Add build option to to set the default panic timeout.
[linux-2.6.git] / kernel / posix-timers.c
index b7b532a..4556182 100644 (file)
@@ -1,5 +1,5 @@
 /*
- * linux/kernel/posix_timers.c
+ * linux/kernel/posix-timers.c
  *
  *
  * 2002-10-15  Posix Clocks & timers
  * POSIX clocks & timers
  */
 #include <linux/mm.h>
-#include <linux/smp_lock.h>
 #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/list.h>
 #include <linux/init.h>
 #include <linux/compiler.h>
 #include <linux/idr.h>
+#include <linux/posix-clock.h>
 #include <linux/posix-timers.h>
 #include <linux/syscalls.h>
 #include <linux/wait.h>
 #include <linux/workqueue.h>
 #include <linux/module.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
@@ -84,7 +69,7 @@ static inline u64  mpy_l_X_l_ll(unsigned long mpy1,unsigned long mpy2)
 /*
  * Lets keep our timers in a slab cache :-)
  */
-static kmem_cache_t *posix_timers_cache;
+static struct kmem_cache *posix_timers_cache;
 static struct idr posix_timers_id;
 static DEFINE_SPINLOCK(idr_lock);
 
@@ -97,6 +82,14 @@ static DEFINE_SPINLOCK(idr_lock);
 #error "SIGEV_THREAD_ID must not share bit with other SIGEV values!"
 #endif
 
+/*
+ * parisc wants ENOTSUP instead of EOPNOTSUPP
+ */
+#ifndef ENOTSUP
+# define ENANOSLEEP_NOTSUP EOPNOTSUPP
+#else
+# define ENANOSLEEP_NOTSUP ENOTSUP
+#endif
 
 /*
  * The timer ID is turned into a timer address by idr_find().
@@ -110,11 +103,7 @@ static DEFINE_SPINLOCK(idr_lock);
 /*
  * CLOCKs: The POSIX standard calls for a couple of clocks and allows us
  *         to implement others.  This structure defines the various
- *         clocks and allows the possibility of adding others.  We
- *         provide an interface to add clocks to the table and expect
- *         the "arch" code to add at least one clock that is high
- *         resolution.  Here we define the standard CLOCK_REALTIME as a
- *         1/HZ resolution clock.
+ *         clocks.
  *
  * RESOLUTION: Clock resolution is used to round up timer and interval
  *         times, NOT to report clock times, which are reported with as
@@ -124,20 +113,13 @@ static DEFINE_SPINLOCK(idr_lock);
  *         necessary code is written.  The standard says we should say
  *         something about this issue in the documentation...
  *
- * FUNCTIONS: The CLOCKs structure defines possible functions to handle
- *         various clock functions.  For clocks that use the standard
- *         system timer code these entries should be NULL.  This will
- *         allow dispatch without the overhead of indirect function
- *         calls.  CLOCKS that depend on other sources (e.g. WWV or GPS)
- *         must supply functions here, even if the function just returns
- *         ENOSYS.  The standard POSIX timer management code assumes the
- *         following: 1.) The k_itimer struct (sched.h) is used for the
- *         timer.  2.) The list, it_lock, it_clock, it_id and it_process
- *         fields are not modified by timer code.
+ * FUNCTIONS: The CLOCKs structure defines possible functions to
+ *         handle various clock functions.
  *
- *          At this time all functions EXCEPT clock_nanosleep can be
- *          redirected by the CLOCKS structure.  Clock_nanosleep is in
- *          there, but the code ignores it.
+ *         The standard POSIX timer management code assumes the
+ *         following: 1.) The k_itimer struct (sched.h) is used for
+ *         the timer.  2.) The list, it_lock, it_clock, it_id and
+ *         it_pid fields are not modified by timer code.
  *
  * Permissions: It is assumed that the clock_settime() function defined
  *         for each clock will take care of permission checks.  Some
@@ -148,95 +130,95 @@ 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 int common_timer_create(struct k_itimer *new_timer);
+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 enum hrtimer_restart posix_timer_fn(struct hrtimer *data);
 
-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 struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags);
 
-static struct k_itimer *lock_timer(timer_t timer_id, unsigned long *flags);
+#define lock_timer(tid, flags)                                            \
+({     struct k_itimer *__timr;                                           \
+       __cond_lock(&__timr->it_lock, __timr = __lock_timer(tid, flags));  \
+       __timr;                                                            \
+})
 
 static inline void unlock_timer(struct k_itimer *timr, unsigned long flags)
 {
        spin_unlock_irqrestore(&timr->it_lock, flags);
 }
 
