2 * Implement CPU time clocks for the POSIX clock interface.
5 #include <linux/sched.h>
6 #include <linux/sched/rt.h>
7 #include <linux/posix-timers.h>
8 #include <linux/errno.h>
9 #include <linux/math64.h>
10 #include <asm/uaccess.h>
11 #include <linux/kernel_stat.h>
12 #include <trace/events/timer.h>
13 #include <linux/random.h>
14 #include <linux/tick.h>
15 #include <linux/workqueue.h>
18 * Called after updating RLIMIT_CPU to run cpu timer and update
19 * tsk->signal->cputime_expires expiration cache if necessary. Needs
20 * siglock protection since other code may update expiration cache as
23 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
25 cputime_t cputime = secs_to_cputime(rlim_new);
27 spin_lock_irq(&task->sighand->siglock);
28 set_process_cpu_timer(task, CPUCLOCK_PROF, &cputime, NULL);
29 spin_unlock_irq(&task->sighand->siglock);
32 static int check_clock(const clockid_t which_clock)
35 struct task_struct *p;
36 const pid_t pid = CPUCLOCK_PID(which_clock);
38 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
45 p = find_task_by_vpid(pid);
46 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
47 same_thread_group(p, current) : has_group_leader_pid(p))) {
55 static inline unsigned long long
56 timespec_to_sample(const clockid_t which_clock, const struct timespec *tp)
58 unsigned long long ret;
60 ret = 0; /* high half always zero when .cpu used */
61 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
62 ret = (unsigned long long)tp->tv_sec * NSEC_PER_SEC + tp->tv_nsec;
64 ret = cputime_to_expires(timespec_to_cputime(tp));
69 static void sample_to_timespec(const clockid_t which_clock,
70 unsigned long long expires,
73 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED)
74 *tp = ns_to_timespec(expires);
76 cputime_to_timespec((__force cputime_t)expires, tp);
80 * Update expiry time from increment, and increase overrun count,
81 * given the current clock sample.
83 static void bump_cpu_timer(struct k_itimer *timer,
84 unsigned long long now)
87 unsigned long long delta, incr;
89 if (timer->it.cpu.incr == 0)
92 if (now < timer->it.cpu.expires)
95 incr = timer->it.cpu.incr;
96 delta = now + incr - timer->it.cpu.expires;
98 /* Don't use (incr*2 < delta), incr*2 might overflow. */
99 for (i = 0; incr < delta - incr; i++)
102 for (; i >= 0; incr >>= 1, i--) {
106 timer->it.cpu.expires += incr;
107 timer->it_overrun += 1 << i;
113 * task_cputime_zero - Check a task_cputime struct for all zero fields.
115 * @cputime: The struct to compare.
117 * Checks @cputime to see if all fields are zero. Returns true if all fields
118 * are zero, false if any field is nonzero.
120 static inline int task_cputime_zero(const struct task_cputime *cputime)
122 if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
127 static inline unsigned long long prof_ticks(struct task_struct *p)
129 cputime_t utime, stime;
131 task_cputime(p, &utime, &stime);
133 return cputime_to_expires(utime + stime);
135 static inline unsigned long long virt_ticks(struct task_struct *p)
139 task_cputime(p, &utime, NULL);
141 return cputime_to_expires(utime);
145 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec *tp)
147 int error = check_clock(which_clock);
150 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
151 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
153 * If sched_clock is using a cycle counter, we
154 * don't have any idea of its true resolution
155 * exported, but it is much more than 1s/HZ.
164 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec *tp)
167 * You can never reset a CPU clock, but we check for other errors
168 * in the call before failing with EPERM.
170 int error = check_clock(which_clock);
179 * Sample a per-thread clock for the given task.
181 static int cpu_clock_sample(const clockid_t which_clock, struct task_struct *p,
182 unsigned long long *sample)
184 switch (CPUCLOCK_WHICH(which_clock)) {
188 *sample = prof_ticks(p);
191 *sample = virt_ticks(p);
194 *sample = task_sched_runtime(p);
200 static void update_gt_cputime(struct task_cputime *a, struct task_cputime *b)
202 if (b->utime > a->utime)
205 if (b->stime > a->stime)
208 if (b->sum_exec_runtime > a->sum_exec_runtime)
209 a->sum_exec_runtime = b->sum_exec_runtime;
212 void thread_group_cputimer(struct task_struct *tsk, struct task_cputime *times)
214 struct thread_group_cputimer *cputimer = &tsk->signal->cputimer;
215 struct task_cputime sum;
218 if (!cputimer->running) {
220 * The POSIX timer interface allows for absolute time expiry
221 * values through the TIMER_ABSTIME flag, therefore we have
222 * to synchronize the timer to the clock every time we start
225 thread_group_cputime(tsk, &sum);
226 raw_spin_lock_irqsave(&cputimer->lock, flags);
227 cputimer->running = 1;
228 update_gt_cputime(&cputimer->cputime, &sum);
230 raw_spin_lock_irqsave(&cputimer->lock, flags);
231 *times = cputimer->cputime;
232 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
236 * Sample a process (thread group) clock for the given group_leader task.
