1 // SPDX-License-Identifier: GPL-2.0
3 * Implement CPU time clocks for the POSIX clock interface.
6 #include <linux/sched/signal.h>
7 #include <linux/sched/cputime.h>
8 #include <linux/posix-timers.h>
9 #include <linux/errno.h>
10 #include <linux/math64.h>
11 #include <linux/uaccess.h>
12 #include <linux/kernel_stat.h>
13 #include <trace/events/timer.h>
14 #include <linux/tick.h>
15 #include <linux/workqueue.h>
16 #include <linux/compat.h>
17 #include <linux/sched/deadline.h>
19 #include "posix-timers.h"
21 static void posix_cpu_timer_rearm(struct k_itimer *timer);
24 * Called after updating RLIMIT_CPU to run cpu timer and update
25 * tsk->signal->cputime_expires expiration cache if necessary. Needs
26 * siglock protection since other code may update expiration cache as
29 void update_rlimit_cpu(struct task_struct *task, unsigned long rlim_new)
31 u64 nsecs = rlim_new * NSEC_PER_SEC;
33 spin_lock_irq(&task->sighand->siglock);
34 set_process_cpu_timer(task, CPUCLOCK_PROF, &nsecs, NULL);
35 spin_unlock_irq(&task->sighand->siglock);
38 static int check_clock(const clockid_t which_clock)
41 struct task_struct *p;
42 const pid_t pid = CPUCLOCK_PID(which_clock);
44 if (CPUCLOCK_WHICH(which_clock) >= CPUCLOCK_MAX)
51 p = find_task_by_vpid(pid);
52 if (!p || !(CPUCLOCK_PERTHREAD(which_clock) ?
53 same_thread_group(p, current) : has_group_leader_pid(p))) {
62 * Update expiry time from increment, and increase overrun count,
63 * given the current clock sample.
65 static void bump_cpu_timer(struct k_itimer *timer, u64 now)
70 if (!timer->it_interval)
73 if (now < timer->it.cpu.expires)
76 incr = timer->it_interval;
77 delta = now + incr - timer->it.cpu.expires;
79 /* Don't use (incr*2 < delta), incr*2 might overflow. */
80 for (i = 0; incr < delta - incr; i++)
83 for (; i >= 0; incr >>= 1, i--) {
87 timer->it.cpu.expires += incr;
88 timer->it_overrun += 1LL << i;
94 * task_cputime_zero - Check a task_cputime struct for all zero fields.
96 * @cputime: The struct to compare.
98 * Checks @cputime to see if all fields are zero. Returns true if all fields
99 * are zero, false if any field is nonzero.
101 static inline int task_cputime_zero(const struct task_cputime *cputime)
103 if (!cputime->utime && !cputime->stime && !cputime->sum_exec_runtime)
108 static inline u64 prof_ticks(struct task_struct *p)
112 task_cputime(p, &utime, &stime);
114 return utime + stime;
116 static inline u64 virt_ticks(struct task_struct *p)
120 task_cputime(p, &utime, &stime);
126 posix_cpu_clock_getres(const clockid_t which_clock, struct timespec64 *tp)
128 int error = check_clock(which_clock);
131 tp->tv_nsec = ((NSEC_PER_SEC + HZ - 1) / HZ);
132 if (CPUCLOCK_WHICH(which_clock) == CPUCLOCK_SCHED) {
134 * If sched_clock is using a cycle counter, we
135 * don't have any idea of its true resolution
136 * exported, but it is much more than 1s/HZ.
145 posix_cpu_clock_set(const clockid_t which_clock, const struct timespec64 *tp)
148 * You can never reset a CPU clock, but we check for other errors
149 * in the call before failing with EPERM.
151 int error = check_clock(which_clock);
160 * Sample a per-thread clock for the given task.
162 static int cpu_clock_sample(const clockid_t which_clock,
163 struct task_struct *p, u64 *sample)
165 switch (CPUCLOCK_WHICH(which_clock)) {
169 *sample = prof_ticks(p);
172 *sample = virt_ticks(p);
175 *sample = task_sched_runtime(p);
182 * Set cputime to sum_cputime if sum_cputime > cputime. Use cmpxchg
183 * to avoid race conditions with concurrent updates to cputime.