+/* Get clock_realtime */
+static int posix_clock_realtime_get(clockid_t which_clock, struct timespec *tp)
+{
+       ktime_get_real_ts(tp);
+       return 0;
+}
+
+/* Set clock_realtime */
+static int posix_clock_realtime_set(const clockid_t which_clock,
+                                   const struct timespec *tp)
+{
+       return do_sys_settimeofday(tp, NULL);
+}
+
+static int posix_clock_realtime_adj(const clockid_t which_clock,
+                                   struct timex *t)
+{
+       return do_adjtimex(t);
+}
+
 /*
- * Call the k_clock hook function if non-null, or the default function.
+ * Get monotonic time for posix timers
  */
-#define CLOCK_DISPATCH(clock, call, arglist) \
-       ((clock) < 0 ? posix_cpu_##call arglist : \
-        (posix_clocks[clock].call != NULL \
-         ? (*posix_clocks[clock].call) arglist : common_##call arglist))
+static int posix_ktime_get_ts(clockid_t which_clock, struct timespec *tp)
+{
+       ktime_get_ts(tp);
+       return 0;
+}
 
 /*
- * Default clock hook functions when the struct k_clock passed
- * to register_posix_clock leaves a function pointer null.
- *
- * The function common_CALL is the default implementation for
- * the function pointer CALL in struct k_clock.
+ * Get monotonic-raw time for posix timers
  */
-
-static inline int common_clock_getres(clockid_t which_clock,
-                                     struct timespec *tp)
+static int posix_get_monotonic_raw(clockid_t which_clock, struct timespec *tp)
 {
-       tp->tv_sec = 0;
-       tp->tv_nsec = posix_clocks[which_clock].res;
+       getrawmonotonic(tp);
        return 0;
 }
 
-static inline int common_clock_get(clockid_t which_clock, struct timespec *tp)
+
+static int posix_get_realtime_coarse(clockid_t which_clock, struct timespec *tp)
 {
-       getnstimeofday(tp);
+       *tp = current_kernel_time();
        return 0;
 }
 
-static inline int common_clock_set(clockid_t which_clock, struct timespec *tp)
+static int posix_get_monotonic_coarse(clockid_t which_clock,
+                                               struct timespec *tp)
 {
-       return do_sys_settimeofday(tp, NULL);
+       *tp = get_monotonic_coarse();
+       return 0;
 }
 
-static inline int common_timer_create(struct k_itimer *new_timer)
+static int posix_get_coarse_res(const clockid_t which_clock, struct timespec *tp)
 {
-       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;
+       *tp = ktime_to_timespec(KTIME_LOW_RES);
        return 0;
 }
 
-/*
- * These ones are defined below.
- */
-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 int posix_get_boottime(const clockid_t which_clock, struct timespec *tp)
 {
-       if (which_clock < 0)    /* CPU clock, posix_cpu_* will check it */
-               return 0;
-       if ((unsigned) which_clock >= MAX_CLOCKS)
-               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_boottime(tp);
+       return 0;
 }
 
 
@@ -245,137 +227,82 @@ static inline int invalid_clockid(clockid_t which_clock)
  */
 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,
+               .clock_get      = posix_clock_realtime_get,
+               .clock_set      = posix_clock_realtime_set,
+               .clock_adj      = posix_clock_realtime_adj,
+               .nsleep         = common_nsleep,
+               .nsleep_restart = hrtimer_nanosleep_restart,
+               .timer_create   = common_timer_create,
+               .timer_set      = common_timer_set,
+               .timer_get      = common_timer_get,
+               .timer_del      = common_timer_del,
+       };
+       struct k_clock clock_monotonic = {
+               .clock_getres   = hrtimer_get_res,
+               .clock_get      = posix_ktime_get_ts,
+               .nsleep         = common_nsleep,
+               .nsleep_restart = hrtimer_nanosleep_restart,
+               .timer_create   = common_timer_create,
+               .timer_set      = common_timer_set,
+               .timer_get      = common_timer_get,
+               .timer_del      = common_timer_del,
+       };
+       struct k_clock clock_monotonic_raw = {
+               .clock_getres   = hrtimer_get_res,
+               .clock_get      = posix_get_monotonic_raw,
+       };
+       struct k_clock clock_realtime_coarse = {
+               .clock_getres   = posix_get_coarse_res,
+               .clock_get      = posix_get_realtime_coarse,
        };
-       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_coarse = {
+               .clock_getres   = posix_get_coarse_res,
+               .clock_get      = posix_get_monotonic_coarse,
+       };
+       struct k_clock clock_boottime = {
+               .clock_getres   = hrtimer_get_res,
+               .clock_get      = posix_get_boottime,
+               .nsleep         = common_nsleep,
+               .nsleep_restart = hrtimer_nanosleep_restart,
+               .timer_create   = common_timer_create,
+               .timer_set      = common_timer_set,
+               .timer_get      = common_timer_get,
+               .timer_del      = common_timer_del,
        };
 