237 * Must be called with task sighand lock held for safe while_each_thread()
240 static int cpu_clock_sample_group(const clockid_t which_clock,
241 struct task_struct *p,
242 unsigned long long *sample)
244 struct task_cputime cputime;
246 switch (CPUCLOCK_WHICH(which_clock)) {
250 thread_group_cputime(p, &cputime);
251 *sample = cputime_to_expires(cputime.utime + cputime.stime);
254 thread_group_cputime(p, &cputime);
255 *sample = cputime_to_expires(cputime.utime);
258 thread_group_cputime(p, &cputime);
259 *sample = cputime.sum_exec_runtime;
265 static int posix_cpu_clock_get_task(struct task_struct *tsk,
266 const clockid_t which_clock,
270 unsigned long long rtn;
272 if (CPUCLOCK_PERTHREAD(which_clock)) {
273 if (same_thread_group(tsk, current))
274 err = cpu_clock_sample(which_clock, tsk, &rtn);
276 if (tsk == current || thread_group_leader(tsk))
277 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
281 sample_to_timespec(which_clock, rtn, tp);
287 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec *tp)
289 const pid_t pid = CPUCLOCK_PID(which_clock);
294 * Special case constant value for our own clocks.
295 * We don't have to do any lookup to find ourselves.
297 err = posix_cpu_clock_get_task(current, which_clock, tp);
300 * Find the given PID, and validate that the caller
301 * should be able to see it.
303 struct task_struct *p;
305 p = find_task_by_vpid(pid);
307 err = posix_cpu_clock_get_task(p, which_clock, tp);
316 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
317 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
318 * new timer already all-zeros initialized.
320 static int posix_cpu_timer_create(struct k_itimer *new_timer)
323 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
324 struct task_struct *p;
326 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
329 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
332 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
336 p = find_task_by_vpid(pid);
337 if (p && !same_thread_group(p, current))
342 p = current->group_leader;
344 p = find_task_by_vpid(pid);
345 if (p && !has_group_leader_pid(p))
349 new_timer->it.cpu.task = p;
361 * Clean up a CPU-clock timer that is about to be destroyed.
362 * This is called from timer deletion with the timer already locked.
363 * If we return TIMER_RETRY, it's necessary to release the timer's lock
364 * and try again. (This happens when the timer is in the middle of firing.)
366 static int posix_cpu_timer_del(struct k_itimer *timer)
370 struct sighand_struct *sighand;
371 struct task_struct *p = timer->it.cpu.task;
373 WARN_ON_ONCE(p == NULL);
376 * Protect against sighand release/switch in exit/exec and process/
377 * thread timer list entry concurrent read/writes.
379 sighand = lock_task_sighand(p, &flags);
380 if (unlikely(sighand == NULL)) {
382 * We raced with the reaping of the task.
383 * The deletion should have cleared us off the list.
385 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
387 if (timer->it.cpu.firing)
390 list_del(&timer->it.cpu.entry);
392 unlock_task_sighand(p, &flags);
401 static void cleanup_timers_list(struct list_head *head)
403 struct cpu_timer_list *timer, *next;
405 list_for_each_entry_safe(timer, next, head, entry)
406 list_del_init(&timer->entry);
410 * Clean out CPU timers still ticking when a thread exited. The task
411 * pointer is cleared, and the expiry time is replaced with the residual
412 * time for later timer_gettime calls to return.
413 * This must be called with the siglock held.
415 static void cleanup_timers(struct list_head *head)
417 cleanup_timers_list(head);
418 cleanup_timers_list(++head);
419 cleanup_timers_list(++head);
423 * These are both called with the siglock held, when the current thread
424 * is being reaped. When the final (leader) thread in the group is reaped,
425 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
427 void posix_cpu_timers_exit(struct task_struct *tsk)
429 add_device_randomness((const void*) &tsk->se.sum_exec_runtime,
430 sizeof(unsigned long long));
431 cleanup_timers(tsk->cpu_timers);
434 void posix_cpu_timers_exit_group(struct task_struct *tsk)
436 cleanup_timers(tsk->signal->cpu_timers);
439 static inline int expires_gt(cputime_t expires, cputime_t new_exp)
441 return expires == 0 || expires > new_exp;
445 * Insert the timer on the appropriate list before any timers that
446 * expire later. This must be called with the sighand lock held.
448 static void arm_timer(struct k_itimer *timer)
450 struct task_struct *p = timer->it.cpu.task;
451 struct list_head *head, *listpos;
452 struct task_cputime *cputime_expires;
453 struct cpu_timer_list *const nt = &timer->it.cpu;
454 struct cpu_timer_list *next;
456 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
457 head = p->cpu_timers;
458 cputime_expires = &p->cputime_expires;
460 head = p->signal->cpu_timers;
461 cputime_expires = &p->signal->cputime_expires;
463 head += CPUCLOCK_WHICH(timer->it_clock);
466 list_for_each_entry(next, head, entry) {
467 if (nt->expires < next->expires)
469 listpos = &next->entry;
471 list_add(&nt->entry, listpos);
473 if (listpos == head) {
474 unsigned long long exp = nt->expires;
477 * We are the new earliest-expiring POSIX 1.b timer, hence
478 * need to update expiration cache. Take into account that
479 * for process timers we share expiration cache with itimers
480 * and RLIMIT_CPU and for thread timers with RLIMIT_RTTIME.
