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;
264 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
266 const pid_t pid = CPUCLOCK_PID(which_clock);
268 unsigned long long rtn;
272 * Special case constant value for our own clocks.
273 * We don't have to do any lookup to find ourselves.
275 if (CPUCLOCK_PERTHREAD(which_clock)) {
277 * Sampling just ourselves we can do with no locking.
279 error = cpu_clock_sample(which_clock,
282 read_lock(&tasklist_lock);
283 error = cpu_clock_sample_group(which_clock,
285 read_unlock(&tasklist_lock);
289 * Find the given PID, and validate that the caller
290 * should be able to see it.
292 struct task_struct *p;
294 p = find_task_by_vpid(pid);
296 if (CPUCLOCK_PERTHREAD(which_clock)) {
297 if (same_thread_group(p, current)) {
298 error = cpu_clock_sample(which_clock,
302 read_lock(&tasklist_lock);
303 if (thread_group_leader(p) && p->sighand) {
305 cpu_clock_sample_group(which_clock,
308 read_unlock(&tasklist_lock);
316 sample_to_timespec(which_clock, rtn, tp);
322 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
323 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
324 * new timer already all-zeros initialized.
326 static int posix_cpu_timer_create(struct k_itimer *new_timer)
329 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
330 struct task_struct *p;
332 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
335 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
338 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
342 p = find_task_by_vpid(pid);
343 if (p && !same_thread_group(p, current))
348 p = current->group_leader;
350 p = find_task_by_vpid(pid);
351 if (p && !has_group_leader_pid(p))
355 new_timer->it.cpu.task = p;
367 * Clean up a CPU-clock timer that is about to be destroyed.
368 * This is called from timer deletion with the timer already locked.
369 * If we return TIMER_RETRY, it's necessary to release the timer's lock
370 * and try again. (This happens when the timer is in the middle of firing.)
372 static int posix_cpu_timer_del(struct k_itimer *timer)
374 struct task_struct *p = timer->it.cpu.task;
377 WARN_ON_ONCE(p == NULL);
379 read_lock(&tasklist_lock);
380 if (unlikely(p->sighand == NULL)) {
382 * We raced with the reaping of the task.
383 * The deletion should have cleared us off the list.
385 BUG_ON(!list_empty(&timer->it.cpu.entry));
387 spin_lock(&p->sighand->siglock);
388 if (timer->it.cpu.firing)
391 list_del(&timer->it.cpu.entry);
392 spin_unlock(&p->sighand->siglock);
394 read_unlock(&tasklist_lock);
402 static void cleanup_timers_list(struct list_head *head)
404 struct cpu_timer_list *timer, *next;
406 list_for_each_entry_safe(timer, next, head, entry)
407 list_del_init(&timer->entry);
411 * Clean out CPU timers still ticking when a thread exited. The task
412 * pointer is cleared, and the expiry time is replaced with the residual
413 * time for later timer_gettime calls to return.
414 * This must be called with the siglock held.
416 static void cleanup_timers(struct list_head *head)
418 cleanup_timers_list(head);
419 cleanup_timers_list(++head);
420 cleanup_timers_list(++head);
424 * These are both called with the siglock held, when the current thread
425 * is being reaped. When the final (leader) thread in the group is reaped,
426 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
428 void posix_cpu_timers_exit(struct task_struct *tsk)
430 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
431 sizeof(unsigned long long));
432 cleanup_timers(tsk->cpu_timers);
435 void posix_cpu_timers_exit_group(struct task_struct *tsk)
437 cleanup_timers(tsk->signal->cpu_timers);
440 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
442 return expires == 0 || expires > new_exp;
446 * Insert the timer on the appropriate list before any timers that
447 * expire later. This must be called with the tasklist_lock held
448 * for reading, interrupts disabled and p->sighand->siglock taken.
