2 * Implement CPU time clocks for the POSIX clock interface.
5 #include <linux/sched.h>
6 #include <linux/posix-timers.h>
7 #include <linux/errno.h>
8 #include <linux/math64.h>
9 #include <asm/uaccess.h>
10 #include <linux/kernel_stat.h>
11 #include <trace/events/timer.h>
12 #include <linux/random.h>
13 #include <linux/tick.h>
14 #include <linux/workqueue.h>
17 * Called after updating RLIMIT_CPU to run cpu timer and update
18 * tsk->signal->cputime_expires expiration cache if necessary. Needs
19 * siglock protection since other code may update expiration cache as
22 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
24 cputime_t cputime = secs_to_cputime(rlim_new);
26 spin_lock_irq(&task->sighand->siglock);
27 set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
28 spin_unlock_irq(&task->sighand->siglock);
31 static int check_clock(const clockid_t which_clock)
34 struct task_struct *p;
35 const pid_t pid = CPUCLOCK_PID(which_clock);
37 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
44 p = find_task_by_vpid(pid);
45 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
46 same_thread_group(p, current) : has_group_leader_pid(p))) {
54 static inline unsigned long long
55 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
57 unsigned long long ret;
59 ret = 0; /* high half always zero when .cpu used */
60 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
61 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
63 ret = cputime_to_expires(timespec_to_cputime(tp));
68 static void sample_to_timespec(const clockid_t which_clock,
69 unsigned long long expires,
72 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
73 *tp = ns_to_timespec(expires);
75 cputime_to_timespec((__force cputime_t)expires, tp);
79 * Update expiry time from increment, and increase overrun count,
80 * given the current clock sample.
82 static void bump_cpu_timer(struct k_itimer *timer,
83 unsigned long long now)
86 unsigned long long delta, incr;
88 if (timer->it.cpu.incr == 0)
91 if (now < timer->it.cpu.expires)
94 incr = timer->it.cpu.incr;
95 delta = now + incr - timer->it.cpu.expires;
97 /* Don't use (incr*2 < delta), incr*2 might overflow. */
98 for (i = 0; incr < delta - incr; i++)
101 for (; i >= 0; incr >>= 1, i--) {
105 timer->it.cpu.expires += incr;
106 timer->it_overrun += 1 << i;
112 * task_cputime_zero - Check a task_cputime struct for all zero fields.
114 * @cputime: The struct to compare.
116 * Checks @cputime to see if all fields are zero. Returns true if all fields
117 * are zero, false if any field is nonzero.
119 static inline int task_cputime_zero(const struct task_cputime *cputime)
121 if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
126 static inline unsigned long long prof_ticks(struct task_struct *p)
128 cputime_t utime, stime;
130 task_cputime(p, &utime, &stime);
132 return cputime_to_expires(utime + stime);
134 static inline unsigned long long virt_ticks(struct task_struct *p)
138 task_cputime(p, &utime, NULL);
140 return cputime_to_expires(utime);
144 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
146 int error = check_clock(which_clock);
149 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
150 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
152 * If sched_clock is using a cycle counter, we
153 * don't have any idea of its true resolution
154 * exported, but it is much more than 1s/HZ.
163 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
166 * You can never reset a CPU clock, but we check for other errors
167 * in the call before failing with EPERM.
169 int error = check_clock(which_clock);
178 * Sample a per-thread clock for the given task.
180 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
181 unsigned long long *sample)
183 switch (CPUCLOCK_WHICH(which_clock)) {
187 *sample = prof_ticks(p);
190 *sample = virt_ticks(p);
193 *sample = task_sched_runtime(p);
199 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
201 if (b->utime > a->utime)
204 if (b->stime > a->stime)
207 if (b->sum_exec_runtime > a->sum_exec_runtime)
208 a->sum_exec_runtime = b->sum_exec_runtime;
211 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
213 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
214 struct task_cputime sum;
217 if (!cputimer->running) {
219 * The POSIX timer interface allows for absolute time expiry
220 * values through the TIMER_ABSTIME flag, therefore we have
221 * to synchronize the timer to the clock every time we start
224 thread_group_cputime(tsk, &sum);
225 raw_spin_lock_irqsave(&cputimer->lock, flags);
226 cputimer->running = 1;
227 update_gt_cputime(&cputimer->cputime, &sum);
229 raw_spin_lock_irqsave(&cputimer->lock, flags);
230 *times = cputimer->cputime;
231 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
235 * Sample a process (thread group) clock for the given group_leader task.