-       register_posix_clock(CLOCK_REALTIME, &clock_realtime);
-       register_posix_clock(CLOCK_MONOTONIC, &clock_monotonic);
+       posix_timers_register_clock(CLOCK_REALTIME, &clock_realtime);
+       posix_timers_register_clock(CLOCK_MONOTONIC, &clock_monotonic);
+       posix_timers_register_clock(CLOCK_MONOTONIC_RAW, &clock_monotonic_raw);
+       posix_timers_register_clock(CLOCK_REALTIME_COARSE, &clock_realtime_coarse);
+       posix_timers_register_clock(CLOCK_MONOTONIC_COARSE, &clock_monotonic_coarse);
+       posix_timers_register_clock(CLOCK_BOOTTIME, &clock_boottime);
 
        posix_timers_cache = kmem_cache_create("posix_timers_cache",
-                                       sizeof (struct k_itimer), 0, 0, NULL, NULL);
+                                       sizeof (struct k_itimer), 0, SLAB_PANIC,
+                                       NULL);
        idr_init(&posix_timers_id);
        return 0;
 }
 
 __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);
-
-               posix_bump_timer(timr, now);
+       timr->it_overrun += (unsigned int) hrtimer_forward(timer,
+                                               timer->base->get_time(),
+                                               timr->it.real.interval);
 
-               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);
 }
 
 /*
@@ -386,7 +313,7 @@ static void schedule_next_timer(struct k_itimer *timr)
  * restarted (i.e. we have flagged this in the sys_private entry of the
  * info block).
  *
- * To protect aginst the timer going away while the interrupt is queued,
+ * To protect against the timer going away while the interrupt is queued,
  * we require that the it_requeue_pending flag be set.
  */
 void do_schedule_next_timer(struct siginfo *info)
@@ -396,54 +323,45 @@ 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);
 }
 
-int posix_timer_event(struct k_itimer *timr,int si_private)
+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;
+       struct task_struct *task;
+       int shared, ret = -1;
        /*
-        * 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.
+        * FIXME: if ->sigq is queued we can race with
+        * dequeue_signal()->do_schedule_next_timer().
+        *
+        * If dequeue_signal() sees the "right" value of
+        * si_sys_private it calls do_schedule_next_timer().
+        * We re-queue ->sigq and drop ->it_lock().
+        * do_schedule_next_timer() locks the timer
+        * and re-schedules it while ->sigq is pending.
+        * Not really bad, but not that we want.
         */
+       timr->sigq->info.si_sys_private = si_private;
 
-       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) {
-               struct task_struct *leader;
-               int 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;
+       rcu_read_lock();
+       task = pid_task(timr->it_pid, PIDTYPE_PID);
+       if (task) {
+               shared = !(timr->it_sigev_notify & SIGEV_THREAD_ID);
+               ret = send_sigqueue(timr->sigq, task, shared);
        }
-
-       return send_group_sigqueue(timr->it_sigev_signo, timr->sigq,
-                                  timr->it_process);
+       rcu_read_unlock();
+       /* If we failed to send the signal the timer stops. */
+       return ret > 0;
 }
 EXPORT_SYMBOL_GPL(posix_timer_event);
 
@@ -454,72 +372,77 @@ 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 enum hrtimer_restart 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;
+       enum hrtimer_restart ret = HRTIMER_NORESTART;
 
+       timr = container_of(timer, struct k_itimer, it.real.timer);
        spin_lock_irqsave(&timr->it_lock, flags);
-       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) {
+                       ktime_t now = hrtimer_cb_get_time(timer);
 
-               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.
+                        * FIXME: What we really want, is to stop this
+                        * timer completely and restart it in case the
+                        * SIG_IGN is removed. This is a non trivial
+                        * change which involves sighand locking
+                        * (sigh !), which we don't want to do late in
+                        * the release cycle.
+                        *
+                        * For now we just let timers with an interval
+                        * less than a jiffie expire every jiffie to
+                        * avoid softirq starvation in case of SIG_IGN
+                        * and a very small interval, which would put
+                        * the timer right back on the softirq pending
+                        * list. By moving now ahead of time we trick
+                        * hrtimer_forward() to expire the timer
+                        * later, while we still maintain the overrun
+                        * accuracy, but have some inconsistency in
+                        * the timer_gettime() case. This is at least
+                        * better than a starved softirq. A more
+                        * complex fix which solves also another related
+                        * inconsistency is already in the pipeline.
                         */
-                       schedule_next_timer(timr);
+#ifdef CONFIG_HIGH_RES_TIMERS
+                       {
+                               ktime_t kj = ktime_set(0, NSEC_PER_SEC / HZ);
+
+                               if (timr->it.real.interval.tv64 < kj.tv64)
+                                       now = ktime_add(now, kj);
+                       }
+#endif
+                       timr->it_overrun += (unsigned int)
+                               hrtimer_forward(timer, now,
+                                               timr->it.real.interval);
+                       ret = HRTIMER_RESTART;
+                       ++timr->it_requeue_pending;
+               }
        }
-       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 pid *good_sigevent(sigevent_t * event)
 {
        struct task_struct *rtn = current->group_leader;
 
        if ((event->sigev_notify & SIGEV_THREAD_ID ) &&
-               (!(rtn = find_task_by_pid(event->sigev_notify_thread_id)) ||
-                rtn->tgid != current->tgid ||
+               (!(rtn = find_task_by_vpid(event->sigev_notify_thread_id)) ||
+                !same_thread_group(rtn, current) ||
                 (event->sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_SIGNAL))
                return NULL;
 