450 static void arm_timer(struct k_itimer *timer)
452 struct task_struct *p = timer->it.cpu.task;
453 struct list_head *head, *listpos;
454 struct task_cputime *cputime_expires;
455 struct cpu_timer_list *const nt = &timer->it.cpu;
456 struct cpu_timer_list *next;
458 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
459 head = p->cpu_timers;
460 cputime_expires = &p->cputime_expires;
462 head = p->signal->cpu_timers;
463 cputime_expires = &p->signal->cputime_expires;
465 head += CPUCLOCK_WHICH(timer->it_clock);
468 list_for_each_entry(next, head, entry) {
469 if (nt->expires < next->expires)
471 listpos = &next->entry;
473 list_add(&nt->entry, listpos);
475 if (listpos == head) {
476 unsigned long long exp = nt->expires;
479 * We are the new earliest-expiring POSIX 1.b timer, hence
480 * need to update expiration cache. Take into account that
481 * for process timers we share expiration cache with itimers
482 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
485 switch (CPUCLOCK_WHICH(timer->it_clock)) {
487 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
488 cputime_expires->prof_exp = expires_to_cputime(exp);
491 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
492 cputime_expires->virt_exp = expires_to_cputime(exp);
495 if (cputime_expires->sched_exp == 0 ||
496 cputime_expires->sched_exp > exp)
497 cputime_expires->sched_exp = exp;
504 * The timer is locked, fire it and arrange for its reload.
506 static void cpu_timer_fire(struct k_itimer *timer)
508 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
510 * User don't want any signal.
512 timer->it.cpu.expires = 0;
513 } else if (unlikely(timer->sigq == NULL)) {
515 * This a special case for clock_nanosleep,
516 * not a normal timer from sys_timer_create.
518 wake_up_process(timer->it_process);
519 timer->it.cpu.expires = 0;
520 } else if (timer->it.cpu.incr == 0) {
522 * One-shot timer. Clear it as soon as it's fired.
524 posix_timer_event(timer, 0);
525 timer->it.cpu.expires = 0;
526 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
528 * The signal did not get queued because the signal
529 * was ignored, so we won't get any callback to
530 * reload the timer. But we need to keep it
531 * ticking in case the signal is deliverable next time.
533 posix_cpu_timer_schedule(timer);
538 * Sample a process (thread group) timer for the given group_leader task.
539 * Must be called with tasklist_lock held for reading.
541 static int cpu_timer_sample_group(const clockid_t which_clock,
542 struct task_struct *p,
543 unsigned long long *sample)
545 struct task_cputime cputime;
547 thread_group_cputimer(p, &cputime);
548 switch (CPUCLOCK_WHICH(which_clock)) {
552 *sample = cputime_to_expires(cputime.utime + cputime.stime);
555 *sample = cputime_to_expires(cputime.utime);
558 *sample = cputime.sum_exec_runtime + task_delta_exec(p);
564 #ifdef CONFIG_NO_HZ_FULL
565 static void nohz_kick_work_fn(struct work_struct *work)
567 tick_nohz_full_kick_all();
570 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
573 * We need the IPIs to be sent from sane process context.
574 * The posix cpu timers are always set with irqs disabled.
576 static void posix_cpu_timer_kick_nohz(void)
578 if (context_tracking_is_enabled())
579 schedule_work(&nohz_kick_work);
582 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
584 if (!task_cputime_zero(&tsk->cputime_expires))
587 if (tsk->signal->cputimer.running)
593 static inline void posix_cpu_timer_kick_nohz(void) { }
597 * Guts of sys_timer_settime for CPU timers.
598 * This is called with the timer locked and interrupts disabled.
599 * If we return TIMER_RETRY, it's necessary to release the timer's lock
600 * and try again. (This happens when the timer is in the middle of firing.)
602 static int posix_cpu_timer_set(struct k_itimer *timer, int flags,
603 struct itimerspec *new, struct itimerspec *old)
605 struct task_struct *p = timer->it.cpu.task;
606 unsigned long long old_expires, new_expires, old_incr, val;
609 WARN_ON_ONCE(p == NULL);
611 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
613 read_lock(&tasklist_lock);
615 * We need the tasklist_lock to protect against reaping that
616 * clears p->sighand. If p has just been reaped, we can no
617 * longer get any information about it at all.