236 * Must be called with tasklist_lock held for reading.
238 static int cpu_clock_sample_group(const clockid_t which_clock,
239 struct task_struct *p,
240 unsigned long long *sample)
242 struct task_cputime cputime;
244 switch (CPUCLOCK_WHICH(which_clock)) {
248 thread_group_cputime(p, &cputime);
249 *sample = cputime_to_expires(cputime.utime + cputime.stime);
252 thread_group_cputime(p, &cputime);
253 *sample = cputime_to_expires(cputime.utime);
256 thread_group_cputime(p, &cputime);
257 *sample = cputime.sum_exec_runtime;
263 static int posix_cpu_clock_get_task(struct task_struct *tsk,
264 const clockid_t which_clock,
268 unsigned long long rtn;
270 if (CPUCLOCK_PERTHREAD(which_clock)) {
271 if (same_thread_group(tsk, current))
272 err = cpu_clock_sample(which_clock, tsk, &rtn);
274 read_lock(&tasklist_lock);
276 if (tsk->sighand && (tsk == current || thread_group_leader(tsk)))
277 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
279 read_unlock(&tasklist_lock);
283 sample_to_timespec(which_clock, rtn, tp);
289 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
291 const pid_t pid = CPUCLOCK_PID(which_clock);
296 * Special case constant value for our own clocks.
297 * We don't have to do any lookup to find ourselves.
299 err = posix_cpu_clock_get_task(current, which_clock, tp);
302 * Find the given PID, and validate that the caller
303 * should be able to see it.
305 struct task_struct *p;
307 p = find_task_by_vpid(pid);
309 err = posix_cpu_clock_get_task(p, which_clock, tp);
318 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
319 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
320 * new timer already all-zeros initialized.
322 static int posix_cpu_timer_create(struct k_itimer *new_timer)
325 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
326 struct task_struct *p;
328 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
331 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
334 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
338 p = find_task_by_vpid(pid);
339 if (p && !same_thread_group(p, current))
344 p = current->group_leader;
346 p = find_task_by_vpid(pid);
347 if (p && !has_group_leader_pid(p))
351 new_timer->it.cpu.task = p;
363 * Clean up a CPU-clock timer that is about to be destroyed.
364 * This is called from timer deletion with the timer already locked.
365 * If we return TIMER_RETRY, it's necessary to release the timer's lock
366 * and try again. (This happens when the timer is in the middle of firing.)
368 static int posix_cpu_timer_del(struct k_itimer *timer)
370 struct task_struct *p = timer->it.cpu.task;
373 WARN_ON_ONCE(p == NULL);
375 read_lock(&tasklist_lock);
376 if (unlikely(p->sighand == NULL)) {
378 * We raced with the reaping of the task.
379 * The deletion should have cleared us off the list.
381 BUG_ON(!list_empty(&timer->it.cpu.entry));
383 spin_lock(&p->sighand->siglock);
384 if (timer->it.cpu.firing)
387 list_del(&timer->it.cpu.entry);
388 spin_unlock(&p->sighand->siglock);
390 read_unlock(&tasklist_lock);
398 static void cleanup_timers_list(struct list_head *head)
400 struct cpu_timer_list *timer, *next;
402 list_for_each_entry_safe(timer, next, head, entry)
403 list_del_init(&timer->entry);
407 * Clean out CPU timers still ticking when a thread exited. The task
408 * pointer is cleared, and the expiry time is replaced with the residual
409 * time for later timer_gettime calls to return.
410 * This must be called with the siglock held.
