2 * Implement CPU time clocks for the POSIX clock interface.
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <asm/uaccess.h>
8 #include <linux/errno.h>
10 static int check_clock(const clockid_t which_clock)
13 struct task_struct *p;
14 const pid_t pid = CPUCLOCK_PID(which_clock);
16 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
22 read_lock(&tasklist_lock);
23 p = find_task_by_pid(pid);
24 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
25 same_thread_group(p, current) : thread_group_leader(p))) {
28 read_unlock(&tasklist_lock);
33 static inline union cpu_time_count
34 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
36 union cpu_time_count ret;
37 ret.sched = 0; /* high half always zero when .cpu used */
38 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
39 ret.sched = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
41 ret.cpu = timespec_to_cputime(tp);
46 static void sample_to_timespec(const clockid_t which_clock,
47 union cpu_time_count cpu,
50 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
51 tp->tv_sec = div_long_long_rem(cpu.sched,
52 NSEC_PER_SEC, &tp->tv_nsec);
54 cputime_to_timespec(cpu.cpu, tp);
58 static inline int cpu_time_before(const clockid_t which_clock,
59 union cpu_time_count now,
60 union cpu_time_count then)
62 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
63 return now.sched < then.sched;
65 return cputime_lt(now.cpu, then.cpu);
68 static inline void cpu_time_add(const clockid_t which_clock,
69 union cpu_time_count *acc,
70 union cpu_time_count val)
72 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
73 acc->sched += val.sched;
75 acc->cpu = cputime_add(acc->cpu, val.cpu);
78 static inline union cpu_time_count cpu_time_sub(const clockid_t which_clock,
79 union cpu_time_count a,
80 union cpu_time_count b)
82 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
85 a.cpu = cputime_sub(a.cpu, b.cpu);
91 * Divide and limit the result to res >= 1
93 * This is necessary to prevent signal delivery starvation, when the result of
94 * the division would be rounded down to 0.
96 static inline cputime_t cputime_div_non_zero(cputime_t time, unsigned long div)
98 cputime_t res = cputime_div(time, div);
100 return max_t(cputime_t, res, 1);
104 * Update expiry time from increment, and increase overrun count,
105 * given the current clock sample.
107 static void bump_cpu_timer(struct k_itimer *timer,
108 union cpu_time_count now)
112 if (timer->it.cpu.incr.sched == 0)
115 if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
116 unsigned long long delta, incr;
118 if (now.sched < timer->it.cpu.expires.sched)
120 incr = timer->it.cpu.incr.sched;
121 delta = now.sched + incr - timer->it.cpu.expires.sched;
122 /* Don't use (incr*2 < delta), incr*2 might overflow. */
123 for (i = 0; incr < delta - incr; i++)
125 for (; i >= 0; incr >>= 1, i--) {
128 timer->it.cpu.expires.sched += incr;
129 timer->it_overrun += 1 << i;
133 cputime_t delta, incr;
135 if (cputime_lt(now.cpu, timer->it.cpu.expires.cpu))
137 incr = timer->it.cpu.incr.cpu;
138 delta = cputime_sub(cputime_add(now.cpu, incr),
139 timer->it.cpu.expires.cpu);
140 /* Don't use (incr*2 < delta), incr*2 might overflow. */
141 for (i = 0; cputime_lt(incr, cputime_sub(delta, incr)); i++)
142 incr = cputime_add(incr, incr);
143 for (; i >= 0; incr = cputime_halve(incr), i--) {
144 if (cputime_lt(delta, incr))
146 timer->it.cpu.expires.cpu =
147 cputime_add(timer->it.cpu.expires.cpu, incr);
148 timer->it_overrun += 1 << i;
149 delta = cputime_sub(delta, incr);
154 static inline cputime_t prof_ticks(struct task_struct *p)
156 return cputime_add(p->utime, p->stime);
158 static inline cputime_t virt_ticks(struct task_struct *p)
162 static inline unsigned long long sched_ns(struct task_struct *p)
164 return task_sched_runtime(p);
167 int posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
169 int error = check_clock(which_clock);
172 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
173 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
175 * If sched_clock is using a cycle counter, we
176 * don't have any idea of its true resolution
177 * exported, but it is much more than 1s/HZ.
185 int posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
188 * You can never reset a CPU clock, but we check for other errors
189 * in the call before failing with EPERM.
191 int error = check_clock(which_clock);
200 * Sample a per-thread clock for the given task.
202 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
203 union cpu_time_count *cpu)
205 switch (CPUCLOCK_WHICH(which_clock)) {
209 cpu->cpu = prof_ticks(p);
212 cpu->cpu = virt_ticks(p);
215 cpu->sched = sched_ns(p);
222 * Sample a process (thread group) clock for the given group_leader task.
223 * Must be called with tasklist_lock held for reading.
224 * Must be called with tasklist_lock held for reading, and p->sighand->siglock.
226 static int cpu_clock_sample_group_locked(unsigned int clock_idx,
227 struct task_struct *p,
228 union cpu_time_count *cpu)
230 struct task_struct *t = p;
235 cpu->cpu = cputime_add(p->signal->utime, p->signal->stime);
237 cpu->cpu = cputime_add(cpu->cpu, prof_ticks(t));
242 cpu->cpu = p->signal->utime;
244 cpu->cpu = cputime_add(cpu->cpu, virt_ticks(t));
249 cpu->sched = p->signal->sum_sched_runtime;
250 /* Add in each other live thread. */
251 while ((t = next_thread(t)) != p) {
252 cpu->sched += t->se.sum_exec_runtime;
254 cpu->sched += sched_ns(p);
261 * Sample a process (thread group) clock for the given group_leader task.