@@ -527,35 +450,54 @@ static inline struct task_struct * good_sigevent(sigevent_t * event)
            ((event->sigev_signo <= 0) || (event->sigev_signo > SIGRTMAX)))
                return NULL;
 
-       return rtn;
+       return task_pid(rtn);
 }
 
-void register_posix_clock(clockid_t clock_id, struct k_clock *new_clock)
+void posix_timers_register_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",
+               printk(KERN_WARNING "POSIX clock register failed for clock_id %d\n",
+                      clock_id);
+               return;
+       }
+
+       if (!new_clock->clock_get) {
+               printk(KERN_WARNING "POSIX clock id %d lacks clock_get()\n",
+                      clock_id);
+               return;
+       }
+       if (!new_clock->clock_getres) {
+               printk(KERN_WARNING "POSIX clock id %d lacks clock_getres()\n",
                       clock_id);
                return;
        }
 
        posix_clocks[clock_id] = *new_clock;
 }
-EXPORT_SYMBOL_GPL(register_posix_clock);
+EXPORT_SYMBOL_GPL(posix_timers_register_clock);
 
 static struct k_itimer * alloc_posix_timer(void)
 {
        struct k_itimer *tmr;
-       tmr = kmem_cache_alloc(posix_timers_cache, GFP_KERNEL);
+       tmr = kmem_cache_zalloc(posix_timers_cache, GFP_KERNEL);
        if (!tmr)
                return tmr;
-       memset(tmr, 0, sizeof (struct k_itimer));
        if (unlikely(!(tmr->sigq = sigqueue_alloc()))) {
                kmem_cache_free(posix_timers_cache, tmr);
-               tmr = NULL;
+               return NULL;
        }
+       memset(&tmr->sigq->info, 0, sizeof(siginfo_t));
        return tmr;
 }
 
+static void k_itimer_rcu_free(struct rcu_head *head)
+{
+       struct k_itimer *tmr = container_of(head, struct k_itimer, it.rcu);
+
+       kmem_cache_free(posix_timers_cache, tmr);
+}
+
 #define IT_ID_SET      1
 #define IT_ID_NOT_SET  0
 static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
@@ -566,30 +508,44 @@ static void release_posix_timer(struct k_itimer *tmr, int it_id_set)
                idr_remove(&posix_timers_id, tmr->it_id);
                spin_unlock_irqrestore(&idr_lock, flags);
        }
+       put_pid(tmr->it_pid);
        sigqueue_free(tmr->sigq);
-       if (unlikely(tmr->it_process) &&
-           tmr->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID))
-               put_task_struct(tmr->it_process);
-       kmem_cache_free(posix_timers_cache, tmr);
+       call_rcu(&tmr->it.rcu, k_itimer_rcu_free);
+}
+
+static struct k_clock *clockid_to_kclock(const clockid_t id)
+{
+       if (id < 0)
+               return (id & CLOCKFD_MASK) == CLOCKFD ?
+                       &clock_posix_dynamic : &clock_posix_cpu;
+
+       if (id >= MAX_CLOCKS || !posix_clocks[id].clock_getres)
+               return NULL;
+       return &posix_clocks[id];
+}
+
+static int common_timer_create(struct k_itimer *new_timer)
+{
+       hrtimer_init(&new_timer->it.real.timer, new_timer->it_clock, 0);
+       return 0;
 }
 
 /* Create a POSIX.1b interval timer. */
 
-asmlinkage long
-sys_timer_create(clockid_t which_clock,
-                struct sigevent __user *timer_event_spec,
-                timer_t __user * created_timer_id)
+SYSCALL_DEFINE3(timer_create, const clockid_t, which_clock,
+               struct sigevent __user *, timer_event_spec,
+               timer_t __user *, created_timer_id)
 {
-       int error = 0;
-       struct k_itimer *new_timer = NULL;
-       int new_timer_id;
-       struct task_struct *process = NULL;
-       unsigned long flags;
+       struct k_clock *kc = clockid_to_kclock(which_clock);
+       struct k_itimer *new_timer;
+       int error, new_timer_id;
        sigevent_t event;
        int it_id_set = IT_ID_NOT_SET;
 