412 static void cleanup_timers(struct list_head *head)
414 cleanup_timers_list(head);
415 cleanup_timers_list(++head);
416 cleanup_timers_list(++head);
420 * These are both called with the siglock held, when the current thread
421 * is being reaped. When the final (leader) thread in the group is reaped,
422 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
424 void posix_cpu_timers_exit(struct task_struct *tsk)
426 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
427 sizeof(unsigned long long));
428 cleanup_timers(tsk->cpu_timers);
431 void posix_cpu_timers_exit_group(struct task_struct *tsk)
433 cleanup_timers(tsk->signal->cpu_timers);
436 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
438 return expires == 0 || expires > new_exp;
442 * Insert the timer on the appropriate list before any timers that
443 * expire later. This must be called with the tasklist_lock held
444 * for reading, interrupts disabled and p->sighand->siglock taken.
446 static void arm_timer(struct k_itimer *timer)
448 struct task_struct *p = timer->it.cpu.task;
449 struct list_head *head, *listpos;
450 struct task_cputime *cputime_expires;
451 struct cpu_timer_list *const nt = &timer->it.cpu;
452 struct cpu_timer_list *next;
454 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
455 head = p->cpu_timers;
456 cputime_expires = &p->cputime_expires;
458 head = p->signal->cpu_timers;
459 cputime_expires = &p->signal->cputime_expires;
461 head += CPUCLOCK_WHICH(timer->it_clock);
464 list_for_each_entry(next, head, entry) {
465 if (nt->expires < next->expires)
467 listpos = &next->entry;
469 list_add(&nt->entry, listpos);
471 if (listpos == head) {
472 unsigned long long exp = nt->expires;
475 * We are the new earliest-expiring POSIX 1.b timer, hence
476 * need to update expiration cache. Take into account that
477 * for process timers we share expiration cache with itimers
478 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
481 switch (CPUCLOCK_WHICH(timer->it_clock)) {
483 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
484 cputime_expires->prof_exp = expires_to_cputime(exp);
487 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
488 cputime_expires->virt_exp = expires_to_cputime(exp);
491 if (cputime_expires->sched_exp == 0 ||
492 cputime_expires->sched_exp > exp)
493 cputime_expires->sched_exp = exp;
500 * The timer is locked, fire it and arrange for its reload.
502 static void cpu_timer_fire(struct k_itimer *timer)
504 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
506 * User don't want any signal.
508 timer->it.cpu.expires = 0;
509 } else if (unlikely(timer->sigq == NULL)) {
511 * This a special case for clock_nanosleep,
512 * not a normal timer from sys_timer_create.
514 wake_up_process(timer->it_process);
515 timer->it.cpu.expires = 0;
516 } else if (timer->it.cpu.incr == 0) {
518 * One-shot timer. Clear it as soon as it's fired.
520 posix_timer_event(timer, 0);
521 timer->it.cpu.expires = 0;
522 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
524 * The signal did not get queued because the signal
525 * was ignored, so we won't get any callback to
526 * reload the timer. But we need to keep it
527 * ticking in case the signal is deliverable next time.
529 posix_cpu_timer_schedule(timer);
534 * Sample a process (thread group) timer for the given group_leader task.
535 * Must be called with tasklist_lock held for reading.
537 static int cpu_timer_sample_group(const clockid_t which_clock,
538 struct task_struct *p,
539 unsigned long long *sample)
541 struct task_cputime cputime;
543 thread_group_cputimer(p, &cputime);
544 switch (CPUCLOCK_WHICH(which_clock)) {
548 *sample = cputime_to_expires(cputime.utime + cputime.stime);
551 *sample = cputime_to_expires(cputime.utime);
554 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
560 #ifdef CONFIG_NO_HZ_FULL
561 static void nohz_kick_work_fn(struct work_struct *work)
563 tick_nohz_full_kick_all();
566 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
569 * We need the IPIs to be sent from sane process context.
570 * The posix cpu timers are always set with irqs disabled.
572 static void posix_cpu_timer_kick_nohz(void)
574 if (context_tracking_is_enabled())
575 schedule_work(&nohz_kick_work);
578 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
580 if (!task_cputime_zero(&tsk->cputime_expires))
583 if (tsk->signal->cputimer.running)
589 static inline void posix_cpu_timer_kick_nohz(void) { }
593 * Guts of sys_timer_settime for CPU timers.
594 * This is called with the timer locked and interrupts disabled.