619 if (unlikely(p->sighand == NULL)) {
620 read_unlock(&tasklist_lock);
625 * Disarm any old timer after extracting its expiry time.
627 BUG_ON(!irqs_disabled());
630 old_incr = timer->it.cpu.incr;
631 spin_lock(&p->sighand->siglock);
632 old_expires = timer->it.cpu.expires;
633 if (unlikely(timer->it.cpu.firing)) {
634 timer->it.cpu.firing = -1;
637 list_del_init(&timer->it.cpu.entry);
640 * We need to sample the current value to convert the new
641 * value from to relative and absolute, and to convert the
642 * old value from absolute to relative. To set a process
643 * timer, we need a sample to balance the thread expiry
644 * times (in arm_timer). With an absolute time, we must
645 * check if it's already passed. In short, we need a sample.
647 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
648 cpu_clock_sample(timer->it_clock, p, &val);
650 cpu_timer_sample_group(timer->it_clock, p, &val);
654 if (old_expires == 0) {
655 old->it_value.tv_sec = 0;
656 old->it_value.tv_nsec = 0;
659 * Update the timer in case it has
660 * overrun already. If it has,
661 * we'll report it as having overrun
662 * and with the next reloaded timer
663 * already ticking, though we are
664 * swallowing that pending
665 * notification here to install the
668 bump_cpu_timer(timer, val);
669 if (val < timer->it.cpu.expires) {
670 old_expires = timer->it.cpu.expires - val;
671 sample_to_timespec(timer->it_clock,
675 old->it_value.tv_nsec = 1;
676 old->it_value.tv_sec = 0;
683 * We are colliding with the timer actually firing.
684 * Punt after filling in the timer's old value, and
685 * disable this firing since we are already reporting
686 * it as an overrun (thanks to bump_cpu_timer above).
688 spin_unlock(&p->sighand->siglock);
689 read_unlock(&tasklist_lock);
693 if (new_expires != 0 && !(flags & TIMER_ABSTIME)) {
698 * Install the new expiry time (or zero).
699 * For a timer with no notification action, we don't actually
700 * arm the timer (we'll just fake it for timer_gettime).
702 timer->it.cpu.expires = new_expires;
703 if (new_expires != 0 && val < new_expires) {
707 spin_unlock(&p->sighand->siglock);
708 read_unlock(&tasklist_lock);
711 * Install the new reload setting, and
712 * set up the signal and overrun bookkeeping.
714 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
718 * This acts as a modification timestamp for the timer,
719 * so any automatic reload attempt will punt on seeing
720 * that we have reset the timer manually.
722 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
724 timer->it_overrun_last = 0;
725 timer->it_overrun = -1;
727 if (new_expires != 0 && !(val < new_expires)) {
729 * The designated time already passed, so we notify
730 * immediately, even if the thread never runs to
731 * accumulate more time on this clock.
733 cpu_timer_fire(timer);
739 sample_to_timespec(timer->it_clock,
740 old_incr, &old->it_interval);
743 posix_cpu_timer_kick_nohz();
747 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
749 unsigned long long now;
750 struct task_struct *p = timer->it.cpu.task;
752 WARN_ON_ONCE(p == NULL);
755 * Easy part: convert the reload time.
757 sample_to_timespec(timer->it_clock,
758 timer->it.cpu.incr, &itp->it_interval);
760 if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */
761 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
766 * Sample the clock to take the difference with the expiry time.
768 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
769 cpu_clock_sample(timer->it_clock, p, &now);
771 read_lock(&tasklist_lock);
772 if (unlikely(p->sighand == NULL)) {
774 * The process has been reaped.
775 * We can't even collect a sample any more.
776 * Call the timer disarmed, nothing else to do.
778 timer->it.cpu.expires = 0;
779 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
781 read_unlock(&tasklist_lock);
783 cpu_timer_sample_group(timer->it_clock, p, &now);
785 read_unlock(&tasklist_lock);
788 if (now < timer->it.cpu.expires) {
789 sample_to_timespec(timer->it_clock,
790 timer->it.cpu.expires - now,
794 * The timer should have expired already, but the firing
795 * hasn't taken place yet. Say it's just about to expire.