185 static inline void __update_gt_cputime(atomic64_t *cputime, u64 sum_cputime)
189 curr_cputime = atomic64_read(cputime);
190 if (sum_cputime > curr_cputime) {
191 if (atomic64_cmpxchg(cputime, curr_cputime, sum_cputime) != curr_cputime)
196 static void update_gt_cputime(struct task_cputime_atomic *cputime_atomic, struct task_cputime *sum)
198 __update_gt_cputime(&cputime_atomic->utime, sum->utime);
199 __update_gt_cputime(&cputime_atomic->stime, sum->stime);
200 __update_gt_cputime(&cputime_atomic->sum_exec_runtime, sum->sum_exec_runtime);
203 /* Sample task_cputime_atomic values in "atomic_timers", store results in "times". */
204 static inline void sample_cputime_atomic(struct task_cputime *times,
205 struct task_cputime_atomic *atomic_times)
207 times->utime = atomic64_read(&atomic_times->utime);
208 times->stime = atomic64_read(&atomic_times->stime);
209 times->sum_exec_runtime = atomic64_read(&atomic_times->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;
217 /* Check if cputimer isn't running. This is accessed without locking. */
218 if (!READ_ONCE(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 it.
224 thread_group_cputime(tsk, &sum);
225 update_gt_cputime(&cputimer->cputime_atomic, &sum);
228 * We're setting cputimer->running without a lock. Ensure
229 * this only gets written to in one operation. We set
230 * running after update_gt_cputime() as a small optimization,
231 * but barriers are not required because update_gt_cputime()
232 * can handle concurrent updates.
234 WRITE_ONCE(cputimer->running, true);
236 sample_cputime_atomic(times, &cputimer->cputime_atomic);
240 * Sample a process (thread group) clock for the given group_leader task.
241 * Must be called with task sighand lock held for safe while_each_thread()
244 static int cpu_clock_sample_group(const clockid_t which_clock,
245 struct task_struct *p,
248 struct task_cputime cputime;
250 switch (CPUCLOCK_WHICH(which_clock)) {
254 thread_group_cputime(p, &cputime);
255 *sample = cputime.utime + cputime.stime;
258 thread_group_cputime(p, &cputime);
259 *sample = cputime.utime;
262 thread_group_cputime(p, &cputime);
263 *sample = cputime.sum_exec_runtime;
269 static int posix_cpu_clock_get_task(struct task_struct *tsk,
270 const clockid_t which_clock,
271 struct timespec64 *tp)
276 if (CPUCLOCK_PERTHREAD(which_clock)) {
277 if (same_thread_group(tsk, current))
278 err = cpu_clock_sample(which_clock, tsk, &rtn);
280 if (tsk == current || thread_group_leader(tsk))
281 err = cpu_clock_sample_group(which_clock, tsk, &rtn);
285 *tp = ns_to_timespec64(rtn);
291 static int posix_cpu_clock_get(const clockid_t which_clock, struct timespec64 *tp)
293 const pid_t pid = CPUCLOCK_PID(which_clock);
298 * Special case constant value for our own clocks.
299 * We don't have to do any lookup to find ourselves.
301 err = posix_cpu_clock_get_task(current, which_clock, tp);
304 * Find the given PID, and validate that the caller
305 * should be able to see it.
307 struct task_struct *p;
309 p = find_task_by_vpid(pid);
311 err = posix_cpu_clock_get_task(p, which_clock, tp);
319 * Validate the clockid_t for a new CPU-clock timer, and initialize the timer.
320 * This is called from sys_timer_create() and do_cpu_nanosleep() with the
321 * new timer already all-zeros initialized.
323 static int posix_cpu_timer_create(struct k_itimer *new_timer)
326 const pid_t pid = CPUCLOCK_PID(new_timer->it_clock);
327 struct task_struct *p;
329 if (CPUCLOCK_WHICH(new_timer->it_clock) >= CPUCLOCK_MAX)
332 new_timer->kclock = &clock_posix_cpu;
334 INIT_LIST_HEAD(&new_timer->it.cpu.entry);
337 if (CPUCLOCK_PERTHREAD(new_timer->it_clock)) {
341 p = find_task_by_vpid(pid);
342 if (p && !same_thread_group(p, current))
347 p = current->group_leader;
349 p = find_task_by_vpid(pid);
350 if (p && !has_group_leader_pid(p))
354 new_timer->it.cpu.task = p;
366 * Clean up a CPU-clock timer that is about to be destroyed.
367 * This is called from timer deletion with the timer already locked.
368 * If we return TIMER_RETRY, it's necessary to release the timer's lock
369 * and try again. (This happens when the timer is in the middle of firing.)
371 static int posix_cpu_timer_del(struct k_itimer *timer)
375 struct sighand_struct *sighand;
376 struct task_struct *p = timer->it.cpu.task;
378 WARN_ON_ONCE(p == NULL);
381 * Protect against sighand release/switch in exit/exec and process/
382 * thread timer list entry concurrent read/writes.