595 * If we return TIMER_RETRY, it's necessary to release the timer's lock
596 * and try again. (This happens when the timer is in the middle of firing.)
598 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
599 struct itimerspec *new, struct itimerspec *old)
601 struct task_struct *p = timer->it.cpu.task;
602 unsigned long long old_expires, new_expires, old_incr, val;
605 WARN_ON_ONCE(p == NULL);
607 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
609 read_lock(&tasklist_lock);
611 * We need the tasklist_lock to protect against reaping that
612 * clears p->sighand. If p has just been reaped, we can no
613 * longer get any information about it at all.
615 if (unlikely(p->sighand == NULL)) {
616 read_unlock(&tasklist_lock);
621 * Disarm any old timer after extracting its expiry time.
623 BUG_ON(!irqs_disabled());
626 old_incr = timer->it.cpu.incr;
627 spin_lock(&p->sighand->siglock);
628 old_expires = timer->it.cpu.expires;
629 if (unlikely(timer->it.cpu.firing)) {
630 timer->it.cpu.firing = -1;
633 list_del_init(&timer->it.cpu.entry);
636 * We need to sample the current value to convert the new
637 * value from to relative and absolute, and to convert the
638 * old value from absolute to relative. To set a process
639 * timer, we need a sample to balance the thread expiry
640 * times (in arm_timer). With an absolute time, we must
641 * check if it's already passed. In short, we need a sample.
643 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
644 cpu_clock_sample(timer->it_clock, p, &val);
646 cpu_timer_sample_group(timer->it_clock, p, &val);
650 if (old_expires == 0) {
651 old->it_value.tv_sec = 0;
652 old->it_value.tv_nsec = 0;
655 * Update the timer in case it has
656 * overrun already. If it has,
657 * we'll report it as having overrun
658 * and with the next reloaded timer
659 * already ticking, though we are
660 * swallowing that pending
661 * notification here to install the
664 bump_cpu_timer(timer, val);
665 if (val < timer->it.cpu.expires) {
666 old_expires = timer->it.cpu.expires - val;
667 sample_to_timespec(timer->it_clock,
671 old->it_value.tv_nsec = 1;
672 old->it_value.tv_sec = 0;
679 * We are colliding with the timer actually firing.
680 * Punt after filling in the timer's old value, and
681 * disable this firing since we are already reporting
682 * it as an overrun (thanks to bump_cpu_timer above).
684 spin_unlock(&p->sighand->siglock);
685 read_unlock(&tasklist_lock);
689 if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
694 * Install the new expiry time (or zero).
695 * For a timer with no notification action, we don't actually
696 * arm the timer (we'll just fake it for timer_gettime).
698 timer->it.cpu.expires = new_expires;
699 if (new_expires != 0 && val < new_expires) {
703 spin_unlock(&p->sighand->siglock);
704 read_unlock(&tasklist_lock);
707 * Install the new reload setting, and
708 * set up the signal and overrun bookkeeping.
710 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
714 * This acts as a modification timestamp for the timer,
715 * so any automatic reload attempt will punt on seeing
716 * that we have reset the timer manually.
718 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
720 timer->it_overrun_last = 0;
721 timer->it_overrun = -1;
723 if (new_expires != 0 && !(val < new_expires)) {
725 * The designated time already passed, so we notify
726 * immediately, even if the thread never runs to
727 * accumulate more time on this clock.
729 cpu_timer_fire(timer);
735 sample_to_timespec(timer->it_clock,
736 old_incr, &old->it_interval);
739 posix_cpu_timer_kick_nohz();
743 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
745 unsigned long long now;
746 struct task_struct *p = timer->it.cpu.task;
748 WARN_ON_ONCE(p == NULL);
751 * Easy part: convert the reload time.
753 sample_to_timespec(timer->it_clock,
754 timer->it.cpu.incr, &itp->it_interval);
756 if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */
757 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
762 * Sample the clock to take the difference with the expiry time.
764 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
765 cpu_clock_sample(timer->it_clock, p, &now);
767 read_lock(&tasklist_lock);
768 if (unlikely(p->sighand == NULL)) {
770 * The process has been reaped.
771 * We can't even collect a sample any more.
772 * Call the timer disarmed, nothing else to do.