483 switch (CPUCLOCK_WHICH(timer->it_clock)) {
485 if (expires_gt(cputime_expires->prof_exp, expires_to_cputime(exp)))
486 cputime_expires->prof_exp = expires_to_cputime(exp);
489 if (expires_gt(cputime_expires->virt_exp, expires_to_cputime(exp)))
490 cputime_expires->virt_exp = expires_to_cputime(exp);
493 if (cputime_expires->sched_exp == 0 ||
494 cputime_expires->sched_exp > exp)
495 cputime_expires->sched_exp = exp;
502 * The timer is locked, fire it and arrange for its reload.
504 static void cpu_timer_fire(struct k_itimer *timer)
506 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
508 * User don't want any signal.
510 timer->it.cpu.expires = 0;
511 } else if (unlikely(timer->sigq == NULL)) {
513 * This a special case for clock_nanosleep,
514 * not a normal timer from sys_timer_create.
516 wake_up_process(timer->it_process);
517 timer->it.cpu.expires = 0;
518 } else if (timer->it.cpu.incr == 0) {
520 * One-shot timer. Clear it as soon as it's fired.
522 posix_timer_event(timer, 0);
523 timer->it.cpu.expires = 0;
524 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
526 * The signal did not get queued because the signal
527 * was ignored, so we won't get any callback to
528 * reload the timer. But we need to keep it
529 * ticking in case the signal is deliverable next time.
531 posix_cpu_timer_schedule(timer);
536 * Sample a process (thread group) timer for the given group_leader task.
537 * Must be called with task sighand lock held for safe while_each_thread()
540 static int cpu_timer_sample_group(const clockid_t which_clock,
541 struct task_struct *p,
542 unsigned long long *sample)
544 struct task_cputime cputime;
546 thread_group_cputimer(p, &cputime);
547 switch (CPUCLOCK_WHICH(which_clock)) {
551 *sample = cputime_to_expires(cputime.utime + cputime.stime);
554 *sample = cputime_to_expires(cputime.utime);
557 *sample = cputime.sum_exec_runtime;
563 #ifdef CONFIG_NO_HZ_FULL
564 static void nohz_kick_work_fn(struct work_struct *work)
566 tick_nohz_full_kick_all();
569 static DECLARE_WORK(nohz_kick_work, nohz_kick_work_fn);
572 * We need the IPIs to be sent from sane process context.
573 * The posix cpu timers are always set with irqs disabled.
575 static void posix_cpu_timer_kick_nohz(void)
577 if (context_tracking_is_enabled())
578 schedule_work(&nohz_kick_work);
581 bool posix_cpu_timers_can_stop_tick(struct task_struct *tsk)
583 if (!task_cputime_zero(&tsk->cputime_expires))
586 if (tsk->signal->cputimer.running)
592 static inline void posix_cpu_timer_kick_nohz(void) { }
596 * Guts of sys_timer_settime for CPU timers.
597 * This is called with the timer locked and interrupts disabled.
598 * If we return TIMER_RETRY, it's necessary to release the timer's lock
599 * and try again. (This happens when the timer is in the middle of firing.)
601 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
602 struct itimerspec *new, struct itimerspec *old)
605 struct sighand_struct *sighand;
606 struct task_struct *p = timer->it.cpu.task;
607 unsigned long long old_expires, new_expires, old_incr, val;
610 WARN_ON_ONCE(p == NULL);
612 new_expires = timespec_to_sample(timer->it_clock, &new->it_value);
615 * Protect against sighand release/switch in exit/exec and p->cpu_timers
616 * and p->signal->cpu_timers read/write in arm_timer()
618 sighand = lock_task_sighand(p, &flags);
620 * If p has just been reaped, we can no
621 * longer get any information about it at all.
623 if (unlikely(sighand == NULL)) {
628 * Disarm any old timer after extracting its expiry time.
630 WARN_ON_ONCE_NONRT(!irqs_disabled());
633 old_incr = timer->it.cpu.incr;
634 old_expires = timer->it.cpu.expires;
635 if (unlikely(timer->it.cpu.firing)) {
636 timer->it.cpu.firing = -1;
639 list_del_init(&timer->it.cpu.entry);
642 * We need to sample the current value to convert the new
643 * value from to relative and absolute, and to convert the
644 * old value from absolute to relative. To set a process
645 * timer, we need a sample to balance the thread expiry
646 * times (in arm_timer). With an absolute time, we must
647 * check if it's already passed. In short, we need a sample.
649 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
650 cpu_clock_sample(timer->it_clock, p, &val);
652 cpu_timer_sample_group(timer->it_clock, p, &val);
656 if (old_expires == 0) {
657 old->it_value.tv_sec = 0;
658 old->it_value.tv_nsec = 0;
661 * Update the timer in case it has
662 * overrun already. If it has,
663 * we'll report it as having overrun
664 * and with the next reloaded timer
665 * already ticking, though we are
666 * swallowing that pending
667 * notification here to install the
670 bump_cpu_timer(timer, val);
671 if (val < timer->it.cpu.expires) {
672 old_expires = timer->it.cpu.expires - val;
673 sample_to_timespec(timer->it_clock,
677 old->it_value.tv_nsec = 1;
678 old->it_value.tv_sec = 0;
685 * We are colliding with the timer actually firing.
686 * Punt after filling in the timer's old value, and
687 * disable this firing since we are already reporting
688 * it as an overrun (thanks to bump_cpu_timer above).