262 * Must be called with tasklist_lock held for reading.
264 static int cpu_clock_sample_group(const clockid_t which_clock,
265 struct task_struct *p,
266 union cpu_time_count *cpu)
270 spin_lock_irqsave(&p->sighand->siglock, flags);
271 ret = cpu_clock_sample_group_locked(CPUCLOCK_WHICH(which_clock), p,
273 spin_unlock_irqrestore(&p->sighand->siglock, flags);
278 int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
280 const pid_t pid = CPUCLOCK_PID(which_clock);
282 union cpu_time_count rtn;
286 * Special case constant value for our own clocks.
287 * We don't have to do any lookup to find ourselves.
289 if (CPUCLOCK_PERTHREAD(which_clock)) {
291 * Sampling just ourselves we can do with no locking.
293 error = cpu_clock_sample(which_clock,
296 read_lock(&tasklist_lock);
297 error = cpu_clock_sample_group(which_clock,
299 read_unlock(&tasklist_lock);
303 * Find the given PID, and validate that the caller
304 * should be able to see it.
306 struct task_struct *p;
308 p = find_task_by_pid(pid);
310 if (CPUCLOCK_PERTHREAD(which_clock)) {
311 if (same_thread_group(p, current)) {
312 error = cpu_clock_sample(which_clock,
316 read_lock(&tasklist_lock);
317 if (thread_group_leader(p) && p->signal) {
319 cpu_clock_sample_group(which_clock,
322 read_unlock(&tasklist_lock);
330 sample_to_timespec(which_clock, rtn, tp);
336 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
337 * This is called from sys_timer_create with the new timer already locked.
339 int posix_cpu_timer_create(struct k_itimer *new_timer)
342 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
343 struct task_struct *p;
345 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
348 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
349 new_timer->it.cpu.incr.sched = 0;
350 new_timer->it.cpu.expires.sched = 0;
352 read_lock(&tasklist_lock);
353 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
357 p = find_task_by_pid(pid);
358 if (p && !same_thread_group(p, current))
363 p = current->group_leader;
365 p = find_task_by_pid(pid);
366 if (p && !thread_group_leader(p))
370 new_timer->it.cpu.task = p;
376 read_unlock(&tasklist_lock);
382 * Clean up a CPU-clock timer that is about to be destroyed.
383 * This is called from timer deletion with the timer already locked.
384 * If we return TIMER_RETRY, it's necessary to release the timer's lock
385 * and try again. (This happens when the timer is in the middle of firing.)
387 int posix_cpu_timer_del(struct k_itimer *timer)
389 struct task_struct *p = timer->it.cpu.task;
392 if (likely(p != NULL)) {
393 read_lock(&tasklist_lock);
394 if (unlikely(p->signal == NULL)) {
396 * We raced with the reaping of the task.
397 * The deletion should have cleared us off the list.
399 BUG_ON(!list_empty(&timer->it.cpu.entry));
401 spin_lock(&p->sighand->siglock);
402 if (timer->it.cpu.firing)
405 list_del(&timer->it.cpu.entry);
406 spin_unlock(&p->sighand->siglock);
408 read_unlock(&tasklist_lock);
418 * Clean out CPU timers still ticking when a thread exited. The task
419 * pointer is cleared, and the expiry time is replaced with the residual
420 * time for later timer_gettime calls to return.
421 * This must be called with the siglock held.
423 static void cleanup_timers(struct list_head *head,
424 cputime_t utime, cputime_t stime,
425 unsigned long long sum_exec_runtime)
427 struct cpu_timer_list *timer, *next;
428 cputime_t ptime = cputime_add(utime, stime);
430 list_for_each_entry_safe(timer, next, head, entry) {
431 list_del_init(&timer->entry);
432 if (cputime_lt(timer->expires.cpu, ptime)) {
433 timer->expires.cpu = cputime_zero;
435 timer->expires.cpu = cputime_sub(timer->expires.cpu,
441 list_for_each_entry_safe(timer, next, head, entry) {
442 list_del_init(&timer->entry);
443 if (cputime_lt(timer->expires.cpu, utime)) {
444 timer->expires.cpu = cputime_zero;
446 timer->expires.cpu = cputime_sub(timer->expires.cpu,
452 list_for_each_entry_safe(timer, next, head, entry) {
453 list_del_init(&timer->entry);
454 if (timer->expires.sched < sum_exec_runtime) {
455 timer->expires.sched = 0;
457 timer->expires.sched -= sum_exec_runtime;
463 * These are both called with the siglock held, when the current thread
464 * is being reaped. When the final (leader) thread in the group is reaped,
465 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
467 void posix_cpu_timers_exit(struct task_struct *tsk)
469 cleanup_timers(tsk->cpu_timers,
470 tsk->utime, tsk->stime, tsk->se.sum_exec_runtime);
473 void posix_cpu_timers_exit_group(struct task_struct *tsk)
475 cleanup_timers(tsk->signal->cpu_timers,
476 cputime_add(tsk->utime, tsk->signal->utime),
477 cputime_add(tsk->stime, tsk->signal->stime),
478 tsk->se.sum_exec_runtime + tsk->signal->sum_sched_runtime);
483 * Set the expiry times of all the threads in the process so one of them
484 * will go off before the process cumulative expiry total is reached.