384 sighand = lock_task_sighand(p, &flags);
385 if (unlikely(sighand == NULL)) {
387 * We raced with the reaping of the task.
388 * The deletion should have cleared us off the list.
390 WARN_ON_ONCE(!list_empty(&timer->it.cpu.entry));
392 if (timer->it.cpu.firing)
395 list_del(&timer->it.cpu.entry);
397 unlock_task_sighand(p, &flags);
406 static void cleanup_timers_list(struct list_head *head)
408 struct cpu_timer_list *timer, *next;
410 list_for_each_entry_safe(timer, next, head, entry)
411 list_del_init(&timer->entry);
415 * Clean out CPU timers still ticking when a thread exited. The task
416 * pointer is cleared, and the expiry time is replaced with the residual
417 * time for later timer_gettime calls to return.
418 * This must be called with the siglock held.
420 static void cleanup_timers(struct list_head *head)
422 cleanup_timers_list(head);
423 cleanup_timers_list(++head);
424 cleanup_timers_list(++head);
428 * These are both called with the siglock held, when the current thread
429 * is being reaped. When the final (leader) thread in the group is reaped,
430 * posix_cpu_timers_exit_group will be called after posix_cpu_timers_exit.
432 void posix_cpu_timers_exit(struct task_struct *tsk)
434 cleanup_timers(tsk->cpu_timers);
436 void posix_cpu_timers_exit_group(struct task_struct *tsk)
438 cleanup_timers(tsk->signal->cpu_timers);
441 static inline int expires_gt(u64 expires, u64 new_exp)
443 return expires == 0 || expires > new_exp;
447 * Insert the timer on the appropriate list before any timers that
448 * expire later. This must be called with the sighand lock held.
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 u64 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, exp))
488 cputime_expires->prof_exp = exp;
491 if (expires_gt(cputime_expires->virt_exp, exp))
492 cputime_expires->virt_exp = exp;
495 if (expires_gt(cputime_expires->sched_exp, exp))
496 cputime_expires->sched_exp = exp;
499 if (CPUCLOCK_PERTHREAD(timer->it_clock))
500 tick_dep_set_task(p, TICK_DEP_BIT_POSIX_TIMER);
502 tick_dep_set_signal(p->signal, TICK_DEP_BIT_POSIX_TIMER);
507 * The timer is locked, fire it and arrange for its reload.
509 static void cpu_timer_fire(struct k_itimer *timer)
511 if ((timer->it_sigev_notify & ~SIGEV_THREAD_ID) == SIGEV_NONE) {
513 * User don't want any signal.
515 timer->it.cpu.expires = 0;
516 } else if (unlikely(timer->sigq == NULL)) {
518 * This a special case for clock_nanosleep,
519 * not a normal timer from sys_timer_create.
521 wake_up_process(timer->it_process);
522 timer->it.cpu.expires = 0;
523 } else if (!timer->it_interval) {
525 * One-shot timer. Clear it as soon as it's fired.
527 posix_timer_event(timer, 0);
528 timer->it.cpu.expires = 0;
529 } else if (posix_timer_event(timer, ++timer->it_requeue_pending)) {
531 * The signal did not get queued because the signal
532 * was ignored, so we won't get any callback to
533 * reload the timer. But we need to keep it
534 * ticking in case the signal is deliverable next time.
536 posix_cpu_timer_rearm(timer);
537 ++timer->it_requeue_pending;
542 * Sample a process (thread group) timer for the given group_leader task.
543 * Must be called with task sighand lock held for safe while_each_thread()
546 static int cpu_timer_sample_group(const clockid_t which_clock,
547 struct task_struct *p, u64 *sample)
549 struct task_cputime cputime;
551 thread_group_cputimer(p, &cputime);
552 switch (CPUCLOCK_WHICH(which_clock)) {
556 *sample = cputime.utime + cputime.stime;
559 *sample = cputime.utime;
562 *sample = cputime.sum_exec_runtime;
569 * Guts of sys_timer_settime for CPU timers.
570 * This is called with the timer locked and interrupts disabled.
571 * If we return TIMER_RETRY, it's necessary to release the timer's lock
572 * and try again. (This happens when the timer is in the middle of firing.)
574 static int posix_cpu_timer_set(struct k_itimer *timer, int timer_flags,
575 struct itimerspec64 *new, struct itimerspec64 *old)
578 struct sighand_struct *sighand;
579 struct task_struct *p = timer->it.cpu.task;
580 u64 old_expires, new_expires, old_incr, val;
583 WARN_ON_ONCE(p == NULL);
586 * Use the to_ktime conversion because that clamps the maximum
587 * value to KTIME_MAX and avoid multiplication overflows.