-       if (invalid_clockid(which_clock))
+       if (!kc)
                return -EINVAL;
+       if (!kc->timer_create)
+               return -EOPNOTSUPP;
 
        new_timer = alloc_posix_timer();
        if (unlikely(!new_timer))
@@ -602,15 +558,13 @@ sys_timer_create(clockid_t which_clock,
                goto out;
        }
        spin_lock_irq(&idr_lock);
-       error = idr_get_new(&posix_timers_id,
-                           (void *) new_timer,
-                           &new_timer_id);
+       error = idr_get_new(&posix_timers_id, new_timer, &new_timer_id);
        spin_unlock_irq(&idr_lock);
-       if (error == -EAGAIN)
-               goto retry;
-       else if (error) {
+       if (error) {
+               if (error == -EAGAIN)
+                       goto retry;
                /*
-                * Wierd looking, but we return EAGAIN if the IDR is
+                * Weird looking, but we return EAGAIN if the IDR is
                 * full (proper POSIX return value for this)
                 */
                error = -EAGAIN;
@@ -621,140 +575,83 @@ sys_timer_create(clockid_t which_clock,
        new_timer->it_id = (timer_t) new_timer_id;
        new_timer->it_clock = which_clock;
        new_timer->it_overrun = -1;
-       error = CLOCK_DISPATCH(which_clock, timer_create, (new_timer));
-       if (error)
-               goto out;
 
-       /*
-        * return the timer_id now.  The next step is hard to
-        * back out if there is an error.
-        */
-       if (copy_to_user(created_timer_id,
-                        &new_timer_id, sizeof (new_timer_id))) {
-               error = -EFAULT;
-               goto out;
-       }
        if (timer_event_spec) {
                if (copy_from_user(&event, timer_event_spec, sizeof (event))) {
                        error = -EFAULT;
                        goto out;
                }
-               new_timer->it_sigev_notify = event.sigev_notify;
-               new_timer->it_sigev_signo = event.sigev_signo;
-               new_timer->it_sigev_value = event.sigev_value;
-
-               read_lock(&tasklist_lock);
-               if ((process = good_sigevent(&event))) {
-                       /*
-                        * We may be setting up this process for another
-                        * thread.  It may be exiting.  To catch this
-                        * case the we check the PF_EXITING flag.  If
-                        * the flag is not set, the siglock will catch
-                        * him before it is too late (in exit_itimers).
-                        *
-                        * The exec case is a bit more invloved but easy
-                        * to code.  If the process is in our thread
-                        * group (and it must be or we would not allow
-                        * it here) and is doing an exec, it will cause
-                        * us to be killed.  In this case it will wait
-                        * for us to die which means we can finish this
-                        * linkage with our last gasp. I.e. no code :)
-                        */
-                       spin_lock_irqsave(&process->sighand->siglock, flags);
-                       if (!(process->flags & PF_EXITING)) {
-                               new_timer->it_process = process;
-                               list_add(&new_timer->list,
-                                        &process->signal->posix_timers);
-                               spin_unlock_irqrestore(&process->sighand->siglock, flags);
-                               if (new_timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID))
-                                       get_task_struct(process);
-                       } else {
-                               spin_unlock_irqrestore(&process->sighand->siglock, flags);
-                               process = NULL;
-                       }
-               }
-               read_unlock(&tasklist_lock);
-               if (!process) {
+               rcu_read_lock();
+               new_timer->it_pid = get_pid(good_sigevent(&event));
+               rcu_read_unlock();
+               if (!new_timer->it_pid) {
                        error = -EINVAL;
                        goto out;
                }
        } else {
-               new_timer->it_sigev_notify = SIGEV_SIGNAL;
-               new_timer->it_sigev_signo = SIGALRM;
-               new_timer->it_sigev_value.sival_int = new_timer->it_id;
-               process = current->group_leader;
-               spin_lock_irqsave(&process->sighand->siglock, flags);
-               new_timer->it_process = process;
-               list_add(&new_timer->list, &process->signal->posix_timers);
-               spin_unlock_irqrestore(&process->sighand->siglock, flags);
+               event.sigev_notify = SIGEV_SIGNAL;
+               event.sigev_signo = SIGALRM;
+               event.sigev_value.sival_int = new_timer->it_id;
+               new_timer->it_pid = get_pid(task_tgid(current));
+       }
+
+       new_timer->it_sigev_notify     = event.sigev_notify;
+       new_timer->sigq->info.si_signo = event.sigev_signo;
+       new_timer->sigq->info.si_value = event.sigev_value;
+       new_timer->sigq->info.si_tid   = new_timer->it_id;
+       new_timer->sigq->info.si_code  = SI_TIMER;
+
+       if (copy_to_user(created_timer_id,
+                        &new_timer_id, sizeof (new_timer_id))) {
+               error = -EFAULT;
+               goto out;
        }
 