690 unlock_task_sighand(p, &flags);
694 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
699 * Install the new expiry time (or zero).
700 * For a timer with no notification action, we don't actually
701 * arm the timer (we'll just fake it for timer_gettime).
703 timer->it.cpu.expires = new_expires;
704 if (new_expires != 0 && val < new_expires) {
708 unlock_task_sighand(p, &flags);
710 * Install the new reload setting, and
711 * set up the signal and overrun bookkeeping.
713 timer->it.cpu.incr = timespec_to_sample(timer->it_clock,
717 * This acts as a modification timestamp for the timer,
718 * so any automatic reload attempt will punt on seeing
719 * that we have reset the timer manually.
721 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
723 timer->it_overrun_last = 0;
724 timer->it_overrun = -1;
726 if (new_expires != 0 && !(val < new_expires)) {
728 * The designated time already passed, so we notify
729 * immediately, even if the thread never runs to
730 * accumulate more time on this clock.
732 cpu_timer_fire(timer);
738 sample_to_timespec(timer->it_clock,
739 old_incr, &old->it_interval);
742 posix_cpu_timer_kick_nohz();
746 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec *itp)
748 unsigned long long now;
749 struct task_struct *p = timer->it.cpu.task;
751 WARN_ON_ONCE(p == NULL);
754 * Easy part: convert the reload time.
756 sample_to_timespec(timer->it_clock,
757 timer->it.cpu.incr, &itp->it_interval);
759 if (timer->it.cpu.expires == 0) { /* Timer not armed at all. */
760 itp->it_value.tv_sec = itp->it_value.tv_nsec = 0;
765 * Sample the clock to take the difference with the expiry time.
767 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
768 cpu_clock_sample(timer->it_clock, p, &now);
770 struct sighand_struct *sighand;
774 * Protect against sighand release/switch in exit/exec and
775 * also make timer sampling safe if it ends up calling
776 * thread_group_cputime().
778 sighand = lock_task_sighand(p, &flags);
779 if (unlikely(sighand == NULL)) {
781 * The process has been reaped.
782 * We can't even collect a sample any more.
783 * Call the timer disarmed, nothing else to do.
785 timer->it.cpu.expires = 0;
786 sample_to_timespec(timer->it_clock, timer->it.cpu.expires,
789 cpu_timer_sample_group(timer->it_clock, p, &now);
790 unlock_task_sighand(p, &flags);
794 if (now < timer->it.cpu.expires) {
795 sample_to_timespec(timer->it_clock,
796 timer->it.cpu.expires - now,
800 * The timer should have expired already, but the firing
801 * hasn't taken place yet. Say it's just about to expire.
803 itp->it_value.tv_nsec = 1;
804 itp->it_value.tv_sec = 0;
808 static unsigned long long
809 check_timers_list(struct list_head *timers,
810 struct list_head *firing,
811 unsigned long long curr)
815 while (!list_empty(timers)) {
816 struct cpu_timer_list *t;
818 t = list_first_entry(timers, struct cpu_timer_list, entry);
820 if (!--maxfire || curr < t->expires)
824 list_move_tail(&t->entry, firing);
831 * Check for any per-thread CPU timers that have fired and move them off
832 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
833 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
835 static void check_thread_timers(struct task_struct *tsk,
836 struct list_head *firing)
838 struct list_head *timers = tsk->cpu_timers;
839 struct signal_struct *const sig = tsk->signal;
840 struct task_cputime *tsk_expires = &tsk->cputime_expires;
841 unsigned long long expires;
844 expires = check_timers_list(timers, firing, prof_ticks(tsk));
845 tsk_expires->prof_exp = expires_to_cputime(expires);
847 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
848 tsk_expires->virt_exp = expires_to_cputime(expires);
850 tsk_expires->sched_exp = check_timers_list(++timers, firing,
851 tsk->se.sum_exec_runtime);
854 * Check for the special case thread timers.
856 soft = ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_cur);
857 if (soft != RLIM_INFINITY) {
859 ACCESS_ONCE(sig->rlim[RLIMIT_RTTIME].rlim_max);
861 if (hard != RLIM_INFINITY &&
862 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
864 * At the hard limit, we just die.
865 * No need to calculate anything else now.
867 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
870 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
872 * At the soft limit, send a SIGXCPU every second.
875 soft += USEC_PER_SEC;
876 sig->rlim[RLIMIT_RTTIME].rlim_cur = soft;
879 "RT Watchdog Timeout: %s[%d]\n",
880 tsk->comm, task_pid_nr(tsk));
881 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
886 static void stop_process_timers(struct signal_struct *sig)
888 struct thread_group_cputimer *cputimer = &sig->cputimer;
891 raw_spin_lock_irqsave(&cputimer->lock, flags);
892 cputimer->running = 0;
893 raw_spin_unlock_irqrestore(&cputimer->lock, flags);
896 static u32 onecputick;
898 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
899 unsigned long long *expires,
900 unsigned long long cur_time, int signo)
905 if (cur_time >= it->expires) {
907 it->expires += it->incr;
908 it->error += it->incr_error;
909 if (it->error >= onecputick) {
910 it->expires -= cputime_one_jiffy;
911 it->error -= onecputick;
917 trace_itimer_expire(signo == SIGPROF ?