589 new_expires = ktime_to_ns(timespec64_to_ktime(new->it_value));
592 * Protect against sighand release/switch in exit/exec and p->cpu_timers
593 * and p->signal->cpu_timers read/write in arm_timer()
595 sighand = lock_task_sighand(p, &flags);
597 * If p has just been reaped, we can no
598 * longer get any information about it at all.
600 if (unlikely(sighand == NULL)) {
605 * Disarm any old timer after extracting its expiry time.
609 old_incr = timer->it_interval;
610 old_expires = timer->it.cpu.expires;
611 if (unlikely(timer->it.cpu.firing)) {
612 timer->it.cpu.firing = -1;
615 list_del_init(&timer->it.cpu.entry);
618 * We need to sample the current value to convert the new
619 * value from to relative and absolute, and to convert the
620 * old value from absolute to relative. To set a process
621 * timer, we need a sample to balance the thread expiry
622 * times (in arm_timer). With an absolute time, we must
623 * check if it's already passed. In short, we need a sample.
625 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
626 cpu_clock_sample(timer->it_clock, p, &val);
628 cpu_timer_sample_group(timer->it_clock, p, &val);
632 if (old_expires == 0) {
633 old->it_value.tv_sec = 0;
634 old->it_value.tv_nsec = 0;
637 * Update the timer in case it has
638 * overrun already. If it has,
639 * we'll report it as having overrun
640 * and with the next reloaded timer
641 * already ticking, though we are
642 * swallowing that pending
643 * notification here to install the
646 bump_cpu_timer(timer, val);
647 if (val < timer->it.cpu.expires) {
648 old_expires = timer->it.cpu.expires - val;
649 old->it_value = ns_to_timespec64(old_expires);
651 old->it_value.tv_nsec = 1;
652 old->it_value.tv_sec = 0;
659 * We are colliding with the timer actually firing.
660 * Punt after filling in the timer's old value, and
661 * disable this firing since we are already reporting
662 * it as an overrun (thanks to bump_cpu_timer above).
664 unlock_task_sighand(p, &flags);
668 if (new_expires != 0 && !(timer_flags & TIMER_ABSTIME)) {
673 * Install the new expiry time (or zero).
674 * For a timer with no notification action, we don't actually
675 * arm the timer (we'll just fake it for timer_gettime).
677 timer->it.cpu.expires = new_expires;
678 if (new_expires != 0 && val < new_expires) {
682 unlock_task_sighand(p, &flags);
684 * Install the new reload setting, and
685 * set up the signal and overrun bookkeeping.
687 timer->it_interval = timespec64_to_ktime(new->it_interval);
690 * This acts as a modification timestamp for the timer,
691 * so any automatic reload attempt will punt on seeing
692 * that we have reset the timer manually.
694 timer->it_requeue_pending = (timer->it_requeue_pending + 2) &
696 timer->it_overrun_last = 0;
697 timer->it_overrun = -1;
699 if (new_expires != 0 && !(val < new_expires)) {
701 * The designated time already passed, so we notify
702 * immediately, even if the thread never runs to
703 * accumulate more time on this clock.
705 cpu_timer_fire(timer);
711 old->it_interval = ns_to_timespec64(old_incr);
716 static void posix_cpu_timer_get(struct k_itimer *timer, struct itimerspec64 *itp)
719 struct task_struct *p = timer->it.cpu.task;
721 WARN_ON_ONCE(p == NULL);
724 * Easy part: convert the reload time.
726 itp->it_interval = ktime_to_timespec64(timer->it_interval);
728 if (!timer->it.cpu.expires)
732 * Sample the clock to take the difference with the expiry time.
734 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
735 cpu_clock_sample(timer->it_clock, p, &now);
737 struct sighand_struct *sighand;
741 * Protect against sighand release/switch in exit/exec and
742 * also make timer sampling safe if it ends up calling
743 * thread_group_cputime().
745 sighand = lock_task_sighand(p, &flags);
746 if (unlikely(sighand == NULL)) {
748 * The process has been reaped.
749 * We can't even collect a sample any more.
750 * Call the timer disarmed, nothing else to do.
752 timer->it.cpu.expires = 0;
755 cpu_timer_sample_group(timer->it_clock, p, &now);
756 unlock_task_sighand(p, &flags);
760 if (now < timer->it.cpu.expires) {
761 itp->it_value = ns_to_timespec64(timer->it.cpu.expires - now);
764 * The timer should have expired already, but the firing
765 * hasn't taken place yet. Say it's just about to expire.