486 static void process_timer_rebalance(struct task_struct *p,
487 unsigned int clock_idx,
488 union cpu_time_count expires,
489 union cpu_time_count val)
491 cputime_t ticks, left;
492 unsigned long long ns, nsleft;
493 struct task_struct *t = p;
494 unsigned int nthreads = atomic_read(&p->signal->live);
504 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
507 if (likely(!(t->flags & PF_EXITING))) {
508 ticks = cputime_add(prof_ticks(t), left);
509 if (cputime_eq(t->it_prof_expires,
511 cputime_gt(t->it_prof_expires, ticks)) {
512 t->it_prof_expires = ticks;
519 left = cputime_div_non_zero(cputime_sub(expires.cpu, val.cpu),
522 if (likely(!(t->flags & PF_EXITING))) {
523 ticks = cputime_add(virt_ticks(t), left);
524 if (cputime_eq(t->it_virt_expires,
526 cputime_gt(t->it_virt_expires, ticks)) {
527 t->it_virt_expires = ticks;
534 nsleft = expires.sched - val.sched;
535 do_div(nsleft, nthreads);
536 nsleft = max_t(unsigned long long, nsleft, 1);
538 if (likely(!(t->flags & PF_EXITING))) {
539 ns = t->se.sum_exec_runtime + nsleft;
540 if (t->it_sched_expires == 0 ||
541 t->it_sched_expires > ns) {
542 t->it_sched_expires = ns;
551 static void clear_dead_task(struct k_itimer *timer, union cpu_time_count now)
554 * That's all for this thread or process.
555 * We leave our residual in expires to be reported.
557 put_task_struct(timer->it.cpu.task);
558 timer->it.cpu.task = NULL;
559 timer->it.cpu.expires = cpu_time_sub(timer->it_clock,
560 timer->it.cpu.expires,
565 * Insert the timer on the appropriate list before any timers that
566 * expire later. This must be called with the tasklist_lock held
567 * for reading, and interrupts disabled.
569 static void arm_timer(struct k_itimer *timer, union cpu_time_count now)
571 struct task_struct *p = timer->it.cpu.task;
572 struct list_head *head, *listpos;
573 struct cpu_timer_list *const nt = &timer->it.cpu;
574 struct cpu_timer_list *next;
577 head = (CPUCLOCK_PERTHREAD(timer->it_clock) ?
578 p->cpu_timers : p->signal->cpu_timers);
579 head += CPUCLOCK_WHICH(timer->it_clock);
581 BUG_ON(!irqs_disabled());
582 spin_lock(&p->sighand->siglock);
585 if (CPUCLOCK_WHICH(timer->it_clock) == CPUCLOCK_SCHED) {
586 list_for_each_entry(next, head, entry) {
587 if (next->expires.sched > nt->expires.sched)
589 listpos = &next->entry;
592 list_for_each_entry(next, head, entry) {
593 if (cputime_gt(next->expires.cpu, nt->expires.cpu))
595 listpos = &next->entry;
598 list_add(&nt->entry, listpos);
600 if (listpos == head) {
602 * We are the new earliest-expiring timer.
603 * If we are a thread timer, there can always
604 * be a process timer telling us to stop earlier.
607 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
608 switch (CPUCLOCK_WHICH(timer->it_clock)) {
612 if (cputime_eq(p->it_prof_expires,
614 cputime_gt(p->it_prof_expires,
616 p->it_prof_expires = nt->expires.cpu;
619 if (cputime_eq(p->it_virt_expires,
621 cputime_gt(p->it_virt_expires,
623 p->it_virt_expires = nt->expires.cpu;
626 if (p->it_sched_expires == 0 ||
627 p->it_sched_expires > nt->expires.sched)
628 p->it_sched_expires = nt->expires.sched;
633 * For a process timer, we must balance
634 * all the live threads' expirations.
636 switch (CPUCLOCK_WHICH(timer->it_clock)) {
640 if (!cputime_eq(p->signal->it_virt_expires,
642 cputime_lt(p->signal->it_virt_expires,
643 timer->it.cpu.expires.cpu))
647 if (!cputime_eq(p->signal->it_prof_expires,
649 cputime_lt(p->signal->it_prof_expires,
650 timer->it.cpu.expires.cpu))
652 i = p->signal->rlim[RLIMIT_CPU].rlim_cur;
653 if (i != RLIM_INFINITY &&
654 i <= cputime_to_secs(timer->it.cpu.expires.cpu))
659 process_timer_rebalance(
661 CPUCLOCK_WHICH(timer->it_clock),
662 timer->it.cpu.expires, now);
668 spin_unlock(&p->sighand->siglock);
672 * The timer is locked, fire it and arrange for its reload.
674 static void cpu_timer_fire(struct k_itimer *timer)
676 if (unlikely(timer->sigq == NULL)) {
678 * This a special case for clock_nanosleep,
679 * not a normal timer from sys_timer_create.
681 wake_up_process(timer->it_process);
682 timer->it.cpu.expires.sched = 0;
683 } else if (timer->it.cpu.incr.sched == 0) {
685 * One-shot timer. Clear it as soon as it's fired.
687 posix_timer_event(timer, 0);
688 timer->it.cpu.expires.sched = 0;
689 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
691 * The signal did not get queued because the signal
692 * was ignored, so we won't get any callback to
693 * reload the timer. But we need to keep it
694 * ticking in case the signal is deliverable next time.