918 ITIMER_PROF : ITIMER_VIRTUAL,
919 tsk->signal->leader_pid, cur_time);
920 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
923 if (it->expires && (!*expires || it->expires < *expires)) {
924 *expires = it->expires;
929 * Check for any per-thread CPU timers that have fired and move them
930 * off the tsk->*_timers list onto the firing list. Per-thread timers
931 * have already been taken off.
933 static void check_process_timers(struct task_struct *tsk,
934 struct list_head *firing)
936 struct signal_struct *const sig = tsk->signal;
937 unsigned long long utime, ptime, virt_expires, prof_expires;
938 unsigned long long sum_sched_runtime, sched_expires;
939 struct list_head *timers = sig->cpu_timers;
940 struct task_cputime cputime;
944 * Collect the current process totals.
946 thread_group_cputimer(tsk, &cputime);
947 utime = cputime_to_expires(cputime.utime);
948 ptime = utime + cputime_to_expires(cputime.stime);
949 sum_sched_runtime = cputime.sum_exec_runtime;
951 prof_expires = check_timers_list(timers, firing, ptime);
952 virt_expires = check_timers_list(++timers, firing, utime);
953 sched_expires = check_timers_list(++timers, firing, sum_sched_runtime);
956 * Check for the special case process timers.
958 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_PROF], &prof_expires, ptime,
960 check_cpu_itimer(tsk, &sig->it[CPUCLOCK_VIRT], &virt_expires, utime,
962 soft = ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_cur);
963 if (soft != RLIM_INFINITY) {
964 unsigned long psecs = cputime_to_secs(ptime);
966 ACCESS_ONCE(sig->rlim[RLIMIT_CPU].rlim_max);
970 * At the hard limit, we just die.
971 * No need to calculate anything else now.
973 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
978 * At the soft limit, send a SIGXCPU every second.
980 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
983 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
986 x = secs_to_cputime(soft);
987 if (!prof_expires || x < prof_expires) {
992 sig->cputime_expires.prof_exp = expires_to_cputime(prof_expires);
993 sig->cputime_expires.virt_exp = expires_to_cputime(virt_expires);
994 sig->cputime_expires.sched_exp = sched_expires;
995 if (task_cputime_zero(&sig->cputime_expires))
996 stop_process_timers(sig);
1000 * This is called from the signal code (via do_schedule_next_timer)
1001 * when the last timer signal was delivered and we have to reload the timer.
1003 void posix_cpu_timer_schedule(struct k_itimer *timer)
1005 struct sighand_struct *sighand;
1006 unsigned long flags;
1007 struct task_struct *p = timer->it.cpu.task;
1008 unsigned long long now;
1010 WARN_ON_ONCE(p == NULL);
1013 * Fetch the current sample and update the timer's expiry time.
1015 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1016 cpu_clock_sample(timer->it_clock, p, &now);
1017 bump_cpu_timer(timer, now);
1018 if (unlikely(p->exit_state))
1021 /* Protect timer list r/w in arm_timer() */
1022 sighand = lock_task_sighand(p, &flags);
1027 * Protect arm_timer() and timer sampling in case of call to
1028 * thread_group_cputime().
1030 sighand = lock_task_sighand(p, &flags);
1031 if (unlikely(sighand == NULL)) {
1033 * The process has been reaped.
1034 * We can't even collect a sample any more.
1036 timer->it.cpu.expires = 0;
1038 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1039 unlock_task_sighand(p, &flags);
1040 /* Optimizations: if the process is dying, no need to rearm */
1043 cpu_timer_sample_group(timer->it_clock, p, &now);
1044 bump_cpu_timer(timer, now);
1045 /* Leave the sighand locked for the call below. */
1049 * Now re-arm for the new expiry time.
1051 WARN_ON_ONCE_NONRT(!irqs_disabled());
1053 unlock_task_sighand(p, &flags);
1055 /* Kick full dynticks CPUs in case they need to tick on the new timer */
1056 posix_cpu_timer_kick_nohz();
1058 timer->it_overrun_last = timer->it_overrun;
1059 timer->it_overrun = -1;
1060 ++timer->it_requeue_pending;
1064 * task_cputime_expired - Compare two task_cputime entities.
1066 * @sample: The task_cputime structure to be checked for expiration.
1067 * @expires: Expiration times, against which @sample will be checked.
1069 * Checks @sample against @expires to see if any field of @sample has expired.
1070 * Returns true if any field of the former is greater than the corresponding
1071 * field of the latter if the latter field is set. Otherwise returns false.
1073 static inline int task_cputime_expired(const struct task_cputime *sample,
1074 const struct task_cputime *expires)
1076 if (expires->utime && sample->utime >= expires->utime)
1078 if (expires->stime && sample->utime + sample->stime >= expires->stime)
1080 if (expires->sum_exec_runtime != 0 &&
1081 sample->sum_exec_runtime >= expires->sum_exec_runtime)
1087 * fastpath_timer_check - POSIX CPU timers fast path.
1089 * @tsk: The task (thread) being checked.
1091 * Check the task and thread group timers. If both are zero (there are no
1092 * timers set) return false. Otherwise snapshot the task and thread group
1093 * timers and compare them with the corresponding expiration times. Return
1094 * true if a timer has expired, else return false.