767 itp->it_value.tv_nsec = 1;
768 itp->it_value.tv_sec = 0;
772 static unsigned long long
773 check_timers_list(struct list_head *timers,
774 struct list_head *firing,
775 unsigned long long curr)
779 while (!list_empty(timers)) {
780 struct cpu_timer_list *t;
782 t = list_first_entry(timers, struct cpu_timer_list, entry);
784 if (!--maxfire || curr < t->expires)
788 list_move_tail(&t->entry, firing);
794 static inline void check_dl_overrun(struct task_struct *tsk)
796 if (tsk->dl.dl_overrun) {
797 tsk->dl.dl_overrun = 0;
798 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
803 * Check for any per-thread CPU timers that have fired and move them off
804 * the tsk->cpu_timers[N] list onto the firing list. Here we update the
805 * tsk->it_*_expires values to reflect the remaining thread CPU timers.
807 static void check_thread_timers(struct task_struct *tsk,
808 struct list_head *firing)
810 struct list_head *timers = tsk->cpu_timers;
811 struct task_cputime *tsk_expires = &tsk->cputime_expires;
816 check_dl_overrun(tsk);
819 * If cputime_expires is zero, then there are no active
820 * per thread CPU timers.
822 if (task_cputime_zero(&tsk->cputime_expires))
825 expires = check_timers_list(timers, firing, prof_ticks(tsk));
826 tsk_expires->prof_exp = expires;
828 expires = check_timers_list(++timers, firing, virt_ticks(tsk));
829 tsk_expires->virt_exp = expires;
831 tsk_expires->sched_exp = check_timers_list(++timers, firing,
832 tsk->se.sum_exec_runtime);
835 * Check for the special case thread timers.
837 soft = task_rlimit(tsk, RLIMIT_RTTIME);
838 if (soft != RLIM_INFINITY) {
839 unsigned long hard = task_rlimit_max(tsk, RLIMIT_RTTIME);
841 if (hard != RLIM_INFINITY &&
842 tsk->rt.timeout > DIV_ROUND_UP(hard, USEC_PER_SEC/HZ)) {
844 * At the hard limit, we just die.
845 * No need to calculate anything else now.
847 if (print_fatal_signals) {
848 pr_info("CPU Watchdog Timeout (hard): %s[%d]\n",
849 tsk->comm, task_pid_nr(tsk));
851 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
854 if (tsk->rt.timeout > DIV_ROUND_UP(soft, USEC_PER_SEC/HZ)) {
856 * At the soft limit, send a SIGXCPU every second.
859 soft += USEC_PER_SEC;
860 tsk->signal->rlim[RLIMIT_RTTIME].rlim_cur =
863 if (print_fatal_signals) {
864 pr_info("RT Watchdog Timeout (soft): %s[%d]\n",
865 tsk->comm, task_pid_nr(tsk));
867 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
870 if (task_cputime_zero(tsk_expires))
871 tick_dep_clear_task(tsk, TICK_DEP_BIT_POSIX_TIMER);
874 static inline void stop_process_timers(struct signal_struct *sig)
876 struct thread_group_cputimer *cputimer = &sig->cputimer;
878 /* Turn off cputimer->running. This is done without locking. */
879 WRITE_ONCE(cputimer->running, false);
880 tick_dep_clear_signal(sig, TICK_DEP_BIT_POSIX_TIMER);
883 static void check_cpu_itimer(struct task_struct *tsk, struct cpu_itimer *it,
884 u64 *expires, u64 cur_time, int signo)
889 if (cur_time >= it->expires) {
891 it->expires += it->incr;
895 trace_itimer_expire(signo == SIGPROF ?
896 ITIMER_PROF : ITIMER_VIRTUAL,
897 task_tgid(tsk), cur_time);
898 __group_send_sig_info(signo, SEND_SIG_PRIV, tsk);
901 if (it->expires && (!*expires || it->expires < *expires))
902 *expires = it->expires;
906 * Check for any per-thread CPU timers that have fired and move them
907 * off the tsk->*_timers list onto the firing list. Per-thread timers
908 * have already been taken off.
910 static void check_process_timers(struct task_struct *tsk,
911 struct list_head *firing)
913 struct signal_struct *const sig = tsk->signal;
914 u64 utime, ptime, virt_expires, prof_expires;
915 u64 sum_sched_runtime, sched_expires;
916 struct list_head *timers = sig->cpu_timers;
917 struct task_cputime cputime;
921 * If cputimer is not running, then there are no active
922 * process wide timers (POSIX 1.b, itimers, RLIMIT_CPU).
924 if (!READ_ONCE(tsk->signal->cputimer.running))
928 * Signify that a thread is checking for process timers.
929 * Write access to this field is protected by the sighand lock.