696 posix_cpu_timer_schedule(timer);
701 * Guts of sys_timer_settime for CPU timers.
702 * This is called with the timer locked and interrupts disabled.
703 * If we return TIMER_RETRY, it's necessary to release the timer's lock
704 * and try again. (This happens when the timer is in the middle of firing.)
706 int posix_cpu_timer_set(struct k_itimer *timer, int flags,
707 struct itimerspec *new, struct itimerspec *old)
709 struct task_struct *p = timer->it.cpu.task;
710 union cpu_time_count old_expires, new_expires, val;
713 if (unlikely(p == NULL)) {
715 * Timer refers to a dead task's clock.
720 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
722 read_lock(&tasklist_lock);
724 * We need the tasklist_lock to protect against reaping that
725 * clears p->signal. If p has just been reaped, we can no
726 * longer get any information about it at all.
728 if (unlikely(p->signal == NULL)) {
729 read_unlock(&tasklist_lock);
731 timer->it.cpu.task = NULL;
736 * Disarm any old timer after extracting its expiry time.
738 BUG_ON(!irqs_disabled());
741 spin_lock(&p->sighand->siglock);
742 old_expires = timer->it.cpu.expires;
743 if (unlikely(timer->it.cpu.firing)) {
744 timer->it.cpu.firing = -1;
747 list_del_init(&timer->it.cpu.entry);
748 spin_unlock(&p->sighand->siglock);
751 * We need to sample the current value to convert the new
752 * value from to relative and absolute, and to convert the
753 * old value from absolute to relative. To set a process
754 * timer, we need a sample to balance the thread expiry
755 * times (in arm_timer). With an absolute time, we must
756 * check if it's already passed. In short, we need a sample.
758 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
759 cpu_clock_sample(timer->it_clock, p, &val);
761 cpu_clock_sample_group(timer->it_clock, p, &val);
765 if (old_expires.sched == 0) {
766 old->it_value.tv_sec = 0;
767 old->it_value.tv_nsec = 0;
770 * Update the timer in case it has
771 * overrun already. If it has,
772 * we'll report it as having overrun
773 * and with the next reloaded timer
774 * already ticking, though we are
775 * swallowing that pending
776 * notification here to install the
779 bump_cpu_timer(timer, val);
780 if (cpu_time_before(timer->it_clock, val,
781 timer->it.cpu.expires)) {
782 old_expires = cpu_time_sub(
784 timer->it.cpu.expires, val);
785 sample_to_timespec(timer->it_clock,
789 old->it_value.tv_nsec = 1;
790 old->it_value.tv_sec = 0;
797 * We are colliding with the timer actually firing.
798 * Punt after filling in the timer's old value, and
799 * disable this firing since we are already reporting
800 * it as an overrun (thanks to bump_cpu_timer above).
802 read_unlock(&tasklist_lock);
806 if (new_expires.sched != 0 && !(flags & TIMER_ABSTIME)) {
807 cpu_time_add(timer->it_clock, &new_expires, val);
811 * Install the new expiry time (or zero).
812 * For a timer with no notification action, we don't actually
813 * arm the timer (we'll just fake it for timer_gettime).
815 timer->it.cpu.expires = new_expires;
816 if (new_expires.sched != 0 &&
817 (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
818 cpu_time_before(timer->it_clock, val, new_expires)) {
819 arm_timer(timer, val);
822 read_unlock(&tasklist_lock);
825 * Install the new reload setting, and
826 * set up the signal and overrun bookkeeping.
828 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
832 * This acts as a modification timestamp for the timer,
833 * so any automatic reload attempt will punt on seeing
834 * that we have reset the timer manually.
836 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
838 timer->it_overrun_last = 0;
839 timer->it_overrun = -1;
841 if (new_expires.sched != 0 &&
842 (timer->it_sigev_notify & ~SIGEV_THREAD_ID) != SIGEV_NONE &&
843 !cpu_time_before(timer->it_clock, val, new_expires)) {
845 * The designated time already passed, so we notify
846 * immediately, even if the thread never runs to
847 * accumulate more time on this clock.
849 cpu_timer_fire(timer);
855 sample_to_timespec(timer->it_clock,
856 timer->it.cpu.incr, &old->it_interval);
861 void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
863 union cpu_time_count now;
864 struct task_struct *p = timer->it.cpu.task;
868 * Easy part: convert the reload time.
870 sample_to_timespec(timer->it_clock,
871 timer->it.cpu.incr, &itp->it_interval);
873 if (timer->it.cpu.expires.sched == 0) { /* Timer not armed at all. */
874 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
878 if (unlikely(p == NULL)) {
880 * This task already died and the timer will never fire.
881 * In this case, expires is actually the dead value.
884 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
890 * Sample the clock to take the difference with the expiry time.
892 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
893 cpu_clock_sample(timer->it_clock, p, &now);
894 clear_dead = p->exit_state;
896 read_lock(&tasklist_lock);
897 if (unlikely(p->signal == NULL)) {
899 * The process has been reaped.
900 * We can't even collect a sample any more.
901 * Call the timer disarmed, nothing else to do.
904 timer->it.cpu.task = NULL;
905 timer->it.cpu.expires.sched = 0;
906 read_unlock(&tasklist_lock);
909 cpu_clock_sample_group(timer->it_clock, p, &now);
910 clear_dead = (unlikely(p->exit_state) &&
911 thread_group_empty(p));
913 read_unlock(&tasklist_lock);
916 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
917 if (timer->it.cpu.incr.sched == 0 &&
918 cpu_time_before(timer->it_clock,
919 timer->it.cpu.expires, now)) {
921 * Do-nothing timer expired and has no reload,
922 * so it's as if it was never set.