-       /*
+       error = kc->timer_create(new_timer);
+       if (error)
+               goto out;
+
+       spin_lock_irq(&current->sighand->siglock);
+       new_timer->it_signal = current->signal;
+       list_add(&new_timer->list, &current->signal->posix_timers);
+       spin_unlock_irq(&current->sighand->siglock);
+
+       return 0;
+       /*
         * In the case of the timer belonging to another task, after
         * the task is unlocked, the timer is owned by the other task
         * and may cease to exist at any time.  Don't use or modify
         * new_timer after the unlock call.
         */
-
 out:
-       if (error)
-               release_posix_timer(new_timer, it_id_set);
-
+       release_posix_timer(new_timer, it_id_set);
        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
  * removal is done under the idr spinlock so we use that here to bridge
  * the find to the timer lock.  To avoid a dead lock, the timer id MUST
  * be release with out holding the timer lock.
  */
-static struct k_itimer * lock_timer(timer_t timer_id, unsigned long *flags)
+static struct k_itimer *__lock_timer(timer_t timer_id, unsigned long *flags)
 {
        struct k_itimer *timr;
-       /*
-        * Watch out here.  We do a irqsave on the idr_lock and pass the
-        * flags part over to the timer lock.  Must not let interrupts in
-        * while we are moving the lock.
-        */
 
-       spin_lock_irqsave(&idr_lock, *flags);
-       timr = (struct k_itimer *) idr_find(&posix_timers_id, (int) timer_id);
+       rcu_read_lock();
+       timr = idr_find(&posix_timers_id, (int)timer_id);
        if (timr) {
-               spin_lock(&timr->it_lock);
-               spin_unlock(&idr_lock);
-
-               if ((timr->it_id != timer_id) || !(timr->it_process) ||
-                               timr->it_process->tgid != current->tgid) {
-                       unlock_timer(timr, *flags);
-                       timr = NULL;
+               spin_lock_irqsave(&timr->it_lock, *flags);
+               if (timr->it_signal == current->signal) {
+                       rcu_read_unlock();
+                       return timr;
                }
-       } else
-               spin_unlock_irqrestore(&idr_lock, *flags);
+               spin_unlock_irqrestore(&timr->it_lock, *flags);
+       }
+       rcu_read_unlock();
 
-       return timr;
+       return NULL;
 }
 
 /*
@@ -776,62 +673,72 @@ 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 += (unsigned int) hrtimer_forward(timer, now, iv);
+
+       remaining = ktime_sub(hrtimer_get_expires(timer), 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. */
-asmlinkage long
-sys_timer_gettime(timer_t timer_id, struct itimerspec __user *setting)
+SYSCALL_DEFINE2(timer_gettime, timer_t, timer_id,
+               struct itimerspec __user *, setting)
 {
-       struct k_itimer *timr;
        struct itimerspec cur_setting;
+       struct k_itimer *timr;
+       struct k_clock *kc;
        unsigned long flags;
+       int ret = 0;
 
        timr = lock_timer(timer_id, &flags);
        if (!timr)
                return -EINVAL;
 
-       CLOCK_DISPATCH(timr->it_clock, timer_get, (timr, &cur_setting));
+       kc = clockid_to_kclock(timr->it_clock);
+       if (WARN_ON_ONCE(!kc || !kc->timer_get))
+               ret = -EINVAL;
+       else
+               kc->timer_get(timr, &cur_setting);
 
        unlock_timer(timr, flags);
 
-       if (copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
+       if (!ret && copy_to_user(setting, &cur_setting, sizeof (cur_setting)))
                return -EFAULT;
 
-       return 0;
+       return ret;
 }
+
 /*
  * 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
@@ -841,13 +748,11 @@ 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)
+SYSCALL_DEFINE1(timer_getoverrun, timer_t, timer_id)
 {
        struct k_itimer *timr;
        int overrun;
-       long flags;
+       unsigned long flags;
 
        timr = lock_timer(timer_id, &flags);
        if (!timr)
@@ -858,167 +763,69 @@ 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 and normalize
-                */
-               set_normalized_timespec(&oc, oc.tv_sec - now.tv_sec,
-                                       oc.tv_nsec - now.tv_nsec);
-       }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.
         */
-       if (try_to_del_timer_sync(&timr->it.real.timer) < 0) {
-#ifdef CONFIG_SMP
-               /*
-                * 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;
-#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_MODE_ABS : HRTIMER_MODE_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);
+       hrtimer_set_expires(timer, 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_MODE_REL) {
+                       hrtimer_add_expires(timer, timer->base->get_time());
+               }
+               return 0;
        }
+
+       hrtimer_start_expires(timer, mode);
        return 0;
 }
 
 /* Set a POSIX.1b interval timer */
-asmlinkage long
-sys_timer_settime(timer_t timer_id, int flags,
-                 const struct itimerspec __user *new_setting,
-                 struct itimerspec __user *old_setting)
+SYSCALL_DEFINE4(timer_settime, timer_t, timer_id, int, flags,
+               const struct itimerspec __user *, new_setting,
+               struct itimerspec __user *, old_setting)
 {
        struct k_itimer *timr;
        struct itimerspec new_spec, old_spec;
        int error = 0;
-       long flag;
+       unsigned long flag;
        struct itimerspec *rtn = old_setting ? &old_spec : NULL;
+       struct k_clock *kc;
 
        if (!new_setting)
                return -EINVAL;
@@ -1026,16 +833,19 @@ 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);
        if (!timr)
                return -EINVAL;
 