1096 static inline int fastpath_timer_check(struct task_struct *tsk)
1098 struct signal_struct *sig;
1099 cputime_t utime, stime;
1101 task_cputime(tsk, &utime, &stime);
1103 if (!task_cputime_zero(&tsk->cputime_expires)) {
1104 struct task_cputime task_sample = {
1107 .sum_exec_runtime = tsk->se.sum_exec_runtime
1110 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1115 if (sig->cputimer.running) {
1116 struct task_cputime group_sample;
1117 unsigned long flags;
1119 raw_spin_lock_irqsave(&sig->cputimer.lock, flags);
1120 group_sample = sig->cputimer.cputime;
1121 raw_spin_unlock_irqrestore(&sig->cputimer.lock, flags);
1123 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1131 * This is called from the timer interrupt handler. The irq handler has
1132 * already updated our counts. We need to check if any timers fire now.
1133 * Interrupts are disabled.
1135 static void __run_posix_cpu_timers(struct task_struct *tsk)
1138 struct k_itimer *timer, *next;
1139 unsigned long flags;
1141 WARN_ON_ONCE_NONRT(!irqs_disabled());
1144 * The fast path checks that there are no expired thread or thread
1145 * group timers. If that's so, just return.
1147 if (!fastpath_timer_check(tsk))
1150 if (!lock_task_sighand(tsk, &flags))
1153 * Here we take off tsk->signal->cpu_timers[N] and
1154 * tsk->cpu_timers[N] all the timers that are firing, and
1155 * put them on the firing list.
1157 check_thread_timers(tsk, &firing);
1159 * If there are any active process wide timers (POSIX 1.b, itimers,
1160 * RLIMIT_CPU) cputimer must be running.
1162 if (tsk->signal->cputimer.running)
1163 check_process_timers(tsk, &firing);
1166 * We must release these locks before taking any timer's lock.
1167 * There is a potential race with timer deletion here, as the
1168 * siglock now protects our private firing list. We have set
1169 * the firing flag in each timer, so that a deletion attempt
1170 * that gets the timer lock before we do will give it up and
1171 * spin until we've taken care of that timer below.
1173 unlock_task_sighand(tsk, &flags);
1176 * Now that all the timers on our list have the firing flag,
1177 * no one will touch their list entries but us. We'll take
1178 * each timer's lock before clearing its firing flag, so no
1179 * timer call will interfere.
1181 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1184 spin_lock(&timer->it_lock);
1185 list_del_init(&timer->it.cpu.entry);
1186 cpu_firing = timer->it.cpu.firing;
1187 timer->it.cpu.firing = 0;
1189 * The firing flag is -1 if we collided with a reset
1190 * of the timer, which already reported this
1191 * almost-firing as an overrun. So don't generate an event.
1193 if (likely(cpu_firing >= 0))
1194 cpu_timer_fire(timer);
1195 spin_unlock(&timer->it_lock);
1199 #ifdef CONFIG_PREEMPT_RT_BASE
1200 #include <linux/kthread.h>
1201 #include <linux/cpu.h>
1202 DEFINE_PER_CPU(struct task_struct *, posix_timer_task);
1203 DEFINE_PER_CPU(struct task_struct *, posix_timer_tasklist);
1205 static int posix_cpu_timers_thread(void *data)
1207 int cpu = (long)data;
1209 BUG_ON(per_cpu(posix_timer_task,cpu) != current);
1211 while (!kthread_should_stop()) {
1212 struct task_struct *tsk = NULL;
1213 struct task_struct *next = NULL;
1215 if (cpu_is_offline(cpu))
1218 /* grab task list */
1219 raw_local_irq_disable();
1220 tsk = per_cpu(posix_timer_tasklist, cpu);
1221 per_cpu(posix_timer_tasklist, cpu) = NULL;
1222 raw_local_irq_enable();
1224 /* its possible the list is empty, just return */
1226 set_current_state(TASK_INTERRUPTIBLE);
1228 __set_current_state(TASK_RUNNING);
1232 /* Process task list */
1235 next = tsk->posix_timer_list;
1237 /* run the task timers, clear its ptr and
1240 __run_posix_cpu_timers(tsk);
1241 tsk->posix_timer_list = NULL;
1242 put_task_struct(tsk);
1244 /* check if this is the last on the list */
1253 /* Wait for kthread_stop */
1254 set_current_state(TASK_INTERRUPTIBLE);
1255 while (!kthread_should_stop()) {
1257 set_current_state(TASK_INTERRUPTIBLE);
1259 __set_current_state(TASK_RUNNING);
1263 static inline int __fastpath_timer_check(struct task_struct *tsk)
1265 /* tsk == current, ensure it is safe to use ->signal/sighand */
1266 if (unlikely(tsk->exit_state))
1269 if (!task_cputime_zero(&tsk->cputime_expires))
1272 if (!task_cputime_zero(&tsk->signal->cputime_expires))
1278 void run_posix_cpu_timers(struct task_struct *tsk)
1280 unsigned long cpu = smp_processor_id();
1281 struct task_struct *tasklist;
1283 BUG_ON(!irqs_disabled());
1284 if(!per_cpu(posix_timer_task, cpu))
1286 /* get per-cpu references */
1287 tasklist = per_cpu(posix_timer_tasklist, cpu);
1289 /* check to see if we're already queued */
1290 if (!tsk->posix_timer_list && __fastpath_timer_check(tsk)) {
1291 get_task_struct(tsk);
1293 tsk->posix_timer_list = tasklist;
1296 * The list is terminated by a self-pointing
1299 tsk->posix_timer_list = tsk;
1301 per_cpu(posix_timer_tasklist, cpu) = tsk;
1303 wake_up_process(per_cpu(posix_timer_task, cpu));
1308 * posix_cpu_thread_call - callback that gets triggered when a CPU is added.