924 timer->it.cpu.expires.sched = 0;
925 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
929 * Account for any expirations and reloads that should
932 bump_cpu_timer(timer, now);
935 if (unlikely(clear_dead)) {
937 * We've noticed that the thread is dead, but
938 * not yet reaped. Take this opportunity to
941 clear_dead_task(timer, now);
945 if (cpu_time_before(timer->it_clock, now, timer->it.cpu.expires)) {
946 sample_to_timespec(timer->it_clock,
947 cpu_time_sub(timer->it_clock,
948 timer->it.cpu.expires, now),
952 * The timer should have expired already, but the firing
953 * hasn't taken place yet. Say it's just about to expire.
955 itp->it_value.tv_nsec = 1;
956 itp->it_value.tv_sec = 0;
961 * Check for any per-thread CPU timers that have fired and move them off
962 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
963 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
965 static void check_thread_timers(struct task_struct *tsk,
966 struct list_head *firing)
969 struct list_head *timers = tsk->cpu_timers;
970 struct signal_struct *const sig = tsk->signal;
973 tsk->it_prof_expires = cputime_zero;
974 while (!list_empty(timers)) {
975 struct cpu_timer_list *t = list_first_entry(timers,
976 struct cpu_timer_list,
978 if (!--maxfire || cputime_lt(prof_ticks(tsk), t->expires.cpu)) {
979 tsk->it_prof_expires = t->expires.cpu;
983 list_move_tail(&t->entry, firing);
988 tsk->it_virt_expires = cputime_zero;
989 while (!list_empty(timers)) {
990 struct cpu_timer_list *t = list_first_entry(timers,
991 struct cpu_timer_list,
993 if (!--maxfire || cputime_lt(virt_ticks(tsk), t->expires.cpu)) {
994 tsk->it_virt_expires = t->expires.cpu;
998 list_move_tail(&t->entry, firing);
1003 tsk->it_sched_expires = 0;
1004 while (!list_empty(timers)) {
1005 struct cpu_timer_list *t = list_first_entry(timers,
1006 struct cpu_timer_list,
1008 if (!--maxfire || tsk->se.sum_exec_runtime < t->expires.sched) {
1009 tsk->it_sched_expires = t->expires.sched;
1013 list_move_tail(&t->entry, firing);
1017 * Check for the special case thread timers.
1019 if (sig->rlim[RLIMIT_RTTIME].rlim_cur != RLIM_INFINITY) {
1020 unsigned long hard = sig->rlim[RLIMIT_RTTIME].rlim_max;
1021 unsigned long *soft = &sig->rlim[RLIMIT_RTTIME].rlim_cur;
1023 if (tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
1025 * At the hard limit, we just die.
1026 * No need to calculate anything else now.
1028 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1031 if (tsk->rt.timeout > DIV_ROUND_UP(*soft, USEC_PER_SEC/HZ)) {
1033 * At the soft limit, send a SIGXCPU every second.
1035 if (sig->rlim[RLIMIT_RTTIME].rlim_cur
1036 < sig->rlim[RLIMIT_RTTIME].rlim_max) {
1037 sig->rlim[RLIMIT_RTTIME].rlim_cur +=
1040 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1046 * Check for any per-thread CPU timers that have fired and move them
1047 * off the tsk->*_timers list onto the firing list. Per-thread timers
1048 * have already been taken off.
1050 static void check_process_timers(struct task_struct *tsk,
1051 struct list_head *firing)
1054 struct signal_struct *const sig = tsk->signal;
1055 cputime_t utime, stime, ptime, virt_expires, prof_expires;
1056 unsigned long long sum_sched_runtime, sched_expires;
1057 struct task_struct *t;
1058 struct list_head *timers = sig->cpu_timers;
1061 * Don't sample the current process CPU clocks if there are no timers.
1063 if (list_empty(&timers[CPUCLOCK_PROF]) &&
1064 cputime_eq(sig->it_prof_expires, cputime_zero) &&
1065 sig->rlim[RLIMIT_CPU].rlim_cur == RLIM_INFINITY &&
1066 list_empty(&timers[CPUCLOCK_VIRT]) &&
1067 cputime_eq(sig->it_virt_expires, cputime_zero) &&
1068 list_empty(&timers[CPUCLOCK_SCHED]))
1072 * Collect the current process totals.
1076 sum_sched_runtime = sig->sum_sched_runtime;
1079 utime = cputime_add(utime, t->utime);
1080 stime = cputime_add(stime, t->stime);
1081 sum_sched_runtime += t->se.sum_exec_runtime;
1084 ptime = cputime_add(utime, stime);
1087 prof_expires = cputime_zero;
1088 while (!list_empty(timers)) {
1089 struct cpu_timer_list *t = list_first_entry(timers,
1090 struct cpu_timer_list,
1092 if (!--maxfire || cputime_lt(ptime, t->expires.cpu)) {
1093 prof_expires = t->expires.cpu;
1097 list_move_tail(&t->entry, firing);
1102 virt_expires = cputime_zero;
1103 while (!list_empty(timers)) {
1104 struct cpu_timer_list *t = list_first_entry(timers,
1105 struct cpu_timer_list,
1107 if (!--maxfire || cputime_lt(utime, t->expires.cpu)) {
1108 virt_expires = t->expires.cpu;
1112 list_move_tail(&t->entry, firing);
1118 while (!list_empty(timers)) {
1119 struct cpu_timer_list *t = list_first_entry(timers,
1120 struct cpu_timer_list,
1122 if (!--maxfire || sum_sched_runtime < t->expires.sched) {
1123 sched_expires = t->expires.sched;
1127 list_move_tail(&t->entry, firing);
1131 * Check for the special case process timers.