774 timer->it.cpu.expires = 0;
775 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
777 read_unlock(&tasklist_lock);
779 cpu_timer_sample_group(timer->it_clock, p, &now);
781 read_unlock(&tasklist_lock);
784 if (now < timer->it.cpu.expires) {
785 sample_to_timespec(timer->it_clock,
786 timer->it.cpu.expires - now,
790 * The timer should have expired already, but the firing
791 * hasn't taken place yet. Say it's just about to expire.
793 itp->it_value.tv_nsec = 1;
794 itp->it_value.tv_sec = 0;
798 static unsigned long long
799 check_timers_list(struct list_head *timers,
800 struct list_head *firing,
801 unsigned long long curr)
805 while (!list_empty(timers)) {
806 struct cpu_timer_list *t;
808 t = list_first_entry(timers, struct cpu_timer_list, entry);
810 if (!--maxfire || curr < t->expires)
814 list_move_tail(&t->entry, firing);
821 * Check for any per-thread CPU timers that have fired and move them off
822 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
823 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
825 static void check_thread_timers(struct task_struct *tsk,
826 struct list_head *firing)
828 struct list_head *timers = tsk->cpu_timers;
829 struct signal_struct *const sig = tsk->signal;
830 struct task_cputime *tsk_expires = &tsk->cputime_expires;
831 unsigned long long expires;
834 expires = check_timers_list(timers, firing, prof_ticks(tsk));
835 tsk_expires->prof_exp = expires_to_cputime(expires);
837 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
838 tsk_expires->virt_exp = expires_to_cputime(expires);
840 tsk_expires->sched_exp = check_timers_list(++timers, firing,
841 tsk->se.sum_exec_runtime);
844 * Check for the special case thread timers.
846 soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
847 if (soft != RLIM_INFINITY) {
849 ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
851 if (hard != RLIM_INFINITY &&
852 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
854 * At the hard limit, we just die.
855 * No need to calculate anything else now.
857 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
860 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
862 * At the soft limit, send a SIGXCPU every second.
865 soft += USEC_PER_SEC;
866 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
869 "RT Watchdog Timeout: %s[%d]\n",
870 tsk->comm, task_pid_nr(tsk));
871 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
876 static void stop_process_timers(struct signal_struct *sig)
878 struct thread_group_cputimer *cputimer = &sig->cputimer;
881 raw_spin_lock_irqsave(&cputimer->lock, flags);
882 cputimer->running = 0;
883 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
886 static u32 onecputick;
888 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
889 unsigned long long *expires,
890 unsigned long long cur_time, int signo)
895 if (cur_time >= it->expires) {
897 it->expires += it->incr;
898 it->error += it->incr_error;
899 if (it->error >= onecputick) {
900 it->expires -= cputime_one_jiffy;
901 it->error -= onecputick;
907 trace_itimer_expire(signo == SIGPROF ?
908 ITIMER_PROF : ITIMER_VIRTUAL,
909 tsk->signal->leader_pid, cur_time);
910 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
913 if (it->expires && (!*expires || it->expires < *expires)) {
914 *expires = it->expires;
919 * Check for any per-thread CPU timers that have fired and move them
920 * off the tsk->*_timers list onto the firing list. Per-thread timers
921 * have already been taken off.
923 static void check_process_timers(struct task_struct *tsk,
924 struct list_head *firing)
926 struct signal_struct *const sig = tsk->signal;
927 unsigned long long utime, ptime, virt_expires, prof_expires;
928 unsigned long long sum_sched_runtime, sched_expires;
929 struct list_head *timers = sig->cpu_timers;
930 struct task_cputime cputime;
934 * Collect the current process totals.
936 thread_group_cputimer(tsk, &cputime);
937 utime = cputime_to_expires(cputime.utime);
938 ptime = utime + cputime_to_expires(cputime.stime);
939 sum_sched_runtime = cputime.sum_exec_runtime;
941 prof_expires = check_timers_list(timers, firing, ptime);
942 virt_expires = check_timers_list(++timers, firing, utime);
943 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
946 * Check for the special case process timers.
948 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
950 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
952 soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
953 if (soft != RLIM_INFINITY) {
954 unsigned long psecs = cputime_to_secs(ptime);
956 ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
960 * At the hard limit, we just die.