931 sig->cputimer.checking_timer = true;
934 * Collect the current process totals.
936 thread_group_cputimer(tsk, &cputime);
937 utime = cputime.utime;
938 ptime = utime + 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 = task_rlimit(tsk, RLIMIT_CPU);
953 if (soft != RLIM_INFINITY) {
954 unsigned long psecs = div_u64(ptime, NSEC_PER_SEC);
955 unsigned long hard = task_rlimit_max(tsk, RLIMIT_CPU);
959 * At the hard limit, we just die.
960 * No need to calculate anything else now.
962 if (print_fatal_signals) {
963 pr_info("RT Watchdog Timeout (hard): %s[%d]\n",
964 tsk->comm, task_pid_nr(tsk));
966 __group_send_sig_info(SIGKILL, SEND_SIG_PRIV, tsk);
971 * At the soft limit, send a SIGXCPU every second.
973 if (print_fatal_signals) {
974 pr_info("CPU Watchdog Timeout (soft): %s[%d]\n",
975 tsk->comm, task_pid_nr(tsk));
977 __group_send_sig_info(SIGXCPU, SEND_SIG_PRIV, tsk);
980 sig->rlim[RLIMIT_CPU].rlim_cur = soft;
983 x = soft * NSEC_PER_SEC;
984 if (!prof_expires || x < prof_expires)
988 sig->cputime_expires.prof_exp = prof_expires;
989 sig->cputime_expires.virt_exp = virt_expires;
990 sig->cputime_expires.sched_exp = sched_expires;
991 if (task_cputime_zero(&sig->cputime_expires))
992 stop_process_timers(sig);
994 sig->cputimer.checking_timer = false;
998 * This is called from the signal code (via posixtimer_rearm)
999 * when the last timer signal was delivered and we have to reload the timer.
1001 static void posix_cpu_timer_rearm(struct k_itimer *timer)
1003 struct sighand_struct *sighand;
1004 unsigned long flags;
1005 struct task_struct *p = timer->it.cpu.task;
1008 WARN_ON_ONCE(p == NULL);
1011 * Fetch the current sample and update the timer's expiry time.
1013 if (CPUCLOCK_PERTHREAD(timer->it_clock)) {
1014 cpu_clock_sample(timer->it_clock, p, &now);
1015 bump_cpu_timer(timer, now);
1016 if (unlikely(p->exit_state))
1019 /* Protect timer list r/w in arm_timer() */
1020 sighand = lock_task_sighand(p, &flags);
1025 * Protect arm_timer() and timer sampling in case of call to
1026 * thread_group_cputime().
1028 sighand = lock_task_sighand(p, &flags);
1029 if (unlikely(sighand == NULL)) {
1031 * The process has been reaped.
1032 * We can't even collect a sample any more.
1034 timer->it.cpu.expires = 0;
1036 } else if (unlikely(p->exit_state) && thread_group_empty(p)) {
1037 /* If the process is dying, no need to rearm */
1040 cpu_timer_sample_group(timer->it_clock, p, &now);
1041 bump_cpu_timer(timer, now);
1042 /* Leave the sighand locked for the call below. */
1046 * Now re-arm for the new expiry time.
1050 unlock_task_sighand(p, &flags);
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;
1090 if (!task_cputime_zero(&tsk->cputime_expires)) {
1091 struct task_cputime task_sample;
1093 task_cputime(tsk, &task_sample.utime, &task_sample.stime);
1094 task_sample.sum_exec_runtime = tsk->se.sum_exec_runtime;
1095 if (task_cputime_expired(&task_sample, &tsk->cputime_expires))
1101 * Check if thread group timers expired when the cputimer is
1102 * running and no other thread in the group is already checking
1103 * for thread group cputimers. These fields are read without the
1104 * sighand lock. However, this is fine because this is meant to
1105 * be a fastpath heuristic to determine whether we should try to
1106 * acquire the sighand lock to check/handle timers.
1108 * In the worst case scenario, if 'running' or 'checking_timer' gets
1109 * set but the current thread doesn't see the change yet, we'll wait
1110 * until the next thread in the group gets a scheduler interrupt to
1111 * handle the timer. This isn't an issue in practice because these
1112 * types of delays with signals actually getting sent are expected.
1114 if (READ_ONCE(sig->cputimer.running) &&
1115 !READ_ONCE(sig->cputimer.checking_timer)) {
1116 struct task_cputime group_sample;
1118 sample_cputime_atomic(&group_sample, &sig->cputimer.cputime_atomic);
1120 if (task_cputime_expired(&group_sample, &sig->cputime_expires))
1124 if (dl_task(tsk) && tsk->dl.dl_overrun)
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 void run_posix_cpu_timers(struct task_struct *tsk)
1138 struct k_itimer *timer, *next;
1139 unsigned long flags;
1141 lockdep_assert_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 check_process_timers(tsk, &firing);
1162 * We must release these locks before taking any timer's lock.