1133 if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1134 if (cputime_ge(ptime, sig->it_prof_expires)) {
1135 /* ITIMER_PROF fires and reloads. */
1136 sig->it_prof_expires = sig->it_prof_incr;
1137 if (!cputime_eq(sig->it_prof_expires, cputime_zero)) {
1138 sig->it_prof_expires = cputime_add(
1139 sig->it_prof_expires, ptime);
1141 __group_send_sig_info(SIGPROF, SEND_SIG_PRIV, tsk);
1143 if (!cputime_eq(sig->it_prof_expires, cputime_zero) &&
1144 (cputime_eq(prof_expires, cputime_zero) ||
1145 cputime_lt(sig->it_prof_expires, prof_expires))) {
1146 prof_expires = sig->it_prof_expires;
1149 if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1150 if (cputime_ge(utime, sig->it_virt_expires)) {
1151 /* ITIMER_VIRTUAL fires and reloads. */
1152 sig->it_virt_expires = sig->it_virt_incr;
1153 if (!cputime_eq(sig->it_virt_expires, cputime_zero)) {
1154 sig->it_virt_expires = cputime_add(
1155 sig->it_virt_expires, utime);
1157 __group_send_sig_info(SIGVTALRM, SEND_SIG_PRIV, tsk);
1159 if (!cputime_eq(sig->it_virt_expires, cputime_zero) &&
1160 (cputime_eq(virt_expires, cputime_zero) ||
1161 cputime_lt(sig->it_virt_expires, virt_expires))) {
1162 virt_expires = sig->it_virt_expires;
1165 if (sig->rlim[RLIMIT_CPU].rlim_cur != RLIM_INFINITY) {
1166 unsigned long psecs = cputime_to_secs(ptime);
1168 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_max) {
1170 * At the hard limit, we just die.
1171 * No need to calculate anything else now.
1173 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
1176 if (psecs >= sig->rlim[RLIMIT_CPU].rlim_cur) {
1178 * At the soft limit, send a SIGXCPU every second.
1180 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
1181 if (sig->rlim[RLIMIT_CPU].rlim_cur
1182 < sig->rlim[RLIMIT_CPU].rlim_max) {
1183 sig->rlim[RLIMIT_CPU].rlim_cur++;
1186 x = secs_to_cputime(sig->rlim[RLIMIT_CPU].rlim_cur);
1187 if (cputime_eq(prof_expires, cputime_zero) ||
1188 cputime_lt(x, prof_expires)) {
1193 if (!cputime_eq(prof_expires, cputime_zero) ||
1194 !cputime_eq(virt_expires, cputime_zero) ||
1195 sched_expires != 0) {
1197 * Rebalance the threads' expiry times for the remaining
1198 * process CPU timers.
1201 cputime_t prof_left, virt_left, ticks;
1202 unsigned long long sched_left, sched;
1203 const unsigned int nthreads = atomic_read(&sig->live);
1208 prof_left = cputime_sub(prof_expires, utime);
1209 prof_left = cputime_sub(prof_left, stime);
1210 prof_left = cputime_div_non_zero(prof_left, nthreads);
1211 virt_left = cputime_sub(virt_expires, utime);
1212 virt_left = cputime_div_non_zero(virt_left, nthreads);
1213 if (sched_expires) {
1214 sched_left = sched_expires - sum_sched_runtime;
1215 do_div(sched_left, nthreads);
1216 sched_left = max_t(unsigned long long, sched_left, 1);
1222 if (unlikely(t->flags & PF_EXITING))
1225 ticks = cputime_add(cputime_add(t->utime, t->stime),
1227 if (!cputime_eq(prof_expires, cputime_zero) &&
1228 (cputime_eq(t->it_prof_expires, cputime_zero) ||
1229 cputime_gt(t->it_prof_expires, ticks))) {
1230 t->it_prof_expires = ticks;
1233 ticks = cputime_add(t->utime, virt_left);
1234 if (!cputime_eq(virt_expires, cputime_zero) &&
1235 (cputime_eq(t->it_virt_expires, cputime_zero) ||
1236 cputime_gt(t->it_virt_expires, ticks))) {
1237 t->it_virt_expires = ticks;
1240 sched = t->se.sum_exec_runtime + sched_left;
1241 if (sched_expires && (t->it_sched_expires == 0 ||
1242 t->it_sched_expires > sched)) {
1243 t->it_sched_expires = sched;
1245 } while ((t = next_thread(t)) != tsk);
1250 * This is called from the signal code (via do_schedule_next_timer)
1251 * when the last timer signal was delivered and we have to reload the timer.
1253 void posix_cpu_timer_schedule(struct k_itimer *timer)
1255 struct task_struct *p = timer->it.cpu.task;
1256 union cpu_time_count now;
1258 if (unlikely(p == NULL))
1260 * The task was cleaned up already, no future firings.
1265 * Fetch the current sample and update the timer's expiry time.