961 * No need to calculate anything else now.
963 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
968 * At the soft limit, send a SIGXCPU every second.
970 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
973 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
976 x = secs_to_cputime(soft);
977 if (!prof_expires || x < prof_expires) {
982 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
983 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
984 sig->cputime_expires.sched_exp = sched_expires;
985 if (task_cputime_zero(&sig->cputime_expires))
986 stop_process_timers(sig);
990 * This is called from the signal code (via do_schedule_next_timer)
991 * when the last timer signal was delivered and we have to reload the timer.
993 void posix_cpu_timer_schedule(struct k_itimer *timer)
995 struct task_struct *p = timer->it.cpu.task;
996 unsigned long long now;
998 WARN_ON_ONCE(p == NULL);
1001 * Fetch the current sample and update the timer's expiry time.
1003 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1004 cpu_clock_sample(timer->it_clock, p, &now);
1005 bump_cpu_timer(timer, now);
1006 if (unlikely(p->exit_state))
1009 read_lock(&tasklist_lock); /* arm_timer needs it. */
1010 spin_lock(&p->sighand->siglock);
1012 read_lock(&tasklist_lock);
1013 if (unlikely(p->sighand == NULL)) {
1015 * The process has been reaped.
1016 * We can't even collect a sample any more.
1018 timer->it.cpu.expires = 0;
1019 read_unlock(&tasklist_lock);
1021 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1022 read_unlock(&tasklist_lock);
1023 /* Optimizations: if the process is dying, no need to rearm */
1026 spin_lock(&p->sighand->siglock);
1027 cpu_timer_sample_group(timer->it_clock, p, &now);
1028 bump_cpu_timer(timer, now);
1029 /* Leave the tasklist_lock locked for the call below. */
1033 * Now re-arm for the new expiry time.
1035 BUG_ON(!irqs_disabled());
1037 spin_unlock(&p->sighand->siglock);
1038 read_unlock(&tasklist_lock);
1040 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1041 posix_cpu_timer_kick_nohz();
1044 timer->it_overrun_last = timer->it_overrun;
1045 timer->it_overrun = -1;
1046 ++timer->it_requeue_pending;
1050 * task_cputime_expired - Compare two task_cputime entities.
1052 * @sample: The task_cputime structure to be checked for expiration.
1053 * @expires: Expiration times, against which @sample will be checked.
1055 * Checks @sample against @expires to see if any field of @sample has expired.
1056 * Returns true if any field of the former is greater than the corresponding
1057 * field of the latter if the latter field is set. Otherwise returns false.
1059 static inline int task_cputime_expired(const struct task_cputime *sample,
1060 const struct task_cputime *expires)
1062 if (expires->utime && sample->utime >= expires->utime)
1064 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1066 if (expires->sum_exec_runtime != 0 &&
1067 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1073 * fastpath_timer_check - POSIX CPU timers fast path.
1075 * @tsk: The task (thread) being checked.
1077 * Check the task and thread group timers. If both are zero (there are no
1078 * timers set) return false. Otherwise snapshot the task and thread group
1079 * timers and compare them with the corresponding expiration times. Return
1080 * true if a timer has expired, else return false.
1082 static inline int fastpath_timer_check(struct task_struct *tsk)
1084 struct signal_struct *sig;
1085 cputime_t utime, stime;
1087 task_cputime(tsk, &utime, &stime);
1089 if (!task_cputime_zero(&tsk->cputime_expires)) {
1090 struct task_cputime task_sample = {
1093 .sum_exec_runtime = tsk->se.sum_exec_runtime
1096 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1101 if (sig->cputimer.running) {
1102 struct task_cputime group_sample;
1104 raw_spin_lock(&sig->cputimer.lock);
1105 group_sample = sig->cputimer.cputime;
1106 raw_spin_unlock(&sig->cputimer.lock);
1108 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1116 * This is called from the timer interrupt handler. The irq handler has
1117 * already updated our counts. We need to check if any timers fire now.
1118 * Interrupts are disabled.