797 itp->it_value.tv_nsec = 1;
798 itp->it_value.tv_sec = 0;
802 static unsigned long long
803 check_timers_list(struct list_head *timers,
804 struct list_head *firing,
805 unsigned long long curr)
809 while (!list_empty(timers)) {
810 struct cpu_timer_list *t;
812 t = list_first_entry(timers, struct cpu_timer_list, entry);
814 if (!--maxfire || curr < t->expires)
818 list_move_tail(&t->entry, firing);
825 * Check for any per-thread CPU timers that have fired and move them off
826 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
827 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
829 static void check_thread_timers(struct task_struct *tsk,
830 struct list_head *firing)
832 struct list_head *timers = tsk->cpu_timers;
833 struct signal_struct *const sig = tsk->signal;
834 struct task_cputime *tsk_expires = &tsk->cputime_expires;
835 unsigned long long expires;
838 expires = check_timers_list(timers, firing, prof_ticks(tsk));
839 tsk_expires->prof_exp = expires_to_cputime(expires);
841 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
842 tsk_expires->virt_exp = expires_to_cputime(expires);
844 tsk_expires->sched_exp = check_timers_list(++timers, firing,
845 tsk->se.sum_exec_runtime);
848 * Check for the special case thread timers.
850 soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
851 if (soft != RLIM_INFINITY) {
853 ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
855 if (hard != RLIM_INFINITY &&
856 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
858 * At the hard limit, we just die.
859 * No need to calculate anything else now.
861 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
864 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
866 * At the soft limit, send a SIGXCPU every second.
869 soft += USEC_PER_SEC;
870 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
873 "RT Watchdog Timeout: %s[%d]\n",
874 tsk->comm, task_pid_nr(tsk));
875 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
880 static void stop_process_timers(struct signal_struct *sig)
882 struct thread_group_cputimer *cputimer = &sig->cputimer;
885 raw_spin_lock_irqsave(&cputimer->lock, flags);
886 cputimer->running = 0;
887 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
890 static u32 onecputick;
892 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
893 unsigned long long *expires,
894 unsigned long long cur_time, int signo)
899 if (cur_time >= it->expires) {
901 it->expires += it->incr;
902 it->error += it->incr_error;
903 if (it->error >= onecputick) {
904 it->expires -= cputime_one_jiffy;
905 it->error -= onecputick;
911 trace_itimer_expire(signo == SIGPROF ?
912 ITIMER_PROF : ITIMER_VIRTUAL,
913 tsk->signal->leader_pid, cur_time);
914 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
917 if (it->expires && (!*expires || it->expires < *expires)) {
918 *expires = it->expires;
923 * Check for any per-thread CPU timers that have fired and move them
924 * off the tsk->*_timers list onto the firing list. Per-thread timers
925 * have already been taken off.
927 static void check_process_timers(struct task_struct *tsk,
928 struct list_head *firing)
930 struct signal_struct *const sig = tsk->signal;
931 unsigned long long utime, ptime, virt_expires, prof_expires;
932 unsigned long long sum_sched_runtime, sched_expires;
933 struct list_head *timers = sig->cpu_timers;
934 struct task_cputime cputime;
938 * Collect the current process totals.
940 thread_group_cputimer(tsk, &cputime);
941 utime = cputime_to_expires(cputime.utime);
942 ptime = utime + cputime_to_expires(cputime.stime);
943 sum_sched_runtime = cputime.sum_exec_runtime;
945 prof_expires = check_timers_list(timers, firing, ptime);
946 virt_expires = check_timers_list(++timers, firing, utime);
947 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
950 * Check for the special case process timers.
952 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
954 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
956 soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
957 if (soft != RLIM_INFINITY) {
958 unsigned long psecs = cputime_to_secs(ptime);
960 ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
964 * At the hard limit, we just die.
965 * No need to calculate anything else now.
967 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
972 * At the soft limit, send a SIGXCPU every second.
