_setting,
-                                                 &old_spec, sizeof (old_spec)))
+       if (old_setting && !error &&
+           copy_to_user(old_setting, &old_spec, sizeof (old_spec)))
                error = -EFAULT;
 
        return error;
@@ -1083,23 +790,10 @@ retry:
 
 static inline int common_timer_del(struct k_itimer *timer)
 {
-       timer->it.real.incr = 0;
-#ifdef CONFIG_SMP
-       if (timer_active(timer) && !del_timer(&timer->it.real.timer))
-               /*
-                * It can only be active if on an other cpu.  Since
-                * we have cleared the interval stuff above, it should
-                * clear once we release the spin lock.  Of course once
-                * we do that anything could happen, including the
-                * complete melt down of the timer.  So return with
-                * a "retry" exit status.
-                */
-               return TIMER_RETRY;
-#else
-       del_timer(&timer->it.real.timer);
-#endif
-       remove_from_abslist(timer);
+       timer->it.real.interval.tv64 = 0;
 
+       if (hrtimer_try_to_cancel(&timer->it.real.timer) < 0)
+               return TIMER_RETRY;
        return 0;
 }
 
@@ -1115,24 +809,16 @@ sys_timer_delete(timer_t timer_id)
        struct k_itimer *timer;
        long flags;
 
-#ifdef CONFIG_SMP
-       int error;
 retry_delete:
-#endif
        timer = lock_timer(timer_id, &flags);
        if (!timer)
                return -EINVAL;
 
-#ifdef CONFIG_SMP
-       error = timer_delete_hook(timer);
-
-       if (error == TIMER_RETRY) {
+       if (timer_delete_hook(timer) == TIMER_RETRY) {
                unlock_timer(timer, flags);
                goto retry_delete;
        }
-#else
-       timer_delete_hook(timer);
-#endif
+
        spin_lock(&current->sighand->siglock);
        list_del(&timer->list);
        spin_unlock(&current->sighand->siglock);
@@ -1149,29 +835,21 @@ retry_delete:
        release_posix_timer(timer, IT_ID_SET);
        return 0;
 }
+
 /*
  * return timer owned by the process, used by exit_itimers
  */
-static inline void itimer_delete(struct k_itimer *timer)
+static void itimer_delete(struct k_itimer *timer)
 {
        unsigned long flags;
 
-#ifdef CONFIG_SMP
-       int error;
 retry_delete:
-#endif
        spin_lock_irqsave(&timer->it_lock, flags);
 
-#ifdef CONFIG_SMP
-       error = timer_delete_hook(timer);
-
-       if (error == TIMER_RETRY) {
+       if (timer_delete_hook(timer) == TIMER_RETRY) {
                unlock_timer(timer, flags);
                goto retry_delete;
        }
-#else
-       timer_delete_hook(timer);
-#endif
        list_del(&timer->list);
        /*
         * This keeps any tasks waiting on the spin lock from thinking
@@ -1187,7 +865,7 @@ retry_delete:
 }
 
 /*
- * This is called by __exit_signal, only when there are no more
+ * This is called by do_exit or de_thread, only when there are no more
  * references to the shared signal_struct.
  */
 void exit_itimers(struct signal_struct *sig)
@@ -1198,76 +876,17 @@ void exit_itimers(struct signal_struct *sig)
                tmr = list_entry(sig->posix_timers.next, struct k_itimer, list);
                itimer_delete(tmr);
        }
-       del_timer_sync(&sig->real_timer);
-}
-
-/*
- * And now for the "clock" calls
- *
- * These functions are called both from timer functions (with the timer
- * spin_lock_irq() held and from clock calls with no locking.  They must
- * use the save flags versions of locks.
- */
-
-/*
- * We do ticks here to avoid the irq lock ( they take sooo long).
- * The seqlock is great here.  Since we a reader, we don't really care
- * if we are interrupted since we don't take lock that will stall us or
- * any other cpu. Voila, no irq lock is needed.
- *
- */
-
-static u64 do_posix_clock_monotonic_gettime_parts(
-       struct timespec *tp, struct timespec *mo)
-{
-       u64 jiff;
-       unsigned int seq;
-
-       do {
-               seq = read_seqbegin(&xtime_lock);
-               getnstimeofday(tp);
-               *mo = wall_to_monotonic;
-               jiff = jiffies_64;
-
-       } while(read_seqretry(&xtime_lock, seq));
-
-       return jiff;
-}
-
-static int do_posix_clock_monotonic_get(clockid_t clock, struct timespec *tp)
-{
-       struct timespec wall_to_mono;
-
-       do_posix_clock_monotonic_gettime_parts(tp, &wall_to_mono);
-
-       tp->tv_sec += wall_to_mono.tv_sec;
-       tp->tv_nsec += wall_to_mono.tv_nsec;
-
-       if ((tp->tv_nsec - NSEC_PER_SEC) > 0) {
-               tp->tv_nsec -= NSEC_PER_SEC;
-               tp->tv_sec++;
-       }
-       return 0;
 }
 