1309 * Here we can start up the necessary migration thread for the new CPU.
1311 static int posix_cpu_thread_call(struct notifier_block *nfb,
1312 unsigned long action, void *hcpu)
1314 int cpu = (long)hcpu;
1315 struct task_struct *p;
1316 struct sched_param param;
1319 case CPU_UP_PREPARE:
1320 p = kthread_create(posix_cpu_timers_thread, hcpu,
1321 "posixcputmr/%d",cpu);
1324 p->flags |= PF_NOFREEZE;
1325 kthread_bind(p, cpu);
1326 /* Must be high prio to avoid getting starved */
1327 param.sched_priority = MAX_RT_PRIO-1;
1328 sched_setscheduler(p, SCHED_FIFO, ¶m);
1329 per_cpu(posix_timer_task,cpu) = p;
1332 /* Strictly unneccessary, as first user will wake it. */
1333 wake_up_process(per_cpu(posix_timer_task,cpu));
1335 #ifdef CONFIG_HOTPLUG_CPU
1336 case CPU_UP_CANCELED:
1337 /* Unbind it from offline cpu so it can run. Fall thru. */
1338 kthread_bind(per_cpu(posix_timer_task, cpu),
1339 cpumask_any(cpu_online_mask));
1340 kthread_stop(per_cpu(posix_timer_task,cpu));
1341 per_cpu(posix_timer_task,cpu) = NULL;
1344 kthread_stop(per_cpu(posix_timer_task,cpu));
1345 per_cpu(posix_timer_task,cpu) = NULL;
1352 /* Register at highest priority so that task migration (migrate_all_tasks)
1353 * happens before everything else.
1355 static struct notifier_block posix_cpu_thread_notifier = {
1356 .notifier_call = posix_cpu_thread_call,
1360 static int __init posix_cpu_thread_init(void)
1362 void *hcpu = (void *)(long)smp_processor_id();
1363 /* Start one for boot CPU. */
1366 /* init the per-cpu posix_timer_tasklets */
1367 for_each_possible_cpu(cpu)
1368 per_cpu(posix_timer_tasklist, cpu) = NULL;
1370 posix_cpu_thread_call(&posix_cpu_thread_notifier, CPU_UP_PREPARE, hcpu);
1371 posix_cpu_thread_call(&posix_cpu_thread_notifier, CPU_ONLINE, hcpu);
1372 register_cpu_notifier(&posix_cpu_thread_notifier);
1375 early_initcall(posix_cpu_thread_init);
1376 #else /* CONFIG_PREEMPT_RT_BASE */
1377 void run_posix_cpu_timers(struct task_struct *tsk)
1379 __run_posix_cpu_timers(tsk);
1381 #endif /* CONFIG_PREEMPT_RT_BASE */
1384 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1385 * The tsk->sighand->siglock must be held by the caller.
1387 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1388 cputime_t *newval, cputime_t *oldval)
1390 unsigned long long now;
1392 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1393 cpu_timer_sample_group(clock_idx, tsk, &now);
1397 * We are setting itimer. The *oldval is absolute and we update
1398 * it to be relative, *newval argument is relative and we update
1399 * it to be absolute.
1402 if (*oldval <= now) {
1403 /* Just about to fire. */
1404 *oldval = cputime_one_jiffy;
1416 * Update expiration cache if we are the earliest timer, or eventually
1417 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1419 switch (clock_idx) {
1421 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1422 tsk->signal->cputime_expires.prof_exp = *newval;
1425 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1426 tsk->signal->cputime_expires.virt_exp = *newval;
1430 posix_cpu_timer_kick_nohz();
1433 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1434 struct timespec *rqtp, struct itimerspec *it)
1436 struct k_itimer timer;
1440 * Set up a temporary timer and then wait for it to go off.
1442 memset(&timer, 0, sizeof timer);
1443 spin_lock_init(&timer.it_lock);
1444 timer.it_clock = which_clock;
1445 timer.it_overrun = -1;
1446 error = posix_cpu_timer_create(&timer);
1447 timer.it_process = current;
1449 static struct itimerspec zero_it;
1451 memset(it, 0, sizeof *it);
1452 it->it_value = *rqtp;
1454 spin_lock_irq(&timer.it_lock);
1455 error = posix_cpu_timer_set(&timer, flags, it, NULL);
1457 spin_unlock_irq(&timer.it_lock);
1461 while (!signal_pending(current)) {
1462 if (timer.it.cpu.expires == 0) {
1464 * Our timer fired and was reset, below
1465 * deletion can not fail.
1467 posix_cpu_timer_del(&timer);
1468 spin_unlock_irq(&timer.it_lock);
1473 * Block until cpu_timer_fire (or a signal) wakes us.
1475 __set_current_state(TASK_INTERRUPTIBLE);
1476 spin_unlock_irq(&timer.it_lock);
1478 spin_lock_irq(&timer.it_lock);
1482 * We were interrupted by a signal.