974 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
977 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
980 x = secs_to_cputime(soft);
981 if (!prof_expires || x < prof_expires) {
986 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
987 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
988 sig->cputime_expires.sched_exp = sched_expires;
989 if (task_cputime_zero(&sig->cputime_expires))
990 stop_process_timers(sig);
994 * This is called from the signal code (via do_schedule_next_timer)
995 * when the last timer signal was delivered and we have to reload the timer.
997 void posix_cpu_timer_schedule(struct k_itimer *timer)
999 struct task_struct *p = timer->it.cpu.task;
1000 unsigned long long now;
1002 WARN_ON_ONCE(p == NULL);
1005 * Fetch the current sample and update the timer's expiry time.
1007 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1008 cpu_clock_sample(timer->it_clock, p, &now);
1009 bump_cpu_timer(timer, now);
1010 if (unlikely(p->exit_state))
1013 read_lock(&tasklist_lock); /* arm_timer needs it. */
1014 spin_lock(&p->sighand->siglock);
1016 read_lock(&tasklist_lock);
1017 if (unlikely(p->sighand == NULL)) {
1019 * The process has been reaped.
1020 * We can't even collect a sample any more.
1022 timer->it.cpu.expires = 0;
1023 read_unlock(&tasklist_lock);
1025 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1026 read_unlock(&tasklist_lock);
1027 /* Optimizations: if the process is dying, no need to rearm */
1030 spin_lock(&p->sighand->siglock);
1031 cpu_timer_sample_group(timer->it_clock, p, &now);
1032 bump_cpu_timer(timer, now);
1033 /* Leave the tasklist_lock locked for the call below. */
1037 * Now re-arm for the new expiry time.
1039 BUG_ON(!irqs_disabled());
1041 spin_unlock(&p->sighand->siglock);
1042 read_unlock(&tasklist_lock);
1044 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1045 posix_cpu_timer_kick_nohz();
1048 timer->it_overrun_last = timer->it_overrun;
1049 timer->it_overrun = -1;
1050 ++timer->it_requeue_pending;
1054 * task_cputime_expired - Compare two task_cputime entities.
1056 * @sample: The task_cputime structure to be checked for expiration.
1057 * @expires: Expiration times, against which @sample will be checked.
1059 * Checks @sample against @expires to see if any field of @sample has expired.
1060 * Returns true if any field of the former is greater than the corresponding
1061 * field of the latter if the latter field is set. Otherwise returns false.
1063 static inline int task_cputime_expired(const struct task_cputime *sample,
1064 const struct task_cputime *expires)
1066 if (expires->utime && sample->utime >= expires->utime)
1068 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1070 if (expires->sum_exec_runtime != 0 &&
1071 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1077 * fastpath_timer_check - POSIX CPU timers fast path.
1079 * @tsk: The task (thread) being checked.
1081 * Check the task and thread group timers. If both are zero (there are no
1082 * timers set) return false. Otherwise snapshot the task and thread group
1083 * timers and compare them with the corresponding expiration times. Return
1084 * true if a timer has expired, else return false.
1086 static inline int fastpath_timer_check(struct task_struct *tsk)
1088 struct signal_struct *sig;
1089 cputime_t utime, stime;
1091 task_cputime(tsk, &utime, &stime);
1093 if (!task_cputime_zero(&tsk->cputime_expires)) {
1094 struct task_cputime task_sample = {
1097 .sum_exec_runtime = tsk->se.sum_exec_runtime
1100 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1105 if (sig->cputimer.running) {
1106 struct task_cputime group_sample;
1108 raw_spin_lock(&sig->cputimer.lock);
1109 group_sample = sig->cputimer.cputime;
1110 raw_spin_unlock(&sig->cputimer.lock);
1112 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1120 * This is called from the timer interrupt handler. The irq handler has
1121 * already updated our counts. We need to check if any timers fire now.
1122 * Interrupts are disabled.
1124 void run_posix_cpu_timers(struct task_struct *tsk)
1127 struct k_itimer *timer, *next;
1128 unsigned long flags;
1130 BUG_ON(!irqs_disabled());
1133 * The fast path checks that there are no expired thread or thread
1134 * group timers. If that's so, just return.