-int do_posix_clock_monotonic_gettime(struct timespec *tp)
-{
-       return do_posix_clock_monotonic_get(CLOCK_MONOTONIC, tp);
-}
-
-int do_posix_clock_nosettime(clockid_t clockid, struct timespec *tp)
+/* Not available / possible... functions */
+int do_posix_clock_nosettime(const clockid_t clockid, struct timespec *tp)
 {
        return -EINVAL;
 }
 EXPORT_SYMBOL_GPL(do_posix_clock_nosettime);
 
-int do_posix_clock_notimer_create(struct k_itimer *timer)
-{
-       return -EINVAL;
-}
-EXPORT_SYMBOL_GPL(do_posix_clock_notimer_create);
-
-int do_posix_clock_nonanosleep(clockid_t clock, int flags, struct timespec *t)
+int do_posix_clock_nonanosleep(const clockid_t clock, int flags,
+                              struct timespec *t, struct timespec __user *r)
 {
 #ifndef ENOTSUP
        return -EOPNOTSUPP;     /* aka ENOTSUP in userland for POSIX */
@@ -1277,8 +896,8 @@ int do_posix_clock_nonanosleep(clockid_t clock, int flags, struct timespec *t)
 }
 EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep);
 
-asmlinkage long
-sys_clock_settime(clockid_t which_clock, const struct timespec __user *tp)
+asmlinkage long sys_clock_settime(const clockid_t which_clock,
+                                 const struct timespec __user *tp)
 {
        struct timespec new_tp;
 
@@ -1291,7 +910,7 @@ sys_clock_settime(clockid_t which_clock, const struct timespec __user *tp)
 }
 
 asmlinkage long
-sys_clock_gettime(clockid_t which_clock, struct timespec __user *tp)
+sys_clock_gettime(const clockid_t which_clock, struct timespec __user *tp)
 {
        struct timespec kernel_tp;
        int error;
@@ -1308,7 +927,7 @@ sys_clock_gettime(clockid_t which_clock, struct timespec __user *tp)
 }
 
 asmlinkage long
-sys_clock_getres(clockid_t which_clock, struct timespec __user *tp)
+sys_clock_getres(const clockid_t which_clock, struct timespec __user *tp)
 {
        struct timespec rtn_tp;
        int error;
@@ -1326,125 +945,22 @@ sys_clock_getres(clockid_t which_clock, struct timespec __user *tp)
        return error;
 }
 