1163 * There is a potential race with timer deletion here, as the
1164 * siglock now protects our private firing list. We have set
1165 * the firing flag in each timer, so that a deletion attempt
1166 * that gets the timer lock before we do will give it up and
1167 * spin until we've taken care of that timer below.
1169 unlock_task_sighand(tsk, &flags);
1172 * Now that all the timers on our list have the firing flag,
1173 * no one will touch their list entries but us. We'll take
1174 * each timer's lock before clearing its firing flag, so no
1175 * timer call will interfere.
1177 list_for_each_entry_safe(timer, next, &firing, it.cpu.entry) {
1180 spin_lock(&timer->it_lock);
1181 list_del_init(&timer->it.cpu.entry);
1182 cpu_firing = timer->it.cpu.firing;
1183 timer->it.cpu.firing = 0;
1185 * The firing flag is -1 if we collided with a reset
1186 * of the timer, which already reported this
1187 * almost-firing as an overrun. So don't generate an event.
1189 if (likely(cpu_firing >= 0))
1190 cpu_timer_fire(timer);
1191 spin_unlock(&timer->it_lock);
1196 * Set one of the process-wide special case CPU timers or RLIMIT_CPU.
1197 * The tsk->sighand->siglock must be held by the caller.
1199 void set_process_cpu_timer(struct task_struct *tsk, unsigned int clock_idx,
1200 u64 *newval, u64 *oldval)
1205 WARN_ON_ONCE(clock_idx == CPUCLOCK_SCHED);
1206 ret = cpu_timer_sample_group(clock_idx, tsk, &now);
1208 if (oldval && ret != -EINVAL) {
1210 * We are setting itimer. The *oldval is absolute and we update
1211 * it to be relative, *newval argument is relative and we update
1212 * it to be absolute.
1215 if (*oldval <= now) {
1216 /* Just about to fire. */
1217 *oldval = TICK_NSEC;
1229 * Update expiration cache if we are the earliest timer, or eventually
1230 * RLIMIT_CPU limit is earlier than prof_exp cpu timer expire.
1232 switch (clock_idx) {
1234 if (expires_gt(tsk->signal->cputime_expires.prof_exp, *newval))
1235 tsk->signal->cputime_expires.prof_exp = *newval;
1238 if (expires_gt(tsk->signal->cputime_expires.virt_exp, *newval))
1239 tsk->signal->cputime_expires.virt_exp = *newval;
1243 tick_dep_set_signal(tsk->signal, TICK_DEP_BIT_POSIX_TIMER);
1246 static int do_cpu_nanosleep(const clockid_t which_clock, int flags,
1247 const struct timespec64 *rqtp)
1249 struct itimerspec64 it;
1250 struct k_itimer timer;
1255 * Set up a temporary timer and then wait for it to go off.
1257 memset(&timer, 0, sizeof timer);
1258 spin_lock_init(&timer.it_lock);
1259 timer.it_clock = which_clock;
1260 timer.it_overrun = -1;
1261 error = posix_cpu_timer_create(&timer);
1262 timer.it_process = current;
1264 static struct itimerspec64 zero_it;
1265 struct restart_block *restart;
1267 memset(&it, 0, sizeof(it));
1268 it.it_value = *rqtp;
1270 spin_lock_irq(&timer.it_lock);
1271 error = posix_cpu_timer_set(&timer, flags, &it, NULL);
1273 spin_unlock_irq(&timer.it_lock);
1277 while (!signal_pending(current)) {
1278 if (timer.it.cpu.expires == 0) {
1280 * Our timer fired and was reset, below
1281 * deletion can not fail.
1283 posix_cpu_timer_del(&timer);
1284 spin_unlock_irq(&timer.it_lock);
1289 * Block until cpu_timer_fire (or a signal) wakes us.
1291 __set_current_state(TASK_INTERRUPTIBLE);
1292 spin_unlock_irq(&timer.it_lock);
1294 spin_lock_irq(&timer.it_lock);
1298 * We were interrupted by a signal.
1300 expires = timer.it.cpu.expires;
1301 error = posix_cpu_timer_set(&timer, 0, &zero_it, &it);
1304 * Timer is now unarmed, deletion can not fail.