1136 if (!fastpath_timer_check(tsk))
1139 if (!lock_task_sighand(tsk, &flags))
1142 * Here we take off tsk->signal->cpu_timers[N] and
1143 * tsk->cpu_timers[N] all the timers that are firing, and
1144 * put them on the firing list.
1146 check_thread_timers(tsk, &firing);
1148 * If there are any active process wide timers (POSIX 1.b, itimers,
1149 * RLIMIT_CPU) cputimer must be running.
1151 if (tsk->signal->cputimer.running)
1152 check_process_timers(tsk, &firing);
1155 * We must release these locks before taking any timer's lock.
1156 * There is a potential race with timer deletion here, as the
1157 * siglock now protects our private firing list. We have set
1158 * the firing flag in each timer, so that a deletion attempt
1159 * that gets the timer lock before we do will give it up and
1160 * spin until we've taken care of that timer below.
1162 unlock_task_sighand(tsk, &flags);
1165 * Now that all the timers on our list have the firing flag,
1166 * no one will touch their list entries but us. We'll take
1167 * each timer's lock before clearing its firing flag, so no
1168 * timer call will interfere.
1170 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1173 spin_lock(&timer->it_lock);
1174 list_del_init(&timer->it.cpu.entry);
1175 cpu_firing = timer->it.cpu.firing;
1176 timer->it.cpu.firing = 0;
1178 * The firing flag is -1 if we collided with a reset
1179 * of the timer, which already reported this
1180 * almost-firing as an overrun. So don't generate an event.
1182 if (likely(cpu_firing >= 0))
1183 cpu_timer_fire(timer);
1184 spin_unlock(&timer->it_lock);
1189 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1190 * The tsk->sighand->siglock must be held by the caller.
1192 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1193 cputime_t *newval, cputime_t *oldval)
1195 unsigned long long now;
1197 BUG_ON(clock_idx == CPUCLOCK_SCHED);
1198 cpu_timer_sample_group(clock_idx, tsk, &now);
1202 * We are setting itimer. The *oldval is absolute and we update
1203 * it to be relative, *newval argument is relative and we update
1204 * it to be absolute.
1207 if (*oldval <= now) {
1208 /* Just about to fire. */
1209 *oldval = cputime_one_jiffy;
1221 * Update expiration cache if we are the earliest timer, or eventually
1222 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1224 switch (clock_idx) {
1226 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1227 tsk->signal->cputime_expires.prof_exp = *newval;
1230 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1231 tsk->signal->cputime_expires.virt_exp = *newval;
1235 posix_cpu_timer_kick_nohz();
1238 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1239 struct timespec *rqtp, struct itimerspec *it)
1241 struct k_itimer timer;
1245 * Set up a temporary timer and then wait for it to go off.
1247 memset(&timer, 0, sizeof timer);
1248 spin_lock_init(&timer.it_lock);
1249 timer.it_clock = which_clock;
1250 timer.it_overrun = -1;
1251 error = posix_cpu_timer_create(&timer);
1252 timer.it_process = current;
1254 static struct itimerspec zero_it;
1256 memset(it, 0, sizeof *it);
1257 it->it_value = *rqtp;
1259 spin_lock_irq(&timer.it_lock);
1260 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1262 spin_unlock_irq(&timer.it_lock);
1266 while (!signal_pending(current)) {
1267 if (timer.it.cpu.expires == 0) {
1269 * Our timer fired and was reset, below
1270 * deletion can not fail.
1272 posix_cpu_timer_del(&timer);
1273 spin_unlock_irq(&timer.it_lock);
1278 * Block until cpu_timer_fire (or a signal) wakes us.
1280 __set_current_state(TASK_INTERRUPTIBLE);
1281 spin_unlock_irq(&timer.it_lock);
1283 spin_lock_irq(&timer.it_lock);
1287 * We were interrupted by a signal.
1289 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1290 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1293 * Timer is now unarmed, deletion can not fail.