1120 void run_posix_cpu_timers(struct task_struct *tsk)
1123 struct k_itimer *timer, *next;
1124 unsigned long flags;
1126 BUG_ON(!irqs_disabled());
1129 * The fast path checks that there are no expired thread or thread
1130 * group timers. If that's so, just return.
1132 if (!fastpath_timer_check(tsk))
1135 if (!lock_task_sighand(tsk, &flags))
1138 * Here we take off tsk->signal->cpu_timers[N] and
1139 * tsk->cpu_timers[N] all the timers that are firing, and
1140 * put them on the firing list.
1142 check_thread_timers(tsk, &firing);
1144 * If there are any active process wide timers (POSIX 1.b, itimers,
1145 * RLIMIT_CPU) cputimer must be running.
1147 if (tsk->signal->cputimer.running)
1148 check_process_timers(tsk, &firing);
1151 * We must release these locks before taking any timer's lock.
1152 * There is a potential race with timer deletion here, as the
1153 * siglock now protects our private firing list. We have set
1154 * the firing flag in each timer, so that a deletion attempt
1155 * that gets the timer lock before we do will give it up and
1156 * spin until we've taken care of that timer below.
1158 unlock_task_sighand(tsk, &flags);
1161 * Now that all the timers on our list have the firing flag,
1162 * no one will touch their list entries but us. We'll take
1163 * each timer's lock before clearing its firing flag, so no
1164 * timer call will interfere.
1166 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1169 spin_lock(&timer->it_lock);
1170 list_del_init(&timer->it.cpu.entry);
1171 cpu_firing = timer->it.cpu.firing;
1172 timer->it.cpu.firing = 0;
1174 * The firing flag is -1 if we collided with a reset
1175 * of the timer, which already reported this
1176 * almost-firing as an overrun. So don't generate an event.
1178 if (likely(cpu_firing >= 0))
1179 cpu_timer_fire(timer);
1180 spin_unlock(&timer->it_lock);
1185 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1186 * The tsk->sighand->siglock must be held by the caller.
1188 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1189 cputime_t *newval, cputime_t *oldval)
1191 unsigned long long now;
1193 BUG_ON(clock_idx == CPUCLOCK_SCHED);
1194 cpu_timer_sample_group(clock_idx, tsk, &now);
1198 * We are setting itimer. The *oldval is absolute and we update
1199 * it to be relative, *newval argument is relative and we update
1200 * it to be absolute.
1203 if (*oldval <= now) {
1204 /* Just about to fire. */
1205 *oldval = cputime_one_jiffy;
1217 * Update expiration cache if we are the earliest timer, or eventually
1218 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1220 switch (clock_idx) {
1222 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1223 tsk->signal->cputime_expires.prof_exp = *newval;
1226 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1227 tsk->signal->cputime_expires.virt_exp = *newval;
1231 posix_cpu_timer_kick_nohz();
1234 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1235 struct timespec *rqtp, struct itimerspec *it)
1237 struct k_itimer timer;
1241 * Set up a temporary timer and then wait for it to go off.
1243 memset(&timer, 0, sizeof timer);
1244 spin_lock_init(&timer.it_lock);
1245 timer.it_clock = which_clock;
1246 timer.it_overrun = -1;
1247 error = posix_cpu_timer_create(&timer);
1248 timer.it_process = current;
1250 static struct itimerspec zero_it;
1252 memset(it, 0, sizeof *it);
1253 it->it_value = *rqtp;
1255 spin_lock_irq(&timer.it_lock);
1256 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1258 spin_unlock_irq(&timer.it_lock);
1262 while (!signal_pending(current)) {
1263 if (timer.it.cpu.expires == 0) {
1265 * Our timer fired and was reset, below
1266 * deletion can not fail.
1268 posix_cpu_timer_del(&timer);
1269 spin_unlock_irq(&timer.it_lock);
1274 * Block until cpu_timer_fire (or a signal) wakes us.
1276 __set_current_state(TASK_INTERRUPTIBLE);
1277 spin_unlock_irq(&timer.it_lock);
1279 spin_lock_irq(&timer.it_lock);
1283 * We were interrupted by a signal.
1285 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1286 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1289 * Timer is now unarmed, deletion can not fail.