1306 posix_cpu_timer_del(&timer);
1308 spin_unlock_irq(&timer.it_lock);
1310 while (error == TIMER_RETRY) {
1312 * We need to handle case when timer was or is in the
1313 * middle of firing. In other cases we already freed
1316 spin_lock_irq(&timer.it_lock);
1317 error = posix_cpu_timer_del(&timer);
1318 spin_unlock_irq(&timer.it_lock);
1321 if ((it.it_value.tv_sec | it.it_value.tv_nsec) == 0) {
1323 * It actually did fire already.
1328 error = -ERESTART_RESTARTBLOCK;
1330 * Report back to the user the time still remaining.
1332 restart = ¤t->restart_block;
1333 restart->nanosleep.expires = expires;
1334 if (restart->nanosleep.type != TT_NONE)
1335 error = nanosleep_copyout(restart, &it.it_value);
1341 static long posix_cpu_nsleep_restart(struct restart_block *restart_block);
1343 static int posix_cpu_nsleep(const clockid_t which_clock, int flags,
1344 const struct timespec64 *rqtp)
1346 struct restart_block *restart_block = ¤t->restart_block;
1350 * Diagnose required errors first.
1352 if (CPUCLOCK_PERTHREAD(which_clock) &&
1353 (CPUCLOCK_PID(which_clock) == 0 ||
1354 CPUCLOCK_PID(which_clock) == task_pid_vnr(current)))
1357 error = do_cpu_nanosleep(which_clock, flags, rqtp);
1359 if (error == -ERESTART_RESTARTBLOCK) {
1361 if (flags & TIMER_ABSTIME)
1362 return -ERESTARTNOHAND;
1364 restart_block->fn = posix_cpu_nsleep_restart;
1365 restart_block->nanosleep.clockid = which_clock;
1370 static long posix_cpu_nsleep_restart(struct restart_block *restart_block)
1372 clockid_t which_clock = restart_block->nanosleep.clockid;
1373 struct timespec64 t;
1375 t = ns_to_timespec64(restart_block->nanosleep.expires);
1377 return do_cpu_nanosleep(which_clock, TIMER_ABSTIME, &t);
1380 #define PROCESS_CLOCK make_process_cpuclock(0, CPUCLOCK_SCHED)
1381 #define THREAD_CLOCK make_thread_cpuclock(0, CPUCLOCK_SCHED)
1383 static int process_cpu_clock_getres(const clockid_t which_clock,
1384 struct timespec64 *tp)
1386 return posix_cpu_clock_getres(PROCESS_CLOCK, tp);
1388 static int process_cpu_clock_get(const clockid_t which_clock,
1389 struct timespec64 *tp)
1391 return posix_cpu_clock_get(PROCESS_CLOCK, tp);
1393 static int process_cpu_timer_create(struct k_itimer *timer)
1395 timer->it_clock = PROCESS_CLOCK;
1396 return posix_cpu_timer_create(timer);
1398 static int process_cpu_nsleep(const clockid_t which_clock, int flags,
1399 const struct timespec64 *rqtp)
1401 return posix_cpu_nsleep(PROCESS_CLOCK, flags, rqtp);
1403 static int thread_cpu_clock_getres(const clockid_t which_clock,
1404 struct timespec64 *tp)
1406 return posix_cpu_clock_getres(THREAD_CLOCK, tp);
1408 static int thread_cpu_clock_get(const clockid_t which_clock,
1409 struct timespec64 *tp)
1411 return posix_cpu_clock_get(THREAD_CLOCK, tp);
1413 static int thread_cpu_timer_create(struct k_itimer *timer)
1415 timer->it_clock = THREAD_CLOCK;
1416 return posix_cpu_timer_create(timer);
1419 const struct k_clock clock_posix_cpu = {
1420 .clock_getres = posix_cpu_clock_getres,
1421 .clock_set = posix_cpu_clock_set,
1422 .clock_get = posix_cpu_clock_get,
1423 .timer_create = posix_cpu_timer_create,
1424 .nsleep = posix_cpu_nsleep,
1425 .timer_set = posix_cpu_timer_set,
1426 .timer_del = posix_cpu_timer_del,
1427 .timer_get = posix_cpu_timer_get,
1428 .timer_rearm = posix_cpu_timer_rearm,
1431 const struct k_clock clock_process = {
1432 .clock_getres = process_cpu_clock_getres,
1433 .clock_get = process_cpu_clock_get,
1434 .timer_create = process_cpu_timer_create,
1435 .nsleep = process_cpu_nsleep,
1438 const struct k_clock clock_thread = {
1439 .clock_getres = thread_cpu_clock_getres,
1440 .clock_get = thread_cpu_clock_get,
1441 .timer_create = thread_cpu_timer_create,