-       error = CLOCK_DISPATCH(timr->it_clock, timer_set,
-                              (timr, flags, &new_spec, rtn));
+       kc = clockid_to_kclock(timr->it_clock);
+       if (WARN_ON_ONCE(!kc || !kc->timer_set))
+               error = -EINVAL;
+       else
+               error = kc->timer_set(timr, flags, &new_spec, rtn);
 
        unlock_timer(timr, flag);
        if (error == TIMER_RETRY) {
@@ -1043,66 +853,47 @@ 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;
 }
 
-static inline int common_timer_del(struct k_itimer *timer)
+static int common_timer_del(struct k_itimer *timer)
 {
-       timer->it.real.incr = 0;
+       timer->it.real.interval.tv64 = 0;
 
-       if (try_to_del_timer_sync(&timer->it.real.timer) < 0) {
-#ifdef CONFIG_SMP
-               /*
-                * 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->it.real.timer) < 0)
                return TIMER_RETRY;
-#endif
-       }
-
-       remove_from_abslist(timer);
-
        return 0;
 }
 
 static inline int timer_delete_hook(struct k_itimer *timer)
 {
-       return CLOCK_DISPATCH(timer->it_clock, timer_del, (timer));
+       struct k_clock *kc = clockid_to_kclock(timer->it_clock);
+
+       if (WARN_ON_ONCE(!kc || !kc->timer_del))
+               return -EINVAL;
+       return kc->timer_del(timer);
 }
 
 /* Delete a POSIX.1b interval timer. */
-asmlinkage long
-sys_timer_delete(timer_t timer_id)
+SYSCALL_DEFINE1(timer_delete, timer_t, timer_id)
 {
        struct k_itimer *timer;
-       long flags;
+       unsigned 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);
@@ -1110,54 +901,40 @@ retry_delete:
         * This keeps any tasks waiting on the spin lock from thinking
         * they got something (see the lock code above).
         */
-       if (timer->it_process) {
-               if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID))
-                       put_task_struct(timer->it_process);
-               timer->it_process = NULL;
-       }
+       timer->it_signal = NULL;
+
        unlock_timer(timer, flags);
        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
         * they got something (see the lock code above).
         */
-       if (timer->it_process) {
-               if (timer->it_sigev_notify == (SIGEV_SIGNAL|SIGEV_THREAD_ID))
-                       put_task_struct(timer->it_process);
-               timer->it_process = NULL;
-       }
+       timer->it_signal = NULL;
+
        unlock_timer(timer, flags);
        release_posix_timer(timer, IT_ID_SET);
 }
 
 /*
- * 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)
@@ -1170,386 +947,123 @@ void exit_itimers(struct signal_struct *sig)
        }
 }
 
-/*
- * 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)
+SYSCALL_DEFINE2(clock_settime, const clockid_t, which_clock,
+               const struct timespec __user *, tp)
 {
-       u64 jiff;
-       unsigned int seq;
+       struct k_clock *kc = clockid_to_kclock(which_clock);
+       struct timespec new_tp;
 
-       do {
-               seq = read_seqbegin(&xtime_lock);
-               getnstimeofday(tp);
-               *mo = wall_to_monotonic;
-               jiff = jiffies_64;
+       if (!kc || !kc->clock_set)
+               return -EINVAL;
 
-       } while(read_seqretry(&xtime_lock, seq));
+       if (copy_from_user(&new_tp, tp, sizeof (*tp)))
+               return -EFAULT;
 
-       return jiff;
+       return kc->clock_set(which_clock, &new_tp);
 }
 
-static int do_posix_clock_monotonic_get(clockid_t clock, struct timespec *tp)
+SYSCALL_DEFINE2(clock_gettime, const clockid_t, which_clock,
+               struct timespec __user *,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;
-}
+       struct k_clock *kc = clockid_to_kclock(which_clock);
+       struct timespec kernel_tp;
+       int error;
 
-int do_posix_clock_monotonic_gettime(struct timespec *tp)
-{
-       return do_posix_clock_monotonic_get(CLOCK_MONOTONIC, tp);
-}
+       if (!kc)
+               return -EINVAL;
 
-int do_posix_clock_nosettime(clockid_t clockid, struct timespec *tp)
-{
-       return -EINVAL;
-}
-EXPORT_SYMBOL_GPL(do_posix_clock_nosettime);
+       error = kc->clock_get(which_clock, &kernel_tp);
 
-int do_posix_clock_notimer_create(struct k_itimer *timer)
-{
-       return -EINVAL;
-}
-EXPORT_SYMBOL_GPL(do_posix_clock_notimer_create);
+       if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
+               error = -EFAULT;
 