1291 posix_cpu_timer_del(&timer);
1293 spin_unlock_irq(&timer.it_lock);
1295 while (error == TIMER_RETRY) {
1297 * We need to handle case when timer was or is in the
1298 * middle of firing. In other cases we already freed
1301 spin_lock_irq(&timer.it_lock);
1302 error = posix_cpu_timer_del(&timer);
1303 spin_unlock_irq(&timer.it_lock);
1306 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1308 * It actually did fire already.
1313 error = -ERESTART_RESTARTBLOCK;
1319 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1321 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1322 struct timespec *rqtp, struct timespec __user *rmtp)
1324 struct restart_block *restart_block =
1325 ¤t_thread_info()->restart_block;
1326 struct itimerspec it;
1330 * Diagnose required errors first.
1332 if (CPUCLOCK_PERTHREAD(which_clock) &&
1333 (CPUCLOCK_PID(which_clock) == 0 ||
1334 CPUCLOCK_PID(which_clock) == current->pid))
1337 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1339 if (error == -ERESTART_RESTARTBLOCK) {
1341 if (flags & TIMER_ABSTIME)
1342 return -ERESTARTNOHAND;
1344 * Report back to the user the time still remaining.
1346 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1349 restart_block->fn = posix_cpu_nsleep_restart;
1350 restart_block->nanosleep.clockid = which_clock;
1351 restart_block->nanosleep.rmtp = rmtp;
1352 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1357 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1359 clockid_t which_clock = restart_block->nanosleep.clockid;
1361 struct itimerspec it;
1364 t = ns_to_timespec(restart_block->nanosleep.expires);
1366 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1368 if (error == -ERESTART_RESTARTBLOCK) {
1369 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1371 * Report back to the user the time still remaining.
1373 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1376 restart_block->nanosleep.expires = timespec_to_ns(&t);
1382 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1383 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1385 static int process_cpu_clock_getres(const clockid_t which_clock,
1386 struct timespec *tp)
1388 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1390 static int process_cpu_clock_get(const clockid_t which_clock,
1391 struct timespec *tp)
1393 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1395 static int process_cpu_timer_create(struct k_itimer *timer)
1397 timer->it_clock = PROCESS_CLOCK;
1398 return posix_cpu_timer_create(timer);
1400 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1401 struct timespec *rqtp,
1402 struct timespec __user *rmtp)
1404 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1406 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1410 static int thread_cpu_clock_getres(const clockid_t which_clock,
1411 struct timespec *tp)
1413 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1415 static int thread_cpu_clock_get(const clockid_t which_clock,
1416 struct timespec *tp)
1418 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1420 static int thread_cpu_timer_create(struct k_itimer *timer)
1422 timer->it_clock = THREAD_CLOCK;
1423 return posix_cpu_timer_create(timer);
1426 struct k_clock clock_posix_cpu = {
1427 .clock_getres = posix_cpu_clock_getres,
1428 .clock_set = posix_cpu_clock_set,
1429 .clock_get = posix_cpu_clock_get,
1430 .timer_create = posix_cpu_timer_create,
1431 .nsleep = posix_cpu_nsleep,
1432 .nsleep_restart = posix_cpu_nsleep_restart,
1433 .timer_set = posix_cpu_timer_set,
1434 .timer_del = posix_cpu_timer_del,
1435 .timer_get = posix_cpu_timer_get,
1438 static __init int init_posix_cpu_timers(void)
1440 struct k_clock process = {
1441 .clock_getres = process_cpu_clock_getres,
1442 .clock_get = process_cpu_clock_get,
1443 .timer_create = process_cpu_timer_create,
1444 .nsleep = process_cpu_nsleep,
1445 .nsleep_restart = process_cpu_nsleep_restart,
1447 struct k_clock thread = {
1448 .clock_getres = thread_cpu_clock_getres,
1449 .clock_get = thread_cpu_clock_get,
1450 .timer_create = thread_cpu_timer_create,
1454 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1455 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1457 cputime_to_timespec(cputime_one_jiffy, &ts);
1458 onecputick = ts.tv_nsec;
1459 WARN_ON(ts.tv_sec != 0);
1463 __initcall(init_posix_cpu_timers);