1267 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1268 cpu_clock_sample(timer->it_clock, p, &now);
1269 bump_cpu_timer(timer, now);
1270 if (unlikely(p->exit_state)) {
1271 clear_dead_task(timer, now);
1274 read_lock(&tasklist_lock); /* arm_timer needs it. */
1276 read_lock(&tasklist_lock);
1277 if (unlikely(p->signal == NULL)) {
1279 * The process has been reaped.
1280 * We can't even collect a sample any more.
1283 timer->it.cpu.task = p = NULL;
1284 timer->it.cpu.expires.sched = 0;
1286 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1288 * We've noticed that the thread is dead, but
1289 * not yet reaped. Take this opportunity to
1290 * drop our task ref.
1292 clear_dead_task(timer, now);
1295 cpu_clock_sample_group(timer->it_clock, p, &now);
1296 bump_cpu_timer(timer, now);
1297 /* Leave the tasklist_lock locked for the call below. */
1301 * Now re-arm for the new expiry time.
1303 arm_timer(timer, now);
1306 read_unlock(&tasklist_lock);
1309 timer->it_overrun_last = timer->it_overrun;
1310 timer->it_overrun = -1;
1311 ++timer->it_requeue_pending;
1315 * This is called from the timer interrupt handler. The irq handler has
1316 * already updated our counts. We need to check if any timers fire now.
1317 * Interrupts are disabled.
1319 void run_posix_cpu_timers(struct task_struct *tsk)
1322 struct k_itimer *timer, *next;
1324 BUG_ON(!irqs_disabled());
1326 #define UNEXPIRED(clock) \
1327 (cputime_eq(tsk->it_##clock##_expires, cputime_zero) || \
1328 cputime_lt(clock##_ticks(tsk), tsk->it_##clock##_expires))
1330 if (UNEXPIRED(prof) && UNEXPIRED(virt) &&
1331 (tsk->it_sched_expires == 0 ||
1332 tsk->se.sum_exec_runtime < tsk->it_sched_expires))
1338 * Double-check with locks held.
1340 read_lock(&tasklist_lock);
1341 if (likely(tsk->signal != NULL)) {
1342 spin_lock(&tsk->sighand->siglock);
1345 * Here we take off tsk->cpu_timers[N] and tsk->signal->cpu_timers[N]
1346 * all the timers that are firing, and put them on the firing list.
1348 check_thread_timers(tsk, &firing);
1349 check_process_timers(tsk, &firing);
1352 * We must release these locks before taking any timer's lock.
1353 * There is a potential race with timer deletion here, as the
1354 * siglock now protects our private firing list. We have set
1355 * the firing flag in each timer, so that a deletion attempt
1356 * that gets the timer lock before we do will give it up and
1357 * spin until we've taken care of that timer below.
1359 spin_unlock(&tsk->sighand->siglock);
1361 read_unlock(&tasklist_lock);
1364 * Now that all the timers on our list have the firing flag,
1365 * noone will touch their list entries but us. We'll take
1366 * each timer's lock before clearing its firing flag, so no
1367 * timer call will interfere.
1369 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1371 spin_lock(&timer->it_lock);
1372 list_del_init(&timer->it.cpu.entry);
1373 firing = timer->it.cpu.firing;
1374 timer->it.cpu.firing = 0;
1376 * The firing flag is -1 if we collided with a reset
1377 * of the timer, which already reported this
1378 * almost-firing as an overrun. So don't generate an event.
1380 if (likely(firing >= 0)) {
1381 cpu_timer_fire(timer);
1383 spin_unlock(&timer->it_lock);
1388 * Set one of the process-wide special case CPU timers.
1389 * The tasklist_lock and tsk->sighand->siglock must be held by the caller.
1390 * The oldval argument is null for the RLIMIT_CPU timer, where *newval is
1391 * absolute; non-null for ITIMER_*, where *newval is relative and we update
1392 * it to be absolute, *oldval is absolute and we update it to be relative.
1394 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1395 cputime_t *newval, cputime_t *oldval)
1397 union cpu_time_count now;
1398 struct list_head *head;
1400 BUG_ON(clock_idx == CPUCLOCK_SCHED);
1401 cpu_clock_sample_group_locked(clock_idx, tsk, &now);
1404 if (!cputime_eq(*oldval, cputime_zero)) {
1405 if (cputime_le(*oldval, now.cpu)) {
1406 /* Just about to fire. */
1407 *oldval = jiffies_to_cputime(1);
1409 *oldval = cputime_sub(*oldval, now.cpu);
1413 if (cputime_eq(*newval, cputime_zero))
1415 *newval = cputime_add(*newval, now.cpu);
1418 * If the RLIMIT_CPU timer will expire before the
1419 * ITIMER_PROF timer, we have nothing else to do.
1421 if (tsk->signal->rlim[RLIMIT_CPU].rlim_cur
1422 < cputime_to_secs(*newval))
1427 * Check whether there are any process timers already set to fire
1428 * before this one. If so, we don't have anything more to do.
1430 head = &tsk->signal->cpu_timers[clock_idx];
1431 if (list_empty(head) ||
1432 cputime_ge(list_first_entry(head,
1433 struct cpu_timer_list, entry)->expires.cpu,
1436 * Rejigger each thread's expiry time so that one will
1437 * notice before we hit the process-cumulative expiry time.