-int do_posix_clock_nonanosleep(clockid_t clock, int flags, struct timespec *t)
-{
-#ifndef ENOTSUP
-       return -EOPNOTSUPP;     /* aka ENOTSUP in userland for POSIX */
-#else  /*  parisc does define it separately.  */
-       return -ENOTSUP;
-#endif
+       return error;
 }
-EXPORT_SYMBOL_GPL(do_posix_clock_nonanosleep);
 
-asmlinkage long
-sys_clock_settime(clockid_t which_clock, const struct timespec __user *tp)
+SYSCALL_DEFINE2(clock_adjtime, const clockid_t, which_clock,
+               struct timex __user *, utx)
 {
-       struct timespec new_tp;
+       struct k_clock *kc = clockid_to_kclock(which_clock);
+       struct timex ktx;
+       int err;
 
-       if (invalid_clockid(which_clock))
+       if (!kc)
                return -EINVAL;
-       if (copy_from_user(&new_tp, tp, sizeof (*tp)))
-               return -EFAULT;
+       if (!kc->clock_adj)
+               return -EOPNOTSUPP;
 
-       return CLOCK_DISPATCH(which_clock, clock_set, (which_clock, &new_tp));
-}
-
-asmlinkage long
-sys_clock_gettime(clockid_t which_clock, struct timespec __user *tp)
-{
-       struct timespec kernel_tp;
-       int error;
+       if (copy_from_user(&ktx, utx, sizeof(ktx)))
+               return -EFAULT;
 
-       if (invalid_clockid(which_clock))
-               return -EINVAL;
-       error = CLOCK_DISPATCH(which_clock, clock_get,
-                              (which_clock, &kernel_tp));
-       if (!error && copy_to_user(tp, &kernel_tp, sizeof (kernel_tp)))
-               error = -EFAULT;
+       err = kc->clock_adj(which_clock, &ktx);
 
-       return error;
+       if (!err && copy_to_user(utx, &ktx, sizeof(ktx)))
+               return -EFAULT;
 
+       return err;
 }
 
-asmlinkage long
-sys_clock_getres(clockid_t which_clock, struct timespec __user *tp)
+SYSCALL_DEFINE2(clock_getres, const clockid_t, which_clock,
+               struct timespec __user *, tp)
 {
+       struct k_clock *kc = clockid_to_kclock(which_clock);
        struct timespec rtn_tp;
        int error;
 
-       if (invalid_clockid(which_clock))
+       if (!kc)
                return -EINVAL;
 
-       error = CLOCK_DISPATCH(which_clock, clock_getres,
-                              (which_clock, &rtn_tp));
+       error = kc->clock_getres(which_clock, &rtn_tp);
 
-       if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp))) {
+       if (!error && tp && copy_to_user(tp, &rtn_tp, sizeof (rtn_tp)))
                error = -EFAULT;
-       }
 
        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_MODE_ABS : HRTIMER_MODE_REL,
+                                which_clock);
 }
 
-long clock_nanosleep_restart(struct restart_block *restart_block);
-
-asmlinkage long
-sys_clock_nanosleep(clockid_t which_clock, int flags,
-                   const struct timespec __user *rqtp,
-                   struct timespec __user *rmtp)
+SYSCALL_DEFINE4(clock_nanosleep, const clockid_t, which_clock, int, flags,
+               const struct timespec __user *, rqtp,
+               struct timespec __user *, rmtp)
 {
+       struct k_clock *kc = clockid_to_kclock(which_clock);
        struct timespec t;
-       struct restart_block *restart_block =
-           &(current_thread_info()->restart_block);
-       int ret;
 
-       if (invalid_clockid(which_clock))
+       if (!kc)
                return -EINVAL;
+       if (!kc->nsleep)
+               return -ENANOSLEEP_NOTSUP;
 
        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;
+       return kc->nsleep(which_clock, flags, &t, rmtp);
 }
 
-
-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.
+ * This will restart clock_nanosleep. This is required only by
+ * compat_clock_nanosleep_restart for now.
  */
-long
-clock_nanosleep_restart(struct restart_block *restart_block)
+long clock_nanosleep_restart(struct restart_block *restart_block)
 {
-       struct timespec t;
-       int ret = common_nsleep(restart_block->arg0, 0, &t);
+       clockid_t which_clock = restart_block->nanosleep.clockid;
+       struct k_clock *kc = clockid_to_kclock(which_clock);
 
-       if ((ret == -ERESTART_RESTARTBLOCK) && restart_block->arg1 &&
-           copy_to_user((struct timespec __user *)(restart_block->arg1), &t,
-                        sizeof (t)))
-               return -EFAULT;
-       return ret;
+       if (WARN_ON_ONCE(!kc || !kc->nsleep_restart))
+               return -EINVAL;
+
+       return kc->nsleep_restart(restart_block);
 }