-static void nanosleep_wake_up(unsigned long __data)
-{
-       struct task_struct *p = (struct task_struct *) __data;
-
-       wake_up_process(p);
-}
-
 /*
- * The standard says that an absolute nanosleep call MUST wake up at
- * the requested time in spite of clock settings.  Here is what we do:
- * For each nanosleep call that needs it (only absolute and not on
- * CLOCK_MONOTONIC* (as it can not be set)) we thread a little structure
- * into the "nanosleep_abs_list".  All we need is the task_struct pointer.
- * When ever the clock is set we just wake up all those tasks.  The rest
- * is done by the while loop in clock_nanosleep().
- *
- * On locking, clock_was_set() is called from update_wall_clock which
- * holds (or has held for it) a write_lock_irq( xtime_lock) and is
- * called from the timer bh code.  Thus we need the irq save locks.
- *
- * Also, on the call from update_wall_clock, that is done as part of a
- * softirq thing.  We don't want to delay the system that much (possibly
- * long list of timers to fix), so we defer that work to keventd.
+ * nanosleep for monotonic and realtime clocks
  */
-
-static DECLARE_WAIT_QUEUE_HEAD(nanosleep_abs_wqueue);
-static DECLARE_WORK(clock_was_set_work, (void(*)(void*))clock_was_set, NULL);
-
-static DECLARE_MUTEX(clock_was_set_lock);
-
-void clock_was_set(void)
+static int common_nsleep(const clockid_t which_clock, int flags,
+                        struct timespec *tsave, struct timespec __user *rmtp)
 {
-       struct k_itimer *timr;
-       struct timespec new_wall_to;
-       LIST_HEAD(cws_list);
-       unsigned long seq;
-
-
-       if (unlikely(in_interrupt())) {
-               schedule_work(&clock_was_set_work);
-               return;
-       }
-       wake_up_all(&nanosleep_abs_wqueue);
-
-       /*
-        * Check if there exist TIMER_ABSTIME timers to correct.
-        *
-        * Notes on locking: This code is run in task context with irq
-        * on.  We CAN be interrupted!  All other usage of the abs list
-        * lock is under the timer lock which holds the irq lock as
-        * well.  We REALLY don't want to scan the whole list with the
-        * interrupt system off, AND we would like a sequence lock on
-        * this code as well.  Since we assume that the clock will not
-        * be set often, it seems ok to take and release the irq lock
-        * for each timer.  In fact add_timer will do this, so this is
-        * not an issue.  So we know when we are done, we will move the
-        * whole list to a new location.  Then as we process each entry,
-        * we will move it to the actual list again.  This way, when our
-        * copy is empty, we are done.  We are not all that concerned
-        * about preemption so we will use a semaphore lock to protect
-        * aginst reentry.  This way we will not stall another
-        * processor.  It is possible that this may delay some timers
-        * that should have expired, given the new clock, but even this
-        * will be minimal as we will always update to the current time,
-        * even if it was set by a task that is waiting for entry to
-        * this code.  Timers that expire too early will be caught by
-        * the expire code and restarted.
-
-        * Absolute timers that repeat are left in the abs list while
-        * waiting for the task to pick up the signal.  This means we
-        * may find timers that are not in the "add_timer" list, but are
-        * in the abs list.  We do the same thing for these, save
-        * putting them back in the "add_timer" list.  (Note, these are
-        * left in the abs list mainly to indicate that they are
-        * ABSOLUTE timers, a fact that is used by the re-arm code, and
-        * for which we have no other flag.)
-
-        */
-
-       down(&clock_was_set_lock);
-       spin_lock_irq(&abs_list.lock);
-       list_splice_init(&abs_list.list, &cws_list);
-       spin_unlock_irq(&abs_list.lock);
-       do {
-               do {
-                       seq = read_seqbegin(&xtime_lock);
-                       new_wall_to =   wall_to_monotonic;
-               } while (read_seqretry(&xtime_lock, seq));
-
-               spin_lock_irq(&abs_list.lock);
-               if (list_empty(&cws_list)) {
-                       spin_unlock_irq(&abs_list.lock);
-                       break;
-               }
-               timr = list_entry(cws_list.next, struct k_itimer,
-                                 it.real.abs_timer_entry);
-
-               list_del_init(&timr->it.real.abs_timer_entry);
-               if (add_clockset_delta(timr, &new_wall_to) &&
-                   del_timer(&timr->it.real.timer))  /* timer run yet? */
-                       add_timer(&timr->it.real.timer);
-               list_add(&timr->it.real.abs_timer_entry, &abs_list.list);
-               spin_unlock_irq(&abs_list.lock);
-       } while (1);
-
-       up(&clock_was_set_lock);
+       return hrtimer_nanosleep(tsave, rmtp, flags & TIMER_ABSTIME ?
+                                HRTIMER_ABS : HRTIMER_REL, which_clock);
 }
 
-long clock_nanosleep_restart(struct restart_block *restart_block);
-
 asmlinkage long
-sys_clock_nanosleep(clockid_t which_clock, int flags,
+sys_clock_nanosleep(const clockid_t which_clock, int flags,
                    const struct timespec __user *rqtp,
                    struct timespec __user *rmtp)
 {
        struct timespec t;
-       struct restart_block *restart_block =
-           &(current_thread_info()->restart_block);
-       int ret;
 
        if (invalid_clockid(which_clock))
                return -EINVAL;
@@ -1452,135 +968,9 @@ sys_clock_nanosleep(clockid_t which_clock, int flags,
        if (copy_from_user(&t, rqtp, sizeof (struct timespec)))
                return -EFAULT;
 
-       if ((unsigned) t.tv_nsec >= NSEC_PER_SEC || t.tv_sec < 0)
+       if (!timespec_valid(&t))
                return -EINVAL;
 