1295 posix_cpu_timer_del(&timer);
1297 spin_unlock_irq(&timer.it_lock);
1299 while (error == TIMER_RETRY) {
1301 * We need to handle case when timer was or is in the
1302 * middle of firing. In other cases we already freed
1305 spin_lock_irq(&timer.it_lock);
1306 error = posix_cpu_timer_del(&timer);
1307 spin_unlock_irq(&timer.it_lock);
1310 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1312 * It actually did fire already.
1317 error = -ERESTART_RESTARTBLOCK;
1323 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1325 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1326 struct timespec *rqtp, struct timespec __user *rmtp)
1328 struct restart_block *restart_block =
1329 ¤t_thread_info()->restart_block;
1330 struct itimerspec it;
1334 * Diagnose required errors first.
1336 if (CPUCLOCK_PERTHREAD(which_clock) &&
1337 (CPUCLOCK_PID(which_clock) == 0 ||
1338 CPUCLOCK_PID(which_clock) == current->pid))
1341 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1343 if (error == -ERESTART_RESTARTBLOCK) {
1345 if (flags & TIMER_ABSTIME)
1346 return -ERESTARTNOHAND;
1348 * Report back to the user the time still remaining.
1350 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1353 restart_block->fn = posix_cpu_nsleep_restart;
1354 restart_block->nanosleep.clockid = which_clock;
1355 restart_block->nanosleep.rmtp = rmtp;
1356 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1361 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1363 clockid_t which_clock = restart_block->nanosleep.clockid;
1365 struct itimerspec it;
1368 t = ns_to_timespec(restart_block->nanosleep.expires);
1370 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1372 if (error == -ERESTART_RESTARTBLOCK) {
1373 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1375 * Report back to the user the time still remaining.
1377 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1380 restart_block->nanosleep.expires = timespec_to_ns(&t);
1386 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1387 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1389 static int process_cpu_clock_getres(const clockid_t which_clock,
1390 struct timespec *tp)
1392 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1394 static int process_cpu_clock_get(const clockid_t which_clock,
1395 struct timespec *tp)
1397 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1399 static int process_cpu_timer_create(struct k_itimer *timer)
1401 timer->it_clock = PROCESS_CLOCK;
1402 return posix_cpu_timer_create(timer);
1404 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1405 struct timespec *rqtp,
1406 struct timespec __user *rmtp)
1408 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1410 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1414 static int thread_cpu_clock_getres(const clockid_t which_clock,
1415 struct timespec *tp)
1417 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1419 static int thread_cpu_clock_get(const clockid_t which_clock,
1420 struct timespec *tp)
1422 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1424 static int thread_cpu_timer_create(struct k_itimer *timer)
1426 timer->it_clock = THREAD_CLOCK;
1427 return posix_cpu_timer_create(timer);
1430 struct k_clock clock_posix_cpu = {
1431 .clock_getres = posix_cpu_clock_getres,
1432 .clock_set = posix_cpu_clock_set,
1433 .clock_get = posix_cpu_clock_get,
1434 .timer_create = posix_cpu_timer_create,
1435 .nsleep = posix_cpu_nsleep,
1436 .nsleep_restart = posix_cpu_nsleep_restart,
1437 .timer_set = posix_cpu_timer_set,
1438 .timer_del = posix_cpu_timer_del,
1439 .timer_get = posix_cpu_timer_get,
1442 static __init int init_posix_cpu_timers(void)
1444 struct k_clock process = {
1445 .clock_getres = process_cpu_clock_getres,
1446 .clock_get = process_cpu_clock_get,
1447 .timer_create = process_cpu_timer_create,
1448 .nsleep = process_cpu_nsleep,
1449 .nsleep_restart = process_cpu_nsleep_restart,
1451 struct k_clock thread = {
1452 .clock_getres = thread_cpu_clock_getres,
1453 .clock_get = thread_cpu_clock_get,
1454 .timer_create = thread_cpu_timer_create,
1458 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1459 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1461 cputime_to_timespec(cputime_one_jiffy, &ts);
1462 onecputick = ts.tv_nsec;
1463 WARN_ON(ts.tv_sec != 0);
1467 __initcall(init_posix_cpu_timers);