1439 union cpu_time_count expires = { .sched = 0 };
1440 expires.cpu = *newval;
1441 process_timer_rebalance(tsk, clock_idx, expires, now);
1445 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1446 struct timespec *rqtp, struct itimerspec *it)
1448 struct k_itimer timer;
1452 * Set up a temporary timer and then wait for it to go off.
1454 memset(&timer, 0, sizeof timer);
1455 spin_lock_init(&timer.it_lock);
1456 timer.it_clock = which_clock;
1457 timer.it_overrun = -1;
1458 error = posix_cpu_timer_create(&timer);
1459 timer.it_process = current;
1461 static struct itimerspec zero_it;
1463 memset(it, 0, sizeof *it);
1464 it->it_value = *rqtp;
1466 spin_lock_irq(&timer.it_lock);
1467 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1469 spin_unlock_irq(&timer.it_lock);
1473 while (!signal_pending(current)) {
1474 if (timer.it.cpu.expires.sched == 0) {
1476 * Our timer fired and was reset.
1478 spin_unlock_irq(&timer.it_lock);
1483 * Block until cpu_timer_fire (or a signal) wakes us.
1485 __set_current_state(TASK_INTERRUPTIBLE);
1486 spin_unlock_irq(&timer.it_lock);
1488 spin_lock_irq(&timer.it_lock);
1492 * We were interrupted by a signal.
1494 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1495 posix_cpu_timer_set(&timer, 0, &zero_it, it);
1496 spin_unlock_irq(&timer.it_lock);
1498 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1500 * It actually did fire already.
1505 error = -ERESTART_RESTARTBLOCK;
1511 int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1512 struct timespec *rqtp, struct timespec __user *rmtp)
1514 struct restart_block *restart_block =
1515 ¤t_thread_info()->restart_block;
1516 struct itimerspec it;
1520 * Diagnose required errors first.
1522 if (CPUCLOCK_PERTHREAD(which_clock) &&
1523 (CPUCLOCK_PID(which_clock) == 0 ||
1524 CPUCLOCK_PID(which_clock) == current->pid))
1527 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1529 if (error == -ERESTART_RESTARTBLOCK) {
1531 if (flags & TIMER_ABSTIME)
1532 return -ERESTARTNOHAND;
1534 * Report back to the user the time still remaining.
1536 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1539 restart_block->fn = posix_cpu_nsleep_restart;
1540 restart_block->arg0 = which_clock;
1541 restart_block->arg1 = (unsigned long) rmtp;
1542 restart_block->arg2 = rqtp->tv_sec;
1543 restart_block->arg3 = rqtp->tv_nsec;
1548 long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1550 clockid_t which_clock = restart_block->arg0;
1551 struct timespec __user *rmtp;
1553 struct itimerspec it;
1556 rmtp = (struct timespec __user *) restart_block->arg1;
1557 t.tv_sec = restart_block->arg2;
1558 t.tv_nsec = restart_block->arg3;
1560 restart_block->fn = do_no_restart_syscall;
1561 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1563 if (error == -ERESTART_RESTARTBLOCK) {
1565 * Report back to the user the time still remaining.
1567 if (rmtp != NULL && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1570 restart_block->fn = posix_cpu_nsleep_restart;
1571 restart_block->arg0 = which_clock;
1572 restart_block->arg1 = (unsigned long) rmtp;
1573 restart_block->arg2 = t.tv_sec;
1574 restart_block->arg3 = t.tv_nsec;
1581 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1582 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1584 static int process_cpu_clock_getres(const clockid_t which_clock,
1585 struct timespec *tp)
1587 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1589 static int process_cpu_clock_get(const clockid_t which_clock,
1590 struct timespec *tp)
1592 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1594 static int process_cpu_timer_create(struct k_itimer *timer)
1596 timer->it_clock = PROCESS_CLOCK;
1597 return posix_cpu_timer_create(timer);
1599 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1600 struct timespec *rqtp,
1601 struct timespec __user *rmtp)
1603 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1605 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1609 static int thread_cpu_clock_getres(const clockid_t which_clock,
1610 struct timespec *tp)
1612 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1614 static int thread_cpu_clock_get(const clockid_t which_clock,
1615 struct timespec *tp)
1617 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1619 static int thread_cpu_timer_create(struct k_itimer *timer)
1621 timer->it_clock = THREAD_CLOCK;
1622 return posix_cpu_timer_create(timer);
1624 static int thread_cpu_nsleep(const clockid_t which_clock, int flags,
1625 struct timespec *rqtp, struct timespec __user *rmtp)
1629 static long thread_cpu_nsleep_restart(struct restart_block *restart_block)
1634 static __init int init_posix_cpu_timers(void)
1636 struct k_clock process = {
1637 .clock_getres = process_cpu_clock_getres,
1638 .clock_get = process_cpu_clock_get,
1639 .clock_set = do_posix_clock_nosettime,
1640 .timer_create = process_cpu_timer_create,
1641 .nsleep = process_cpu_nsleep,
1642 .nsleep_restart = process_cpu_nsleep_restart,
1644 struct k_clock thread = {
1645 .clock_getres = thread_cpu_clock_getres,
1646 .clock_get = thread_cpu_clock_get,
1647 .clock_set = do_posix_clock_nosettime,
1648 .timer_create = thread_cpu_timer_create,
1649 .nsleep = thread_cpu_nsleep,
1650 .nsleep_restart = thread_cpu_nsleep_restart,
1653 register_posix_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1654 register_posix_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1658 __initcall(init_posix_cpu_timers);