-       /*
-        * Do this here as nsleep function does not have the real address.
-        */
-       restart_block->arg1 = (unsigned long)rmtp;
-
-       ret = CLOCK_DISPATCH(which_clock, nsleep, (which_clock, flags, &t));
-
-       if ((ret == -ERESTART_RESTARTBLOCK) && rmtp &&
-                                       copy_to_user(rmtp, &t, sizeof (t)))
-               return -EFAULT;
-       return ret;
-}
-
-
-static int common_nsleep(clockid_t which_clock,
-                        int flags, struct timespec *tsave)
-{
-       struct timespec t, dum;
-       struct timer_list new_timer;
-       DECLARE_WAITQUEUE(abs_wqueue, current);
-       u64 rq_time = (u64)0;
-       s64 left;
-       int abs;
-       struct restart_block *restart_block =
-           &current_thread_info()->restart_block;
-
-       abs_wqueue.flags = 0;
-       init_timer(&new_timer);
-       new_timer.expires = 0;
-       new_timer.data = (unsigned long) current;
-       new_timer.function = nanosleep_wake_up;
-       abs = flags & TIMER_ABSTIME;
-
-       if (restart_block->fn == clock_nanosleep_restart) {
-               /*
-                * Interrupted by a non-delivered signal, pick up remaining
-                * time and continue.  Remaining time is in arg2 & 3.
-                */
-               restart_block->fn = do_no_restart_syscall;
-
-               rq_time = restart_block->arg3;
-               rq_time = (rq_time << 32) + restart_block->arg2;
-               if (!rq_time)
-                       return -EINTR;
-               left = rq_time - get_jiffies_64();
-               if (left <= (s64)0)
-                       return 0;       /* Already passed */
-       }
-
-       if (abs && (posix_clocks[which_clock].clock_get !=
-                           posix_clocks[CLOCK_MONOTONIC].clock_get))
-               add_wait_queue(&nanosleep_abs_wqueue, &abs_wqueue);
-
-       do {
-               t = *tsave;
-               if (abs || !rq_time) {
-                       adjust_abs_time(&posix_clocks[which_clock], &t, abs,
-                                       &rq_time, &dum);
-               }
-
-               left = rq_time - get_jiffies_64();
-               if (left >= (s64)MAX_JIFFY_OFFSET)
-                       left = (s64)MAX_JIFFY_OFFSET;
-               if (left < (s64)0)
-                       break;
-
-               new_timer.expires = jiffies + left;
-               __set_current_state(TASK_INTERRUPTIBLE);
-               add_timer(&new_timer);
-
-               schedule();
-
-               del_timer_sync(&new_timer);
-               left = rq_time - get_jiffies_64();
-       } while (left > (s64)0 && !test_thread_flag(TIF_SIGPENDING));
-
-       if (abs_wqueue.task_list.next)
-               finish_wait(&nanosleep_abs_wqueue, &abs_wqueue);
-
-       if (left > (s64)0) {
-
-               /*
-                * Always restart abs calls from scratch to pick up any
-                * clock shifting that happened while we are away.
-                */
-               if (abs)
-                       return -ERESTARTNOHAND;
-
-               left *= TICK_NSEC;
-               tsave->tv_sec = div_long_long_rem(left, 
-                                                 NSEC_PER_SEC, 
-                                                 &tsave->tv_nsec);
-               /*
-                * Restart works by saving the time remaing in 
-                * arg2 & 3 (it is 64-bits of jiffies).  The other
-                * info we need is the clock_id (saved in arg0). 
-                * The sys_call interface needs the users 
-                * timespec return address which _it_ saves in arg1.
-                * Since we have cast the nanosleep call to a clock_nanosleep
-                * both can be restarted with the same code.
-                */
-               restart_block->fn = clock_nanosleep_restart;
-               restart_block->arg0 = which_clock;
-               /*
-                * Caller sets arg1
-                */
-               restart_block->arg2 = rq_time & 0xffffffffLL;
-               restart_block->arg3 = rq_time >> 32;
-
-               return -ERESTART_RESTARTBLOCK;
-       }
-
-       return 0;
-}
-/*
- * This will restart clock_nanosleep.
- */
-long
-clock_nanosleep_restart(struct restart_block *restart_block)
-{
-       struct timespec t;
-       int ret = common_nsleep(restart_block->arg0, 0, &t);
-
-       if ((ret == -ERESTART_RESTARTBLOCK) && restart_block->arg1 &&
-           copy_to_user((struct timespec __user *)(restart_block->arg1), &t,
-                        sizeof (t)))
-               return -EFAULT;
-       return ret;
+       return CLOCK_DISPATCH(which_clock, nsleep,
+                             (which_clock, flags, &t, rmtp));
 }