1484 sample_to_timespec(which_clock, timer.it.cpu.expires, rqtp);
1485 error = posix_cpu_timer_set(&timer, 0, &zero_it, it);
1488 * Timer is now unarmed, deletion can not fail.
1490 posix_cpu_timer_del(&timer);
1492 spin_unlock_irq(&timer.it_lock);
1494 while (error == TIMER_RETRY) {
1496 * We need to handle case when timer was or is in the
1497 * middle of firing. In other cases we already freed
1500 spin_lock_irq(&timer.it_lock);
1501 error = posix_cpu_timer_del(&timer);
1502 spin_unlock_irq(&timer.it_lock);
1505 if ((it->it_value.tv_sec | it->it_value.tv_nsec) == 0) {
1507 * It actually did fire already.
1512 error = -ERESTART_RESTARTBLOCK;
1518 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1520 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1521 struct timespec *rqtp, struct timespec __user *rmtp)
1523 struct restart_block *restart_block = ¤t->restart_block;
1524 struct itimerspec it;
1528 * Diagnose required errors first.
1530 if (CPUCLOCK_PERTHREAD(which_clock) &&
1531 (CPUCLOCK_PID(which_clock) == 0 ||
1532 CPUCLOCK_PID(which_clock) == current->pid))
1535 error = do_cpu_nanosleep(which_clock, flags, rqtp, &it);
1537 if (error == -ERESTART_RESTARTBLOCK) {
1539 if (flags & TIMER_ABSTIME)
1540 return -ERESTARTNOHAND;
1542 * Report back to the user the time still remaining.
1544 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1547 restart_block->fn = posix_cpu_nsleep_restart;
1548 restart_block->nanosleep.clockid = which_clock;
1549 restart_block->nanosleep.rmtp = rmtp;
1550 restart_block->nanosleep.expires = timespec_to_ns(rqtp);
1555 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1557 clockid_t which_clock = restart_block->nanosleep.clockid;
1559 struct itimerspec it;
1562 t = ns_to_timespec(restart_block->nanosleep.expires);
1564 error = do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t, &it);
1566 if (error == -ERESTART_RESTARTBLOCK) {
1567 struct timespec __user *rmtp = restart_block->nanosleep.rmtp;
1569 * Report back to the user the time still remaining.
1571 if (rmtp && copy_to_user(rmtp, &it.it_value, sizeof *rmtp))
1574 restart_block->nanosleep.expires = timespec_to_ns(&t);
1580 #define PROCESS_CLOCK MAKE_PROCESS_CPUCLOCK(0, CPUCLOCK_SCHED)
1581 #define THREAD_CLOCK MAKE_THREAD_CPUCLOCK(0, CPUCLOCK_SCHED)
1583 static int process_cpu_clock_getres(const clockid_t which_clock,
1584 struct timespec *tp)
1586 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1588 static int process_cpu_clock_get(const clockid_t which_clock,
1589 struct timespec *tp)
1591 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1593 static int process_cpu_timer_create(struct k_itimer *timer)
1595 timer->it_clock = PROCESS_CLOCK;
1596 return posix_cpu_timer_create(timer);
1598 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1599 struct timespec *rqtp,
1600 struct timespec __user *rmtp)
1602 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp, rmtp);
1604 static long process_cpu_nsleep_restart(struct restart_block *restart_block)
1608 static int thread_cpu_clock_getres(const clockid_t which_clock,
1609 struct timespec *tp)
1611 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1613 static int thread_cpu_clock_get(const clockid_t which_clock,
1614 struct timespec *tp)
1616 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1618 static int thread_cpu_timer_create(struct k_itimer *timer)
1620 timer->it_clock = THREAD_CLOCK;
1621 return posix_cpu_timer_create(timer);
1624 struct k_clock clock_posix_cpu = {
1625 .clock_getres = posix_cpu_clock_getres,
1626 .clock_set = posix_cpu_clock_set,
1627 .clock_get = posix_cpu_clock_get,
1628 .timer_create = posix_cpu_timer_create,
1629 .nsleep = posix_cpu_nsleep,
1630 .nsleep_restart = posix_cpu_nsleep_restart,
1631 .timer_set = posix_cpu_timer_set,
1632 .timer_del = posix_cpu_timer_del,
1633 .timer_get = posix_cpu_timer_get,
1636 static __init int init_posix_cpu_timers(void)
1638 struct k_clock process = {
1639 .clock_getres = process_cpu_clock_getres,
1640 .clock_get = process_cpu_clock_get,
1641 .timer_create = process_cpu_timer_create,
1642 .nsleep = process_cpu_nsleep,
1643 .nsleep_restart = process_cpu_nsleep_restart,
1645 struct k_clock thread = {
1646 .clock_getres = thread_cpu_clock_getres,
1647 .clock_get = thread_cpu_clock_get,
1648 .timer_create = thread_cpu_timer_create,
1652 posix_timers_register_clock(CLOCK_PROCESS_CPUTIME_ID, &process);
1653 posix_timers_register_clock(CLOCK_THREAD_CPUTIME_ID, &thread);
1655 cputime_to_timespec(cputime_one_jiffy, &ts);
1656 onecputick = ts.tv_nsec;
1657 WARN_ON(ts.tv_sec != 0);
1661 __initcall(init_posix_cpu_timers);